[PATCH] D45336: Apply accumulator to fadd/fmul experimental vector reductions (PR36734)

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon created this revision.
RKSimon added reviewers: aemerson, chandlerc, hfinkel, rengolin, mkuper, ABataev, spatel.

The llvm.experimental.vector.reduce.fadd/fmul intrinsic expansions were ignoring the accumulator scalar argument which should be added/multiplied to the scalar value from the vector reduction.

For (fast) shuffle reductions we should be able to apply the accumulator at the end of the sequence.

NOTE: I am currently working on a second patch for scalarizing strict fadd/fmul vector reductions (PR36732)


Repository:
  rL LLVM

https://reviews.llvm.org/D45336

Files:
  include/llvm/Transforms/Utils/LoopUtils.h
  lib/CodeGen/ExpandReductions.cpp
  lib/Transforms/Utils/LoopUtils.cpp
  test/CodeGen/Generic/expand-experimental-reductions.ll
  test/CodeGen/X86/vector-reduce-fadd-fast.ll
  test/CodeGen/X86/vector-reduce-fmul-fast.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D45336.141225.patch
Type: text/x-patch
Size: 41298 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180405/afe82036/attachment.bin>