Merge branch 'main' into paulfradgley-semaphore_export_types

CMakeLists.txt

@@ -105,8 +105,6 @@ if(CMAKE_COMPILER_IS_GNUCC OR "${CMAKE_CXX_COMPILER_ID}" MATCHES "(Apple)?Clang"
         add_cxx_flag_if_supported(-Wall)
         # Suppress warnings that currently trigger on the code base.
         # This list should shrink over time when warnings are fixed.
-        add_cxx_flag_if_supported(-Wno-unused-but-set-variable)
-        add_cxx_flag_if_supported(-Wno-sometimes-uninitialized)
         add_cxx_flag_if_supported(-Wno-sign-compare)
     endif()
     add_cxx_flag_if_supported(-Wno-narrowing)

presubmit.sh

@@ -77,7 +77,6 @@ cmake .. -G Ninja \
       -DBUILD_WSI_XLIB_SUPPORT=OFF \
       -DBUILD_WSI_XCB_SUPPORT=OFF \
       -DBUILD_WSI_WAYLAND_SUPPORT=OFF \
-      -DUSE_GAS=OFF \
       -C helper.cmake ..
 cmake --build . -j2
 

test_common/harness/compat.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _COMPAT_H_
-#define _COMPAT_H_
+#ifndef COMPAT_H_
+#define COMPAT_H_
 
 #if defined(_WIN32) && defined(_MSC_VER)
 #include <Windows.h>
@@ -398,4 +398,4 @@ EXTERN_C int __builtin_clz(unsigned int pattern);
 #define sleep(sec) Sleep((sec)*1000)
 #endif
 
-#endif // _COMPAT_H_
+#endif // COMPAT_H_

test_common/harness/crc32.h

@@ -15,8 +15,8 @@ Agreement or Khronos Conformance Test Source License Agreement as
 executed between Khronos and the recipient.
 ******************************************************************/
 
-#ifndef _CRC32_H_
-#define _CRC32_H_
+#ifndef CRC32_H_
+#define CRC32_H_
 
 #include <stdint.h>
 #include <stddef.h>

test_conformance/SVM/CMakeLists.txt

@@ -17,4 +17,6 @@ set(${MODULE_NAME}_SOURCES
     test_migrate.cpp
 )
 
+set_gnulike_module_compile_flags("-Wno-sometimes-uninitialized")
+
 include(../CMakeCommon.txt)

test_conformance/allocations/allocation_execute.cpp

@@ -79,20 +79,30 @@ int check_image(cl_command_queue queue, cl_mem mem) {
         return -1;
     }
 
-    if (type == CL_MEM_OBJECT_BUFFER) {
+    switch (type)
+    {
+        case CL_MEM_OBJECT_BUFFER:
             log_error("Expected image object, not buffer.\n");
             return -1;
-    } else if (type == CL_MEM_OBJECT_IMAGE2D) {
-        error = clGetImageInfo(mem, CL_IMAGE_WIDTH, sizeof(width), &width, NULL);
-        if (error) {
-            print_error(error, "clGetMemObjectInfo failed for CL_IMAGE_WIDTH.");
+        case CL_MEM_OBJECT_IMAGE2D:
+            error = clGetImageInfo(mem, CL_IMAGE_WIDTH, sizeof(width), &width,
+                                   NULL);
+            if (error)
+            {
+                print_error(error,
+                            "clGetMemObjectInfo failed for CL_IMAGE_WIDTH.");
                 return -1;
             }
-        error = clGetImageInfo(mem, CL_IMAGE_HEIGHT, sizeof(height), &height, NULL);
-        if (error) {
-            print_error(error, "clGetMemObjectInfo failed for CL_IMAGE_HEIGHT.");
+            error = clGetImageInfo(mem, CL_IMAGE_HEIGHT, sizeof(height),
+                                   &height, NULL);
+            if (error)
+            {
+                print_error(error,
+                            "clGetMemObjectInfo failed for CL_IMAGE_HEIGHT.");
                 return -1;
             }
+            break;
+        default: log_error("unexpected object type"); return -1;
     }
 
 

test_conformance/api/test_null_buffer_arg.cpp

@@ -64,16 +64,21 @@ static int test_setargs_and_execution(cl_command_queue queue, cl_kernel kernel,
     cl_int status;
     const char *typestr;
 
-    if (type == NON_NULL_PATH) {
+    switch (type)
+    {
+        case NON_NULL_PATH:
             status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &test_buf);
             typestr = "non-NULL";
-    } else if (type == ADDROF_NULL_PATH) {
+            break;
+        case ADDROF_NULL_PATH:
             test_buf = NULL;
             status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &test_buf);
             typestr = "&NULL";
-    } else if (type == NULL_PATH) {
+            break;
+        case NULL_PATH:
             status = clSetKernelArg(kernel, 0, sizeof(cl_mem), NULL);
             typestr = "NULL";
+            break;
     }
 
     log_info("Testing setKernelArgs with %s buffer.\n", typestr);

test_conformance/atomics/test_indexed_cases.cpp

@@ -201,7 +201,6 @@ int add_index_bin_test(size_t *global_threads, cl_command_queue queue,
     int number_of_bins = number_of_items / divisor;
     int max_counts_per_bin = divisor * 2;
 
-    int fail = 0;
     int err;
 
     clProgramWrapper program;
@@ -345,7 +344,6 @@ int add_index_bin_test(size_t *global_threads, cl_command_queue queue,
     {
         log_error("add_index_bin_test FAILED to set kernel arguments: %d\n",
                   err);
-        fail = 1;
         return -1;
     }
 
@@ -354,7 +352,7 @@ int add_index_bin_test(size_t *global_threads, cl_command_queue queue,
     if (err)
     {
         log_error("add_index_bin_test FAILED to execute kernel: %d\n", err);
-        fail = 1;
+        return -1;
     }
 
     cl_int *final_bin_assignments =
@@ -372,7 +370,7 @@ int add_index_bin_test(size_t *global_threads, cl_command_queue queue,
     if (err)
     {
         log_error("add_index_bin_test FAILED to read back bins: %d\n", err);
-        fail = 1;
+        return -1;
     }
 
     cl_int *final_bin_counts =
@@ -390,7 +388,7 @@ int add_index_bin_test(size_t *global_threads, cl_command_queue queue,
     {
         log_error("add_index_bin_test FAILED to read back bin_counters: %d\n",
                   err);
-        fail = 1;
+        return -1;
     }
 
     // Verification.

test_conformance/basic/CMakeLists.txt

@@ -11,7 +11,7 @@ set(${MODULE_NAME}_SOURCES
     test_multireadimageonefmt.cpp test_multireadimagemultifmt.cpp
     test_imagedim.cpp
     test_vloadstore.cpp
-    test_int2float.cpp test_float2int.cpp
+    test_int2float.cpp
     test_createkernelsinprogram.cpp
     test_hostptr.cpp
     test_explicit_s2v.cpp
@@ -70,4 +70,6 @@ if(APPLE)
     list(APPEND ${MODULE_NAME}_SOURCES test_queue_priority.cpp)
 endif(APPLE)
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../CMakeCommon.txt)

test_conformance/basic/test_astype.cpp

@@ -15,61 +15,39 @@
 //
 #include "harness/compat.h"
 
+#include <limits.h>
 #include <stdio.h>
 #include <string.h>
-#include <limits.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <vector>
 
-
-#include "procs.h"
 #include "harness/conversions.h"
 #include "harness/typeWrappers.h"
 
+#include "procs.h"
+#include "utils.h"
 
-static const char *astype_kernel_pattern =
-"%s\n"
+// clang-format off
+
+static char extension[128] = { 0 };
+static char strLoad[128] = { 0 };
+static char strStore[128] = { 0 };
+static const char *regLoad = "as_%s%s(src[tid]);\n";
+static const char *v3Load = "as_%s%s(vload3(tid,(__global %s*)src));\n";
+static const char *regStore = "dst[tid] = tmp;\n";
+static const char *v3Store = "vstore3(tmp, tid, (__global %s*)dst);\n";
+
+static const char* astype_kernel_pattern[] = {
+extension,
 "__kernel void test_fn( __global %s%s *src, __global %s%s *dst )\n"
 "{\n"
-"    int tid = get_global_id( 0 );\n"
-"    %s%s tmp = as_%s%s( src[ tid ] );\n"
-"   dst[ tid ] = tmp;\n"
-"}\n";
-
-static const char *astype_kernel_pattern_V3srcV3dst =
-"%s\n"
-"__kernel void test_fn( __global %s *src, __global %s *dst )\n"
-"{\n"
-"    int tid = get_global_id( 0 );\n"
-"    %s%s tmp = as_%s%s( vload3(tid,src) );\n"
-"   vstore3(tmp,tid,dst);\n"
-"}\n";
-// in the printf, remove the third and fifth argument, each of which
-// should be a "3", when copying from the printf for astype_kernel_pattern
-
-static const char *astype_kernel_pattern_V3dst =
-"%s\n"
-"__kernel void test_fn( __global %s%s *src, __global %s *dst )\n"
-"{\n"
-"    int tid = get_global_id( 0 );\n"
-"    %s3 tmp = as_%s3( src[ tid ] );\n"
-"   vstore3(tmp,tid,dst);\n"
-"}\n";
-// in the printf, remove the fifth argument, which
-// should be a "3", when copying from the printf for astype_kernel_pattern
-
-
-static const char *astype_kernel_pattern_V3src =
-"%s\n"
-"__kernel void test_fn( __global %s *src, __global %s%s *dst )\n"
-"{\n"
-"    int tid = get_global_id( 0 );\n"
-"    %s%s tmp = as_%s%s( vload3(tid,src) );\n"
-"   dst[ tid ] = tmp;\n"
-"}\n";
-// in the printf, remove the third argument, which
-// should be a "3", when copying from the printf for astype_kernel_pattern
+"    int tid = get_global_id( 0 );\n",
+"    %s%s tmp = ", strLoad,
+"    ", strStore,
+"}\n"};
 
+// clang-format on
 
 int test_astype_set( cl_device_id device, cl_context context, cl_command_queue queue, ExplicitType inVecType, ExplicitType outVecType,
                     unsigned int vecSize, unsigned int outVecSize,
@@ -81,68 +59,60 @@ int test_astype_set( cl_device_id device, cl_context context, cl_command_queue q
     clKernelWrapper kernel;
     clMemWrapper streams[ 2 ];
 
-    char programSrc[ 10240 ];
     size_t threads[ 1 ], localThreads[ 1 ];
     size_t typeSize = get_explicit_type_size( inVecType );
     size_t outTypeSize = get_explicit_type_size(outVecType);
     char sizeNames[][ 3 ] = { "", "", "2", "3", "4", "", "", "", "8", "", "", "", "", "", "", "", "16" };
-    MTdata d;
+    MTdataHolder d(gRandomSeed);
 
+    std::ostringstream sstr;
+    if (outVecType == kDouble || inVecType == kDouble)
+        sstr << "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
 
+    if (outVecType == kHalf || inVecType == kHalf)
+        sstr << "#pragma OPENCL EXTENSION cl_khr_fp16 : enable\n";
 
-    // Create program
-    if(outVecSize == 3 && vecSize == 3) {
-        // astype_kernel_pattern_V3srcV3dst
-        sprintf( programSrc, astype_kernel_pattern_V3srcV3dst,
-                (outVecType == kDouble || inVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-                get_explicit_type_name( inVecType ), // sizeNames[ vecSize ],
-                get_explicit_type_name( outVecType ), // sizeNames[ outVecSize ],
+    strcpy(extension, sstr.str().c_str());
+
+    if (vecSize == 3)
+        std::snprintf(strLoad, sizeof(strLoad), v3Load,
                       get_explicit_type_name(outVecType), sizeNames[outVecSize],
-                get_explicit_type_name( outVecType ), sizeNames[ outVecSize ] );
-    } else if(outVecSize == 3) {
-        // astype_kernel_pattern_V3dst
-        sprintf( programSrc, astype_kernel_pattern_V3dst,
-                (outVecType == kDouble || inVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-                get_explicit_type_name( inVecType ), sizeNames[ vecSize ],
-                get_explicit_type_name( outVecType ),
+                      get_explicit_type_name(inVecType));
+    else
+        std::snprintf(strLoad, sizeof(strLoad), regLoad,
                       get_explicit_type_name(outVecType),
+                      sizeNames[outVecSize]);
+
+    if (outVecSize == 3)
+        std::snprintf(strStore, sizeof(strStore), v3Store,
                       get_explicit_type_name(outVecType));
+    else
+        std::snprintf(strStore, sizeof(strStore), "%s", regStore);
 
-    } else if(vecSize == 3) {
-        // astype_kernel_pattern_V3src
-        sprintf( programSrc, astype_kernel_pattern_V3src,
-                (outVecType == kDouble || inVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-                get_explicit_type_name( inVecType ),// sizeNames[ vecSize ],
-                get_explicit_type_name( outVecType ), sizeNames[ outVecSize ],
+    auto str =
+        concat_kernel(astype_kernel_pattern,
+                      sizeof(astype_kernel_pattern) / sizeof(const char *));
+    std::string kernelSource =
+        str_sprintf(str, get_explicit_type_name(inVecType), sizeNames[vecSize],
                     get_explicit_type_name(outVecType), sizeNames[outVecSize],
                     get_explicit_type_name(outVecType), sizeNames[outVecSize]);
-    } else {
-        sprintf( programSrc, astype_kernel_pattern,
-                (outVecType == kDouble || inVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-                get_explicit_type_name( inVecType ), sizeNames[ vecSize ],
-                get_explicit_type_name( outVecType ), sizeNames[ outVecSize ],
-                get_explicit_type_name( outVecType ), sizeNames[ outVecSize ],
-                get_explicit_type_name( outVecType ), sizeNames[ outVecSize ]);
-    }
 
-    const char *ptr = programSrc;
+    const char *ptr = kernelSource.c_str();
     error = create_single_kernel_helper( context, &program, &kernel, 1, &ptr, "test_fn" );
     test_error( error, "Unable to create testing kernel" );
 
-
     // Create some input values
     size_t inBufferSize = sizeof(char)* numElements * get_explicit_type_size( inVecType ) * vecSize;
-    char *inBuffer = (char*)malloc( inBufferSize );
+    std::vector<char> inBuffer(inBufferSize);
     size_t outBufferSize = sizeof(char)* numElements * get_explicit_type_size( outVecType ) *outVecSize;
-    char *outBuffer = (char*)malloc( outBufferSize );
+    std::vector<char> outBuffer(outBufferSize);
 
-    d = init_genrand( gRandomSeed );
-    generate_random_data( inVecType, numElements * vecSize,
-                         d, inBuffer );
-    free_mtdata(d); d = NULL;
+    generate_random_data(inVecType, numElements * vecSize, d,
+                         &inBuffer.front());
 
     // Create I/O streams and set arguments
-    streams[ 0 ] = clCreateBuffer( context, CL_MEM_COPY_HOST_PTR, inBufferSize, inBuffer, &error );
+    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, inBufferSize,
+                                &inBuffer.front(), &error);
     test_error( error, "Unable to create I/O stream" );
     streams[ 1 ] = clCreateBuffer( context, CL_MEM_READ_WRITE, outBufferSize, NULL, &error );
     test_error( error, "Unable to create I/O stream" );
@@ -161,15 +131,15 @@ int test_astype_set( cl_device_id device, cl_context context, cl_command_queue q
     error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, localThreads, 0, NULL, NULL );
     test_error( error, "Unable to run kernel" );
 
-
     // Get the results and compare
     // The beauty is that astype is supposed to return the bit pattern as a different type, which means
     // the output should have the exact same bit pattern as the input. No interpretation necessary!
-    error = clEnqueueReadBuffer( queue, streams[ 1 ], CL_TRUE, 0, outBufferSize, outBuffer, 0, NULL, NULL );
+    error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, outBufferSize,
+                                &outBuffer.front(), 0, NULL, NULL);
     test_error( error, "Unable to read results" );
 
-    char *expected = inBuffer;
-    char *actual = outBuffer;
+    char *expected = &inBuffer.front();
+    char *actual = &outBuffer.front();
     size_t compSize = typeSize*vecSize;
     if(outTypeSize*outVecSize < compSize) {
         compSize = outTypeSize*outVecSize;
@@ -178,8 +148,6 @@ int test_astype_set( cl_device_id device, cl_context context, cl_command_queue q
     if(outVecSize == 4 && vecSize == 3)
     {
         // as_type4(vec3) should compile but produce undefined results??
-        free(inBuffer);
-        free(outBuffer);
         return 0;
     }
 
@@ -188,8 +156,6 @@ int test_astype_set( cl_device_id device, cl_context context, cl_command_queue q
         // as_typen(vecm) should compile and run but produce
         // implementation-defined results for m != n
         // and n*sizeof(type) = sizeof(vecm)
-        free(inBuffer);
-        free(outBuffer);
         return 0;
     }
 
@@ -203,17 +169,14 @@ int test_astype_set( cl_device_id device, cl_context context, cl_command_queue q
                       GetDataVectorString( expected, typeSize, vecSize, expectedString ),
                       GetDataVectorString( actual, typeSize, vecSize, actualString ) );
             log_error("Src is :\n%s\n----\n%d threads %d localthreads\n",
-                      programSrc, (int)threads[0],(int) localThreads[0]);
-            free(inBuffer);
-            free(outBuffer);
+                      kernelSource.c_str(), (int)threads[0],
+                      (int)localThreads[0]);
             return 1;
         }
         expected += typeSize * vecSize;
         actual += outTypeSize * outVecSize;
     }
 
-    free(inBuffer);
-    free(outBuffer);
     return 0;
 }
 
@@ -223,31 +186,39 @@ int test_astype(cl_device_id device, cl_context context, cl_command_queue queue,
     // legal in OpenCL 1.0, the result is dependent on the device it runs on, which means there's no actual way
     // for us to verify what is "valid". So the only thing we can test are types that match in size independent
     // of the element count (char -> uchar, etc)
-    ExplicitType vecTypes[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble, kNumExplicitTypes };
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
+    const std::vector<ExplicitType> vecTypes = { kChar,   kUChar, kShort,
+                                                 kUShort, kInt,   kUInt,
+                                                 kLong,   kULong, kFloat,
+                                                 kHalf,   kDouble };
+    const unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
     unsigned int inTypeIdx, outTypeIdx, sizeIdx, outSizeIdx;
     size_t inTypeSize, outTypeSize;
     int error = 0;
 
-    for( inTypeIdx = 0; vecTypes[ inTypeIdx ] != kNumExplicitTypes; inTypeIdx++ )
+    bool fp16Support = is_extension_available(device, "cl_khr_fp16");
+    bool fp64Support = is_extension_available(device, "cl_khr_fp64");
+
+    auto skip_type = [&](ExplicitType et) {
+        if ((et == kLong || et == kULong) && !gHasLong)
+            return true;
+        else if (et == kDouble && !fp64Support)
+            return true;
+        else if (et == kHalf && !fp16Support)
+            return true;
+        return false;
+    };
+
+    for (inTypeIdx = 0; inTypeIdx < vecTypes.size(); inTypeIdx++)
     {
         inTypeSize = get_explicit_type_size(vecTypes[inTypeIdx]);
 
-        if( vecTypes[ inTypeIdx ] == kDouble && !is_extension_available( device, "cl_khr_fp64" ) )
-            continue;
+        if (skip_type(vecTypes[inTypeIdx])) continue;
 
-        if (( vecTypes[ inTypeIdx ] == kLong || vecTypes[ inTypeIdx ] == kULong ) && !gHasLong )
-            continue;
-
-        for( outTypeIdx = 0; vecTypes[ outTypeIdx ] != kNumExplicitTypes; outTypeIdx++ )
+        for (outTypeIdx = 0; outTypeIdx < vecTypes.size(); outTypeIdx++)
         {
             outTypeSize = get_explicit_type_size(vecTypes[outTypeIdx]);
-            if( vecTypes[ outTypeIdx ] == kDouble && !is_extension_available( device, "cl_khr_fp64" ) ) {
-                continue;
-            }
 
-            if (( vecTypes[ outTypeIdx ] == kLong || vecTypes[ outTypeIdx ] == kULong ) && !gHasLong )
-                continue;
+            if (skip_type(vecTypes[outTypeIdx])) continue;
 
             // change this check
             if( inTypeIdx == outTypeIdx ) {
@@ -259,7 +230,6 @@ int test_astype(cl_device_id device, cl_context context, cl_command_queue queue,
 
             for( sizeIdx = 0; vecSizes[ sizeIdx ] != 0; sizeIdx++ )
             {
-
                 for(outSizeIdx = 0; vecSizes[outSizeIdx] != 0; outSizeIdx++)
                 {
                     if(vecSizes[sizeIdx]*inTypeSize !=
@@ -268,10 +238,7 @@ int test_astype(cl_device_id device, cl_context context, cl_command_queue queue,
                         continue;
                     }
                     error += test_astype_set( device, context, queue, vecTypes[ inTypeIdx ], vecTypes[ outTypeIdx ], vecSizes[ sizeIdx ], vecSizes[outSizeIdx], n_elems );
-
-
                 }
-
             }
             if(get_explicit_type_size(vecTypes[inTypeIdx]) ==
                get_explicit_type_size(vecTypes[outTypeIdx])) {

test_conformance/basic/test_async_copy.cpp

@@ -20,8 +20,7 @@
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
-
-
+#include <vector>
 
 #include "procs.h"
 #include "harness/conversions.h"
@@ -86,8 +85,7 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
     clKernelWrapper kernel;
     clMemWrapper streams[ 2 ];
     size_t threads[ 1 ], localThreads[ 1 ];
-    void *inBuffer, *outBuffer;
-    MTdata d;
+    MTdataHolder d(gRandomSeed);
     char vecNameString[64]; vecNameString[0] = 0;
     if (vecSize == 1)
         sprintf(vecNameString, "%s", get_explicit_type_name(vecType));
@@ -109,9 +107,15 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
     char programSource[4096]; programSource[0]=0;
     char *programPtr;
 
-    sprintf(programSource, kernelCode,
-            vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-            vecNameString, vecNameString, vecNameString, vecNameString, get_explicit_type_name(vecType), vecNameString, vecNameString);
+    std::string extStr = "";
+    if (vecType == kDouble)
+        extStr = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable";
+    else if (vecType == kHalf)
+        extStr = "#pragma OPENCL EXTENSION cl_khr_fp16 : enable";
+
+    sprintf(programSource, kernelCode, extStr.c_str(), vecNameString,
+            vecNameString, vecNameString, vecNameString,
+            get_explicit_type_name(vecType), vecNameString, vecNameString);
     //log_info("program: %s\n", programSource);
     programPtr = programSource;
 
@@ -150,9 +154,10 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
     size_t globalBufferSize = numberOfLocalWorkgroups*localBufferSize;
     size_t globalWorkgroupSize = numberOfLocalWorkgroups*localWorkgroupSize;
 
-    inBuffer = (void*)malloc(globalBufferSize);
-    outBuffer = (void*)malloc(globalBufferSize);
-    memset(outBuffer, 0, globalBufferSize);
+    std::vector<unsigned char> inBuffer(globalBufferSize);
+    std::vector<unsigned char> outBuffer(globalBufferSize);
+
+    outBuffer.assign(globalBufferSize, 0);
 
     cl_int copiesPerWorkItemInt, copiesPerWorkgroup;
     copiesPerWorkItemInt = (int)numberOfCopiesPerWorkitem;
@@ -164,13 +169,15 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
     threads[0] = globalWorkgroupSize;
     localThreads[0] = localWorkgroupSize;
 
-    d = init_genrand( gRandomSeed );
-    generate_random_data( vecType, globalBufferSize/get_explicit_type_size(vecType), d, inBuffer );
-    free_mtdata(d); d = NULL;
+    generate_random_data(vecType,
+                         globalBufferSize / get_explicit_type_size(vecType), d,
+                         &inBuffer.front());
 
-    streams[ 0 ] = clCreateBuffer( context, CL_MEM_COPY_HOST_PTR, globalBufferSize, inBuffer, &error );
+    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, globalBufferSize,
+                                &inBuffer.front(), &error);
     test_error( error, "Unable to create input buffer" );
-    streams[ 1 ] = clCreateBuffer( context, CL_MEM_COPY_HOST_PTR, globalBufferSize, outBuffer, &error );
+    streams[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, globalBufferSize,
+                                &outBuffer.front(), &error);
     test_error( error, "Unable to create output buffer" );
 
     error = clSetKernelArg( kernel, 0, sizeof( streams[ 0 ] ), &streams[ 0 ] );
@@ -189,16 +196,18 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
     test_error( error, "Unable to queue kernel" );
 
     // Read
-    error = clEnqueueReadBuffer( queue, streams[ 1 ], CL_TRUE, 0, globalBufferSize, outBuffer, 0, NULL, NULL );
+    error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, globalBufferSize,
+                                &outBuffer.front(), 0, NULL, NULL);
     test_error( error, "Unable to read results" );
 
     // Verify
     int failuresPrinted = 0;
-    if( memcmp( inBuffer, outBuffer, globalBufferSize ) != 0 )
+    if (memcmp(&inBuffer.front(), &outBuffer.front(), globalBufferSize) != 0)
     {
         size_t typeSize = get_explicit_type_size(vecType)* vecSize;
-        unsigned char * inchar = (unsigned char*)inBuffer;
-        unsigned char * outchar = (unsigned char*)outBuffer;
+        unsigned char *inchar = static_cast<unsigned char *>(&inBuffer.front());
+        unsigned char *outchar =
+            static_cast<unsigned char *>(&outBuffer.front());
         for (int i=0; i< (int)globalBufferSize; i+=(int)elementSize) {
             if (memcmp( ((char *)inchar)+i, ((char *)outchar)+i, typeSize) != 0 )
             {
@@ -226,26 +235,29 @@ int test_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue,
         }
     }
 
-    free(inBuffer);
-    free(outBuffer);
-
     return failuresPrinted ? -1 : 0;
 }
 
 int test_copy_all_types(cl_device_id deviceID, cl_context context, cl_command_queue queue, const char *kernelCode) {
-    ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble, kNumExplicitTypes };
+    const std::vector<ExplicitType> vecType = { kChar,  kUChar, kShort, kUShort,
+                                                kInt,   kUInt,  kLong,  kULong,
+                                                kFloat, kHalf,  kDouble };
     unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
     unsigned int size, typeIndex;
 
     int errors = 0;
 
-    for( typeIndex = 0; vecType[ typeIndex ] != kNumExplicitTypes; typeIndex++ )
-    {
-        if( vecType[ typeIndex ] == kDouble && !is_extension_available( deviceID, "cl_khr_fp64" ) )
-            continue;
+    bool fp16Support = is_extension_available(deviceID, "cl_khr_fp16");
+    bool fp64Support = is_extension_available(deviceID, "cl_khr_fp64");
 
+    for (typeIndex = 0; typeIndex < vecType.size(); typeIndex++)
+    {
         if (( vecType[ typeIndex ] == kLong || vecType[ typeIndex ] == kULong ) && !gHasLong )
             continue;
+        else if (vecType[typeIndex] == kDouble && !fp64Support)
+            continue;
+        else if (vecType[typeIndex] == kHalf && !fp16Support)
+            continue;
 
         for( size = 0; vecSizes[ size ] != 0; size++ )
         {
@@ -259,9 +271,6 @@ int test_copy_all_types(cl_device_id deviceID, cl_context context, cl_command_qu
     return 0;
 }
 
-
-
-
 int test_async_copy_global_to_local(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
 {
     return test_copy_all_types( deviceID, context, queue, async_global_to_local_kernel );

test_conformance/basic/test_async_copy2D.cpp

@@ -27,17 +27,25 @@
 
 static const char *async_global_to_local_kernel2D = R"OpenCLC(
 #pragma OPENCL EXTENSION cl_khr_extended_async_copies : enable
-%s // optional pragma string
 
-__kernel void test_fn(const __global %s *src, __global %s *dst,
-                      __local %s *localBuffer, int numElementsPerLine,
+#define STRUCT_SIZE %d
+typedef struct __attribute__((packed))
+{
+    uchar byte[STRUCT_SIZE];
+} VarSizeStruct __attribute__((aligned(1)));
+
+
+__kernel void test_fn(const __global VarSizeStruct *src, __global VarSizeStruct *dst,
+                      __local VarSizeStruct *localBuffer, int numElementsPerLine,
                       int lineCopiesPerWorkgroup, int lineCopiesPerWorkItem,
                       int srcStride, int dstStride) {
   // Zero the local storage first
   for (int i = 0; i < lineCopiesPerWorkItem; i++) {
     for (int j = 0; j < numElementsPerLine; j++) {
       const int index = (get_local_id(0) * lineCopiesPerWorkItem + i) * dstStride + j;
-      localBuffer[index] = (%s)(%s)0;
+      for (int k = 0; k < STRUCT_SIZE; k++) {
+        localBuffer[index].byte[k] = 0;
+      }
     }
   }
 
@@ -45,7 +53,7 @@ __kernel void test_fn(const __global %s *src, __global %s *dst,
   // try the copy
   barrier( CLK_LOCAL_MEM_FENCE );
   event_t event = async_work_group_copy_2D2D(localBuffer, 0, src,
-    lineCopiesPerWorkgroup * get_group_id(0) * srcStride, sizeof(%s),
+    lineCopiesPerWorkgroup * get_group_id(0) * srcStride, sizeof(VarSizeStruct),
     (size_t)numElementsPerLine, (size_t)lineCopiesPerWorkgroup, srcStride, dstStride, 0);
 
   // Wait for the copy to complete, then verify by manually copying to the dest
@@ -63,16 +71,24 @@ __kernel void test_fn(const __global %s *src, __global %s *dst,
 
 static const char *async_local_to_global_kernel2D = R"OpenCLC(
 #pragma OPENCL EXTENSION cl_khr_extended_async_copies : enable
-%s // optional pragma string
 
-__kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *localBuffer,
+#define STRUCT_SIZE %d
+typedef struct __attribute__((packed))
+{
+    uchar byte[STRUCT_SIZE];
+} VarSizeStruct __attribute__((aligned(1)));
+
+
+__kernel void test_fn(const __global VarSizeStruct *src, __global VarSizeStruct *dst, __local VarSizeStruct *localBuffer,
                       int numElementsPerLine, int lineCopiesPerWorkgroup,
                       int lineCopiesPerWorkItem, int srcStride, int dstStride) {
   // Zero the local storage first
   for (int i = 0; i < lineCopiesPerWorkItem; i++) {
     for (int j = 0; j < numElementsPerLine; j++) {
       const int index = (get_local_id(0) * lineCopiesPerWorkItem + i) * srcStride + j;
-      localBuffer[index] = (%s)(%s)0;
+      for (int k = 0; k < STRUCT_SIZE; k++) {
+        localBuffer[index].byte[k] = 0;
+      }
     }
   }
 
@@ -90,36 +106,22 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
   // Do this to verify all kernels are done copying to the local buffer before we try the copy
   barrier(CLK_LOCAL_MEM_FENCE);
   event_t event = async_work_group_copy_2D2D(dst, lineCopiesPerWorkgroup * get_group_id(0) * dstStride,
-    localBuffer, 0, sizeof(%s), (size_t)numElementsPerLine, (size_t)lineCopiesPerWorkgroup, srcStride,
+    localBuffer, 0, sizeof(VarSizeStruct), (size_t)numElementsPerLine, (size_t)lineCopiesPerWorkgroup, srcStride,
    dstStride, 0 );
 
   wait_group_events(1, &event);
 };
 )OpenCLC";
 
-int test_copy2D(cl_device_id deviceID, cl_context context,
-                cl_command_queue queue, const char *kernelCode,
-                ExplicitType vecType, int vecSize, int srcMargin, int dstMargin,
-                bool localIsDst)
+int test_copy2D(const cl_device_id deviceID, const cl_context context,
+                const cl_command_queue queue, const char *const kernelCode,
+                const size_t elementSize, const int srcMargin,
+                const int dstMargin, const bool localIsDst)
 {
     int error;
-    clProgramWrapper program;
-    clKernelWrapper kernel;
-    clMemWrapper streams[2];
-    size_t threads[1], localThreads[1];
-    void *inBuffer, *outBuffer, *outBufferCopy;
-    MTdata d;
-    char vecNameString[64];
-    vecNameString[0] = 0;
-    if (vecSize == 1)
-        sprintf(vecNameString, "%s", get_explicit_type_name(vecType));
-    else
-        sprintf(vecNameString, "%s%d", get_explicit_type_name(vecType),
-                vecSize);
 
-    size_t elementSize = get_explicit_type_size(vecType) * vecSize;
-    log_info("Testing %s with srcMargin = %d, dstMargin = %d\n", vecNameString,
-             srcMargin, dstMargin);
+    log_info("Testing %d byte element with srcMargin = %d, dstMargin = %d\n",
+             elementSize, srcMargin, dstMargin);
 
     cl_long max_local_mem_size;
     error =
@@ -139,6 +141,13 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
     test_error(error,
                "clGetDeviceInfo for CL_DEVICE_MAX_MEM_ALLOC_SIZE failed.");
 
+    cl_long max_work_group_size;
+    error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_GROUP_SIZE,
+                            sizeof(max_work_group_size), &max_work_group_size,
+                            NULL);
+    test_error(error,
+               "clGetDeviceInfo for CL_DEVICE_MAX_WORK_GROUP_SIZE failed.");
+
     if (max_alloc_size > max_global_mem_size / 2)
         max_alloc_size = max_global_mem_size / 2;
 
@@ -149,20 +158,17 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
     test_error(error,
                "clGetDeviceInfo for CL_DEVICE_MAX_COMPUTE_UNITS failed.");
 
-    char programSource[4096];
-    programSource[0] = 0;
-    char *programPtr;
+    char programSource[4096] = { 0 };
+    const char *programPtr = programSource;
 
-    sprintf(programSource, kernelCode,
-            vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
-                               : "",
-            vecNameString, vecNameString, vecNameString, vecNameString,
-            get_explicit_type_name(vecType), vecNameString);
+    sprintf(programSource, kernelCode, elementSize);
     // log_info("program: %s\n", programSource);
-    programPtr = programSource;
+
+    clProgramWrapper program;
+    clKernelWrapper kernel;
 
     error = create_single_kernel_helper(context, &program, &kernel, 1,
-                                        (const char **)&programPtr, "test_fn");
+                                        &programPtr, "test_fn");
     test_error(error, "Unable to create testing kernel");
 
     size_t max_workgroup_size;
@@ -188,9 +194,6 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
     const cl_int dstStride = numElementsPerLine + dstMargin;
     const cl_int srcStride = numElementsPerLine + srcMargin;
 
-    elementSize =
-        get_explicit_type_size(vecType) * ((vecSize == 3) ? 4 : vecSize);
-
     const size_t lineCopiesPerWorkItem = 13;
     const size_t localStorageSpacePerWorkitem = lineCopiesPerWorkItem
         * elementSize * (localIsDst ? dstStride : srcStride);
@@ -208,7 +211,6 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
     if (maxLocalWorkgroupSize > max_workgroup_size)
         localWorkgroupSize = max_workgroup_size;
 
-
     const size_t maxTotalLinesIn =
         (max_alloc_size / elementSize + srcMargin) / srcStride;
     const size_t maxTotalLinesOut =
@@ -231,9 +233,17 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
     const size_t globalWorkgroupSize =
         numberOfLocalWorkgroups * localWorkgroupSize;
 
-    inBuffer = (void *)malloc(inBufferSize);
-    outBuffer = (void *)malloc(outBufferSize);
-    outBufferCopy = (void *)malloc(outBufferSize);
+    if ((localBufferSize / 4) > max_work_group_size)
+    {
+        log_info("Skipping due to resource requirements local:%db  "
+                 "max_work_group_size:%d\n",
+                 localBufferSize, max_work_group_size);
+        return 0;
+    }
+
+    void *const inBuffer = (void *)malloc(inBufferSize);
+    void *const outBuffer = (void *)malloc(outBufferSize);
+    void *const outBufferCopy = (void *)malloc(outBufferSize);
 
     const cl_int lineCopiesPerWorkItemInt =
         static_cast<cl_int>(lineCopiesPerWorkItem);
@@ -250,18 +260,20 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
         (int)inBufferSize, (int)outBufferSize, lineCopiesPerWorkgroup,
         lineCopiesPerWorkItemInt);
 
+    size_t threads[1], localThreads[1];
+
     threads[0] = globalWorkgroupSize;
     localThreads[0] = localWorkgroupSize;
 
-    d = init_genrand(gRandomSeed);
-    generate_random_data(
-        vecType, inBufferSize / get_explicit_type_size(vecType), d, inBuffer);
-    generate_random_data(
-        vecType, outBufferSize / get_explicit_type_size(vecType), d, outBuffer);
+    MTdata d = init_genrand(gRandomSeed);
+    generate_random_data(kChar, inBufferSize, d, inBuffer);
+    generate_random_data(kChar, outBufferSize, d, outBuffer);
     free_mtdata(d);
     d = NULL;
     memcpy(outBufferCopy, outBuffer, outBufferSize);
 
+    clMemWrapper streams[2];
+
     streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, inBufferSize,
                                 inBuffer, &error);
     test_error(error, "Unable to create input buffer");
@@ -301,8 +313,7 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
 
     // Verify
     int failuresPrinted = 0;
-    // Verify
-    size_t typeSize = get_explicit_type_size(vecType) * vecSize;
+
     for (int i = 0;
          i < (int)globalWorkgroupSize * lineCopiesPerWorkItem * elementSize;
          i += elementSize)
@@ -313,13 +324,12 @@ int test_copy2D(cl_device_id deviceID, cl_context context,
             int inIdx = i * srcStride + j;
             int outIdx = i * dstStride + j;
             if (memcmp(((char *)inBuffer) + inIdx, ((char *)outBuffer) + outIdx,
-                       typeSize)
+                       elementSize)
                 != 0)
             {
                 unsigned char *inchar = (unsigned char *)inBuffer + inIdx;
                 unsigned char *outchar = (unsigned char *)outBuffer + outIdx;
-                char values[4096];
-                values[0] = 0;
+                char values[4096] = { 0 };
 
                 if (failuresPrinted == 0)
                 {
@@ -382,16 +392,14 @@ int test_copy2D_all_types(cl_device_id deviceID, cl_context context,
                           cl_command_queue queue, const char *kernelCode,
                           bool localIsDst)
 {
-    ExplicitType vecType[] = {
-        kChar,  kUChar, kShort,  kUShort,          kInt, kUInt, kLong,
-        kULong, kFloat, kDouble, kNumExplicitTypes
-    };
+    const unsigned int elemSizes[] = { 1, 2,  3,  4,  5,  6, 7,
+                                       8, 13, 16, 32, 47, 64 };
     // The margins below represent the number of elements between the end of
     // one line and the start of the next. The strides are equivalent to the
     // length of the line plus the chosen margin.
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
-    unsigned int smallTypesMarginSizes[] = { 0, 10, 100 };
-    unsigned int size, typeIndex, srcMargin, dstMargin;
+    // These have to be multipliers, because the margin must be a multiple of
+    // element size.
+    const unsigned int marginMultipliers[] = { 0, 10, 100 };
 
     int errors = 0;
 
@@ -399,55 +407,27 @@ int test_copy2D_all_types(cl_device_id deviceID, cl_context context,
     {
         log_info(
             "Device does not support extended async copies. Skipping test.\n");
-        return 0;
     }
-
-    for (typeIndex = 0; vecType[typeIndex] != kNumExplicitTypes; typeIndex++)
+    else
     {
-        if (vecType[typeIndex] == kDouble
-            && !is_extension_available(deviceID, "cl_khr_fp64"))
-            continue;
-
-        if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong)
-            && !gHasLong)
-            continue;
-
-        for (size = 0; vecSizes[size] != 0; size++)
+        for (const unsigned int elemSize : elemSizes)
         {
-            if (get_explicit_type_size(vecType[typeIndex]) * vecSizes[size]
-                <= 2) // small type
+            for (const unsigned int srcMarginMultiplier : marginMultipliers)
             {
-                for (srcMargin = 0; srcMargin < sizeof(smallTypesMarginSizes)
-                         / sizeof(smallTypesMarginSizes[0]);
-                     srcMargin++)
-                {
-                    for (dstMargin = 0;
-                         dstMargin < sizeof(smallTypesMarginSizes)
-                             / sizeof(smallTypesMarginSizes[0]);
-                         dstMargin++)
+                for (const unsigned int dstMarginMultiplier : marginMultipliers)
                 {
                     if (test_copy2D(deviceID, context, queue, kernelCode,
-                                        vecType[typeIndex], vecSizes[size],
-                                        smallTypesMarginSizes[srcMargin],
-                                        smallTypesMarginSizes[dstMargin],
-                                        localIsDst))
+                                    elemSize, srcMarginMultiplier * elemSize,
+                                    dstMarginMultiplier * elemSize, localIsDst))
                     {
                         errors++;
                     }
                 }
             }
         }
-            // not a small type, check only zero stride
-            else if (test_copy2D(deviceID, context, queue, kernelCode,
-                                 vecType[typeIndex], vecSizes[size], 0, 0,
-                                 localIsDst))
-            {
-                errors++;
     }
-        }
-    }
-    if (errors) return -1;
-    return 0;
+
+    return errors ? -1 : 0;
 }
 
 int test_async_copy_global_to_local2D(cl_device_id deviceID, cl_context context,

test_conformance/basic/test_async_copy3D.cpp

@@ -27,9 +27,14 @@
 
 static const char *async_global_to_local_kernel3D = R"OpenCLC(
 #pragma OPENCL EXTENSION cl_khr_extended_async_copies : enable
-%s // optional pragma string
 
-__kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *localBuffer,
+#define STRUCT_SIZE %d
+typedef struct __attribute__((packed))
+{
+    uchar byte[STRUCT_SIZE];
+} VarSizeStruct __attribute__((aligned(1)));
+
+__kernel void test_fn(const __global VarSizeStruct *src, __global VarSizeStruct *dst, __local VarSizeStruct *localBuffer,
                       int numElementsPerLine, int numLines, int planesCopiesPerWorkgroup,
                       int planesCopiesPerWorkItem, int srcLineStride,
                       int dstLineStride, int srcPlaneStride, int dstPlaneStride ) {
@@ -38,7 +43,9 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
     for (int j = 0; j < numLines; j++) {
       for (int k = 0; k < numElementsPerLine; k++) {
         const int index = (get_local_id(0) * planesCopiesPerWorkItem + i) * dstPlaneStride + j * dstLineStride + k;
-        localBuffer[index] = (%s)(%s)0;
+        for (int k = 0; k < STRUCT_SIZE; k++) {
+          localBuffer[index].byte[k] = 0;
+        }
       }
     }
   }
@@ -48,7 +55,7 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
 
   event_t event = async_work_group_copy_3D3D(localBuffer, 0, src,
     planesCopiesPerWorkgroup * get_group_id(0) * srcPlaneStride,
-    sizeof(%s), (size_t)numElementsPerLine, (size_t)numLines,
+    sizeof(VarSizeStruct), (size_t)numElementsPerLine, (size_t)numLines,
     planesCopiesPerWorkgroup, srcLineStride, srcPlaneStride, dstLineStride,
     dstPlaneStride, 0);
 
