Upstream merge 56

[dorde-antic]

Overview

Moves copy of LLVM to this commit :
ROCm/llvm-project@659e9cd04732

Required some changes in rocMLIR due to following commit:
ROCm/llvm-project@878d359#diff-3cbcc8f6c740f2d6e16f5a0c19daf4bb8224ad92d9e430fc10c935587a67dcce

• Resolves https://github.com/ROCm/rocMLIR-internal/issues/1883

Diffs, FAT update

Files that are different between rocm/llvm-project and external/llvm-project

Only in llvm-project/clang/include/clang/Format: .clang-format
Only in llvm-project/clang/lib/Format: .clang-format
Only in llvm-project/clang/unittests/Format: .clang-format
Only in llvm-project/: diff-rup-upstream-merge56.txt
Only in llvm-project/: .git
Files llvm-project/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp and rocMLIR/external/llvm-project/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp differ
Files llvm-project/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp and rocMLIR/external/llvm-project/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp differ
Only in rocMLIR/external/llvm-project/llvm/test/CodeGen/AMDGPU: bug-multi-operands-to-update-after-fold.mir
Only in rocMLIR/external/llvm-project/llvm/utils/lit/lit/formats: __pycache__
Only in rocMLIR/external/llvm-project/llvm/utils/lit/lit/llvm: __pycache__
Only in rocMLIR/external/llvm-project/llvm/utils/lit/lit: __pycache__
Files llvm-project/mlir/include/mlir/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.h and rocMLIR/external/llvm-project/mlir/include/mlir/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.h differ
Files llvm-project/mlir/include/mlir/Conversion/GPUToROCDL/GPUToROCDLPass.h and rocMLIR/external/llvm-project/mlir/include/mlir/Conversion/GPUToROCDL/GPUToROCDLPass.h differ
Files llvm-project/mlir/include/mlir/Dialect/Bufferization/Transforms/Bufferize.h and rocMLIR/external/llvm-project/mlir/include/mlir/Dialect/Bufferization/Transforms/Bufferize.h differ
Files llvm-project/mlir/include/mlir/Dialect/Tosa/IR/TosaOpBase.td and rocMLIR/external/llvm-project/mlir/include/mlir/Dialect/Tosa/IR/TosaOpBase.td differ
Files llvm-project/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td and rocMLIR/external/llvm-project/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td differ
Only in rocMLIR/external/llvm-project/mlir/include/mlir/ExecutionEngine: CpuSystemDetect.h
Only in rocMLIR/external/llvm-project/mlir/include/mlir/ExecutionEngine: RocmDeviceName.h
Only in rocMLIR/external/llvm-project/mlir/include/mlir/ExecutionEngine: RocmSystemDetect.h
Only in rocMLIR/external/llvm-project/mlir/include/mlir/ExecutionEngine: SystemDevices.h
Files llvm-project/mlir/include/mlir/Interfaces/CallInterfaces.h and rocMLIR/external/llvm-project/mlir/include/mlir/Interfaces/CallInterfaces.h differ
Files llvm-project/mlir/include/mlir/Interfaces/CallInterfaces.td and rocMLIR/external/llvm-project/mlir/include/mlir/Interfaces/CallInterfaces.td differ
Files llvm-project/mlir/lib/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.cpp and rocMLIR/external/llvm-project/mlir/lib/Conversion/AMDGPUToROCDL/AMDGPUToROCDL.cpp differ
Files llvm-project/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp and rocMLIR/external/llvm-project/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp differ
Files llvm-project/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp and rocMLIR/external/llvm-project/mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp differ
Files llvm-project/mlir/lib/Dialect/AMDGPU/IR/AMDGPUDialect.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/AMDGPU/IR/AMDGPUDialect.cpp differ
Files llvm-project/mlir/lib/Dialect/Bufferization/Transforms/Bufferize.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Bufferization/Transforms/Bufferize.cpp differ
Files llvm-project/mlir/lib/Dialect/Bufferization/Transforms/BufferResultsToOutParams.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Bufferization/Transforms/BufferResultsToOutParams.cpp differ
Files llvm-project/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Tosa/IR/TosaCanonicalizations.cpp differ
Files llvm-project/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp differ
Files llvm-project/mlir/lib/Dialect/Tosa/Transforms/TosaDecomposeTransposeConv.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Tosa/Transforms/TosaDecomposeTransposeConv.cpp differ
Files llvm-project/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp and rocMLIR/external/llvm-project/mlir/lib/Dialect/Tosa/Transforms/TosaValidation.cpp differ
Files llvm-project/mlir/lib/ExecutionEngine/CMakeLists.txt and rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine/CMakeLists.txt differ
Only in rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine: CpuSystemDetect.cpp
Only in rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine: RocmDeviceName.cpp
Only in rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine: RocmSystemDetect.cpp
Only in rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine: SystemDevices.cpp
Only in rocMLIR/external/llvm-project/mlir/test/Conversion/AMDGPUToROCDL: ldsbarrier-hack.mlir
Files llvm-project/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-named.mlir and rocMLIR/external/llvm-project/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-named.mlir differ
Files llvm-project/mlir/test/Dialect/Tosa/invalid.mlir and rocMLIR/external/llvm-project/mlir/test/Dialect/Tosa/invalid.mlir differ
Files llvm-project/mlir/test/Dialect/Tosa/verifier.mlir and rocMLIR/external/llvm-project/mlir/test/Dialect/Tosa/verifier.mlir differ
Only in llvm-project/mlir/test/Integration/GPU/ROCM: gpu-to-hsaco.mlir
Only in llvm-project/mlir/test/Integration/GPU/ROCM: printf.mlir
Only in llvm-project/mlir/test/Integration/GPU/ROCM: two-modules.mlir
Only in llvm-project/mlir/test/Integration/GPU/ROCM: vecadd.mlir
Only in llvm-project/mlir/test/Integration/GPU/ROCM: vector-transferops.mlir

