[pull] main from llvm:main #5594
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
![pull[bot]](https://avatars.githubusercontent.com/in/12910?s=60&v=4){kind=small}
Implement olMemFill to support filling device memory with arbitrary length patterns. AMDGPU support will be added in a follow-up PR.
Fix regression introduced by #154102 - the way offload-tblgen handles names has changed
- use llvm::endian::read<> to read bit/little endian.
- Range check against size of the lookup tables instead of hardcoded
numbers.
- Make lookup tables constexpr.
- Drop {} for single-statement if-else.
This change adds intrinsics for MMA sparse. The implementation is based on PTX ISA version 8.8.
Using the default constructor makes this header work in both c++17 and
c++20 codebases. Without this, a c++20 codebase will break like this:
```c++
external/llvm-project/mlir/include/mlir/IR/Remarks.h:66:12: error: no matching constructor for initialization of 'RemarkOpts'
66 | return RemarkOpts{n, {}, {}, {}};
| ^ ~~~~~~~~~~~~~~~
external/llvm-project/mlir/include/mlir/IR/Remarks.h:58:8: note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 4 were provided
58 | struct RemarkOpts {
| ^~~~~~~~~~
external/llvm-project/mlir/include/mlir/IR/Remarks.h:58:8: note: candidate constructor (the implicit move constructor) not viable: requires 1 argument, but 4 were provided
58 | struct RemarkOpts {
| ^~~~~~~~~~
external/llvm-project/mlir/include/mlir/IR/Remarks.h:63:3: note: candidate constructor not viable: requires 0 arguments, but 4 were provided
63 | RemarkOpts() = delete;
| ^
external/llvm-project/mlir/include/mlir/IR/Remarks.h:65:31: error: constexpr function's return type 'RemarkOpts' is not a literal type
65 | static constexpr RemarkOpts name(StringRef n) {
| ^
external/llvm-project/mlir/include/mlir/IR/Remarks.h:58:8: note: 'RemarkOpts' is not literal because it is not an aggregate and has no constexpr constructors other than copy or move constructors
58 | struct RemarkOpts {
| ^
```
These tests don't work due to limitations in backend support, so it's better to mark them uniformly unsupported on AIX and z/OS.
…b` (#154630) In Windows, on a MSVC environment (e.g. when linking against the UCRT), `-nostdlib` is used (for example, by CMake) to prevent linking in non-existent `glibc`. However, an unintended side-effect is that we end up never linking in the HIP RT in these circumstances, even when `--hip-link` is explicitly specified. This breaks `hipstdpar`, where we implicitly link in the HIP RT when `--hipstdpar` is passed as a link flag. To fix this, we relax the restriction on linking the HIP RT, for known MSVC environments.
…n `modulemap`s (#148959) This PR teaches the modulemap parsing logic to report warnings that default to errors if the parsing logic sees duplicating link declarations in the same module. Specifically, duplicating link declarations means multiple link declarations with the same string-literal in the same module. No errors are reported if a same link declaration exist in a submodule and its enclosing module. The warning can be disabled with `-Wno-module-link-redeclaration`. rdar://155880064
Adds a utility getter to `warp_execute_on_lane_0` which simplifies access to the op's terminator. Uses are refactored to utilize the new terminator getter.
This patch fixes: mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp:132:3: error: default label in switch which covers all enumeration values [-Werror,-Wcovered-switch-default]
…#153993) This PR adds the following basic math functions for BFloat16 type along with the tests: - nextafterbf16 - nextdownbf16 - nexttowardbf16 - nextupbf16 --------- Signed-off-by: Krishna Pandey <[email protected]> Co-authored-by: OverMighty <[email protected]>
Constant fold the NVVM intrinsics for add, mul, div, fma with specific rounding modes.
This migrates the CompletionHandler to structured types and adds a new CompletionItem and CompletionItemType to the general types. --------- Co-authored-by: Ebuka Ezike <[email protected]>
Make it clear that other object file formats (e.g. ELF) do not use this field.
Fixes #150550. With the test case ``` void f(unsigned char *x, unsigned char *y, int n) { // should have been vectorized into avgr_u instead of seperated vectorized add and logical right shift for (int i = 0; i < n; i++) x[i] = (x[i] + y[i] + 1) / 2; } ``` the backend failed to recognize that this can be reduced to avgr_u since the loop vectorizer doesn't transform into the existing pattern in tablegen. This PR sets AVGCEIL_U as legal for v8i16 and v16i8 and selects it to avgr_u in the tablegen file.
getAsExpr() already returns Expr *.
getTargetFlags() already returns TargetFlagsType.
…rontend (#154376) This PR adds workdistribute mlir op in omp dialect and also in llvm frontend. The work in this PR is c-p and updated from @ivanradanov commits from coexecute implementation: flang_workdistribute_iwomp_2024
…oops (#155077) We cannot guarantee the validity of the interchange if the loops have iter_args, since the dependence analysis does not take them into account. Conservatively return false in such cases. Add an option to check permutation validity in test-loop-permutation pass to test this change. Signed-off-by: Prathamesh Tagore <[email protected]>
Improve const-correctness of `CheckerContext` API by defining the missing `const` overloads to its accessor member functions.