@@ -69,9 +76,14 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
 
 static const char *async_local_to_global_kernel3D = R"OpenCLC(
 #pragma OPENCL EXTENSION cl_khr_extended_async_copies : enable
-%s // optional pragma string
 
-__kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *localBuffer,
+#define STRUCT_SIZE %d
+typedef struct __attribute__((packed))
+{
+    uchar byte[STRUCT_SIZE];
+} VarSizeStruct __attribute__((aligned(1)));
+
+__kernel void test_fn(const __global VarSizeStruct *src, __global VarSizeStruct *dst, __local VarSizeStruct *localBuffer,
                       int numElementsPerLine, int numLines, int planesCopiesPerWorkgroup,
                       int planesCopiesPerWorkItem, int srcLineStride,
                       int dstLineStride, int srcPlaneStride, int dstPlaneStride) {
@@ -80,7 +92,9 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
     for (int j = 0; j < numLines; j++) {
       for (int k = 0; k < numElementsPerLine; k++) {
         const int index = (get_local_id(0) * planesCopiesPerWorkItem + i) * srcPlaneStride + j * srcLineStride + k;
-        localBuffer[index] = (%s)(%s)0;
+        for (int k = 0; k < STRUCT_SIZE; k++) {
+          localBuffer[index].byte[k] = 0;
+        }
       }
     }
   }
@@ -103,38 +117,25 @@ __kernel void test_fn(const __global %s *src, __global %s *dst, __local %s *loca
 
   event_t event = async_work_group_copy_3D3D(dst,
     planesCopiesPerWorkgroup * get_group_id(0) * dstPlaneStride, localBuffer, 0,
-    sizeof(%s), (size_t)numElementsPerLine, (size_t)numLines, planesCopiesPerWorkgroup,
+    sizeof(VarSizeStruct), (size_t)numElementsPerLine, (size_t)numLines, planesCopiesPerWorkgroup,
     srcLineStride, srcPlaneStride, dstLineStride, dstPlaneStride, 0);
 
   wait_group_events(1, &event);
 }
 )OpenCLC";
 
-int test_copy3D(cl_device_id deviceID, cl_context context,
-                cl_command_queue queue, const char *kernelCode,
-                ExplicitType vecType, int vecSize, int srcLineMargin,
-                int dstLineMargin, int srcPlaneMargin, int dstPlaneMargin,
-                bool localIsDst)
+int test_copy3D(const cl_device_id deviceID, const cl_context context,
+                const cl_command_queue queue, const char *const kernelCode,
+                const size_t elementSize, const int srcLineMargin,
+                const int dstLineMargin, const int srcPlaneMargin,
+                const int dstPlaneMargin, const bool localIsDst)
 {
     int error;
-    clProgramWrapper program;
-    clKernelWrapper kernel;
-    clMemWrapper streams[2];
-    size_t threads[1], localThreads[1];
-    void *inBuffer, *outBuffer, *outBufferCopy;
-    MTdata d;
-    char vecNameString[64];
-    vecNameString[0] = 0;
-    if (vecSize == 1)
-        sprintf(vecNameString, "%s", get_explicit_type_name(vecType));
-    else
-        sprintf(vecNameString, "%s%d", get_explicit_type_name(vecType),
-                vecSize);
 
-    size_t elementSize = get_explicit_type_size(vecType) * vecSize;
-    log_info("Testing %s with srcLineMargin = %d, dstLineMargin = %d, "
+    log_info(
+        "Testing %d byte element with srcLineMargin = %d, dstLineMargin = %d, "
         "srcPlaneMargin = %d, dstPlaneMargin = %d\n",
-             vecNameString, srcLineMargin, dstLineMargin, srcPlaneMargin,
+        elementSize, srcLineMargin, dstLineMargin, srcPlaneMargin,
         dstPlaneMargin);
 
     cl_long max_local_mem_size;
@@ -165,20 +166,16 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
     test_error(error,
                "clGetDeviceInfo for CL_DEVICE_MAX_COMPUTE_UNITS failed.");
 
-    char programSource[4096];
-    programSource[0] = 0;
-    char *programPtr;
+    char programSource[4096] = { 0 };
+    const char *programPtr = programSource;
 
-    sprintf(programSource, kernelCode,
-            vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
-                               : "",
-            vecNameString, vecNameString, vecNameString, vecNameString,
-            get_explicit_type_name(vecType), vecNameString, vecNameString);
+    sprintf(programSource, kernelCode, elementSize);
     // log_info("program: %s\n", programSource);
-    programPtr = programSource;
+    clProgramWrapper program;
+    clKernelWrapper kernel;
 
     error = create_single_kernel_helper(context, &program, &kernel, 1,
-                                        (const char **)&programPtr, "test_fn");
+                                        &programPtr, "test_fn");
     test_error(error, "Unable to create testing kernel");
 
     size_t max_workgroup_size;
@@ -196,6 +193,13 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
     test_error(error,
                "clGetDeviceInfo failed for CL_DEVICE_MAX_WORK_ITEM_SIZES");
 
+    cl_long max_work_group_size;
+    error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_GROUP_SIZE,
+                            sizeof(max_work_group_size), &max_work_group_size,
+                            NULL);
+    test_error(error,
+               "clGetDeviceInfo for CL_DEVICE_MAX_WORK_GROUP_SIZE failed.");
+
     // Pick the minimum of the device and the kernel
     if (max_workgroup_size > max_local_workgroup_size[0])
         max_workgroup_size = max_local_workgroup_size[0];
@@ -208,8 +212,6 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
     const cl_int dstPlaneStride = (numLines * dstLineStride) + dstPlaneMargin;
     const cl_int srcPlaneStride = (numLines * srcLineStride) + srcPlaneMargin;
 
-    elementSize =
-        get_explicit_type_size(vecType) * ((vecSize == 3) ? 4 : vecSize);
     const size_t planesCopiesPerWorkItem = 2;
     const size_t localStorageSpacePerWorkitem = elementSize
         * planesCopiesPerWorkItem
@@ -251,9 +253,17 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
     const size_t globalWorkgroupSize =
         numberOfLocalWorkgroups * localWorkgroupSize;
 
-    inBuffer = (void *)malloc(inBufferSize);
-    outBuffer = (void *)malloc(outBufferSize);
-    outBufferCopy = (void *)malloc(outBufferSize);
+    if ((localBufferSize / 4) > max_work_group_size)
+    {
+        log_info("Skipping due to resource requirements local:%db  "
+                 "max_work_group_size:%d\n",
+                 localBufferSize, max_work_group_size);
+        return 0;
+    }
+
+    void *const inBuffer = (void *)malloc(inBufferSize);
+    void *const outBuffer = (void *)malloc(outBufferSize);
+    void *const outBufferCopy = (void *)malloc(outBufferSize);
 
     const cl_int planesCopiesPerWorkItemInt =
         static_cast<cl_int>(planesCopiesPerWorkItem);
@@ -270,18 +280,20 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
              (int)localBufferSize, (int)inBufferSize, (int)outBufferSize,
              planesCopiesPerWorkgroup, planesCopiesPerWorkItemInt);
 
+    size_t threads[1], localThreads[1];
+
     threads[0] = globalWorkgroupSize;
     localThreads[0] = localWorkgroupSize;
 
-    d = init_genrand(gRandomSeed);
-    generate_random_data(
-        vecType, inBufferSize / get_explicit_type_size(vecType), d, inBuffer);
-    generate_random_data(
-        vecType, outBufferSize / get_explicit_type_size(vecType), d, outBuffer);
+    MTdata d = init_genrand(gRandomSeed);
+    generate_random_data(kChar, inBufferSize, d, inBuffer);
+    generate_random_data(kChar, outBufferSize, d, outBuffer);
     free_mtdata(d);
     d = NULL;
     memcpy(outBufferCopy, outBuffer, outBufferSize);
 
+    clMemWrapper streams[2];
+
     streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, inBufferSize,
                                 inBuffer, &error);
     test_error(error, "Unable to create input buffer");
@@ -327,8 +339,7 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
 
     // Verify
     int failuresPrinted = 0;
-    // Verify
-    size_t typeSize = get_explicit_type_size(vecType) * vecSize;
+
     for (int i = 0;
          i < (int)globalWorkgroupSize * planesCopiesPerWorkItem * elementSize;
          i += elementSize)
@@ -341,14 +352,13 @@ int test_copy3D(cl_device_id deviceID, cl_context context,
                 int inIdx = i * srcPlaneStride + j * srcLineStride + k;
                 int outIdx = i * dstPlaneStride + j * dstLineStride + k;
                 if (memcmp(((char *)inBuffer) + inIdx,
-                           ((char *)outBuffer) + outIdx, typeSize)
+                           ((char *)outBuffer) + outIdx, elementSize)
                     != 0)
                 {
                     unsigned char *inchar = (unsigned char *)inBuffer + inIdx;
                     unsigned char *outchar =
                         (unsigned char *)outBuffer + outIdx;
-                    char values[4096];
-                    values[0] = 0;
+                    char values[4096] = { 0 };
 
                     if (failuresPrinted == 0)
                     {
@@ -439,17 +449,14 @@ int test_copy3D_all_types(cl_device_id deviceID, cl_context context,
                           cl_command_queue queue, const char *kernelCode,
                           bool localIsDst)
 {
-    ExplicitType vecType[] = {
-        kChar,  kUChar, kShort,  kUShort,          kInt, kUInt, kLong,
-        kULong, kFloat, kDouble, kNumExplicitTypes
-    };
+    const unsigned int elemSizes[] = { 1, 2,  3,  4,  5,  6, 7,
+                                       8, 13, 16, 32, 47, 64 };
     // The margins below represent the number of elements between the end of
-    // one line or plane and the start of the next. The strides are equivalent
-    // to the size of the line or plane plus the chosen margin.
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
-    unsigned int smallTypesMarginSizes[] = { 0, 10, 100 };
-    unsigned int size, typeIndex, srcLineMargin, dstLineMargin, srcPlaneMargin,
-        dstPlaneMargin;
+    // one line and the start of the next. The strides are equivalent to the
+    // size of the line or plane plus the chosen margin.
+    // These have to be multipliers, because the margin must be a multiple of
+    // element size.
+    const unsigned int marginMultipliers[] = { 0, 10, 100 };
 
     int errors = 0;
 
@@ -457,51 +464,28 @@ int test_copy3D_all_types(cl_device_id deviceID, cl_context context,
     {
         log_info(
             "Device does not support extended async copies. Skipping test.\n");
-        return 0;
     }
-
-    for (typeIndex = 0; vecType[typeIndex] != kNumExplicitTypes; typeIndex++)
+    else
     {
-        if (vecType[typeIndex] == kDouble
-            && !is_extension_available(deviceID, "cl_khr_fp64"))
-            continue;
-
-        if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong)
-            && !gHasLong)
-            continue;
-
-        for (size = 0; vecSizes[size] != 0; size++)
+        for (const unsigned int elemSize : elemSizes)
         {
-            if (get_explicit_type_size(vecType[typeIndex]) * vecSizes[size]
-                <= 2) // small type
+            for (const unsigned int srcLineMarginMultiplier : marginMultipliers)
             {
-                for (srcLineMargin = 0;
-                     srcLineMargin < sizeof(smallTypesMarginSizes)
-                         / sizeof(smallTypesMarginSizes[0]);
-                     srcLineMargin++)
+                for (const unsigned int dstLineMarginMultiplier :
+                     marginMultipliers)
                 {
-                    for (dstLineMargin = 0;
-                         dstLineMargin < sizeof(smallTypesMarginSizes)
-                             / sizeof(smallTypesMarginSizes[0]);
-                         dstLineMargin++)
+                    for (const unsigned int srcPlaneMarginMultiplier :
+                         marginMultipliers)
                     {
-                        for (srcPlaneMargin = 0;
-                             srcPlaneMargin < sizeof(smallTypesMarginSizes)
-                                 / sizeof(smallTypesMarginSizes[0]);
-                             srcPlaneMargin++)
+                        for (const unsigned int dstPlaneMarginMultiplier :
+                             marginMultipliers)
                         {
-                            for (dstPlaneMargin = 0;
-                                 dstPlaneMargin < sizeof(smallTypesMarginSizes)
-                                     / sizeof(smallTypesMarginSizes[0]);
-                                 dstPlaneMargin++)
-                            {
-                                if (test_copy3D(
-                                        deviceID, context, queue, kernelCode,
-                                        vecType[typeIndex], vecSizes[size],
-                                        smallTypesMarginSizes[srcLineMargin],
-                                        smallTypesMarginSizes[dstLineMargin],
-                                        smallTypesMarginSizes[srcPlaneMargin],
-                                        smallTypesMarginSizes[dstPlaneMargin],
+                            if (test_copy3D(deviceID, context, queue,
+                                            kernelCode, elemSize,
+                                            srcLineMarginMultiplier * elemSize,
+                                            dstLineMarginMultiplier * elemSize,
+                                            srcPlaneMarginMultiplier * elemSize,
+                                            dstPlaneMarginMultiplier * elemSize,
                                             localIsDst))
                             {
                                 errors++;
@@ -511,14 +495,6 @@ int test_copy3D_all_types(cl_device_id deviceID, cl_context context,
                 }
             }
         }
-            // not a small type, check only zero stride
-            else if (test_copy3D(deviceID, context, queue, kernelCode,
-                                 vecType[typeIndex], vecSizes[size], 0, 0, 0, 0,
-                                 localIsDst))
-            {
-                errors++;
-            }
-        }
     }
     if (errors) return -1;
     return 0;

test_conformance/basic/test_async_strided_copy.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -20,15 +20,16 @@
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
-
-
+#include <vector>
 
 #include "procs.h"
 #include "harness/conversions.h"
 
+// clang-format off
+
 static const char *async_strided_global_to_local_kernel =
 "%s\n" // optional pragma string
-"%s__kernel void test_fn( const __global %s *src, __global %s *dst, __local %s *localBuffer, int copiesPerWorkgroup, int copiesPerWorkItem, int stride )\n"
+"__kernel void test_fn( const __global %s *src, __global %s *dst, __local %s *localBuffer, int copiesPerWorkgroup, int copiesPerWorkItem, int stride )\n"
 "{\n"
 " int i;\n"
 // Zero the local storage first
@@ -46,7 +47,7 @@ static const char *async_strided_global_to_local_kernel =
 
 static const char *async_strided_local_to_global_kernel =
 "%s\n" // optional pragma string
-"%s__kernel void test_fn( const __global %s *src, __global %s *dst, __local %s *localBuffer, int copiesPerWorkgroup, int copiesPerWorkItem, int stride )\n"
+"__kernel void test_fn( const __global %s *src, __global %s *dst, __local %s *localBuffer, int copiesPerWorkgroup, int copiesPerWorkItem, int stride )\n"
 "{\n"
 " int i;\n"
 // Zero the local storage first
@@ -63,6 +64,7 @@ static const char *async_strided_local_to_global_kernel =
 " wait_group_events( 1, &event );\n"
 "}\n" ;
 
+// clang-format on
 
 int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queue queue, const char *kernelCode, ExplicitType vecType, int vecSize, int stride)
 {
@@ -71,8 +73,7 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
     clKernelWrapper kernel;
     clMemWrapper streams[ 2 ];
     size_t threads[ 1 ], localThreads[ 1 ];
-    void *inBuffer, *outBuffer;
-    MTdata d;
+    MTdataHolder d(gRandomSeed);
     char vecNameString[64]; vecNameString[0] = 0;
 
     if (vecSize == 1)
@@ -94,10 +95,15 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
     char programSource[4096]; programSource[0]=0;
     char *programPtr;
 
-    sprintf(programSource, kernelCode,
-        vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
-        "",
-        vecNameString, vecNameString, vecNameString, vecNameString, get_explicit_type_name(vecType), vecNameString, vecNameString);
+    std::string extStr = "";
+    if (vecType == kDouble)
+        extStr = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable";
+    else if (vecType == kHalf)
+        extStr = "#pragma OPENCL EXTENSION cl_khr_fp16 : enable";
+
+    sprintf(programSource, kernelCode, extStr.c_str(), vecNameString,
+            vecNameString, vecNameString, vecNameString,
+            get_explicit_type_name(vecType), vecNameString, vecNameString);
     //log_info("program: %s\n", programSource);
     programPtr = programSource;
 
@@ -151,9 +157,9 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
     size_t globalBufferSize = numberOfLocalWorkgroups*localBufferSize*stride;
     size_t globalWorkgroupSize = numberOfLocalWorkgroups*localWorkgroupSize;
 
-    inBuffer = (void*)malloc(globalBufferSize);
-    outBuffer = (void*)malloc(globalBufferSize);
-    memset(outBuffer, 0, globalBufferSize);
+    std::vector<unsigned char> inBuffer(globalBufferSize);
+    std::vector<unsigned char> outBuffer(globalBufferSize);
+    memset(outBuffer.data(), 0, globalBufferSize);
 
     cl_int copiesPerWorkItemInt, copiesPerWorkgroup;
     copiesPerWorkItemInt = (int)numberOfCopiesPerWorkitem;
@@ -165,13 +171,15 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
     threads[0] = globalWorkgroupSize;
     localThreads[0] = localWorkgroupSize;
 
-    d = init_genrand( gRandomSeed );
-    generate_random_data( vecType, globalBufferSize/get_explicit_type_size(vecType), d, inBuffer );
-    free_mtdata(d); d = NULL;
+    generate_random_data(vecType,
+                         globalBufferSize / get_explicit_type_size(vecType), d,
+                         inBuffer.data());
 
-    streams[ 0 ] = clCreateBuffer( context, CL_MEM_COPY_HOST_PTR, globalBufferSize, inBuffer, &error );
+    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, globalBufferSize,
+                                inBuffer.data(), &error);
     test_error( error, "Unable to create input buffer" );
-    streams[ 1 ] = clCreateBuffer( context, CL_MEM_COPY_HOST_PTR, globalBufferSize, outBuffer, &error );
+    streams[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR, globalBufferSize,
+                                outBuffer.data(), &error);
     test_error( error, "Unable to create output buffer" );
 
     error = clSetKernelArg( kernel, 0, sizeof( streams[ 0 ] ), &streams[ 0 ] );
@@ -192,17 +200,20 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
     test_error( error, "Unable to queue kernel" );
 
     // Read
-    error = clEnqueueReadBuffer( queue, streams[ 1 ], CL_TRUE, 0, globalBufferSize, outBuffer, 0, NULL, NULL );
+    error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, globalBufferSize,
+                                outBuffer.data(), 0, NULL, NULL);
     test_error( error, "Unable to read results" );
 
     // Verify
     size_t typeSize = get_explicit_type_size(vecType)* vecSize;
     for (int i=0; i<(int)globalBufferSize; i+=(int)elementSize*(int)stride)
     {
-        if (memcmp( ((char *)inBuffer)+i, ((char *)outBuffer)+i, typeSize) != 0 )
+        if (memcmp(&inBuffer.at(i), &outBuffer.at(i), typeSize) != 0)
         {
-            unsigned char * inchar = (unsigned char*)inBuffer + i;
-            unsigned char * outchar = (unsigned char*)outBuffer + i;
+            unsigned char *inchar =
+                static_cast<unsigned char *>(&inBuffer.at(i));
+            unsigned char *outchar =
+                static_cast<unsigned char *>(&outBuffer.at(i));
             char values[4096];
             values[0] = 0;
 
@@ -215,34 +226,35 @@ int test_strided_copy(cl_device_id deviceID, cl_context context, cl_command_queu
                 sprintf(values + strlen( values), "%2x ", outchar[j]);
             sprintf(values + strlen(values), "]");
             log_error("%s\n", values);
-            free(inBuffer);
-            free(outBuffer);
             return -1;
         }
     }
 
-    free(inBuffer);
-    free(outBuffer);
-
     return 0;
 }
 
 int test_strided_copy_all_types(cl_device_id deviceID, cl_context context, cl_command_queue queue, const char *kernelCode)
 {
-    ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble, kNumExplicitTypes };
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
-    unsigned int strideSizes[] = { 1, 3, 4, 5, 0 };
+    const std::vector<ExplicitType> vecType = { kChar,  kUChar, kShort, kUShort,
+                                                kInt,   kUInt,  kLong,  kULong,
+                                                kFloat, kHalf,  kDouble };
+    const unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
+    const unsigned int strideSizes[] = { 1, 3, 4, 5, 0 };
     unsigned int size, typeIndex, stride;
 
     int errors = 0;
 
-    for( typeIndex = 0; vecType[ typeIndex ] != kNumExplicitTypes; typeIndex++ )
-    {
-        if( vecType[ typeIndex ] == kDouble && !is_extension_available( deviceID, "cl_khr_fp64" ) )
-            continue;
+    bool fp16Support = is_extension_available(deviceID, "cl_khr_fp16");
+    bool fp64Support = is_extension_available(deviceID, "cl_khr_fp64");
 
+    for (typeIndex = 0; typeIndex < vecType.size(); typeIndex++)
+    {
         if (( vecType[ typeIndex ] == kLong || vecType[ typeIndex ] == kULong ) && !gHasLong )
             continue;
+        else if (vecType[typeIndex] == kDouble && !fp64Support)
+            continue;
+        else if (vecType[typeIndex] == kHalf && !fp16Support)
+            continue;
 
         for( size = 0; vecSizes[ size ] != 0; size++ )
         {
@@ -260,9 +272,6 @@ int test_strided_copy_all_types(cl_device_id deviceID, cl_context context, cl_co
     return 0;
 }
 
-
-
-
 int test_async_strided_copy_global_to_local(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
 {
     return test_strided_copy_all_types( deviceID, context, queue, async_strided_global_to_local_kernel );

test_conformance/basic/test_get_linear_ids.cpp

@@ -104,15 +104,19 @@ test_get_linear_ids(cl_device_id device, cl_context context, cl_command_queue qu
 
         switch (dims) {
         case 1:
-            log_info("  testing offset=%u global=%u local=%u...\n", gwo[0], gws[0], lws[0]);
+            log_info("  testing offset=%zu global=%zu local=%zu...\n", gwo[0],
+                     gws[0], lws[0]);
             break;
         case 2:
-            log_info("  testing offset=(%u,%u) global=(%u,%u) local=(%u,%u)...\n",
+            log_info("  testing offset=(%zu,%zu) global=(%zu,%zu) "
+                     "local=(%zu,%zu)...\n",
                      gwo[0], gwo[1], gws[0], gws[1], lws[0], lws[1]);
             break;
         case 3:
-            log_info("  testing offset=(%u,%u,%u) global=(%u,%u,%u) local=(%u,%u,%u)...\n",
-                    gwo[0], gwo[1], gwo[2], gws[0], gws[1], gws[2], lws[0], lws[1], lws[2]);
+            log_info("  testing offset=(%zu,%zu,%zu) global=(%zu,%zu,%zu) "
+                     "local=(%zu,%zu,%zu)...\n",
+                     gwo[0], gwo[1], gwo[2], gws[0], gws[1], gws[2], lws[0],
+                     lws[1], lws[2]);
             break;
         }
 

test_conformance/basic/test_hiloeo.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -13,14 +13,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
-
+#include <iomanip>
+#include <limits.h>
 #include <stdio.h>
 #include <string.h>
-#include <limits.h>
 #include <sys/types.h>
 #include <sys/stat.h>
-
+#include <vector>
 
 #include "procs.h"
 
@@ -31,9 +30,10 @@ int odd_offset( int index, int vectorSize ) { return index * 2 + 1; }
 
 typedef int (*OffsetFunc)( int index, int vectorSize );
 static const OffsetFunc offsetFuncs[4] = { hi_offset, lo_offset, even_offset, odd_offset };
-typedef int (*verifyFunc)( const void *, const void *, const void *, int n, const char *sizeName );
 static const char *operatorToUse_names[] = { "hi", "lo", "even", "odd" };
-static const char *test_str_names[] = { "char", "uchar", "short", "ushort", "int", "uint", "long", "ulong", "float", "double" };
+static const char *test_str_names[] = { "char", "uchar", "short", "ushort",
+                                        "int",  "uint",  "long",  "ulong",
+                                        "half", "float", "double" };
 
 static const unsigned int vector_sizes[] =     { 1, 2, 3, 4, 8, 16};
 static const unsigned int vector_aligns[] =    { 1, 2, 4, 4, 8, 16};
@@ -45,43 +45,41 @@ static const unsigned int out_vector_idx[] =   { 0, 0, 1, 1, 3, 4};
 // strcat(gentype, vector_size_names[out_vector_idx[i]]);
 static const char *vector_size_names[] = { "", "2", "3", "4", "8", "16"};
 
-static const size_t  kSizes[] = { 1, 1, 2, 2, 4, 4, 8, 8, 4, 8 };
+static const size_t kSizes[] = { 1, 1, 2, 2, 4, 4, 8, 8, 2, 4, 8 };
 static int CheckResults( void *in, void *out, size_t elementCount, int type, int vectorSize, int operatorToUse );
 
 int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
 {
-    cl_int *input_ptr, *output_ptr, *p;
     int err;
-    cl_uint i;
     int hasDouble = is_extension_available( device, "cl_khr_fp64" );
+    int hasHalf = is_extension_available(device, "cl_khr_fp16");
     cl_uint vectorSize, operatorToUse;
     cl_uint type;
-    MTdata d;
+    MTdataHolder d(gRandomSeed);
 
     int expressionMode;
     int numExpressionModes = 2;
 
     size_t length = sizeof(cl_int) * 4 * n_elems;
 
-    input_ptr   = (cl_int*)malloc(length);
-    output_ptr  = (cl_int*)malloc(length);
+    std::vector<cl_int> input_ptr(4 * n_elems);
+    std::vector<cl_int> output_ptr(4 * n_elems);
 
-    p = input_ptr;
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<4 * (cl_uint) n_elems; i++)
-        p[i] = genrand_int32(d);
-    free_mtdata(d); d = NULL;
+    for (cl_uint i = 0; i < 4 * (cl_uint)n_elems; i++)
+        input_ptr[i] = genrand_int32(d);
 
     for( type = 0; type < sizeof( test_str_names ) / sizeof( test_str_names[0] ); type++ )
     {
         // Note: restrict the element count here so we don't end up overrunning the output buffer if we're compensating for 32-bit writes
         size_t elementCount = length / kSizes[type];
-        cl_mem streams[2];
+        clMemWrapper streams[2];
 
         // skip double if unavailable
         if( !hasDouble && ( 0 == strcmp( test_str_names[type], "double" )))
             continue;
 
+        if (!hasHalf && (0 == strcmp(test_str_names[type], "half"))) continue;
+
         if( !gHasLong &&
             (( 0 == strcmp( test_str_names[type], "long" )) ||
             ( 0 == strcmp( test_str_names[type], "ulong" ))))
@@ -104,12 +102,9 @@ int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue,
             return -1;
         }
 
-        err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, length, input_ptr, 0, NULL, NULL);
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueWriteBuffer failed\n");
-            return -1;
-        }
+        err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, length,
+                                   input_ptr.data(), 0, NULL, NULL);
+        test_error(err, "clEnqueueWriteBuffer failed\n");
 
         for( operatorToUse = 0; operatorToUse < sizeof( operatorToUse_names ) / sizeof( operatorToUse_names[0] ); operatorToUse++ )
         {
@@ -118,8 +113,8 @@ int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue,
             for( vectorSize = 1; vectorSize < sizeof( vector_size_names ) / sizeof( vector_size_names[0] ); vectorSize++ ) {
                 for(expressionMode = 0; expressionMode < numExpressionModes; ++expressionMode) {
 
-                    cl_program program = NULL;
-                    cl_kernel kernel = NULL;
+                    clProgramWrapper program;
+                    clKernelWrapper kernel;
                     cl_uint outVectorSize = out_vector_idx[vectorSize];
                     char expression[1024];
 
@@ -139,92 +134,64 @@ int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue,
                         "}\n"
                     };
 
-                    if(expressionMode == 0) {
-                        sprintf(expression, "srcA[tid]");
-                    } else if(expressionMode == 1) {
-                        switch(vector_sizes[vectorSize]) {
-                            case 16:
-                                sprintf(expression,
-                                        "((%s16)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3, srcA[tid].s4, srcA[tid].s5, srcA[tid].s6, srcA[tid].s7, srcA[tid].s8, srcA[tid].s9, srcA[tid].sA, srcA[tid].sB, srcA[tid].sC, srcA[tid].sD, srcA[tid].sE, srcA[tid].sf))",
-                                        test_str_names[type]
-                                        );
-                                break;
-                            case 8:
-                                sprintf(expression,
-                                        "((%s8)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3, srcA[tid].s4, srcA[tid].s5, srcA[tid].s6, srcA[tid].s7))",
-                                        test_str_names[type]
-                                        );
-                                break;
-                            case 4:
-                                sprintf(expression,
-                                        "((%s4)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2, srcA[tid].s3))",
-                                        test_str_names[type]
-                                        );
-                                break;
-                            case 3:
-                                sprintf(expression,
-                                        "((%s3)(srcA[tid].s0, srcA[tid].s1, srcA[tid].s2))",
-                                        test_str_names[type]
-                                        );
-                                break;
-                            case 2:
-                                sprintf(expression,
-                                        "((%s2)(srcA[tid].s0, srcA[tid].s1))",
-                                        test_str_names[type]
-                                        );
-                                break;
-                            default :
-                                sprintf(expression, "srcA[tid]");
-                                log_info("Default\n");
+                    if (expressionMode == 1 && vector_sizes[vectorSize] != 1)
+                    {
+                        std::ostringstream sstr;
+                        const char *index_chars[] = { "0", "1", "2", "3",
+                                                      "4", "5", "6", "7",
+                                                      "8", "9", "A", "B",
+                                                      "C", "D", "E", "f" };
+                        sstr << "((" << test_str_names[type]
+                             << std::to_string(vector_sizes[vectorSize])
+                             << ")(";
+                        for (unsigned i = 0; i < vector_sizes[vectorSize]; i++)
+                            sstr << " srcA[tid].s" << index_chars[i] << ",";
+                        sstr.seekp(-1, sstr.cur);
+                        sstr << "))";
+                        std::snprintf(expression, sizeof(expression), "%s",
+                                      sstr.str().c_str());
                     }
-                    } else {
-                        sprintf(expression, "srcA[tid]");
+                    else
+                    {
+                        std::snprintf(expression, sizeof(expression),
+                                      "srcA[tid]");
                     }
 
                     if (0 == strcmp( test_str_names[type], "double" ))
                         source[0] = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
 
+                    if (0 == strcmp(test_str_names[type], "half"))
+                        source[0] =
+                            "#pragma OPENCL EXTENSION cl_khr_fp16 : enable\n";
+
                     char kernelName[128];
                     snprintf( kernelName, sizeof( kernelName ), "test_%s_%s%s", operatorToUse_names[ operatorToUse ], test_str_names[type], vector_size_names[vectorSize] );
                     err = create_single_kernel_helper(context, &program, &kernel, sizeof( source ) / sizeof( source[0] ), source, kernelName );
-                    if (err)
-                        return -1;
+                    test_error(err, "create_single_kernel_helper failed\n");
 
                     err  = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0]);
                     err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1]);
-                    if (err != CL_SUCCESS)
-                    {
-                        log_error("clSetKernelArgs failed\n");
-                        return -1;
-                    }
+                    test_error(err, "clSetKernelArg failed\n");
 
                     //Wipe the output buffer clean
                     uint32_t pattern = 0xdeadbeef;
-                    memset_pattern4( output_ptr, &pattern, length );
-                    err = clEnqueueWriteBuffer(queue, streams[1], CL_TRUE, 0, length, output_ptr, 0, NULL, NULL);
-                    if (err != CL_SUCCESS)
-                    {
-                        log_error("clEnqueueWriteBuffer failed\n");
-                        return -1;
-                    }
+                    memset_pattern4(output_ptr.data(), &pattern, length);
+                    err = clEnqueueWriteBuffer(queue, streams[1], CL_TRUE, 0,
+                                               length, output_ptr.data(), 0,
+                                               NULL, NULL);
+                    test_error(err, "clEnqueueWriteBuffer failed\n");
 
                     size_t size = elementCount / (vector_aligns[vectorSize]);
                     err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &size, NULL, 0, NULL, NULL);
-                    if (err != CL_SUCCESS)
-                    {
-                        log_error("clEnqueueNDRangeKernel failed\n");
-                        return -1;
-                    }
+                    test_error(err, "clEnqueueNDRangeKernel failed\n");
 
-                    err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, length, output_ptr, 0, NULL, NULL);
-                    if (err != CL_SUCCESS)
-                    {
-                        log_error("clEnqueueReadBuffer failed\n");
-                        return -1;
-                    }
+                    err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0,
+                                              length, output_ptr.data(), 0,
+                                              NULL, NULL);
+                    test_error(err, "clEnqueueReadBuffer failed\n");
 
-                    char *inP = (char *)input_ptr;
-                    char *outP = (char *)output_ptr;
+                    char *inP = (char *)input_ptr.data();
+                    char *outP = (char *)output_ptr.data();
                     outP += kSizes[type] * ( ( vector_sizes[outVectorSize] ) -
                                             ( vector_sizes[ out_vector_idx[vectorSize] ] ) );
                     // was                outP += kSizes[type] * ( ( 1 << outVectorSize ) - ( 1 << ( vectorSize - 1 ) ) );
@@ -240,41 +207,69 @@ int test_hiloeo(cl_device_id device, cl_context context, cl_command_queue queue,
                         inP += kSizes[type] * ( vector_aligns[vectorSize] );
                         outP += kSizes[type] * ( vector_aligns[outVectorSize] );
                     }
-
-                    clReleaseKernel( kernel );
-                    clReleaseProgram( program );
                     log_info( "." );
                     fflush( stdout );
                 }
             }
         }
-
-        clReleaseMemObject( streams[0] );
-        clReleaseMemObject( streams[1] );
         log_info( "done\n" );
     }
 
     log_info("HiLoEO test passed\n");
-
-    free(input_ptr);
-    free(output_ptr);
-
     return err;
 }
 
-static int CheckResults( void *in, void *out, size_t elementCount, int type, int vectorSize, int operatorToUse )
+template <typename T>
+cl_int verify(void *in, void *out, size_t elementCount, int type,
+              int vectorSize, int operatorToUse, size_t cmpVectorSize)
 {
+    size_t halfVectorSize = vector_sizes[out_vector_idx[vectorSize]];
+    size_t elementSize = kSizes[type];
+    OffsetFunc f = offsetFuncs[operatorToUse];
     cl_ulong array[8];
     void *p = array;
-    size_t halfVectorSize  = vector_sizes[out_vector_idx[vectorSize]];
+
+    std::ostringstream ss;
+
+    T *i = (T *)in, *o = (T *)out;
+
+    for (cl_uint k = 0; k < elementCount; k++)
+    {
+        T *o2 = (T *)p;
+        for (size_t j = 0; j < halfVectorSize; j++)
+            o2[j] = i[f((int)j, (int)halfVectorSize * 2)];
+
+        if (memcmp(o, o2, elementSize * cmpVectorSize))
+        {
+            ss << "\n"
+               << k << ") Failure for" << test_str_names[type]
+               << vector_size_names[vectorSize] << '.'
+               << operatorToUse_names[operatorToUse] << " { "
+               << "0x" << std::setfill('0') << std::setw(elementSize * 2)
+               << std::hex << i[0];
+
+            for (size_t j = 1; j < halfVectorSize * 2; j++) ss << ", " << i[j];
+            ss << " } --> { " << o[0];
+            for (size_t j = 1; j < halfVectorSize; j++) ss << ", " << o[j];
+            ss << " }\n";
+            return -1;
+        }
+        i += 2 * halfVectorSize;
+        o += halfVectorSize;
+    }
+    return 0;
+}
+
+static int CheckResults(void *in, void *out, size_t elementCount, int type,
+                        int vectorSize, int operatorToUse)
+{
     size_t cmpVectorSize = vector_sizes[out_vector_idx[vectorSize]];
-    // was 1 << (vectorSize-1);
-    OffsetFunc f = offsetFuncs[ operatorToUse ];
     size_t elementSize = kSizes[type];
 
-    if(vector_size_names[vectorSize][0] == '3') {
-        if(operatorToUse_names[operatorToUse][0] == 'h' ||
-           operatorToUse_names[operatorToUse][0] == 'o') // hi or odd
+    if (vector_size_names[vectorSize][0] == '3')
+    {
+        if (operatorToUse_names[operatorToUse][0] == 'h'
+            || operatorToUse_names[operatorToUse][0] == 'o') // hi or odd
         {
             cmpVectorSize = 1; // special case for vec3 ignored values
         }
@@ -283,137 +278,17 @@ static int CheckResults( void *in, void *out, size_t elementCount, int type, int
     switch (elementSize)
     {
         case 1:
-        {
-            char *i = (char*)in;
-            char *o = (char*)out;
-            size_t j;
-            cl_uint k;
-            OffsetFunc f = offsetFuncs[ operatorToUse ];
-
-            for( k = 0; k  < elementCount; k++ )
-            {
-                char *o2 = (char*)p;
-                for( j = 0; j < halfVectorSize; j++ )
-                    o2[j] = i[ f((int)j, (int)halfVectorSize*2) ];
-
-                if( memcmp( o, o2, elementSize * cmpVectorSize ) )
-                {
-                    log_info( "\n%d) Failure for %s%s.%s { %d", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] );
-                    for( j = 1; j < halfVectorSize * 2; j++ )
-                        log_info( ", %d", i[j] );
-                    log_info( " } --> { %d", o[0] );
-                    for( j = 1; j < halfVectorSize; j++ )
-                        log_info( ", %d", o[j] );
-                    log_info( " }\n" );
-                    return -1;
-                }
-                i += 2 * halfVectorSize;
-                o += halfVectorSize;
-            }
-        }
-            break;
-
+            return verify<char>(in, out, elementCount, type, vectorSize,
+                                operatorToUse, cmpVectorSize);
         case 2:
-        {
-            short *i = (short*)in;
-            short *o = (short*)out;
-            size_t j;
-            cl_uint k;
-
-            for( k = 0; k  < elementCount; k++ )
-            {
-                short *o2 = (short*)p;
-                for( j = 0; j < halfVectorSize; j++ )
-                    o2[j] = i[ f((int)j, (int)halfVectorSize*2) ];
-
-                if( memcmp( o, o2, elementSize * cmpVectorSize ) )
-                {
-                    log_info( "\n%d) Failure for %s%s.%s { %d", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] );
-                    for( j = 1; j < halfVectorSize * 2; j++ )
-                        log_info( ", %d", i[j] );
-                    log_info( " } --> { %d", o[0] );
-                    for( j = 1; j < halfVectorSize; j++ )
-                        log_info( ", %d", o[j] );
-                    log_info( " }\n" );
-                    return -1;
-                }
-                i += 2 * halfVectorSize;
-                o += halfVectorSize;
-            }
-        }
-            break;
-
+            return verify<short>(in, out, elementCount, type, vectorSize,
+                                 operatorToUse, cmpVectorSize);
         case 4:
-        {
-            int *i = (int*)in;
-            int *o = (int*)out;
-            size_t j;
-            cl_uint k;
-
-            for( k = 0; k  < elementCount; k++ )
-            {
-                int *o2 = (int *)p;
-                for( j = 0; j < halfVectorSize; j++ )
-                    o2[j] = i[ f((int)j, (int)halfVectorSize*2) ];
-
-                for( j = 0; j < cmpVectorSize; j++ )
-        {
-            /* Allow float nans to be binary different */
-            if( memcmp( &o[j], &o2[j], elementSize ) && !((strcmp(test_str_names[type], "float") == 0) && isnan(((float *)o)[j]) && isnan(((float *)o2)[j])))
-            {
-                log_info( "\n%d) Failure for %s%s.%s { 0x%8.8x", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] );
-            for( j = 1; j < halfVectorSize * 2; j++ )
-                log_info( ", 0x%8.8x", i[j] );
-            log_info( " } --> { 0x%8.8x", o[0] );
-            for( j = 1; j < halfVectorSize; j++ )
-                log_info( ", 0x%8.8x", o[j] );
-            log_info( " }\n" );
-            return -1;
-            }
-        }
-        i += 2 * halfVectorSize;
-        o += halfVectorSize;
-            }
-        }
-            break;
-
+            return verify<int>(in, out, elementCount, type, vectorSize,
+                               operatorToUse, cmpVectorSize);
         case 8:
-        {
-            cl_ulong *i = (cl_ulong*)in;
-            cl_ulong *o = (cl_ulong*)out;
-            size_t j;
-            cl_uint k;
-
-            for( k = 0; k  < elementCount; k++ )
-            {
-                cl_ulong *o2 = (cl_ulong*)p;
-                for( j = 0; j < halfVectorSize; j++ )
-                    o2[j] = i[ f((int)j, (int)halfVectorSize*2) ];
-
-                if( memcmp( o, o2, elementSize * cmpVectorSize ) )
-                {
-                    log_info( "\n%d) Failure for %s%s.%s { 0x%16.16llx", k, test_str_names[type], vector_size_names[ vectorSize ], operatorToUse_names[ operatorToUse ], i[0] );
-                    for( j = 1; j < halfVectorSize * 2; j++ )
-                        log_info( ", 0x%16.16llx", i[j] );
-                    log_info( " } --> { 0x%16.16llx", o[0] );
-                    for( j = 1; j < halfVectorSize; j++ )
-                        log_info( ", 0x%16.16llx", o[j] );
-                    log_info( " }\n" );
-                    return -1;
-                }
-                i += 2 * halfVectorSize;
-                o += halfVectorSize;
+            return verify<cl_ulong>(in, out, elementCount, type, vectorSize,
+                                    operatorToUse, cmpVectorSize);
+        default: log_info("Internal error. Unknown data type\n"); return -2;
     }
 }
-            break;
-
-        default:
-            log_info( "Internal error. Unknown data type\n" );
-            return -2;
-    }
-
-    return 0;
-}
-
-
-

test_conformance/basic/test_int2float.cpp

@@ -21,123 +21,120 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 
+#include <algorithm>
+#include <vector>
 
 #include "procs.h"
 
-const char *int2float_kernel_code =
-"__kernel void test_int2float(__global int *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = (float)src[tid];\n"
-"\n"
-"}\n";
-
-
-int
-verify_int2float(cl_int *inptr, cl_float *outptr, int n)
+namespace {
+const char *int2float_kernel_code = R"(
+__kernel void test_X2Y(__global TYPE_X *src, __global TYPE_Y *dst)
 {
-    int     i;
+    int  tid = get_global_id(0);
 
-    for (i=0; i<n; i++)
+    dst[tid] = (TYPE_Y)src[tid];
+
+})";
+
+template <typename T> const char *Type2str() { return ""; }
+template <> const char *Type2str<cl_int>() { return "int"; }
+template <> const char *Type2str<cl_float>() { return "float"; }
+
+template <typename T> void generate_random_inputs(std::vector<T> &v)
 {
-        if (outptr[i] != (float)inptr[i])
+    RandomSeed seed(gRandomSeed);
+
+    auto random_generator = [&seed]() {
+        return get_random_float(-MAKE_HEX_FLOAT(0x1.0p31f, 0x1, 31),
+                                MAKE_HEX_FLOAT(0x1.0p31f, 0x1, 31), seed);
+    };
+
+    std::generate(v.begin(), v.end(), random_generator);
+}
+
+template <typename Tx, typename Ty> bool equal_value(Tx a, Ty b)
 {
-            log_error("INT2FLOAT test failed\n");
+    return a == (Tx)b;
+}
+
+template <typename Tx, typename Ty>
+int verify_X2Y(std::vector<Tx> input, std::vector<Ty> output,
+               const char *test_name)
+{
+
+    if (!std::equal(output.begin(), output.end(), input.begin(),
+                    equal_value<Tx, Ty>))
+    {
+        log_error("%s test failed\n", test_name);
         return -1;
     }
-    }
 
