Skip to content

Commit 432b2ab

Browse files
alexey-bataevalexey-bataev
committed
[SLP]Test for min/max reductions bug, NFC.
1 parent 3ba1b1c commit 432b2ab

File tree

1 file changed

+121
-1
lines changed

1 file changed

+121
-1
lines changed

‎llvm/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll

Lines changed: 121 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
22
; RUN: opt < %s -mtriple=x86_64-unknown-linux -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,SSE
33
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=corei7-avx -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,AVX
4-
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=core-avx2 -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,AVX
4+
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=core-avx2 -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,DEFAULT,AVX2
55
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skx -slp-vectorizer -S -slp-threshold=-100 | FileCheck %s --check-prefixes=CHECK,THRESH
66

77
@arr = local_unnamed_addr global [32 x i32] zeroinitializer, align 16
@@ -777,6 +777,25 @@ define i32 @maxi8_mutiple_uses(i32) {
777777
; AVX-NEXT: store i32 [[TMP14]], i32* @var, align 8
778778
; AVX-NEXT: ret i32 [[TMP13]]
779779
;
780+
; AVX2-LABEL: @maxi8_mutiple_uses(
781+
; AVX2-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
782+
; AVX2-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
783+
; AVX2-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
784+
; AVX2-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
785+
; AVX2-NEXT: [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
786+
; AVX2-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
787+
; AVX2-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP6]])
788+
; AVX2-NEXT: [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], [[TMP7]]
789+
; AVX2-NEXT: [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 [[TMP7]]
790+
; AVX2-NEXT: [[OP_EXTRA:%.*]] = icmp sgt i32 [[TMP10]], [[TMP5]]
791+
; AVX2-NEXT: [[OP_EXTRA1:%.*]] = select i1 [[OP_EXTRA]], i32 [[TMP10]], i32 [[TMP5]]
792+
; AVX2-NEXT: [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
793+
; AVX2-NEXT: [[TMP12:%.*]] = icmp sgt i32 [[OP_EXTRA1]], [[TMP11]]
794+
; AVX2-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[OP_EXTRA1]], i32 [[TMP11]]
795+
; AVX2-NEXT: [[TMP14:%.*]] = select i1 [[TMP4]], i32 3, i32 4
796+
; AVX2-NEXT: store i32 [[TMP14]], i32* @var, align 8
797+
; AVX2-NEXT: ret i32 [[TMP13]]
798+
;
780799
; THRESH-LABEL: @maxi8_mutiple_uses(
781800
; THRESH-NEXT: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
782801
; THRESH-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
@@ -829,6 +848,88 @@ define i32 @maxi8_mutiple_uses(i32) {
829848
ret i32 %23
830849
}
831850

