aboutsummaryrefslogtreecommitdiff
AgeCommit message (Collapse)AuthorFilesLines
2026-01-13compacted the virt-reg numbersusers/cdevadas/precommit-subreg-reload-testChristudasan Devadasan1-14/+14
2026-01-13[AMDGPU] Test precommit for subreg reloadChristudasan Devadasan1-0/+37
This test currently fails due to insufficient registers during allocation. Once the subreg reload is implemented, it will begin to pass as the partial reload help mitigate register pressure.
2026-01-13[AMDGPU] Put back ProperlyAlighedRC helper functionsusers/cdevadas/putback-ProperlyAlighedRCChristudasan Devadasan2-0/+27
Putting back the functions that are recently deleted as they were found unused. They are needed for implementing subreg reload during RA.
2026-01-13[CodeGen] Enhance createFrom for sub-reg aware cloningusers/cdevadas/enhance-createFrom-functionChristudasan Devadasan2-3/+14
Instead of just cloning the virtual register, this function now creates a new virtual register derived from a subregister class of the original value.
2026-01-13suggestions incorporated.users/cdevadas/make-AMDGPURewriteAGPRCopyMFMA-subreg-reload-awareChristudasan Devadasan1-2/+2
2026-01-13[AMDGPU] Make AMDGPURewriteAGPRCopyMFMA aware of subreg reloadChristudasan Devadasan3-1/+54
AMDGPURewriteAGPRCopyMFMA pass is currently not subreg-aware. In particular, the logic that optimizes spills into COPY instructions assumes full register reloads. This becomes problematic when the reload instruction partially restores a tuple register. This patch introduces the necessary changes to make this pass subreg-aware, for a future patch that implements subreg reload during RA.
2026-01-13fixed a comment.users/cdevadas/make-getNumSubRegsForSpillOp-member-functionChristudasan Devadasan1-1/+1
2026-01-13moved the implementation to SIInstrInfo.Christudasan Devadasan4-151/+150
2026-01-13[AMDGPU] Make getNumSubRegsForSpillOp externally available (NFC).Christudasan Devadasan2-3/+5
2026-01-13incorporated review comments.users/cdevadas/add-spill-offset-to-sgpr-spill-pseudosChristudasan Devadasan1-3/+3
2026-01-13[AMDGPU] Introduce Offset field in SGPR spill PseudosChristudasan Devadasan41-165/+167
Currently, SGPR spill pseudo-instructions lack an offset field to represent non-zero stack offsets. This patch introduces an additional offset field to SGPR spill pseudo-instructions and updates all relevant passes that handle spill lowering to support this new field. This field is essential for a future patch that implements subreg reload of tuple registers from their stack location during RA.
2026-01-13[CodeGen][InlineSpiller] Add SubReg argument to loadRegFromStackSlot for ↵Christudasan Devadasan48-53/+63
subreg-reload (#175581) This preparatory patch introduces an additional argument to the target hook loadRegFromStackSlot. Ths is essential for targets to handle subregister-specific reload in the future. See how this is used for AMDGPU target with PR #175002.
2026-01-13[LV][NFC] Follow-up fix for #173262 (#175513)Mel Chen1-7/+6
2026-01-13[UniformityAnalysis] Jump over reducible cycles when locating join blocks ↵Sameer Sahasrabuddhe2-24/+78
(#174938) When locating the join blocks of a divergent block, the algorithm relies on pseudo-edges from the header of a reducible cycle to the cycle exits. This was missed in the actual traversal, producing unnecessary joins inside the reducible cycle. This caused an assert in the included test, which expected that if a join existed in a reducible cycle for a divergent branch outside the cycle, then it must be header. This fixes the reverted commit from #174117
2026-01-13[flang][CUDA] Apply implicit managed attribute when `-gpu=mem:managed` is ↵Zhen Wang3-2/+175
used. (#175648) When `-gpu=mem:managed` is used, allocatable arrays without explicit CUDA data attributes are implicitly treated as managed. The `-gpu=mem:managed` flag to enable this feature is currently only supported in `bbc`.
2026-01-12[NFC][Offload] Rename a function (#175673)Hansang Bae1-6/+6
Renamed a function as suggested in #175664.
2026-01-13[LLVM] Ignore two Cursor specific files. (#175683)Daniil Fukalov1-0/+2
2026-01-12[RISCV] Add the missing SEW search table field to vector FMA instructions ↵Min-Yih Hsu4-210/+210
(#175646) We split vector floating point FMA (pseudo) instructions' opcodes by SEW since c6b7944be4dfbb1fb35301c670812726845acaa7 , but forgot to populate their `SEW` field, which is used by various search tables. This results in incorrect pseudo instruction opcodes lookup -- and to a larger extent, incorrect scheduling class lookups -- in llvm-mca. This patch fixes such issue.
