300e1293de
We are going to remove the old 'perfect shuffle' optimization since it brings performance penalty in hot loop around vectors. For example, in following loop sharing the same mask: %v.1 = shufflevector ... <0,1,2,3,8,9,10,11,16,17,18,19,24,25,26,27> %v.2 = shufflevector ... <0,1,2,3,8,9,10,11,16,17,18,19,24,25,26,27> The generated instructions will be `vmrglw-vmrghw-vmrglw-vmrghw` instead of `vperm-vperm`. In some large loop cases, this causes 20%+ performance penalty. The original attempt to resolve this is to pre-record masks of every shufflevector operation in DAG, but that is somewhat complex and brings unnecessary computation (to scan all nodes) in optimization. Here we disable it by default. There're indeed some cases becoming worse after this, which will be fixed in a more careful way in future patches. Reviewed By: jsji Differential Revision: https://reviews.llvm.org/D121082
39 lines
1.9 KiB
LLVM
39 lines
1.9 KiB
LLVM
; RUN: llc -verify-machineinstrs < %s -mtriple=ppc32-- -mcpu=g5 -ppc-disable-perfect-shuffle=false | not grep vperm
|
|
|
|
; TODO: Fix this case when disabling perfect shuffle
|
|
|
|
define <4 x float> @test_uu72(<4 x float>* %P1, <4 x float>* %P2) {
|
|
%V1 = load <4 x float>, <4 x float>* %P1 ; <<4 x float>> [#uses=1]
|
|
%V2 = load <4 x float>, <4 x float>* %P2 ; <<4 x float>> [#uses=1]
|
|
%V3 = shufflevector <4 x float> %V1, <4 x float> %V2, <4 x i32> < i32 undef, i32 undef, i32 7, i32 2 > ; <<4 x float>> [#uses=1]
|
|
ret <4 x float> %V3
|
|
}
|
|
|
|
define <4 x float> @test_30u5(<4 x float>* %P1, <4 x float>* %P2) {
|
|
%V1 = load <4 x float>, <4 x float>* %P1 ; <<4 x float>> [#uses=1]
|
|
%V2 = load <4 x float>, <4 x float>* %P2 ; <<4 x float>> [#uses=1]
|
|
%V3 = shufflevector <4 x float> %V1, <4 x float> %V2, <4 x i32> < i32 3, i32 0, i32 undef, i32 5 > ; <<4 x float>> [#uses=1]
|
|
ret <4 x float> %V3
|
|
}
|
|
|
|
define <4 x float> @test_3u73(<4 x float>* %P1, <4 x float>* %P2) {
|
|
%V1 = load <4 x float>, <4 x float>* %P1 ; <<4 x float>> [#uses=1]
|
|
%V2 = load <4 x float>, <4 x float>* %P2 ; <<4 x float>> [#uses=1]
|
|
%V3 = shufflevector <4 x float> %V1, <4 x float> %V2, <4 x i32> < i32 3, i32 undef, i32 7, i32 3 > ; <<4 x float>> [#uses=1]
|
|
ret <4 x float> %V3
|
|
}
|
|
|
|
define <4 x float> @test_3774(<4 x float>* %P1, <4 x float>* %P2) {
|
|
%V1 = load <4 x float>, <4 x float>* %P1 ; <<4 x float>> [#uses=1]
|
|
%V2 = load <4 x float>, <4 x float>* %P2 ; <<4 x float>> [#uses=1]
|
|
%V3 = shufflevector <4 x float> %V1, <4 x float> %V2, <4 x i32> < i32 3, i32 7, i32 7, i32 4 > ; <<4 x float>> [#uses=1]
|
|
ret <4 x float> %V3
|
|
}
|
|
|
|
define <4 x float> @test_4450(<4 x float>* %P1, <4 x float>* %P2) {
|
|
%V1 = load <4 x float>, <4 x float>* %P1 ; <<4 x float>> [#uses=1]
|
|
%V2 = load <4 x float>, <4 x float>* %P2 ; <<4 x float>> [#uses=1]
|
|
%V3 = shufflevector <4 x float> %V1, <4 x float> %V2, <4 x i32> < i32 4, i32 4, i32 5, i32 0 > ; <<4 x float>> [#uses=1]
|
|
ret <4 x float> %V3
|
|
}
|