851+
define i32 @maxi8_mutiple_uses2(i32) {
852+
; SSE-LABEL: @maxi8_mutiple_uses2(
853+
; SSE-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
854+
; SSE-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
855+
; SSE-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
856+
; SSE-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
857+
; SSE-NEXT: [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
858+
; SSE-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
859+
; SSE-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
860+
; SSE-NEXT: [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
861+
; SSE-NEXT: [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
862+
; SSE-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
863+
; SSE-NEXT: [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
864+
; SSE-NEXT: [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
865+
; SSE-NEXT: [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
866+
; SSE-NEXT: [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
867+
; SSE-NEXT: [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
868+
; SSE-NEXT: [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
869+
; SSE-NEXT: [[TMP18:%.*]] = select i1 [[TMP10]], i32 3, i32 4
870+
; SSE-NEXT: store i32 [[TMP18]], i32* @var, align 8
871+
; SSE-NEXT: ret i32 [[TMP17]]
872+
;
873+
; AVX-LABEL: @maxi8_mutiple_uses2(
874+
; AVX-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
875+
; AVX-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP2]])
876+
; AVX-NEXT: [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
877+
; AVX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
878+
; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP3]], i32 [[TMP4]]
879+
; AVX-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
880+
; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
881+
; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 [[TMP7]]
882+
; AVX-NEXT: [[TMP10:%.*]] = select i1 undef, i32 3, i32 4
883+
; AVX-NEXT: store i32 [[TMP10]], i32* @var, align 8
884+
; AVX-NEXT: ret i32 [[TMP9]]
885+
;
886+
; AVX2-LABEL: @maxi8_mutiple_uses2(
887+
; AVX2-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
888+
; AVX2-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP2]])
889+
; AVX2-NEXT: [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
890+
; AVX2-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
891+
; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP3]], i32 [[TMP4]]
892+
; AVX2-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
893+
; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
894+
; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 [[TMP7]]
895+
; AVX2-NEXT: [[TMP10:%.*]] = select i1 undef, i32 3, i32 4
896+
; AVX2-NEXT: store i32 [[TMP10]], i32* @var, align 8
897+
; AVX2-NEXT: ret i32 [[TMP9]]
898+
;
899+
; THRESH-LABEL: @maxi8_mutiple_uses2(
900+
; THRESH-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
901+
; THRESH-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP2]])
902+
; THRESH-NEXT: [[TMP4:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
903+
; THRESH-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
904+
; THRESH-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP3]], i32 [[TMP4]]
905+
; THRESH-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
906+
; THRESH-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
907+
; THRESH-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 [[TMP7]]
908+
; THRESH-NEXT: [[TMP10:%.*]] = select i1 undef, i32 3, i32 4
909+
; THRESH-NEXT: store i32 [[TMP10]], i32* @var, align 8
910+
; THRESH-NEXT: ret i32 [[TMP9]]
911+
;
912+
%2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
913+
%3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
914+
%4 = icmp sgt i32 %2, %3
915+
%5 = select i1 %4, i32 %2, i32 %3
916+
%6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
917+
%7 = icmp sgt i32 %5, %6
918+
%8 = select i1 %7, i32 %5, i32 %6
919+
%9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
920+
%10 = icmp sgt i32 %8, %9
921+
%11 = select i1 %10, i32 %8, i32 %9
922+
%12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
923+
%13 = icmp sgt i32 %11, %12
924+
%14 = select i1 %13, i32 %11, i32 %12
925+
%15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
926+
%16 = icmp sgt i32 %14, %15
927+
%17 = select i1 %16, i32 %14, i32 %15
928+
%18 = select i1 %10, i32 3, i32 4
929+
store i32 %18, i32* @var, align 8
930+
ret i32 %17
931+
}
932+
832933
define i32 @maxi8_wrong_parent(i32) {
833934
; SSE-LABEL: @maxi8_wrong_parent(
834935
; SSE-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
@@ -876,6 +977,25 @@ define i32 @maxi8_wrong_parent(i32) {
876977
; AVX-NEXT: [[OP_EXTRA1:%.*]] = select i1 [[OP_EXTRA]], i32 [[TMP13]], i32 [[TMP5]]
877978
; AVX-NEXT: ret i32 [[OP_EXTRA1]]
878979
;
980+
; AVX2-LABEL: @maxi8_wrong_parent(
981+
; AVX2-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
982+
; AVX2-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
983+
; AVX2-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
984+
; AVX2-NEXT: br label [[PP:%.*]]
985+
; AVX2: pp:
986+
; AVX2-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
987+
; AVX2-NEXT: [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
988+
; AVX2-NEXT: [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
989+
; AVX2-NEXT: [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
990+
; AVX2-NEXT: [[TMP9:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP6]])
991+
; AVX2-NEXT: [[TMP10:%.*]] = icmp sgt i32 [[TMP9]], [[TMP7]]
992+
; AVX2-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP9]], i32 [[TMP7]]
993+
; AVX2-NEXT: [[TMP12:%.*]] = icmp sgt i32 [[TMP11]], [[TMP8]]
994+
; AVX2-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
995+
; AVX2-NEXT: [[OP_EXTRA:%.*]] = icmp sgt i32 [[TMP13]], [[TMP5]]
996+
; AVX2-NEXT: [[OP_EXTRA1:%.*]] = select i1 [[OP_EXTRA]], i32 [[TMP13]], i32 [[TMP5]]
997+
; AVX2-NEXT: ret i32 [[OP_EXTRA1]]
998+
;
879999
; THRESH-LABEL: @maxi8_wrong_parent(
8801000
; THRESH-NEXT: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
8811001
; THRESH-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0

0 commit comments

Comments
 (0)