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

SWDEV-280473 - Support HSAIL shared library build

Browse Source 
This change makes HSAIL usage similar to that of Comgr. By default, the
runtime will statically link against it, however if HSAIL_DYN_DLL is
defined, then the runtime will try to dynamically load HSAIL.

Currently stick to statically linking to HSAIL. In a feature patch the
dynamic loading behaviour will be enabled.

Change-Id: I6a78a4375975cf847f236b200404c8cf941d012b
This commit is contained in:
Vladislav Sytchenko
2021-04-13 11:36:32 -04:00
parent 759cf19600
commit c7b50bb890
19 changed files with 662 additions and 214 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/device/pal/paldevice.cpp
+14 -4
View File
@@ -35,7 +35,7 @@
#include "palLib.h"
#include "palPlatform.h"
#include "palDevice.h"
#include "acl.h"
#include "hsailctx.hpp"
#include "vdi_common.hpp"
@@ -246,6 +246,11 @@ bool NullDevice::create(const char* palName, const amd::Isa& isa, Pal::GfxIpLeve
return false;
}
if (!ValidateHsail()) {
LogPrintfError("HSAIL initialization failed for offline PAL device %s", isa.targetId());
return false;
}
if (!amd::Device::create(isa)) {
LogPrintfError("Unable to setup device for PAL offline device %s", isa.targetId());
return false;
@@ -272,7 +277,7 @@ bool NullDevice::create(const char* palName, const amd::Isa& isa, Pal::GfxIpLeve
AMD_OCL_SC_LIB};
// Initialize the compiler handle
acl_error error;
compiler_ = aclCompilerInit(&opts, &error);
compiler_ = amd::Hsail::CompilerInit(&opts, &error);
if (error != ACL_SUCCESS) {
LogPrintfError("Error initializing the compiler for offline PAL device %s", isa.targetId());
return false;
@@ -915,6 +920,11 @@ bool Device::create(Pal::IDevice* device) {
return false;
}
if (!ValidateHsail()) {
LogError("Hsail initialization failed!");
return false;
}
computeEnginesId_.resize(std::min(numComputeEngines(), settings().numComputeRings_));
amd::Context::Info info = {0};
@@ -972,7 +982,7 @@ bool Device::create(Pal::IDevice* device) {
AMD_OCL_SC_LIB};
// Initialize the compiler handle
acl_error error;
compiler_ = aclCompilerInit(&opts, &error);
compiler_ = amd::Hsail::CompilerInit(&opts, &error);
if (error != ACL_SUCCESS) {
LogError("Error initializing the compiler");
return false;
@@ -1327,7 +1337,7 @@ void Device::tearDown() {
#if defined(WITH_COMPILER_LIB)
if (compiler_ != nullptr) {
aclCompilerFini(compiler_);
amd::Hsail::CompilerFini(compiler_);
compiler_ = nullptr;
}
#endif // defined(WITH_COMPILER_LIB)
rocclr/device/pal/palkernel.cpp
+30 -30
View File
@@ -25,7 +25,7 @@
#include "device/pal/palsched.hpp"
#include "platform/commandqueue.hpp"
#include "utils/options.hpp"
#include "acl.h"
#include "hsailctx.hpp"
#include <string>
#include <memory>
#include <fstream>
@@ -134,9 +134,9 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
if (palNullDevice().settings().svmFineGrainSystem_) {
options.append(" -sc-xnack-iommu");
}
error = aclCompile(palNullDevice().compiler(), prog().binaryElf(), options.c_str(), ACL_TYPE_CG,
ACL_TYPE_ISA, nullptr);
buildLog_ += aclGetCompilerLog(palNullDevice().compiler());
error = amd::Hsail::Compile(palNullDevice().compiler(), prog().binaryElf(), options.c_str(), ACL_TYPE_CG,
ACL_TYPE_ISA, nullptr);
buildLog_ += amd::Hsail::GetCompilerLog(palNullDevice().compiler());
if (error != ACL_SUCCESS) {
LogError("Failed to finalize kernel");
return false;
@@ -149,8 +149,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
// Pull out metadata from the ELF
size_t sizeOfArgList;
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_ARGUMENT_ARRAY,
openClKernelName.c_str(), nullptr, &sizeOfArgList);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_ARGUMENT_ARRAY,
