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

P4 to Git Change 1567428 by gandryey@gera-w8 on 2018/06/12 18:39:23

Browse Source 
SWDEV-79445 - OCL generic changes and code clean-up
	- Optimize setup of kernel arguments. Stage 2.
	- Add HW ABI support in the abstraction layer
	- Remove arguments parsing loop from the kernel launch. Memory processing will be responsible for dependency tracking and  patching of arguments.

	http://ocltc.amd.com/reviews/r/15122/

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/device.cpp#221 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#307 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#325 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.cpp#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#53 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#17 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.hpp#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#107 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#53 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#36 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.cpp#30 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.hpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/program.cpp#95 edit


[ROCm/clr commit: 1be400ff01]
This commit is contained in:
foreman
2018-06-12 18:57:20 -04:00
parent 028339d9be
commit 665aab7ca4
13 changed files with 430 additions and 320 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/pal/palblit.cpp
+10 -4
View File
@@ -943,24 +943,30 @@ static void setArgument(amd::Kernel* kernel, size_t index, size_t size, const vo
uint32_t uint32_value = 0;
uint64_t uint64_value = 0;
size_t argSize = desc.size_;
if (desc.type_ == T_POINTER && desc.size_ != 0) {
if ((value == NULL) || (static_cast<const cl_mem*>(value) == NULL)) {
LP64_SWITCH(uint32_value, uint64_value) = 0;
reinterpret_cast<Memory**>(kernel->parameters().values() +
kernel->parameters().memoryObjOffset())[desc.info_.arrayIndex_] = nullptr;
} else {
// convert cl_mem to amd::Memory*, return false if invalid.
LP64_SWITCH(uint32_value, uint64_value) = static_cast<uintptr_t>((
*static_cast<Memory* const*>(value))->vmAddress());
*static_cast<Memory* const*>(value))->virtualAddress());
reinterpret_cast<Memory**>(kernel->parameters().values() +
kernel->parameters().memoryObjOffset())[desc.info_.arrayIndex_] =
*static_cast<Memory* const*>(value);
// Note: Special case for image SRD, which is 64 bit always
if (LP64_SWITCH(true, false) &&
(desc.info_.oclObject_ == amd::KernelParameterDescriptor::ImageObject)) {
uint64_value = uint32_value;
argSize = sizeof(uint64_t);
}
}
} else if (desc.type_ == T_SAMPLER) {
assert(false && "No sampler support in blit manager! Use internal samplers!");
} else
switch (desc.size_) {
switch (argSize) {
case 1:
uint32_value = *static_cast<const uint8_t*>(value);
break;
@@ -977,7 +983,7 @@ static void setArgument(amd::Kernel* kernel, size_t index, size_t size, const vo
break;
}
switch (desc.size_) {
switch (argSize) {
case 0 /*local mem*/:
*static_cast<size_t*>(param) = size;
break;