SWDEV-82596 - HSA HLC: Create AMDInline pass
The generic llvm inlining heuristcs do not work well for GPU.
In particular we have a common problem in several tests:
If we have a pointer to private array passed into a function it will not be optimized out, leaving scratch usage.
The pass increases the inline threshold to allow inliniting in this case.
Also that we can move at least some AMD inlining customizations into this file from the common code.
Inline hint threshold is moved in this change.
Performance impact on ocltst sha256, 32 bit, Fiji:
AMDIL HSAIL Diff HSAIL+Inliner Diff Diff
before to AMDIL to HSAIL to AMDIL
OCLPerfSHA256[ 0] 43.843 40.894 0.93 69.910 1.71 1.59
OCLPerfSHA256[ 1] 53.611 51.083 0.95 80.919 1.58 1.51
OCLPerfSHA256[ 2] 52.127 51.528 0.99 80.640 1.56 1.55
OCLPerfSHA256[ 3] 60.952 57.027 0.94 68.615 1.20 1.13
OCLPerfSHA256[ 4] 76.173 70.150 0.92 80.582 1.15 1.06
OCLPerfSHA256[ 5] 75.886 70.264 0.93 81.000 1.15 1.07
Testing: smoke, precheckin, ocltst sha256
Reviewed by Danill Fukalov
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/opt_level.cpp#28 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/InitializePasses.h#93 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/LinkAllPasses.h#49 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/Transforms/IPO.h#32 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/AMDInline.cpp#1 add
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/CMakeLists.txt#24 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/IPO.cpp#32 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/Inliner.cpp#42 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/tools/opt/amdopt.inc#28 edit
SWDEV-82054 - [CQE OCL][QR][LNX] RQ Conformance "Integer_Ops" test is crashing on CPU; Faulty CL#1206023.
In llvm32 by default llvm::DisablePrettyStackTrace is off, which causes a trap handler installed by default and interferes with the trap handler in runtime, causing unhandled SIGFPE exceptions when executing conformance/integer_ops on certain cpu.
To fix this, put stack trace duping under an env var AMD_DUMP_STACK_TRACE and set llvm::DisablePrettyStackTrace=true by default.
Here env var is used because there is still no elf binary at this stage to pass the compiler option to if_aclCompilerInit.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/common/v0_8/if_acl.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#86 edit
SWDEV-81805 - Fix compiler lib bug: incorrect type name %opencl.pipe_t0 is generated when using clang.
Clang does not return llvm::Module. It saves the bitcode to a memory buffer and passed back to compiler lib, then bitcode reader is used to get llvm::Module. Clang and bitcode reader uses the same LLVMContext which is created earlier in aclCompileInternal. Since named struct types are shared between modules in LLVMContext. When bitcode reader loads the module, name collision happens for named struct types, which causes them to be postfixed with a number, e.g. %opencl.pipe_t => %opencl.pipe_t0.
This causes failure in SPIR-V drop-in conformance test.
The fix is to let clang uses a separate LLVMContext.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/frontend_clang.cpp#26 edit
SWDEV-67990 - SPIR-V: Fix duplicate creation of opaque types in SPIR-V/LLVM translation.
Fix mem leak in drop-in LLVM/SPIR-V translator.
Add warning for undefined function in linker.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/linker.cpp#135 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/SPIRV/SPIRVReader.cpp#5 edit
SWDEV-80864 - HSAIL Metadata Workgroup Size Hint and Vec Type Hint added to HSAIL Runtime
Runtime changes required for the use of these two metadata:
- Runtime's gpukernel.cpp requires new aclQueries during HSAILKernel::Init
- One for quering WorkGroupSizeHint's array
- Two for size of VecTypeHint and fetching VecTypeHint's string
- initArgList needs to be moved to end of HSAILKernel::init to allow createSignature to get non empty values
- Compiler lib's workgroup hint (wsh) needs to match runtime's type (size_t)
- In Kernel constructor, instead of using memset which corrupts std::string, specifically set default workGroupInfo struct's variables
Also fixed wavesPerSimdHint to use size_t to match runtime.