-    log_info("INT2FLOAT test passed\n");
+    log_info("%s test passed\n", test_name);
     return 0;
 }
-
-int
-test_int2float(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
+template <typename Tx, typename Ty>
+int test_X2Y(cl_device_id device, cl_context context, cl_command_queue queue,
+             int num_elements, const char *test_name)
 {
-    cl_mem            streams[2];
-    cl_int            *input_ptr;
-    cl_float        *output_ptr;
-    cl_program        program;
-    cl_kernel        kernel;
-    size_t    threads[1];
+    clMemWrapper streams[2];
+    clProgramWrapper program;
+    clKernelWrapper kernel;
     int err;
-    int                i;
-    MTdata          d;
 
-    input_ptr = (cl_int*)malloc(sizeof(cl_int) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
+
+    std::vector<Tx> input(num_elements);
+    std::vector<Ty> output(num_elements);
+
     streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_int) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
+                                sizeof(Tx) * num_elements, nullptr, &err);
+    test_error(err, "clCreateBuffer failed.");
     streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
+                                sizeof(Ty) * num_elements, nullptr, &err);
+    test_error(err, "clCreateBuffer failed.");
 
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-        input_ptr[i] = (cl_int)get_random_float(-MAKE_HEX_FLOAT( 0x1.0p31f, 0x1, 31), MAKE_HEX_FLOAT( 0x1.0p31f, 0x1, 31), d);
-    free_mtdata(d); d = NULL;
+    generate_random_inputs(input);
 
-    err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, sizeof(cl_int)*num_elements, (void *)input_ptr, 0, NULL, NULL);
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
+    err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0,
+                               sizeof(Tx) * num_elements, input.data(), 0,
+                               nullptr, nullptr);
+    test_error(err, "clEnqueueWriteBuffer failed.");
 
-    err = create_single_kernel_helper(context, &program, &kernel, 1, &int2float_kernel_code, "test_int2float");
-    if (err != CL_SUCCESS)
-    {
-        log_error("create_single_kernel_helper failed\n");
-        return -1;
-    }
+    std::string build_options;
+    build_options.append("-DTYPE_X=").append(Type2str<Tx>());
+    build_options.append(" -DTYPE_Y=").append(Type2str<Ty>());
+    err = create_single_kernel_helper(context, &program, &kernel, 1,
+                                      &int2float_kernel_code, "test_X2Y",
+                                      build_options.c_str());
+    test_error(err, "create_single_kernel_helper failed.");
 
     err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0]);
     err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1]);
-    if (err != CL_SUCCESS)
-    {
-        log_error("clSetKernelArgs failed\n");
-        return -1;
-    }
+    test_error(err, "clSetKernelArg failed.");
 
-    threads[0] = (size_t)num_elements;
-    err = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clEnqueueNDRangeKernel failed\n");
-        return -1;
-    }
+    size_t threads[] = { (size_t)num_elements };
+    err = clEnqueueNDRangeKernel(queue, kernel, 1, nullptr, threads, nullptr, 0,
+                                 nullptr, nullptr);
+    test_error(err, "clEnqueueNDRangeKernel failed.");
 
-    err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clEnqueueReadBuffer failed\n");
-        return -1;
-    }
+    err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0,
+                              sizeof(Ty) * num_elements, output.data(), 0,
+                              nullptr, nullptr);
+    test_error(err, "clEnqueueReadBuffer failed.");
 
-    err = verify_int2float(input_ptr, output_ptr, num_elements);
-
-    // cleanup
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseKernel(kernel);
-    clReleaseProgram(program);
-    free(input_ptr);
-    free(output_ptr);
+    err = verify_X2Y(input, output, test_name);
 
     return err;
 }
-
-
-
-
-
+}
+int test_int2float(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int num_elements)
+{
+    return test_X2Y<cl_int, cl_float>(device, context, queue, num_elements,
+                                      "INT2FLOAT");
+}
+int test_float2int(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int num_elements)
+{
+    return test_X2Y<cl_float, cl_int>(device, context, queue, num_elements,
+                                      "FLOAT2INT");
+}

test_conformance/basic/test_progvar.cpp

@@ -25,7 +25,6 @@
 
 #define ALIGNMENT 128
 
-#define OPTIONS "-cl-std=CL2.0"
 
 // NUM_ROUNDS must be at least 1.
 // It determines how many sets of random data we push through the global
@@ -439,6 +438,7 @@ static int l_capacity(cl_device_id device, cl_context context,
 static int l_user_type(cl_device_id device, cl_context context,
                        cl_command_queue queue, bool separate_compile);
 
+static std::string get_build_options(cl_device_id device);
 
 ////////////////////
 // File scope function definitions
@@ -1116,9 +1116,8 @@ static int l_write_read_for_type(cl_device_id device, cl_context context,
     clProgramWrapper program;
     clKernelWrapper writer;
 
-    status = create_single_kernel_helper_with_build_options(
-        context, &program, &writer, ksrc.num_str(), ksrc.strs(), "writer",
-        OPTIONS);
+    status = create_single_kernel_helper(context, &program, &writer,
+                                         ksrc.num_str(), ksrc.strs(), "writer");
     test_error_ret(status, "Failed to create program for read-after-write test",
                    status);
 
@@ -1326,9 +1325,8 @@ static int l_init_write_read_for_type(cl_device_id device, cl_context context,
     clProgramWrapper program;
     clKernelWrapper writer;
 
-    status = create_single_kernel_helper_with_build_options(
-        context, &program, &writer, ksrc.num_str(), ksrc.strs(), "writer",
-        OPTIONS);
+    status = create_single_kernel_helper(context, &program, &writer,
+                                         ksrc.num_str(), ksrc.strs(), "writer");
     test_error_ret(status,
                    "Failed to create program for init-read-after-write test",
                    status);
@@ -1581,9 +1579,9 @@ static int l_capacity(cl_device_id device, cl_context context,
     clProgramWrapper program;
     clKernelWrapper get_max_size;
 
-    status = create_single_kernel_helper_with_build_options(
-        context, &program, &get_max_size, ksrc.num_str(), ksrc.strs(),
-        "get_max_size", OPTIONS);
+    status = create_single_kernel_helper(context, &program, &get_max_size,
+                                         ksrc.num_str(), ksrc.strs(),
+                                         "get_max_size");
     test_error_ret(status, "Failed to create program for capacity test",
                    status);
 
@@ -1737,6 +1735,8 @@ static int l_user_type(cl_device_id device, cl_context context,
 
     clProgramWrapper program;
 
+    const std::string options = get_build_options(device);
+
     if (separate_compile)
     {
         // Separate compilation flow.
@@ -1757,15 +1757,15 @@ static int l_user_type(cl_device_id device, cl_context context,
                        "Failed to create writer program for user type test",
                        status);
 
-        status = clCompileProgram(writer_program, 1, &device, OPTIONS, 0, 0, 0,
-                                  0, 0);
+        status = clCompileProgram(writer_program, 1, &device, options.c_str(),
+                                  0, 0, 0, 0, 0);
         if (check_error(
                 status,
                 "Failed to compile writer program for user type test (%s)",
                 IGetErrorString(status)))
         {
             print_build_log(writer_program, 1, &device, wksrc.num_str(),
-                            wksrc.strs(), wksrc.lengths(), OPTIONS);
+                            wksrc.strs(), wksrc.lengths(), options.c_str());
             return status;
         }
 
@@ -1775,15 +1775,15 @@ static int l_user_type(cl_device_id device, cl_context context,
                        "Failed to create reader program for user type test",
                        status);
 
-        status = clCompileProgram(reader_program, 1, &device, OPTIONS, 0, 0, 0,
-                                  0, 0);
+        status = clCompileProgram(reader_program, 1, &device, options.c_str(),
+                                  0, 0, 0, 0, 0);
         if (check_error(
                 status,
                 "Failed to compile reader program for user type test (%s)",
                 IGetErrorString(status)))
         {
             print_build_log(reader_program, 1, &device, rksrc.num_str(),
-                            rksrc.strs(), rksrc.lengths(), OPTIONS);
+                            rksrc.strs(), rksrc.lengths(), options.c_str());
             return status;
         }
 
@@ -1813,23 +1813,23 @@ static int l_user_type(cl_device_id device, cl_context context,
         int status = CL_SUCCESS;
 
         status = create_single_kernel_helper_create_program(
-            context, &program, ksrc.num_str(), ksrc.strs(), OPTIONS);
+            context, &program, ksrc.num_str(), ksrc.strs(), options.c_str());
         if (check_error(status,
                         "Failed to build program for user type test (%s)",
                         IGetErrorString(status)))
         {
             print_build_log(program, 1, &device, ksrc.num_str(), ksrc.strs(),
-                            ksrc.lengths(), OPTIONS);
+                            ksrc.lengths(), options.c_str());
             return status;
         }
 
-        status = clBuildProgram(program, 1, &device, OPTIONS, 0, 0);
+        status = clBuildProgram(program, 1, &device, options.c_str(), 0, 0);
         if (check_error(status,
                         "Failed to compile program for user type test (%s)",
                         IGetErrorString(status)))
         {
             print_build_log(program, 1, &device, ksrc.num_str(), ksrc.strs(),
-                            ksrc.lengths(), OPTIONS);
+                            ksrc.lengths(), options.c_str());
             return status;
         }
     }
@@ -1935,6 +1935,14 @@ static int l_user_type(cl_device_id device, cl_context context,
     return err;
 }
 
+static std::string get_build_options(cl_device_id device)
+{
+    std::string options = "-cl-std=CL";
+    Version latest_cl_c_version = get_device_latest_cl_c_version(device);
+    options += latest_cl_c_version.to_string();
+    return options;
+}
+
 // Determines whether its valid to skip this test based on the driver version
 // and the features it optionally supports.
 // Whether the test should be skipped is writen into the out paramter skip.
@@ -2102,9 +2110,9 @@ int test_progvar_func_scope(cl_device_id device, cl_context context,
     clProgramWrapper program;
     clKernelWrapper test_bump;
 
-    status = create_single_kernel_helper_with_build_options(
-        context, &program, &test_bump, ksrc.num_str(), ksrc.strs(), "test_bump",
-        OPTIONS);
+    status =
+        create_single_kernel_helper(context, &program, &test_bump,
+                                    ksrc.num_str(), ksrc.strs(), "test_bump");
     test_error_ret(status,
                    "Failed to create program for function static variable test",
                    status);

test_conformance/basic/test_vector_swizzle.cpp

@@ -22,6 +22,8 @@
 #include "procs.h"
 #include "harness/testHarness.h"
 
+static std::string pragma_extension;
+
 template <int N> struct TestInfo
 {
 };
@@ -629,7 +631,9 @@ static int test_vectype(const char* type_name, cl_device_id device,
         clProgramWrapper program;
         clKernelWrapper kernel;
 
-        const char* xyzw_source = TestInfo<N>::kernel_source_xyzw;
+        std::string program_src =
+            pragma_extension + std::string(TestInfo<N>::kernel_source_xyzw);
+        const char* xyzw_source = program_src.c_str();
         error = create_single_kernel_helper(
             context, &program, &kernel, 1, &xyzw_source,
             "test_vector_swizzle_xyzw", buildOptions.c_str());
@@ -643,7 +647,9 @@ static int test_vectype(const char* type_name, cl_device_id device,
         clProgramWrapper program;
         clKernelWrapper kernel;
 
-        const char* sN_source = TestInfo<N>::kernel_source_sN;
+        std::string program_src =
+            pragma_extension + std::string(TestInfo<N>::kernel_source_sN);
+        const char* sN_source = program_src.c_str();
         error = create_single_kernel_helper(
             context, &program, &kernel, 1, &sN_source, "test_vector_swizzle_sN",
             buildOptions.c_str());
@@ -660,7 +666,9 @@ static int test_vectype(const char* type_name, cl_device_id device,
         const Version device_version = get_device_cl_version(device);
         if (device_version >= Version(3, 0))
         {
-            const char* rgba_source = TestInfo<N>::kernel_source_rgba;
+            std::string program_src =
+                pragma_extension + std::string(TestInfo<N>::kernel_source_rgba);
+            const char* rgba_source = program_src.c_str();
             error = create_single_kernel_helper(
                 context, &program, &kernel, 1, &rgba_source,
                 "test_vector_swizzle_rgba", buildOptions.c_str());
@@ -689,6 +697,7 @@ int test_vector_swizzle(cl_device_id device, cl_context context,
                         cl_command_queue queue, int num_elements)
 {
     int hasDouble = is_extension_available(device, "cl_khr_fp64");
+    int hasHalf = is_extension_available(device, "cl_khr_fp16");
 
     int result = TEST_PASS;
     result |= test_type<cl_char>("char", device, context, queue);
@@ -703,8 +712,14 @@ int test_vector_swizzle(cl_device_id device, cl_context context,
         result |= test_type<cl_ulong>("ulong", device, context, queue);
     }
     result |= test_type<cl_float>("float", device, context, queue);
+    if (hasHalf)
+    {
+        pragma_extension = "#pragma OPENCL EXTENSION cl_khr_fp16 : enable\n";
+        result |= test_type<cl_half>("half", device, context, queue);
+    }
     if (hasDouble)
     {
+        pragma_extension = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
         result |= test_type<cl_double>("double", device, context, queue);
     }
     return result;

test_conformance/basic/utils.h

@@ -0,0 +1,41 @@
+//
+// Copyright (c) 2023 The Khronos Group Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+#ifndef BASIC_UTILS_H
+#define BASIC_UTILS_H
+
+#include <memory>
+#include <string>
+
+inline std::string concat_kernel(const char *sstr[], int num)
+{
+    std::string res;
+    for (int i = 0; i < num; i++) res += std::string(sstr[i]);
+    return res;
+}
+
+template <typename... Args>
+inline std::string str_sprintf(const std::string &str, Args... args)
+{
+    int str_size = std::snprintf(nullptr, 0, str.c_str(), args...) + 1;
+    if (str_size <= 0) throw std::runtime_error("Formatting error.");
+    size_t s = static_cast<size_t>(str_size);
+    std::unique_ptr<char[]> buffer(new char[s]);
+    std::snprintf(buffer.get(), s, str.c_str(), args...);
+    return std::string(buffer.get(), buffer.get() + s - 1);
+}
+
+#endif // BASIC_UTIL_H

test_conformance/c11_atomics/common.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _COMMON_H_
-#define _COMMON_H_
+#ifndef COMMON_H_
+#define COMMON_H_
 
 #include "harness/testHarness.h"
 #include "harness/typeWrappers.h"
@@ -1567,4 +1567,4 @@ int CBasicTest<HostAtomicType, HostDataType>::ExecuteSingleTest(
     return 0;
 }
 
-#endif //_COMMON_H_
+#endif // COMMON_H_

test_conformance/c11_atomics/host_atomics.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _HOST_ATOMICS_H_
-#define _HOST_ATOMICS_H_
+#ifndef HOST_ATOMICS_H_
+#define HOST_ATOMICS_H_
 
 #include "harness/testHarness.h"
 
@@ -247,4 +247,4 @@ CorrespondingType host_atomic_fetch_max(volatile AtomicType *a, CorrespondingTyp
 bool host_atomic_flag_test_and_set(volatile HOST_ATOMIC_FLAG *a, TExplicitMemoryOrderType order);
 void host_atomic_flag_clear(volatile HOST_ATOMIC_FLAG *a, TExplicitMemoryOrderType order);
 
-#endif //_HOST_ATOMICS_H_
+#endif // HOST_ATOMICS_H_

test_conformance/commonfns/CMakeLists.txt

@@ -3,22 +3,10 @@ set(MODULE_NAME COMMONFNS)
 set(${MODULE_NAME}_SOURCES
     main.cpp
     test_clamp.cpp
-    test_degrees.cpp
-    test_max.cpp
-    test_maxf.cpp
-    test_min.cpp
-    test_minf.cpp
+    test_unary_fn.cpp
     test_mix.cpp
-    test_radians.cpp
     test_step.cpp
-    test_stepf.cpp
     test_smoothstep.cpp
-    test_smoothstepf.cpp
-    test_sign.cpp
-    test_fmax.cpp
-    test_fmin.cpp
-    test_fmaxf.cpp
-    test_fminf.cpp
     test_binary_fn.cpp
 )
 

test_conformance/commonfns/main.cpp

@@ -13,11 +13,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
 
 #include <stdio.h>
 #include <string.h>
 #include "procs.h"
+#include "test_base.h"
+
+std::map<size_t, std::string> BaseFunctionTest::type2name;
 
 int g_arrVecSizes[kVectorSizeCount + kStrangeVectorSizeCount];
 int g_arrStrangeVectorSizes[kStrangeVectorSizeCount] = {3};
@@ -32,25 +34,13 @@ static void initVecSizes() {
     }
 }
 
-
 test_definition test_list[] = {
-    ADD_TEST( clamp ),
-    ADD_TEST( degrees ),
-    ADD_TEST( fmax ),
-    ADD_TEST( fmaxf ),
-    ADD_TEST( fmin ),
-    ADD_TEST( fminf ),
-    ADD_TEST( max ),
-    ADD_TEST( maxf ),
-    ADD_TEST( min ),
-    ADD_TEST( minf ),
-    ADD_TEST( mix ),
-    ADD_TEST( radians ),
-    ADD_TEST( step ),
-    ADD_TEST( stepf ),
-    ADD_TEST( smoothstep ),
-    ADD_TEST( smoothstepf ),
-    ADD_TEST( sign ),
+    ADD_TEST(clamp),      ADD_TEST(degrees),     ADD_TEST(fmax),
+    ADD_TEST(fmaxf),      ADD_TEST(fmin),        ADD_TEST(fminf),
+    ADD_TEST(max),        ADD_TEST(maxf),        ADD_TEST(min),
+    ADD_TEST(minf),       ADD_TEST(mix),         ADD_TEST(mixf),
+    ADD_TEST(radians),    ADD_TEST(step),        ADD_TEST(stepf),
+    ADD_TEST(smoothstep), ADD_TEST(smoothstepf), ADD_TEST(sign),
 };
 
 const int test_num = ARRAY_SIZE( test_list );
@@ -58,6 +48,14 @@ const int test_num = ARRAY_SIZE( test_list );
 int main(int argc, const char *argv[])
 {
     initVecSizes();
+
+    if (BaseFunctionTest::type2name.empty())
+    {
+        BaseFunctionTest::type2name[sizeof(half)] = "half";
+        BaseFunctionTest::type2name[sizeof(float)] = "float";
+        BaseFunctionTest::type2name[sizeof(double)] = "double";
+    }
+
     return runTestHarness(argc, argv, test_num, test_list, false, 0);
 }
 

test_conformance/commonfns/procs.h

@@ -37,6 +37,8 @@ extern int        test_maxf(cl_device_id device, cl_context context, cl_command_
 extern int        test_min(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 extern int        test_minf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 extern int        test_mix(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int test_mixf(cl_device_id device, cl_context context,
+                     cl_command_queue queue, int num_elements);
 extern int        test_radians(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 extern int        test_step(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 extern int        test_stepf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
@@ -44,11 +46,4 @@ extern int        test_smoothstep(cl_device_id device, cl_context context, cl_co
 extern int        test_smoothstepf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 extern int        test_sign(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
 
-typedef int     (*binary_verify_float_fn)( float *x, float *y, float *out, int numElements, int vecSize );
-typedef int     (*binary_verify_double_fn)( double *x, double *y, double *out, int numElements, int vecSize );
-
-extern int      test_binary_fn( cl_device_id device, cl_context context, cl_command_queue queue, int n_elems,
-                           const char *fnName, bool vectorSecondParam,
-                           binary_verify_float_fn floatVerifyFn, binary_verify_double_fn doubleVerifyFn );
-
 

test_conformance/commonfns/test_base.h

@@ -0,0 +1,193 @@
+// Copyright (c) 2023 The Khronos Group Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+#ifndef TEST_COMMONFNS_BASE_H
+#define TEST_COMMONFNS_BASE_H
+
+#include <vector>
+#include <map>
+#include <memory>
+
+#include <CL/cl_half.h>
+#include <CL/cl_ext.h>
+
+#include "harness/deviceInfo.h"
+#include "harness/testHarness.h"
+#include "harness/typeWrappers.h"
+
+
+template <typename T>
+using VerifyFuncBinary = int (*)(const T *const, const T *const, const T *const,
+                                 const int num, const int vs, const int vp);
+
+
+template <typename T>
+using VerifyFuncUnary = int (*)(const T *const, const T *const, const int num);
+
+
+using half = cl_half;
+
+
+struct BaseFunctionTest
+{
+    BaseFunctionTest(cl_device_id device, cl_context context,
+                     cl_command_queue queue, int num_elems, const char *fn,
+                     bool vsp)
+        : device(device), context(context), queue(queue), num_elems(num_elems),
+          fnName(fn), vecParam(vsp)
+    {}
+
+    // Test body returning an OpenCL error code
+    virtual cl_int Run() = 0;
+
+    cl_device_id device;
+    cl_context context;
+    cl_command_queue queue;
+
+    int num_elems;
+    std::string fnName;
+    bool vecParam;
+
+    static std::map<size_t, std::string> type2name;
+};
+
+
+struct MinTest : BaseFunctionTest
+{
+    MinTest(cl_device_id device, cl_context context, cl_command_queue queue,
+            int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct MaxTest : BaseFunctionTest
+{
+    MaxTest(cl_device_id device, cl_context context, cl_command_queue queue,
+            int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct ClampTest : BaseFunctionTest
+{
+    ClampTest(cl_device_id device, cl_context context, cl_command_queue queue,
+              int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct DegreesTest : BaseFunctionTest
+{
+    DegreesTest(cl_device_id device, cl_context context, cl_command_queue queue,
+                int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct RadiansTest : BaseFunctionTest
+{
+    RadiansTest(cl_device_id device, cl_context context, cl_command_queue queue,
+                int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct SignTest : BaseFunctionTest
+{
+    SignTest(cl_device_id device, cl_context context, cl_command_queue queue,
+             int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct SmoothstepTest : BaseFunctionTest
+{
+    SmoothstepTest(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int num_elems, const char *fn,
+                   bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct StepTest : BaseFunctionTest
+{
+    StepTest(cl_device_id device, cl_context context, cl_command_queue queue,
+             int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+struct MixTest : BaseFunctionTest
+{
+    MixTest(cl_device_id device, cl_context context, cl_command_queue queue,
+            int num_elems, const char *fn, bool vsp)
+        : BaseFunctionTest(device, context, queue, num_elems, fn, vsp)
+    {}
+
+    cl_int Run() override;
+};
+
+
+template <typename... Args>
+std::string string_format(const std::string &format, Args... args)
+{
+    int sformat = std::snprintf(nullptr, 0, format.c_str(), args...) + 1;
+    if (sformat <= 0)
+        throw std::runtime_error("string_format: string processing error.");
+    auto format_size = static_cast<size_t>(sformat);
+    std::unique_ptr<char[]> buffer(new char[format_size]);
+    std::snprintf(buffer.get(), format_size, format.c_str(), args...);
+    return std::string(buffer.get(), buffer.get() + format_size - 1);
+}
+
+
+template <class T>
+int MakeAndRunTest(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int num_elements,
+                   const char *fn = "", bool vsp = false)
+{
+    auto test_fixture = T(device, context, queue, num_elements, fn, vsp);
+
+    cl_int error = test_fixture.Run();
+    test_error_ret(error, "Test Failed", TEST_FAIL);
+
+    return TEST_PASS;
+}
+
+#endif // TEST_COMMONFNS_BASE_H

test_conformance/commonfns/test_binary_fn.cpp

@@ -13,14 +13,18 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
 
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <vector>
+
+#include "harness/deviceInfo.h"
+#include "harness/typeWrappers.h"
 
 #include "procs.h"
+#include "test_base.h"
 
 const char *binary_fn_code_pattern =
 "%s\n" /* optional pragma */
@@ -49,216 +53,286 @@ const char *binary_fn_code_pattern_v3_scalar =
 "    vstore3(%s(vload3(tid,x), y[tid] ), tid, dst);\n"
 "}\n";
 
-int test_binary_fn( cl_device_id device, cl_context context, cl_command_queue queue, int n_elems,
-                    const char *fnName, bool vectorSecondParam,
-                    binary_verify_float_fn floatVerifyFn, binary_verify_double_fn doubleVerifyFn )
+
+template <typename T>
+int test_binary_fn(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int n_elems,
+                   const std::string& fnName, bool vecSecParam,
+                   VerifyFuncBinary<T> verifyFn)
 {
-    cl_mem      streams[6];
-    cl_float      *input_ptr[2], *output_ptr;
-    cl_double     *input_ptr_double[2], *output_ptr_double=NULL;
-    cl_program  *program;
-    cl_kernel   *kernel;
-    size_t threads[1];
-    int num_elements;
-    int err;
-    int i, j;
-    MTdata d;
+    clMemWrapper streams[3];
+    std::vector<T> input_ptr[2], output_ptr;
 
-      program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount*2);
-      kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount*2);
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+    int err, i, j;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
 
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
 
-    int test_double = 0;
-    if(is_extension_available( device, "cl_khr_fp64" ))
-    {
-        log_info("Testing doubles.\n");
-        test_double = 1;
-    }
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
 
-    for( i = 0; i < 2; i++ )
-    {
-        input_ptr[i] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-        if (test_double) input_ptr_double[i] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    }
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    if (test_double) output_ptr_double = (cl_double*)malloc(sizeof(cl_double) * num_elements);
+    int num_elements = n_elems * (1 << (kTotalVecCount - 1));
+
+    for (i = 0; i < 2; i++) input_ptr[i].resize(num_elements);
+    output_ptr.resize(num_elements);
 
     for( i = 0; i < 3; i++ )
     {
-        streams[i] =
-            clCreateBuffer(context, CL_MEM_READ_WRITE,
-                           sizeof(cl_float) * num_elements, NULL, &err);
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
         test_error( err, "clCreateBuffer failed");
     }
 
-    if (test_double)
-        for( i = 3; i < 6; i++ )
+    std::string pragma_str;
+    if (std::is_same<T, float>::value)
     {
-            streams[i] =
-                clCreateBuffer(context, CL_MEM_READ_WRITE,
-                               sizeof(cl_double) * num_elements, NULL, &err);
-            test_error(err, "clCreateBuffer failed");
-        }
-
-    d = init_genrand( gRandomSeed );
         for (j = 0; j < num_elements; j++)
         {
             input_ptr[0][j] = get_random_float(-0x20000000, 0x20000000, d);
             input_ptr[1][j] = get_random_float(-0x20000000, 0x20000000, d);
-        if (test_double)
+        }
+    }
+    else if (std::is_same<T, double>::value)
     {
-            input_ptr_double[0][j] = get_random_double(-0x20000000, 0x20000000, d);
-            input_ptr_double[1][j] = get_random_double(-0x20000000, 0x20000000, d);
+        pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
+        for (j = 0; j < num_elements; j++)
+        {
+            input_ptr[0][j] = get_random_double(-0x20000000, 0x20000000, d);
+            input_ptr[1][j] = get_random_double(-0x20000000, 0x20000000, d);
         }
     }
-    free_mtdata(d);     d = NULL;
 
     for (i = 0; i < 2; i++)
     {
-        err = clEnqueueWriteBuffer( queue, streams[ i ], CL_TRUE, 0, sizeof( cl_float ) * num_elements, input_ptr[ i ], 0, NULL, NULL );
-        test_error( err, "Unable to write input buffer" );
-
-        if (test_double)
-        {
-          err = clEnqueueWriteBuffer( queue, streams[ 3 + i ], CL_TRUE, 0, sizeof( cl_double ) * num_elements, input_ptr_double[ i ], 0, NULL, NULL );
+        err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
+                                   sizeof(T) * num_elements,
+                                   &input_ptr[i].front(), 0, NULL, NULL);
         test_error(err, "Unable to write input buffer");
     }
-    }
 
-    for( i = 0; i < kTotalVecCount; i++ )
-    {
-        char programSrc[ 10240 ];
     char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
 
-        if(i >= kVectorSizeCount) {
-            // do vec3 print
-
-            if(vectorSecondParam) {
-            sprintf( programSrc,binary_fn_code_pattern_v3, "", "float", "float", "float", fnName );
-        } else  {
-            sprintf( programSrc,binary_fn_code_pattern_v3_scalar, "", "float", "float", "float", fnName );
+    for (i = 0; i < kTotalVecCount; i++)
+    {
+        std::string kernelSource;
+        if (i >= kVectorSizeCount)
+        {
+            if (vecSecParam)
+            {
+                std::string str = binary_fn_code_pattern_v3;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), fnName.c_str());
             }
-        } else  {
+            else
+            {
+                std::string str = binary_fn_code_pattern_v3_scalar;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), fnName.c_str());
+            }
+        }
+        else
+        {
             // do regular
-            sprintf( programSrc, binary_fn_code_pattern, "", "float", vecSizeNames[ i ], "float", vectorSecondParam ? vecSizeNames[ i ] : "", "float", vecSizeNames[ i ], fnName );
+            std::string str = binary_fn_code_pattern;
+            kernelSource = string_format(
+                str, pragma_str.c_str(), tname.c_str(), vecSizeNames[i],
+                tname.c_str(), vecSecParam ? vecSizeNames[i] : "",
+                tname.c_str(), vecSizeNames[i], fnName.c_str());
         }
-        const char *ptr = programSrc;
-        err = create_single_kernel_helper( context, &program[ i ], &kernel[ i ], 1, &ptr, "test_fn" );
+        const char* programPtr = kernelSource.c_str();
+        err = create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
+                                          (const char**)&programPtr, "test_fn");
         test_error(err, "Unable to create kernel");
 
-        if (test_double)
-        {
-        if(i >= kVectorSizeCount) {
-        if(vectorSecondParam) {
-            sprintf( programSrc, binary_fn_code_pattern_v3, "#pragma OPENCL EXTENSION cl_khr_fp64 : enable",
-            "double",  "double",  "double",  fnName );
-        } else {
-
-        sprintf( programSrc, binary_fn_code_pattern_v3_scalar, "#pragma OPENCL EXTENSION cl_khr_fp64 : enable",
-                 "double",  "double",  "double",  fnName );
-        }
-        } else {
-        sprintf( programSrc, binary_fn_code_pattern, "#pragma OPENCL EXTENSION cl_khr_fp64 : enable",
-            "double", vecSizeNames[ i ], "double", vectorSecondParam ? vecSizeNames[ i ] : "", "double", vecSizeNames[ i ], fnName );
-        }
-            ptr = programSrc;
-            err = create_single_kernel_helper( context, &program[ kTotalVecCount + i ], &kernel[ kTotalVecCount + i ], 1, &ptr, "test_fn" );
-            test_error( err, "Unable to create kernel" );
-        }
-    }
-
-    for( i = 0; i < kTotalVecCount; i++ )
-    {
         for( j = 0; j < 3; j++ )
         {
-            err = clSetKernelArg( kernel[ i ], j, sizeof( streams[ j ] ), &streams[ j ] );
+            err =
+                clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
             test_error( err, "Unable to set kernel argument" );
         }
 
-        threads[0] = (size_t)n_elems;
+        size_t threads = (size_t)n_elems;
 
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
         test_error( err, "Unable to execute kernel" );
 
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
+        err = clEnqueueReadBuffer(queue, streams[2], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
         test_error( err, "Unable to read results" );
 
-
-
-        if( floatVerifyFn( input_ptr[0], input_ptr[1], output_ptr, n_elems, ((g_arrVecSizes[i])) ) )
+        if (verifyFn((T*)&input_ptr[0].front(), (T*)&input_ptr[1].front(),
+                     &output_ptr[0], n_elems, g_arrVecSizes[i],
+                     vecSecParam ? 1 : 0))
         {
-            log_error(" float%d%s test failed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", float");
+            log_error("%s %s%d%s test failed\n", fnName.c_str(), tname.c_str(),
+                      ((g_arrVecSizes[i])),
+                      vecSecParam ? "" : std::string(", " + tname).c_str());
             err = -1;
         }
         else
         {
-            log_info(" float%d%s test passed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", float");
+            log_info("%s %s%d%s test passed\n", fnName.c_str(), tname.c_str(),
+                     ((g_arrVecSizes[i])),
+                     vecSecParam ? "" : std::string(", " + tname).c_str());
             err = 0;
         }
 
         if (err)
             break;
     }
-
-    if (test_double)
-    {
-        for( i = 0; i < kTotalVecCount; i++ )
-        {
-            for( j = 0; j < 3; j++ )
-            {
-                err = clSetKernelArg( kernel[ kTotalVecCount + i ], j, sizeof( streams[ 3 + j ] ), &streams[ 3 + j ] );
-                test_error( err, "Unable to set kernel argument" );
-            }
-
-            threads[0] = (size_t)n_elems;
-
-            err = clEnqueueNDRangeKernel( queue, kernel[kTotalVecCount + i], 1, NULL, threads, NULL, 0, NULL, NULL );
-            test_error( err, "Unable to execute kernel" );
-
-            err = clEnqueueReadBuffer( queue, streams[5], CL_TRUE, 0, sizeof(cl_double)*num_elements, (void *)output_ptr_double, 0, NULL, NULL );
-            test_error( err, "Unable to read results" );
-
-            if( doubleVerifyFn( input_ptr_double[0], input_ptr_double[1], output_ptr_double, n_elems, ((g_arrVecSizes[i]))))
-            {
-                log_error(" double%d%s test failed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", double");
-                err = -1;
-            }
-            else
-            {
-                log_info(" double%d%s test passed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", double");
-                err = 0;
-            }
-
-            if (err)
-            break;
-        }
-    }
-
-
-    for( i = 0; i < ((test_double) ? 6 : 3); i++ )
-    {
-        clReleaseMemObject(streams[i]);
-    }
-    for (i=0; i < ((test_double) ? kTotalVecCount * 2 : kTotalVecCount) ; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-      free(program);
-      free(kernel);
-
-    if (test_double)
-    {
-        free(input_ptr_double[0]);
-        free(input_ptr_double[1]);
-        free(output_ptr_double);
-    }
-
     return err;
 }
 
+namespace {
 
+template <typename T>
+int max_verify(const T* const x, const T* const y, const T* const out,
+               int numElements, int vecSize, int vecParam)
+{
+    for (int i = 0; i < numElements; i++)
+    {
+        for (int j = 0; j < vecSize; j++)
+        {
+            int k = i * vecSize + j;
+            int l = (k * vecParam + i * (1 - vecParam));
+            T v = (x[k] < y[l]) ? y[l] : x[k];
+            if (v != out[k])
+            {
+                log_error(
+                    "x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is "
+                    "vector %d, element %d, for vector size %d)\n",
+                    k, x[k], l, y[l], k, out[k], v, k, i, j, vecSize);
+                return -1;
+            }
+        }
+    }
+    return 0;
+}
+
+template <typename T>
+int min_verify(const T* const x, const T* const y, const T* const out,
+               int numElements, int vecSize, int vecParam)
+{
+    for (int i = 0; i < numElements; i++)
+    {
+        for (int j = 0; j < vecSize; j++)
+        {
+            int k = i * vecSize + j;
+            int l = (k * vecParam + i * (1 - vecParam));
+            T v = (x[k] > y[l]) ? y[l] : x[k];
+            if (v != out[k])
+            {
+                log_error(
+                    "x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is "
+                    "vector %d, element %d, for vector size %d)\n",
+                    k, x[k], l, y[l], k, out[k], v, k, i, j, vecSize);
+                return -1;
+            }
+        }
+    }
+    return 0;
+}
+
+}
+
+cl_int MaxTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error = test_binary_fn<float>(device, context, queue, num_elems,
+                                  fnName.c_str(), vecParam, max_verify<float>);
+    test_error(error, "MaxTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_binary_fn<double>(device, context, queue, num_elems,
+                                       fnName.c_str(), vecParam,
+                                       max_verify<double>);
+        test_error(error, "MaxTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+cl_int MinTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error = test_binary_fn<float>(device, context, queue, num_elems,
+                                  fnName.c_str(), vecParam, min_verify<float>);
+    test_error(error, "MinTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_binary_fn<double>(device, context, queue, num_elems,
+                                       fnName.c_str(), vecParam,
+                                       min_verify<double>);
+        test_error(error, "MinTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+int test_min(cl_device_id device, cl_context context, cl_command_queue queue,
+             int n_elems)
+{
+    return MakeAndRunTest<MinTest>(device, context, queue, n_elems, "min",
+                                   true);
+}
+
+int test_minf(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<MinTest>(device, context, queue, n_elems, "min",
+                                   false);
+}
+
+int test_fmin(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<MinTest>(device, context, queue, n_elems, "fmin",
+                                   true);
+}
+
+int test_fminf(cl_device_id device, cl_context context, cl_command_queue queue,
+               int n_elems)
+{
+    return MakeAndRunTest<MinTest>(device, context, queue, n_elems, "fmin",
+                                   false);
+}
+
+int test_max(cl_device_id device, cl_context context, cl_command_queue queue,
+             int n_elems)
+{
+    return MakeAndRunTest<MaxTest>(device, context, queue, n_elems, "max",
+                                   true);
+}
+
+int test_maxf(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<MaxTest>(device, context, queue, n_elems, "max",
+                                   false);
+}
+
+int test_fmax(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<MaxTest>(device, context, queue, n_elems, "fmax",
+                                   true);
+}
+
+int test_fmaxf(cl_device_id device, cl_context context, cl_command_queue queue,
+               int n_elems)
+{
+    return MakeAndRunTest<MaxTest>(device, context, queue, n_elems, "fmax",
+                                   false);
+}

test_conformance/commonfns/test_clamp.cpp

@@ -13,23 +13,29 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
 
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <vector>
+
+#include "harness/deviceInfo.h"
+#include "harness/typeWrappers.h"
 
 #include "procs.h"
+#include "test_base.h"
+
 
 #ifndef M_PI
 #define M_PI 3.14159265358979323846264338327950288
 #endif
 
+
 #define CLAMP_KERNEL(type)                                                     \
-    const char *clamp_##type##_kernel_code =                \
-    EMIT_PRAGMA_DIRECTIVE                        \
-    "__kernel void test_clamp(__global " #type " *x, __global " #type " *minval, __global " #type " *maxval, __global " #type " *dst)\n" \
+    const char *clamp_##type##_kernel_code = EMIT_PRAGMA_DIRECTIVE             \
+        "__kernel void test_clamp(__global " #type " *x, __global " #type      \
+        " *minval, __global " #type " *maxval, __global " #type " *dst)\n"     \
         "{\n"                                                                  \
         "    int  tid = get_global_id(0);\n"                                   \
         "\n"                                                                   \
@@ -37,9 +43,10 @@
         "}\n";
 
 #define CLAMP_KERNEL_V(type, size)                                             \
-    const char *clamp_##type##size##_kernel_code =            \
-    EMIT_PRAGMA_DIRECTIVE                        \
-    "__kernel void test_clamp(__global " #type #size " *x, __global " #type #size " *minval, __global " #type #size " *maxval, __global " #type #size " *dst)\n" \
+    const char *clamp_##type##size##_kernel_code = EMIT_PRAGMA_DIRECTIVE       \
+        "__kernel void test_clamp(__global " #type #size                       \
+        " *x, __global " #type #size " *minval, __global " #type #size         \
+        " *maxval, __global " #type #size " *dst)\n"                           \
         "{\n"                                                                  \
         "    int  tid = get_global_id(0);\n"                                   \
         "\n"                                                                   \
@@ -47,15 +54,17 @@
         "}\n";
 
 #define CLAMP_KERNEL_V3(type, size)                                            \
-    const char *clamp_##type##size##_kernel_code =            \
-    EMIT_PRAGMA_DIRECTIVE                        \
-    "__kernel void test_clamp(__global " #type " *x, __global " #type " *minval, __global " #type " *maxval, __global " #type " *dst)\n" \
+    const char *clamp_##type##size##_kernel_code = EMIT_PRAGMA_DIRECTIVE       \
+        "__kernel void test_clamp(__global " #type " *x, __global " #type      \
+        " *minval, __global " #type " *maxval, __global " #type " *dst)\n"     \
         "{\n"                                                                  \
         "    int  tid = get_global_id(0);\n"                                   \
         "\n"                                                                   \
-    "    vstore3(clamp(vload3(tid, x), vload3(tid,minval), vload3(tid,maxval)), tid, dst);\n"    \
+        "    vstore3(clamp(vload3(tid, x), vload3(tid,minval), "               \
+        "vload3(tid,maxval)), tid, dst);\n"                                    \
         "}\n";
 
+
 #define EMIT_PRAGMA_DIRECTIVE " "
 CLAMP_KERNEL(float)
 CLAMP_KERNEL_V(float, 2)
@@ -74,242 +83,182 @@ CLAMP_KERNEL_V( double, 16 )
 CLAMP_KERNEL_V3(double, 3)
 #undef EMIT_PRAGMA_DIRECTIVE
 
-const char *clamp_float_codes[] = { clamp_float_kernel_code, clamp_float2_kernel_code, clamp_float4_kernel_code, clamp_float8_kernel_code, clamp_float16_kernel_code, clamp_float3_kernel_code };
-const char *clamp_double_codes[] = { clamp_double_kernel_code, clamp_double2_kernel_code, clamp_double4_kernel_code, clamp_double8_kernel_code, clamp_double16_kernel_code, clamp_double3_kernel_code };
+const char *clamp_float_codes[] = {
+    clamp_float_kernel_code,   clamp_float2_kernel_code,
+    clamp_float4_kernel_code,  clamp_float8_kernel_code,
+    clamp_float16_kernel_code, clamp_float3_kernel_code
+};
+const char *clamp_double_codes[] = {
+    clamp_double_kernel_code,   clamp_double2_kernel_code,
+    clamp_double4_kernel_code,  clamp_double8_kernel_code,
+    clamp_double16_kernel_code, clamp_double3_kernel_code
+};
 
-static int verify_clamp(float *x, float *minval, float *maxval, float *outptr, int n)
-{
-    float       t;
-    int         i;
+namespace {
 
