ahmed/OpenCL-CTS
1
0
Fork 0
mirror of https://github.com/KhronosGroup/OpenCL-CTS.git synced 2026-03-19 06:09:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

math_brute_force: Use clKernelWrapper in threaded tests (#1562)

Browse Source 
Simplify code by relying on RAII to free resources.  Reduce code
duplication.

This commit only affects tests that use `BuildKernelInfo`, which are
the multi-threaded tests.  Another patch will deal with the
single-threaded tests, i.e., those using `BuildKernelInfo2`.

Original patch by Marco Antognini.

Signed-off-by: Marco Antognini <marco.antognini@arm.com>
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>

Signed-off-by: Marco Antognini <marco.antognini@arm.com>
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
This commit is contained in:
Sven van Haastregt
2022-11-27 21:25:32 +00:00
committed by GitHub
parent e9d2abf705
commit 6ecf824122
16 changed files with 171 additions and 444 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
test_conformance/math_brute_force/common.cpp
View File

@@ -85,6 +85,28 @@ void EmitEnableExtension(std::ostringstream &kernel, ParameterType type)
}
}
std::string GetBuildOptions(bool relaxed_mode)
{
std::ostringstream options;
if (gForceFTZ)
{
options << " -cl-denorms-are-zero";
}
if (gFloatCapabilities & CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT)
{
options << " -cl-fp32-correctly-rounded-divide-sqrt";
}
if (relaxed_mode)
{
options << " -cl-fast-relaxed-math";
}
return options.str();
}
} // anonymous namespace
std::string GetKernelName(int vector_size_index)
@@ -530,3 +552,50 @@ __kernel void )", kernel_name.c_str(), R"((__global RETTYPE_SCALAR* out,
return kernel.str();
}
cl_int BuildKernels(BuildKernelInfo &info, cl_uint job_id,
SourceGenerator generator)
{
// Generate the kernel code.
cl_uint vector_size_index = gMinVectorSizeIndex + job_id;
auto kernel_name = GetKernelName(vector_size_index);
auto source = generator(kernel_name, info.nameInCode, vector_size_index);
std::array<const char *, 1> sources{ source.c_str() };
// Create the program.
clProgramWrapper &program = info.programs[vector_size_index];
auto options = GetBuildOptions(info.relaxedMode);
int error =
create_single_kernel_helper(gContext, &program, nullptr, sources.size(),
sources.data(), nullptr, options.c_str());
if (error != CL_SUCCESS)
{
vlog_error("\t\tFAILED -- Failed to create program. (%d)\n", error);
return error;
}
// Create a kernel for each thread. cl_kernels aren't thread safe, so make
// one for every thread
auto &kernels = info.kernels[vector_size_index];
assert(kernels.empty() && "Dirty BuildKernelInfo");
kernels.resize(info.threadCount);
for (auto &kernel : kernels)
{
kernel = clCreateKernel(program, kernel_name.c_str(), &error);
if (!kernel || error != CL_SUCCESS)
{
vlog_error("\t\tFAILED -- clCreateKernel() failed: (%d)\n", error);
size_t log_size;
clGetProgramBuildInfo(program, gDevice, CL_PROGRAM_BUILD_LOG, 0,
nullptr, &log_size);
std::string buffer;
buffer.resize(log_size + 1);
clGetProgramBuildInfo(program, gDevice, CL_PROGRAM_BUILD_LOG,
log_size, &buffer[0], NULL);
vlog_error("Log: %s\n", buffer.c_str());
return error;
}
}
return CL_SUCCESS;
}