2026-01-12[cmake] Make CMAKE_BUILD_TYPE=Release the default (#174520)Reid Kleckner1-8/+3
Currently, we report a fatal error if the user leaves CMAKE_BUILD_TYPE blank. This was implemented in https://reviews.llvm.org/D124153 / 350bdf9227ceb , based on this RFC: https://discourse.llvm.org/t/rfc-select-a-better-linker-by-default-or-warn-about-using-bfd/61899/1 Tom Stellard mentioned that he'd like to revisit this on Discord, and Aiden, myself, and apparently most people on the original RFC agree, so I'm proposing we do it. However, on the review, several folks objected and insisted that Debug was a better default. I want to reopen the question. I think we've made the wrong tradeoff. I wish Debug builds worked out of the box on most systems, but they don't, and LLVM has only gotten bigger over the last four years, making the build scalability problems of Debug builds worse. I think we should optimize our build configuration for new developers, not experienced longtime contributors who are invested enough to tweak the build to their liking. With this PR, we emit a warning, and set the build type to Release, which has a higher likelihood of success for first-time users. Making the build work out of the box is very important for making LLVM development more accessible to new contributors, so it seems worth smoothing over this rough edge. A separate possible improvement would be to set LLVM_ENABLE_ASSERTIONS=ON, but that is out of scope for this PR.
2026-01-12Remove cleanup of incorrect output in test dir (#171256)Mircea Trofin1-1/+0
This follows #171255 , removing the cleanup line.
2026-01-12[profcheck] Fix encoding of 0 loopEstimatedTrip count (#174896)Mircea Trofin4-9/+12
We currently encode an estimated trip count of 0 as the latch having branch probabilities 0-0. That's an invalid pair of weights. The probability of a branch is computed as a fraction of its corresponding weight and the sum of the weights. In fact, `BranchProbabilityInfo::calcMetadataWeights` will convert this to a 1-1, meaning 50% - 50%, which isn't quite what we want. To indicate the loop is never taken, we just need to initialize the exit probability to non-zero (hence, 1) Related: https://reviews.llvm.org/D67905 Issue #147390
2026-01-13[AMDGPU]Add specific instruction feature for multicast load (#175503)Shoreshen7-14/+28
2026-01-13[orc-rt] Add Maintainers.md. (#175691)Lang Hames1-0/+10
2026-01-13[RISCV] Adjust base cost for Xqcilo loads/stores in RISCVMakeCompressible ↵Sudharsan Veeravalli2-12/+86
(#175572) We only need two uses in Xqcilo load/store instructions for the base adjustment to be profitable as compared to three uses in the base load/store instructions.
2026-01-12[CIR][X86] Add support for `intersect` builtins (#172554)Vishruth Thimmaiah3-10/+302
adds support for the `__builtin_ia32_vp2intersect_d`/`__builtin_ia32_vp2intersect_q` x86 builtins. Part of #167765 --------- Signed-off-by: vishruth-thimmaiah <vishruththimmaiah@gmail.com>
2026-01-12[FIX] Add target requirement in ↵Shilei Tian1-0/+1
`clang/test/CodeGen/AMDGPU/nullptr-in-different-address-spaces.cpp`
2026-01-12[flang] Turn -Werror back off for Flang build (#175689)Peter Klausler2-2/+1
Different build environments are picking up warnings that my testing didn't expose; turn -Werror back off. (And also delete an unused data member that was triggering some MSVC warnings.)
2026-01-13[RISCV] Use LD_RV32/SD_RV32 for spills and reloads when Zilsd is enabled ↵Sudharsan Veeravalli3-42/+385
(#153595) We are currently only using `PseudoRV32ZdinxSD/LD` for spills and reloads when the register class is `GPRPairRegClass` . However, we can use `LD_RV32/SD_RV32` when the `Zilsd` extension is enabled and certain alignment requirements are met.
