[llvm] r313042 - [SLP] Test with mutiple uses of conditional op and wrong parent.
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 12 09:15:04 PDT 2017
Author: abataev
Date: Tue Sep 12 09:15:04 2017
New Revision: 313042
URL: http://llvm.org/viewvc/llvm-project?rev=313042&view=rev
Log:
[SLP] Test with mutiple uses of conditional op and wrong parent.
Modified:
llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll
Modified: llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll?rev=313042&r1=313041&r2=313042&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll (original)
+++ llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll Tue Sep 12 09:15:04 2017
@@ -2,10 +2,11 @@
; RUN: opt < %s -mtriple=x86_64-unknown-linux -slp-vectorizer -S | FileCheck %s
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=corei7-avx -slp-vectorizer -S | FileCheck %s --check-prefix=AVX
; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=core-avx2 -slp-vectorizer -S | FileCheck %s --check-prefix=AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skx -slp-vectorizer -S | FileCheck %s --check-prefix=SKX
+; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skx -slp-vectorizer -S -slp-threshold=-100 | FileCheck %s --check-prefix=SKX
@arr = local_unnamed_addr global [32 x i32] zeroinitializer, align 16
@arr1 = local_unnamed_addr global [32 x float] zeroinitializer, align 16
+ at var = global i32 zeroinitializer, align 8
define i32 @maxi8(i32) {
; CHECK-LABEL: @maxi8(
@@ -35,45 +36,111 @@ define i32 @maxi8(i32) {
;
; AVX-LABEL: @maxi8(
; AVX-NEXT: [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP24:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP25:%.*]] = icmp sgt <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x i32> [[BIN_RDX]], <8 x i32> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP26:%.*]] = icmp sgt <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x i32> [[BIN_RDX2]], <8 x i32> [[RDX_SHUF3]]
-; AVX-NEXT: [[TMP27:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
-; AVX: ret i32 [[TMP27]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
+; AVX-NEXT: [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX-NEXT: ret i32 [[TMP24]]
;
; AVX2-LABEL: @maxi8(
; AVX2-NEXT: [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP24:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP25:%.*]] = icmp sgt <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x i32> [[BIN_RDX]], <8 x i32> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP26:%.*]] = icmp sgt <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x i32> [[BIN_RDX2]], <8 x i32> [[RDX_SHUF3]]
-; AVX2-NEXT: [[TMP27:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
-; AVX2: ret i32 [[TMP27]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
+; AVX2-NEXT: [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX2-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX2-NEXT: ret i32 [[TMP24]]
;
; SKX-LABEL: @maxi8(
; SKX-NEXT: [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP24:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP25:%.*]] = icmp sgt <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x i32> [[BIN_RDX]], <8 x i32> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP26:%.*]] = icmp sgt <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x i32> [[BIN_RDX2]], <8 x i32> [[RDX_SHUF3]]
-; SKX-NEXT: [[TMP27:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
-; SKX: ret i32 [[TMP27]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
+; SKX-NEXT: [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; SKX-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; SKX-NEXT: ret i32 [[TMP24]]
;
%2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
%3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
@@ -152,54 +219,192 @@ define i32 @maxi16(i32) {
;
; AVX-LABEL: @maxi16(
; AVX-NEXT: [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP48:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[BIN_RDX]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP49:%.*]] = icmp sgt <16 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x i32> [[BIN_RDX]], <16 x i32> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x i32> [[BIN_RDX2]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP50:%.*]] = icmp sgt <16 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x i32> [[BIN_RDX2]], <16 x i32> [[RDX_SHUF3]]
-; AVX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x i32> [[BIN_RDX4]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x i32> [[BIN_RDX4]], <16 x i32> [[RDX_SHUF5]]
-; AVX-NEXT: [[TMP52:%.*]] = extractelement <16 x i32> [[BIN_RDX6]], i32 0
-; AVX: ret i32 [[TMP52]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX-NEXT: [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
+; AVX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; AVX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; AVX-NEXT: [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
+; AVX-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; AVX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; AVX-NEXT: [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
+; AVX-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; AVX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; AVX-NEXT: [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
+; AVX-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; AVX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; AVX-NEXT: [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
+; AVX-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; AVX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; AVX-NEXT: [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
+; AVX-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; AVX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; AVX-NEXT: [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
+; AVX-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; AVX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; AVX-NEXT: [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
+; AVX-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
+; AVX-NEXT: [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; AVX-NEXT: ret i32 [[TMP48]]
;
; AVX2-LABEL: @maxi16(
; AVX2-NEXT: [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP48:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[BIN_RDX]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP49:%.*]] = icmp sgt <16 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x i32> [[BIN_RDX]], <16 x i32> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x i32> [[BIN_RDX2]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP50:%.*]] = icmp sgt <16 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x i32> [[BIN_RDX2]], <16 x i32> [[RDX_SHUF3]]
-; AVX2-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x i32> [[BIN_RDX4]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX2-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x i32> [[BIN_RDX4]], <16 x i32> [[RDX_SHUF5]]
-; AVX2-NEXT: [[TMP52:%.*]] = extractelement <16 x i32> [[BIN_RDX6]], i32 0
-; AVX2: ret i32 [[TMP52]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX2-NEXT: [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
+; AVX2-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; AVX2-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; AVX2-NEXT: [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
+; AVX2-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; AVX2-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; AVX2-NEXT: [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
+; AVX2-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; AVX2-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; AVX2-NEXT: [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
+; AVX2-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; AVX2-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; AVX2-NEXT: [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
+; AVX2-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; AVX2-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; AVX2-NEXT: [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
+; AVX2-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; AVX2-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; AVX2-NEXT: [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
