Files
OpenCL-CTS/test_conformance/compiler/test_build_options.cpp
Romaric Jodin df46a38e31 Better support for -cl-uniform-work-group-size (#2564)
Add support for build options available only after a certain version:
- -cl-uniform-work-group-size after 2.0
- -cl-no-subgroup-ifp after 2.1

Add specific test for cl-uniform-work-group-size
- Check that test can be executed when work group size is uniform.
- Check that test returns the proper error code when work group size is
not uniform.

Ref #2563
2025-11-26 16:19:33 -08:00

474 lines
17 KiB
C++

//
// Copyright (c) 2017 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "testBase.h"
#include "harness/os_helpers.h"
#include "harness/testHarness.h"
#include <array>
const char *preprocessor_test_kernel[] = {
"__kernel void sample_test(__global int *dst)\n"
"{\n"
" dst[0] = TEST_MACRO;\n"
"\n"
"}\n" };
const char *preprocessor_existence_test_kernel[] = {
"__kernel void sample_test(__global int *dst)\n"
"{\n"
"#ifdef TEST_MACRO\n"
" dst[0] = 42;\n"
"#else\n"
" dst[0] = 24;\n"
"#endif\n"
"\n"
"}\n" };
const char *include_test_kernel[] = {
"#include \"./testIncludeFile.h\"\n"
"__kernel void sample_test(__global int *dst)\n"
"{\n"
" dst[0] = HEADER_FOUND;\n"
"\n"
"}\n" };
const char *options_test_kernel[] = { "__kernel void sample_test() {}\n" };
std::array optimization_options{
std::pair{ "-cl-single-precision-constant", Version(1, 0) },
std::pair{ "-cl-denorms-are-zero", Version(1, 0) },
std::pair{ "-cl-opt-disable", Version(1, 0) },
std::pair{ "-cl-mad-enable", Version(1, 0) },
std::pair{ "-cl-no-signed-zeros", Version(1, 0) },
std::pair{ "-cl-unsafe-math-optimizations", Version(1, 0) },
std::pair{ "-cl-finite-math-only", Version(1, 0) },
std::pair{ "-cl-fast-relaxed-math", Version(1, 0) },
std::pair{ "-w", Version(1, 0) },
std::pair{ "-Werror", Version(1, 0) },
std::pair{ "-cl-uniform-work-group-size", Version(2, 0) },
std::pair{ "-cl-no-subgroup-ifp", Version(2, 1) },
};
cl_int get_result_from_program( cl_context context, cl_command_queue queue, cl_program program, cl_int *outValue )
{
cl_int error;
clKernelWrapper kernel = clCreateKernel( program, "sample_test", &error );
test_error( error, "Unable to create kernel from program" );
clMemWrapper outStream;
outStream = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_int), NULL,
&error);
test_error( error, "Unable to create test buffer" );
error = clSetKernelArg( kernel, 0, sizeof( outStream ), &outStream );
test_error( error, "Unable to set kernel argument" );
size_t threads[1] = { 1 };
error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, NULL, 0, NULL, NULL );
test_error( error, "Unable to execute test kernel" );
error = clEnqueueReadBuffer( queue, outStream, true, 0, sizeof( cl_int ), outValue, 0, NULL, NULL );
test_error( error, "Unable to read output array!" );
return CL_SUCCESS;
}
REGISTER_TEST(options_build_optimizations)
{
int error;
cl_build_status status;
Version version = get_device_cl_version(device);
for (const auto &optimization_option : optimization_options)
{
if (version < optimization_option.second)
{
continue;
}
const char *option = optimization_option.first;
clProgramWrapper program;
error = create_single_kernel_helper_create_program(
context, &program, 1, options_test_kernel, option);
if (program == NULL || error != CL_SUCCESS)
{
log_error("ERROR: Unable to create reference program!\n");
return -1;
}
/* Build with the macro defined */
log_info("Testing optimization option '%s'\n", option);
error = clBuildProgram(program, 1, &device, option, NULL, NULL);
test_error(error, "Test program did not properly build");
error = clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_STATUS,
sizeof(status), &status, NULL);
test_error(error, "Unable to get program build status");
if ((int)status != CL_BUILD_SUCCESS)
{
log_info(
"Building with optimization option '%s' failed to compile!\n",
option);
print_error(error,
"Failed to build with optimization defined") return -1;
}
}
return 0;
}
REGISTER_TEST(options_build_macro)
{
int error;
clProgramWrapper program;
cl_build_status status;
program = clCreateProgramWithSource( context, 1, preprocessor_test_kernel, NULL, &error );
if( program == NULL || error != CL_SUCCESS )
{
log_error( "ERROR: Unable to create reference program!\n" );
return -1;
}
/* Build with the macro defined */
error = clBuildProgram(program, 1, &device, "-DTEST_MACRO=1 ", NULL, NULL);
test_error( error, "Test program did not properly build" );
error = clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_STATUS,
sizeof(status), &status, NULL);
test_error( error, "Unable to get program build status" );
if( (int)status != CL_BUILD_SUCCESS )
{
print_error( error, "Failed to build with macro defined" );
return -1;
}
// Go ahead and run the program to verify results
cl_int firstResult, secondResult;
error = get_result_from_program( context, queue, program, &firstResult );
test_error( error, "Unable to get result from first program" );
if( firstResult != 1 )
{
log_error( "ERROR: Result from first program did not validate! (Expected 1, got %d)\n", firstResult );
