Lewis Crawford ea3fdc5972

Avoid maxnum(sNaN, x) optimizations / folds (#170181) ...

The behaviour of constant-folding `maxnum(sNaN, x)` and `minnum(sNaN,
x)` has become controversial, and there are ongoing discussions about
which behaviour we want to specify in the LLVM IR LangRef.

See:
  - https://github.com/llvm/llvm-project/issues/170082
  - https://github.com/llvm/llvm-project/pull/168838
  - https://github.com/llvm/llvm-project/pull/138451
  - https://github.com/llvm/llvm-project/pull/170067
-
https://discourse.llvm.org/t/rfc-a-consistent-set-of-semantics-for-the-floating-point-minimum-and-maximum-operations/89006

This patch removes optimizations and constant-folding support for
`maxnum(sNaN, x)` but keeps it folded/optimized for `qNaN`. This should
allow for some more flexibility so the implementation can conform to
either the old or new version of the semantics specified without any
changes.

As far as I am aware, optimizations involving constant `sNaN` should
generally be edge-cases that rarely occur, so here should hopefully be
very little real-world performance impact from disabling these
optimizations.

2025-12-02 12:43:03 +00:00

..

CMakeLists.txt

[SelectionDAG] Wire up -gen-sdnode-info TableGen backend (#125358)

2025-04-06 13:14:37 +03:00

DAGCombiner.cpp

Avoid maxnum(sNaN, x) optimizations / folds (#170181)

2025-12-02 12:43:03 +00:00

FastISel.cpp

CodeGen: Remove TRI argument from getRegClass (#158225)

2025-11-10 15:43:55 -08:00

FunctionLoweringInfo.cpp

[RemoveDIs][NFC] Remove dbg intrinsic handling code from SelectionDAG ISel (#144702)

2025-06-18 16:04:18 +01:00

InstrEmitter.cpp

Add IR and codegen support for deactivation symbols.

2025-11-26 12:37:09 -08:00

InstrEmitter.h

[SelectionDAG] Pass SDValue to InstrEmitter::EmitCopyFromReg. NFC (#153485)

2025-08-13 21:46:48 -07:00

LegalizeDAG.cpp

DAG: Move expandMultipleResultFPLibCall to TargetLowering (NFC) (#166988)

2025-11-12 03:50:33 +00:00

LegalizeFloatTypes.cpp

DAG: Move expandMultipleResultFPLibCall to TargetLowering (NFC) (#166988)

2025-11-12 03:50:33 +00:00

LegalizeIntegerTypes.cpp

[SDAG] Fix whitespace errors (NFC) (#168897)

2025-11-20 16:44:31 +00:00

LegalizeTypes.cpp

[SelectionDAG] Use getShiftAmountConstant. (#158395)

2025-09-12 19:49:48 -07:00

LegalizeTypes.h

[SelectionDAG][RISCV] Support STACK/PATCHPOINT in SoftenFloatOperand. (#165922)

2025-10-31 23:31:10 -07:00

LegalizeTypesGeneric.cpp

[SelectionDAG] Rename MemSDNode::getOriginalAlign to getBaseAlign. NFC (#139930)

2025-05-16 09:37:02 -07:00

LegalizeVectorOps.cpp

[SelectionDAG] Fix AArch64 machine verifier bug when expanding LOOP_DEPENDENCE_MASK (#168221)

2025-11-15 21:12:11 +00:00

LegalizeVectorTypes.cpp

[LLVM][CodeGen] Remove failure cases when widening EXTRACT/INSERT_SUBVECTOR. (#162308)

2025-12-01 12:32:58 +00:00

MatchContext.h

…

ResourcePriorityQueue.cpp

[llvm] Support building with c++23 (#154372)

2025-08-29 12:52:07 +00:00

ScheduleDAGFast.cpp

[CodeGen] Use range-based for loops (NFC) (#138488)

2025-05-05 10:08:49 -07:00

ScheduleDAGRRList.cpp

CodeGen: Remove TRI argument from getRegClass (#158225)

2025-11-10 15:43:55 -08:00

ScheduleDAGSDNodes.cpp

DAG: Remove TargetLowering::checkForPhysRegDependency (#161787)

2025-10-03 15:58:51 +00:00

ScheduleDAGSDNodes.h

[SDAG] Use BatchAAResults for querying alias analysis (AA) results (#123934)

2025-01-23 09:16:09 +00:00

ScheduleDAGVLIW.cpp

[SDAG] Use BatchAAResults for querying alias analysis (AA) results (#123934)

2025-01-23 09:16:09 +00:00

SDNodeDbgValue.h

[llvm] Use llvm::copy (NFC) (#137470)

2025-04-26 15:50:38 -07:00

SDNodeInfo.cpp

[SelectionDAG] Verify SDTCisVT and SDTCVecEltisVT constraints (#150125)

2025-11-16 18:26:03 +03:00

SelectionDAG.cpp

Avoid maxnum(sNaN, x) optimizations / folds (#170181)

2025-12-02 12:43:03 +00:00

SelectionDAGAddressAnalysis.cpp

[CodeGen] Remove handling for lifetime.start/end on non-alloca (#149838)

2025-07-22 09:44:59 +02:00

SelectionDAGBuilder.cpp

[SelectionDAG] Add SelectionDAG::getTypeSize. NFC (#169764)

2025-12-01 10:33:50 +00:00

SelectionDAGBuilder.h

DAG: Fix constructing a temporary TargetTransformInfo instance (#168480)

2025-11-20 01:19:23 -05:00

SelectionDAGDumper.cpp

Add llvm.vector.partial.reduce.fadd intrinsic (#159776)

2025-11-07 15:36:54 +00:00

SelectionDAGISel.cpp

Add IR and codegen support for deactivation symbols.

2025-11-26 12:37:09 -08:00

SelectionDAGPrinter.cpp

…

SelectionDAGTargetInfo.cpp

[SelectionDAG] Wire up -gen-sdnode-info TableGen backend (#125358)

2025-04-06 13:14:37 +03:00

StatepointLowering.cpp

[StatepointLowering] Use FrameIndex instead of TargetFrameIndex (#153555)

2025-08-15 08:02:31 +01:00

StatepointLowering.h

…

TargetLowering.cpp

[SelectionDAG] Add SelectionDAG::getTypeSize. NFC (#169764)

2025-12-01 10:33:50 +00:00