P4 to Git Change 1294024 by lmoriche@lmoriche_opencl_dev on 2016/07/21 12:33:32
SWDEV-94640 - Re-submit CL#1292110 after integration from fsa_foundation:
- [OCL-LC-ROCm] OpenCL Runtime Library Implements OpenCL runtime API. Add HSA virtual device to ORCA.
Rename hsa_foundation to ROCm
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/Makefile#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/build/Makefile#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/build/Makefile.rocm#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/mesa_glinterop.h#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocappprofile.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocappprofile.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocbinary.cpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocbinary.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocblit.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompilerlib.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompilerlib.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#4 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#4 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocglinterop.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocglinterop.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roclib.cpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roclib.h#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprintf.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprintf.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocregisters.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.hpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#3 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#3 add
[ROCm/clr commit: 5ba83829e2]
Dieser Commit ist enthalten in:
@@ -0,0 +1,100 @@
|
||||
//
|
||||
// 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 "rocsettings.hpp"
|
||||
#include "device/rocm/rocglinterop.hpp"
|
||||
|
||||
namespace roc {
|
||||
|
||||
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;
|
||||
|
||||
enableLocalMemory_ = HSA_LOCAL_MEMORY_ENABLE;
|
||||
enableImageHandle_ = true;
|
||||
|
||||
maxWorkGroupSize_ = 256;
|
||||
maxWorkGroupSize2DX_ = 16;
|
||||
maxWorkGroupSize2DY_ = 16;
|
||||
maxWorkGroupSize3DX_ = 4;
|
||||
maxWorkGroupSize3DY_ = 4;
|
||||
maxWorkGroupSize3DZ_ = 4;
|
||||
|
||||
kernargPoolSize_ = HSA_KERNARG_POOL_SIZE;
|
||||
signalPoolSize_ = HSA_SIGNAL_POOL_SIZE;
|
||||
|
||||
// Determine if user is requesting Non-Coherent mode
|
||||
// for system memory. By default system memory is
|
||||
// operates or is programmed to be in Coherent mode.
|
||||
// Users can turn it off for hardware that does not
|
||||
// support this feature naturally
|
||||
char *nonCoherentMode = NULL;
|
||||
nonCoherentMode = getenv("OPENCL_USE_NC_MEMORY_POLICY");
|
||||
enableNCMode_ = (nonCoherentMode)? true : false;
|
||||
|
||||
// Determine if user wishes to disable support for
|
||||
// partial dispatch. By default support for partial
|
||||
// dispatch is enabled. Users can turn it off for
|
||||
// devices that do not support this feature.
|
||||
//
|
||||
// @note Update appropriate field of device::Settings
|
||||
char *partialDispatch = NULL;
|
||||
partialDispatch = getenv("OPENCL_DISABLE_PARTIAL_DISPATCH");
|
||||
enablePartialDispatch_ = (partialDispatch) ? false : true;
|
||||
partialDispatch_ = (partialDispatch) ? false : true;
|
||||
}
|
||||
|
||||
bool
|
||||
Settings::create(bool doublePrecision)
|
||||
{
|
||||
customHostAllocator_ = true;
|
||||
|
||||
// Enable extensions
|
||||
enableExtension(ClKhrByteAddressableStore);
|
||||
enableExtension(ClKhrGlobalInt32BaseAtomics);
|
||||
enableExtension(ClKhrGlobalInt32ExtendedAtomics);
|
||||
enableExtension(ClKhrLocalInt32BaseAtomics);
|
||||
enableExtension(ClKhrLocalInt32ExtendedAtomics);
|
||||
enableExtension(ClKhr3DImageWrites);
|
||||
enableExtension(ClAmdMediaOps);
|
||||
enableExtension(ClAmdMediaOps2);
|
||||
if(MesaInterop::Supported())
|
||||
enableExtension(ClKhrGlSharing);
|
||||
|
||||
// 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(ClKhrDepthImages);
|
||||
supportDepthsRGB_ = true;
|
||||
|
||||
// Override current device settings
|
||||
override();
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
void
|
||||
Settings::override()
|
||||
{
|
||||
}
|
||||
|
||||
} // namespace roc
|
||||
|
||||
#endif // WITHOUT_GPU_BACKEND
|
||||
In neuem Issue referenzieren
Einen Benutzer sperren