Initial open source release of OpenCL 2.2 CTS.

test_conformance/commonfns/CMakeLists.txt

@@ -0,0 +1,33 @@
+set(MODULE_NAME COMMONFNS)
+
+set(${MODULE_NAME}_SOURCES
+    main.c
+    test_clamp.c
+    test_degrees.c
+    test_max.c
+    test_maxf.c
+    test_min.c
+    test_minf.c
+    test_mix.c
+    test_radians.c
+    test_step.c
+    test_stepf.c
+    test_smoothstep.c
+    test_smoothstepf.c
+    test_sign.c
+    test_fmax.c
+    test_fmin.c
+    test_fmaxf.c
+    test_fminf.c
+    test_binary_fn.c
+    ../../test_common/harness/errorHelpers.c
+    ../../test_common/harness/threadTesting.c
+    ../../test_common/harness/testHarness.c
+    ../../test_common/harness/kernelHelpers.c
+    ../../test_common/harness/mt19937.c
+    ../../test_common/harness/conversions.c
+    ../../test_common/harness/msvc9.c
+    ../../test_common/harness/parseParameters.cpp
+)
+
+include(../CMakeCommon.txt)

test_conformance/commonfns/Jamfile

@@ -0,0 +1,33 @@
+project
+    : requirements
+      <toolset>gcc:<cflags>-xc++
+      <toolset>msvc:<cflags>"/TP"
+    ;
+
+exe test_commonfns
+    : main.c
+      test_binary_fn.c
+      test_clamp.c
+      test_degrees.c
+      test_fmax.c
+      test_fmaxf.c
+      test_fmin.c
+      test_fminf.c
+      test_max.c
+      test_maxf.c
+      test_min.c
+      test_minf.c
+      test_mix.c
+      test_radians.c
+      test_sign.c
+      test_smoothstep.c
+      test_smoothstepf.c
+      test_step.c
+      test_stepf.c
+    ;
+
+install dist
+    : test_commonfns
+    : <variant>debug:<location>$(DIST)/debug/tests/test_conformance/commonfns
+      <variant>release:<location>$(DIST)/release/tests/test_conformance/commonfns
+    ;

test_conformance/commonfns/Makefile

@@ -0,0 +1,45 @@
+ifdef BUILD_WITH_ATF
+ATF = -framework ATF
+USE_ATF = -DUSE_ATF
+endif
+
+SRCS = main.c test_clamp.c test_degrees.c \
+			test_max.c test_maxf.c test_min.c test_minf.c \
+            test_mix.c test_radians.c test_step.c test_stepf.c\
+            test_smoothstep.c test_smoothstepf.c test_sign.c \
+			test_fmax.c test_fmin.c test_fmaxf.c test_fminf.c test_binary_fn.c \
+		  ../../test_common/harness/errorHelpers.c \
+		  ../../test_common/harness/threadTesting.c \
+		  ../../test_common/harness/testHarness.c \
+                  ../../test_common/harness/conversions.c \
+                  ../../test_common/harness/mt19937.c \
+		  ../../test_common/harness/kernelHelpers.c 
+
+SOURCES = $(abspath $(SRCS))
+
+LIBPATH += -L/System/Library/Frameworks/OpenCL.framework/Libraries
+LIBPATH += -L.
+FRAMEWORK = $(SOURCES)
+HEADERS = 
+TARGET = test_commonfns
+INCLUDE = 
+COMPILERFLAGS = -c -Wall -g -Wshorten-64-to-32 -Os
+CC = c++
+CFLAGS = $(COMPILERFLAGS) ${RC_CFLAGS} ${USE_ATF}
+CXXFLAGS = $(COMPILERFLAGS) ${RC_CFLAGS} ${USE_ATF} $(DEFINES:%=-D%) $(INCLUDE)
+LIBRARIES = -framework OpenCL -framework OpenGL -framework GLUT -framework AppKit ${RC_CFLAGS} ${ATF}
+
+OBJECTS := ${SOURCES:.c=.o}
+OBJECTS := ${OBJECTS:.cpp=.o}
+
+TARGETOBJECT =
+all: $(TARGET)
+
+$(TARGET): $(OBJECTS)
+	$(CC) $(RC_CFLAGS) $(OBJECTS) -o $@ $(LIBPATH) $(LIBRARIES)
+
+clean:
+	rm -f $(TARGET) $(OBJECTS)
+
+.DEFAULT:
+	@echo The target \"$@\" does not exist in Makefile.