+; AVX2-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; AVX2-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; AVX2-NEXT: [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
+; AVX2-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
+; AVX2-NEXT: [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX2-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; AVX2-NEXT: ret i32 [[TMP48]]
;
; SKX-LABEL: @maxi16(
; SKX-NEXT: [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP48:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[BIN_RDX]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP49:%.*]] = icmp sgt <16 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x i32> [[BIN_RDX]], <16 x i32> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x i32> [[BIN_RDX2]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP50:%.*]] = icmp sgt <16 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x i32> [[BIN_RDX2]], <16 x i32> [[RDX_SHUF3]]
-; SKX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x i32> [[BIN_RDX4]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP51:%.*]] = icmp sgt <16 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; SKX-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x i32> [[BIN_RDX4]], <16 x i32> [[RDX_SHUF5]]
-; SKX-NEXT: [[TMP52:%.*]] = extractelement <16 x i32> [[BIN_RDX6]], i32 0
-; SKX: ret i32 [[TMP52]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; SKX-NEXT: [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
+; SKX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; SKX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; SKX-NEXT: [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
+; SKX-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; SKX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; SKX-NEXT: [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
+; SKX-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; SKX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; SKX-NEXT: [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
+; SKX-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; SKX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; SKX-NEXT: [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
+; SKX-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; SKX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; SKX-NEXT: [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
+; SKX-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; SKX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; SKX-NEXT: [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
+; SKX-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; SKX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; SKX-NEXT: [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
+; SKX-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
+; SKX-NEXT: [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; SKX-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; SKX-NEXT: ret i32 [[TMP48]]
;
%2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
%3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
@@ -253,83 +458,459 @@ define i32 @maxi16(i32) {
define i32 @maxi32(i32) {
; CHECK-LABEL: @maxi32(
; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; CHECK: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT: [[TMP96:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
-; CHECK-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
-; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[BIN_RDX]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT: [[TMP97:%.*]] = icmp sgt <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; CHECK-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x i32> [[BIN_RDX]], <32 x i32> [[RDX_SHUF1]]
-; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x i32> [[BIN_RDX2]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT: [[TMP98:%.*]] = icmp sgt <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; CHECK-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x i32> [[BIN_RDX2]], <32 x i32> [[RDX_SHUF3]]
-; CHECK-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x i32> [[BIN_RDX4]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT: [[TMP99:%.*]] = icmp sgt <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; CHECK-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x i32> [[BIN_RDX4]], <32 x i32> [[RDX_SHUF5]]
-; CHECK-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[BIN_RDX6]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT: [[TMP100:%.*]] = icmp sgt <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
-; CHECK-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x i32> [[BIN_RDX6]], <32 x i32> [[RDX_SHUF7]]
-; CHECK-NEXT: [[TMP101:%.*]] = extractelement <32 x i32> [[BIN_RDX8]], i32 0
-; CHECK: ret i32 [[TMP101]]
+; CHECK-NEXT: [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
+; CHECK-NEXT: [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
+; CHECK-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; CHECK-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; CHECK-NEXT: [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
+; CHECK-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; CHECK-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; CHECK-NEXT: [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
+; CHECK-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; CHECK-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
+; CHECK-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; CHECK-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; CHECK-NEXT: [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
+; CHECK-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; CHECK-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; CHECK-NEXT: [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
+; CHECK-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; CHECK-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; CHECK-NEXT: [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
+; CHECK-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; CHECK-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; CHECK-NEXT: [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
+; CHECK-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; CHECK-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; CHECK-NEXT: [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
+; CHECK-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; CHECK-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; CHECK-NEXT: [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
+; CHECK-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; CHECK-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; CHECK-NEXT: [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
+; CHECK-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; CHECK-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; CHECK-NEXT: [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
+; CHECK-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; CHECK-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; CHECK-NEXT: [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
+; CHECK-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; CHECK-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; CHECK-NEXT: [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
+; CHECK-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; CHECK-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; CHECK-NEXT: [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
+; CHECK-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; CHECK-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; CHECK-NEXT: [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
+; CHECK-NEXT: [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
+; CHECK-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
+; CHECK-NEXT: [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
+; CHECK-NEXT: [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
+; CHECK-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
+; CHECK-NEXT: [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
+; CHECK-NEXT: [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
+; CHECK-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
+; CHECK-NEXT: [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