…kend (#154672) This patch adds support for reading the global timer low register in the NVVM dialect and NVPTX backend. This change includes adding the `NVVM_GlobalTimerLoOp` operation to NVVM dialect and `int_nvvm_read_ptx_sreg_globaltimer_lo` intrinsic to the NVPTX backend. All the lit tests have been added.
This was not checking the alignment requirement for 64-bit operands which accept inline immediates. Not all custom operand types were handled in the switch, so round out with explicit handling of all enum values, and change the default to use the default checks for unhandled cases. Fixes #155095
Regenerate these with a newer UTC version, so that the function signature is included. Otherwise we can get some very confusing naming on updates.
Fix nvgpu mlir file integration test. This PR fixes the bug by removing memref.get_global and then using memref.view.
I noticed a typo in the directory name `refwrap.comparissons`, then did a quick pass to fix typos elsewhere in the tests. All fixes were manual (some carefully search-and-replaced); I used [cspell](https://www.npmjs.com/package/cspell) to find them.
This is meant as the inverse of getNamedOperandIdx and returns the OpName for a given operand index for a given opcode. --------- Co-authored-by: Matt Arsenault <[email protected]>
Initialize the `OldToLValue` member with the actual old value of `ToLValue`. Pointed out by Shafik in #153601 (comment)
InstCombine tries to convert `freeze(inst(op))` to `inst(freeze(op))`. Currently, this is limited to the case where a single operand needs to be frozen, and all other operands are guaranteed non-poison. This patch allows the transform even if multiple operands need to be frozen. The existing limitation makes sure that we do not increase the total number of freezes, but it also means that that we may fail to eliminate freezes (via poison flag dropping) and may prevent optimizations (as analysis generally can't look past freeze). Overall, I believe that aggressively pushing freezes upwards is more beneficial than harmful. This is the middle-end version of #145939 in DAGCombine (which is currently reverted for SDAG-specific reasons).
#137975) An authenticated pointer can be explicitly checked by the compiler via a sequence of instructions that executes BRK on failure. It is important to recognize such BRK instruction as checking every register (as it is expected to immediately trigger an abnormal program termination) to prevent false positive reports about authentication oracles: autia x2, x3 autia x0, x1 ; neither x0 nor x2 are checked at this point eor x16, x0, x0, lsl #1 tbz x16, #62, on_success ; marks x0 as checked ; end of BB: for x2 to be checked here, it must be checked in both ; successor basic blocks on_failure: brk 0xc470 on_success: ; x2 is checked ldr x1, [x2] ; marks x2 as checked
…atten attribute (#154801) Fixes #149866 --------- Co-authored-by: Aaron Ballman <[email protected]>
) We were sizing the table appropriately for the number of LibcallImpls, but many of those have identical names which were pushing up the collision count unnecessarily. This ends up decreasing the table size slightly, and makes it a bit faster. BM_LookupRuntimeLibcallByNameRandomCalls improves by ~25% and BM_LookupRuntimeLibcallByNameSampleData by ~5%. As a secondary change, align the table size up to the next power of 2. This makes the table larger than before, but improves the sample data benchmark by an additional 5%.
…cs (#152575) Previously, the alignment of pointer arithmetics was inferred from the pointee type, losing the alignment information from its operands: https://github.com/llvm/llvm-project/blob/503c0908c3450d228debd64baecf41df8f58476e/clang/lib/CodeGen/CGExpr.cpp#L1446-L1449 This patch preserves alignment information for pointer arithmetics `P +/- C`, to match the behavior of identical array subscript `&P[C]`: https://godbolt.org/z/xx1hfTrx4. Closes #152330. Although the motivating case can be fixed by #145733, the alignment cannot be recovered without a dominating memory access with larger alignment.
We need to use the base offset in both cases. Also, add additional assertions to make sure we don't miss this case again. Fixes #155132
Some buildbots build unittests with `-Werror,-Wsign-compare`.
Improve the documentation of `replaceUsesOfBlockArgument` to clarify its semantics is rollback mode. Add an assertion to make sure that the same block argument is not replaced multiple times. That's an API violation and messes with the internal state of the conversion driver. This commit is in preparation of adding full support for `RewriterBase::replaceAllUsesWith`.
This reverts commit 0f4db1a. This breaks the armv8-quick builder: https://lab.llvm.org/buildbot/#/builders/154/builds/20676
…154377) This PR adds workdistribute parser and semantic support in flang. The work in this PR is c-p and updated from @ivanradanov commits from coexecute implementation: flang_workdistribute_iwomp_2024
Following #144344, #152207, #151690, this PR adds the alignment attribute to the following operations in the vector dialect: * `compressstore` * `expandload` * `vector.scatter` * `vector.gather` --------- Co-authored-by: Jakub Kuderski <[email protected]>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.3)
Can you help keep this open source service alive? 💖 Please sponsor : )