-    for (i=0; i<n; i++)
+
+template <typename T>
+int verify_clamp(const T *const x, const T *const minval, const T *const maxval,
+                 const T *const outptr, int n)
 {
-        t = fminf( fmaxf( x[ i ], minval[ i ] ), maxval[ i ] );
+    T t;
+    for (int i = 0; i < n; i++)
+    {
+        t = std::min(std::max(x[i], minval[i]), maxval[i]);
         if (t != outptr[i])
         {
-            log_error( "%d) verification error: clamp( %a, %a, %a) = *%a vs. %a\n", i, x[i], minval[i], maxval[i], t, outptr[i] );
+            log_error(
+                "%d) verification error: clamp( %a, %a, %a) = *%a vs. %a\n", i,
+                x[i], minval[i], maxval[i], t, outptr[i]);
             return -1;
         }
     }
 
     return 0;
 }
-
-static int verify_clamp_double(double *x, double *minval, double *maxval, double *outptr, int n)
-{
-    double       t;
-    int         i;
-
-    for (i=0; i<n; i++)
-    {
-        t = fmin( fmax( x[ i ], minval[ i ] ), maxval[ i ] );
-        if (t != outptr[i])
-        {
-            log_error( "%d) verification error: clamp( %a, %a, %a) = *%a vs. %a\n", i, x[i], minval[i], maxval[i], t, outptr[i] );
-            return -1;
-        }
-    }
-
-    return 0;
-}
-
-int
-test_clamp(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem      streams[8];
-    cl_float      *input_ptr[3], *output_ptr;
-    cl_double     *input_ptr_double[3], *output_ptr_double = NULL;
-    cl_program  *program;
-    cl_kernel   *kernel;
-    size_t threads[1];
-    int num_elements;
-    int err;
-    int i, j;
-    MTdata d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount*2);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount*2);
-
-    num_elements = n_elems * (1 << (kVectorSizeCount-1));
-
-    int test_double = 0;
-    if(is_extension_available( device, "cl_khr_fp64" )) {
-    log_info("Testing doubles.\n");
-      test_double = 1;
 }
 
 
-    // why does this go from 0 to 2?? -- Oh, I see, there are four function
-    // arguments to the function, and 3 of them are inputs?
-    for( i = 0; i < 3; i++ )
+template <typename T>
+int test_clamp_fn(cl_device_id device, cl_context context,
+                  cl_command_queue queue, int n_elems)
 {
-        input_ptr[i] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-        if (test_double) input_ptr_double[i] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    }
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    if (test_double) output_ptr_double = (cl_double*)malloc(sizeof(cl_double) * num_elements);
+    clMemWrapper streams[4];
+    std::vector<T> input_ptr[3], output_ptr;
+
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+
+    int err, i, j;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
+
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
+
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
+
+    int num_elements = n_elems * (1 << (kVectorSizeCount - 1));
+
+    for (i = 0; i < 3; i++) input_ptr[i].resize(num_elements);
+    output_ptr.resize(num_elements);
 
-    // why does this go from 0 to 3?
     for (i = 0; i < 4; i++)
     {
-        streams[i] =
-            clCreateBuffer(context, CL_MEM_READ_WRITE,
-                           sizeof(cl_float) * num_elements, NULL, NULL);
-        if (!streams[0])
-        {
-            log_error("clCreateBuffer failed\n");
-            return -1;
-        }
-    }
-    if (test_double)
-    for( i = 4; i < 8; i++ )
-        {
-            streams[i] =
-                clCreateBuffer(context, CL_MEM_READ_WRITE,
-                               sizeof(cl_double) * num_elements, NULL, NULL);
-            if (!streams[0])
-            {
-            log_error("clCreateBuffer failed\n");
-            return -1;
-            }
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
+        test_error(err, "clCreateBuffer failed");
     }
 
-    d = init_genrand( gRandomSeed );
+    if (std::is_same<T, float>::value)
+    {
         for (j = 0; j < num_elements; j++)
         {
-        input_ptr[0][j] = get_random_float(-0x20000000, 0x20000000, d);
-        input_ptr[1][j] = get_random_float(-0x20000000, 0x20000000, d);
-        input_ptr[2][j] = get_random_float(input_ptr[1][j], 0x20000000, d);
-
-        if (test_double) {
-        input_ptr_double[0][j] = get_random_double(-0x20000000, 0x20000000, d);
-        input_ptr_double[1][j] = get_random_double(-0x20000000, 0x20000000, d);
-        input_ptr_double[2][j] = get_random_double(input_ptr_double[1][j], 0x20000000, d);
+            input_ptr[0][j] = get_random_float(-0x200000, 0x200000, d);
+            input_ptr[1][j] = get_random_float(-0x200000, 0x200000, d);
+            input_ptr[2][j] = get_random_float(input_ptr[1][j], 0x200000, d);
+        }
+    }
+    else if (std::is_same<T, double>::value)
+    {
+        for (j = 0; j < num_elements; j++)
+        {
+            input_ptr[0][j] = get_random_double(-0x20000000, 0x20000000, d);
+            input_ptr[1][j] = get_random_double(-0x20000000, 0x20000000, d);
+            input_ptr[2][j] = get_random_double(input_ptr[1][j], 0x20000000, d);
         }
     }
-    free_mtdata(d); d = NULL;
 
     for (i = 0; i < 3; i++)
     {
-        err = clEnqueueWriteBuffer( queue, streams[ i ], CL_TRUE, 0, sizeof( cl_float ) * num_elements, input_ptr[ i ], 0, NULL, NULL );
+        err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
+                                   sizeof(T) * num_elements,
+                                   &input_ptr[i].front(), 0, NULL, NULL);
         test_error(err, "Unable to write input buffer");
-
-        if (test_double) {
-        err = clEnqueueWriteBuffer( queue, streams[ 4 + i ], CL_TRUE, 0, sizeof( cl_double ) * num_elements, input_ptr_double[ i ], 0, NULL, NULL );
-        test_error( err, "Unable to write input buffer" );
-        }
     }
 
     for (i = 0; i < kTotalVecCount; i++)
     {
-        err = create_single_kernel_helper( context, &program[ i ], &kernel[ i ], 1, &clamp_float_codes[ i ], "test_clamp" );
-        test_error( err, "Unable to create kernel" );
-
-        log_info("Just made a program for float, i=%d, size=%d, in slot %d\n", i, g_arrVecSizes[i], i);
-        fflush(stdout);
-
-        if (test_double) {
-        err = create_single_kernel_helper( context, &program[ kTotalVecCount + i ], &kernel[ kTotalVecCount + i ], 1, &clamp_double_codes[ i ], "test_clamp" );
-        log_info("Just made a program for double, i=%d, size=%d, in slot %d\n", i, g_arrVecSizes[i], kTotalVecCount+i);
-        fflush(stdout);
-        test_error( err, "Unable to create kernel" );
-        }
-    }
-
-    for( i = 0; i < kTotalVecCount; i++ )
+        if (std::is_same<T, float>::value)
         {
+            err = create_single_kernel_helper(
+                context, &programs[i], &kernels[i], 1, &clamp_float_codes[i],
+                "test_clamp");
+            test_error(err, "Unable to create kernel");
+        }
+        else if (std::is_same<T, double>::value)
+        {
+            err = create_single_kernel_helper(
+                context, &programs[i], &kernels[i], 1, &clamp_double_codes[i],
+                "test_clamp");
+            test_error(err, "Unable to create kernel");
+        }
+
+        log_info("Just made a program for float, i=%d, size=%d, in slot %d\n",
+                 i, g_arrVecSizes[i], i);
+        fflush(stdout);
+
         for (j = 0; j < 4; j++)
         {
-            err = clSetKernelArg( kernel[ i ], j, sizeof( streams[ j ] ), &streams[ j ] );
+            err =
+                clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
             test_error(err, "Unable to set kernel argument");
         }
 
-        threads[0] = (size_t)n_elems;
+        size_t threads = (size_t)n_elems;
 
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
         test_error(err, "Unable to execute kernel");
 
-        err = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
+        err = clEnqueueReadBuffer(queue, streams[3], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
         test_error(err, "Unable to read results");
 
-        if (verify_clamp(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems*((g_arrVecSizes[i]))))
+        if (verify_clamp<T>((T *)&input_ptr[0].front(),
+                            (T *)&input_ptr[1].front(),
+                            (T *)&input_ptr[2].front(), (T *)&output_ptr[0],
+                            n_elems * ((g_arrVecSizes[i]))))
         {
-            log_error("CLAMP float%d test failed\n", ((g_arrVecSizes[i])));
+            log_error("CLAMP %s%d test failed\n", tname.c_str(),
+                      ((g_arrVecSizes[i])));
             err = -1;
         }
         else
         {
-            log_info("CLAMP float%d test passed\n", ((g_arrVecSizes[i])));
+            log_info("CLAMP %s%d test passed\n", tname.c_str(),
+                     ((g_arrVecSizes[i])));
             err = 0;
         }
 
-
-
-        if (err)
-        break;
-    }
-
-    // If the device supports double precision then test that
-    if (test_double)
-    {
-        for( ; i < 2*kTotalVecCount; i++ )
-        {
-
-            log_info("Start of test_double loop, i is %d\n", i);
-            for( j = 0; j < 4; j++ )
-            {
-                err = clSetKernelArg( kernel[i], j, sizeof( streams[j+4] ), &streams[j+4] );
-                test_error( err, "Unable to set kernel argument" );
-            }
-
-            threads[0] = (size_t)n_elems;
-
-            err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-            test_error( err, "Unable to execute kernel" );
-
-            err = clEnqueueReadBuffer( queue, streams[7], CL_TRUE, 0, sizeof(cl_double)*num_elements, (void *)output_ptr_double, 0, NULL, NULL );
-            test_error( err, "Unable to read results" );
-
-            if (verify_clamp_double(input_ptr_double[0], input_ptr_double[1], input_ptr_double[2], output_ptr_double, n_elems*g_arrVecSizes[(i-kTotalVecCount)]))
-            {
-                log_error("CLAMP double%d test failed\n", g_arrVecSizes[(i-kTotalVecCount)]);
-                err = -1;
-            }
-            else
-            {
-                log_info("CLAMP double%d test passed\n", g_arrVecSizes[(i-kTotalVecCount)]);
-                err = 0;
-            }
-
-            if (err)
-            break;
-        }
-    }
-
-
-    for( i = 0; i < ((test_double) ? 8 : 4); i++ )
-    {
-        clReleaseMemObject(streams[i]);
-    }
-    for (i=0; i < ((test_double) ? kTotalVecCount * 2-1 : kTotalVecCount); i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(input_ptr[2]);
-    free(output_ptr);
-    free(program);
-    free(kernel);
-    if (test_double) {
-        free(input_ptr_double[0]);
-        free(input_ptr_double[1]);
-        free(input_ptr_double[2]);
-        free(output_ptr_double);
+        if (err) break;
     }
 
     return err;
 }
 
 
+cl_int ClampTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error = test_clamp_fn<float>(device, context, queue, num_elems);
+    test_error(error, "ClampTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_clamp_fn<double>(device, context, queue, num_elems);
+        test_error(error, "ClampTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+int test_clamp(cl_device_id device, cl_context context, cl_command_queue queue,
+               int n_elems)
+{
+    return MakeAndRunTest<ClampTest>(device, context, queue, n_elems);
+}

test_conformance/commonfns/test_degrees.cpp

@@ -1,470 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-#ifndef M_PI
-#define M_PI    3.14159265358979323846264338327950288
-#endif
-
-static int test_degrees_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
-
-
-const char *degrees_kernel_code =
-"__kernel void test_degrees(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees2_kernel_code =
-"__kernel void test_degrees2(__global float2 *src, __global float2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees4_kernel_code =
-"__kernel void test_degrees4(__global float4 *src, __global float4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees8_kernel_code =
-"__kernel void test_degrees8(__global float8 *src, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees16_kernel_code =
-"__kernel void test_degrees16(__global float16 *src, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees3_kernel_code =
-"__kernel void test_degrees3(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(degrees(vload3(tid,src)),tid,dst);\n"
-"}\n";
-
-
-#define MAX_ERR  2.0f
-
-static int
-verify_degrees(float *inptr, float *outptr, int n)
-{
-    float error, max_error = 0.0f;
-    double   r, max_val = NAN;
-    int     i, j, max_index = 0;
-
-    for (i=0,j=0; i<n; i++,j++)
-    {
-        r = (180.0 / M_PI) * inptr[i];
-        error = Ulp_Error( outptr[i], r );
-        if( fabsf(error) > max_error)
-        {
-            max_error = error;
-            max_index = i;
-            max_val = r;
-            if( fabsf(error) > MAX_ERR)
-            {
-                log_error( "%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n", i, inptr[i], r, outptr[i], r, outptr[i], error );
-                return 1;
-            }
-        }
-    }
-
-    log_info( "degrees: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n", max_error, max_index, max_val, outptr[max_index], max_val, outptr[max_index] );
-
-    return 0;
-}
-
-int
-test_degrees(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[2];
-    cl_float     *input_ptr[1], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t threads[1];
-    int          num_elements;
-    int          err;
-    int          i;
-    MTdata        d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float((float)(-100000.f * M_PI), (float)(100000.f * M_PI) ,d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &degrees_kernel_code, "test_degrees" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &degrees2_kernel_code, "test_degrees2" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &degrees4_kernel_code, "test_degrees4" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &degrees8_kernel_code, "test_degrees8" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &degrees16_kernel_code, "test_degrees16" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &degrees3_kernel_code, "test_degrees3" );
-    if (err)
-        return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-
-        // Line below is troublesome...
-        threads[0] = (size_t)num_elements / ((g_arrVecSizes[i]));
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        cl_uint dead = 0xdeaddead;
-        memset_pattern4(output_ptr, &dead, sizeof(cl_float)*num_elements);
-        err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_degrees(input_ptr[0], output_ptr, n_elems*(i+1)))
-        {
-            log_error("DEGREES float%d test failed\n",((g_arrVecSizes[i])));
-            err = -1;
-        }
-        else
-        {
-            log_info("DEGREES float%d test passed\n", ((g_arrVecSizes[i])));
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    for (i=0; i < kTotalVecCount; i++) {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(output_ptr);
-
-    if( err )
-        return err;
-
-    if( ! is_extension_available( device, "cl_khr_fp64" ) )
-    {
-        log_info( "Skipping double -- cl_khr_fp64 is not supported by this device.\n" );
-        return 0;
-    }
-
-    return test_degrees_double( device, context, queue, n_elems);
-}
-
-#pragma mark -
-
-const char *degrees_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees2_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees2_double(__global double2 *src, __global double2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees4_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees4_double(__global double4 *src, __global double4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees8_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees8_double(__global double8 *src, __global double8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees16_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees16_double(__global double16 *src, __global double16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = degrees(src[tid]);\n"
-"}\n";
-
-const char *degrees3_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_degrees3_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(degrees(vload3(tid,src)),tid,dst);\n"
-"}\n";
-
-
-#define MAX_ERR  2.0f
-
-static int
-verify_degrees_double(double *inptr, double *outptr, int n)
-{
-    float error, max_error = 0.0f;
-    double   r, max_val = NAN;
-    int     i, j, max_index = 0;
-
-    for (i=0,j=0; i<n; i++,j++)
-    {
-        r = (180.0L / 3.14159265358979323846264338327950288L) * inptr[i];
-        error = Ulp_Error_Double( outptr[i], r );
-        if( fabsf(error) > max_error)
-        {
-            max_error = error;
-            max_index = i;
-            max_val = r;
-            if( fabsf(error) > MAX_ERR)
-            {
-                log_error( "%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n", i, inptr[i], r, outptr[i], r, outptr[i], error );
-                return 1;
-            }
-        }
-    }
-
-    log_info( "degreesd: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n", max_error, max_index, max_val, outptr[max_index], max_val, outptr[max_index] );
-
-    return 0;
-}
-
-static int
-test_degrees_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[2];
-    cl_double    *input_ptr[1], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t threads[1];
-    int          num_elements;
-    int          err;
-    int          i;
-    MTdata        d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    // TODO: line below is clearly wrong
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-        p[i] = get_random_double((-100000. * M_PI), (100000. * M_PI) ,d);
-
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &degrees_kernel_code_double, "test_degrees_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &degrees2_kernel_code_double, "test_degrees2_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &degrees4_kernel_code_double, "test_degrees4_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &degrees8_kernel_code_double, "test_degrees8_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &degrees16_kernel_code_double, "test_degrees16_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &degrees3_kernel_code_double, "test_degrees3_double" );
-    if (err)
-        return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-
-        // Line below is troublesome...
-        threads[0] = (size_t)num_elements / ((g_arrVecSizes[i]));
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        cl_uint dead = 0xdeaddead;
-        memset_pattern4(output_ptr, &dead, sizeof(cl_double)*num_elements);
-        err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_degrees_double(input_ptr[0], output_ptr, n_elems*(i+1)))
-        {
-            log_error("DEGREES double%d test failed\n",((g_arrVecSizes[i])));
-            err = -1;
-        }
-        else
-        {
-            log_info("DEGREES double%d test passed\n", ((g_arrVecSizes[i])));
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    for (i=0; i < kTotalVecCount; i++) {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(output_ptr);
-
-    return err;
-}
-
-
-

test_conformance/commonfns/test_fmax.cpp

@@ -1,233 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static const char *fmax_kernel_code =
-    "__kernel void test_fmax(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax2_kernel_code =
-    "__kernel void test_fmax2(__global float2 *srcA, __global float2 *srcB, __global float2 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax4_kernel_code =
-    "__kernel void test_fmax4(__global float4 *srcA, __global float4 *srcB, __global float4 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax8_kernel_code =
-    "__kernel void test_fmax8(__global float8 *srcA, __global float8 *srcB, __global float8 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax16_kernel_code =
-    "__kernel void test_fmax16(__global float16 *srcA, __global float16 *srcB, __global float16 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-
-static const char *fmax3_kernel_code =
-    "__kernel void test_fmax3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    vstore3(fmax(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
-    "}\n";
-
-static int
-verify_fmax(float *inptrA, float *inptrB, float *outptr, int n)
-{
-    float       r;
-    int         i;
-
-    for (i=0; i<n; i++)
-    {
-        r = (inptrA[i] >= inptrB[i]) ? inptrA[i] : inptrB[i];
-        if (r != outptr[i])
-        return -1;
-    }
-
-    return 0;
-}
-
-int
-test_fmax(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[3];
-    cl_float     *input_ptr[2], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    d = init_genrand( gRandomSeed );
-    p = input_ptr[0];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000, d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000,d );
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmax_kernel_code, "test_fmax" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmax2_kernel_code, "test_fmax2" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmax4_kernel_code, "test_fmax4" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmax8_kernel_code, "test_fmax8" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmax16_kernel_code, "test_fmax16" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmax3_kernel_code, "test_fmax3" );
-    if (err)
-    return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_fmax(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i]))))
-        {
-            log_error("FMAX float%d test failed\n", (g_arrVecSizes[i]));
-            err = -1;
-        }
-        else
-        {
-            log_info("FMAX float%d test passed\n", (g_arrVecSizes[i]));
-            err = 0;
-        }
-
-        if (err)
-        break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    return err;
-}
-
-

test_conformance/commonfns/test_fmaxf.cpp

@@ -1,244 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static const char *fmax_kernel_code =
-    "__kernel void test_fmax(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax2_kernel_code =
-    "__kernel void test_fmax2(__global float2 *srcA, __global float *srcB, __global float2 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax4_kernel_code =
-    "__kernel void test_fmax4(__global float4 *srcA, __global float *srcB, __global float4 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax8_kernel_code =
-    "__kernel void test_fmax8(__global float8 *srcA, __global float *srcB, __global float8 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax16_kernel_code =
-    "__kernel void test_fmax16(__global float16 *srcA, __global float *srcB, __global float16 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmax(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmax3_kernel_code =
-    "__kernel void test_fmax3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    vstore3(fmax(vload3(tid,srcA), srcB[tid]),tid,dst);\n"
-    "}\n";
-
-static int
-verify_fmax(float *inptrA, float *inptrB, float *outptr, int n, int veclen)
-{
-    float       r;
-    int         i, j;
-
-    for (i=0; i<n; ) {
-        int ii = i/veclen;
-        for (j=0; j<veclen && i<n; ++j, ++i) {
-            r = (inptrA[i] >= inptrB[ii]) ? inptrA[i] : inptrB[ii];
-            if (r != outptr[i]) {
-                log_info("Verify noted discrepancy at %d (of %d) (vec %d, pos %d)\n",
-                         i,n,ii,j);
-                log_info("SHould be %f, is %f\n", r, outptr[i]);
-                log_info("Taking max of (%f,%f)\n", inptrA[i], inptrB[i]);
-                return -1;
-            }
-        }
-    }
-
-    return 0;
-}
-
-int
-test_fmaxf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[3];
-    cl_float    *input_ptr[2], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-        {
-            log_error("clCreateBuffer failed\n");
-            return -1;
-        }
-        streams[1] =
-            clCreateBuffer(context, CL_MEM_READ_WRITE,
-                           sizeof(cl_float) * num_elements, NULL, NULL);
-        if (!streams[1])
-        {
-            log_error("clCreateBuffer failed\n");
-            return -1;
-        }
-        streams[2] =
-            clCreateBuffer(context, CL_MEM_READ_WRITE,
-                           sizeof(cl_float) * num_elements, NULL, NULL);
-        if (!streams[2])
-        {
-            log_error("clCreateBuffer failed\n");
-            return -1;
-        }
-
-    d = init_genrand( gRandomSeed );
-    p = input_ptr[0];
-    for (i=0; i<num_elements; i++)
-        {
-            p[i] = get_random_float(-0x20000000, 0x20000000, d);
-        }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-        {
-            p[i] = get_random_float(-0x20000000, 0x20000000, d);
-        }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements,
-                                (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-        {
-            log_error("clWriteArray failed\n");
-            return -1;
-        }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements,
-                                (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-        {
-            log_error("clWriteArray failed\n");
-            return -1;
-        }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmax_kernel_code, "test_fmax" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmax2_kernel_code, "test_fmax2" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmax4_kernel_code, "test_fmax4" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmax8_kernel_code, "test_fmax8" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmax16_kernel_code, "test_fmax16" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmax3_kernel_code, "test_fmax3" );
-    if (err)
-        return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-        {
-            err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-            err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-            err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-            if (err != CL_SUCCESS)
-                {
-                    log_error("clSetKernelArgs failed\n");
-                    return -1;
-                }
-        }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i < kTotalVecCount; i++)
-        {
-            err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-            if (err != CL_SUCCESS)
-                {
-                    log_error("clEnqueueNDRangeKernel failed\n");
-                    return -1;
-                }
-
-            err = clEnqueueReadBuffer(queue, streams[2], true, 0, sizeof(cl_float)*num_elements,
-                                      output_ptr, 0, NULL, NULL);
-            if (err != CL_SUCCESS)
-                {
-                    log_error("clEnqueueReadBuffer failed\n");
-                    return -1;
-                }
-
-            if (verify_fmax(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i])), (g_arrVecSizes[i])))
-                {
-                    log_error("FMAX float%d,float test failed\n", (g_arrVecSizes[i]));
-                    err = -1;
-                }
-            else
-                {
-                    log_info("FMAX float%d,float test passed\n", (g_arrVecSizes[i]));
-                    err = 0;
-                }
-
-            if (err)
-                break;
-        }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i < kTotalVecCount; i++)
-        {
-            clReleaseKernel(kernel[i]);
-            clReleaseProgram(program[i]);
-        }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    return err;
-}
-
-

test_conformance/commonfns/test_fmin.cpp

@@ -1,238 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static const char *fmin_kernel_code =
-    "__kernel void test_fmin(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin2_kernel_code =
-    "__kernel void test_fmin2(__global float2 *srcA, __global float2 *srcB, __global float2 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin4_kernel_code =
-    "__kernel void test_fmin4(__global float4 *srcA, __global float4 *srcB, __global float4 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin8_kernel_code =
-    "__kernel void test_fmin8(__global float8 *srcA, __global float8 *srcB, __global float8 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin16_kernel_code =
-    "__kernel void test_fmin16(__global float16 *srcA, __global float16 *srcB, __global float16 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-
-static const char *fmin3_kernel_code =
-    "__kernel void test_fmin3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    vstore3(fmin(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
-    "}\n";
-
-int
-verify_fmin(float *inptrA, float *inptrB, float *outptr, int n)
-{
-    float       r;
-    int         i;
-
-    for (i=0; i<n; i++)
-    {
-        r = (inptrA[i] > inptrB[i]) ? inptrB[i] : inptrA[i];
-        if (r != outptr[i])
-        return -1;
-    }
-
-    return 0;
-}
-
-int
-test_fmin(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[3];
-    cl_float    *input_ptr[2], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));;
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    d = init_genrand( gRandomSeed );
-    p = input_ptr[0];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000, d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000, d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements,
-                (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements,
-                (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmin_kernel_code, "test_fmin" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmin2_kernel_code, "test_fmin2" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmin4_kernel_code, "test_fmin4" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmin8_kernel_code, "test_fmin8" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmin16_kernel_code, "test_fmin16" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmin3_kernel_code, "test_fmin3" );
-    if (err)
-    return -1;
-
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_fmin(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i]))))
-        {
-            log_error("FMIN float%d test failed\n", (g_arrVecSizes[i]));
-            err = -1;
-        }
-        else
-        {
-            log_info("FMIN float%d test passed\n", (g_arrVecSizes[i]));
-            err = 0;
-        }
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    return err;
-}
-
-

test_conformance/commonfns/test_fminf.cpp

@@ -1,236 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static const char *fmin_kernel_code =
-    "__kernel void test_fmin(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin2_kernel_code =
-    "__kernel void test_fmin2(__global float2 *srcA, __global float *srcB, __global float2 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin4_kernel_code =
-    "__kernel void test_fmin4(__global float4 *srcA, __global float *srcB, __global float4 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin8_kernel_code =
-    "__kernel void test_fmin8(__global float8 *srcA, __global float *srcB, __global float8 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin16_kernel_code =
-    "__kernel void test_fmin16(__global float16 *srcA, __global float *srcB, __global float16 *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    dst[tid] = fmin(srcA[tid], srcB[tid]);\n"
-    "}\n";
-
-static const char *fmin3_kernel_code =
-    "__kernel void test_fmin3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-    "{\n"
-    "    int  tid = get_global_id(0);\n"
-    "    vstore3(fmin(vload3(tid,srcA), srcB[tid]),tid,dst);\n"
-    "}\n";
-
-static int
-verify_fmin(float *inptrA, float *inptrB, float *outptr, int n, int veclen)
-{
-    float       r;
-    int         i, j;
-
-    for (i=0; i<n; ) {
-    int ii = i/veclen;
-    for (j=0; j<veclen && i<n; ++j, ++i) {
-        r = (inptrA[i] > inptrB[ii]) ? inptrB[ii] : inptrA[i];
-        if (r != outptr[i])
-        return -1;
-    }
-    }
-
-    return 0;
-}
-
-int
-test_fminf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[3];
-    cl_float     *input_ptr[2], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata      d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    d = init_genrand( gRandomSeed );
-    p = input_ptr[0];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000, d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x20000000, 0x20000000, d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements,
-                (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements,
-                (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &fmin_kernel_code, "test_fmin" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &fmin2_kernel_code, "test_fmin2" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &fmin4_kernel_code, "test_fmin4" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &fmin8_kernel_code, "test_fmin8" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &fmin16_kernel_code, "test_fmin16" );
-    if (err)
-    return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &fmin3_kernel_code, "test_fmin3" );
-    if (err)
-    return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_fmin(input_ptr[0], input_ptr[1], output_ptr, n_elems*((g_arrVecSizes[i])), (g_arrVecSizes[i])))
-        {
-            log_error("fmin float%d,float test failed\n", (g_arrVecSizes[i]));
-            err = -1;
-        }
-        else
-        {
-            log_info("fmin float%d,float test passed\n", (g_arrVecSizes[i]));
-            err = 0;
-        }
-
-        if (err)
-        break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    return err;
-}
-
-

test_conformance/commonfns/test_max.cpp

@@ -1,60 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static int max_verify_float( float *x, float *y, float *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements * vecSize; i++ )
-    {
-        float v = ( x[ i ] < y[ i ] ) ? y[ i ] : x[ i ];
-        if( v != out[ i ] )
-        {
-            log_error("x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is vector %d, element %d, for vector size %d)\n",
-                i, x[i], i, y[i], i, out[i], v, i, i/vecSize, i%vecSize, vecSize);
-            return -1;
-        }
-    }
-    return 0;
-}
-
-static int max_verify_double( double *x, double *y, double *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements * vecSize; i++ )
-    {
-        double v = ( x[ i ] < y[ i ] ) ? y[ i ] : x[ i ];
-        if( v != out[ i ] )
-        {
-            log_error("x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is vector %d, element %d, for vector size %d)\n",
-                i, x[i], i, y[i], i, out[i], v, i, i/vecSize, i%vecSize, vecSize);
-            return -1;
-        }
-    }
-    return 0;
-}
-
-int test_max(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    return test_binary_fn( device, context, queue, n_elems, "max", true, max_verify_float, max_verify_double );
-}
-
-

test_conformance/commonfns/test_maxf.cpp

@@ -1,64 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static int max_verify_float( float *x, float *y, float *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements; i++ )
-    {
-        for( int j = 0; j < vecSize; j++ )
-        {
-            float v = ( x[ i * vecSize + j ] < y[ i ] ) ? y[ i ] : x[ i * vecSize + j ];
-            if( v != out[ i * vecSize + j ] )
-            {
-                log_error( "Failure for vector size %d at position %d, element %d:\n\t max(%a, %a) = *%a vs %a\n", vecSize, i, j, x[ i * vecSize + j ], y[i], v,  out[ i * vecSize + j ] );
-                return -1;
-            }
-        }
-    }
-    return 0;
-}
-
-static int max_verify_double( double *x, double *y, double *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements; i++ )
-    {
-        for( int j = 0; j < vecSize; j++ )
-        {
-            double v = ( x[ i * vecSize + j ] < y[ i ] ) ? y[ i ] : x[ i * vecSize + j ];
-            if(    v != out[ i * vecSize + j ] )
-            {
-                log_error( "Failure for vector size %d at position %d, element %d:\n\t max(%a, %a) = *%a vs %a\n", vecSize, i, j, x[ i * vecSize + j ], y[i], v,  out[ i * vecSize + j ] );
-                return -1;
-            }
-        }
-    }
-    return 0;
-}
-
-int test_maxf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    return test_binary_fn( device, context, queue, n_elems, "max", false, max_verify_float, max_verify_double );
-}
-
-

test_conformance/commonfns/test_min.cpp

@@ -1,56 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static int min_verify_float( float *x, float *y, float *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements * vecSize; i++ )
-    {
-        float v = ( y[ i ] < x[ i ] ) ? y[ i ] : x[ i ];
-        if( v != out[ i ] ) {
-      log_error("x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is vector %d, element %d, for vector size %d)\n", i, x[i], i, y[i], i, out[i], v, i, i/vecSize, i%vecSize, vecSize);
-            return -1;
-    }
-    }
-    return 0;
-}
-
-static int min_verify_double( double *x, double *y, double *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements * vecSize; i++ )
-    {
-        double v = ( y[ i ] < x[ i ] ) ? y[ i ] : x[ i ];
-        if( v != out[ i ] ) {
-      log_error("x[%d]=%g y[%d]=%g out[%d]=%g, expected %g. (index %d is vector %d, element %d, for vector size %d)\n", i, x[i], i, y[i], i, out[i], v, i, i/vecSize, i%vecSize, vecSize);
-            return -1;
-    }
-    }
-    return 0;
-}
-
-int test_min(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    return test_binary_fn( device, context, queue, n_elems, "min", true, min_verify_float, min_verify_double );
-}
-
-

test_conformance/commonfns/test_minf.cpp

@@ -1,70 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-#include "harness/errorHelpers.h"
-
-static int min_verify_float( float *x, float *y, float *out, int numElements, int vecSize )
-{
-    for( int i = 0; i < numElements; i++ )
-    {
-        for( int j = 0; j < vecSize; j++ )
-        {
-            float v = ( y[ i ] < x[ i * vecSize + j ] ) ? y[ i ] : x[ i * vecSize + j ];
-            if( v != out[ i * vecSize + j ] )
-            {
-                log_error( "Failure for vector size %d at position %d, element %d:\n\t min(%a, %a) = *%a vs %a\n", vecSize, i, j, x[ i * vecSize + j ], y[i], v,  out[ i * vecSize + j ] );
-                return -1;
-            }
-        }
-    }
-    return 0;
-}
-
-static int min_verify_double( double *x, double *y, double *out, int numElements, int vecSize )
-{
-    int maxFail = 1;
-    int numFails = 0;
-    for( int i = 0; i < numElements; i++ )
-    {
-        for( int j = 0; j < vecSize; j++ )
-        {
-            double v = ( y[ i ] < x[ i * vecSize + j ] ) ? y[ i ] : x[ i * vecSize + j ];
-            if(    v != out[ i * vecSize + j ] )
-            {
-                log_error( "Failure for vector size %d at position %d, element %d:\n\t min(%a, %a) = *%a vs %a\n", vecSize, i, j, x[ i * vecSize + j ], y[i], v,  out[ i * vecSize + j ] );
-                ++numFails;
-                if(numFails >= maxFail) {
-                return -1;
-            }
-        }
-    }
-    }
-    return 0;
-}
-
-int test_minf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    return test_binary_fn( device, context, queue, n_elems, "min", false, min_verify_float, min_verify_double );
-}
-
-

test_conformance/commonfns/test_mix.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -13,179 +13,265 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
-
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 
 #include "procs.h"
+#include "test_base.h"
 
-const char *mix_kernel_code =
-"__kernel void test_mix(__global float *srcA, __global float *srcB, __global float *srcC, __global float *dst)\n"
+
+const char *mix_fn_code_pattern =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s%s *x, __global %s%s *y, __global %s%s "
+    "*a, __global %s%s *dst)\n"
+    "{\n"
+    "    int  tid = get_global_id(0);\n"
+    "    dst[tid] = mix(x[tid], y[tid], a[tid]);\n"
+    "}\n";
+
+const char *mix_fn_code_pattern_v3 =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *x, __global %s *y, __global %s *a, "
+    "__global %s *dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
     "\n"
-"    dst[tid] = mix(srcA[tid], srcB[tid], srcC[tid]);\n"
+    "    vstore3(mix(vload3(tid, x), vload3(tid, y), vload3(tid, a)), tid, "
+    "dst);\n"
     "}\n";
 
+const char *mix_fn_code_pattern_v3_scalar =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *x, __global %s *y, __global %s *a, "
+    "__global %s *dst)\n"
+    "{\n"
+    "    int  tid = get_global_id(0);\n"
+    "\n"
+    "    vstore3(mix(vload3(tid, x), vload3(tid, y), a[tid]), tid, dst);\n"
+    "}\n";
+
+
 #define MAX_ERR 1e-3
 
-float
-verify_mix(float *inptrA, float *inptrB, float *inptrC, float *outptr, int n)
+namespace {
+
+
+template <typename T>
+int verify_mix(const T *const inptrX, const T *const inptrY,
+               const T *const inptrA, const T *const outptr, const int n,
+               const int veclen, const bool vecParam)
 {
-    float       r, delta, max_err = 0.0f;
+    T r;
+    float delta = 0.0f;
     int i;
 
-    for (i=0; i<n; i++)
+    if (vecParam)
     {
-        r = inptrA[i] + ((inptrB[i] - inptrA[i]) * inptrC[i]);
-        delta = fabsf(r - outptr[i]) / r;
-        if(delta > max_err) max_err = delta;
-    }
-    return max_err;
-}
-
-int
-test_mix(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
+        for (i = 0; i < n * veclen; i++)
         {
-    cl_mem            streams[4];
-    cl_float        *input_ptr[3], *output_ptr, *p;
-    cl_program        program;
-    cl_kernel        kernel;
-    size_t    threads[1];
-    float            max_err;
-    int                err;
-    int                i;
-    MTdata          d;
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[2] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
+            r = inptrX[i] + ((inptrY[i] - inptrX[i]) * inptrA[i]);
+            delta = fabs(double(r - outptr[i])) / r;
+            if (delta > MAX_ERR)
             {
-        log_error("clCreateBuffer failed\n");
+                log_error(
+                    "%d) verification error: mix(%a, %a, %a) = *%a vs. %a\n", i,
+                    inptrX[i], inptrY[i], inptrA[i], r, outptr[i]);
                 return -1;
             }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
+        }
+    }
+    else
     {
-        log_error("clCreateBuffer failed\n");
+        for (int i = 0; i < n; ++i)
+        {
+            int ii = i / veclen;
+            int vi = i * veclen;
+            for (int j = 0; j < veclen; ++j, ++vi)
+            {
+                r = inptrX[vi] + ((inptrY[vi] - inptrX[vi]) * inptrA[i]);
+                delta = fabs(double(r - outptr[vi])) / r;
+                if (delta > MAX_ERR)
+                {
+                    log_error("{%d, element %d}) verification error: mix(%a, "
+                              "%a, %a) = *%a vs. %a\n",
+                              ii, j, inptrX[vi], inptrY[vi], inptrA[i], r,
+                              outptr[vi]);
                     return -1;
                 }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
+            }
+        }
     }
 
-    streams[3] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[3])
+    return 0;
+}
+} // namespace
+
+
+template <typename T>
+int test_mix_fn(cl_device_id device, cl_context context, cl_command_queue queue,
+                int n_elems, bool vecParam)
 {
-        log_error("clCreateBuffer failed\n");
-        return -1;
+    clMemWrapper streams[4];
+    std::vector<T> input_ptr[3], output_ptr;
+
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+
+    int err, i;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
+
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
+
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
+
+    int num_elements = n_elems * (1 << (kTotalVecCount - 1));
+
+
+    for (i = 0; i < 3; i++) input_ptr[i].resize(num_elements);
+    output_ptr.resize(num_elements);
+
+    for (i = 0; i < 4; i++)
+    {
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
+        test_error(err, "clCreateBuffer failed");
     }
 
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
     for (i = 0; i < num_elements; i++)
     {
-        p[i] =  (float) genrand_real1(d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = (float) genrand_real1(d);
-    }
-    p = input_ptr[2];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = (float) genrand_real1(d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[2], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
+        input_ptr[0][i] = (T)genrand_real1(d);
+        input_ptr[1][i] = (T)genrand_real1(d);
+        input_ptr[2][i] = (T)genrand_real1(d);
     }
 
-    err = create_single_kernel_helper( context, &program, &kernel, 1, &mix_kernel_code, "test_mix" );
-    test_error( err, "Unable to create test kernel" );
-
-  err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0] );
-  err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1] );
-  err |= clSetKernelArg(kernel, 2, sizeof streams[2], &streams[2] );
-  err |= clSetKernelArg(kernel, 3, sizeof streams[3], &streams[3] );
-    if (err != CL_SUCCESS)
+    std::string pragma_str;
+    if (std::is_same<T, double>::value)
     {
-        log_error("clSetKernelArgs failed\n");
-        return -1;
+        pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
     }
 
-    threads[0] = (size_t)num_elements;
-    err = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
+    for (i = 0; i < 3; i++)
     {
-        log_error("clEnqueueNDRangeKernel failed\n");
-        return -1;
+        err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
+                                   sizeof(T) * num_elements,
+                                   &input_ptr[i].front(), 0, NULL, NULL);
+        test_error(err, "Unable to write input buffer");
     }
 
-    err = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
+    char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
+
+    for (i = 0; i < kTotalVecCount; i++)
     {
-        log_error("clEnqueueReadBuffer failed\n");
-        return -1;
+        std::string kernelSource;
+        if (i >= kVectorSizeCount)
+        {
+            if (vecParam)
+            {
+                std::string str = mix_fn_code_pattern_v3;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), tname.c_str());
+            }
+            else
+            {
+                std::string str = mix_fn_code_pattern_v3_scalar;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), tname.c_str());
+            }
+        }
+        else
+        {
+            // regular path
+            std::string str = mix_fn_code_pattern;
+            kernelSource =
+                string_format(str, pragma_str.c_str(), tname.c_str(),
+                              vecSizeNames[i], tname.c_str(), vecSizeNames[i],
+                              tname.c_str(), vecParam ? vecSizeNames[i] : "",
+                              tname.c_str(), vecSizeNames[i]);
+        }
+        const char *programPtr = kernelSource.c_str();
+        err =
+            create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
+                                        (const char **)&programPtr, "test_fn");
+        test_error(err, "Unable to create kernel");
+
+        for (int j = 0; j < 4; j++)
+        {
+            err =
+                clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
+            test_error(err, "Unable to set kernel argument");
         }
 
-    max_err = verify_mix(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, num_elements);
-    if (max_err > MAX_ERR)
+        size_t threads = (size_t)n_elems;
+
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
+        test_error(err, "Unable to execute kernel");
+
+        err = clEnqueueReadBuffer(queue, streams[3], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
+        test_error(err, "Unable to read results");
+
+        if (verify_mix(&input_ptr[0].front(), &input_ptr[1].front(),
+                       &input_ptr[2].front(), &output_ptr.front(), n_elems,
+                       g_arrVecSizes[i], vecParam))
         {
-        log_error("MIX test failed %g max err\n", max_err);
+            log_error("mix %s%d%s test failed\n", tname.c_str(),
+                      ((g_arrVecSizes[i])),
+                      vecParam ? "" : std::string(", " + tname).c_str());
             err = -1;
         }
         else
         {
-        log_info("MIX test passed %g max err\n", max_err);
+            log_info("mix %s%d%s test passed\n", tname.c_str(),
+                     ((g_arrVecSizes[i])),
+                     vecParam ? "" : std::string(", " + tname).c_str());
             err = 0;
         }
 
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    clReleaseMemObject(streams[3]);
-    clReleaseKernel(kernel);
-    clReleaseProgram(program);
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(input_ptr[2]);
-    free(output_ptr);
+        if (err) break;
+    }
 
     return err;
 }
 
 
+cl_int MixTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error = test_mix_fn<float>(device, context, queue, num_elems, vecParam);
+    test_error(error, "MixTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error =
+            test_mix_fn<double>(device, context, queue, num_elems, vecParam);
+        test_error(error, "MixTest::Run<double> failed");
+    }
+
+    return error;
+}
 
 
+int test_mix(cl_device_id device, cl_context context, cl_command_queue queue,
+             int n_elems)
+{
+    return MakeAndRunTest<MixTest>(device, context, queue, n_elems, "mix",
+                                   true);
+}
 