llvm_diffs.txt
mlir_diffs.txt

There are no changes in librockcompiler_deps.cmake after executing get_fat_library_deps_list.pl

CI checks and manual tests

PR CI: Done ✅
Nightly CI: Done - passed benchmark and report performance, but failed on MIGraphX ❌. Then, in new run (rekicked migraphx only), it passed MIGraphX too ✅
Weekly CI: in progress⌛

Manual tests:

• Navi2x: waiting for Navi2x ⌛
• MI350/MI355: Done ✅ one failing test (won't block CI since we don't use MI350/MI355 in CI atm) https://github.com/ROCm/rocMLIR-internal/issues/1959

[dhernandez0]

failed to legalize operation 'rock.threadwise_gemm' that was explicitly marked illegal

I've fixed that. The issue was that vector::TransferReadOp build() uses ub::PoisonOp for the padding (I think) if it's undefined. We just need to add ub::PoisonOp as legal op.

We never use the mask or padding when we create TransferReadOp, it's always inbounds and without padding. So, it's fine to have a poison value for padding.

[umangyadav]

@dorde-antic can you add diff list similar to how i did it earlier ?
Also there are some merge conflicts.
#1899

[dorde-antic]

@dorde-antic can you add diff list similar to how i did it earlier

Yes, is there some tool or script that generates the diff list as one that you provided or?

Also there are some merge conflicts. #1899

Seems that it has happened today after updating branch from our develop

[umangyadav]

Yes, is there some tool or script that generates the diff list as one that you provided or?

To get list of files that are different you can just run

diff -qr path/to/folderA path/to/folderB

to get diff of contents that are different

diff -rup /path/to/folderA path/to/folderB

[umangyadav]

Changest look good to me, Wait for CI to pass before hitting merge button.

Also test manually on navi2x and MI350.