+; CHECK-NEXT: [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
+; CHECK-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
+; CHECK-NEXT: [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
+; CHECK-NEXT: [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
+; CHECK-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
+; CHECK-NEXT: [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
+; CHECK-NEXT: [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
+; CHECK-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
+; CHECK-NEXT: [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
+; CHECK-NEXT: [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
+; CHECK-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
+; CHECK-NEXT: [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
+; CHECK-NEXT: [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
+; CHECK-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
+; CHECK-NEXT: [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
+; CHECK-NEXT: [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
+; CHECK-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
+; CHECK-NEXT: [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
+; CHECK-NEXT: [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
+; CHECK-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
+; CHECK-NEXT: [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
+; CHECK-NEXT: [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
+; CHECK-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
+; CHECK-NEXT: [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
+; CHECK-NEXT: [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
+; CHECK-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
+; CHECK-NEXT: [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
+; CHECK-NEXT: [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
+; CHECK-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
+; CHECK-NEXT: [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
+; CHECK-NEXT: [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
+; CHECK-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
+; CHECK-NEXT: [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
+; CHECK-NEXT: [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
+; CHECK-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
+; CHECK-NEXT: [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
+; CHECK-NEXT: [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
+; CHECK-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
+; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; CHECK-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
+; CHECK-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; CHECK-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
+; CHECK-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; CHECK-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
+; CHECK-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT: [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; CHECK-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
+; CHECK-NEXT: [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; CHECK-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
+; CHECK-NEXT: ret i32 [[TMP96]]
;
; AVX-LABEL: @maxi32(
; AVX-NEXT: [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP96:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[BIN_RDX]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP97:%.*]] = icmp sgt <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x i32> [[BIN_RDX]], <32 x i32> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x i32> [[BIN_RDX2]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP98:%.*]] = icmp sgt <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x i32> [[BIN_RDX2]], <32 x i32> [[RDX_SHUF3]]
-; AVX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x i32> [[BIN_RDX4]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP99:%.*]] = icmp sgt <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x i32> [[BIN_RDX4]], <32 x i32> [[RDX_SHUF5]]
-; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[BIN_RDX6]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP100:%.*]] = icmp sgt <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
-; AVX-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x i32> [[BIN_RDX6]], <32 x i32> [[RDX_SHUF7]]
-; AVX-NEXT: [[TMP101:%.*]] = extractelement <32 x i32> [[BIN_RDX8]], i32 0
-; AVX: ret i32 [[TMP101]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX-NEXT: [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
+; AVX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; AVX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; AVX-NEXT: [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
+; AVX-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; AVX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; AVX-NEXT: [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
+; AVX-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; AVX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; AVX-NEXT: [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
+; AVX-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; AVX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; AVX-NEXT: [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
+; AVX-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; AVX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; AVX-NEXT: [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
+; AVX-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; AVX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; AVX-NEXT: [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
+; AVX-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; AVX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; AVX-NEXT: [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
+; AVX-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; AVX-NEXT: [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
+; AVX-NEXT: [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
+; AVX-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
+; AVX-NEXT: [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
+; AVX-NEXT: [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
+; AVX-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
+; AVX-NEXT: [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
+; AVX-NEXT: [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
+; AVX-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
+; AVX-NEXT: [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
+; AVX-NEXT: [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
+; AVX-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
+; AVX-NEXT: [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
+; AVX-NEXT: [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
+; AVX-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
+; AVX-NEXT: [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
+; AVX-NEXT: [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
+; AVX-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
+; AVX-NEXT: [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
+; AVX-NEXT: [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
+; AVX-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
+; AVX-NEXT: [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
+; AVX-NEXT: [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
+; AVX-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
+; AVX-NEXT: [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
+; AVX-NEXT: [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
+; AVX-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
+; AVX-NEXT: [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
+; AVX-NEXT: [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
+; AVX-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
+; AVX-NEXT: [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
+; AVX-NEXT: [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
+; AVX-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
+; AVX-NEXT: [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
+; AVX-NEXT: [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
+; AVX-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
+; AVX-NEXT: [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