openClKernelName.c_str(), nullptr, &sizeOfArgList);
if (error != ACL_SUCCESS) {
return false;
}
@@ -159,8 +159,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
if (nullptr == aclArgList) {
return false;
}
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_ARGUMENT_ARRAY,
openClKernelName.c_str(), aclArgList, &sizeOfArgList);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_ARGUMENT_ARRAY,
openClKernelName.c_str(), aclArgList, &sizeOfArgList);
if (error != ACL_SUCCESS) {
return false;
}
@@ -169,13 +169,13 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
delete[] aclArgList;
size_t sizeOfWorkGroupSize;
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE,
openClKernelName.c_str(), nullptr, &sizeOfWorkGroupSize);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE,
openClKernelName.c_str(), nullptr, &sizeOfWorkGroupSize);
if (error != ACL_SUCCESS) {
return false;
}
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE,
openClKernelName.c_str(), workGroupInfo_.compileSize_, &sizeOfWorkGroupSize);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE,
openClKernelName.c_str(), workGroupInfo_.compileSize_, &sizeOfWorkGroupSize);
if (error != ACL_SUCCESS) {
return false;
}
@@ -192,8 +192,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
// Pull out printf metadata from the ELF
size_t sizeOfPrintfList;
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_GPU_PRINTF_ARRAY,
openClKernelName.c_str(), nullptr, &sizeOfPrintfList);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_GPU_PRINTF_ARRAY,
openClKernelName.c_str(), nullptr, &sizeOfPrintfList);
if (error != ACL_SUCCESS) {
return false;
}
@@ -204,8 +204,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
if (nullptr == aclPrintfList) {
return false;
}
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_GPU_PRINTF_ARRAY,
openClKernelName.c_str(), aclPrintfList, &sizeOfPrintfList);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_GPU_PRINTF_ARRAY,
openClKernelName.c_str(), aclPrintfList, &sizeOfPrintfList);
if (error != ACL_SUCCESS) {
return false;
}
@@ -218,8 +218,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
aclMetadata md;
md.enqueue_kernel = false;
size_t sizeOfDeviceEnqueue = sizeof(md.enqueue_kernel);
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_DEVICE_ENQUEUE,
openClKernelName.c_str(), &md.enqueue_kernel, &sizeOfDeviceEnqueue);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_DEVICE_ENQUEUE,
openClKernelName.c_str(), &md.enqueue_kernel, &sizeOfDeviceEnqueue);
if (error != ACL_SUCCESS) {
return false;
}
@@ -227,17 +227,17 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
md.kernel_index = -1;
size_t sizeOfIndex = sizeof(md.kernel_index);
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_KERNEL_INDEX,
openClKernelName.c_str(), &md.kernel_index, &sizeOfIndex);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_KERNEL_INDEX,
openClKernelName.c_str(), &md.kernel_index, &sizeOfIndex);
if (error != ACL_SUCCESS) {
return false;
}
index_ = md.kernel_index;
size_t sizeOfWavesPerSimdHint = sizeof(workGroupInfo_.wavesPerSimdHint_);
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WAVES_PER_SIMD_HINT,
openClKernelName.c_str(), &workGroupInfo_.wavesPerSimdHint_,
&sizeOfWavesPerSimdHint);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WAVES_PER_SIMD_HINT,
openClKernelName.c_str(), &workGroupInfo_.wavesPerSimdHint_,
&sizeOfWavesPerSimdHint);
if (error != ACL_SUCCESS) {
return false;
}
@@ -245,16 +245,16 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
waveLimiter_.enable();
size_t sizeOfWorkGroupSizeHint = sizeof(workGroupInfo_.compileSizeHint_);
