LLVM and SPIRV-LLVM-Translator pulldown (WW52) #12255
Conversation
…) (#76168) Also, for consistency make the ZeroOp lowering switch on the ArmSMETileType, rather than the element bit width.
Follow up to the discussion from #75258, and serves as an alternate solution for #74670. Set the location to Unknown for deduplicated / moved / materialized constants by OperationFolder. This makes sure that the folded constants don't end up with an arbitrary location of one of the original ops that became it, and that hoisted ops don't confuse the stepping order.
This reverts commit 199a0f9. Fixed the left-shift of signed integer which was causing UB.
… (#75827) Xcode 14.3.1 seems to have dropped these flags so we are creating unit tests to reproduce the issue.
Depositing value into the lowest byte/word is a common code pattern. This patch improves the code generation for it to avoid redundant AND and OR operations.
This patch fixes: flang/lib/Optimizer/Transforms/StackArrays.cpp:452:7: error: ignoring return value of function declared with 'nodiscard' attribute [-Werror,-Wunused-result]
This reverts commit 9f0f558. Fix expensive checks failure by properly marking register def for ADR.
By looking at whether a global is large instead of looking at the code model. This also fixes references to large data in the small code model. We now always fold any 32-bit offset into the addressing mode with the large code model since it uses 64-bit relocations.
Extends fold-arith-extf-into-vector-contract.mlir by adding a test case for scalable vectors.
…own large section name (#74381)" This reverts commit 19fff85. Now that explicit large globals are handled properly in the small code model.
Renaming a member variable from "Endoding" to "Encoding". Also replace inlined code for "isNormalized" with a call to the function, so that if the definition of normalization ever changes, we only need to change the one place.
…T. (#75737) Add SME2 DOT builtins.
… (#75630) Improve tests for atomic loads and stores, mainly by testing 128-bit atomic load and store instructions both with and w/out natural alignment.
TestGlobalModuleCache.py, a recently added test, tries to update a source file in the build directory, but it assumes the file is writable. In our distributed build and test system, this is not always true, so the test often fails with a write permissions error. This change fixes that by setting the permissions on the file to be writable before attempting to write to it.
It makes them easier to read.
…… (#76184) … scalar result. (#75820)" This reverts commit 701f647. The commit breaks some uses of the 'maxloc' intrinsic. See PR #75820
This diff speeds up CDSplit by not considering any hot-warm splitting point that could break a fall-through branch from a basic block to its most likely successor. Co-authored-by: spupyrev <spupyrev@fb.com>
…d (#76188) LLVM ObjectFile currently records the start offsets of sections as the start of the section header, whereas most other tools (WABT, emscripten, wasm-tools) record it as the start of the section content, after the header. This affects binutils tools such as objdump and nm, but not compilation/assembly (since that is driven by symbols and assembler labels which already have their values inside the section payload rather in the header. This patch updates LLVM to match the other tools.
This patch fixes the erroneous multiple-target requirement in Fortran
offloading tests. Additionally, it adds two new variables
(test_flags_clang, test_flags_flang) to lit.cfg so that
compiler-specific flags for Clang and Flang can be specified.
This patch re-lands: #74543. The error was caused by having:
```
config.substitutions.append(("%flags", config.test_flags))
config.substitutions.append(("%flags_clang", config.test_flags_clang))
config.substitutions.append(("%flags_flang", config.test_flags_flang))
```
when instead it has to be:
```
config.substitutions.append(("%flags_clang", config.test_flags_clang))
config.substitutions.append(("%flags_flang", config.test_flags_flang))
config.substitutions.append(("%flags", config.test_flags))
```
because LIT replaces with the first longest sub-string match.
This commit implements conditional compilation for ASan helper code. As convey to me by @EricWF, string benchmarks with UBSan have been experiencing significant performance hit after the commit with ASan string annotations. This is likely due to the fact that no-op ASan code is not optimized out with UBSan. To address this issue, this commit conditionalizes the inclusion of ASan helper function bodies using `#ifdef` directives. This approach allows us to selectively include only the ASan code when it's actually required, thereby enhancing optimizations and improving performance. While issue was noticed in string benchmarks, I expect same overhead (just less noticeable) in other containers, therefore `std::vector` and `std::deque` have same changes. To see impact of that change run `string.libcxx.out` with UBSan and `--benchmark_filter=BM_StringAssign` or `--benchmark_filter=BM_StringConstruct`.
…rate passthru operand. (#75682) ISD::VP_MERGE treats the false operand as the source for elements past VL. The vmerge instruction encodes 3 registers and treats the vd register as the source for the tail. This patch adds a new ISD opcode that models the tail source explicitly. During lowering we copy the false operand to this operand. I think we can merge RISCVISD::VSELECT_VL with this new opcode by using an UNDEF passthru, but I'll save that for another patch.
…w's getopt.h (#76137) We previously were defining _BSD_SOURCE right before including getopt.h. However, on mingw-w64, getopt.h is also transitively included by unistd.h, and unistd.h can be transitively included by many headers (recently, by some libc++ headers). Therefore, to be safe, we need to define _BSD_SOURCE before including any header. Thus do this in CMake. This fixes llvm/llvm-project#76050.
llvm/llvm-project#73077 added -Wswitch-default diagnostic but it produced false positives in templates. This PR will address that. llvm/llvm-project#75943
We can't distinguish UAR and UAS, but by definition UAR is already UAS.
|
This is ready for review.
|
@jsji, 5c5358f changes SYCL runtime, so you should tag @intel/llvm-reviewers-runtime team. |
I don't know either. Note that this is NOT exposed in our GitHub Action CI, it was exposed in the RHEL build. |
I see. CI behavior is strange, but I suggest we make 5c5358f change in a separate PR. |
Oh, had another look. This should be due to llvm/llvm-project@898320d, so this is exposed by pulldown. |
jsji
force-pushed
the
llvmspirv_pulldown
branch
from
5c5358f to
c99ffd5
Compare
|
The build passes even without 5c5358f. 😕 |
|
/merge |
|
Fri 29 Dec 2023 02:22:46 AM UTC --- Start to merge the commit into sycl branch. It will take several minutes. |
|
Fri 29 Dec 2023 02:27:08 AM UTC --- Merge the branch in this PR to base automatically. Will close the PR later. |
The build passed because it already picked up the fix in #12260. |
Pre-commit has started before #12260 was merged. It looks like some build job started after #12260 was merged and it pulled most recent commits from the SYCL branch. I expected pre-commit to use the same SYCL branch commit for all build configuration for consistent results. That's why I was confused. I can imagine a situation when Linux and Windows builds will use different commits from the SYCL branch and generate inconsistent pre-commit reports. |
Yes, it would be great if all the jobs triggered by the same commit/PR use the same |
LLVM: llvm/llvm-project@bc8c4bb
SPIRV-LLVM-Translator: KhronosGroup/SPIRV-LLVM-Translator@e63fdb8