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

P4 to Git Change 1748674 by cpaquot@cpaquot-ocl-lc-lnx on 2019/02/26 11:31:59

Browse Source 
SWDEV-145570 - [HIP] Implemented texture object for ROCm backend
	Needed to implement sampler object and return the getHsaImageObject
	for roc::Image::cpuSrd.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#118 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#34 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.hpp#15 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#73 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#22 edit


[ROCm/clr commit: 2732ad0300]
This commit is contained in:
foreman
2019-02-26 11:47:55 -05:00
parent 5aab4688e9
commit 1d0572dd0f
5 changed files with 92 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/rocm/rocvirtual.cpp
+2 -68
View File
@@ -186,35 +186,6 @@ void VirtualGPU::MemoryDependency::clear(bool all) {
}
}
static void fillSampleDescriptor(hsa_ext_sampler_descriptor_t& samplerDescriptor,
const amd::Sampler& sampler) {
samplerDescriptor.filter_mode = sampler.filterMode() == CL_FILTER_NEAREST
? HSA_EXT_SAMPLER_FILTER_MODE_NEAREST
: HSA_EXT_SAMPLER_FILTER_MODE_LINEAR;
samplerDescriptor.coordinate_mode = sampler.normalizedCoords()
? HSA_EXT_SAMPLER_COORDINATE_MODE_NORMALIZED
: HSA_EXT_SAMPLER_COORDINATE_MODE_UNNORMALIZED;
switch (sampler.addressingMode()) {
case CL_ADDRESS_CLAMP_TO_EDGE:
samplerDescriptor.address_mode = HSA_EXT_SAMPLER_ADDRESSING_MODE_CLAMP_TO_EDGE;
break;
case CL_ADDRESS_REPEAT:
samplerDescriptor.address_mode = HSA_EXT_SAMPLER_ADDRESSING_MODE_REPEAT;
break;
case CL_ADDRESS_CLAMP:
samplerDescriptor.address_mode = HSA_EXT_SAMPLER_ADDRESSING_MODE_CLAMP_TO_BORDER;
break;
case CL_ADDRESS_MIRRORED_REPEAT:
samplerDescriptor.address_mode = HSA_EXT_SAMPLER_ADDRESSING_MODE_MIRRORED_REPEAT;
break;
case CL_ADDRESS_NONE:
samplerDescriptor.address_mode = HSA_EXT_SAMPLER_ADDRESSING_MODE_UNDEFINED;
break;
default:
return;
}
}
bool VirtualGPU::processMemObjects(const amd::Kernel& kernel, const_address params, size_t& ldsAddress) {
Kernel& hsaKernel = const_cast<Kernel&>(static_cast<const Kernel&>(*(kernel.getDeviceKernel(dev()))));
const amd::KernelSignature& signature = kernel.signature();
@@ -395,36 +366,10 @@ bool VirtualGPU::processMemObjects(const amd::Kernel& kernel, const_address para
const amd::Sampler* sampler = reinterpret_cast<amd::Sampler* const*>(params +
kernelParams.samplerObjOffset())[index];
hsa_ext_sampler_descriptor_t samplerDescriptor;
fillSampleDescriptor(samplerDescriptor, *sampler);
device::Sampler* devSampler = sampler->getDeviceSampler(dev());
hsa_ext_sampler_t hsa_sampler;
hsa_status_t status =
hsa_ext_sampler_create(dev().getBackendDevice(), &samplerDescriptor, &hsa_sampler);
if (status != HSA_STATUS_SUCCESS) {
// Wait on a kernel if one is outstanding
releaseGpuMemoryFence();
// Release the sampler handles allocated for the various
// on one or more kernel submissions
for (const auto& it: samplerList_) {
if (hsa_ext_sampler_destroy(gpu_device_, it) != HSA_STATUS_SUCCESS) {
LogWarning("Error destroying device sampler object!");
}
}
samplerList_.clear();
status = hsa_ext_sampler_create(dev().getBackendDevice(), &samplerDescriptor, &hsa_sampler);
if (status != HSA_STATUS_SUCCESS) {
LogError("Error creating device sampler object!");
return false;
}
}
uint64_t sampler_srd = hsa_sampler.handle;
uint64_t sampler_srd = devSampler->hwSrd();
WriteAqlArgAt(const_cast<address>(params), &sampler_srd, sizeof(sampler_srd), desc.offset_);
samplerList_.push_back(hsa_sampler);
// TODO: destroy sampler.
}
}
@@ -896,17 +841,6 @@ void VirtualGPU::updateCommandsState(amd::Command* list) {
current->release();
current = next;
}
// Release the sampler handles allocated for the various
// on one or more kernel submissions
for (const auto& it: samplerList_) {
if (hsa_ext_sampler_destroy(gpu_device_, it) != HSA_STATUS_SUCCESS) {
LogWarning("Error destroying device sampler object!");
}
}
samplerList_.clear();
return;
}
void VirtualGPU::submitReadMemory(amd::ReadMemoryCommand& cmd) {