[all-commits] [llvm/llvm-project] 0db5d2: [LV] Use index type for base pointer computation i...

[Mel Chen via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 0db5d27170ba624b1b409cc68a1868023b4c41ef
      https://github.com/llvm/llvm-project/commit/0db5d27170ba624b1b409cc68a1868023b4c41ef
  Author: Mel Chen <mel.chen at sifive.com>
  Date:   2026-06-08 (Mon, 08 Jun 2026)

  Changed paths:
    M llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
    M llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/masked_gather_scatter.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/strided-access-wide-stride.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll

  Log Message:
  -----------
  [LV] Use index type for base pointer computation in convertToStridedAccesses (#201070)

The base pointer for strided accesses was computed as:
 ```
   offset = canonicalIV * stride
   base_ptr = ptradd start, offset
 ```
On a 64-bit target, if the canonical IV type is i32, the GEP operation
for ptradd will first sign-extend the offset to i64. Once the offset
multiplication has already overflowed in i32, it will ultimately result
in an incorrect base address.

This patch fixes this by extending the canonical IV to the index type
before the offset multiplication.

Based on #199647
Fixes #199640



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