[dorde-antic]

The tuning weekly task is failing due to known issues that are unrelated to this PR (CI detected errors in the output). Should we consider modifying the tuning weekly check to exclude log and output file validation?

[dorde-antic]

The tuning weekly task is failing due to known issues that are unrelated to this PR (CI detected errors in the output). Should we consider modifying the tuning weekly check to exclude log and output file validation?

try locally failing commands from log

[dorde-antic]

loc("<stdin>":75:7): error: Currently, attention/gemm+gemm/conv+gemm op is only supported on GPUs with matrix accelerator extentions
Lowering failed

for mixr-gemm-gemm-add-f16.e2e.mlir on Navi2

PR that updates tests is included, but for some reason mlir/test/fusion/e2e/lit.local.cfg was not updated even though develop was merged into upstream_merge_56 @umangyadav

[dorde-antic]

Failed to create MemoryBuffer for: /home/djoantic/rocMLIR/rnd_e2e/%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext
Error: No such file or directory
Failed to create MemoryBuffer for: /home/djoantic/rocMLIR/rnd_e2e/%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext
Error: No such file or directory
Failed to create MemoryBuffer for: /home/djoantic/rocMLIR/rnd_e2e/%linalg_test_lib_dir/libmlir_runner_utils%shlibext
Error: No such file or directory
Failed to create MemoryBuffer for: /home/djoantic/rocMLIR/rnd_e2e/%linalg_test_lib_dir/libmlir_float16_utils%shlibext
Error: No such file or directory
Failed to create MemoryBuffer for: /home/djoantic/rocMLIR/rnd_e2e/%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext
Error: No such file or directory
JIT session error: Symbols not found: [ __truncsfbf2, printMemrefF32, mgpuStreamWaitEvent, mgpuEventCreate, mgpuEventDestroy, mgpuEventRecord, mgpuLaunchKernel, mgpuMemAlloc, mgpuMemcpy, mgpuModuleGetFunction, mgpuModuleLoad, mgpuModuleUnload, mgpuStreamCreate, mgpuStreamDestroy, mgpuStreamSynchronize ]
Failure value returned from cantFail wrapped call
Failed to materialize symbols: { (main, { _mlir__memcpy_bf16_f32_15, _mlir_dot_add_module_load, dot_add_wrapper, _memcpy_bf16_f32_15, main, _mlir_dot_add_module_unload, $.LLVMDialectModule.__inits.0, _mlir_dot_add_wrapper, _mlir_main, _memcpy_bf16_f32_20, dot_add, __orc_init_func.LLVMDialectModule, _mlir_dot_add, _memcpy_bf16_f32_12, __orc_deinit_func.LLVMDialectModule, _mlir__memcpy_bf16_f32_20, _mlir__memcpy_bf16_f32_12 }) }
UNREACHABLE executed at /home/djoantic/rocMLIR/external/llvm-project/llvm/include/llvm/Support/Error.h:779!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: ./bin/xmir-runner --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext,%linalg_test_lib_dir/libmlir_c_runner_utils%shlibext --entry-point-result=void
 #0 0x00007f502356a528 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) /home/djoantic/rocMLIR/external/llvm-project/llvm/lib/Support/Unix/Signals.inc:819:13
 #1 0x00007f5023567fe3 llvm::sys::RunSignalHandlers() /home/djoantic/rocMLIR/external/llvm-project/llvm/lib/Support/Signals.cpp:105:18
 #2 0x00007f502356b291 SignalHandler(int, siginfo_t*, void*) /home/djoantic/rocMLIR/external/llvm-project/llvm/lib/Support/Unix/Signals.inc:0:3
 #3 0x00007f5022e36520 (/lib/x86_64-linux-gnu/libc.so.6+0x42520)
 #4 0x00007f5022e8a9fc __pthread_kill_implementation ./nptl/pthread_kill.c:44:76
 #5 0x00007f5022e8a9fc __pthread_kill_internal ./nptl/pthread_kill.c:78:10
 #6 0x00007f5022e8a9fc pthread_kill ./nptl/pthread_kill.c:89:10