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE_HINT,
openClKernelName.c_str(), workGroupInfo_.compileSizeHint_,
&sizeOfWorkGroupSizeHint);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_WORK_GROUP_SIZE_HINT,
openClKernelName.c_str(), workGroupInfo_.compileSizeHint_,
&sizeOfWorkGroupSizeHint);
if (error != ACL_SUCCESS) {
return false;
}
size_t sizeOfVecTypeHint;
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_VEC_TYPE_HINT,
openClKernelName.c_str(), NULL, &sizeOfVecTypeHint);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_VEC_TYPE_HINT,
openClKernelName.c_str(), NULL, &sizeOfVecTypeHint);
if (error != ACL_SUCCESS) {
return false;
}
@@ -264,8 +264,8 @@ bool HSAILKernel::init(amd::hsa::loader::Symbol* sym, bool finalize) {
if (NULL == VecTypeHint) {
return false;
}
error = aclQueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_VEC_TYPE_HINT,
openClKernelName.c_str(), VecTypeHint, &sizeOfVecTypeHint);
error = amd::Hsail::QueryInfo(palNullDevice().compiler(), prog().binaryElf(), RT_VEC_TYPE_HINT,
openClKernelName.c_str(), VecTypeHint, &sizeOfVecTypeHint);
if (error != ACL_SUCCESS) {
return false;
}
rocclr/device/pal/palprogram.cpp
+10 -10
View File
@@ -209,12 +209,12 @@ HSAILProgram::~HSAILProgram() {
}
#if defined(WITH_COMPILER_LIB)
if (rawBinary_ != nullptr) {
aclFreeMem(binaryElf_, rawBinary_);
amd::Hsail::FreeMem(binaryElf_, rawBinary_);
}
acl_error error;
// Free the elf binary
if (binaryElf_ != nullptr) {
error = aclBinaryFini(binaryElf_);
error = amd::Hsail::BinaryFini(binaryElf_);
if (error != ACL_SUCCESS) {
LogWarning("Error while destroying the acl binary \n");
}
@@ -274,16 +274,16 @@ bool HSAILProgram::setKernels(amd::option::Options* options, void* binary, size_
}
size_t kernelNamesSize = 0;
acl_error errorCode = aclQueryInfo(palNullDevice().compiler(), binaryElf_, RT_KERNEL_NAMES, nullptr,
nullptr, &kernelNamesSize);
acl_error errorCode = amd::Hsail::QueryInfo(palNullDevice().compiler(), binaryElf_, RT_KERNEL_NAMES, nullptr,
nullptr, &kernelNamesSize);
if (errorCode != ACL_SUCCESS) {
buildLog_ += "Error: Querying of kernel names size from the binary failed.\n";
return false;
}
if (kernelNamesSize > 0) {
char* kernelNames = new char[kernelNamesSize];
errorCode = aclQueryInfo(palNullDevice().compiler(), binaryElf_, RT_KERNEL_NAMES, nullptr, kernelNames,
&kernelNamesSize);
errorCode = amd::Hsail::QueryInfo(palNullDevice().compiler(), binaryElf_, RT_KERNEL_NAMES, nullptr, kernelNames,
&kernelNamesSize);
if (errorCode != ACL_SUCCESS) {
buildLog_ += "Error: Querying of kernel names from the binary failed.\n";
delete[] kernelNames;
@@ -360,8 +360,8 @@ void HSAILProgram::fillResListWithKernels(VirtualGPU& gpu) const { gpu.addVmMemo
const aclTargetInfo& HSAILProgram::info() {
#if defined(WITH_COMPILER_LIB)
acl_error err;
info_ = aclGetTargetInfo(palNullDevice().settings().use64BitPtr_ ? "hsail64" : "hsail",
device().isa().hsailName(), &err);
info_ = amd::Hsail::GetTargetInfo(palNullDevice().settings().use64BitPtr_ ? "hsail64" : "hsail",
device().isa().hsailName(), &err);
if (err != ACL_SUCCESS) {
LogWarning("aclGetTargetInfo failed");
}
@@ -374,11 +374,11 @@ bool HSAILProgram::saveBinaryAndSetType(type_t type) {
// Write binary to memory
if (rawBinary_ != nullptr) {
// Free memory containing rawBinary
aclFreeMem(binaryElf_, rawBinary_);
amd::Hsail::FreeMem(binaryElf_, rawBinary_);
rawBinary_ = nullptr;
}
size_t size = 0;
if (aclWriteToMem(binaryElf_, &rawBinary_, &size) != ACL_SUCCESS) {
if (amd::Hsail::WriteToMem(binaryElf_, &rawBinary_, &size) != ACL_SUCCESS) {
buildLog_ += "Failed to write binary to memory \n";
return false;
}