2026-01-13[Clang][AMDGPU] Get correct nullptr value for AS3 and AS5 (#175610)Shilei Tian3-5/+77
2026-01-13[Darwin] [compiler-rt] Normalize DYLD_LIBRARY_PATH to workaround LD bug ↵Andrew Haberlandt1-0/+6
(#175685) There is an issue in certain versions of LD which causes the wrong libLTO to be used if the DYLD_LIBRARY_PATH is not normalized. Will fix these failures: ``` AddressSanitizer-x86_64-darwin.TestCases/Darwin.odr-lto.cpp AddressSanitizer-x86_64h-darwin.TestCases/Darwin.odr-lto.cpp ``` https://green.lab.llvm.org/job/llvm.org/job/clang-stage1-cmake-RA-incremental/13428/ rdar://168024431
2026-01-12[flang] Temporarily disabled a check in a test that was breaking buildbots ↵Eugene Epshteyn1-3/+5
(#175684) PR https://github.com/llvm/llvm-project/pull/175383 had breaking test Semantics/OpenMP/linear-clause01.f90 I disabled problematic part of the test for now to let the builds pass. I will file the issue for PR author to fix the test.
2026-01-13[clang][Tooling] Fix assertion failure when processing CUDA files (#173762)mitchell3-8/+26
Running clang-tidy on CUDA files without specifying `--cuda-host-only` or `--cuda-device-only` would trigger an assertion failure in `Actions.size() > 1`, a related discussion: https://github.com/llvm/llvm-project/pull/173699#discussion_r2649279975. This occurred because the Clang Driver generates a single top-level `OffloadAction` in `-fsyntax-only`, `-E`, `-M`. This commit removes the overly strict assertions. Closes #173777
2026-01-13[llvm-jitlink] Remove flaky testcase. (#175680)Lang Hames1-41/+0
On some systems, backtraces contain addresses with their high bits set*. These high bits prevent symbolication using the JIT symbol table. Since this test is for a best-effort debugging / diagnosis tool it seems best to remove the test until/unless we can get it passing on all systems, or find some way to identify systems that will fail. See discussion in https://github.com/llvm/llvm-project/pull/175537. * Note that the test does not use PAC or pointer tagging -- the high bits are coming from somewhere else. Possibly libunwind, but that is just speculation.
2026-01-12workflows/commit-access-review: Automatically create the issue (#175400)Tom Stellard2-0/+51
We had been doing this manually, and this will be a lot more convenient.
2026-01-12[flang] Add missing module file hashes (#175636)Peter Klausler2-10/+51
The compiler emits "!need$" lines to module files only for modules needed by the module's outermost scope, but misses dependences on other modules that might be USE'd in inner scopes. Fixes https://github.com/llvm/llvm-project/issues/175611.
2026-01-12[flang] Fix spurious NaN result from infinite Kahan summation (#175373)Peter Klausler10-40/+78
There are six instances of Kahan's extended precision summation algorithm in flang/flang-rt, and they share a bug: the calculation of the correction value produces a Nan due to the subtraction Inf-Inf after the accumulation saturates to Inf. This leads to the surprising Nan result from SUM([Inf, 0.]). This bug doesn't affect run-time calculation of SUM when optimization is enabled -- lowering emits an open-coded SUM that lacks Kahan summation -- but it does affect compilation-time folding and -O0 runtime results. Fix the one instance of Kahan summation in the runtime, and consolidate the other five instances in Evaluate into one new member function, also corrected. Fixes https://github.com/llvm/llvm-project/issues/89528.
2026-01-12[flang] Handle useless NAMELIST in interface block (#175214)Peter Klausler2-0/+13
NAMELIST has no useful purpose in an interface block, but it's allowed. Fix a crash due to our deferred handling of NAMELIST groups in the execution part (which doesn't exist in an interface block). Fixes https://github.com/llvm/llvm-project/issues/175207.
2026-01-12[flang][NFC] Document that the integrity of parentheses is not violated ↵Peter Klausler1-0/+21
(#175071) Add language to flang/docs/Extensions.md to explain why "A+(B*C)" must round the result of the multiplication, when REAL and the -ffast-math option is not used.
2026-01-12[llvm][flang] Silence warning, resume -Werror builds of flang (#175053)Peter Klausler3-8/+4
Add (void) uses of two parameters to dodge a C++ compiler warning that has broken -Werror builds of flang since 9-28-25, and restore that option as the default for flang builds.
2026-01-12[lldb][TypeSystem] Call Type::isIntegralType from ↵Michael Buch2-9/+54
TypeSystemClang::IsIntegerType (#175669) Instead of re-implementing `Type::isIntegralType`, call it explicitly. This means we get support for `BitIntType` out-of-the-box. We don't use `IsIntegerType` here because we want to abide by the language-specific notions of an integer type (which differ between C++ and C). The slight behaviour change here is that `IsIntegerType` will now treat complete enumerations as integers in C. This is correct according to the C standard.