return -1;
}
// Rebuild with a different value for the define macro, to make sure caching behaves properly
error = clBuildProgram(program, 1, &device, "-DTEST_MACRO=5 ", NULL, NULL);
test_error( error, "Test program did not properly rebuild" );
error = get_result_from_program( context, queue, program, &secondResult );
test_error( error, "Unable to get result from second program" );
if( secondResult != 5 )
{
if( secondResult == firstResult )
log_error( "ERROR: Program result did not change with device macro change (program was not recompiled)!\n" );
else
log_error( "ERROR: Result from second program did not validate! (Expected 5, got %d)\n", secondResult );
return -1;
}
return 0;
}
REGISTER_TEST(options_build_macro_existence)
{
int error;
clProgramWrapper program;
// In this case, the program should still run without the macro, but it should give a different result
program = clCreateProgramWithSource( context, 1, preprocessor_existence_test_kernel, NULL, &error );
if( program == NULL || error != CL_SUCCESS )
{
log_error( "ERROR: Unable to create reference program!\n" );
return -1;
}
/* Build without the macro defined */
error = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
test_error( error, "Test program did not properly build" );
// Go ahead and run the program to verify results
cl_int firstResult, secondResult;
error = get_result_from_program( context, queue, program, &firstResult );
test_error( error, "Unable to get result from first program" );
if( firstResult != 24 )
{
log_error( "ERROR: Result from first program did not validate! (Expected 24, got %d)\n", firstResult );
return -1;
}
// Now compile again with the macro defined and verify a change in results
error = clBuildProgram(program, 1, &device, "-DTEST_MACRO", NULL, NULL);
test_error( error, "Test program did not properly build" );
error = get_result_from_program( context, queue, program, &secondResult );
test_error( error, "Unable to get result from second program" );
if( secondResult != 42 )
{
if( secondResult == firstResult )
log_error( "ERROR: Program result did not change with device macro addition (program was not recompiled)!\n" );
else
log_error( "ERROR: Result from second program did not validate! (Expected 42, got %d)\n", secondResult );
return -1;
}
return 0;
}
REGISTER_TEST(options_include_directory)
{
int error;
std::string sep = dir_sep();
std::string path = exe_dir(); // Directory where test executable is located.
std::string include_dir;
clProgramWrapper program;
cl_build_status status;
/* Try compiling the program first without the directory included Should fail. */
program = clCreateProgramWithSource( context, 1, include_test_kernel, NULL, &error );
if( program == NULL || error != CL_SUCCESS )
{
log_error( "ERROR: Unable to create reference program!\n" );
return -1;
}
/* Build with the include directory defined */
include_dir = "-I " + path + sep + "includeTestDirectory";
// log_info("%s\n", include_dir);
error =
clBuildProgram(program, 1, &device, include_dir.c_str(), NULL, NULL);
test_error( error, "Test program did not properly build" );
error = clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_STATUS,
sizeof(status), &status, NULL);
test_error( error, "Unable to get program build status" );
if( (int)status != CL_BUILD_SUCCESS )
{
print_error( error, "Failed to build with include directory" );
return -1;
}
// Go ahead and run the program to verify results
cl_int firstResult, secondResult;
error = get_result_from_program( context, queue, program, &firstResult );
test_error( error, "Unable to get result from first program" );
if( firstResult != 12 )
{
log_error( "ERROR: Result from first program did not validate! (Expected 12, got %d)\n", firstResult );
return -1;
}
// Rebuild with a different include directory
include_dir = "-I " + path + sep + "secondIncludeTestDirectory";
error =
clBuildProgram(program, 1, &device, include_dir.c_str(), NULL, NULL);
test_error( error, "Test program did not properly rebuild" );
error = get_result_from_program( context, queue, program, &secondResult );
test_error( error, "Unable to get result from second program" );
if( secondResult != 42 )
{
if( secondResult == firstResult )
log_error( "ERROR: Program result did not change with include path change (program was not recompiled)!\n" );
else
log_error( "ERROR: Result from second program did not validate! (Expected 42, got %d)\n", secondResult );
return -1;
}
return 0;
}
const char *denorm_test_kernel[] = {
"__kernel void sample_test( float src1, float src2, __global float *dst)\n"
"{\n"
" dst[ 0 ] = src1 + src2;\n"
"\n"
"}\n" };
cl_int get_float_result_from_program( cl_context context, cl_command_queue queue, cl_program program, cl_float inA, cl_float inB, cl_float *outValue )
{
cl_int error;
clKernelWrapper kernel = clCreateKernel( program, "sample_test", &error );
test_error( error, "Unable to create kernel from program" );
clMemWrapper outStream = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(cl_float), NULL, &error);
test_error( error, "Unable to create test buffer" );
error = clSetKernelArg( kernel, 0, sizeof( cl_float ), &inA );
