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

P4 to Git Change 1559096 by gandryey@gera-w8 on 2018/05/24 10:34:47

Browse Source 
SWDEV-79445 - OCL generic changes and code clean-up
	- Remove obsolete logic for aliased kernels, since it was a pre-SI optimization

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#418 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#98 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#52 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.hpp#21 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/program.cpp#93 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/program.hpp#44 edit
This commit is contained in:
foreman
2018-05-24 10:45:00 -04:00
parent a245f46203
commit bcea58f07f
7 changed files with 14 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/gpu/gpuvirtual.cpp
+3 -5
View File
@@ -1879,9 +1879,8 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
GpuEvent gpuEvent;
groupOffset = 0;
// Get the GPU kernel object with optimization enabled
bool noAlias = true;
device::Kernel* devKernel = const_cast<device::Kernel*>(kernel.getDeviceKernel(dev(), noAlias));
// Get the GPU kernel object
device::Kernel* devKernel = const_cast<device::Kernel*>(kernel.getDeviceKernel(dev()));
Kernel& gpuKernelOpt = static_cast<gpu::Kernel&>(*devKernel);
if (gpuKernelOpt.hsa()) {
@@ -2922,9 +2921,8 @@ void VirtualGPU::profileEvent(EngineType engine, bool type) const {
bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address params,
bool nativeMem, std::vector<const Memory*>* memList) {
static const bool NoAlias = true;
const HSAILKernel& hsaKernel =
static_cast<const HSAILKernel&>(*(kernel.getDeviceKernel(dev(), NoAlias)));
static_cast<const HSAILKernel&>(*(kernel.getDeviceKernel(dev())));
const amd::KernelSignature& signature = kernel.signature();
const amd::KernelParameters& kernelParams = kernel.parameters();
rocclr/runtime/device/pal/palvirtual.cpp
+1 -2
View File
@@ -2905,9 +2905,8 @@ void VirtualGPU::profileEvent(EngineType engine, bool type) const {
bool VirtualGPU::processMemObjectsHSA(const amd::Kernel& kernel, const_address params,
bool nativeMem) {
static const bool NoAlias = true;
const HSAILKernel& hsaKernel =
static_cast<const HSAILKernel&>(*(kernel.getDeviceKernel(dev(), NoAlias)));
static_cast<const HSAILKernel&>(*(kernel.getDeviceKernel(dev())));
const amd::KernelSignature& signature = kernel.signature();
const amd::KernelParameters& kernelParams = kernel.parameters();
std::vector<const Memory*> memList;
rocclr/runtime/device/rocm/rocvirtual.cpp
+1 -2
View File
@@ -171,8 +171,7 @@ void VirtualGPU::MemoryDependency::clear(bool all) {
}
bool VirtualGPU::processMemObjects(const amd::Kernel& kernel, const_address params) {
static const bool NoAlias = true;
const Kernel& hsaKernel = static_cast<const Kernel&>(*(kernel.getDeviceKernel(dev(), NoAlias)));
const Kernel& hsaKernel = static_cast<const Kernel&>(*(kernel.getDeviceKernel(dev())));
const amd::KernelSignature& signature = kernel.signature();
const amd::KernelParameters& kernelParams = kernel.parameters();