AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

P4 to Git Change 1271191 by gandryey@gera-w8 on 2016/05/19 18:42:50

Browse Source 
SWDEV-86035 - Add PAL backend to OpenCL
	- Adds SDMA worakround for pagefault
	- Removes directSRD and hsail flags
	- Fixes a BSOD with the latets PAL on Fiji. KMD expects a valid UMD client

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#315 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#225 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#345 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.hpp#96 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/src/core/os/win/winPlatform.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#7 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#3 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#3 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#254 edit
This commit is contained in:
foreman
2016-05-19 18:59:20 -04:00
parent 4e2cfbc872
commit ffcdad85e5
11 changed files with 114 additions and 162 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/pal/paldevice.cpp
+39 -54
View File
@@ -149,19 +149,14 @@ NullDevice::create(Pal::GfxIpLevel ipLevel)
device::Program*
NullDevice::createProgram(amd::option::Options* options)
{
device::Program* nullProgram;
if (settings().hsail_) {
nullProgram = new HSAILProgram(*this);
}
else {
// AMDIL path
ShouldNotReachHere();
}
if (nullProgram == nullptr) {
device::Program* program;
program = new HSAILProgram(*this);
if (program == nullptr) {
LogError("Memory allocation has failed!");
}
return nullProgram;
return program;
}
void NullDevice::fillDeviceInfo(
@@ -643,19 +638,26 @@ Device::create(Pal::IDevice* device)
// Update HW info for the device
hwInfo_ = &DeviceInfo[static_cast<uint>(properties().revision)];
// Find the number of available engines
numComputeEngines_ =
properties().engineProperties[Pal::QueueTypeCompute].engineCount;
numDmaEngines_ =
properties().engineProperties[Pal::QueueTypeDma].engineCount;
Pal::PalPublicSettings*const palSettings = iDev()->GetPublicSettings();
// Modify settings here
// palSettings ...
palSettings->textureOptLevel = Pal::TextureFilterOptimizationsDisabled;
// Commit the new settings for the device
result = iDev()->CommitSettingsAndInit();
if (result == Pal::Result::Success) {
Pal::DeviceFinalizeInfo finalizeInfo = {};
// Request 2 compute engines
finalizeInfo.engineCounts[Pal::QueueTypeCompute] = 2;
// Request 2 SDMA engines
finalizeInfo.engineCounts[Pal::QueueTypeDma] = 2;
// Request all compute engines
finalizeInfo.engineCounts[Pal::QueueTypeCompute] = numComputeEngines_;
// Request all SDMA engines
finalizeInfo.engineCounts[Pal::QueueTypeDma] = numDmaEngines_;
result = iDev()->Finalize(finalizeInfo);
}
@@ -670,12 +672,6 @@ Device::create(Pal::IDevice* device)
appProfile_.reportAsOCL12Device())) {
return false;
}
// Find the number of available engines
numComputeEngines_ =
properties().engineProperties[Pal::QueueTypeCompute].engineCount;
numDmaEngines_ =
properties().engineProperties[Pal::QueueTypeDma].engineCount;
numComputeEngines_ = std::min(numComputeEngines_, settings().numComputeRings_);
amd::Context::Info info = {0};
@@ -816,22 +812,20 @@ Device::initializeHeapResources()
}
// Delay compilation due to brig_loader memory allocation
if (settings().hsail_ || (settings().oclVersion_ == OpenCL20)) {
const char* scheduler = nullptr;
const char* ocl20 = nullptr;
if (settings().oclVersion_ == OpenCL20) {
scheduler = SchedulerSourceCode;
ocl20 = "-cl-std=CL2.0";
}
blitProgram_ = new BlitProgram(context_);
// Create blit programs
if (blitProgram_ == nullptr ||
!blitProgram_->create(this, scheduler, ocl20)) {
delete blitProgram_;
blitProgram_ = nullptr;
LogError("Couldn't create blit kernels!");
return false;
}
const char* scheduler = nullptr;
const char* ocl20 = nullptr;
if (settings().oclVersion_ == OpenCL20) {
scheduler = SchedulerSourceCode;
ocl20 = "-cl-std=CL2.0";
}
blitProgram_ = new BlitProgram(context_);
// Create blit programs
if (blitProgram_ == nullptr ||
!blitProgram_->create(this, scheduler, ocl20)) {
delete blitProgram_;
blitProgram_ = nullptr;
LogError("Couldn't create blit kernels!");
return false;
}
// Create a synchronized transfer queue
@@ -900,20 +894,13 @@ Device::createVirtualDevice(
device::Program*
Device::createProgram(amd::option::Options* options)
{
device::Program* gpuProgram;
if (settings().hsail_) {
gpuProgram = new HSAILProgram(*this);
}
else {
ShouldNotReachHere();
//AMDIL
//gpuProgram = new Program(*this);
}
if (gpuProgram == nullptr) {
device::Program* program;
program = new HSAILProgram(*this);
if (program == nullptr) {
LogError("We failed memory allocation for program!");
}
return gpuProgram;
return program;
}
//! Requested devices list as configured by the GPU_DEVICE_ORDINAL
@@ -1410,14 +1397,12 @@ bool
Device::createSampler(const amd::Sampler& owner, device::Sampler** sampler) const
{
*sampler = nullptr;
if (settings().hsail_ || (settings().oclVersion_ >= OpenCL20)) {
Sampler* gpuSampler = new Sampler(*this);
if ((nullptr == gpuSampler) || !gpuSampler->create(owner)) {
delete gpuSampler;
return false;
}
*sampler = gpuSampler;
Sampler* gpuSampler = new Sampler(*this);
if ((nullptr == gpuSampler) || !gpuSampler->create(owner)) {
delete gpuSampler;
return false;
}
*sampler = gpuSampler;
return true;
}