test_conformance/commonfns/main.c

@@ -0,0 +1,87 @@
+//
+// 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 "../../test_common/harness/compat.h"
+
+#include <stdio.h>
+#include <string.h>
+#include "procs.h"
+
+int g_arrVecSizes[kVectorSizeCount + kStrangeVectorSizeCount];
+int g_arrStrangeVectorSizes[kStrangeVectorSizeCount] = {3};
+
+static void initVecSizes() {
+    int i;
+    for(i = 0; i < kVectorSizeCount; ++i) {
+    g_arrVecSizes[i] = (1<<i);
+    }
+    for(; i < kVectorSizeCount + kStrangeVectorSizeCount; ++i) {
+    g_arrVecSizes[i] = g_arrStrangeVectorSizes[i-kVectorSizeCount];
+    }
+}
+
+
+basefn    commonfn_list[] = {
+                test_clamp,
+                test_degrees,
+                test_fmax,
+                test_fmaxf,
+                test_fmin,
+                test_fminf,
+                test_max,
+                test_maxf,
+                test_min,
+                test_minf,
+                test_mix,
+                test_radians,
+                test_step,
+                test_stepf,
+                test_smoothstep,
+                test_smoothstepf,
+                test_sign,
+};
+
+const char *commonfn_names[] = {
+    "clamp",
+    "degrees",
+    "fmax",
+    "fmaxf",
+    "fmin",
+    "fminf",
+    "max",
+    "maxf",
+    "min",
+    "minf",
+    "mix",
+    "radians",
+    "step",
+    "stepf",
+    "smoothstep",
+    "smoothstepf",
+    "sign",
+};
+
+ct_assert((sizeof(commonfn_names) / sizeof(commonfn_names[0])) == (sizeof(commonfn_list) / sizeof(commonfn_list[0])));
+
+int    num_commonfns = sizeof(commonfn_names) / sizeof(char *);
+
+int
+main(int argc, const char *argv[])
+{
+    initVecSizes();
+    return runTestHarness( argc, argv, num_commonfns, commonfn_list, commonfn_names, false, false, 0 );
+}
+
+

test_conformance/commonfns/procs.h

@@ -0,0 +1,54 @@
+//
+// 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 "../../test_common/harness/testHarness.h"
+#include "../../test_common/harness/kernelHelpers.h"
+#include "../../test_common/harness/errorHelpers.h"
+#include "../../test_common/harness/conversions.h"
+#include "../../test_common/harness/mt19937.h"
+
+#define kVectorSizeCount 5
+#define kStrangeVectorSizeCount 1
+#define kTotalVecCount (kVectorSizeCount + kStrangeVectorSizeCount)
+
+extern int g_arrVecSizes[kVectorSizeCount + kStrangeVectorSizeCount];
+// int g_arrStrangeVectorSizes[kStrangeVectorSizeCount] = {3};
+
+extern int        test_clamp(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_degrees(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_fmax(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_fmaxf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_fmin(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_fminf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_max(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+extern int        test_maxf(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+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_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);
+extern int        test_smoothstep(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements);
+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_binary_fn.c

@@ -0,0 +1,260 @@
+//
+// 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 "../../test_common/harness/compat.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include "procs.h"
+
+const char *binary_fn_code_pattern =
+"%s\n" /* optional pragma */
+"__kernel void test_fn(__global %s%s *x, __global %s%s *y, __global %s%s *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    dst[tid] = %s(x[tid], y[tid]);\n"
+"}\n";
+
+const char *binary_fn_code_pattern_v3 =
+"%s\n" /* optional pragma */
+"__kernel void test_fn(__global %s *x, __global %s *y, __global %s *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    vstore3(%s(vload3(tid,x), vload3(tid,y) ), tid, dst);\n"
+"}\n";
+
+const char *binary_fn_code_pattern_v3_scalar =
+"%s\n" /* optional pragma */
+"__kernel void test_fn(__global %s *x, __global %s *y, __global %s *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    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 )
+{
+    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;
+
+      program = (cl_program*)malloc(sizeof(cl_program)*kTotalVecCount*2);
+      kernel = (cl_kernel*)malloc(sizeof(cl_kernel)*kTotalVecCount*2);
+
+    num_elements = n_elems * (1 << (kTotalVecCount-1));
+
+    int test_double = 0;
+    if(is_extension_available( device, "cl_khr_fp64" ))
+    {
+        log_info("Testing doubles.\n");
+        test_double = 1;
+    }
+
+    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);
+
+    for( i = 0; i < 3; i++ )
+    {
+        streams[ i ] = clCreateBuffer( context, (cl_mem_flags)(CL_MEM_READ_WRITE),  sizeof(cl_float) * num_elements, NULL, &err );
+        test_error( err, "clCreateBuffer failed");
+    }
+
+    if (test_double)
+        for( i = 3; i < 6; i++ )
+        {
+          streams[ i ] = clCreateBuffer( context, (cl_mem_flags)(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)
+        {
+            input_ptr_double[0][j] = get_random_double(-0x20000000, 0x20000000, d);
+            input_ptr_double[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 );
+          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 );
+            }
+        } else  {
+            // do regular
+            sprintf( programSrc, binary_fn_code_pattern, "", "float", vecSizeNames[ i ], "float", vectorSecondParam ? vecSizeNames[ i ] : "", "float", vecSizeNames[ i ], fnName );
+        }
+        const char *ptr = programSrc;
+        err = create_single_kernel_helper( context, &program[ i ], &kernel[ i ], 1, &ptr, "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 ] );
+            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[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 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])) ) )
+        {
+            log_error(" float%d%s test failed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", float");
+            err = -1;
+        }
+        else
+        {
+            log_info(" float%d%s test passed\n", ((g_arrVecSizes[i])), vectorSecondParam ? "" : ", float");
+            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;
+}
+
+