+
+int test_mixf(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<MixTest>(device, context, queue, n_elems, "mix",
+                                   false);
+}

test_conformance/commonfns/test_radians.cpp

@@ -1,468 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-#ifndef M_PI
-#define M_PI    3.14159265358979323846264338327950288
-#endif
-
-static int test_radians_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
-
-
-const char *radians_kernel_code =
-"__kernel void test_radians(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians2_kernel_code =
-"__kernel void test_radians2(__global float2 *src, __global float2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians4_kernel_code =
-"__kernel void test_radians4(__global float4 *src, __global float4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians8_kernel_code =
-"__kernel void test_radians8(__global float8 *src, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians16_kernel_code =
-"__kernel void test_radians16(__global float16 *src, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians3_kernel_code =
-"__kernel void test_radians3(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(radians(vload3(tid,src)),tid,dst);\n"
-"}\n";
-
-
-#define MAX_ERR  2.0f
-
-static float
-verify_radians(float *inptr, float *outptr, int n)
-{
-    float error, max_error = 0.0f;
-    double   r, max_val = NAN;
-    int     i, j, max_index = 0;
-
-    for (i=0,j=0; i<n; i++,j++)
-    {
-        r = (M_PI / 180.0) * inptr[i];
-        error = Ulp_Error( outptr[i], r );
-        if( fabsf(error) > max_error)
-        {
-            max_error = error;
-            max_index = i;
-            max_val = r;
-            if( fabsf(error) > MAX_ERR)
-            {
-                log_error( "%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n", i, inptr[i], r, outptr[i], r, outptr[i], error );
-                return 1;
-            }
-        }
-    }
-
-    log_info( "radians: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n", max_error, max_index, max_val, outptr[max_index], max_val, outptr[max_index] );
-
-    return 0;
-}
-
-
-int
-test_radians(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[2];
-    cl_float     *input_ptr[1], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t       threads[1];
-    int          num_elements;
-    int          err;
-    int          i;
-    MTdata       d;
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float((float)(-100000.f * M_PI), (float)(100000.f * M_PI) ,d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &radians_kernel_code, "test_radians" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &radians2_kernel_code, "test_radians2" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &radians4_kernel_code, "test_radians4" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &radians8_kernel_code, "test_radians8" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &radians16_kernel_code, "test_radians16" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &radians3_kernel_code, "test_radians3" );
-    if (err)
-        return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        threads[0] = (size_t)num_elements / ((g_arrVecSizes[i]));
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        cl_uint dead = 0xdeaddead;
-        memset_pattern4(output_ptr, &dead, sizeof(cl_float)*num_elements);
-        err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_radians(input_ptr[0], output_ptr, n_elems*(i+1)))
-        {
-            log_error("RADIANS float%d test failed\n",((g_arrVecSizes[i])));
-            err = -1;
-        }
-        else
-        {
-            log_info("RADIANS float%d test passed\n", ((g_arrVecSizes[i])));
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    for (i=0; i < kTotalVecCount; i++) {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(output_ptr);
-    if( err )
-        return err;
-
-    if( ! is_extension_available( device, "cl_khr_fp64" ) )
-    {
-        log_info( "Skipping double -- cl_khr_fp64 is not supported by this device.\n" );
-        return 0;
-    }
-
-    return test_radians_double( device,  context,  queue,  n_elems);
-}
-
-
-
-#pragma mark -
-
-const char *radians_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians2_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians2_double(__global double2 *src, __global double2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians4_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians4_double(__global double4 *src, __global double4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians8_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians8_double(__global double8 *src, __global double8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians16_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians16_double(__global double16 *src, __global double16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = radians(src[tid]);\n"
-"}\n";
-
-const char *radians3_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_radians3_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(radians(vload3(tid,src)),tid,dst);\n"
-"}\n";
-
-
-#define MAX_ERR  2.0f
-
-static double
-verify_radians_double(double *inptr, double *outptr, int n)
-{
-    float error, max_error = 0.0f;
-    double   r, max_val = NAN;
-    int     i, j, max_index = 0;
-
-    for (i=0,j=0; i<n; i++,j++)
-    {
-        r = (3.14159265358979323846264338327950288L / 180.0L) * inptr[i];
-        error = Ulp_Error_Double( outptr[i], r );
-        if( fabsf(error) > max_error)
-        {
-            max_error = error;
-            max_index = i;
-            max_val = r;
-            if( fabsf(error) > MAX_ERR)
-            {
-                log_error( "%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n", i, inptr[i], r, outptr[i], r, outptr[i], error );
-                return 1;
-            }
-        }
-    }
-
-    log_info( "radiansd: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n", max_error, max_index, max_val, outptr[max_index], max_val, outptr[max_index] );
-
-    return 0;
-}
-
-
-int
-test_radians_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem       streams[2];
-    cl_double     *input_ptr[1], *output_ptr, *p;
-    cl_program   *program;
-    cl_kernel    *kernel;
-    size_t       threads[1];
-    int          num_elements;
-    int          err;
-    int          i;
-    MTdata       d;
-
-
-    program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount);
-    kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount);
-
-    //TODO: line below is clearly wrong
-    num_elements = n_elems * (1 << (kTotalVecCount-1));
-
-    input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-        p[i] = get_random_double((float)(-100000.0 * M_PI), (float)(100000.0 * M_PI) ,d);
-
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &radians_kernel_code_double, "test_radians_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &radians2_kernel_code_double, "test_radians2_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &radians4_kernel_code_double, "test_radians4_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &radians8_kernel_code_double, "test_radians8_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &radians16_kernel_code_double, "test_radians16_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &radians3_kernel_code_double, "test_radians3_double" );
-    if (err)
-        return -1;
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    for (i=0; i < kTotalVecCount; i++)
-    {
-        threads[0] = (size_t)num_elements / ((g_arrVecSizes[i]));
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        cl_uint dead = 0xdeaddead;
-        memset_pattern4(output_ptr, &dead, sizeof(cl_double)*num_elements);
-        err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        if (verify_radians_double(input_ptr[0], output_ptr, n_elems*(i+1)))
-        {
-            log_error("RADIANS double%d test failed\n",((g_arrVecSizes[i])));
-            err = -1;
-        }
-        else
-        {
-            log_info("RADIANS double%d test passed\n", ((g_arrVecSizes[i])));
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    for (i=0; i < kTotalVecCount; i++) {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(program);
-    free(kernel);
-    free(input_ptr[0]);
-    free(output_ptr);
-
-    return err;
-}
-

test_conformance/commonfns/test_sign.cpp

@@ -1,437 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static int
-test_sign_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
-
-
-const char *sign_kernel_code =
-"__kernel void test_sign(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign2_kernel_code =
-"__kernel void test_sign2(__global float2 *src, __global float2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign4_kernel_code =
-"__kernel void test_sign4(__global float4 *src, __global float4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign8_kernel_code =
-"__kernel void test_sign8(__global float8 *src, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign16_kernel_code =
-"__kernel void test_sign16(__global float16 *src, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign3_kernel_code =
-"__kernel void test_sign3(__global float *src, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(sign(vload3(tid,src)), tid, dst);\n"
-"}\n";
-
-
-
-static int
-verify_sign(float *inptr, float *outptr, int n)
-{
-  float       r;
-  int         i;
-
-  for (i=0; i<n; i++)
-  {
-    if (inptr[i] > 0.0f)
-      r = 1.0f;
-    else if (inptr[i] < 0.0f)
-      r = -1.0f;
-    else
-      r = 0.0f;
-    if (r != outptr[i])
-      return -1;
-  }
-
-  return 0;
-}
-
-static const char *fn_names[] = { "SIGN float", "SIGN float2", "SIGN float4", "SIGN float8", "SIGN float16", "SIGN float3" };
-
-int
-test_sign(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-  cl_mem      streams[2];
-  cl_float    *input_ptr[1], *output_ptr, *p;
-  cl_program  program[kTotalVecCount];
-  cl_kernel   kernel[kTotalVecCount];
-  size_t  threads[1];
-  int num_elements;
-  int err;
-  int i;
-  MTdata    d;
-
-  num_elements = n_elems * 16;
-
-  input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[0])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[1])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  d = init_genrand( gRandomSeed );
-  p = input_ptr[0];
-  for (i=0; i<num_elements; i++)
-  {
-    p[i] = get_random_float(-0x20000000, 0x20000000, d);
-  }
-  free_mtdata(d);   d = NULL;
-
-
-  err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-
-  err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &sign_kernel_code, "test_sign" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &sign2_kernel_code, "test_sign2" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &sign4_kernel_code, "test_sign4" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &sign8_kernel_code, "test_sign8" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &sign16_kernel_code, "test_sign16" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &sign3_kernel_code, "test_sign3" );
-  if (err)
-    return -1;
-
-  for (i=0; i<kTotalVecCount; i++)
-  {
-      err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-      err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-      if (err != CL_SUCCESS)
-    {
-      log_error("clSetKernelArgs failed\n");
-      return -1;
-    }
-  }
-
-  threads[0] = (size_t)n_elems;
-  for (i=0; i<kTotalVecCount; i++) // change this so we test all
-  {
-    err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueNDRangeKernel failed\n");
-      return -1;
-    }
-
-    err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueReadBuffer failed\n");
-      return -1;
-    }
-
-    if (verify_sign(input_ptr[0], output_ptr, n_elems*(i+1)))
-    {
-      log_error("%s test failed\n", fn_names[i]);
-      err = -1;
-    }
-    else
-    {
-      log_info("%s test passed\n", fn_names[i]);
-      err = 0;
-    }
-
-    if (err)
-      break;
-  }
-
-  clReleaseMemObject(streams[0]);
-  clReleaseMemObject(streams[1]);
-  for (i=0; i<kTotalVecCount; i++)
-  {
-    clReleaseKernel(kernel[i]);
-    clReleaseProgram(program[i]);
-  }
-  free(input_ptr[0]);
-  free(output_ptr);
-
-  if (err) return err;
-
-  if (!is_extension_available(device, "cl_khr_fp64"))
-  {
-      log_info("skipping double test -- cl_khr_fp64 not supported.\n");
-      return 0;
-  }
-
-    return test_sign_double( device, context, queue, n_elems);
-}
-
-#pragma mark -
-
-const char *sign_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign2_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign2_double(__global double2 *src, __global double2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign4_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign4_double(__global double4 *src, __global double4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign8_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign8_double(__global double8 *src, __global double8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign16_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign16_double(__global double16 *src, __global double16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = sign(src[tid]);\n"
-"}\n";
-
-const char *sign3_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_sign3_double(__global double *src, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(sign(vload3(tid,src)), tid, dst);\n"
-"}\n";
-
-
-static int
-verify_sign_double(double *inptr, double *outptr, int n)
-{
-  double       r;
-  int         i;
-
-  for (i=0; i<n; i++)
-  {
-    if (inptr[i] > 0.0)
-      r = 1.0;
-    else if (inptr[i] < 0.0)
-      r = -1.0;
-    else
-      r = 0.0f;
-    if (r != outptr[i])
-      return -1;
-  }
-
-  return 0;
-}
-
-static const char *fn_names_double[] = { "SIGN double", "SIGN double2", "SIGN double4", "SIGN double8", "SIGN double16", "SIGN double3" };
-
-int
-test_sign_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-  cl_mem      streams[2];
-  cl_double    *input_ptr[1], *output_ptr, *p;
-  cl_program  program[kTotalVecCount];
-  cl_kernel   kernel[kTotalVecCount];
-  size_t  threads[1];
-  int num_elements;
-  int err;
-  int i;
-  MTdata    d;
-
-  num_elements = n_elems * 16;
-
-  input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-  output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-  streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_double) * num_elements, NULL, NULL);
-  if (!streams[0])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_double) * num_elements, NULL, NULL);
-  if (!streams[1])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  d = init_genrand( gRandomSeed );
-  p = input_ptr[0];
-  for (i=0; i<num_elements; i++)
-    p[i] = get_random_double(-0x20000000, 0x20000000, d);
-
-  free_mtdata(d);   d = NULL;
-
-
-  err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-
-  err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &sign_kernel_code_double, "test_sign_double" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &sign2_kernel_code_double, "test_sign2_double" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &sign4_kernel_code_double, "test_sign4_double" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &sign8_kernel_code_double, "test_sign8_double" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &sign16_kernel_code_double, "test_sign16_double" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &sign3_kernel_code_double, "test_sign3_double" );
-  if (err)
-    return -1;
-
-  for (i=0; i<kTotalVecCount; i++)
-  {
-      err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-      err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-      if (err != CL_SUCCESS)
-    {
-      log_error("clSetKernelArgs failed\n");
-      return -1;
-    }
-  }
-
-  threads[0] = (size_t)n_elems;
-  for (i=0; i<kTotalVecCount; i++) // this hsould be changed
-  {
-    err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueNDRangeKernel failed\n");
-      return -1;
-    }
-
-    err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueReadBuffer failed\n");
-      return -1;
-    }
-
-    if (verify_sign_double(input_ptr[0], output_ptr, n_elems*(i+1)))
-    {
-      log_error("%s test failed\n", fn_names_double[i]);
-      err = -1;
-    }
-    else
-    {
-      log_info("%s test passed\n", fn_names_double[i]);
-      err = 0;
-    }
-
-    if (err)
-      break;
-  }
-
-  clReleaseMemObject(streams[0]);
-  clReleaseMemObject(streams[1]);
-  for (i=0; i<kTotalVecCount; i++)
-  {
-    clReleaseKernel(kernel[i]);
-    clReleaseProgram(program[i]);
-  }
-  free(input_ptr[0]);
-  free(output_ptr);
-
-  return err;
-}
-
-

test_conformance/commonfns/test_smoothstep.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -13,72 +13,63 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
-
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 
 #include "procs.h"
+#include "test_base.h"
 
-static const char *smoothstep_kernel_code =
-"__kernel void test_smoothstep(__global float *edge0, __global float *edge1, __global float *x, __global float *dst)\n"
+
+const char *smoothstep_fn_code_pattern =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s%s *e0, __global %s%s *e1, __global %s%s "
+    "*x, __global %s%s *dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
     "\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
+    "    dst[tid] = smoothstep(e0[tid], e1[tid], x[tid]);\n"
     "}\n";
 
-static const char *smoothstep2_kernel_code =
-"__kernel void test_smoothstep2(__global float2 *edge0, __global float2 *edge1, __global float2 *x, __global float2 *dst)\n"
+const char *smoothstep_fn_code_pattern_v3 =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *e0, __global %s *e1, __global %s *x, "
+    "__global %s *dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
     "\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
+    "    vstore3(smoothstep(vload3(tid,e0), vload3(tid,e1), vload3(tid,x)), "
+    "tid, dst);\n"
     "}\n";
 
-static const char *smoothstep4_kernel_code =
-"__kernel void test_smoothstep4(__global float4 *edge0, __global float4 *edge1, __global float4 *x, __global float4 *dst)\n"
+const char *smoothstep_fn_code_pattern_v3_scalar =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *e0, __global %s *e1, __global %s *x, "
+    "__global %s *dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
     "\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
+    "    vstore3(smoothstep(e0[tid], e1[tid], vload3(tid,x)), tid, dst);\n"
     "}\n";
 
-static const char *smoothstep8_kernel_code =
-"__kernel void test_smoothstep8(__global float8 *edge0, __global float8 *edge1, __global float8 *x, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
-"}\n";
-
-static const char *smoothstep16_kernel_code =
-"__kernel void test_smoothstep16(__global float16 *edge0, __global float16 *edge1, __global float16 *x, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
-"}\n";
-
-static const char *smoothstep3_kernel_code =
-"__kernel void test_smoothstep3(__global float *edge0, __global float *edge1, __global float *x, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(smoothstep(vload3(tid,edge0),vload3(tid,edge1),vload3(tid,x)), tid, dst);\n"
-"}\n";
 
 #define MAX_ERR (1e-5f)
 
-static float
-verify_smoothstep(float *edge0, float *edge1, float *x, float *outptr, int n)
-{
-  float       r, t, delta, max_err = 0.0f;
-  int         i;
+namespace {
 
-  for (i=0; i<n; i++)
+
+template <typename T>
+int verify_smoothstep(const T *const edge0, const T *const edge1,
+                      const T *const x, const T *const outptr, const int n,
+                      const int veclen, const bool vecParam)
+{
+    T r, t;
+    float delta = 0;
+
+    if (vecParam)
+    {
+        for (int i = 0; i < n * veclen; i++)
         {
             t = (x[i] - edge0[i]) / (edge1[i] - edge0[i]);
             if (t < 0.0f)
@@ -87,196 +78,218 @@ verify_smoothstep(float *edge0, float *edge1, float *x, float *outptr, int n)
                 t = 1.0f;
             r = t * t * (3.0f - 2.0f * t);
             delta = (float)fabs(r - outptr[i]);
-    if (delta > max_err)
-      max_err = delta;
-  }
-
-  return max_err;
-}
-
-const static char *fn_names[] = { "SMOOTHSTEP float", "SMOOTHSTEP float2", "SMOOTHSTEP float4", "SMOOTHSTEP float8", "SMOOTHSTEP float16", "SMOOTHSTEP float3" };
-
-int
-test_smoothstep(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
+            if (delta > MAX_ERR)
             {
-  cl_mem      streams[4];
-  cl_float    *input_ptr[3], *output_ptr, *p, *p_edge0;
-  cl_program  program[kTotalVecCount];
-  cl_kernel   kernel[kTotalVecCount];
-  size_t  threads[1];
-  float max_err;
-  int num_elements;
-  int err;
-  int i;
-  MTdata d;
-
-  num_elements = n_elems * 16;
-
-  input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  input_ptr[2] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[0])
-  {
-    log_error("clCreateBuffer failed\n");
+                log_error("%d) verification error: smoothstep(%a, %a, %a) = "
+                          "*%a vs. %a\n",
+                          i, x[i], edge0[i], edge1[i], r, outptr[i]);
                 return -1;
             }
-  streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[1])
+        }
+    }
+    else
     {
-    log_error("clCreateBuffer failed\n");
+        for (int i = 0; i < n; ++i)
+        {
+            int ii = i / veclen;
+            int vi = i * veclen;
+            for (int j = 0; j < veclen; ++j, ++vi)
+            {
+                t = (x[vi] - edge0[i]) / (edge1[i] - edge0[i]);
+                if (t < 0.0f)
+                    t = 0.0f;
+                else if (t > 1.0f)
+                    t = 1.0f;
+                r = t * t * (3.0f - 2.0f * t);
+                delta = (float)fabs(r - outptr[vi]);
+                if (delta > MAX_ERR)
+                {
+                    log_error("{%d, element %d}) verification error: "
+                              "smoothstep(%a, %a, %a) = *%a vs. %a\n",
+                              ii, j, x[vi], edge0[i], edge1[i], r, outptr[vi]);
                     return -1;
                 }
-  streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[2])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
+            }
+        }
+    }
+    return 0;
 }
 
-  streams[3] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[3])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
 }
 
-  p = input_ptr[0];
-  d = init_genrand( gRandomSeed );
+
+template <typename T>
+int test_smoothstep_fn(cl_device_id device, cl_context context,
+                       cl_command_queue queue, int n_elems, bool vecParam)
+{
+    clMemWrapper streams[4];
+    std::vector<T> input_ptr[3], output_ptr;
+
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+
+    int err, i;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
+
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
+
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
+
+    int num_elements = n_elems * (1 << (kTotalVecCount - 1));
+
+    for (i = 0; i < 3; i++) input_ptr[i].resize(num_elements);
+    output_ptr.resize(num_elements);
+
+    for (i = 0; i < 4; i++)
+    {
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
+        test_error(err, "clCreateBuffer failed");
+    }
+
+    std::string pragma_str;
+    if (std::is_same<T, float>::value)
+    {
         for (i = 0; i < num_elements; i++)
         {
-    p[i] = get_random_float(-0x00400000, 0x00400000, d);
+            input_ptr[0][i] = get_random_float(-0x00200000, 0x00010000, d);
+            input_ptr[1][i] = get_random_float(input_ptr[0][i], 0x00200000, d);
+            input_ptr[2][i] = get_random_float(-0x20000000, 0x20000000, d);
         }
-
-  p = input_ptr[1];
-  p_edge0 = input_ptr[0];
+    }
+    else if (std::is_same<T, double>::value)
+    {
+        pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
         for (i = 0; i < num_elements; i++)
         {
-    float edge0 = p_edge0[i];
-    float edge1;
-    do {
-      edge1 = get_random_float(-0x00400000, 0x00400000, d);
-      if (edge0 < edge1)
-        break;
-    } while (1);
-    p[i] = edge1;
+            input_ptr[0][i] = get_random_double(-0x00200000, 0x00010000, d);
+            input_ptr[1][i] = get_random_double(input_ptr[0][i], 0x00200000, d);
+            input_ptr[2][i] = get_random_double(-0x20000000, 0x20000000, d);
+        }
     }
 
-  p = input_ptr[2];
-  for (i=0; i<num_elements; i++)
+    for (i = 0; i < 3; i++)
     {
-    p[i] = get_random_float(-0x00400000, 0x00400000, d);
-  }
-  free_mtdata(d);
-  d = NULL;
-
-  err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-  err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-  err = clEnqueueWriteBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[2], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
+        err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
+                                   sizeof(T) * num_elements,
+                                   &input_ptr[i].front(), 0, NULL, NULL);
+        test_error(err, "Unable to write input buffer");
     }
 
-  err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &smoothstep_kernel_code, "test_smoothstep" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &smoothstep2_kernel_code, "test_smoothstep2" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &smoothstep4_kernel_code, "test_smoothstep4" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &smoothstep8_kernel_code, "test_smoothstep8" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &smoothstep16_kernel_code, "test_smoothstep16" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &smoothstep3_kernel_code, "test_smoothstep3" );
-  if (err)
-    return -1;
+    char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
 
     for (i = 0; i < kTotalVecCount; i++)
     {
-      err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-      err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-      err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-      err |= clSetKernelArg(kernel[i], 3, sizeof streams[3], &streams[3] );
-      if (err != CL_SUCCESS)
+        std::string kernelSource;
+        if (i >= kVectorSizeCount)
         {
-      log_error("clSetKernelArgs failed\n");
-      return -1;
+            if (vecParam)
+            {
+                std::string str = smoothstep_fn_code_pattern_v3;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), tname.c_str());
+            }
+            else
+            {
+                std::string str = smoothstep_fn_code_pattern_v3_scalar;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str(), tname.c_str());
             }
         }
-
-
-  threads[0] = (size_t)n_elems;
-  for (i=0; i<kTotalVecCount; i++)
+        else
         {
-    err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
+            // regular path
+            std::string str = smoothstep_fn_code_pattern;
+            kernelSource =
+                string_format(str, pragma_str.c_str(), tname.c_str(),
+                              vecParam ? vecSizeNames[i] : "", tname.c_str(),
+                              vecParam ? vecSizeNames[i] : "", tname.c_str(),
+                              vecSizeNames[i], tname.c_str(), vecSizeNames[i]);
+        }
+        const char *programPtr = kernelSource.c_str();
+        err =
+            create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
+                                        (const char **)&programPtr, "test_fn");
+        test_error(err, "Unable to create kernel");
+
+        for (int j = 0; j < 4; j++)
         {
-      log_error("clEnqueueNDRangeKernel failed\n");
-      return -1;
+            err =
+                clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
+            test_error(err, "Unable to set kernel argument");
         }
 
+        size_t threads = (size_t)n_elems;
 
-    err = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
+        test_error(err, "Unable to execute kernel");
+
+        err = clEnqueueReadBuffer(queue, streams[3], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
+        test_error(err, "Unable to read results");
+
+        if (verify_smoothstep((T *)&input_ptr[0].front(),
+                              (T *)&input_ptr[1].front(),
+                              (T *)&input_ptr[2].front(), &output_ptr[0],
+                              n_elems, g_arrVecSizes[i], vecParam))
         {
-      log_error("clEnqueueReadBuffer failed\n");
-      return -1;
-    }
-
-    max_err = verify_smoothstep(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems * g_arrVecSizes[i]);
-
-    if (max_err > MAX_ERR)
-    {
-      log_error("%s test failed %g max err\n", fn_names[i], max_err);
+            log_error("smoothstep %s%d%s test failed\n", tname.c_str(),
+                      ((g_arrVecSizes[i])),
+                      vecParam ? "" : std::string(", " + tname).c_str());
             err = -1;
         }
         else
         {
-      log_info("%s test passed %g max err\n", fn_names[i], max_err);
+            log_info("smoothstep %s%d%s test passed\n", tname.c_str(),
+                     ((g_arrVecSizes[i])),
+                     vecParam ? "" : std::string(", " + tname).c_str());
             err = 0;
         }
 
-    if (err)
-      break;
+        if (err) break;
     }
 
-  clReleaseMemObject(streams[0]);
-  clReleaseMemObject(streams[1]);
-  clReleaseMemObject(streams[2]);
-  clReleaseMemObject(streams[3]);
-  for (i=0; i<kTotalVecCount; i++)
-  {
-    clReleaseKernel(kernel[i]);
-    clReleaseProgram(program[i]);
-  }
-  free(input_ptr[0]);
-  free(input_ptr[1]);
-  free(input_ptr[2]);
-  free(output_ptr);
-
     return err;
 }
 
 
+cl_int SmoothstepTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error =
+        test_smoothstep_fn<float>(device, context, queue, num_elems, vecParam);
+    test_error(error, "SmoothstepTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_smoothstep_fn<double>(device, context, queue, num_elems,
+                                           vecParam);
+        test_error(error, "SmoothstepTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+int test_smoothstep(cl_device_id device, cl_context context,
+                    cl_command_queue queue, int n_elems)
+{
+    return MakeAndRunTest<SmoothstepTest>(device, context, queue, n_elems,
+                                          "smoothstep", true);
+}
+
+
+int test_smoothstepf(cl_device_id device, cl_context context,
+                     cl_command_queue queue, int n_elems)
+{
+    return MakeAndRunTest<SmoothstepTest>(device, context, queue, n_elems,
+                                          "smoothstep", false);
+}

test_conformance/commonfns/test_smoothstepf.cpp

@@ -1,259 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static const char *smoothstep_kernel_code =
-"__kernel void test_smoothstep(__global float *edge0, __global float *edge1, __global float *x, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
-"}\n";
-
-static const char *smoothstep2_kernel_code =
-"__kernel void test_smoothstep2f(__global float *edge0, __global float *edge1, __global float2 *x, __global float2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
-"}\n";
-
-static const char *smoothstep4_kernel_code =
-"__kernel void test_smoothstep4f(__global float *edge0, __global float *edge1, __global float4 *x, __global float4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = smoothstep(edge0[tid], edge1[tid], x[tid]);\n"
-"}\n";
-
-#define MAX_ERR (1e-5f)
-
-float verify_smoothstep(float *edge0, float *edge1, float *x, float *outptr,
-                        int n, int veclen)
-{
-  float       r, t, delta, max_err = 0.0f;
-  int         i, j;
-
-  for (i = 0; i < n; ++i) {
-    int vi = i * veclen;
-    for (j = 0; j < veclen; ++j, ++vi) {
-      t = (x[vi] - edge0[i]) / (edge1[i] - edge0[i]);
-      if (t < 0.0f)
-        t = 0.0f;
-      else if (t > 1.0f)
-        t = 1.0f;
-      r = t * t * (3.0f - 2.0f * t);
-      delta = (float)fabs(r - outptr[vi]);
-      if (delta > max_err)
-        max_err = delta;
-    }
-  }
-  return max_err;
-}
-
-const static char *fn_names[] = { "SMOOTHSTEP float", "SMOOTHSTEP float2", "SMOOTHSTEP float4"};
-
-int
-test_smoothstepf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-  cl_mem      streams[4];
-  cl_float    *input_ptr[3], *output_ptr, *p, *p_edge0;
-  cl_program  program[3];
-  cl_kernel   kernel[3];
-  size_t  threads[1];
-  float max_err = 0.0f;
-  int num_elements;
-  int err;
-  int i;
-  MTdata d;
-
-  num_elements = n_elems * 4;
-
-  input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  input_ptr[2] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-  streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[0])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-  streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[1])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-  streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[2])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  streams[3] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                              sizeof(cl_float) * num_elements, NULL, NULL);
-  if (!streams[3])
-  {
-    log_error("clCreateBuffer failed\n");
-    return -1;
-  }
-
-  d = init_genrand( gRandomSeed );
-  p = input_ptr[0];
-  for (i=0; i<num_elements; i++)
-  {
-    p[i] = get_random_float(-0x00200000, 0x00200000, d);
-  }
-
-  p = input_ptr[1];
-  p_edge0 = input_ptr[0];
-  for (i=0; i<num_elements; i++)
-  {
-    float edge0 = p_edge0[i];
-    float edge1;
-    do {
-      edge1 = get_random_float( -0x00200000, 0x00200000, d);
-      if (edge0 < edge1)
-        break;
-    } while (1);
-    p[i] = edge1;
-  }
-
-  p = input_ptr[2];
-  for (i=0; i<num_elements; i++)
-  {
-    p[i] = get_random_float(-0x00200000, 0x00200000, d);
-  }
-  free_mtdata(d);
-  d = NULL;
-
-  err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-  err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-  err = clEnqueueWriteBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[2], 0, NULL, NULL );
-  if (err != CL_SUCCESS)
-  {
-    log_error("clWriteArray failed\n");
-    return -1;
-  }
-
-  err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &smoothstep_kernel_code, "test_smoothstep" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &smoothstep2_kernel_code, "test_smoothstep2f" );
-  if (err)
-    return -1;
-  err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &smoothstep4_kernel_code, "test_smoothstep4f" );
-  if (err)
-    return -1;
-
-  for (i=0; i<3; i++)
-  {
-      err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-      err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-      err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-      err |= clSetKernelArg(kernel[i], 3, sizeof streams[3], &streams[3] );
-      if (err != CL_SUCCESS)
-    {
-      log_error("clSetKernelArgs failed\n");
-      return -1;
-    }
-  }
-
-  threads[0] = (size_t)n_elems;
-  for (i=0; i<3; i++)
-  {
-    err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueNDRangeKernel failed\n");
-      return -1;
-    }
-
-    err = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-      log_error("clEnqueueReadBuffer failed\n");
-      return -1;
-    }
-
-    switch (i)
-    {
-      case 0:
-        max_err = verify_smoothstep(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems, 1);
-        break;
-      case 1:
-        max_err = verify_smoothstep(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems, 2);
-        break;
-      case 2:
-        max_err = verify_smoothstep(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, n_elems, 4);
-        break;
-    }
-
-    if (max_err > MAX_ERR)
-    {
-      log_error("%s test failed %g max err\n", fn_names[i], max_err);
-      err = -1;
-    }
-    else
-    {
-      log_info("%s test passed %g max err\n", fn_names[i], max_err);
-      err = 0;
-    }
-
-    if (err)
-      break;
-  }
-
-  clReleaseMemObject(streams[0]);
-  clReleaseMemObject(streams[1]);
-  clReleaseMemObject(streams[2]);
-  clReleaseMemObject(streams[3]);
-  for (i=0; i<3; i++)
-  {
-    clReleaseKernel(kernel[i]);
-    clReleaseProgram(program[i]);
-  }
-  free(input_ptr[0]);
-  free(input_ptr[1]);
-  free(input_ptr[2]);
-  free(output_ptr);
-
-  return err;
-}
-
-

test_conformance/commonfns/test_step.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -13,524 +13,252 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#include "harness/compat.h"
-
 #include <stdio.h>
 #include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 
 #include "procs.h"
-
-static int
-test_step_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
+#include "test_base.h"
 
 
-const char *step_kernel_code =
-"__kernel void test_step(__global float *srcA, __global float *srcB, __global float *dst)\n"
+const char *step_fn_code_pattern = "%s\n" /* optional pragma */
+                                   "__kernel void test_fn(__global %s%s *edge, "
+                                   "__global %s%s *x, __global %s%s *dst)\n"
                                    "{\n"
                                    "    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
+                                   "    dst[tid] = step(edge[tid], x[tid]);\n"
                                    "}\n";
 
-const char *step2_kernel_code =
-"__kernel void test_step2(__global float2 *srcA, __global float2 *srcB, __global float2 *dst)\n"
+const char *step_fn_code_pattern_v3 =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *edge, __global %s *x, __global %s "
+    "*dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
+    "    vstore3(step(vload3(tid,edge), vload3(tid,x)), tid, dst);\n"
     "}\n";
 
-const char *step4_kernel_code =
-"__kernel void test_step4(__global float4 *srcA, __global float4 *srcB, __global float4 *dst)\n"
+const char *step_fn_code_pattern_v3_scalar =
+    "%s\n" /* optional pragma */
+    "__kernel void test_fn(__global %s *edge, __global %s *x, __global %s "
+    "*dst)\n"
     "{\n"
     "    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step8_kernel_code =
-"__kernel void test_step8(__global float8 *srcA, __global float8 *srcB, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step16_kernel_code =
-"__kernel void test_step16(__global float16 *srcA, __global float16 *srcB, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step3_kernel_code =
-"__kernel void test_step3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(step(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
+    "    vstore3(step(edge[tid], vload3(tid,x)), tid, dst);\n"
     "}\n";
 
 
-int
-verify_step(float *inptrA, float *inptrB, float *outptr, int n)
+namespace {
+
+template <typename T>
+int verify_step(const T *const inptrA, const T *const inptrB,
+                const T *const outptr, const int n, const int veclen,
+                const bool vecParam)
 {
-    float       r;
-    int         i;
+    T r;
 
-    for (i=0; i<n; i++)
+    if (vecParam)
     {
-        r = (inptrB[i] < inptrA[i]) ? 0.0f : 1.0f;
-        if (r != outptr[i])
-            return -1;
-    }
-
-    return 0;
-}
-
-int
-test_step(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem      streams[3];
-    cl_float    *input_ptr[2], *output_ptr, *p;
-  cl_program  program[kTotalVecCount];
-  cl_kernel   kernel[kTotalVecCount];
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-  num_elements = n_elems * 16;
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x40000000, 0x40000000, d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x40000000, 0x40000000, d);
-    }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code, "test_step" );
-    if (err) return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code, "test_step2" );
-    if (err) return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code, "test_step4" );
-    if (err) return -1;
-    err = create_single_kernel_helper(context, &program[3], &kernel[3], 1,
-                                      &step8_kernel_code, "test_step8");
-    if (err) return -1;
-    err = create_single_kernel_helper(context, &program[4], &kernel[4], 1,
-                                      &step16_kernel_code, "test_step16");
-    if (err) return -1;
-    err = create_single_kernel_helper(context, &program[5], &kernel[5], 1,
-                                      &step3_kernel_code, "test_step3");
-    if (err) return -1;
-
-  for (i=0; i <kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-  for (i=0; i<kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        switch (i)
-        {
-            case 0:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems);
-                if (err)
-                    log_error("STEP float test failed\n");
-                else
-                    log_info("STEP float test passed\n");
-                break;
-
-            case 1:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*2);
-                if (err)
-                    log_error("STEP float2 test failed\n");
-                else
-                    log_info("STEP float2 test passed\n");
-                break;
-
-            case 2:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*4);
-                if (err)
-                    log_error("STEP float4 test failed\n");
-                else
-                    log_info("STEP float4 test passed\n");
-                break;
-
-        case 3:
-        err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*8);
-        if (err)
-          log_error("STEP float8 test failed\n");
-        else
-          log_info("STEP float8 test passed\n");
-        break;
-
-        case 4:
-        err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*16);
-        if (err)
-          log_error("STEP float16 test failed\n");
-        else
-          log_info("STEP float16 test passed\n");
-        break;
-
-        case 5:
-        err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*3);
-        if (err)
-          log_error("STEP float3 test failed\n");
-        else
-          log_info("STEP float3 test passed\n");
-        break;
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-  for (i=0; i<kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    if( err )
-        return err;
-
-    if( ! is_extension_available( device, "cl_khr_fp64" ))
-        return 0;
-
-    return test_step_double( device, context, queue, n_elems);
-}
-
-
-#pragma mark -
-
-const char *step_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step2_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step2_double(__global double2 *srcA, __global double2 *srcB, __global double2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step4_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step4_double(__global double4 *srcA, __global double4 *srcB, __global double4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step8_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step8_double(__global double8 *srcA, __global double8 *srcB, __global double8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step16_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step16_double(__global double16 *srcA, __global double16 *srcB, __global double16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-const char *step3_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step3_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(step(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
-"}\n";
-
-
-int
-verify_step_double(double *inptrA, double *inptrB, double *outptr, int n)
-{
-    double       r;
-    int         i;
-
-    for (i=0; i<n; i++)
+        for (int i = 0; i < n * veclen; i++)
         {
             r = (inptrB[i] < inptrA[i]) ? 0.0 : 1.0;
+            if (r != outptr[i]) return -1;
+        }
+    }
+    else
+    {
+        for (int i = 0; i < n;)
+        {
+            int ii = i / veclen;
+            for (int j = 0; j < veclen && i < n; ++j, ++i)
+            {
+                r = (inptrB[i] < inptrA[ii]) ? 0.0f : 1.0f;
                 if (r != outptr[i])
+                {
+                    log_error("Failure @ {%d, element %d}: step(%a,%a) -> *%a "
+                              "vs %a\n",
+                              ii, j, inptrA[ii], inptrB[i], r, outptr[i]);
                     return -1;
                 }
+            }
+        }
+    }
 
     return 0;
 }
 
-static int
-test_step_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem      streams[3];
-    cl_double    *input_ptr[2], *output_ptr, *p;
-    cl_program  program[kTotalVecCount];
-    cl_kernel   kernel[kTotalVecCount];
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-    num_elements = n_elems * 16;
-
-    input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    input_ptr[1] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
 }
 
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
+
+template <typename T>
+int test_step_fn(cl_device_id device, cl_context context,
+                 cl_command_queue queue, int n_elems, bool vecParam)
+{
+    clMemWrapper streams[3];
+    std::vector<T> input_ptr[2], output_ptr;
+
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+
+    int err, i;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
+
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
+    int num_elements = n_elems * (1 << (kTotalVecCount - 1));
+
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
+
+    for (i = 0; i < 2; i++) input_ptr[i].resize(num_elements);
+    output_ptr.resize(num_elements);
+
+    for (i = 0; i < 3; i++)
+    {
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
+        test_error(err, "clCreateBuffer failed");
+    }
+
+    std::string pragma_str;
+    if (std::is_same<T, float>::value)
+    {
         for (i = 0; i < num_elements; i++)
         {
-        p[i] = get_random_double(-0x40000000, 0x40000000, d);
+            input_ptr[0][i] = get_random_float(-0x40000000, 0x40000000, d);
+            input_ptr[1][i] = get_random_float(-0x40000000, 0x40000000, d);
         }
-    p = input_ptr[1];
+    }
+    else if (std::is_same<T, double>::value)
+    {
+        pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
         for (i = 0; i < num_elements; i++)
         {
-        p[i] = get_random_double(-0x40000000, 0x40000000, d);
+            input_ptr[0][i] = get_random_double(-0x40000000, 0x40000000, d);
+            input_ptr[1][i] = get_random_double(-0x40000000, 0x40000000, d);
         }
-    free_mtdata(d); d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
     }
 
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code_double, "test_step_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code_double, "test_step2_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code_double, "test_step4_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &step8_kernel_code_double, "test_step8_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &step16_kernel_code_double, "test_step16_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &step3_kernel_code_double, "test_step3_double" );
-    if (err)
-        return -1;
+    for (i = 0; i < 2; i++)
+    {
+        err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
+                                   sizeof(T) * num_elements,
+                                   &input_ptr[i].front(), 0, NULL, NULL);
+        test_error(err, "Unable to write input buffer");
+    }
+
+    char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
 
     for (i = 0; i < kTotalVecCount; i++)
     {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
+        std::string kernelSource;
+        if (i >= kVectorSizeCount)
         {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i<kTotalVecCount; i++)
+            if (vecParam)
             {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
+                std::string str = step_fn_code_pattern_v3;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str());
+            }
+            else
             {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
+                std::string str = step_fn_code_pattern_v3_scalar;
+                kernelSource =
+                    string_format(str, pragma_str.c_str(), tname.c_str(),
+                                  tname.c_str(), tname.c_str());
             }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
+        }
+        else
         {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
+            // regular path
+            std::string str = step_fn_code_pattern;
+            kernelSource =
+                string_format(str, pragma_str.c_str(), tname.c_str(),
+                              vecParam ? vecSizeNames[i] : "", tname.c_str(),
+                              vecSizeNames[i], tname.c_str(), vecSizeNames[i]);
         }
+        const char *programPtr = kernelSource.c_str();
+        err =
+            create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
+                                        (const char **)&programPtr, "test_fn");
+        test_error(err, "Unable to create kernel");
 
-        switch (i)
+        for (int j = 0; j < 3; j++)
         {
-            case 0:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems);
-                if (err)
-                    log_error("STEP double test failed\n");
-                else
-                    log_info("STEP double test passed\n");
-                break;
-
-            case 1:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*2);
-                if (err)
-                    log_error("STEP double2 test failed\n");
-                else
-                    log_info("STEP double2 test passed\n");
-                break;
-
-            case 2:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*4);
-                if (err)
-                    log_error("STEP double4 test failed\n");
-                else
-                    log_info("STEP double4 test passed\n");
-                break;
-
-        case 3:
-        err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*8);
-        if (err)
-          log_error("STEP double8 test failed\n");
-        else
-          log_info("STEP double8 test passed\n");
-        break;
-
-        case 4:
-        err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*16);
-        if (err)
-          log_error("STEP double16 test failed\n");
-        else
-          log_info("STEP double16 test passed\n");
-        break;
-
-        case 5:
-        err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*3);
-        if (err)
-          log_error("STEP double3 test failed\n");
-        else
-          log_info("STEP double3 test passed\n");
-        break;
+            err =
+                clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
+            test_error(err, "Unable to set kernel argument");
         }
 
-        if (err)
-            break;
-    }
+        size_t threads = (size_t)n_elems;
 