test_error( error, "Unable to set kernel argument" );
error = clSetKernelArg( kernel, 1, sizeof( cl_float ), &inB );
test_error( error, "Unable to set kernel argument" );
error = clSetKernelArg( kernel, 2, sizeof( outStream ), &outStream );
test_error( error, "Unable to set kernel argument" );
size_t threads[1] = { 1 };
error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, NULL, 0, NULL, NULL );
test_error( error, "Unable to execute test kernel" );
error = clEnqueueReadBuffer( queue, outStream, true, 0, sizeof( cl_float ), outValue, 0, NULL, NULL );
test_error( error, "Unable to read output array!" );
return CL_SUCCESS;
}
REGISTER_TEST(options_denorm_cache)
{
int error;
clProgramWrapper program;
cl_build_status status;
// If denorms aren't even supported, testing this flag is pointless
cl_device_fp_config floatCaps = 0;
error = clGetDeviceInfo(device, CL_DEVICE_SINGLE_FP_CONFIG,
sizeof(floatCaps), &floatCaps, NULL);
test_error( error, "Unable to get device FP config" );
if( ( floatCaps & CL_FP_DENORM ) == 0 )
{
log_info( "Device does not support denormalized single-precision floats; skipping test.\n" );
return 0;
}
program = clCreateProgramWithSource( context, 1, denorm_test_kernel, NULL, &error );
test_error( error, "Unable to create test program" );
// Build first WITH the denorm flush flag
error =
clBuildProgram(program, 1, &device, "-cl-denorms-are-zero", NULL, NULL);
test_error( error, "Test program did not properly build" );
error = clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_STATUS,
sizeof(status), &status, NULL);
test_error( error, "Unable to get program build status" );
if( (int)status != CL_BUILD_SUCCESS )
{
print_error( error, "Failed to build with include directory" );
return -1;
}
// Note: the following in floating point is a subnormal number, thus adding two of them together
// should give us a subnormalized result. If denormals are flushed to zero, however, it'll give us zero instead
uint32_t intSubnormal = 0x00000001;
cl_float *input = (cl_float *)&intSubnormal;
cl_float firstResult, secondResult;
error = get_float_result_from_program( context, queue, program, *input, *input, &firstResult );
test_error( error, "Unable to get result from first program" );
// Note: since -cl-denorms-are-zero is a HINT, not a requirement, the result we got could
// either be subnormal (hint ignored) or zero (hint respected). Since either is technically
// valid, there isn't anything we can to do validate results for now
// Rebuild without flushing flag set
error = clBuildProgram(program, 1, &device, NULL, NULL, NULL);
test_error( error, "Test program did not properly rebuild" );
error = get_float_result_from_program( context, queue, program, *input, *input, &secondResult );
test_error( error, "Unable to get result from second program" );
// Now, there are three possiblities here:
// 1. The denorms-are-zero hint is not respected, in which case the first and second result will be identical
// 2. The hint is respected, and the program was properly rebuilt, in which case the first result will be zero and the second non-zero
// 3. The hint is respected, but the program was not properly rebuilt, in which case both results will be zero
// 3 is the only error condition we need to look for
uint32_t *fPtr = (uint32_t *)&firstResult;
uint32_t *sPtr = (uint32_t *)&secondResult;
if( ( *fPtr == 0 ) && ( *sPtr == 0 ) )
{
log_error( "ERROR: Program result didn't change when -cl-denorms-are-zero flag was removed.\n"
"First result (should be zero): 0x%08x, Second result (should be non-zero): 0x%08x\n",
*fPtr, *sPtr );
return -1;
}
return 0;
}
REGISTER_TEST(options_uniform_work_group_size)
{
if (get_device_cl_version(device) < Version(2, 0))
{
return TEST_SKIPPED_ITSELF;
}
const char *options = "-cl-uniform-work-group-size";
clProgramWrapper program;
int error = create_single_kernel_helper_create_program(
context, &program, 1, options_test_kernel, options);
if (program == NULL || error != CL_SUCCESS)
{
log_error("Error: Unable to create reference program!\n");
return TEST_FAIL;
}
error = clBuildProgram(program, 1, &device, options, NULL, NULL);
test_error(error, "Test program did not properly build");
clKernelWrapper kernel = clCreateKernel(program, "sample_test", &error);
test_error(error, "Unable to create kernel");
size_t global_work_size = 4;
size_t uniform_local_work_size = 2;
error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global_work_size,
&uniform_local_work_size, 0, NULL, NULL);
test_error(error,
"Unable to enqueue NDRange kernel with uniform work group size");
error = clFinish(queue);
test_error(error, "Unable to finish");
size_t non_uniform_local_work_size = 3;
error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global_work_size,
&non_uniform_local_work_size, 0, NULL, NULL);
if (error != CL_INVALID_WORK_GROUP_SIZE)
{
log_error(
"Error: expected error 'CL_INVALID_WORK_GROUP_SIZE' (got '%s') "
"trying to enqueue kernel compiled with '%s' with non-uniform work "
"group size\n",
IGetErrorString(error), options);
return TEST_FAIL;
}
return TEST_PASS;
}