 #7 0x00007f5022e36476 gsignal ./signal/../sysdeps/posix/raise.c:27:6
 #8 0x00007f5022e1c7f3 abort ./stdlib/abort.c:81:7
 #9 0x00007f50234b389f (/home/djoantic/rocMLIR/rnd_e2e/external/llvm-project/llvm/lib/libLLVMSupport.so.21.0git+0x15989f)
#10 0x00007f502d89ff30 std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::_M_create(unsigned long&, unsigned long) /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/basic_string.tcc:138:2
#11 0x00007f502d89ff30 void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::_M_construct<char const*>(char const*, char const*, std::forward_iterator_tag) /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/basic_string.tcc:219:14
#12 0x00007f502d89ff30 void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::_M_construct_aux<char const*>(char const*, char const*, std::__false_type) /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/basic_string.h:255:11
#13 0x00007f502d89ff30 void std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::_M_construct<char const*>(char const*, char const*) /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/basic_string.h:274:4
#14 0x00007f502d89ff30 std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::basic_string(char const*, unsigned long, std::allocator<char> const&) /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/basic_string.h:521:9
#15 0x00007f502d89ff30 llvm::StringRef::str[abi:cxx11]() const /home/djoantic/rocMLIR/external/llvm-project/llvm/include/llvm/ADT/StringRef.h:234:14
#16 0x00007f502d89ff30 mlir::ExecutionEngine::create(mlir::Operation*, mlir::ExecutionEngineOptions const&, std::unique_ptr<llvm::TargetMachine, std::default_delete<llvm::TargetMachine>>) /home/djoantic/rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine/ExecutionEngine.cpp:0:0
#17 0x00007f502fe451fb std::unique_ptr<llvm::TargetMachine, std::default_delete<llvm::TargetMachine>>::~unique_ptr() /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/unique_ptr.h:360:6
#18 0x00007f502fe451fb compileAndExecute((anonymous namespace)::Options&, mlir::Operation*, llvm::StringRef, (anonymous namespace)::CompileAndExecuteConfig, void**, std::unique_ptr<llvm::TargetMachine, std::default_delete<llvm::TargetMachine>>) /home/djoantic/rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine/JitRunner.cpp:199:7
#19 0x00007f502fe42c86 compileAndExecuteVoidFunction((anonymous namespace)::Options&, mlir::Operation*, llvm::StringRef, (anonymous namespace)::CompileAndExecuteConfig, std::unique_ptr<llvm::TargetMachine, std::default_delete<llvm::TargetMachine>>) /home/djoantic/rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine/JitRunner.cpp:239:10
#20 0x00007f502fe40fc5 std::unique_ptr<llvm::TargetMachine, std::default_delete<llvm::TargetMachine>>::~unique_ptr() /usr/lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/unique_ptr.h:360:6
#21 0x00007f502fe40fc5 mlir::JitRunnerMain(int, char**, mlir::DialectRegistry const&, mlir::JitRunnerConfig) /home/djoantic/rocMLIR/external/llvm-project/mlir/lib/ExecutionEngine/JitRunner.cpp:396:17
#22 0x000000000020a9e2 main /home/djoantic/rocMLIR/mlir/tools/xmir-runner/xmir-runner.cpp:121:10
#23 0x00007f5022e1dd90 __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:58:16
#24 0x00007f5022e1de40 call_init ./csu/../csu/libc-start.c:128:20
#25 0x00007f5022e1de40 __libc_start_main ./csu/../csu/libc-start.c:379:5
#26 0x000000000020a625 _start (./bin/xmir-runner+0x20a625)
Aborted (core dumped)

for mixr-gemm-a dd-bf16.e2e.mlir on Navi2