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i<kTotalVecCount; i++)
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
+        test_error(err, "Unable to execute kernel");
+
+        err = clEnqueueReadBuffer(queue, streams[2], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
+        test_error(err, "Unable to read results");
+
+        err = verify_step(&input_ptr[0].front(), &input_ptr[1].front(),
+                          &output_ptr.front(), n_elems, g_arrVecSizes[i],
+                          vecParam);
+        if (err)
         {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
+            log_error("step %s%d%s test failed\n", tname.c_str(),
+                      ((g_arrVecSizes[i])),
+                      vecParam ? "" : std::string(", " + tname).c_str());
+            err = -1;
+        }
+        else
+        {
+            log_info("step %s%d%s test passed\n", tname.c_str(),
+                     ((g_arrVecSizes[i])),
+                     vecParam ? "" : std::string(", " + tname).c_str());
+            err = 0;
+        }
+
+        if (err)
+            break;
     }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
 
     return err;
 }
 
+
+cl_int StepTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+
+    error = test_step_fn<float>(device, context, queue, num_elems, vecParam);
+    test_error(error, "StepTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error =
+            test_step_fn<double>(device, context, queue, num_elems, vecParam);
+        test_error(error, "StepTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+int test_step(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<StepTest>(device, context, queue, n_elems, "step",
+                                    true);
+}
+
+
+int test_stepf(cl_device_id device, cl_context context, cl_command_queue queue,
+               int n_elems)
+{
+    return MakeAndRunTest<StepTest>(device, context, queue, n_elems, "step",
+                                    false);
+}

test_conformance/commonfns/test_stepf.cpp

@@ -1,546 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "harness/compat.h"
-
-#include <stdio.h>
-#include <string.h>
-#include <sys/types.h>
-#include <sys/stat.h>
-
-#include "procs.h"
-
-static int test_stepf_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
-
-
-static const char *step_kernel_code =
-"__kernel void test_step(__global float *srcA, __global float *srcB, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step2_kernel_code =
-"__kernel void test_step2(__global float *srcA, __global float2 *srcB, __global float2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step4_kernel_code =
-"__kernel void test_step4(__global float *srcA, __global float4 *srcB, __global float4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step8_kernel_code =
-"__kernel void test_step8(__global float *srcA, __global float8 *srcB, __global float8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step16_kernel_code =
-"__kernel void test_step16(__global float *srcA, __global float16 *srcB, __global float16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step3_kernel_code =
-"__kernel void test_step3(__global float *srcA, __global float *srcB, __global float *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(step(srcA[tid], vload3(tid,srcB)) ,tid,dst);\n"
-"}\n";
-
-
-static int
-verify_step( cl_float *inptrA, cl_float *inptrB, cl_float *outptr, int n, int veclen)
-{
-    float       r;
-    int         i, j;
-
-    for (i=0; i<n; ) {
-        int ii = i/veclen;
-        for (j=0; j<veclen && i<n; ++j, ++i) {
-            r = (inptrB[i] < inptrA[ii]) ? 0.0f : 1.0f;
-            if (r != outptr[i])
-            {
-                log_error( "Failure @ {%d, element %d}: step(%a,%a) -> *%a vs %a\n", ii, j, inptrA[ii], inptrB[i], r, outptr[i] );
-                return -1;
-            }
-        }
-    }
-
-    return 0;
-}
-
-int test_stepf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem      streams[3];
-    cl_float    *input_ptr[2], *output_ptr, *p;
-    cl_program  program[kTotalVecCount];
-    cl_kernel   kernel[kTotalVecCount];
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata d;
-    num_elements = n_elems * 16;
-
-    input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_float) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x40000000, 0x40000000, d);
-    }
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-    {
-        p[i] = get_random_float(-0x40000000, 0x40000000, d);
-    }
-    free_mtdata(d);   d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code, "test_step" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code, "test_step2" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code, "test_step4" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &step8_kernel_code, "test_step8" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &step16_kernel_code, "test_step16" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &step3_kernel_code, "test_step3" );
-    if (err)
-        return -1;
-
-    for (i=0; i <kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        switch (i)
-        {
-            case 0:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems, 1);
-                if (err)
-                    log_error("STEP float test failed\n");
-                else
-                    log_info("STEP float test passed\n");
-                break;
-
-            case 1:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*2, 2);
-                if (err)
-                    log_error("STEP float2 test failed\n");
-                else
-                    log_info("STEP float2 test passed\n");
-                break;
-
-            case 2:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*4, 4);
-                if (err)
-                    log_error("STEP float4 test failed\n");
-                else
-                    log_info("STEP float4 test passed\n");
-                break;
-
-            case 3:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*8, 8);
-                if (err)
-                    log_error("STEP float8 test failed\n");
-                else
-                    log_info("STEP float8 test passed\n");
-                break;
-
-            case 4:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*16, 16);
-                if (err)
-                    log_error("STEP float16 test failed\n");
-                else
-                    log_info("STEP float16 test passed\n");
-                break;
-
-            case 5:
-                err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*3, 3);
-                if (err)
-                    log_error("STEP float3 test failed\n");
-                else
-                    log_info("STEP float3 test passed\n");
-                break;
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    if(err)
-        return err;
-
-    if( ! is_extension_available( device, "cl_khr_fp64" ))
-    {
-        log_info( "Device does not support cl_khr_fp64.  Skipping double precision tests.\n" );
-        return 0;
-    }
-
-    return test_stepf_double( device, context, queue, n_elems);
-}
-
-#pragma mark -
-
-static const char *step_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step2_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step2_double(__global double *srcA, __global double2 *srcB, __global double2 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step4_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step4_double(__global double *srcA, __global double4 *srcB, __global double4 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step8_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step8_double(__global double *srcA, __global double8 *srcB, __global double8 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step16_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step16_double(__global double *srcA, __global double16 *srcB, __global double16 *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    dst[tid] = step(srcA[tid], srcB[tid]);\n"
-"}\n";
-
-static const char *step3_kernel_code_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void test_step3_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"\n"
-"    vstore3(step(srcA[tid], vload3(tid,srcB)) ,tid,dst);\n"
-"}\n";
-
-
-static int
-verify_step_double(cl_double *inptrA, cl_double *inptrB, cl_double *outptr, int n, int veclen)
-{
-    double r;
-    int    i, j;
-
-    for (i=0; i<n; ) {
-        int ii = i/veclen;
-        for (j=0; j<veclen && i<n; ++j, ++i) {
-            r = (inptrB[i] < inptrA[ii]) ? 0.0 : 1.0;
-            if (r != outptr[i])
-            {
-                log_error( "Failure @ {%d, element %d}: step(%a,%a) -> *%a vs %a\n", ii, j, inptrA[ii], inptrB[i], r, outptr[i] );
-                return -1;
-            }
-        }
-    }
-
-    return 0;
-}
-
-int test_stepf_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
-{
-    cl_mem      streams[3];
-    cl_double  *input_ptr[2], *output_ptr, *p;
-    cl_program  program[kTotalVecCount];
-    cl_kernel   kernel[kTotalVecCount];
-    size_t  threads[1];
-    int num_elements;
-    int err;
-    int i;
-    MTdata    d;
-    num_elements = n_elems * 16;
-
-    input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    input_ptr[1] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
-    streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[0])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[1])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
-                                sizeof(cl_double) * num_elements, NULL, NULL);
-    if (!streams[2])
-    {
-        log_error("clCreateBuffer failed\n");
-        return -1;
-    }
-
-    p = input_ptr[0];
-    d = init_genrand( gRandomSeed );
-    for (i=0; i<num_elements; i++)
-        p[i] = get_random_double(-0x40000000, 0x40000000, d);
-
-    p = input_ptr[1];
-    for (i=0; i<num_elements; i++)
-        p[i] = get_random_double(-0x40000000, 0x40000000, d);
-
-    free_mtdata(d);   d = NULL;
-
-    err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-    err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
-    if (err != CL_SUCCESS)
-    {
-        log_error("clWriteArray failed\n");
-        return -1;
-    }
-
-    err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code_double, "test_step_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code_double, "test_step2_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code_double, "test_step4_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &step8_kernel_code_double, "test_step8_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &step16_kernel_code_double, "test_step16_double" );
-    if (err)
-        return -1;
-    err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &step3_kernel_code_double, "test_step3_double" );
-    if (err)
-        return -1;
-
-    for (i=0; i <kTotalVecCount; i++)
-    {
-        err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
-        err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
-        err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clSetKernelArgs failed\n");
-            return -1;
-        }
-    }
-
-    threads[0] = (size_t)n_elems;
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueNDRangeKernel failed\n");
-            return -1;
-        }
-
-        err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
-        if (err != CL_SUCCESS)
-        {
-            log_error("clEnqueueReadBuffer failed\n");
-            return -1;
-        }
-
-        switch (i)
-        {
-            case 0:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems, 1);
-                if (err)
-                    log_error("STEP double test failed\n");
-                else
-                    log_info("STEP double test passed\n");
-                break;
-
-            case 1:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*2, 2);
-                if (err)
-                    log_error("STEP double2 test failed\n");
-                else
-                    log_info("STEP double2 test passed\n");
-                break;
-
-            case 2:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*4, 4);
-                if (err)
-                    log_error("STEP double4 test failed\n");
-                else
-                    log_info("STEP double4 test passed\n");
-                break;
-
-            case 3:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*8, 8);
-                if (err)
-                    log_error("STEP double8 test failed\n");
-                else
-                    log_info("STEP double8 test passed\n");
-                break;
-
-            case 4:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*16, 16);
-                if (err)
-                    log_error("STEP double16 test failed\n");
-                else
-                    log_info("STEP double16 test passed\n");
-                break;
-
-            case 5:
-                err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*3, 3);
-                if (err)
-                    log_error("STEP double3 test failed\n");
-                else
-                    log_info("STEP double3 test passed\n");
-                break;
-        }
-
-        if (err)
-            break;
-    }
-
-    clReleaseMemObject(streams[0]);
-    clReleaseMemObject(streams[1]);
-    clReleaseMemObject(streams[2]);
-    for (i=0; i<kTotalVecCount; i++)
-    {
-        clReleaseKernel(kernel[i]);
-        clReleaseProgram(program[i]);
-    }
-    free(input_ptr[0]);
-    free(input_ptr[1]);
-    free(output_ptr);
-
-    return err;
-}
-

test_conformance/commonfns/test_unary_fn.cpp

@@ -0,0 +1,365 @@
+//
+// Copyright (c) 2023 The Khronos Group Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+#include <stdio.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include <vector>
+
+#include "harness/deviceInfo.h"
+#include "harness/typeWrappers.h"
+
+#include "procs.h"
+#include "test_base.h"
+
+#ifndef M_PI
+#define M_PI 3.14159265358979323846264338327950288
+#endif
+
+
+// clang-format off
+const char *unary_fn_code_pattern =
+"%s\n" /* optional pragma */
+"__kernel void test_fn(__global %s%s *src, __global %s%s *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    dst[tid] = %s(src[tid]);\n"
+"}\n";
+
+const char *unary_fn_code_pattern_v3 =
+"%s\n" /* optional pragma */
+"__kernel void test_fn(__global %s *src, __global %s *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    vstore3(%s(vload3(tid,src)), tid, dst);\n"
+"}\n";
+// clang-format on
+
+
+#define MAX_ERR 2.0f
+
+namespace {
+
+
+template <typename T> float UlpFn(const T &val, const double &r)
+{
+    if (std::is_same<T, double>::value)
+        return Ulp_Error_Double(val, r);
+    else if (std::is_same<T, float>::value)
+        return Ulp_Error(val, r);
+    else if (std::is_same<T, half>::value)
+        return Ulp_Error(val, r);
+}
+
+
+template <typename T>
+int verify_degrees(const T *const inptr, const T *const outptr, int n)
+{
+    float error, max_error = 0.0f;
+    double r, max_val = NAN;
+    int max_index = 0;
+
+    for (int i = 0, j = 0; i < n; i++, j++)
+    {
+        r = (180.0 / M_PI) * inptr[i];
+
+        error = UlpFn(outptr[i], r);
+
+        if (fabsf(error) > max_error)
+        {
+            max_error = error;
+            max_index = i;
+            max_val = r;
+            if (fabsf(error) > MAX_ERR)
+            {
+                log_error("%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n",
+                          i, inptr[i], r, outptr[i], r, outptr[i], error);
+                return 1;
+            }
+        }
+    }
+
+    log_info("degrees: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n",
+             max_error, max_index, max_val, outptr[max_index], max_val,
+             outptr[max_index]);
+
+    return 0;
+}
+
+
+template <typename T>
+int verify_radians(const T *const inptr, const T *const outptr, int n)
+{
+    float error, max_error = 0.0f;
+    double r, max_val = NAN;
+    int max_index = 0;
+
+    for (int i = 0, j = 0; i < n; i++, j++)
+    {
+        r = (M_PI / 180.0) * inptr[i];
+        error = Ulp_Error(outptr[i], r);
+        if (fabsf(error) > max_error)
+        {
+            max_error = error;
+            max_index = i;
+            max_val = r;
+            if (fabsf(error) > MAX_ERR)
+            {
+                log_error("%d) Error @ %a: *%a vs %a  (*%g vs %g) ulps: %f\n",
+                          i, inptr[i], r, outptr[i], r, outptr[i], error);
+                return 1;
+            }
+        }
+    }
+
+    log_info("radians: Max error %f ulps at %d: *%a vs %a  (*%g vs %g)\n",
+             max_error, max_index, max_val, outptr[max_index], max_val,
+             outptr[max_index]);
+
+    return 0;
+}
+
+
+template <typename T>
+int verify_sign(const T *const inptr, const T *const outptr, int n)
+{
+    T r = 0;
+    for (int i = 0; i < n; i++)
+    {
+        if (inptr[i] > 0.0f)
+            r = 1.0;
+        else if (inptr[i] < 0.0f)
+            r = -1.0;
+        else
+            r = 0.0;
+        if (r != outptr[i]) return -1;
+    }
+    return 0;
+}
+
+}
+
+
+template <typename T>
+int test_unary_fn(cl_device_id device, cl_context context,
+                  cl_command_queue queue, int n_elems,
+                  const std::string &fnName, VerifyFuncUnary<T> verifyFn)
+{
+    clMemWrapper streams[2];
+    std::vector<T> input_ptr, output_ptr;
+
+    std::vector<clProgramWrapper> programs;
+    std::vector<clKernelWrapper> kernels;
+
+    int err, i;
+    MTdataHolder d = MTdataHolder(gRandomSeed);
+
+    assert(BaseFunctionTest::type2name.find(sizeof(T))
+           != BaseFunctionTest::type2name.end());
+    auto tname = BaseFunctionTest::type2name[sizeof(T)];
+
+    programs.resize(kTotalVecCount);
+    kernels.resize(kTotalVecCount);
+
+    int num_elements = n_elems * (1 << (kTotalVecCount - 1));
+
+    input_ptr.resize(num_elements);
+    output_ptr.resize(num_elements);
+
+    for (i = 0; i < 2; i++)
+    {
+        streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                    sizeof(T) * num_elements, NULL, &err);
+        test_error(err, "clCreateBuffer failed");
+    }
+
+    std::string pragma_str;
+    if (std::is_same<T, float>::value)
+    {
+        for (int j = 0; j < num_elements; j++)
+        {
+            input_ptr[j] = get_random_float((float)(-100000.f * M_PI),
+                                            (float)(100000.f * M_PI), d);
+        }
+    }
+    else if (std::is_same<T, double>::value)
+    {
+        pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
+        for (int j = 0; j < num_elements; j++)
+        {
+            input_ptr[j] =
+                get_random_double(-100000.0 * M_PI, 100000.0 * M_PI, d);
+        }
+    }
+
+    err = clEnqueueWriteBuffer(queue, streams[0], true, 0,
+                               sizeof(T) * num_elements, &input_ptr.front(), 0,
+                               NULL, NULL);
+    if (err != CL_SUCCESS)
+    {
+        log_error("clEnqueueWriteBuffer failed\n");
+        return -1;
+    }
+
+    for (i = 0; i < kTotalVecCount; i++)
+    {
+        std::string kernelSource;
+        char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
+
+        if (i >= kVectorSizeCount)
+        {
+            std::string str = unary_fn_code_pattern_v3;
+            kernelSource = string_format(str, pragma_str.c_str(), tname.c_str(),
+                                         tname.c_str(), fnName.c_str());
+        }
+        else
+        {
+            std::string str = unary_fn_code_pattern;
+            kernelSource = string_format(str, pragma_str.c_str(), tname.c_str(),
+                                         vecSizeNames[i], tname.c_str(),
+                                         vecSizeNames[i], fnName.c_str());
+        }
+
+        /* Create kernels */
+        const char *programPtr = kernelSource.c_str();
+        err =
+            create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
+                                        (const char **)&programPtr, "test_fn");
+
+        err = clSetKernelArg(kernels[i], 0, sizeof streams[0], &streams[0]);
+        err |= clSetKernelArg(kernels[i], 1, sizeof streams[1], &streams[1]);
+        if (err != CL_SUCCESS)
+        {
+            log_error("clSetKernelArgs failed\n");
+            return -1;
+        }
+
+        // Line below is troublesome...
+        size_t threads = (size_t)num_elements / ((g_arrVecSizes[i]));
+        err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
+                                     0, NULL, NULL);
+        if (err != CL_SUCCESS)
+        {
+            log_error("clEnqueueNDRangeKernel failed\n");
+            return -1;
+        }
+
+        cl_uint dead = 42;
+        memset_pattern4(&output_ptr[0], &dead, sizeof(T) * num_elements);
+        err = clEnqueueReadBuffer(queue, streams[1], true, 0,
+                                  sizeof(T) * num_elements, &output_ptr[0], 0,
+                                  NULL, NULL);
+        if (err != CL_SUCCESS)
+        {
+            log_error("clEnqueueReadBuffer failed\n");
+            return -1;
+        }
+
+        if (verifyFn((T *)&input_ptr.front(), (T *)&output_ptr.front(),
+                     n_elems * (i + 1)))
+        {
+            log_error("%s %s%d test failed\n", fnName.c_str(), tname.c_str(),
+                      ((g_arrVecSizes[i])));
+            err = -1;
+        }
+        else
+        {
+            log_info("%s %s%d test passed\n", fnName.c_str(), tname.c_str(),
+                     ((g_arrVecSizes[i])));
+        }
+
+        if (err) break;
+    }
+
+    return err;
+}
+
+
+cl_int DegreesTest::Run()
+{
+    cl_int error = test_unary_fn<float>(device, context, queue, num_elems,
+                                        fnName.c_str(), verify_degrees<float>);
+    test_error(error, "DegreesTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_unary_fn<double>(device, context, queue, num_elems,
+                                      fnName.c_str(), verify_degrees<double>);
+        test_error(error, "DegreesTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+cl_int RadiansTest::Run()
+{
+    cl_int error = test_unary_fn<float>(device, context, queue, num_elems,
+                                        fnName.c_str(), verify_radians<float>);
+    test_error(error, "RadiansTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_unary_fn<double>(device, context, queue, num_elems,
+                                      fnName.c_str(), verify_radians<double>);
+        test_error(error, "RadiansTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+cl_int SignTest::Run()
+{
+    cl_int error = test_unary_fn<float>(device, context, queue, num_elems,
+                                        fnName.c_str(), verify_sign<float>);
+    test_error(error, "SignTest::Run<float> failed");
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+    {
+        error = test_unary_fn<double>(device, context, queue, num_elems,
+                                      fnName.c_str(), verify_sign<double>);
+        test_error(error, "SignTest::Run<double> failed");
+    }
+
+    return error;
+}
+
+
+int test_degrees(cl_device_id device, cl_context context,
+                 cl_command_queue queue, int n_elems)
+{
+    return MakeAndRunTest<DegreesTest>(device, context, queue, n_elems,
+                                       "degrees");
+}
+
+
+int test_radians(cl_device_id device, cl_context context,
+                 cl_command_queue queue, int n_elems)
+{
+    return MakeAndRunTest<RadiansTest>(device, context, queue, n_elems,
+                                       "radians");
+}
+
+
+int test_sign(cl_device_id device, cl_context context, cl_command_queue queue,
+              int n_elems)
+{
+    return MakeAndRunTest<SignTest>(device, context, queue, n_elems, "sign");
+}

test_conformance/computeinfo/main.cpp

@@ -439,8 +439,8 @@ int getPlatformConfigInfo(cl_platform_id platform, config_info* info)
                 err = clGetPlatformInfo(platform, info->opcode, config_size_set,
                                         &info->config.cl_name_version_single,
                                         &config_size_ret);
-            }
                 size_err = config_size_set != config_size_ret;
+            }
             break;
         default:
             log_error("Unknown config type: %d\n", info->config_type);
@@ -585,8 +585,8 @@ int getConfigInfo(cl_device_id device, config_info* info)
                 err = clGetDeviceInfo(device, info->opcode, config_size_set,
                                       &info->config.cl_name_version_single,
                                       &config_size_ret);
-            }
                 size_err = config_size_set != config_size_ret;
+            }
             break;
         default:
             log_error("Unknown config type: %d\n", info->config_type);

test_conformance/conversions/CMakeLists.txt

@@ -16,4 +16,6 @@ set_source_files_properties(
         COMPILE_FLAGS -march=i686)
 endif(NOT CMAKE_CL_64 AND NOT MSVC AND NOT ANDROID)
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../CMakeCommon.txt)

test_conformance/d3d10/harness.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _HARNESS_H_
-#define _HARNESS_H_
+#ifndef HARNESS_H_
+#define HARNESS_H_
 
 #define _CRT_SECURE_NO_WARNINGS
 

test_conformance/device_timer/CMakeLists.txt

@@ -5,4 +5,6 @@ set(${MODULE_NAME}_SOURCES
     test_device_timer.cpp
 )
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../CMakeCommon.txt)

test_conformance/extensions/cl_khr_command_buffer/basic_command_buffer.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#ifndef _CL_KHR_BASIC_COMMAND_BUFFER_H
-#define _CL_KHR_BASIC_COMMAND_BUFFER_H
+#ifndef CL_KHR_BASIC_COMMAND_BUFFER_H
+#define CL_KHR_BASIC_COMMAND_BUFFER_H
 
 #include "command_buffer_test_base.h"
 #include "harness/typeWrappers.h"
@@ -28,7 +28,7 @@
     {                                                                          \
         if (reference != result)                                               \
         {                                                                      \
-            log_error("Expected %d was %d at index %u\n", reference, result,   \
+            log_error("Expected %d was %d at index %zu\n", reference, result,  \
                       index);                                                  \
             return TEST_FAIL;                                                  \
         }                                                                      \
@@ -99,4 +99,4 @@ int MakeAndRunTest(cl_device_id device, cl_context context,
     return TEST_PASS;
 }
 
-#endif // _CL_KHR_BASIC_COMMAND_BUFFER_H
+#endif // CL_KHR_BASIC_COMMAND_BUFFER_H

test_conformance/extensions/cl_khr_command_buffer/cl_khr_command_buffer_mutable_dispatch/mutable_command_basic.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#ifndef _CL_KHR_MUTABLE_COMMAND_BASIC_H
-#define _CL_KHR_MUTABLE_COMMAND_BASIC_H
+#ifndef CL_KHR_MUTABLE_COMMAND_BASIC_H
+#define CL_KHR_MUTABLE_COMMAND_BASIC_H
 
 #include "../basic_command_buffer.h"
 #include "../command_buffer_test_base.h"
@@ -104,4 +104,4 @@ struct BasicMutableCommandBufferTest : BasicCommandBufferTest
     const size_t global_work_size = 4 * sizeof(cl_int);
 };
 
-#endif //_CL_KHR_MUTABLE_COMMAND_BASIC_H
+#endif // CL_KHR_MUTABLE_COMMAND_BASIC_H

test_conformance/extensions/cl_khr_command_buffer/cl_khr_command_buffer_mutable_dispatch/procs.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H
-#define _CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H
+#ifndef CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H
+#define CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H
 
 #include <CL/cl.h>
 
@@ -59,4 +59,4 @@ extern int test_mutable_command_info_global_work_size(cl_device_id device,
                                                       cl_context context,
                                                       cl_command_queue queue,
                                                       int num_elements);
-#endif /*_CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H*/
+#endif // CL_KHR_COMMAND_BUFFER_MUTABLE_DISPATCH_PROCS_H

test_conformance/extensions/cl_khr_command_buffer/command_buffer_get_command_buffer_info.cpp

@@ -240,9 +240,10 @@ struct CommandBufferGetCommandBufferInfo : public BasicCommandBufferTest
         clEventWrapper trigger_event = clCreateUserEvent(context, &error);
         test_error(error, "clCreateUserEvent failed");
 
+        clEventWrapper execute_event;
         // enqueued command buffer blocked on user event
         error = clEnqueueCommandBufferKHR(0, nullptr, command_buffer, 1,
-                                          &trigger_event, nullptr);
+                                          &trigger_event, &execute_event);
         test_error(error, "clEnqueueCommandBufferKHR failed");
 
         // verify pending state
@@ -255,6 +256,13 @@ struct CommandBufferGetCommandBufferInfo : public BasicCommandBufferTest
 
         test_error(signal_error, "clSetUserEventStatus failed");
 
+        error = clWaitForEvents(1, &execute_event);
+        test_error(error, "Unable to wait for execute event");
+
+        // verify executable state
+        error = verify_state(CL_COMMAND_BUFFER_STATE_EXECUTABLE_KHR);
+        test_error(error, "verify_state failed");
+
         return CL_SUCCESS;
     }
 

test_conformance/extensions/cl_khr_command_buffer/command_buffer_test_base.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#ifndef _CL_KHR_COMMAND_BUFFER_TEST_BASE_H
-#define _CL_KHR_COMMAND_BUFFER_TEST_BASE_H
+#ifndef CL_KHR_COMMAND_BUFFER_TEST_BASE_H
+#define CL_KHR_COMMAND_BUFFER_TEST_BASE_H
 
 #include <CL/cl_ext.h>
 #include "harness/deviceInfo.h"
@@ -174,4 +174,4 @@ public:
     }
 
 
-#endif // _CL_KHR_COMMAND_BUFFER_TEST_BASE_H
+#endif // CL_KHR_COMMAND_BUFFER_TEST_BASE_H

test_conformance/extensions/cl_khr_command_buffer/procs.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _CL_KHR_COMMAND_BUFFER_PROCS_H
-#define _CL_KHR_COMMAND_BUFFER_PROCS_H
+#ifndef CL_KHR_COMMAND_BUFFER_PROCS_H
+#define CL_KHR_COMMAND_BUFFER_PROCS_H
 
 #include <CL/cl.h>
 
@@ -131,4 +131,4 @@ extern int test_event_info_reference_count(cl_device_id device,
                                            cl_command_queue queue,
                                            int num_elements);
 
-#endif /*_CL_KHR_COMMAND_BUFFER_PROCS_H*/
+#endif // CL_KHR_COMMAND_BUFFER_PROCS_H

test_conformance/extensions/cl_khr_external_semaphore/procs.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _CL_KHR_EXTERNAL_SEMAPHORE_PROCS_H
-#define _CL_KHR_EXTERNAL_SEMAPHORE_PROCS_H
+#ifndef CL_KHR_EXTERNAL_SEMAPHORE_PROCS_H
+#define CL_KHR_EXTERNAL_SEMAPHORE_PROCS_H
 
 #include <CL/cl.h>
 
@@ -79,4 +79,4 @@ extern int test_external_semaphores_invalid_command(cl_device_id deviceID,
                                                     cl_context context,
                                                     cl_command_queue queue,
                                                     int num_elements);
-#endif /* CL_KHR_EXTERNAL_SEMAPHORE */
+#endif // CL_KHR_EXTERNAL_SEMAPHORE_PROCS_H

test_conformance/extensions/cl_khr_semaphore/main.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -34,11 +34,7 @@ test_definition test_list[] = {
     ADD_TEST_VERSION(semaphores_multi_signal, Version(1, 2)),
     ADD_TEST_VERSION(semaphores_multi_wait, Version(1, 2)),
     ADD_TEST_VERSION(semaphores_queries, Version(1, 2)),
-    ADD_TEST_VERSION(semaphores_order_1, Version(1, 2)),
-    ADD_TEST_VERSION(semaphores_order_2, Version(1, 2)),
-    ADD_TEST_VERSION(semaphores_order_3, Version(1, 2)),
     ADD_TEST_VERSION(semaphores_import_export_fd, Version(1, 2)),
-    ADD_TEST_VERSION(semaphores_invalid_command, Version(1, 2)),
 };
 
 const int test_num = ARRAY_SIZE(test_list);

test_conformance/extensions/cl_khr_semaphore/procs.h

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2017 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -41,17 +41,7 @@ extern int test_semaphores_multi_wait(cl_device_id deviceID, cl_context context,
                                       cl_command_queue queue, int num_elements);
 extern int test_semaphores_queries(cl_device_id deviceID, cl_context context,
                                    cl_command_queue queue, int num_elements);
-extern int test_semaphores_order_1(cl_device_id deviceID, cl_context context,
-                                   cl_command_queue queue, int num_elements);
-extern int test_semaphores_order_2(cl_device_id deviceID, cl_context context,
-                                   cl_command_queue queue, int num_elements);
-extern int test_semaphores_order_3(cl_device_id deviceID, cl_context context,
-                                   cl_command_queue queue, int num_elements);
 extern int test_semaphores_import_export_fd(cl_device_id deviceID,
                                             cl_context context,
                                             cl_command_queue queue,
                                             int num_elements);
-extern int test_semaphores_invalid_command(cl_device_id deviceID,
-                                           cl_context context,
-                                           cl_command_queue queue,
-                                           int num_elements);

test_conformance/extensions/cl_khr_semaphore/test_semaphores.cpp

@@ -1,5 +1,5 @@
 //
-// Copyright (c) 2022 The Khronos Group Inc.
+// Copyright (c) 2023 The Khronos Group Inc.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -646,303 +646,6 @@ int test_semaphores_queries(cl_device_id deviceID, cl_context context,
     return TEST_PASS;
 }
 
-// Confirm that it is possible to enqueue a signal of wait and signal in any
-// order as soon as the submission order (after deferred dependencies) is
-// correct. Case: first one deferred wait, then one non deferred signal.
-int test_semaphores_order_1(cl_device_id deviceID, cl_context context,
-                            cl_command_queue defaultQueue, int num_elements)
-{
-    cl_int err;
-
-    if (!is_extension_available(deviceID, "cl_khr_semaphore"))
-    {
-        log_info("cl_khr_semaphore is not supported on this platoform. "
-                 "Skipping test.\n");
-        return TEST_SKIPPED_ITSELF;
-    }
-
-    // Obtain pointers to semaphore's API
-    GET_PFN(deviceID, clCreateSemaphoreWithPropertiesKHR);
-    GET_PFN(deviceID, clEnqueueSignalSemaphoresKHR);
-    GET_PFN(deviceID, clEnqueueWaitSemaphoresKHR);
-    GET_PFN(deviceID, clReleaseSemaphoreKHR);
-
-    // Create ooo queue
-    clCommandQueueWrapper queue = clCreateCommandQueue(
-        context, deviceID, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err);
-    test_error(err, "Could not create command queue");
-
-    // Create semaphore
-    cl_semaphore_properties_khr sema_props[] = {
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_KHR),
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_BINARY_KHR),
-        0
-    };
-    cl_semaphore_khr sema =
-        clCreateSemaphoreWithPropertiesKHR(context, sema_props, &err);
-    test_error(err, "Could not create semaphore");
-
-    // Create user event
-    clEventWrapper user_event = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    // Wait semaphore (dependency on user_event)
-    clEventWrapper wait_event;
-    err = clEnqueueWaitSemaphoresKHR(queue, 1, &sema, nullptr, 1, &user_event,
-                                     &wait_event);
-    test_error(err, "Could not wait semaphore");
-
-    // Signal semaphore
-    clEventWrapper signal_event;
-    err = clEnqueueSignalSemaphoresKHR(queue, 1, &sema, nullptr, 0, nullptr,
-                                       &signal_event);
-    test_error(err, "Could not signal semaphore");
-
-    // Flush and delay
-    err = clFlush(queue);
-    test_error(err, "Could not flush queue");
-    std::this_thread::sleep_for(std::chrono::seconds(FLUSH_DELAY_S));
-
-    // Ensure signal event is completed while wait event is not
-    test_assert_event_complete(signal_event);
-    test_assert_event_inprogress(wait_event);
-
-    // Complete user_event
-    err = clSetUserEventStatus(user_event, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Finish
-    err = clFinish(queue);
-    test_error(err, "Could not finish queue");
-
-    // Ensure all events are completed
-    test_assert_event_complete(signal_event);
-    test_assert_event_complete(wait_event);
-
-    // Release semaphore
-    err = clReleaseSemaphoreKHR(sema);
-    test_error(err, "Could not release semaphore");
-
-    return TEST_PASS;
-}
-
-// Confirm that it is possible to enqueue a signal of wait and signal in any
-// order as soon as the submission order (after deferred dependencies) is
-// correct. Case: first two deferred signals, then one deferred wait. Unblock
-// signal, then unblock wait. When wait completes, unblock the other signal.
-int test_semaphores_order_2(cl_device_id deviceID, cl_context context,
-                            cl_command_queue defaultQueue, int num_elements)
-{
-    cl_int err;
-
-    if (!is_extension_available(deviceID, "cl_khr_semaphore"))
-    {
-        log_info("cl_khr_semaphore is not supported on this platoform. "
-                 "Skipping test.\n");
-        return TEST_SKIPPED_ITSELF;
-    }
-
-    // Obtain pointers to semaphore's API
-    GET_PFN(deviceID, clCreateSemaphoreWithPropertiesKHR);
-    GET_PFN(deviceID, clEnqueueSignalSemaphoresKHR);
-    GET_PFN(deviceID, clEnqueueWaitSemaphoresKHR);
-    GET_PFN(deviceID, clReleaseSemaphoreKHR);
-
-    // Create ooo queue
-    clCommandQueueWrapper queue = clCreateCommandQueue(
-        context, deviceID, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err);
-    test_error(err, "Could not create command queue");
-
-    // Create semaphore
-    cl_semaphore_properties_khr sema_props[] = {
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_KHR),
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_BINARY_KHR),
-        0
-    };
-    cl_semaphore_khr sema =
-        clCreateSemaphoreWithPropertiesKHR(context, sema_props, &err);
-    test_error(err, "Could not create semaphore");
-
-    // Create user events
-    clEventWrapper user_event_1 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    clEventWrapper user_event_2 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    clEventWrapper user_event_3 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    // Signal semaphore (dependency on user_event_1)
-    clEventWrapper signal_1_event;
-    err = clEnqueueSignalSemaphoresKHR(queue, 1, &sema, nullptr, 1,
-                                       &user_event_1, &signal_1_event);
-    test_error(err, "Could not signal semaphore");
-
-    // Signal semaphore (dependency on user_event_2)
-    clEventWrapper signal_2_event;
-    err = clEnqueueSignalSemaphoresKHR(queue, 1, &sema, nullptr, 1,
-                                       &user_event_2, &signal_2_event);
-    test_error(err, "Could not signal semaphore");
-
-    // Wait semaphore (dependency on user_event_3)
-    clEventWrapper wait_event;
-    err = clEnqueueWaitSemaphoresKHR(queue, 1, &sema, nullptr, 1, &user_event_3,
-                                     &wait_event);
-    test_error(err, "Could not wait semaphore");
-
-    // Complete user_event_1
-    err = clSetUserEventStatus(user_event_1, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Complete user_event_3
-    err = clSetUserEventStatus(user_event_3, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Flush and delay
-    err = clFlush(queue);
-    test_error(err, "Could not flush queue");
-    std::this_thread::sleep_for(std::chrono::seconds(FLUSH_DELAY_S));
-
-    // Ensure all events are completed except for second signal
-    test_assert_event_complete(signal_1_event);
-    test_assert_event_inprogress(signal_2_event);
-    test_assert_event_complete(wait_event);
-
-    // Complete user_event_2
-    err = clSetUserEventStatus(user_event_2, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Finish
-    err = clFinish(queue);
-    test_error(err, "Could not finish queue");
-
-    // Ensure all events are completed
-    test_assert_event_complete(signal_1_event);
-    test_assert_event_complete(signal_2_event);
-    test_assert_event_complete(wait_event);
-
-    // Release semaphore
-    err = clReleaseSemaphoreKHR(sema);
-    test_error(err, "Could not release semaphore");
-
-    return TEST_PASS;
-}
-
-// Confirm that it is possible to enqueue a signal of wait and signal in any
-// order as soon as the submission order (after deferred dependencies) is
-// correct. Case: first two deferred signals, then two deferred waits. Unblock
-// one signal and one wait (both blocked by the same user event). When wait
-// completes, unblock the other signal. Then unblock the other wait.
-int test_semaphores_order_3(cl_device_id deviceID, cl_context context,
-                            cl_command_queue defaultQueue, int num_elements)
-{
-    cl_int err;
-
-    if (!is_extension_available(deviceID, "cl_khr_semaphore"))
-    {
-        log_info("cl_khr_semaphore is not supported on this platoform. "
-                 "Skipping test.\n");
-        return TEST_SKIPPED_ITSELF;
-    }
-
-    // Obtain pointers to semaphore's API
-    GET_PFN(deviceID, clCreateSemaphoreWithPropertiesKHR);
-    GET_PFN(deviceID, clEnqueueSignalSemaphoresKHR);
-    GET_PFN(deviceID, clEnqueueWaitSemaphoresKHR);
-    GET_PFN(deviceID, clReleaseSemaphoreKHR);
-
-    // Create ooo queue
-    clCommandQueueWrapper queue = clCreateCommandQueue(
-        context, deviceID, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err);
-    test_error(err, "Could not create command queue");
-
-    // Create semaphore
-    cl_semaphore_properties_khr sema_props[] = {
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_KHR),
-        static_cast<cl_semaphore_properties_khr>(CL_SEMAPHORE_TYPE_BINARY_KHR),
-        0
-    };
-    cl_semaphore_khr sema =
-        clCreateSemaphoreWithPropertiesKHR(context, sema_props, &err);
-    test_error(err, "Could not create semaphore");
-
-    // Create user events
-    clEventWrapper user_event_1 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    clEventWrapper user_event_2 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    clEventWrapper user_event_3 = clCreateUserEvent(context, &err);
-    test_error(err, "Could not create user event");
-
-    // Signal semaphore (dependency on user_event_1)
-    clEventWrapper signal_1_event;
-    err = clEnqueueSignalSemaphoresKHR(queue, 1, &sema, nullptr, 1,
-                                       &user_event_1, &signal_1_event);
-    test_error(err, "Could not signal semaphore");
-
-    // Signal semaphore (dependency on user_event_2)
-    clEventWrapper signal_2_event;
-    err = clEnqueueSignalSemaphoresKHR(queue, 1, &sema, nullptr, 1,
-                                       &user_event_2, &signal_2_event);
-    test_error(err, "Could not signal semaphore");
-
-    // Wait semaphore (dependency on user_event_3)
-    clEventWrapper wait_1_event;
-    err = clEnqueueWaitSemaphoresKHR(queue, 1, &sema, nullptr, 1, &user_event_3,
-                                     &wait_1_event);
-    test_error(err, "Could not wait semaphore");
-
-    // Wait semaphore (dependency on user_event_2)
-    clEventWrapper wait_2_event;
-    err = clEnqueueWaitSemaphoresKHR(queue, 1, &sema, nullptr, 1, &user_event_2,
-                                     &wait_2_event);
-    test_error(err, "Could not wait semaphore");
-
-    // Complete user_event_2
-    err = clSetUserEventStatus(user_event_2, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Flush and delay
-    err = clFlush(queue);
-    test_error(err, "Could not flush queue");
-    std::this_thread::sleep_for(std::chrono::seconds(FLUSH_DELAY_S));
-
-    // Ensure only second signal and second wait completed
-    cl_event event_list[] = { signal_2_event, wait_2_event };
-    err = clWaitForEvents(2, event_list);
-    test_error(err, "Could not wait for events");
-
-    test_assert_event_inprogress(signal_1_event);
-    test_assert_event_inprogress(wait_1_event);
-
-    // Complete user_event_1
-    err = clSetUserEventStatus(user_event_1, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Complete user_event_3
-    err = clSetUserEventStatus(user_event_3, CL_COMPLETE);
-    test_error(err, "Could not set user event to CL_COMPLETE");
-
-    // Finish
-    err = clFinish(queue);
-    test_error(err, "Could not finish queue");
-
-    // Ensure all events are completed
-    test_assert_event_complete(signal_1_event);
-    test_assert_event_complete(signal_2_event);
-    test_assert_event_complete(wait_1_event);
-    test_assert_event_complete(wait_2_event);
-
-    // Release semaphore
-    err = clReleaseSemaphoreKHR(sema);
-    test_error(err, "Could not release semaphore");
-
-    return TEST_PASS;
-}
-
 // Test it is possible to export a semaphore to a sync fd and import the same
 // sync fd to a new semaphore
 int test_semaphores_import_export_fd(cl_device_id deviceID, cl_context context,

test_conformance/images/clCopyImage/CMakeLists.txt

@@ -15,5 +15,7 @@ set(${MODULE_NAME}_SOURCES
     ../common.cpp
 )
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../../CMakeCommon.txt)
 

test_conformance/images/clCopyImage/test_copy_generic.cpp

@@ -289,12 +289,6 @@ cl_mem create_image( cl_context context, cl_command_queue queue, BufferOwningPtr
     return img;
 }
 
-// WARNING -- not thread safe
-BufferOwningPtr<char> srcData;
-BufferOwningPtr<char> dstData;
-BufferOwningPtr<char> srcHost;
-BufferOwningPtr<char> dstHost;
-
 int test_copy_image_generic( cl_context context, cl_command_queue queue, image_descriptor *srcImageInfo, image_descriptor *dstImageInfo,
                             const size_t sourcePos[], const size_t destPos[], const size_t regionSize[], MTdata d )
 {
@@ -302,6 +296,11 @@ int test_copy_image_generic( cl_context context, cl_command_queue queue, image_d
 
     clMemWrapper srcImage, dstImage;
 
+    BufferOwningPtr<char> srcData;
+    BufferOwningPtr<char> dstData;
+    BufferOwningPtr<char> srcHost;
+    BufferOwningPtr<char> dstHost;
+
     if( gDebugTrace )
         log_info( " ++ Entering inner test loop...\n" );
 

test_conformance/images/clCopyImage/test_loops.cpp

@@ -41,60 +41,52 @@ int test_image_type( cl_device_id device, cl_context context, cl_command_queue q
         }
     }
 
-    if( testMethod == k1D )
+    switch (testMethod)
     {
+        case k1D:
             name = "1D -> 1D";
             imageType = CL_MEM_OBJECT_IMAGE1D;
-    }
-    else if( testMethod == k2D )
-    {
+            break;
+        case k2D:
             name = "2D -> 2D";
             imageType = CL_MEM_OBJECT_IMAGE2D;
-    }
-    else if( testMethod == k3D )
-    {
+            break;
+        case k3D:
             name = "3D -> 3D";
             imageType = CL_MEM_OBJECT_IMAGE3D;
-    }
-    else if( testMethod == k1DArray )
-    {
+            break;
+        case k1DArray:
             name = "1D array -> 1D array";
             imageType = CL_MEM_OBJECT_IMAGE1D_ARRAY;
-    }
-    else if( testMethod == k2DArray )
-    {
+            break;
+        case k2DArray:
             name = "2D array -> 2D array";
             imageType = CL_MEM_OBJECT_IMAGE2D_ARRAY;
-    }
-    else if( testMethod == k2DTo3D )
-    {
+            break;
+        case k2DTo3D:
             name = "2D -> 3D";
             imageType = CL_MEM_OBJECT_IMAGE3D;
-    }
-    else if( testMethod == k3DTo2D )
-    {
+            break;
+        case k3DTo2D:
             name = "3D -> 2D";
             imageType = CL_MEM_OBJECT_IMAGE3D;
-    }
-    else if( testMethod == k2DArrayTo2D )
-    {
+            break;
+        case k2DArrayTo2D:
             name = "2D array -> 2D";
             imageType = CL_MEM_OBJECT_IMAGE2D_ARRAY;
-    }
-    else if( testMethod == k2DTo2DArray )
-    {
+            break;
+        case k2DTo2DArray:
             name = "2D -> 2D array";
             imageType = CL_MEM_OBJECT_IMAGE2D_ARRAY;
-    }
-    else if( testMethod == k2DArrayTo3D )
-    {
+            break;
+        case k2DArrayTo3D:
             name = "2D array -> 3D";
             imageType = CL_MEM_OBJECT_IMAGE3D;
-    }
-    else if( testMethod == k3DTo2DArray )
-    {
+            break;
+        case k3DTo2DArray:
             name = "3D -> 2D array";
             imageType = CL_MEM_OBJECT_IMAGE3D;
+            break;
     }
 
     if(gTestMipmaps)

test_conformance/images/clFillImage/test_loops.cpp

@@ -33,35 +33,34 @@ int test_image_type( cl_device_id device, cl_context context, cl_command_queue q
     cl_mem_object_type imageType;
     test_func test_fn;
 