Updated CLAssumptionCheck.cpp since aclMetadata structure was modified.
Note: This is the runtime counterpart to submitted CL#1204512. (Post Review#8808, SWDEV-79695)
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/include/v0_8/aclStructs.h#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#22 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#260 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#308 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLAssumptionCheck.cpp#48 edit
SWDEV-77584 - Compiler Lib: Preparations for enabling HSAIL on OpenCL 1.2 by default. Adding -legacy and -binary_is_spirv.
-legacy option will be used for forcing AMDIL path after switching HSAIL by default for OpenCL.
-binary_is_spirv option will be used for indicating that the binary is constructed from SPIRV.
[Testing] pre-checkin:
http://ocltc.amd.com:8111/viewModification.html?modId=61541&personal=true&buildTypeId=&tab=vcsModificationBuilds&show_all_builds=true
[Reviewer] Stanislav Mekhanoshin
http://ocltc.amd.com/reviews/r/8850
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/utils/OPTIONS.def#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/OPTIONS.def#132 edit
SWDEV-80261 - Enable dump stack trace in debug build when assert/abort/segfault to facilitate debugging.
It is useful to see the stack trace when program assert(), especially in teamcity when some issues are not reproducible locally. This change enables that.
Only enabled for debug build.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/common/v0_8/if_acl.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#83 edit
SWDEV-79695 - Checkin again after fixed build issue with ocltst changes. Merged all changes in compiler lib to legacy-lib.
HSAIL Metadata Workgroup Size Hint and Vec Type Hint added to HSAIL
HSAIL does not handle kernel attributes for work_group_size_hint or vec_type_hint.
Adding these two kernel attributes to metadata requires multiple changes:
HSAIL: Added parsing functions to HSAILParseMetadata.cpp, and similar files
Properly translates to .hsail files as 'wsh' and 'vth'
MDParser: Flex scanner now reads new hsail metadata, 'wsh' and 'vth'
Acl_metadata: added the two new metadata
Uint32_t wsh[3]; // Added case: RT_WORK_GROUP_SIZE_HINT
Char *vth; // Added case: RT_VEC_TYPE_HINT
Since vth is std::string, it was required to serialize, and then de-serialize for aclMetadata
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/common/v0_8/if_acl.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/gpu/MDParser/AMDILMDInterface.h#3 integrate
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/gpu/MDParser/AMDILMDParser.l#4 integrate
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/gpu/MDParser/AMDILMDTypes.cpp#3 integrate
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/gpu/MDParser/lex.yy.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/backends/gpu/metadata.cpp#4 integrate
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/include/v0_8/aclEnums.h#4 integrate
... //depot/stg/opencl/drivers/opencl/compiler/legacy-lib/include/v0_8/aclStructs.h#4 integrate
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#81 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDInterface.h#6 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDParser.l#7 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDTypes.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/lex.yy.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/metadata.cpp#12 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclEnums.h#26 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#21 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILKernel.h#15 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILKernelManager.cpp#40 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILModuleInfo.cpp#18 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILParseMetadata.cpp#16 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLAssumptionCheck.cpp#47 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLEnumCheck.cpp#51 edit
SWDEV-79695 - HSAIL Metadata Workgroup Size Hint and Vec Type Hint added to HSAIL
HSAIL does not handle kernel attributes for work_group_size_hint or vec_type_hint.
