shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/VectorCombine/AMDGPU/extract-insert-i8.ll
Nicolai Hähnle 88bd56597c
VectorCombine: Improve the insert/extract fold in the narrowing case (#168820) 
Keeping the extracted element in a natural position in the narrowed
vector has two beneficial effects:

1. It makes the narrowing shuffles cheaper (at least on AMDGPU), which
allows the insert/extract fold to trigger.
2. It makes the narrowing shuffles in a chain of extract/insert
compatible, which allows foldLengthChangingShuffles to successfully
recognize a chain that can be folded.

There are minor X86 test changes that look reasonable to me. The IR
change for AVX2 in
llvm/test/Transforms/VectorCombine/X86/extract-insert-poison.ll
doesn't change the assembly generated by `llc -mtriple=x86_64--
-mattr=AVX2`
at all.
2025-12-15 11:25:51 -08:00

114 lines
7.0 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S -mtriple=amdgcn-- -mcpu=gfx1250 -passes=vector-combine < %s | FileCheck -check-prefix=OPT %s
define <32 x i8> @extract_insert_chain(<8 x i8> %in0, <8 x i8> %in1, <8 x i8> %in2, <8 x i8> %in3) {
; OPT-LABEL: define <32 x i8> @extract_insert_chain(
; OPT-SAME: <8 x i8> [[IN0:%.*]], <8 x i8> [[IN1:%.*]], <8 x i8> [[IN2:%.*]], <8 x i8> [[IN3:%.*]]) #[[ATTR0:[0-9]+]] {
; OPT-NEXT: [[ENTRY:.*:]]
; OPT-NEXT: [[O_1_7:%.*]] = shufflevector <8 x i8> [[IN0]], <8 x i8> [[IN1]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; OPT-NEXT: [[TMP2:%.*]] = shufflevector <8 x i8> [[IN2]], <8 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; OPT-NEXT: [[O_2_7:%.*]] = shufflevector <32 x i8> [[O_1_7]], <32 x i8> [[TMP2]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
; OPT-NEXT: [[TMP3:%.*]] = shufflevector <8 x i8> [[IN3]], <8 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
; OPT-NEXT: [[O_3_7:%.*]] = shufflevector <32 x i8> [[O_2_7]], <32 x i8> [[TMP3]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>
; OPT-NEXT: ret <32 x i8> [[O_3_7]]
;
entry:
%i.0.0 = extractelement <8 x i8> %in0, i64 0
%i.0.1 = extractelement <8 x i8> %in0, i64 1
%i.0.2 = extractelement <8 x i8> %in0, i64 2
%i.0.3 = extractelement <8 x i8> %in0, i64 3
%i.0.4 = extractelement <8 x i8> %in0, i64 4
%i.0.5 = extractelement <8 x i8> %in0, i64 5
%i.0.6 = extractelement <8 x i8> %in0, i64 6
%i.0.7 = extractelement <8 x i8> %in0, i64 7
%i.1.0 = extractelement <8 x i8> %in1, i64 0
%i.1.1 = extractelement <8 x i8> %in1, i64 1
%i.1.2 = extractelement <8 x i8> %in1, i64 2
%i.1.3 = extractelement <8 x i8> %in1, i64 3
%i.1.4 = extractelement <8 x i8> %in1, i64 4
%i.1.5 = extractelement <8 x i8> %in1, i64 5
%i.1.6 = extractelement <8 x i8> %in1, i64 6
%i.1.7 = extractelement <8 x i8> %in1, i64 7
%i.2.0 = extractelement <8 x i8> %in2, i64 0
%i.2.1 = extractelement <8 x i8> %in2, i64 1
%i.2.2 = extractelement <8 x i8> %in2, i64 2
%i.2.3 = extractelement <8 x i8> %in2, i64 3
%i.2.4 = extractelement <8 x i8> %in2, i64 4
%i.2.5 = extractelement <8 x i8> %in2, i64 5
%i.2.6 = extractelement <8 x i8> %in2, i64 6
%i.2.7 = extractelement <8 x i8> %in2, i64 7
%i.3.0 = extractelement <8 x i8> %in3, i64 0
%i.3.1 = extractelement <8 x i8> %in3, i64 1
%i.3.2 = extractelement <8 x i8> %in3, i64 2
%i.3.3 = extractelement <8 x i8> %in3, i64 3
