AI/rocm-systems
2
0
派生 0
代码 工单 合并请求 工作流 软件包 项目 发布 百科 活动

P4 to Git Change 1177220 by emankov@em-hsa-amd on 2015/08/05 06:08:39

浏览代码 
ECR #333753 - ORCA RT/Compiler Lib/aoc2: AMD HSA Code Object Import feature (part II) - arbitrary hidden (extra) kernargs support

	Only HSAIL path is affected. It doesn't affect blit kernels.

	To use offline by aoc2:
	aoc2 -hsacodeobject=<importing_code_object_filename> -numhiddenkernargs=<num> -cl-std=CL2.0 -march=hsail(-64) -mdevice=Bonaire <source_cl_filename>

	To use online by setting env:
	AMD_DEBUG_HSA_NUM_HIDDEN_KERNARGS=<num>

	where num >= 0. If num == 0, then no additional arguments will be added on RT for every kernel. The default value is unchanged and equal to 6 for now.

	Misc:
	+ get rid of PRE & POST defines in Compiler Lib, as they started to conflict with ugl\gl\gs\hwl\ headers with the same defines.
	+ minor copy/paste eliminations & typo fixes
	+ ocltst complib tests update

	Testing: pre check-in, manually based on ocl sdk MatrixMultiplication

	Reviewers: Brian Sumner, German Andryeyev, Nikolay Haustov, Artem Tamazov

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#72 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/hsail_be.cpp#49 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/metadata.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclDefs.h#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclEnums.h#19 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#17 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/bif_section_labels.hpp#21 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.h#20 edit
... //depot/stg/opencl/drivers/opencl/compiler/tools/aoc2/aoc2.cpp#74 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.cpp#181 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#249 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#291 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#113 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#199 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#369 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsaprogram.cpp#38 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsakernel.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsakernel.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsaprogram.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsavirtual.cpp#43 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLAssumptionCheck.cpp#43 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLEnumCheck.cpp#44 edit
这个提交包含在：
foreman
2015-08-05 06:18:33 -04:00
父节点 881eeab3b8
当前提交 81b331f4c5
共修改 14 个文件，包含 214 行新增和 93 行删除
下载 Patch 文件 下载 Diff 文件 展开所有文件 折叠所有文件
rocclr/runtime/device/gpu/gpuprogram.cpp
+18 -7
查看文件
@@ -1974,7 +1974,7 @@ HSAILProgram::getNextCompilationStageFromBinary(amd::option::Options* options) {
break;
const oclBIFSymbolStruct* symbol = findBIF30SymStruct(symOpenclCompilerOptions);
assert(symbol && "symbol not found");
std::string symName = std::string(symbol->str[PRE]) + std::string(symbol->str[POST]);
std::string symName = std::string(symbol->str[bif::PRE]) + std::string(symbol->str[bif::POST]);
size_t symSize = 0;
const void *opts = aclExtractSymbol(dev().hsaCompiler(),
binaryElf_, &symSize, aclCOMMENT, symName.c_str(), &errorCode);
@@ -2095,21 +2095,21 @@ HSAILProgram::linkImpl(amd::option::Options* options)
}
hsa_status_t status = executable_->LoadCodeObject(agent, code_object, NULL);
if (status != HSA_STATUS_SUCCESS) {
buildLog_ += "Error while HSA Loader phase: loading HSA Code Object \n";
buildLog_ += "Error while HSA Loader phase: loading HSA Code Object\n";
return false;
}
}
size_t kernelNamesSize = 0;
errorCode = aclQueryInfo(dev().hsaCompiler(), binaryElf_, RT_KERNEL_NAMES, NULL, NULL, &kernelNamesSize);
if (errorCode != ACL_SUCCESS) {
buildLog_ += "Error while Finalization phase: kernel names query from the ELF failed\n";
buildLog_ += "Error while Finalization phase: Kernel names size querying from the ELF failed\n";
return false;
}
if (!isNull() && kernelNamesSize > 0) {
char* kernelNames = new char[kernelNamesSize];
errorCode = aclQueryInfo(dev().hsaCompiler(), binaryElf_, RT_KERNEL_NAMES, NULL, kernelNames, &kernelNamesSize);
if (errorCode != ACL_SUCCESS) {
buildLog_ += "Error while Finalization phase: kernel's Metadata is corrupted in the ELF\n";
buildLog_ += "Error while Finalization phase: Kernel names querying from the ELF failed\n";
delete kernelNames;
return false;
}
@@ -2117,11 +2117,22 @@ HSAILProgram::linkImpl(amd::option::Options* options)
delete kernelNames;
std::vector<std::string>::iterator it = vKernels.begin();
bool dynamicParallelism = false;
aclMetadata md;
md.numHiddenKernelArgs = 0;
size_t sizeOfnumHiddenKernelArgs = sizeof(md.numHiddenKernelArgs);
for (it; it != vKernels.end(); ++it) {
std::string kernelName = *it;
HSAILKernel *aKernel = new HSAILKernel(kernelName, this, options->origOptionStr + hsailOptions());
std::string kernelName(*it);
std::string openclKernelName = Kernel::openclMangledName(kernelName);
errorCode = aclQueryInfo(dev().hsaCompiler(), binaryElf_, RT_NUM_KERNEL_HIDDEN_ARGS,
openclKernelName.c_str(), &md.numHiddenKernelArgs, &sizeOfnumHiddenKernelArgs);
if (errorCode != ACL_SUCCESS) {
buildLog_ += "Error while Finalization phase: Kernel extra arguments count querying from the ELF failed\n";
return false;
}
HSAILKernel *aKernel = new HSAILKernel(kernelName, this, options->origOptionStr + hsailOptions(),
md.numHiddenKernelArgs);
kernels()[kernelName] = aKernel;
amd::hsa::loader::Symbol *sym = executable_->GetSymbol("", Kernel::openclMangledName(kernelName).c_str(), agent, 0);
amd::hsa::loader::Symbol *sym = executable_->GetSymbol("", openclKernelName.c_str(), agent, 0);
if (!sym) {
LogError("Failed to get kernel ISA code");
return false;