[PATCH] D88154: Initial support for vectorization using Libmvec (GLIBC vector math library).

Fri Oct 16 07:38:36 PDT 2020

spatel added a subscriber: aeubanks.
spatel added a comment.

In D88154#2328312 <https://reviews.llvm.org/D88154#2328312>, @venkataramanan.kumar.llvm wrote:

> I can add few more float type tests with meta data for VF=8.   please let me know your suggestions.

I may be missing some subtlety of the vectorizer behavior. Can we vary the test types + metadata in 1 file,s o that there is coverage for something like this v2f64 call : `TLI_DEFINE_VECFUNC("llvm.sin.f64", "_ZGVbN2v_sin", 2)`?
I'm just trying to make sure we don't fall into some blind-spot by only testing VF=4.

================
Comment at: llvm/test/Transforms/LoopVectorize/X86/libm-vector-calls-finite.ll:2
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -vector-library=LIBMVEC-X86 -inject-tli-mappings -loop-vectorize -S < %s | FileCheck %s
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
----------------
fhahn wrote:
> fpetrogalli wrote:
> > `-inject-tli-mappings` is not required here, as a pass itself is required by the loop vectorizer.
> I guess it still doesn't hurt to be explicit. Also, can you add a line for the new pass manager?
We need to be explicit about that pass with new-pass-manager as shown here:
df5576a

cc @aeubanks as I'm not sure if we want to update tests with NPM RUN lines or if we want to silently transition whenever the default gets changed.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D88154/new/

https://reviews.llvm.org/D88154