-    if ( testMethod == k1D )
+    switch (testMethod)
     {
+        case k1D:
             name = "1D Image Fill";
             imageType = CL_MEM_OBJECT_IMAGE1D;
             test_fn = &test_fill_image_set_1D;
-    }
-    else if ( testMethod == k2D )
-    {
+            break;
+        case k2D:
             name = "2D Image Fill";
             imageType = CL_MEM_OBJECT_IMAGE2D;
             test_fn = &test_fill_image_set_2D;
-    }
-    else if ( testMethod == k1DArray )
-    {
+            break;
+        case k1DArray:
             name = "1D Image Array Fill";
             imageType = CL_MEM_OBJECT_IMAGE1D_ARRAY;
             test_fn = &test_fill_image_set_1D_array;
-    }
-    else if ( testMethod == k2DArray )
-    {
+            break;
+        case k2DArray:
             name = "2D Image Array Fill";
             imageType = CL_MEM_OBJECT_IMAGE2D_ARRAY;
             test_fn = &test_fill_image_set_2D_array;
-    }
-    else if ( testMethod == k3D )
-    {
+            break;
+        case k3D:
             name = "3D Image Fill";
             imageType = CL_MEM_OBJECT_IMAGE3D;
             test_fn = &test_fill_image_set_3D;
+            break;
+        default: log_error("Unhandled method\n"); return -1;
     }
 
     log_info( "Running %s tests...\n", name );

test_conformance/images/clReadWriteImage/CMakeLists.txt

@@ -11,5 +11,7 @@ set(${MODULE_NAME}_SOURCES
     ../common.cpp
 )
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../../CMakeCommon.txt)
 

test_conformance/images/kernel_read_write/CMakeLists.txt

@@ -21,7 +21,7 @@ set(${MODULE_NAME}_SOURCES
 
 # Make unused variables not fatal in this module; see
 # https://github.com/KhronosGroup/OpenCL-CTS/issues/1484
-set_gnulike_module_compile_flags("-Wno-error=unused-variable")
+set_gnulike_module_compile_flags("-Wno-error=unused-variable -Wno-unused-but-set-variable")
 
 include(../../CMakeCommon.txt)
 

test_conformance/images/kernel_read_write/test_cl_ext_image_buffer.hpp

@@ -14,8 +14,8 @@
 // limitations under the License.
 //
 
-#ifndef _TEST_CL_EXT_IMAGE_BUFFER
-#define _TEST_CL_EXT_IMAGE_BUFFER
+#ifndef TEST_CL_EXT_IMAGE_BUFFER
+#define TEST_CL_EXT_IMAGE_BUFFER
 
 #define TEST_IMAGE_SIZE 20
 
@@ -121,4 +121,4 @@ static inline void image_desc_init(cl_image_desc* desc,
     }
 }
 
-#endif /* _TEST_CL_EXT_IMAGE_BUFFER */
+#endif // TEST_CL_EXT_IMAGE_BUFFER

test_conformance/integer_ops/test_absdiff.cpp

@@ -22,6 +22,17 @@
 
 #include "procs.h"
 
+template <class Integer>
+static typename std::make_unsigned<Integer>::type abs_diff(Integer a, Integer b)
+{
+    using Unsigned = typename std::make_unsigned<Integer>::type;
+    Unsigned ua = a;
+    Unsigned ub = b;
+    Unsigned diff = ua - ub;
+    if (a < b) diff = -diff;
+    return diff;
+}
+
 static int verify_absdiff_char( const void *p, const void *q, const void *r, size_t n, const char *sizeName, size_t vecSize )
 {
     const cl_char *inA = (const cl_char *)p;
@@ -30,9 +41,7 @@ static int verify_absdiff_char( const void *p, const void *q, const void *r, siz
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_uchar r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_uchar r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (char%s) 0x%2.2x, (char%s) 0x%2.2x) = *0x%2.2x vs 0x%2.2x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -47,9 +56,7 @@ static int verify_absdiff_uchar( const void *p, const void *q, const void *r, si
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_uchar r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_uchar r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (uchar%s) 0x%2.2x, (uchar%s) 0x%2.2x) = *0x%2.2x vs 0x%2.2x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -64,9 +71,7 @@ static int verify_absdiff_short( const void *p, const void *q, const void *r, si
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_ushort r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_ushort r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (short%s) 0x%4.4x, (short%s) 0x%4.4x) = *0x%4.4x vs 0x%4.4x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -81,9 +86,7 @@ static int verify_absdiff_ushort( const void *p, const void *q, const void *r, s
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_ushort r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_ushort r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (ushort%s) 0x%4.4x, (ushort%s) 0x%4.4x) = *0x%4.4x vs 0x%4.4x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -98,9 +101,7 @@ static int verify_absdiff_int( const void *p, const void *q, const void *r, size
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_uint r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_uint r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         {
             log_info( "%ld) Failure for absdiff( (int%s) 0x%8.8x, (int%s) 0x%8.8x) = *0x%8.8x vs 0x%8.8x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] );
@@ -118,9 +119,7 @@ static int verify_absdiff_uint( const void *p, const void *q, const void *r, siz
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_uint r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_uint r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (uint%s) 0x%8.8x, (uint%s) 0x%8.8x) = *0x%8.8x vs 0x%8.8x\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -135,9 +134,7 @@ static int verify_absdiff_long( const void *p, const void *q, const void *r, siz
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_ulong r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_ulong r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (long%s) 0x%16.16llx, (long%s) 0x%16.16llx) = *0x%16.16llx vs 0x%16.16llx\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }
@@ -152,9 +149,7 @@ static int verify_absdiff_ulong( const void *p, const void *q, const void *r, si
     size_t i;
     for( i = 0; i < n; i++ )
     {
-        cl_ulong r = inA[i] - inB[i];
-        if( inB[i] > inA[i] )
-            r = inB[i] - inA[i];
+        cl_ulong r = abs_diff(inA[i], inB[i]);
         if( r != outptr[i] )
         { log_info( "%ld) Failure for absdiff( (ulong%s) 0x%16.16llx, (ulong%s) 0x%16.16llx) = *0x%16.16llx vs 0x%16.16llx\n", i, sizeName, inA[i], sizeName, inB[i], r, outptr[i] ); return -1; }
     }

test_conformance/math_brute_force/common.cpp

@@ -67,13 +67,16 @@ void EmitDefineUndef(std::ostringstream &kernel, const char *name,
     kernel << "#define " << name << " " << GetUndefValue(type) << '\n';
 }
 
-void EmitEnableExtension(std::ostringstream &kernel, ParameterType type)
+void EmitEnableExtension(std::ostringstream &kernel,
+                         const std::initializer_list<ParameterType> &types)
+{
+    bool needsFp64 = false;
+
+    for (const auto &type : types)
     {
         switch (type)
         {
-        case ParameterType::Double:
-            kernel << "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
-            break;
+            case ParameterType::Double: needsFp64 = true; break;
 
             case ParameterType::Float:
             case ParameterType::Int:
@@ -85,6 +88,9 @@ void EmitEnableExtension(std::ostringstream &kernel, ParameterType type)
         }
     }
 
+    if (needsFp64) kernel << "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
+}
+
 std::string GetBuildOptions(bool relaxed_mode)
 {
     std::ostringstream options;
@@ -123,7 +129,7 @@ std::string GetUnaryKernel(const std::string &kernel_name, const char *builtin,
     EmitDefineType(kernel, "RETTYPE", retType, vector_size_index);
     EmitDefineType(kernel, "TYPE1", type1, vector_size_index);
     EmitDefineUndef(kernel, "UNDEF1", type1);
-    EmitEnableExtension(kernel, type1);
+    EmitEnableExtension(kernel, { retType, type1 });
 
     // clang-format off
     const char *kernel_nonvec3[] = { R"(
@@ -199,7 +205,7 @@ std::string GetUnaryKernel(const std::string &kernel_name, const char *builtin,
     EmitDefineType(kernel, "TYPE1", type1, vector_size_index);
     EmitDefineUndef(kernel, "UNDEF1", type1);
     EmitDefineUndef(kernel, "UNDEFR2", retType2);
-    EmitEnableExtension(kernel, type1);
+    EmitEnableExtension(kernel, { retType1, retType2, type1 });
 
     // clang-format off
     const char *kernel_nonvec3[] = { R"(
@@ -282,7 +288,7 @@ std::string GetBinaryKernel(const std::string &kernel_name, const char *builtin,
     EmitDefineType(kernel, "TYPE2", type2, vector_size_index);
     EmitDefineUndef(kernel, "UNDEF1", type1);
     EmitDefineUndef(kernel, "UNDEF2", type2);
-    EmitEnableExtension(kernel, type1);
+    EmitEnableExtension(kernel, { retType, type1, type2 });
 
     const bool is_vec3 = sizeValues[vector_size_index] == 3;
 
@@ -384,7 +390,7 @@ std::string GetBinaryKernel(const std::string &kernel_name, const char *builtin,
     EmitDefineUndef(kernel, "UNDEF1", type1);
     EmitDefineUndef(kernel, "UNDEF2", type2);
     EmitDefineUndef(kernel, "UNDEFR2", retType2);
-    EmitEnableExtension(kernel, type1);
+    EmitEnableExtension(kernel, { retType1, retType2, type1, type2 });
 
     // clang-format off
     const char *kernel_nonvec3[] = { R"(
@@ -476,7 +482,7 @@ std::string GetTernaryKernel(const std::string &kernel_name,
     EmitDefineUndef(kernel, "UNDEF1", type1);
     EmitDefineUndef(kernel, "UNDEF2", type2);
     EmitDefineUndef(kernel, "UNDEF3", type3);
-    EmitEnableExtension(kernel, type1);
+    EmitEnableExtension(kernel, { retType, type1, type2, type3 });
 
     // clang-format off
     const char *kernel_nonvec3[] = { R"(
@@ -585,14 +591,6 @@ cl_int BuildKernels(BuildKernelInfo &info, cl_uint job_id,
         if (!kernel || error != CL_SUCCESS)
         {
             vlog_error("\t\tFAILED -- clCreateKernel() failed: (%d)\n", error);
-            size_t log_size;
-            clGetProgramBuildInfo(program, gDevice, CL_PROGRAM_BUILD_LOG, 0,
-                                  nullptr, &log_size);
-            std::string buffer;
-            buffer.resize(log_size + 1);
-            clGetProgramBuildInfo(program, gDevice, CL_PROGRAM_BUILD_LOG,
-                                  log_size, &buffer[0], NULL);
-            vlog_error("Log: %s\n", buffer.c_str());
             return error;
         }
     }

test_conformance/mem_host_flags/CMakeLists.txt

@@ -6,4 +6,6 @@ set(${MODULE_NAME}_SOURCES
     mem_host_image.cpp
 )
 
+set_gnulike_module_compile_flags("-Wno-unused-but-set-variable")
+
 include(../CMakeCommon.txt)

test_conformance/mem_host_flags/C_host_memory_block.h

@@ -24,14 +24,14 @@
 
 template <class T> class C_host_memory_block {
 public:
-    int num_elements;
+    size_t num_elements;
     int element_size;
     T *pData;
 
     C_host_memory_block();
     ~C_host_memory_block();
-    void Init(int num_elem, T &value);
-    void Init(int num_elem);
+    void Init(size_t num_elem, T &value);
+    void Init(size_t num_elem);
     void Set_to(T &val);
     void Set_to_zero();
     bool Equal_to(T &val);
@@ -40,7 +40,7 @@ public:
     bool Equal_rect(C_host_memory_block<T> &another, size_t *host_origin,
                     size_t *region, size_t host_row_pitch,
                     size_t host_slice_pitch);
-    bool Equal(T *pData, int num_elements);
+    bool Equal(T *pData, size_t num_elements);
 
     bool Equal_rect_from_orig(C_host_memory_block<T> &another, size_t *soffset,
                               size_t *region, size_t host_row_pitch,
@@ -63,20 +63,20 @@ template <class T> C_host_memory_block<T>::~C_host_memory_block()
     num_elements = 0;
 }
 
-template <class T> void C_host_memory_block<T>::Init(int num_elem, T &value)
+template <class T> void C_host_memory_block<T>::Init(size_t num_elem, T &value)
 {
     if (pData != NULL) delete[] pData;
     pData = new T[num_elem];
-    for (int i = 0; i < num_elem; i++) pData[i] = value;
+    for (size_t i = 0; i < num_elem; i++) pData[i] = value;
 
     num_elements = num_elem;
 }
 
-template <class T> void C_host_memory_block<T>::Init(int num_elem)
+template <class T> void C_host_memory_block<T>::Init(size_t num_elem)
 {
     if (pData != NULL) delete[] pData;
     pData = new T[num_elem];
-    for (int i = 0; i < num_elem; i++) pData[i] = (T)i;
+    for (size_t i = 0; i < num_elem; i++) pData[i] = (T)i;
 
     num_elements = num_elem;
 }
@@ -88,14 +88,14 @@ template <class T> void C_host_memory_block<T>::Set_to_zero()
 
 template <class T> void C_host_memory_block<T>::Set_to(T &val)
 {
-    for (int i = 0; i < num_elements; i++) pData[i] = val;
+    for (size_t i = 0; i < num_elements; i++) pData[i] = val;
 }
 
 template <class T> bool C_host_memory_block<T>::Equal_to(T &val)
 {
-    int count = 0;
+    size_t count = 0;
 
-    for (int i = 0; i < num_elements; i++)
+    for (size_t i = 0; i < num_elements; i++)
     {
         if (pData[i] == val) count++;
     }
@@ -106,9 +106,9 @@ template <class T> bool C_host_memory_block<T>::Equal_to(T &val)
 template <class T>
 bool C_host_memory_block<T>::Equal(C_host_memory_block<T> &another)
 {
-    int count = 0;
+    size_t count = 0;
 
-    for (int i = 0; i < num_elements; i++)
+    for (size_t i = 0; i < num_elements; i++)
     {
         if (pData[i] == another.pData[i]) count++;
     }
@@ -117,13 +117,13 @@ bool C_host_memory_block<T>::Equal(C_host_memory_block<T> &another)
 }
 
 template <class T>
-bool C_host_memory_block<T>::Equal(T *pIn_Data, int Innum_elements)
+bool C_host_memory_block<T>::Equal(T *pIn_Data, size_t Innum_elements)
 {
     if (this->num_elements != Innum_elements) return false;
 
-    int count = 0;
+    size_t count = 0;
 
-    for (int i = 0; i < num_elements; i++)
+    for (size_t i = 0; i < num_elements; i++)
     {
         if (pData[i] == pIn_Data[i]) count++;
     }
@@ -134,7 +134,7 @@ bool C_host_memory_block<T>::Equal(T *pIn_Data, int Innum_elements)
 template <class T> size_t C_host_memory_block<T>::Count(T &val)
 {
     size_t count = 0;
-    for (int i = 0; i < num_elements; i++)
+    for (size_t i = 0; i < num_elements; i++)
     {
         if (pData[i] == val) count++;
     }

test_conformance/mem_host_flags/checker.h

@@ -219,7 +219,7 @@ cl_int cBuffer_checker<T>::SetupASSubBuffer(cl_mem_flags parent_buffer_flag)
         err = CL_SUCCESS;
     }
 
-    cl_mem_flags f;
+    cl_mem_flags f = 0;
     if (parent_buffer_flag & CL_MEM_HOST_READ_ONLY)
         f = CL_MEM_HOST_READ_ONLY;
     else if (parent_buffer_flag & CL_MEM_HOST_WRITE_ONLY)

test_conformance/non_uniform_work_group/TestNonUniformWorkGroup.cpp

@@ -448,13 +448,8 @@ void TestNonUniformWorkGroup::verifyData (DataContainerAttrib * reference, DataC
 }
 
 void TestNonUniformWorkGroup::calculateExpectedValues () {
-  size_t nonRemainderGlobalSize[MAX_DIMS];
   size_t numberOfPossibleRegions[MAX_DIMS];
 
-  nonRemainderGlobalSize[0] = _globalSize[0] - (_globalSize[0] % _enqueuedLocalSize[0]);
-  nonRemainderGlobalSize[1] = _globalSize[1] - (_globalSize[1] % _enqueuedLocalSize[1]);
-  nonRemainderGlobalSize[2] = _globalSize[2] - (_globalSize[2] % _enqueuedLocalSize[2]);
-
   numberOfPossibleRegions[0] = (_globalSize[0]>1)?2:1;
   numberOfPossibleRegions[1] = (_globalSize[1]>1)?2:1;
   numberOfPossibleRegions[2] = (_globalSize[2]>1)?2:1;
@@ -502,6 +497,11 @@ size_t TestNonUniformWorkGroup::getMaxLocalWorkgroupSize (const cl_device_id &de
   if (TestNonUniformWorkGroup::_maxLocalWorkgroupSize == 0) {
     err = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_GROUP_SIZE,
       sizeof(TestNonUniformWorkGroup::_maxLocalWorkgroupSize), &TestNonUniformWorkGroup::_maxLocalWorkgroupSize, NULL);
+    if (err)
+    {
+        log_error("clGetDeviceInfo failed\n");
+        return 0;
+    }
   }
 
   return TestNonUniformWorkGroup::_maxLocalWorkgroupSize;

test_conformance/non_uniform_work_group/TestNonUniformWorkGroup.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _TESTNONUNIFORMWORKGROUP_H
-#define _TESTNONUNIFORMWORKGROUP_H
+#ifndef TESTNONUNIFORMWORKGROUP_H
+#define TESTNONUNIFORMWORKGROUP_H
 
 #include "procs.h"
 #include <vector>
@@ -147,5 +147,4 @@ private:
   unsigned int _overallCounter;
 };
 
-#endif // _TESTNONUNIFORMWORKGROUP_H
-
+#endif // TESTNONUNIFORMWORKGROUP_H

test_conformance/non_uniform_work_group/tools.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _TOOLS_H
-#define _TOOLS_H
+#ifndef TOOLS_H
+#define TOOLS_H
 
 #include "procs.h"
 #include <vector>
@@ -106,4 +106,4 @@ namespace Error {
   };
 
 }
-#endif // _TOOLS_H
+#endif // TOOLS_H

test_conformance/pipes/kernels.h

@@ -13,8 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 //
-#ifndef _KERNELS_H_
-#define _KERNELS_H_
+#ifndef KERNELS_H_
+#define KERNELS_H_
 
 static const char* pipe_readwrite_struct_kernel_code = {
     "typedef struct{\n"
@@ -127,4 +127,4 @@ static const char* pipe_convenience_readwrite_struct_kernel_code = {
     "    read_pipe(in_pipe, &dst[gid]);\n"
     "}\n" };
 
-#endif //_KERNELS_H_
+#endif // KERNELS_H_

test_conformance/printf/test_printf.cpp

@@ -268,7 +268,7 @@ static cl_program makePrintfProgram(cl_kernel *kernel_ptr, const cl_context cont
     };
 
     //Update testname
-    sprintf(testname,"%s%d","test",testId);
+    std::snprintf(testname, sizeof(testname), "%s%d", "test", testId);
 
     if (allTestCase[testId]->_type == TYPE_HALF
         || allTestCase[testId]->_type == TYPE_HALF_LIMITS)
@@ -278,13 +278,18 @@ static cl_program makePrintfProgram(cl_kernel *kernel_ptr, const cl_context cont
     //Update addrSpaceArgument and addrSpacePAddArgument types, based on FULL_PROFILE/EMBEDDED_PROFILE
     if(allTestCase[testId]->_type == TYPE_ADDRESS_SPACE)
     {
-        sprintf(addrSpaceArgument, "%s",allTestCase[testId]->_genParameters[testNum].addrSpaceArgumentTypeQualifier);
+        std::snprintf(addrSpaceArgument, sizeof(addrSpaceArgument), "%s",
+                      allTestCase[testId]
+                          ->_genParameters[testNum]
+                          .addrSpaceArgumentTypeQualifier);
 
-        sprintf(addrSpacePAddArgument, "%s", allTestCase[testId]->_genParameters[testNum].addrSpacePAdd);
+        std::snprintf(
+            addrSpacePAddArgument, sizeof(addrSpacePAddArgument), "%s",
+            allTestCase[testId]->_genParameters[testNum].addrSpacePAdd);
     }
 
     if (strlen(addrSpaceArgument) == 0)
-        sprintf(addrSpaceArgument,"void");
+        std::snprintf(addrSpaceArgument, sizeof(addrSpaceArgument), "void");
 
     // create program based on its type
 

test_conformance/relationals/CMakeLists.txt

@@ -3,8 +3,7 @@ set(MODULE_NAME RELATIONALS)
 set(${MODULE_NAME}_SOURCES
     main.cpp
     test_relationals.cpp
-    test_comparisons_float.cpp
-    test_comparisons_double.cpp
+    test_comparisons_fp.cpp
     test_shuffles.cpp
 )
 

test_conformance/relationals/test_comparisons_double.cpp

@@ -1,363 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "testBase.h"
-#include "harness/conversions.h"
-#include "harness/typeWrappers.h"
-
-#define TEST_SIZE 512
-
-const char *equivTestKernelPattern_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void sample_test(__global double%s *sourceA, __global double%s *sourceB, __global long%s *destValues, __global long%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
-"    destValuesB[tid] = sourceA[tid] %s sourceB[tid];\n"
-"\n"
-"}\n";
-
-const char *equivTestKernelPatternLessGreater_double =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void sample_test(__global double%s *sourceA, __global double%s *sourceB, __global long%s *destValues, __global long%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
-"    destValuesB[tid] = (sourceA[tid] < sourceB[tid]) | (sourceA[tid] > sourceB[tid]);\n"
-"\n"
-"}\n";
-
-
-const char *equivTestKernelPattern_double3 =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void sample_test(__global double%s *sourceA, __global double%s *sourceB, __global long%s *destValues, __global long%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    double3 sampA = vload3(tid, (__global double *)sourceA);\n"
-"    double3 sampB = vload3(tid, (__global double *)sourceB);\n"
-"    vstore3(%s( sampA, sampB ), tid, (__global long *)destValues);\n"
-"    vstore3(( sampA %s sampB ), tid, (__global long *)destValuesB);\n"
-"\n"
-"}\n";
-
-const char *equivTestKernelPatternLessGreater_double3 =
-"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
-"__kernel void sample_test(__global double%s *sourceA, __global double%s *sourceB, __global long%s *destValues, __global long%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    double3 sampA = vload3(tid, (__global double *)sourceA);\n"
-"    double3 sampB = vload3(tid, (__global double *)sourceB);\n"
-"    vstore3(%s( sampA, sampB ), tid, (__global long *)destValues);\n"
-"    vstore3(( sampA < sampB ) | (sampA > sampB), tid, (__global long *)destValuesB);\n"
-"\n"
-"}\n";
-
-
-typedef bool (*equivVerifyFn)( double inDataA, double inDataB );
-
-void verify_equiv_values_double( unsigned int vecSize, double *inDataA, double *inDataB, cl_long *outData, equivVerifyFn verifyFn )
-{
-    unsigned int i;
-    cl_long trueResult;
-    bool result;
-
-    trueResult = ( vecSize == 1 ) ? 1 : -1;
-    for( i = 0; i < vecSize; i++ )
-    {
-        result = verifyFn( inDataA[ i ], inDataB[ i ] );
-        outData[ i ] = result ? trueResult : 0;
-    }
-}
-
-void generate_equiv_test_data_double( double *outData, unsigned int vecSize, bool alpha, MTdata d )
-{
-    unsigned int i;
-
-    generate_random_data( kDouble, vecSize * TEST_SIZE, d, outData );
-
-    // Fill the first few vectors with NAN in each vector element (or the second set if we're alpha, so we can test either case)
-    if( alpha )
-        outData += vecSize * vecSize;
-    for( i = 0; i < vecSize; i++ )
-    {
-        outData[ 0 ] = NAN;
-        outData += vecSize + 1;
-    }
-    // Make sure the third set is filled regardless, to test the case where both have NANs
-    if( !alpha )
-        outData += vecSize * vecSize;
-    for( i = 0; i < vecSize; i++ )
-    {
-        outData[ 0 ] = NAN;
-        outData += vecSize + 1;
-    }
-}
-
-int test_equiv_kernel_double(cl_context context, cl_command_queue queue, const char *fnName, const char *opName,
-                             unsigned int vecSize, equivVerifyFn verifyFn, MTdata d )
-{
-    clProgramWrapper program;
-    clKernelWrapper kernel;
-    clMemWrapper streams[4];
-    double inDataA[TEST_SIZE * 16], inDataB[ TEST_SIZE * 16 ];
-    cl_long outData[TEST_SIZE * 16], expected[16];
-    int error, i, j;
-    size_t threads[1], localThreads[1];
-    char kernelSource[10240];
-    char *programPtr;
-    char sizeName[4];
-
-
-    /* Create the source */
-    if( vecSize == 1 )
-        sizeName[ 0 ] = 0;
-    else
-        sprintf( sizeName, "%d", vecSize );
-
-    if(DENSE_PACK_VECS && vecSize == 3) {
-        if (strcmp(fnName, "islessgreater")) {
-            sprintf( kernelSource, equivTestKernelPattern_double3, sizeName, sizeName, sizeName, sizeName, fnName, opName );
-        } else {
-            sprintf( kernelSource, equivTestKernelPatternLessGreater_double3, sizeName, sizeName, sizeName, sizeName, fnName );
-        }
-    } else {
-        if (strcmp(fnName, "islessgreater")) {
-            sprintf( kernelSource, equivTestKernelPattern_double, sizeName, sizeName, sizeName, sizeName, fnName, opName );
-        } else {
-            sprintf( kernelSource, equivTestKernelPatternLessGreater_double, sizeName, sizeName, sizeName, sizeName, fnName );
-        }
-    }
-
-    /* Create kernels */
-    programPtr = kernelSource;
-    if( create_single_kernel_helper( context, &program, &kernel, 1, (const char **)&programPtr, "sample_test" ) )
-    {
-        return -1;
-    }
-
-    /* Generate some streams */
-    generate_equiv_test_data_double( inDataA, vecSize, true, d );
-    generate_equiv_test_data_double( inDataB, vecSize, false, d );
-
-    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
-                                sizeof(cl_double) * vecSize * TEST_SIZE,
-                                &inDataA, &error);
-    if( streams[0] == NULL )
-    {
-        print_error( error, "Creating input array A failed!\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
-                                sizeof(cl_double) * vecSize * TEST_SIZE,
-                                &inDataB, &error);
-    if( streams[1] == NULL )
-    {
-        print_error( error, "Creating input array A failed!\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer( context, CL_MEM_READ_WRITE, sizeof( cl_long ) * vecSize * TEST_SIZE, NULL, &error);
-    if( streams[2] == NULL )
-    {
-        print_error( error, "Creating output array failed!\n");
-        return -1;
-    }
-    streams[3] = clCreateBuffer( context, CL_MEM_READ_WRITE, sizeof( cl_long ) * vecSize * TEST_SIZE, NULL, &error);
-    if( streams[3] == NULL )
-    {
-        print_error( error, "Creating output array failed!\n");
-        return -1;
-    }
-
-
-    /* Assign streams and execute */
-    error = clSetKernelArg( kernel, 0, sizeof( streams[0] ), &streams[0] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-    error = clSetKernelArg( kernel, 1, sizeof( streams[1] ), &streams[1] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-    error = clSetKernelArg( kernel, 2, sizeof( streams[2] ), &streams[2] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-    error = clSetKernelArg( kernel, 3, sizeof( streams[3] ), &streams[3] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-
-
-    /* Run the kernel */
-    threads[0] = TEST_SIZE;
-
-    error = get_max_common_work_group_size( context, kernel, threads[0], &localThreads[0] );
-    test_error( error, "Unable to get work group size to use" );
-
-    error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, localThreads, 0, NULL, NULL );
-    test_error( error, "Unable to execute test kernel" );
-
-    /* Now get the results */
-    error = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof( cl_long ) * TEST_SIZE * vecSize, outData, 0, NULL, NULL );
-    test_error( error, "Unable to read output array!" );
-
-    /* And verify! */
-    for( i = 0; i < TEST_SIZE; i++ )
-    {
-        verify_equiv_values_double( vecSize, &inDataA[ i * vecSize ], &inDataB[ i * vecSize ], expected, verifyFn);
-
-        for( j = 0; j < (int)vecSize; j++ )
-        {
-            if( expected[ j ] != outData[ i * vecSize + j ] )
-            {
-                log_error( "ERROR: Data sample %d:%d at size %d does not validate! Expected %lld, got %lld, source %f,%f\n",
-                          i, j, vecSize, expected[ j ], outData[ i * vecSize + j ], inDataA[i*vecSize + j], inDataB[i*vecSize + j] );
-                return -1;
-            }
-        }
-    }
-
-    /* Now get the results */
-    error = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof( cl_long ) * TEST_SIZE * vecSize, outData, 0, NULL, NULL );
-    test_error( error, "Unable to read output array!" );
-
-    /* And verify! */
-    for( i = 0; i < TEST_SIZE; i++ )
-    {
-        verify_equiv_values_double( vecSize, &inDataA[ i * vecSize ], &inDataB[ i * vecSize ], expected, verifyFn);
-
-        for( j = 0; j < (int)vecSize; j++ )
-        {
-            if( expected[ j ] != outData[ i * vecSize + j ] )
-            {
-                log_error( "ERROR: Data sample %d:%d at size %d does not validate! Expected %lld, got %lld, source %f,%f\n",
-                          i, j, vecSize, expected[ j ], outData[ i * vecSize + j ], inDataA[i*vecSize + j], inDataB[i*vecSize + j] );
-                return -1;
-            }
-        }
-    }
-
-    return 0;
-}
-
-int test_equiv_kernel_set_double(cl_device_id device, cl_context context, cl_command_queue queue, const char *fnName, const char *opName, equivVerifyFn verifyFn, MTdata d )
-{
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
-    unsigned int index;
-    int retVal = 0;
-
-    if (!is_extension_available(device, "cl_khr_fp64")) {
-        log_info("Extension cl_khr_fp64 not supported; skipping double tests.\n");
-        return 0;
-    }
-    log_info("Testing doubles.\n");
-
-    for( index = 0; vecSizes[ index ] != 0; index++ )
-    {
-        // Test!
-        if( test_equiv_kernel_double(context, queue, fnName, opName, vecSizes[ index ], verifyFn, d ) != 0 )
-        {
-            log_error( "   Vector double%d FAILED\n", vecSizes[ index ] );
-            retVal = -1;
-        }
-    }
-
-    return retVal;
-}
-
-bool isequal_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return valueA == valueB;
-}
-
-int test_relational_isequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "isequal", "==", isequal_verify_fn_double, seed );
-}
-
-bool isnotequal_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return true;
-    return valueA != valueB;
-}
-
-int test_relational_isnotequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "isnotequal", "!=", isnotequal_verify_fn_double, seed );
-}
-
-bool isgreater_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return valueA > valueB;
-}
-
-int test_relational_isgreater_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "isgreater", ">", isgreater_verify_fn_double, seed );
-}
-
-bool isgreaterequal_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return valueA >= valueB;
-}
-
-int test_relational_isgreaterequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "isgreaterequal", ">=", isgreaterequal_verify_fn_double, seed );
-}
-
-bool isless_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return valueA < valueB;
-}
-
-int test_relational_isless_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "isless", "<", isless_verify_fn_double, seed );
-}
-
-bool islessequal_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return valueA <= valueB;
-}
-
-int test_relational_islessequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "islessequal", "<=", islessequal_verify_fn_double, seed );
-}
-
-bool islessgreater_verify_fn_double( double valueA, double valueB )
-{
-    if( isnan( valueA ) || isnan( valueB ) )
-        return false;
-    return ( valueA < valueB ) || ( valueA > valueB );
-}
-
-int test_relational_islessgreater_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed(gRandomSeed);
-    return test_equiv_kernel_set_double( device, context, queue, "islessgreater", "<>", islessgreater_verify_fn_double, seed );
-}
-
-

test_conformance/relationals/test_comparisons_float.cpp

@@ -1,362 +0,0 @@
-//
-// Copyright (c) 2017 The Khronos Group Inc.
-// 
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//    http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-#include "testBase.h"
-#include "harness/conversions.h"
-#include "harness/typeWrappers.h"
-
-#define TEST_SIZE 512
-
-const char *equivTestKernelPattern_float =
-"__kernel void sample_test(__global float%s *sourceA, __global float%s *sourceB, __global int%s *destValues, __global int%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
-"    destValuesB[tid] = sourceA[tid] %s sourceB[tid];\n"
-"\n"
-"}\n";
-
-const char *equivTestKernelPatternLessGreater_float =
-"__kernel void sample_test(__global float%s *sourceA, __global float%s *sourceB, __global int%s *destValues, __global int%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
-"    destValuesB[tid] = (sourceA[tid] < sourceB[tid]) | (sourceA[tid] > sourceB[tid]);\n"
-"\n"
-"}\n";
-
-
-const char *equivTestKernelPattern_float3 =
-"__kernel void sample_test(__global float%s *sourceA, __global float%s *sourceB, __global int%s *destValues, __global int%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    float3 sampA = vload3(tid, (__global float *)sourceA);\n"
-"    float3 sampB = vload3(tid, (__global float *)sourceB);\n"
-"    vstore3(%s( sampA, sampB ), tid, (__global int *)destValues);\n"
-"    vstore3(( sampA %s sampB ), tid, (__global int *)destValuesB);\n"
-"\n"
-"}\n";
-
-const char *equivTestKernelPatternLessGreater_float3 =
-"__kernel void sample_test(__global float%s *sourceA, __global float%s *sourceB, __global int%s *destValues, __global int%s *destValuesB)\n"
-"{\n"
-"    int  tid = get_global_id(0);\n"
-"    float3 sampA = vload3(tid, (__global float *)sourceA);\n"
-"    float3 sampB = vload3(tid, (__global float *)sourceB);\n"
-"    vstore3(%s( sampA, sampB ), tid, (__global int *)destValues);\n"
-"    vstore3(( sampA < sampB ) | (sampA > sampB), tid, (__global int *)destValuesB);\n"
-"\n"
-"}\n";
-
-typedef bool (*equivVerifyFn)( float inDataA, float inDataB );
-
-int IsFloatInfinity(float x)
-{
-    return isinf(x);
-}
-
-int IsFloatNaN(float x)
-{
-    return isnan(x);
-}
-
-void verify_equiv_values_float( unsigned int vecSize, float *inDataA, float *inDataB, int *outData, equivVerifyFn verifyFn )
-{
-    unsigned int i;
-    int trueResult;
-    bool result;
-
-    trueResult = ( vecSize == 1 ) ? 1 : -1;
-    for( i = 0; i < vecSize; i++ )
-    {
-        result = verifyFn( inDataA[ i ], inDataB[ i ] );
-        outData[ i ] = result ? trueResult : 0;
-    }
-}
-
-void generate_equiv_test_data_float( float *outData, unsigned int vecSize, bool alpha, MTdata d )
-{
-    unsigned int i;
-
-    generate_random_data( kFloat, vecSize * TEST_SIZE, d, outData );
-
-    // Fill the first few vectors with NAN in each vector element (or the second set if we're alpha, so we can test either case)
-    if( alpha )
-        outData += vecSize * vecSize;
-    for( i = 0; i < vecSize; i++ )
-    {
-        outData[ 0 ] = NAN;
-        outData += vecSize + 1;
-    }
-    // Make sure the third set is filled regardless, to test the case where both have NANs
-    if( !alpha )
-        outData += vecSize * vecSize;
-    for( i = 0; i < vecSize; i++ )
-    {
-        outData[ 0 ] = NAN;
-        outData += vecSize + 1;
-    }
-}
-
-int test_equiv_kernel_float(cl_context context, cl_command_queue queue, const char *fnName, const char *opName,
-                       unsigned int vecSize, equivVerifyFn verifyFn, MTdata d )
-{
-    clProgramWrapper program;
-    clKernelWrapper kernel;
-    clMemWrapper streams[4];
-    float inDataA[TEST_SIZE * 16], inDataB[ TEST_SIZE * 16 ];
-    int outData[TEST_SIZE * 16], expected[16];
-    int error, i, j;
-    size_t threads[1], localThreads[1];
-    char kernelSource[10240];
-    char *programPtr;
-    char sizeName[4];
-
-
-    /* Create the source */
-    if( vecSize == 1 )
-        sizeName[ 0 ] = 0;
-    else
-        sprintf( sizeName, "%d", vecSize );
-
-
-    if(DENSE_PACK_VECS && vecSize == 3) {
-  if (strcmp(fnName, "islessgreater")) {
-            sprintf( kernelSource, equivTestKernelPattern_float3, sizeName, sizeName, sizeName, sizeName, fnName, opName );
-        } else {
-            sprintf( kernelSource, equivTestKernelPatternLessGreater_float3, sizeName, sizeName, sizeName, sizeName, fnName );
-        }
-    } else {
-        if (strcmp(fnName, "islessgreater")) {
-          sprintf( kernelSource, equivTestKernelPattern_float, sizeName, sizeName, sizeName, sizeName, fnName, opName );
-  } else {
-    sprintf( kernelSource, equivTestKernelPatternLessGreater_float, sizeName, sizeName, sizeName, sizeName, fnName );
-  }
-    }
-
-    /* Create kernels */
-    programPtr = kernelSource;
-    if( create_single_kernel_helper( context, &program, &kernel, 1, (const char **)&programPtr, "sample_test" ) )
-    {
-        return -1;
-    }
-
-    /* Generate some streams */
-    generate_equiv_test_data_float( inDataA, vecSize, true, d );
-    generate_equiv_test_data_float( inDataB, vecSize, false, d );
-
-    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
-                                sizeof(cl_float) * vecSize * TEST_SIZE,
-                                &inDataA, &error);
-    if( streams[0] == NULL )
-    {
-        print_error( error, "Creating input array A failed!\n");
-        return -1;
-    }
-    streams[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
-                                sizeof(cl_float) * vecSize * TEST_SIZE,
-                                &inDataB, &error);
-    if( streams[1] == NULL )
-    {
-        print_error( error, "Creating input array A failed!\n");
-        return -1;
-    }
-    streams[2] = clCreateBuffer( context, CL_MEM_READ_WRITE, sizeof( cl_int ) * vecSize * TEST_SIZE, NULL, &error);
-    if( streams[2] == NULL )
-    {
-        print_error( error, "Creating output array failed!\n");
-        return -1;
-    }
-  streams[3] = clCreateBuffer( context, CL_MEM_READ_WRITE, sizeof( cl_int ) * vecSize * TEST_SIZE, NULL, &error);
-    if( streams[3] == NULL )
-    {
-        print_error( error, "Creating output array failed!\n");
-        return -1;
-    }
-
-
-    /* Assign streams and execute */
-    error = clSetKernelArg( kernel, 0, sizeof( streams[0] ), &streams[0] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-    error = clSetKernelArg( kernel, 1, sizeof( streams[1] ), &streams[1] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-    error = clSetKernelArg( kernel, 2, sizeof( streams[2] ), &streams[2] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-  error = clSetKernelArg( kernel, 3, sizeof( streams[3] ), &streams[3] );
-    test_error( error, "Unable to set indexed kernel arguments" );
-
-
-    /* Run the kernel */
-    threads[0] = TEST_SIZE;
-
-    error = get_max_common_work_group_size( context, kernel, threads[0], &localThreads[0] );
-    test_error( error, "Unable to get work group size to use" );
-
-    error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, localThreads, 0, NULL, NULL );
-    test_error( error, "Unable to execute test kernel" );
-
-  /* Now get the results */
-    error = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof( int ) * TEST_SIZE * vecSize, outData, 0, NULL, NULL );
-    test_error( error, "Unable to read output array!" );
-
-  /* And verify! */
-  for( i = 0; i < TEST_SIZE; i++ )
-  {
-        verify_equiv_values_float( vecSize, &inDataA[ i * vecSize ], &inDataB[ i * vecSize ], expected, verifyFn);
-
-        for( j = 0; j < (int)vecSize; j++ )
-        {
-            if( expected[ j ] != outData[ i * vecSize + j ] )
-            {
-                log_error( "ERROR: Data sample %d:%d at size %d does not validate! Expected %d, got %d, source %f,%f\n",
-                  i, j, vecSize, expected[ j ], outData[ i * vecSize + j ], inDataA[i*vecSize + j], inDataB[i*vecSize + j] );
-                return -1;
-            }
-        }
-  }
-
-  /* Now get the results */
-    error = clEnqueueReadBuffer( queue, streams[3], true, 0, sizeof( int ) * TEST_SIZE * vecSize, outData, 0, NULL, NULL );
-    test_error( error, "Unable to read output array!" );
-
-  /* And verify! */
-    int fail = 0;
-    for( i = 0; i < TEST_SIZE; i++ )
-    {
-        verify_equiv_values_float( vecSize, &inDataA[ i * vecSize ], &inDataB[ i * vecSize ], expected, verifyFn);
-
-        for( j = 0; j < (int)vecSize; j++ )
-        {
-            if( expected[ j ] != outData[ i * vecSize + j ] )
-            {
-                if (gInfNanSupport == 0)
-                {
-                    if (IsFloatNaN(inDataA[i*vecSize + j]) || IsFloatNaN (inDataB[i*vecSize + j]))
-                    {
-                        fail = 0;
-                    }
-                    else
-                        fail = 1;
-                }
-                if (fail)
-                {
-                    log_error( "ERROR: Data sample %d:%d at size %d does not validate! Expected %d, got %d, source %f,%f\n",
-                      i, j, vecSize, expected[ j ], outData[ i * vecSize + j ], inDataA[i*vecSize + j], inDataB[i*vecSize + j] );
-                    return -1;
-                }
-            }
-        }
-  }
-
-  return 0;
-}
-
-int test_equiv_kernel_set_float(cl_context context, cl_command_queue queue, const char *fnName, const char *opName, equivVerifyFn verifyFn, MTdata d )
-{
-    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
-    unsigned int index;
-    int retVal = 0;
-
-    for( index = 0; vecSizes[ index ] != 0; index++ )
-    {
-        // Test!
-        if( test_equiv_kernel_float(context, queue, fnName, opName, vecSizes[ index ], verifyFn, d ) != 0 )
-        {
-            log_error( "   Vector float%d FAILED\n", vecSizes[ index ] );
-            retVal = -1;
-        }
-    }
-
-    return retVal;
-}
-
-bool isequal_verify_fn_float( float valueA, float valueB )
-{
-    return valueA == valueB;
-}
-
-int test_relational_isequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "isequal", "==", isequal_verify_fn_float, seed );
-}
-
-bool isnotequal_verify_fn_float( float valueA, float valueB )
-{
-    return valueA != valueB;
-}
-
-int test_relational_isnotequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "isnotequal", "!=", isnotequal_verify_fn_float, seed );
-}
-
-bool isgreater_verify_fn_float( float valueA, float valueB )
-{
-    return valueA > valueB;
-}
-
-int test_relational_isgreater_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "isgreater", ">", isgreater_verify_fn_float, seed );
-}
-
-bool isgreaterequal_verify_fn_float( float valueA, float valueB )
-{
-    return valueA >= valueB;
-}
-
-int test_relational_isgreaterequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "isgreaterequal", ">=", isgreaterequal_verify_fn_float, seed );
-}
-
-bool isless_verify_fn_float( float valueA, float valueB )
-{
-    return valueA < valueB;
-}
-
-int test_relational_isless_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "isless", "<", isless_verify_fn_float, seed );
-}
-
-bool islessequal_verify_fn_float( float valueA, float valueB )
-{
-    return valueA <= valueB;
-}
-
-int test_relational_islessequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "islessequal", "<=", islessequal_verify_fn_float, seed );
-}
-
-bool islessgreater_verify_fn_float( float valueA, float valueB )
-{
-    return ( valueA < valueB ) || ( valueA > valueB );
-}
-
-int test_relational_islessgreater_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    RandomSeed seed( gRandomSeed );
-    return test_equiv_kernel_set_float( context, queue, "islessgreater", "<>", islessgreater_verify_fn_float, seed );
-}
-
-