%i.3.4 = extractelement <8 x i8> %in3, i64 4
%i.3.5 = extractelement <8 x i8> %in3, i64 5
%i.3.6 = extractelement <8 x i8> %in3, i64 6
%i.3.7 = extractelement <8 x i8> %in3, i64 7
%o.0.0 = insertelement <32 x i8> poison, i8 %i.0.0, i32 0
%o.0.1 = insertelement <32 x i8> %o.0.0, i8 %i.0.1, i32 1
%o.0.2 = insertelement <32 x i8> %o.0.1, i8 %i.0.2, i32 2
%o.0.3 = insertelement <32 x i8> %o.0.2, i8 %i.0.3, i32 3
%o.0.4 = insertelement <32 x i8> %o.0.3, i8 %i.0.4, i32 4
%o.0.5 = insertelement <32 x i8> %o.0.4, i8 %i.0.5, i32 5
%o.0.6 = insertelement <32 x i8> %o.0.5, i8 %i.0.6, i32 6
%o.0.7 = insertelement <32 x i8> %o.0.6, i8 %i.0.7, i32 7
%o.1.0 = insertelement <32 x i8> %o.0.7, i8 %i.1.0, i32 8
%o.1.1 = insertelement <32 x i8> %o.1.0, i8 %i.1.1, i32 9
%o.1.2 = insertelement <32 x i8> %o.1.1, i8 %i.1.2, i32 10
%o.1.3 = insertelement <32 x i8> %o.1.2, i8 %i.1.3, i32 11
%o.1.4 = insertelement <32 x i8> %o.1.3, i8 %i.1.4, i32 12
%o.1.5 = insertelement <32 x i8> %o.1.4, i8 %i.1.5, i32 13
%o.1.6 = insertelement <32 x i8> %o.1.5, i8 %i.1.6, i32 14
%o.1.7 = insertelement <32 x i8> %o.1.6, i8 %i.1.7, i32 15
%o.2.0 = insertelement <32 x i8> %o.1.7, i8 %i.2.0, i32 16
%o.2.1 = insertelement <32 x i8> %o.2.0, i8 %i.2.1, i32 17
%o.2.2 = insertelement <32 x i8> %o.2.1, i8 %i.2.2, i32 18
%o.2.3 = insertelement <32 x i8> %o.2.2, i8 %i.2.3, i32 19
%o.2.4 = insertelement <32 x i8> %o.2.3, i8 %i.2.4, i32 20
%o.2.5 = insertelement <32 x i8> %o.2.4, i8 %i.2.5, i32 21
%o.2.6 = insertelement <32 x i8> %o.2.5, i8 %i.2.6, i32 22
%o.2.7 = insertelement <32 x i8> %o.2.6, i8 %i.2.7, i32 23
%o.3.0 = insertelement <32 x i8> %o.2.7, i8 %i.3.0, i32 24
%o.3.1 = insertelement <32 x i8> %o.3.0, i8 %i.3.1, i32 25
%o.3.2 = insertelement <32 x i8> %o.3.1, i8 %i.3.2, i32 26
%o.3.3 = insertelement <32 x i8> %o.3.2, i8 %i.3.3, i32 27
%o.3.4 = insertelement <32 x i8> %o.3.3, i8 %i.3.4, i32 28
%o.3.5 = insertelement <32 x i8> %o.3.4, i8 %i.3.5, i32 29
%o.3.6 = insertelement <32 x i8> %o.3.5, i8 %i.3.6, i32 30
%o.3.7 = insertelement <32 x i8> %o.3.6, i8 %i.3.7, i32 31
ret <32 x i8> %o.3.7
}
define <8 x i8> @extract_insert_chain_shortening(<32 x i8> %in) {
; OPT-LABEL: define <8 x i8> @extract_insert_chain_shortening(
; OPT-SAME: <32 x i8> [[IN:%.*]]) #[[ATTR0]] {
; OPT-NEXT: [[TMP1:%.*]] = shufflevector <32 x i8> [[IN]], <32 x i8> poison, <8 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
; OPT-NEXT: ret <8 x i8> [[TMP1]]
;
%i.0 = extractelement <32 x i8> %in, i64 16
%i.1 = extractelement <32 x i8> %in, i64 17
%i.2 = extractelement <32 x i8> %in, i64 18
%i.3 = extractelement <32 x i8> %in, i64 19
%i.4 = extractelement <32 x i8> %in, i64 20
%i.5 = extractelement <32 x i8> %in, i64 21
%i.6 = extractelement <32 x i8> %in, i64 22
%i.7 = extractelement <32 x i8> %in, i64 23
%o.0 = insertelement <8 x i8> poison, i8 %i.0, i32 0
%o.1 = insertelement <8 x i8> %o.0, i8 %i.1, i32 1
%o.2 = insertelement <8 x i8> %o.1, i8 %i.2, i32 2
%o.3 = insertelement <8 x i8> %o.2, i8 %i.3, i32 3
%o.4 = insertelement <8 x i8> %o.3, i8 %i.4, i32 4
%o.5 = insertelement <8 x i8> %o.4, i8 %i.5, i32 5
%o.6 = insertelement <8 x i8> %o.5, i8 %i.6, i32 6
%o.7 = insertelement <8 x i8> %o.6, i8 %i.7, i32 7
ret <8 x i8> %o.7
}
Reference in New Issue View Git Blame Copy Permalink