Adding these two kernel attributes to metadata requires multiple changes:
HSAIL: Added parsing functions to HSAILParseMetadata.cpp, and similar files
Properly translates to .hsail files as 'wsh' and 'vth'
MDParser: Flex scanner now reads new hsail metadata, 'wsh' and 'vth'
Acl_metadata: added the two new metadata
Uint32_t wsh[3]; // Added case: RT_WORK_GROUP_SIZE_HINT
Char *vth; // Added case: RT_VEC_TYPE_HINT
Since vth is std::string, it was required to serialize, and then de-serialize for aclMetadata
TODO: Runtime changes required for the use of these two metadata:
Runtime's gpukernel.cpp requires 4 new aclQueries during HSAILKernel::Init
Runtime changes are showing issues when testing
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#79 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDInterface.h#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDParser.l#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/AMDILMDTypes.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/MDParser/lex.yy.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/metadata.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclEnums.h#24 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#19 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILKernel.h#13 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILKernelManager.cpp#38 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILModuleInfo.cpp#16 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Target/HSAIL/HSAILParseMetadata.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLAssumptionCheck.cpp#45 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLEnumCheck.cpp#49 edit
SWDEV-77174 - Change compiler lib def and sc lib def to be consistent.
SC interface depends on some compile options which need to be consistent between compiler library and SC.
This change will add the SC interface dependent build options to opencldefs so that they are shared between compiler lib def and sc lib def.
Both compilier lib def and sc lib def will include scinterfacedefs, which will generate consistent compile options required by sc interface based on variables defined in opencldefs.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/complibdefs#45 edit
... //depot/stg/opencl/drivers/opencl/opencldefs#163 edit
SWDEV-2 - Change OpenCL version number from 1898 to 1899.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/utils/versions.hpp#1645 edit
SWDEV-77584 - HSA HLC: fixed reflection metadata generation on HSAIL OCL 1.2 path
We are producing 6 extra arguments, but metadata was produced only for 3.
Removed KE_OCL12_NUM_ARGS define to avoid confusion.
Testing: smoke, precheckin
Reviewed by Yaxun Liu
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/AMDOpenCLKernenv.h#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/Scalar/AMDInsertOpenCLKernenv.cpp#10 edit
SWDEV-77164 - Remove libelf differences from original elftoolchain. These erroneously make ELF_C_RDWR not to load existing section headers.
Testing: pre-checkin
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/loaders/elf/utils/libelf/elf_scn.c#4 edit
ECR #304775 - Modify caching library code to be used by HSAIL runtime
1. Remove dependencies on shader compiler version, amd::Os and amd::option
2. Let version be one the of input
3. Change all prefix "Kernel" to prefix "String", e.g., class name is modified from KernelCache to StringCache, etc.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/compiler/tools/caching/cache.cpp#1 branch
... //depot/stg/opencl/drivers/opencl/compiler/tools/caching/cache.hpp#1 branch
ECR #333753 - Compiler Lib: Bug 10985 - Print compiler options as comment in .cl file when dumping using -save-temps
Affects all targets.
Also fixes rare problem with dumping cl as a raw binary.
Testing: pre check-in
Reviewers: Brian Sumner, Nikolay Haustov, Matthew Arsenault
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/frontend.cpp#36 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/frontend.hpp#10 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/frontend_clang.cpp#23 edit
EPR #425389 - Back out changelist 1181925
Although the compiler library sources are split, the build does not yet use this, so the wrong default value is being used for AMDIL vs. HSAIL
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/OPTIONS.def#130 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/AMDLLVMContextHook.h#28 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/Transforms/IPO/AMDOptOptions.h#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/AMDOptOptions.cpp#10 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/AMDPassManagerBuilder.cpp#56 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/tools/opt/amdopt.inc#26 edit
ECR #354633 - SPIR-V: Let aoc2 load and save SPIR-V.
E.g.
aoc2 -march=hsail-64 -cl-std=CL2.0 -srctospv testReadf.cl
compile a cl to SPIR-V binary and save it as .spv
aoc2 -march=hsail-64 -cl-std=CL2.0 -spirv work_group_any.spv
load a SPIR-V binary and compiles it to ISA and save it to elf in .bin
Changed the option for round-trip translation of SPIR-V to -round-trip-spirv.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/acl.cpp#35 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/frontend_clang.cpp#22 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/linker.cpp#133 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#74 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclEnums.h#22 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/OPTIONS.def#129 edit
... //depot/stg/opencl/drivers/opencl/compiler/tools/aoc2/aoc2.cpp#76 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/complib/CLEnumCheck.cpp#47 edit
ECR #354633 - SPIR-V: Add -D for blowfish for compiler lib, which is required by ValidateBinaryImage as it is moved from runtime to compiler lib.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/complibdefs#42 edit
ECR #304775 - Remove HLC_Unroll_* variables.
