[all-commits] [llvm/llvm-project] 6f98b3: [AMDGPU][SDAG] DAGCombine PTRADD -> disjoint OR

[Fabian Ritter via All-commits]

  Branch: refs/heads/users/ritter-x2a/06-27-_amdgpu_sdag_dagcombine_ptradd_-_disjoint_or
  Home:   https://github.com/llvm/llvm-project
  Commit: 6f98b3a7b6207c50e13d1a549e2098d93875be7b
      https://github.com/llvm/llvm-project/commit/6f98b3a7b6207c50e13d1a549e2098d93875be7b
  Author: Fabian Ritter <fabian.ritter at amd.com>
  Date:   2025-07-18 (Fri, 18 Jul 2025)

  Changed paths:
    M llvm/lib/Target/AMDGPU/SIISelLowering.cpp
    M llvm/test/CodeGen/AMDGPU/ptradd-sdag-optimizations.ll

  Log Message:
  -----------
  [AMDGPU][SDAG] DAGCombine PTRADD -> disjoint OR

If we can't fold a PTRADD's offset into its users, lowering them to
disjoint ORs is preferable: Often, a 32-bit OR instruction suffices
where we'd otherwise use a pair of 32-bit additions with carry.

This needs to be a DAGCombine (and not a selection rule) because its
main purpose is to enable subsequent DAGCombines for bitwise operations.
We don't want to just turn PTRADDs into disjoint ORs whenever that's
sound because this transform loses the information that the operation
implements pointer arithmetic, which we will soon need to fold offsets
into FLAT instructions. Currently, disjoint ORs can still be used for
offset folding, so that part of the logic can't be tested.

The PR contains a hacky workaround for a situation where an AssertAlign
operand of a PTRADD is not DAGCombined before the PTRADD, causing the
PTRADD to be turned into a disjoint OR although reassociating it with
the operand of the AssertAlign would be better. This wouldn't be a
problem if the DAGCombiner ensured that a node is only processed after
all its operands have been processed.

For SWDEV-516125.


  Commit: 8be2fb1e0fd1108873caff68c7c3566131ce0dcd
      https://github.com/llvm/llvm-project/commit/8be2fb1e0fd1108873caff68c7c3566131ce0dcd
  Author: Fabian Ritter <fabian.ritter at amd.com>
  Date:   2025-07-18 (Fri, 18 Jul 2025)

  Changed paths:
    M llvm/test/CodeGen/AMDGPU/ptradd-sdag-optimizations.ll

  Log Message:
  -----------
  Remove unnecessary attributes in test


  Commit: 44a37868de3a4d3ffc5971a00f5854693d696d21
      https://github.com/llvm/llvm-project/commit/44a37868de3a4d3ffc5971a00f5854693d696d21
  Author: Fabian Ritter <fabian.ritter at amd.com>
  Date:   2025-07-18 (Fri, 18 Jul 2025)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    M llvm/lib/Target/AMDGPU/SIISelLowering.cpp

  Log Message:
  -----------
  Move ptradd -> disjoint OR combine to generic combines


Compare: https://github.com/llvm/llvm-project/compare/f7b627f5bb3b...44a37868de3a

To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications