AI/rocm-systems
2
0
Vork 0
Code Kwesties Pull-aanvragen Actions Paketten Projecten Publicaties Wiki Activiteit

SWDEV-277566 - HSAIL: put back the original setKernels logic for now

Bladeren bron 
In HSAIL path, kernel akc info is obtained after code object
loading, and kernel signature creation requires the akc info
when one of the kernel argument is a reference object.

Change-Id: I9cdb1dbf2c72f4620b0b6e46a88402a2473c3e97
This commit is contained in:
Jason Tang
2021-05-20 10:58:07 -04:00
bovenliggende a81756bba3
commit 8cac880779
3 gewijzigde bestanden met toevoegingen van 49 en 60 verwijderingen
Download Patch-bestand Download Diff-bestand Expand all files Collapse all files
rocclr/device/pal/palprogram.cpp
+30 -27
Bestand weergeven
@@ -246,6 +246,34 @@ inline static std::vector<std::string> splitSpaceSeparatedString(char* str) {
bool HSAILProgram::createKernels(void* binary, size_t binSize, bool useUniformWorkGroupSize,
bool internalKernel) {
#if defined(WITH_COMPILER_LIB)
// Stop compilation if it is an offline device - PAL runtime does not
// support ISA compiled offline
if (!device().isOnline()) {
return true;
}
// ACL_TYPE_CG stage is not performed for offline compilation
executable_ = loader_->CreateExecutable(HSA_PROFILE_FULL, nullptr);
if (executable_ == nullptr) {
buildLog_ += "Error: Executable for AMD HSA Code Object isn't created.\n";
return false;
}
size_t size = binSize;
hsa_code_object_t code_object;
code_object.handle = reinterpret_cast<uint64_t>(binary);
hsa_agent_t agent = {amd::Device::toHandle(&(device()))};
hsa_status_t status = executable_->LoadCodeObject(agent, code_object, nullptr);
if (status != HSA_STATUS_SUCCESS) {
buildLog_ += "Error: AMD HSA Code Object loading failed.\n";
return false;
}
status = executable_->Freeze(nullptr);
if (status != HSA_STATUS_SUCCESS) {
buildLog_ += "Error: AMD HSA Code Object freeze failed.\n";
return false;
}
size_t kernelNamesSize = 0;
acl_error errorCode = amd::Hsail::QueryInfo(palNullDevice().compiler(), binaryElf_,
RT_KERNEL_NAMES, nullptr, nullptr, &kernelNamesSize);
@@ -278,6 +306,8 @@ bool HSAILProgram::createKernels(void* binary, size_t binSize, bool useUniformWo
aKernel->setUniformWorkGroupSize(useUniformWorkGroupSize);
}
}
DestroySegmentCpuAccess();
#endif // defined(WITH_COMPILER_LIB)
return true;
}
@@ -285,35 +315,10 @@ bool HSAILProgram::createKernels(void* binary, size_t binSize, bool useUniformWo
bool HSAILProgram::setKernels(void* binary, size_t binSize,
amd::Os::FileDesc fdesc, size_t foffset, std::string uri) {
#if defined(WITH_COMPILER_LIB)
// Stop compilation if it is an offline device - PAL runtime does not
// support ISA compiled offline
if (!device().isOnline()) {
return true;
}
// ACL_TYPE_CG stage is not performed for offline compilation
executable_ = loader_->CreateExecutable(HSA_PROFILE_FULL, nullptr);
if (executable_ == nullptr) {
buildLog_ += "Error: Executable for AMD HSA Code Object isn't created.\n";
return false;
}
size_t size = binSize;
hsa_code_object_t code_object;
code_object.handle = reinterpret_cast<uint64_t>(binary);
hsa_agent_t agent = {amd::Device::toHandle(&(device()))};
hsa_status_t status = executable_->LoadCodeObject(agent, code_object, nullptr);
if (status != HSA_STATUS_SUCCESS) {
buildLog_ += "Error: AMD HSA Code Object loading failed.\n";
return false;
}
status = executable_->Freeze(nullptr);
if (status != HSA_STATUS_SUCCESS) {
buildLog_ += "Error: AMD HSA Code Object freeze failed.\n";
return false;
}
bool dynamicParallelism = false;
for (auto& kit : kernels()) {
HSAILKernel* aKernel = static_cast<HSAILKernel*>(kit.second);
@@ -332,8 +337,6 @@ bool HSAILProgram::setKernels(void* binary, size_t binSize,
if (!isNull() && dynamicParallelism && !allocKernelTable()) {
return false;
}
DestroySegmentCpuAccess();
#endif // defined(WITH_COMPILER_LIB)
return true;
}