test_conformance/commonfns/test_clamp.c

@@ -0,0 +1,311 @@
+//
+// 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 "../../test_common/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
+
+#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" \
+    "{\n"                                \
+    "    int  tid = get_global_id(0);\n"                \
+    "\n"                                \
+    "    dst[tid] = clamp(x[tid], minval[tid], maxval[tid]);\n"    \
+    "}\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" \
+    "{\n"                                \
+    "    int  tid = get_global_id(0);\n"                \
+    "\n"                                \
+    "    dst[tid] = clamp(x[tid], minval[tid], maxval[tid]);\n"    \
+    "}\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" \
+    "{\n"                                \
+    "    int  tid = get_global_id(0);\n"                \
+    "\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 )
+CLAMP_KERNEL_V( float, 4 )
+CLAMP_KERNEL_V( float, 8 )
+CLAMP_KERNEL_V( float, 16 )
+CLAMP_KERNEL_V3( float, 3)
+#undef EMIT_PRAGMA_DIRECTIVE
+
+#define EMIT_PRAGMA_DIRECTIVE "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
+CLAMP_KERNEL( double )
+CLAMP_KERNEL_V( double, 2 )
+CLAMP_KERNEL_V( double, 4 )
+CLAMP_KERNEL_V( double, 8 )
+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 };
+
+static int verify_clamp(float *x, float *minval, float *maxval, float *outptr, int n)
+{
+    float       t;
+    int         i;
+
+    for (i=0; i<n; i++)
+    {
+        t = fminf( fmaxf( 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;
+}
+
+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++ )
+    {
+        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);
+
+    // why does this go from 0 to 3?
+    for( i = 0; i < 4; i++ )
+    {
+        streams[ i ] = clCreateBuffer( context, (cl_mem_flags)(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_flags)(CL_MEM_READ_WRITE),  sizeof(cl_double) * num_elements, NULL, NULL );
+        if (!streams[0])
+            {
+            log_error("clCreateBuffer failed\n");
+            return -1;
+            }
+        }
+
+    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);
+        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);
+        }
+    }
+    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 );
+        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++ )
+    {
+        for( j = 0; j < 4; j++ )
+        {
+            err = clSetKernelArg( kernel[ i ], j, sizeof( streams[ j ] ), &streams[ j ] );
+            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[3], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 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]))))
+        {
+            log_error("CLAMP float%d test failed\n", ((g_arrVecSizes[i])));
+            err = -1;
+        }
+        else
+        {
+            log_info("CLAMP float%d test passed\n", ((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);
+    }
+
+    return err;
+}
+
+

test_conformance/commonfns/test_degrees.c

@@ -0,0 +1,472 @@
+//
+// 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 "../../test_common/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;
+    void        *values[2];
+    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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[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;
+    void        *values[2];
+    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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[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.c

@@ -0,0 +1,235 @@
+//
+// 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 "../../test_common/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;
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    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.c

@@ -0,0 +1,243 @@
+//
+// 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 "../../test_common/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;
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    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.c

@@ -0,0 +1,239 @@
+//
+// 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 "../../test_common/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;
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    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.c

@@ -0,0 +1,237 @@
+//
+// 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 "../../test_common/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;
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    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.c

@@ -0,0 +1,60 @@
+//
+// 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 "../../test_common/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.c

@@ -0,0 +1,64 @@
+//
+// 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 "../../test_common/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.c

@@ -0,0 +1,56 @@
+//
+// 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 "../../test_common/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.c

@@ -0,0 +1,70 @@
+//
+// 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 "../../test_common/harness/compat.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include "procs.h"
+#include "../../test_common/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.c

