[PATCH] D138670: [AArch64][SME]: Generate streaming-compatible code for int/fp compares.

David Sherwood via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Nov 25 05:46:17 PST 2022


david-arm added inline comments.


================
Comment at: llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-compares.ll:39
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl16
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
----------------
I'm a bit confused by the old CHECK lines - it looks like this already used SVE perfectly fine before? Also, it looks like this is assuming a minimum SVE vector length of 256 bits because there is only a single `fcmeq` instruction for a <16 x half> operation.

I noticed similar things in the other tests too - would you mind taking a look into this and finding out what's happening here?


================
Comment at: llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-fp-compares.ll:10
 
 define <2 x i16> @fcmp_oeq_v2f16(<2 x half> %op1, <2 x half> %op2) #0 {
   %cmp = fcmp oeq <2 x half> %op1, %op2
----------------
hassnaa-arm wrote:
> sdesmalen wrote:
> > Why does this file not have CHECK lines for each test?
> because this is a new test, and the test file was crashing, so I couldn't generate new code for the precursory patch.
In this case @hassnaa-arm I think it probably makes sense to add this test only in this patch, otherwise the pre-commit patch will cause build failures. I know that if you commit them in sequence quickly it shouldn't matter much, but if for anyone reason you have to revert this patch then you'll also have to revert the pre-commit too.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D138670/new/

https://reviews.llvm.org/D138670



More information about the llvm-commits mailing list