test_conformance/relationals/test_comparisons_fp.cpp

@@ -0,0 +1,663 @@
+//
+// Copyright (c) 2022 The Khronos Group Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+#include <cstdint>
+#include <functional>
+#include <iostream>
+#include <map>
+#include <memory>
+#include <stdexcept>
+#include <vector>
+
+#include <CL/cl_half.h>
+
+#include "test_comparisons_fp.h"
+
+#define TEST_SIZE 512
+
+static char ftype[32] = { 0 };
+static char ftype_vec[32] = { 0 };
+static char itype[32] = { 0 };
+static char itype_vec[32] = { 0 };
+static char extension[128] = { 0 };
+
+// clang-format off
+// for readability sake keep this section unformatted
+const char* equivTestKernPat[] = {
+extension,
+"__kernel void sample_test(__global ", ftype_vec, " *sourceA, __global ", ftype_vec,
+" *sourceB, __global ", itype_vec, " *destValues, __global ", itype_vec, " *destValuesB)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
+"    destValuesB[tid] = sourceA[tid] %s sourceB[tid];\n"
+"}\n"};
+
+const char* equivTestKernPatLessGreater[] = {
+extension,
+"__kernel void sample_test(__global ", ftype_vec, " *sourceA, __global ", ftype_vec,
+" *sourceB, __global ", itype_vec, " *destValues, __global ", itype_vec, " *destValuesB)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"    destValues[tid] = %s( sourceA[tid], sourceB[tid] );\n"
+"    destValuesB[tid] = (sourceA[tid] < sourceB[tid]) | (sourceA[tid] > sourceB[tid]);\n"
+"}\n"};
+
+const char* equivTestKerPat_3[] = {
+extension,
+"__kernel void sample_test(__global ", ftype_vec, " *sourceA, __global ", ftype_vec,
+" *sourceB, __global ", itype_vec, " *destValues, __global ", itype_vec, " *destValuesB)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"    ",ftype_vec," sampA = vload3(tid, (__global ",ftype," *)sourceA);\n"
+"    ",ftype_vec," sampB = vload3(tid, (__global ",ftype," *)sourceB);\n"
+"    vstore3(%s( sampA, sampB ), tid, (__global ",itype," *)destValues);\n"
+"    vstore3(( sampA %s sampB ), tid, (__global ",itype," *)destValuesB);\n"
+"}\n"};
+
+const char* equivTestKerPatLessGreater_3[] = {
+extension,
+"__kernel void sample_test(__global ", ftype_vec, " *sourceA, __global ", ftype_vec,
+" *sourceB, __global ", itype_vec, " *destValues, __global ", itype_vec, " *destValuesB)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"    ", ftype_vec, " sampA = vload3(tid, (__global ", ftype, " *)sourceA);\n"
+"    ", ftype_vec, " sampB = vload3(tid, (__global ", ftype, " *)sourceB);\n"
+"    vstore3(%s( sampA, sampB ), tid, (__global ", itype, " *)destValues);\n"
+"    vstore3(( sampA < sampB ) | (sampA > sampB), tid, (__global ", itype, " *)destValuesB);\n"
+"}\n"
+};
+// clang-format on
+
+
+std::string concat_kernel(const char* sstr[], int num)
+{
+    std::string res;
+    for (int i = 0; i < num; i++) res += std::string(sstr[i]);
+    return res;
+}
+
+template <typename... Args>
+std::string string_format(const std::string& format, Args... args)
+{
+    int size_s = std::snprintf(nullptr, 0, format.c_str(), args...)
+        + 1; // Extra space for '\0'
+    if (size_s <= 0)
+    {
+        throw std::runtime_error("Error during formatting.");
+    }
+    auto size = static_cast<size_t>(size_s);
+    std::unique_ptr<char[]> buf(new char[size]);
+    std::snprintf(buf.get(), size, format.c_str(), args...);
+    return std::string(buf.get(),
+                       buf.get() + size - 1); // We don't want the '\0' inside
+}
+
+template <typename T, typename F> bool verify(const T& A, const T& B)
+{
+    return F()(A, B);
+}
+
+RelationalsFPTest::RelationalsFPTest(cl_context context, cl_device_id device,
+                                     cl_command_queue queue, const char* fn,
+                                     const char* op)
+    : context(context), device(device), queue(queue), fnName(fn), opName(op),
+      halfFlushDenormsToZero(0)
+{
+    // hardcoded for now, to be changed into typeid().name solution in future
+    // for now C++ spec doesn't guarantee human readable type name
+
+    eqTypeNames = { { kHalf, "short" },
+                    { kFloat, "int" },
+                    { kDouble, "long" } };
+}
+
+template <typename T>
+void RelationalsFPTest::generate_equiv_test_data(T* outData,
+                                                 unsigned int vecSize,
+                                                 bool alpha,
+                                                 const RelTestParams<T>& param,
+                                                 const MTdata& d)
+{
+    unsigned int i;
+
+    generate_random_data(param.dataType, vecSize * TEST_SIZE, d, outData);
+
+    // Fill the first few vectors with NAN in each vector element (or the second
+    // set if we're alpha, so we can test either case)
+    if (alpha) outData += vecSize * vecSize;
+    for (i = 0; i < vecSize; i++)
+    {
+        outData[0] = param.nan;
+        outData += vecSize + 1;
+    }
+    // Make sure the third set is filled regardless, to test the case where both
+    // have NANs
+    if (!alpha) outData += vecSize * vecSize;
+    for (i = 0; i < vecSize; i++)
+    {
+        outData[0] = param.nan;
+        outData += vecSize + 1;
+    }
+}
+
+template <typename T, typename U>
+void RelationalsFPTest::verify_equiv_values(unsigned int vecSize,
+                                            const T* const inDataA,
+                                            const T* const inDataB,
+                                            U* const outData,
+                                            const VerifyFunc<T>& verifyFn)
+{
+    unsigned int i;
+    int trueResult;
+    bool result;
+
+    trueResult = (vecSize == 1) ? 1 : -1;
+    for (i = 0; i < vecSize; i++)
+    {
+        result = verifyFn(inDataA[i], inDataB[i]);
+        outData[i] = result ? trueResult : 0;
+    }
+}
+
+template <typename T>
+int RelationalsFPTest::test_equiv_kernel(unsigned int vecSize,
+                                         const RelTestParams<T>& param,
+                                         const MTdata& d)
+{
+    clProgramWrapper program;
+    clKernelWrapper kernel;
+    clMemWrapper streams[4];
+    T inDataA[TEST_SIZE * 16], inDataB[TEST_SIZE * 16];
+
+    // support half, float, double equivalents - otherwise assert
+    typedef typename std::conditional<
+        (sizeof(T) == sizeof(std::int16_t)), std::int16_t,
+        typename std::conditional<(sizeof(T) == sizeof(std::int32_t)),
+                                  std::int32_t, std::int64_t>::type>::type U;
+
+    U outData[TEST_SIZE * 16], expected[16];
+    int error, i, j;
+    size_t threads[1], localThreads[1];
+    std::string kernelSource;
+    char sizeName[4];
+
+    /* Create the source */
+    if (vecSize == 1)
+        sizeName[0] = 0;
+    else
+        sprintf(sizeName, "%d", vecSize);
+
+    if (eqTypeNames.find(param.dataType) == eqTypeNames.end())
+        log_error(
+            "RelationalsFPTest::test_equiv_kernel: unsupported fp data type");
+
+    sprintf(ftype, "%s", get_explicit_type_name(param.dataType));
+    sprintf(ftype_vec, "%s%s", get_explicit_type_name(param.dataType),
+            sizeName);
+
+    sprintf(itype, "%s", eqTypeNames[param.dataType].c_str());
+    sprintf(itype_vec, "%s%s", eqTypeNames[param.dataType].c_str(), sizeName);
+
+    if (std::is_same<T, double>::value)
+        strcpy(extension, "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n");
+    else if (std::is_same<T, cl_half>::value)
+        strcpy(extension, "#pragma OPENCL EXTENSION cl_khr_fp16 : enable\n");
+    else
+        extension[0] = '\0';
+
+    if (DENSE_PACK_VECS && vecSize == 3)
+    {
+        if (strcmp(fnName.c_str(), "islessgreater"))
+        {
+            auto str =
+                concat_kernel(equivTestKerPat_3,
+                              sizeof(equivTestKerPat_3) / sizeof(const char*));
+            kernelSource = string_format(str, fnName.c_str(), opName.c_str());
+        }
+        else
+        {
+            auto str = concat_kernel(equivTestKerPatLessGreater_3,
+                                     sizeof(equivTestKerPatLessGreater_3)
+                                         / sizeof(const char*));
+            kernelSource = string_format(str, fnName.c_str());
+        }
+    }
+    else
+    {
+        if (strcmp(fnName.c_str(), "islessgreater"))
+        {
+            auto str =
+                concat_kernel(equivTestKernPat,
+                              sizeof(equivTestKernPat) / sizeof(const char*));
+            kernelSource = string_format(str, fnName.c_str(), opName.c_str());
+        }
+        else
+        {
+            auto str = concat_kernel(equivTestKernPatLessGreater,
+                                     sizeof(equivTestKernPatLessGreater)
+                                         / sizeof(const char*));
+            kernelSource = string_format(str, fnName.c_str());
+        }
+    }
+
+    /* Create kernels */
+    const char* programPtr = kernelSource.c_str();
+    if (create_single_kernel_helper(context, &program, &kernel, 1,
+                                    (const char**)&programPtr, "sample_test"))
+    {
+        return -1;
+    }
+
+    /* Generate some streams */
+    generate_equiv_test_data<T>(inDataA, vecSize, true, param, d);
+    generate_equiv_test_data<T>(inDataB, vecSize, false, param, d);
+
+    streams[0] =
+        clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
+                       sizeof(T) * vecSize * TEST_SIZE, &inDataA, &error);
+    if (streams[0] == NULL)
+    {
+        print_error(error, "Creating input array A failed!\n");
+        return -1;
+    }
+    streams[1] =
+        clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
+                       sizeof(T) * vecSize * TEST_SIZE, &inDataB, &error);
+    if (streams[1] == NULL)
+    {
+        print_error(error, "Creating input array A failed!\n");
+        return -1;
+    }
+    streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                sizeof(U) * vecSize * TEST_SIZE, NULL, &error);
+    if (streams[2] == NULL)
+    {
+        print_error(error, "Creating output array failed!\n");
+        return -1;
+    }
+    streams[3] = clCreateBuffer(context, CL_MEM_READ_WRITE,
+                                sizeof(U) * vecSize * TEST_SIZE, NULL, &error);
+    if (streams[3] == NULL)
+    {
+        print_error(error, "Creating output array failed!\n");
+        return -1;
+    }
+
+    /* Assign streams and execute */
+    error = clSetKernelArg(kernel, 0, sizeof(streams[0]), &streams[0]);
+    test_error(error, "Unable to set indexed kernel arguments");
+    error = clSetKernelArg(kernel, 1, sizeof(streams[1]), &streams[1]);
+    test_error(error, "Unable to set indexed kernel arguments");
+    error = clSetKernelArg(kernel, 2, sizeof(streams[2]), &streams[2]);
+    test_error(error, "Unable to set indexed kernel arguments");
+    error = clSetKernelArg(kernel, 3, sizeof(streams[3]), &streams[3]);
+    test_error(error, "Unable to set indexed kernel arguments");
+
+    /* Run the kernel */
+    threads[0] = TEST_SIZE;
+
+    error = get_max_common_work_group_size(context, kernel, threads[0],
+                                           &localThreads[0]);
+    test_error(error, "Unable to get work group size to use");
+
+    error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, threads,
+                                   localThreads, 0, NULL, NULL);
+    test_error(error, "Unable to execute test kernel");
+
+    /* Now get the results */
+    error = clEnqueueReadBuffer(queue, streams[2], true, 0,
+                                sizeof(U) * TEST_SIZE * vecSize, outData, 0,
+                                NULL, NULL);
+    test_error(error, "Unable to read output array!");
+
+    auto verror_msg = [](const int& i, const int& j, const unsigned& vs,
+                         const U& e, const U& o, const T& iA, const T& iB) {
+        std::stringstream sstr;
+        sstr << "ERROR: Data sample " << i << ":" << j << " at size " << vs
+             << " does not validate! Expected " << e << ", got " << o
+             << ", source " << iA << ":" << iB << std::endl;
+        log_error(sstr.str().c_str());
+    };
+
+    /* And verify! */
+    for (i = 0; i < TEST_SIZE; i++)
+    {
+        verify_equiv_values<T, U>(vecSize, &inDataA[i * vecSize],
+                                  &inDataB[i * vecSize], expected,
+                                  param.verifyFn);
+
+        for (j = 0; j < (int)vecSize; j++)
+        {
+            if (expected[j] != outData[i * vecSize + j])
+            {
+                bool acceptFail = true;
+                if (std::is_same<T, cl_half>::value)
+                {
+                    bool in_denorm = IsHalfSubnormal(inDataA[i * vecSize + j])
+                        || IsHalfSubnormal(inDataB[i * vecSize + j]);
+
+                    if (halfFlushDenormsToZero && in_denorm)
+                    {
+                        acceptFail = false;
+                    }
+                }
+
+                if (acceptFail)
+                {
+                    verror_msg(
+                        i, j, vecSize, expected[j], outData[i * vecSize + j],
+                        inDataA[i * vecSize + j], inDataB[i * vecSize + j]);
+                    return -1;
+                }
+            }
+        }
+    }
+
+    /* Now get the results */
+    error = clEnqueueReadBuffer(queue, streams[3], true, 0,
+                                sizeof(U) * TEST_SIZE * vecSize, outData, 0,
+                                NULL, NULL);
+    test_error(error, "Unable to read output array!");
+
+    /* And verify! */
+    int fail = 0;
+    for (i = 0; i < TEST_SIZE; i++)
+    {
+        verify_equiv_values<T, U>(vecSize, &inDataA[i * vecSize],
+                                  &inDataB[i * vecSize], expected,
+                                  param.verifyFn);
+
+        for (j = 0; j < (int)vecSize; j++)
+        {
+            if (expected[j] != outData[i * vecSize + j])
+            {
+                if (std::is_same<T, float>::value)
+                {
+                    if (gInfNanSupport == 0)
+                    {
+                        if (isnan(inDataA[i * vecSize + j])
+                            || isnan(inDataB[i * vecSize + j]))
+                            fail = 0;
+                        else
+                            fail = 1;
+                    }
+                    if (fail)
+                    {
+                        verror_msg(i, j, vecSize, expected[j],
+                                   outData[i * vecSize + j],
+                                   inDataA[i * vecSize + j],
+                                   inDataB[i * vecSize + j]);
+                        return -1;
+                    }
+                }
+                else if (std::is_same<T, cl_half>::value)
+                {
+                    bool in_denorm = IsHalfSubnormal(inDataA[i * vecSize + j])
+                        || IsHalfSubnormal(inDataB[i * vecSize + j]);
+
+                    if (!(halfFlushDenormsToZero && in_denorm))
+                    {
+                        verror_msg(i, j, vecSize, expected[j],
+                                   outData[i * vecSize + j],
+                                   inDataA[i * vecSize + j],
+                                   inDataB[i * vecSize + j]);
+                        return -1;
+                    }
+                }
+                else
+                {
+                    verror_msg(
+                        i, j, vecSize, expected[j], outData[i * vecSize + j],
+                        inDataA[i * vecSize + j], inDataB[i * vecSize + j]);
+                    return -1;
+                }
+            }
+        }
+    }
+    return 0;
+}
+
+template <typename T>
+int RelationalsFPTest::test_relational(int numElements,
+                                       const RelTestParams<T>& param)
+{
+    RandomSeed seed(gRandomSeed);
+    unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
+    unsigned int index;
+    int retVal = 0;
+
+    for (index = 0; vecSizes[index] != 0; index++)
+    {
+        // Test!
+        if (test_equiv_kernel<T>(vecSizes[index], param, seed) != 0)
+        {
+            log_error("   Vector %s%d FAILED\n", ftype, vecSizes[index]);
+            retVal = -1;
+        }
+    }
+    return retVal;
+}
+
+cl_int RelationalsFPTest::SetUp(int elements)
+{
+    if (is_extension_available(device, "cl_khr_fp16"))
+    {
+        cl_device_fp_config config = 0;
+        cl_int error = clGetDeviceInfo(device, CL_DEVICE_HALF_FP_CONFIG,
+                                       sizeof(config), &config, NULL);
+        test_error(error, "Unable to get device CL_DEVICE_HALF_FP_CONFIG");
+
+        halfFlushDenormsToZero = (0 == (config & CL_FP_DENORM));
+        log_info("Supports half precision denormals: %s\n",
+                 halfFlushDenormsToZero ? "NO" : "YES");
+    }
+
+    return CL_SUCCESS;
+}
+
+cl_int RelationalsFPTest::Run()
+{
+    cl_int error = CL_SUCCESS;
+    for (auto&& param : params)
+    {
+        switch (param->dataType)
+        {
+            case kHalf:
+                error = test_relational<cl_half>(
+                    num_elements, *((RelTestParams<cl_half>*)param.get()));
+                break;
+            case kFloat:
+                error = test_relational<float>(
+                    num_elements, *((RelTestParams<float>*)param.get()));
+                break;
+            case kDouble:
+                error = test_relational<double>(
+                    num_elements, *((RelTestParams<double>*)param.get()));
+                break;
+            default:
+                test_error(-1, "RelationalsFPTest::Run: incorrect fp type");
+                break;
+        }
+        test_error(error, "RelationalsFPTest::Run: test_relational failed");
+    }
+    return CL_SUCCESS;
+}
+
+cl_int IsEqualFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_equals_to>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::equal_to<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::equal_to<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsNotEqualFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_not_equals_to>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::not_equal_to<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::not_equal_to<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsGreaterFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_greater>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::greater<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::greater<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsGreaterEqualFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_greater_equal>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::greater_equal<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::greater_equal<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsLessFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_less>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::less<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::less<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsLessEqualFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_less_equal>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, std::less_equal<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, std::less_equal<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+cl_int IsLessGreaterFPTest::SetUp(int elements)
+{
+    num_elements = elements;
+    if (is_extension_available(device, "cl_khr_fp16"))
+        params.emplace_back(new RelTestParams<cl_half>(
+            &verify<cl_half, half_less_greater>, kHalf, HALF_NAN));
+
+    params.emplace_back(new RelTestParams<float>(
+        &verify<float, less_greater<float>>, kFloat, NAN));
+
+    if (is_extension_available(device, "cl_khr_fp64"))
+        params.emplace_back(new RelTestParams<double>(
+            &verify<double, less_greater<double>>, kDouble, NAN));
+
+    return RelationalsFPTest::SetUp(elements);
+}
+
+int test_relational_isequal(cl_device_id device, cl_context context,
+                            cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsEqualFPTest>(device, context, queue, numElements);
+}
+
+int test_relational_isnotequal(cl_device_id device, cl_context context,
+                               cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsNotEqualFPTest>(device, context, queue,
+                                            numElements);
+}
+
+int test_relational_isgreater(cl_device_id device, cl_context context,
+                              cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsGreaterFPTest>(device, context, queue, numElements);
+}
+
+int test_relational_isgreaterequal(cl_device_id device, cl_context context,
+                                   cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsGreaterEqualFPTest>(device, context, queue,
+                                                numElements);
+}
+
+int test_relational_isless(cl_device_id device, cl_context context,
+                           cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsLessFPTest>(device, context, queue, numElements);
+}
+
+int test_relational_islessequal(cl_device_id device, cl_context context,
+                                cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsLessEqualFPTest>(device, context, queue,
+                                             numElements);
+}
+
+int test_relational_islessgreater(cl_device_id device, cl_context context,
+                                  cl_command_queue queue, int numElements)
+{
+    return MakeAndRunTest<IsLessGreaterFPTest>(device, context, queue,
+                                               numElements);
+}

test_conformance/relationals/test_comparisons_fp.h

@@ -0,0 +1,228 @@
+//
+// Copyright (c) 2022 The Khronos Group Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+#ifndef TEST_COMPARISONS_FP_H
+#define TEST_COMPARISONS_FP_H
+
+#include <map>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include <CL/cl_half.h>
+
+#include "testBase.h"
+
+#define HALF_NAN 0x7e00
+template <typename T> using VerifyFunc = bool (*)(const T &, const T &);
+
+struct RelTestBase
+{
+    explicit RelTestBase(const ExplicitTypes &dt): dataType(dt) {}
+    virtual ~RelTestBase() = default;
+    ExplicitTypes dataType;
+};
+
+template <typename T> struct RelTestParams : public RelTestBase
+{
+    RelTestParams(const VerifyFunc<T> &vfn, const ExplicitTypes &dt,
+                  const T &nan_)
+        : RelTestBase(dt), verifyFn(vfn), nan(nan_)
+    {}
+
+    VerifyFunc<T> verifyFn;
+    T nan;
+};
+
+struct RelationalsFPTest
+{
+    RelationalsFPTest(cl_context context, cl_device_id device,
+                      cl_command_queue queue, const char *fn, const char *op);
+
+    virtual cl_int SetUp(int elements);
+
+    // Test body returning an OpenCL error code
+    virtual cl_int Run();
+
+    template <typename T>
+    void generate_equiv_test_data(T *, unsigned int, bool,
+                                  const RelTestParams<T> &, const MTdata &);
+
+    template <typename T, typename U>
+    void verify_equiv_values(unsigned int, const T *const, const T *const,
+                             U *const, const VerifyFunc<T> &);
+
+    template <typename T>
+    int test_equiv_kernel(unsigned int vecSize, const RelTestParams<T> &param,
+                          const MTdata &d);
+
+    template <typename T>
+    int test_relational(int numElements, const RelTestParams<T> &param);
+
+protected:
+    cl_context context;
+    cl_device_id device;
+    cl_command_queue queue;
+
+    std::string fnName;
+    std::string opName;
+
+    std::vector<std::unique_ptr<RelTestBase>> params;
+    std::map<ExplicitTypes, std::string> eqTypeNames;
+    size_t num_elements;
+
+    int halfFlushDenormsToZero;
+};
+
+struct IsEqualFPTest : public RelationalsFPTest
+{
+    IsEqualFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "isequal", "==")
+    {}
+    cl_int SetUp(int elements) override;
+
+    // for correct handling nan/inf we need fp value
+    struct half_equals_to
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) == cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsNotEqualFPTest : public RelationalsFPTest
+{
+    IsNotEqualFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "isnotequal", "!=")
+    {}
+    cl_int SetUp(int elements) override;
+
+    // for correct handling nan/inf we need fp value
+    struct half_not_equals_to
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) != cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsGreaterFPTest : public RelationalsFPTest
+{
+    IsGreaterFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "isgreater", ">")
+    {}
+    cl_int SetUp(int elements) override;
+
+    struct half_greater
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) > cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsGreaterEqualFPTest : public RelationalsFPTest
+{
+    IsGreaterEqualFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "isgreaterequal", ">=")
+    {}
+    cl_int SetUp(int elements) override;
+
+    struct half_greater_equal
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) >= cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsLessFPTest : public RelationalsFPTest
+{
+    IsLessFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "isless", "<")
+    {}
+    cl_int SetUp(int elements) override;
+
+    struct half_less
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) < cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsLessEqualFPTest : public RelationalsFPTest
+{
+    IsLessEqualFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "islessequal", "<=")
+    {}
+    cl_int SetUp(int elements) override;
+
+    struct half_less_equal
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            return cl_half_to_float(lhs) <= cl_half_to_float(rhs);
+        }
+    };
+};
+
+struct IsLessGreaterFPTest : public RelationalsFPTest
+{
+    IsLessGreaterFPTest(cl_device_id d, cl_context c, cl_command_queue q)
+        : RelationalsFPTest(c, d, q, "islessgreater", "<>")
+    {}
+    cl_int SetUp(int elements) override;
+
+    struct half_less_greater
+    {
+        bool operator()(const cl_half &lhs, const cl_half &rhs) const
+        {
+            float flhs = cl_half_to_float(lhs), frhs = cl_half_to_float(rhs);
+            return (flhs < frhs) || (flhs > frhs);
+        }
+    };
+
+    template <typename T> struct less_greater
+    {
+        bool operator()(const T &lhs, const T &rhs) const
+        {
+            return (lhs < rhs) || (lhs > rhs);
+        }
+    };
+};
+
+template <class T>
+int MakeAndRunTest(cl_device_id device, cl_context context,
+                   cl_command_queue queue, int num_elements)
+{
+    auto test_fixture = T(device, context, queue);
+
+    cl_int error = test_fixture.SetUp(num_elements);
+    test_error_ret(error, "Error in test initialization", TEST_FAIL);
+
+    error = test_fixture.Run();
+    test_error_ret(error, "Test Failed", TEST_FAIL);
+
+    return TEST_PASS;
+}
+
+#endif // TEST_COMPARISONS_FP_H

test_conformance/relationals/test_relationals.cpp

@@ -18,8 +18,11 @@
 #include "harness/typeWrappers.h"
 #include "harness/testHarness.h"
 
+// clang-format off
+
 const char *anyAllTestKernelPattern =
 "%s\n" // optional pragma
+"%s\n" // optional pragma
 "__kernel void sample_test(__global %s%s *sourceA, __global int *destValues)\n"
 "{\n"
 "    int  tid = get_global_id(0);\n"
@@ -29,6 +32,7 @@ const char *anyAllTestKernelPattern =
 
 const char *anyAllTestKernelPatternVload =
 "%s\n" // optional pragma
+"%s\n" // optional pragma
 "__kernel void sample_test(__global %s%s *sourceA, __global int *destValues)\n"
 "{\n"
 "    int  tid = get_global_id(0);\n"
@@ -36,6 +40,8 @@ const char *anyAllTestKernelPatternVload =
 "\n"
 "}\n";
 
+// clang-format on
+
 #define TEST_SIZE 512
 
 typedef int (*anyAllVerifyFn)( ExplicitType vecType, unsigned int vecSize, void *inData );
@@ -67,13 +73,21 @@ int test_any_all_kernel(cl_context context, cl_command_queue queue,
              get_explicit_type_name( vecType ), sizeName);
     if(DENSE_PACK_VECS && vecSize == 3) {
         // anyAllTestKernelPatternVload
-        sprintf( kernelSource, anyAllTestKernelPatternVload,
-                vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
+        sprintf(
+            kernelSource, anyAllTestKernelPatternVload,
+            vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
+                               : "",
+            vecType == kHalf ? "#pragma OPENCL EXTENSION cl_khr_fp16 : enable"
+                             : "",
             get_explicit_type_name(vecType), sizeName, fnName,
             get_explicit_type_name(vecType));
     } else {
-        sprintf( kernelSource, anyAllTestKernelPattern,
-                vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
+        sprintf(
+            kernelSource, anyAllTestKernelPattern,
+            vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
+                               : "",
+            vecType == kHalf ? "#pragma OPENCL EXTENSION cl_khr_fp16 : enable"
+                             : "",
             get_explicit_type_name(vecType), sizeName, fnName);
     }
     /* Create kernels */
@@ -282,8 +296,11 @@ int test_relational_all(cl_device_id device, cl_context context, cl_command_queu
     return retVal;
 }
 
+// clang-format off
+
 const char *selectTestKernelPattern =
 "%s\n" // optional pragma
+"%s\n" // optional pragma
 "__kernel void sample_test(__global %s%s *sourceA, __global %s%s *sourceB, __global %s%s *sourceC, __global %s%s *destValues)\n"
 "{\n"
 "    int  tid = get_global_id(0);\n"
@@ -294,6 +311,7 @@ const char *selectTestKernelPattern =
 
 const char *selectTestKernelPatternVload =
 "%s\n" // optional pragma
+"%s\n" // optional pragma
 "__kernel void sample_test(__global %s%s *sourceA, __global %s%s *sourceB, __global %s%s *sourceC, __global %s%s *destValues)\n"
 "{\n"
 "    int  tid = get_global_id(0);\n"
@@ -302,6 +320,8 @@ const char *selectTestKernelPatternVload =
 "\n"
 "}\n";
 
+// clang-format on
+
 typedef void (*selectVerifyFn)( ExplicitType vecType, ExplicitType testVecType, unsigned int vecSize, void *inDataA, void *inDataB, void *inDataTest, void *outData );
 
 int test_select_kernel(cl_context context, cl_command_queue queue, const char *fnName,
@@ -336,25 +356,33 @@ int test_select_kernel(cl_context context, cl_command_queue queue, const char *f
     if(DENSE_PACK_VECS && vecSize == 3) {
         // anyAllTestKernelPatternVload
         sprintf(kernelSource, selectTestKernelPatternVload,
-                (vecType == kDouble || testVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
+                (vecType == kDouble || testVecType == kDouble)
+                    ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
+                    : "",
+                (vecType == kHalf || testVecType == kHalf)
+                    ? "#pragma OPENCL EXTENSION cl_khr_fp16 : enable"
+                    : "",
                 get_explicit_type_name(vecType), sizeName,
                 get_explicit_type_name(vecType), sizeName,
                 get_explicit_type_name(testVecType), sizeName,
                 get_explicit_type_name(vecType), outSizeName,
-                get_explicit_type_name( vecType ), sizeName,
-                fnName,
+                get_explicit_type_name(vecType), sizeName, fnName,
                 get_explicit_type_name(vecType),
                 get_explicit_type_name(vecType),
                 get_explicit_type_name(vecType),
                 get_explicit_type_name(testVecType));
     } else {
         sprintf(kernelSource, selectTestKernelPattern,
-                (vecType == kDouble || testVecType == kDouble) ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable" : "",
+                (vecType == kDouble || testVecType == kDouble)
+                    ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
+                    : "",
+                (vecType == kHalf || testVecType == kHalf)
+                    ? "#pragma OPENCL EXTENSION cl_khr_fp16 : enable"
+                    : "",
                 get_explicit_type_name(vecType), sizeName,
                 get_explicit_type_name(vecType), sizeName,
                 get_explicit_type_name(testVecType), sizeName,
-                get_explicit_type_name( vecType ), outSizeName,
-                fnName );
+                get_explicit_type_name(vecType), outSizeName, fnName);
     }
 
     /* Create kernels */
@@ -500,14 +528,17 @@ void bitselect_verify_fn( ExplicitType vecType, ExplicitType testVecType, unsign
 
 int test_relational_bitselect(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
 {
-    ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble };
+    constexpr ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort,
+                                         kInt,  kUInt,  kLong,  kULong,
+                                         kHalf, kFloat, kDouble };
+    constexpr auto vecTypeSize = sizeof(vecType) / sizeof(ExplicitType);
     unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
     unsigned int index, typeIndex;
     int retVal = 0;
     RandomSeed seed( gRandomSeed );
 
 
-    for( typeIndex = 0; typeIndex < 10; typeIndex++ )
+    for (typeIndex = 0; typeIndex < vecTypeSize; typeIndex++)
     {
         if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong) && !gHasLong)
             continue;
@@ -522,6 +553,19 @@ int test_relational_bitselect(cl_device_id device, cl_context context, cl_comman
             else
                 log_info("Testing doubles.\n");
         }
+
+        if (vecType[typeIndex] == kHalf)
+        {
+            if (!is_extension_available(device, "cl_khr_fp16"))
+            {
+                log_info("Extension cl_khr_fp16 not supported; skipping half "
+                         "tests.\n");
+                continue;
+            }
+            else
+                log_info("Testing halfs.\n");
+        }
+
         for( index = 0; vecSizes[ index ] != 0; index++ )
         {
             // Test!
@@ -584,14 +628,18 @@ void select_signed_verify_fn( ExplicitType vecType, ExplicitType testVecType, un
 
 int test_relational_select_signed(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
 {
-    ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble };
+    constexpr ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort,
+                                         kInt,  kUInt,  kLong,  kULong,
+                                         kHalf, kFloat, kDouble };
+    constexpr auto vecTypeSize = sizeof(vecType) / sizeof(ExplicitType);
+
     ExplicitType testVecType[] = { kChar, kShort, kInt, kLong, kNumExplicitTypes };
     unsigned int vecSizes[] = { 1, 2, 4, 8, 16, 0 };
     unsigned int index, typeIndex, testTypeIndex;
     int retVal = 0;
     RandomSeed seed( gRandomSeed );
 
-    for( typeIndex = 0; typeIndex < 10; typeIndex++ )
+    for (typeIndex = 0; typeIndex < vecTypeSize; typeIndex++)
     {
         if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong) && !gHasLong)
             continue;
@@ -604,6 +652,19 @@ int test_relational_select_signed(cl_device_id device, cl_context context, cl_co
                 log_info("Testing doubles.\n");
             }
         }
+        if (vecType[typeIndex] == kHalf)
+        {
+            if (!is_extension_available(device, "cl_khr_fp16"))
+            {
+                log_info("Extension cl_khr_fp16 not supported; skipping half "
+                         "tests.\n");
+                continue;
+            }
+            else
+            {
+                log_info("Testing halfs.\n");
+            }
+        }
         for( testTypeIndex = 0; testVecType[ testTypeIndex ] != kNumExplicitTypes; testTypeIndex++ )
         {
             if( testVecType[ testTypeIndex ] != vecType[ typeIndex ] )
@@ -673,7 +734,11 @@ void select_unsigned_verify_fn( ExplicitType vecType, ExplicitType testVecType,
 
 int test_relational_select_unsigned(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
 {
-    ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong, kULong, kFloat, kDouble };
+    constexpr ExplicitType vecType[] = { kChar, kUChar, kShort, kUShort,
+                                         kInt,  kUInt,  kLong,  kULong,
+                                         kHalf, kFloat, kDouble };
+    constexpr auto vecTypeSize = sizeof(vecType) / sizeof(ExplicitType);
+
     ExplicitType testVecType[] = { kUChar, kUShort, kUInt, kULong, kNumExplicitTypes };
     unsigned int vecSizes[] = { 1, 2, 4, 8, 16, 0 };
     unsigned int index, typeIndex, testTypeIndex;
@@ -681,7 +746,7 @@ int test_relational_select_unsigned(cl_device_id device, cl_context context, cl_
     RandomSeed seed(gRandomSeed);
 
 
-    for( typeIndex = 0; typeIndex < 10; typeIndex++ )
+    for (typeIndex = 0; typeIndex < vecTypeSize; typeIndex++)
     {
         if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong) && !gHasLong)
             continue;
@@ -694,6 +759,19 @@ int test_relational_select_unsigned(cl_device_id device, cl_context context, cl_
                 log_info("Testing doubles.\n");
             }
         }
+        if (vecType[typeIndex] == kHalf)
+        {
+            if (!is_extension_available(device, "cl_khr_fp16"))
+            {
+                log_info("Extension cl_khr_fp16 not supported; skipping half "
+                         "tests.\n");
+                continue;
+            }
+            else
+            {
+                log_info("Testing halfs.\n");
+            }
+        }
         for( testTypeIndex = 0; testVecType[ testTypeIndex ] != kNumExplicitTypes; testTypeIndex++ )
         {
             if( testVecType[ testTypeIndex ] != vecType[ typeIndex ] )
@@ -714,85 +792,3 @@ int test_relational_select_unsigned(cl_device_id device, cl_context context, cl_
 
     return retVal;
 }
-
-
-
-extern int test_relational_isequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isnotequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isgreater_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isgreaterequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isless_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_islessequal_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_islessgreater_float(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isnotequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isgreater_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isgreaterequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_isless_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_islessequal_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-extern int test_relational_islessgreater_double(cl_device_id device, cl_context context, cl_command_queue queue, int numElements );
-
-
-int test_relational_isequal(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_isequal_float( device, context, queue, numElements );
-    err |= test_relational_isequal_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_isnotequal(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_isnotequal_float( device, context, queue, numElements );
-    err |= test_relational_isnotequal_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_isgreater(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_isgreater_float( device, context, queue, numElements );
-    err |= test_relational_isgreater_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_isgreaterequal(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_isgreaterequal_float( device, context, queue, numElements );
-    err |= test_relational_isgreaterequal_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_isless(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_isless_float( device, context, queue, numElements );
-    err |= test_relational_isless_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_islessequal(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_islessequal_float( device, context, queue, numElements );
-    err |= test_relational_islessequal_double( device, context, queue, numElements );
-    return err;
-}
-
-
-int test_relational_islessgreater(cl_device_id device, cl_context context, cl_command_queue queue, int numElements )
-{
-    int err = 0;
-    err |= test_relational_islessgreater_float( device, context, queue, numElements );
-    err |= test_relational_islessgreater_double( device, context, queue, numElements );
-    return err;
-}
-
-

test_conformance/select/test_select.cpp

@@ -303,6 +303,10 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
     cl_mem dest = NULL;
     void *ref = NULL;
     void *sref = NULL;
+    void *src1_host = NULL;
+    void *src2_host = NULL;
+    void *cmp_host = NULL;
+    void *dest_host = NULL;
 
     cl_ulong blocks = type_size[stype] * 0x100000000ULL / BUFFER_SIZE;
     size_t block_elements = BUFFER_SIZE / type_size[stype];
@@ -359,6 +363,30 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
     dest = clCreateBuffer( context, CL_MEM_WRITE_ONLY, BUFFER_SIZE, NULL, &err );
     if( err ) { log_error( "Error: could not allocate dest buffer\n" );  ++s_test_fail; goto exit; }
 
+    src1_host = malloc(BUFFER_SIZE);
+    if (NULL == src1_host)
+    {
+        log_error("Error: could not allocate src1_host buffer\n");
+        goto exit;
+    }
+    src2_host = malloc(BUFFER_SIZE);
+    if (NULL == src2_host)
+    {
+        log_error("Error: could not allocate src2_host buffer\n");
+        goto exit;
+    }
+    cmp_host = malloc(BUFFER_SIZE);
+    if (NULL == cmp_host)
+    {
+        log_error("Error: could not allocate cmp_host buffer\n");
+        goto exit;
+    }
+    dest_host = malloc(BUFFER_SIZE);
+    if (NULL == dest_host)
+    {
+        log_error("Error: could not allocate dest_host buffer\n");
+        goto exit;
+    }
 
     // We block the test as we are running over the range of compare values
     // "block the test" means "break the test into blocks"
@@ -387,13 +415,6 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
         // Setup the input data to change for each block
         initCmpBuffer(s3, cmptype, i * cmp_stride, block_elements);
 
-        // Create the reference result
-        Select sfunc = (cmptype == ctype[stype][0]) ? vrefSelects[stype][0] : vrefSelects[stype][1];
-        (*sfunc)(ref, s1, s2, s3, block_elements);
-
-        sfunc = (cmptype == ctype[stype][0]) ? refSelects[stype][0] : refSelects[stype][1];
-        (*sfunc)(sref, s1, s2, s3, block_elements);
-
         if( (err = clEnqueueUnmapMemObject( queue, src1, s1, 0, NULL, NULL )))
         { log_error( "Error: coult not unmap src1\n" );  ++s_test_fail; goto exit; }
         if( (err = clEnqueueUnmapMemObject( queue, src2, s2, 0, NULL, NULL )))
@@ -401,6 +422,40 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
         if( (err = clEnqueueUnmapMemObject( queue, cmp, s3, 0, NULL, NULL )))
         { log_error( "Error: coult not unmap cmp\n" );  ++s_test_fail; goto exit; }
 
+        // Create the reference result
+        err = clEnqueueReadBuffer(queue, src1, CL_TRUE, 0, BUFFER_SIZE,
+                                  src1_host, 0, NULL, NULL);
+        if (err)
+        {
+            log_error("Error: Reading buffer from src1 to src1_host failed\n");
+            ++s_test_fail;
+            goto exit;
+        }
+        err = clEnqueueReadBuffer(queue, src2, CL_TRUE, 0, BUFFER_SIZE,
+                                  src2_host, 0, NULL, NULL);
+        if (err)
+        {
+            log_error("Error: Reading buffer from src2 to src2_host failed\n");
+            ++s_test_fail;
+            goto exit;
+        }
+        err = clEnqueueReadBuffer(queue, cmp, CL_TRUE, 0, BUFFER_SIZE, cmp_host,
+                                  0, NULL, NULL);
+        if (err)
+        {
+            log_error("Error: Reading buffer from cmp to cmp_host failed\n");
+            ++s_test_fail;
+            goto exit;
+        }
+
+        Select sfunc = (cmptype == ctype[stype][0]) ? vrefSelects[stype][0]
+                                                    : vrefSelects[stype][1];
+        (*sfunc)(ref, src1_host, src2_host, cmp_host, block_elements);
+
+        sfunc = (cmptype == ctype[stype][0]) ? refSelects[stype][0]
+                                             : refSelects[stype][1];
+        (*sfunc)(sref, src1_host, src2_host, cmp_host, block_elements);
+
         for (vecsize = 0; vecsize < VECTOR_SIZE_COUNT; ++vecsize)
         {
             size_t vector_size = element_count[vecsize] * type_size[stype];
@@ -415,7 +470,6 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
             if((err = clSetKernelArg(kernels[vecsize], 3,  sizeof cmp, &cmp) ))
             { log_error( "Error: Cannot set kernel arg dest! %d\n", err ); ++s_test_fail; goto exit; }
 
-
             // Wipe destination
             void *d = clEnqueueMapBuffer( queue, dest, CL_TRUE, CL_MAP_WRITE, 0, BUFFER_SIZE, 0, NULL, NULL, &err );
             if( err ){ log_error( "Error: Could not map dest" );  ++s_test_fail; goto exit; }
@@ -429,18 +483,22 @@ static int doTest(cl_command_queue queue, cl_context context, Type stype, Type c
                 goto exit;
             }
 
-            d = clEnqueueMapBuffer( queue, dest, CL_TRUE, CL_MAP_READ, 0, BUFFER_SIZE, 0, NULL, NULL, &err );
-            if( err ){ log_error( "Error: Could not map dest # 2" );  ++s_test_fail; goto exit; }
-
-            if ((*checkResults[stype])(d, vecsize == 0 ? sref : ref, block_elements, element_count[vecsize])!=0){
-                log_error("vec_size:%d indx: 0x%16.16llx\n", (int)element_count[vecsize], i);
+            err = clEnqueueReadBuffer(queue, dest, CL_TRUE, 0, BUFFER_SIZE,
+                                      dest_host, 0, NULL, NULL);
+            if (err)
+            {
+                log_error(
+                    "Error: Reading buffer from dest to dest_host failed\n");
                 ++s_test_fail;
                 goto exit;
             }
 
-            if( (err = clEnqueueUnmapMemObject( queue, dest, d, 0, NULL, NULL ) ) )
+            if ((*checkResults[stype])(dest_host, vecsize == 0 ? sref : ref,
+                                       block_elements, element_count[vecsize])
+                != 0)
             {
-                log_error( "Error: Could not unmap dest" );
+                log_error("vec_size:%d indx: 0x%16.16llx\n",
+                          (int)element_count[vecsize], i);
                 ++s_test_fail;
                 goto exit;
             }
@@ -459,6 +517,10 @@ exit:
     if( dest)   clReleaseMemObject( dest );
     if( ref )   free(ref );
     if( sref )  free(sref );
+    if (src1_host) free(src1_host);
+    if (src2_host) free(src2_host);
+    if (cmp_host) free(cmp_host);
+    if (dest_host) free(dest_host);
 
     for (vecsize = 0; vecsize < VECTOR_SIZE_COUNT; vecsize++) {
         clReleaseKernel(kernels[vecsize]);