[PATCH] D127548: [Matrix] Add cost tests for fast fadd reduction
Florian Hahn via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jun 13 09:37:31 PDT 2022
fhahn added a comment.
Thanks! Could you also add tests for reductions with just the `reassoicate` fast-math flag?
Also, this is not really specific to matrix support in LLVM, but applies generally to the AArch64 cost modeling. So it would be better to use `[AArch64]` in the description.
================
Comment at: llvm/test/Analysis/CostModel/AArch64/reduce-fadd.ll:25
+
+define void @strict_fp_reductions_fast() {
+; CHECK-LABEL: 'strict_fp_reductions_fast'
----------------
I think the `strict` part here refers to the ordering requirement, i.e. strict sequential order.
So it would probably be better to name the function `@fast_fp_reductions`.
================
Comment at: llvm/test/Analysis/CostModel/AArch64/reduce-fadd.ll:40
+ %fadd_v2f64 = call fast double @llvm.vector.reduce.fadd.v2f64(double 0.0, <2 x double> undef)
+ %fadd_v4f64 = call fast double @llvm.vector.reduce.fadd.v4f64(double 0.0, <4 x double> undef)
+
----------------
can you also throw in some vectors with some odd numbers of elements.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D127548/new/
https://reviews.llvm.org/D127548
More information about the llvm-commits
mailing list