2026-01-12[CIR] Implement simple folding for unary operations (#174882)Andy Kaylor11-47/+289
This extends the UnaryOp folder to handle plus, minus, and not operations on constant operands. This is in preparation for a change that will attempt to fold these unary operations as they are generated, but this change only performs the folding via the cir-canonicalize pass.
2026-01-12[llvm][RISCV] Suppress unused `IsMulH` warning. (#175653)Chenguang Wang1-2/+3
Introduced in commit d28daddd. `IsMulH` is only used in assert(), and triggers unused variable warnings in non-debug builds.
2026-01-12[X86][NewPM] Port x86-pre-tile-config (#175649)Aiden Grossman4-23/+57
Standard pass porting. Used callbacks to get MLI so we do not compute it in the common case where we have no AMX registers. Also moved the call to releaseMemory into runOnMachineFunction through a scope exit rather than calling it through the pass manager so we can get consistent behavior across both PMs. No test coverage added in this one as we also need x86-tile-config to be able to run any tests.
2026-01-12[Offload] Fix level_zero plugin build (#175664)Hansang Bae1-2/+0
Build has been broken when OMPTARGET_DEBUG is undefined.
2026-01-12KnownFPClass: Make LLVM_ABI checker happy (#175654)Matt Arsenault1-1/+1
Warned on PRs that happened to touch nearby lines.
2026-01-12[lldb][TypeSystemClang] Simplify TypeSystemClang::IsEnumerationType ↵Michael Buch2-5/+59
implementation (#175560) We were calling into `IsIntegerType` to determine the signedness of the enum. Calling the relevant `clang::Type` API is simpler. This shouldn't have any observable behaviour change. We were lacking unit-test coverage for this. Added some tests that pass before and after this change.
2026-01-12[VPlan] Merge Select with previous cases in ::computeCost (NFC).Florian Hahn1-1/+0
Merge cases calling the same helper, as suggested post-commit in https://github.com/llvm/llvm-project/pull/174234
2026-01-12[X86][NewPM] Port x86-tile-config (#175647)Aiden Grossman5-19/+46
Standard porting. Use callbacks to get the needed analyses to make the pass portable between Legacy/New PMs and to prevent computing anything if we do not have any AMX registers in the function. No test coverage for now as amx-greedy-ra.ll is the only test that references this pass and needs pass pipeline setup in order to work which I plan on getting to this week.
2026-01-12[lldb][NativePDB] NFC: Add language-agnostic hooks for AST population (#175624)Leonard Grey3-12/+39
Currently, `SymbolFileNativePDB` calls several `PdbAstBuilder` methods for side-effects to ensure the AST is populated. This change adds new void-returning methods for `SymbolFileNativePDB` to use as a hook instead, so that it doesn't depend on Clang-specific parts of `PdbAstBuilder`'s interface. This is part of the work to allow language-agnostic `PdbAstBuilder` (see RFC: https://discourse.llvm.org/t/rfc-lldb-make-pdbastbuilder-language-agnostic/89117)
2026-01-12[lsan] Fix flaky test in swapcontext.cpp (#175635)Thurston Dang1-0/+10
LSan, by design, can have false negatives, making it unreliable to check that the leak was found in the stack-allocated case: ``` ==123685==Scanning STACK range 0x7ffe6e554ca0-0x7ffe6e557000. ==123685==0x7ffe6e554de0: found 0x51e0000009f0 pointing into chunk 0x51e000000000-0x51e000000c00 of size 3072. ==123685==0x7ffe6e554e30: found 0x51e000000c00 pointing into chunk 0x51e000000c00-0x51e000001668 of size 2664. <- this prevented the leak from being found ``` This has led to flakiness on the buildbots e.g., https://lab.llvm.org/buildbot/#/builders/66/builds/24669 ``` # | /home/b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/lsan/TestCases/swapcontext.cpp:44:11: error: CHECK: expected string not found in input # | // CHECK: SUMMARY: {{.*}}Sanitizer: 2664 byte(s) leaked in 1 allocation(s) ... Failed Tests (2): LeakSanitizer-HWAddressSanitizer-x86_64 :: TestCases/swapcontext.cpp LeakSanitizer-Standalone-x86_64 :: TestCases/swapcontext.cpp ``` This patch fixes the issue by clearing the buffer, as suggested by Vitaly.
generated by cgit v1.2.3 (git 2.39.1) at 2026-05-16 06:46:47 +0000