[all-commits] [llvm/llvm-project] 7cc6b8: [RISCV][CostModel] Model vrgather.vv as being quad...

[Philip Reames via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 7cc6b80d9a97602bd8f3bc5333cd26020a3726da
      https://github.com/llvm/llvm-project/commit/7cc6b80d9a97602bd8f3bc5333cd26020a3726da
  Author: Philip Reames <preames at rivosinc.com>
  Date:   2023-07-18 (Tue, 18 Jul 2023)

  Changed paths:
    M llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
    M llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
    M llvm/test/Analysis/CostModel/RISCV/rvv-shuffle.ll
    M llvm/test/Analysis/CostModel/RISCV/shuffle-interleave.ll
    M llvm/test/Analysis/CostModel/RISCV/shuffle-permute.ll
    M llvm/test/Analysis/CostModel/RISCV/shuffle-reverse.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/interleaved-accesses.ll
    M llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll

  Log Message:
  -----------
  [RISCV][CostModel] Model vrgather.vv as being quadradic in LMUL

vrgather.vv across multiple vector registers (i.e. LMUL > 1) requires all to all data movement. This includes two conceptual sets of changes:

    For permutes, we were modeling these as being linear in LMUL.
    For reverse, we were modeling them as being fixed cost in LMUL.

Both were wrong, and have been adjusted to O(LMUL^2).  Noticed via code inspection while looking at something else.

Its worth asking whether we should be lowering reverse to something other than a vrgather at high LMULs. That shuffle is quite expensive.  (Future work)

Differential Revision: https://reviews.llvm.org/D152019