@@ -0,0 +1,195 @@
+//
+// 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 "../../test_common/harness/compat.h"
+
+#include <stdio.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include "procs.h"
+
+const char *mix_kernel_code =
+"__kernel void test_mix(__global float *srcA, __global float *srcB, __global float *srcC, __global float *dst)\n"
+"{\n"
+"    int  tid = get_global_id(0);\n"
+"\n"
+"    dst[tid] = mix(srcA[tid], srcB[tid], srcC[tid]);\n"
+"}\n";
+
+#define MAX_ERR 1e-3
+
+float
+verify_mix(float *inptrA, float *inptrB, float *inptrC, float *outptr, int n)
+{
+    float       r, delta, max_err = 0.0f;
+    int         i;
+
+    for (i=0; i<n; i++)
+    {
+        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)
+{
+    cl_mem            streams[4];
+    cl_float        *input_ptr[3], *output_ptr, *p;
+    cl_program        program;
+    cl_kernel        kernel;
+    void            *values[4];
+    size_t            lengths[1];
+    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_flags)(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_flags)(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_flags)(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_flags)(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 );
+    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;
+    }
+
+    lengths[0] = strlen(mix_kernel_code);
+    err = create_single_kernel_helper( context, &program, &kernel, 1, &mix_kernel_code, "test_mix" );
+    test_error( err, "Unable to create test kernel" );
+
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    values[3] = streams[3];
+  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)
+    {
+        log_error("clSetKernelArgs failed\n");
+        return -1;
+    }
+
+    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;
+    }
+
+    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;
+    }
+
+    max_err = verify_mix(input_ptr[0], input_ptr[1], input_ptr[2], output_ptr, num_elements);
+    if (max_err > MAX_ERR)
+    {
+        log_error("MIX test failed %g max err\n", max_err);
+        err = -1;
+    }
+    else
+    {
+        log_info("MIX test passed %g max err\n", max_err);
+        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);
+
+    return err;
+}
+
+
+
+
+

test_conformance/commonfns/test_radians.c

@@ -0,0 +1,470 @@
+//
+// 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 "../../test_common/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;
+    void         *values[2];
+    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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[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;
+    void         *values[2];
+    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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[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.c

@@ -0,0 +1,440 @@
+//
+// 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 "../../test_common/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];
+  void        *values[2];
+  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_flags)(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_flags)(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;
+
+  values[0] = streams[0];
+  values[1] = streams[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];
+  void        *values[2];
+  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_flags)(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_flags)(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;
+
+  values[0] = streams[0];
+  values[1] = streams[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.c

@@ -0,0 +1,278 @@
+//
+// 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 "../../test_common/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_smoothstep2(__global float2 *edge0, __global float2 *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_smoothstep4(__global float4 *edge0, __global float4 *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";
+
+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;
+
+  for (i=0; i<n; i++)
+  {
+    t = (x[i] - 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[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)
+{
+  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_flags)(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_flags)(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_flags)(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_flags)(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 );
+  for (i=0; i<num_elements; i++)
+  {
+    p[i] = get_random_float(-0x00400000, 0x00400000, 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(-0x00400000, 0x00400000, 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(-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 = 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;
+
+  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)
+    {
+      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[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;
+    }
+
+    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);
+      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<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;
+}
+
+

test_conformance/commonfns/test_smoothstepf.c

@@ -0,0 +1,255 @@
+//
+// 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 "../../test_common/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)
+
+extern "C" 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_flags)(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_flags)(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_flags)(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_flags)(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.c

@@ -0,0 +1,541 @@
+//
+// 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 "../../test_common/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);
+
+
+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";
+
+const char *step2_kernel_code =
+"__kernel void test_step2(__global float2 *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";
+
+const char *step4_kernel_code =
+"__kernel void test_step4(__global float4 *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";
+
+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"
+"}\n";
+
+
+int
+verify_step(float *inptrA, float *inptrB, float *outptr, int n)
+{
+    float       r;
+    int         i;
+
+    for (i=0; i<n; i++)
+    {
+        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];
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+  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++)
+    {
+        r = (inptrB[i] < inptrA[i]) ? 0.0 : 1.0;
+        if (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];
+    void        *values[3];
+    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_flags)(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_flags)(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_flags)(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;
+
+    values[0] = streams[0];
+    values[1] = streams[1];
+    values[2] = streams[2];
+    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);
+                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;
+        }
+
+        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_stepf.c

@@ -0,0 +1,540 @@
+//
+// 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 "../../test_common/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_flags)(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_flags)(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_flags)(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_flags)(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_flags)(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_flags)(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;
+}
+