HLC_Unroll_Scratch_Threshold was unused. The others have equivalent settings in the AMDLLVMContextHook, so consistently use that version. The patches to opt were already had different set of command line flags for the same options.
This changes two of the defaults in compiler library and the equivalent flags in opt to match the values which were actually in use so this shoudn't change the current behavior. The unroll threshold default and allow partial unrolling defaults were changed to the actually used default values. Eventually all of these custom options should be removed, because in current LLVM these can be controlled per loop by the TargetTransformInfo, and all have equivalent cl::opts already.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/OPTIONS.def#128 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/AMDLLVMContextHook.h#27 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/Transforms/IPO/AMDOptOptions.h#7 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/AMDOptOptions.cpp#9 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Transforms/IPO/AMDPassManagerBuilder.cpp#55 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/tools/opt/amdopt.inc#25 edit
ECR #304775 - Preparation for kernel caching feature
force line ending on files to one style. Precheckin does not capture this failure as it converts all text files to linux format so that both linux and windows can compile the code during precheckin.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.hpp#4 edit
ECR #304775 - Preparation for kernel caching feature
1. Each device have a separate cache directory
2. It logs caching errors, so we can debug the cache and/or detect collisions
3. Implementeded cache size tracking, so we can evict old data when cache files are too large
4. Added file/path access permission control on both windows and linux
5. Have read/write file lock protection
6. -kcache-disable flag can be used to turn on/off the caching functionality
7. AMD_FORCE_KCACHE_TEST env variable is used for internal testing
8. For the stage we want to cache, call getCacheEntry() followed by makeCacheEntry() if the get fails; otherwise directly return cached data.
- After the compiler library code is refactored, getCacheEntry() and makeCacheEntry() will be wrapped into one function call, so that only one call is needed at the place we want to cache
TO DO:
1. Use it in the compiler library code
- Waiting for the decision on how many stages we want to cache, i) 1-stage caching: source->ISA; or ii) 3-stage caching: source->LLVM IR, LLVM IR->IL, IL->ISA
2. Tracking of timestamps for cache entries
- LRU eviction when cache grows too large
- Suggestion from Laurent: Regarding tracking timestamps for LRU eviction: Random eviction would probably perform as well as LRU and does not require timestamps.
3. Track cache entries per application
ReviewBoardURL = http://ocltc.amd.com/reviews/r/8194/
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.cpp#3 add
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/cache.hpp#3 add
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/OPTIONS.def#127 edit
ECR #304775 - Move AMDExportKernelNature to compiler library.
This is part of the effort to move all extra passes out of the LLVM directory and into compiler library so we can build with a pure copy of LLVM. This is far from all of the sources required to move, but I'm trying to do the migration in moderately sized pieces.
Duplicates AMDSymbolNames.h utility header to avoid having a circular dependency between LLVM and compiler library. Eventually when all extra sources are moved into compiler library, there should only be one copy left.
Affected files ...
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/optimizer.cpp#27 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/linker/include/AMDPasses.h#1 add
... //depot/stg/opencl/drivers/opencl/compiler/lib/linker/include/AMDSymbolName.h#1 branch
... //depot/stg/opencl/drivers/opencl/compiler/lib/linker/lib/AMDExportKernelNature.cpp#1 move/add
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/Analysis/Passes.h#56 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/InitializePasses.h#81 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Analysis/AMDExportKernelNature.cpp#16 move/delete
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Analysis/Analysis.cpp#34 edit
foreman