foreman
b6c34a714c
ECR #354633 - SPIR-V: Rename clCreateProgramWithIL as clCreateProgramWithILKHR and define it as an extension API. Add API to OCL 1.2/2.1. Add cl_khr_il_program extension. Fix SPIR-V conformance and ocltst. Affected files ... ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.def.in#15 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.map.in#18 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_context.cpp#44 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_icd.cpp#24 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_icd_amd.h#17 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_program.cpp#37 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl1.2/CL/cl_ext.h#9 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl2.0/CL/cl.h#21 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl2.0/CL/cl_ext.h#16 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/headers/opencl2.1/CL/cl.h#3 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/icd/OpenCL.def.in#12 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/icd/icd_dispatch.c#28 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/icd/icd_dispatch.h#23 edit ... //depot/stg/opencl/drivers/opencl/api/opencl/khronos/icd/icd_exports.map.in#8 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpusettings.cpp#30 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#252 edit ... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsasettings.cpp#37 edit ... //depot/stg/opencl/drivers/opencl/tests/conformance/devel/2.0/test_conformance/spirv/main.cpp#7 edit ... //depot/stg/opencl/drivers/opencl/tests/conformance/devel/2.0/test_conformance/spirv/run_services.cpp#3 edit ... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/spir/SPIRVBasic.cpp#3 edit
82 lines
2.0 KiB
C++
82 lines
2.0 KiB
C++
//
|
|
// Copyright (c) 2010 Advanced Micro Devices, Inc. All rights reserved.
|
|
//
|
|
|
|
#ifndef WITHOUT_GPU_BACKEND
|
|
|
|
#include "top.hpp"
|
|
#include "os/os.hpp"
|
|
#include "device/device.hpp"
|
|
#include "hsasettings.hpp"
|
|
|
|
namespace oclhsa {
|
|
|
|
Settings::Settings()
|
|
{
|
|
// Initialize the HSA device default settings
|
|
|
|
// Set this to true when we drop the flag
|
|
doublePrecision_ = ::CL_KHR_FP64;
|
|
pollCompletion_ = ENVVAR_HSA_POLL_KERNEL_COMPLETION;
|
|
|
|
// Enable "local" memory in HSA
|
|
enableLocalMemory_ = HSA_LOCAL_MEMORY_ENABLE;
|
|
enableSvm32BitsAtomics_ = HSA_ENABLE_ATOMICS_32B;
|
|
|
|
maxWorkGroupSize_ = 256;
|
|
maxWorkGroupSize2DX_ = 16;
|
|
maxWorkGroupSize2DY_ = 16;
|
|
maxWorkGroupSize3DX_ = 4;
|
|
maxWorkGroupSize3DY_ = 4;
|
|
maxWorkGroupSize3DZ_ = 4;
|
|
}
|
|
|
|
bool
|
|
Settings::create(bool doublePrecision)
|
|
{
|
|
largeHostMemAlloc_ = true;
|
|
customHostAllocator_ = true;
|
|
|
|
// Enable extensions
|
|
enableExtension(ClKhrByteAddressableStore);
|
|
enableExtension(ClKhrGlobalInt32BaseAtomics);
|
|
enableExtension(ClKhrGlobalInt32ExtendedAtomics);
|
|
enableExtension(ClKhrLocalInt32BaseAtomics);
|
|
enableExtension(ClKhrLocalInt32ExtendedAtomics);
|
|
enableExtension(ClExtAtomicCounters32);
|
|
//enableExtension(ClKhr3DImageWrites);
|
|
enableExtension(ClKhrGlSharing);
|
|
enableExtension(ClAmdMediaOps);
|
|
enableExtension(ClAmdMediaOps2);
|
|
#if defined(_WIN32)
|
|
enableExtension(ClKhrD3d10Sharing);
|
|
enableExtension(ClKhrD3d11Sharing);
|
|
#endif // _WIN32
|
|
//enableExtension(ClKhrImage2dFromBuffer);
|
|
//enableExtension(ClAmdImage2dFromBufferReadOnly);
|
|
// Make sure device supports doubles
|
|
doublePrecision_ &= doublePrecision;
|
|
|
|
if (doublePrecision_) {
|
|
// Enable KHR double precision extension
|
|
enableExtension(ClKhrFp64);
|
|
// Also enable AMD double precision extension?
|
|
enableExtension(ClAmdFp64);
|
|
}
|
|
enableExtension(ClKhrIlProgram);
|
|
|
|
// Override current device settings
|
|
override();
|
|
|
|
return true;
|
|
}
|
|
|
|
void
|
|
Settings::override()
|
|
{
|
|
}
|
|
|
|
} // namespace oclhsa
|
|
|
|
#endif // WITHOUT_GPU_BACKEND
|