+; AVX-NEXT: [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
+; AVX-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
+; AVX-NEXT: [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
+; AVX-NEXT: [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
+; AVX-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
+; AVX-NEXT: [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
+; AVX-NEXT: [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
+; AVX-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
+; AVX-NEXT: [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
+; AVX-NEXT: [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
+; AVX-NEXT: [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
+; AVX-NEXT: ret i32 [[TMP96]]
;
; AVX2-LABEL: @maxi32(
; AVX2-NEXT: [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP96:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[BIN_RDX]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP97:%.*]] = icmp sgt <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x i32> [[BIN_RDX]], <32 x i32> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x i32> [[BIN_RDX2]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP98:%.*]] = icmp sgt <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x i32> [[BIN_RDX2]], <32 x i32> [[RDX_SHUF3]]
-; AVX2-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x i32> [[BIN_RDX4]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP99:%.*]] = icmp sgt <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX2-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x i32> [[BIN_RDX4]], <32 x i32> [[RDX_SHUF5]]
-; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[BIN_RDX6]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP100:%.*]] = icmp sgt <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
-; AVX2-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x i32> [[BIN_RDX6]], <32 x i32> [[RDX_SHUF7]]
-; AVX2-NEXT: [[TMP101:%.*]] = extractelement <32 x i32> [[BIN_RDX8]], i32 0
-; AVX2: ret i32 [[TMP101]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; AVX2-NEXT: [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
+; AVX2-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; AVX2-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; AVX2-NEXT: [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
+; AVX2-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; AVX2-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; AVX2-NEXT: [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
+; AVX2-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; AVX2-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; AVX2-NEXT: [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
+; AVX2-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; AVX2-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; AVX2-NEXT: [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
+; AVX2-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; AVX2-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; AVX2-NEXT: [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
+; AVX2-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; AVX2-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; AVX2-NEXT: [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
+; AVX2-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; AVX2-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; AVX2-NEXT: [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
+; AVX2-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX2-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; AVX2-NEXT: [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
+; AVX2-NEXT: [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
+; AVX2-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
+; AVX2-NEXT: [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
+; AVX2-NEXT: [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
+; AVX2-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
+; AVX2-NEXT: [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
+; AVX2-NEXT: [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
+; AVX2-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
+; AVX2-NEXT: [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
+; AVX2-NEXT: [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
+; AVX2-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
+; AVX2-NEXT: [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
+; AVX2-NEXT: [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
+; AVX2-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
+; AVX2-NEXT: [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
+; AVX2-NEXT: [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
+; AVX2-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
+; AVX2-NEXT: [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
+; AVX2-NEXT: [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
+; AVX2-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
+; AVX2-NEXT: [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
+; AVX2-NEXT: [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
+; AVX2-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
+; AVX2-NEXT: [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
+; AVX2-NEXT: [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
+; AVX2-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
+; AVX2-NEXT: [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
+; AVX2-NEXT: [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
+; AVX2-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
+; AVX2-NEXT: [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
+; AVX2-NEXT: [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
+; AVX2-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
+; AVX2-NEXT: [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
+; AVX2-NEXT: [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
+; AVX2-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
+; AVX2-NEXT: [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
+; AVX2-NEXT: [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
+; AVX2-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
+; AVX2-NEXT: [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
+; AVX2-NEXT: [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
+; AVX2-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
+; AVX2-NEXT: [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
+; AVX2-NEXT: [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
+; AVX2-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
+; AVX2-NEXT: [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
+; AVX2-NEXT: [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
+; AVX2-NEXT: [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX2-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
+; AVX2-NEXT: ret i32 [[TMP96]]
;
; SKX-LABEL: @maxi32(
; SKX-NEXT: [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP96:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[BIN_RDX]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP97:%.*]] = icmp sgt <32 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x i32> [[BIN_RDX]], <32 x i32> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x i32> [[BIN_RDX2]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP98:%.*]] = icmp sgt <32 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x i32> [[BIN_RDX2]], <32 x i32> [[RDX_SHUF3]]
-; SKX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x i32> [[BIN_RDX4]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP99:%.*]] = icmp sgt <32 x i32> [[BIN_RDX4]], [[RDX_SHUF5]]
-; SKX-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x i32> [[BIN_RDX4]], <32 x i32> [[RDX_SHUF5]]
-; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[BIN_RDX6]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP100:%.*]] = icmp sgt <32 x i32> [[BIN_RDX6]], [[RDX_SHUF7]]
-; SKX-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x i32> [[BIN_RDX6]], <32 x i32> [[RDX_SHUF7]]
-; SKX-NEXT: [[TMP101:%.*]] = extractelement <32 x i32> [[BIN_RDX8]], i32 0
-; SKX: ret i32 [[TMP101]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
+; SKX-NEXT: [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
+; SKX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
+; SKX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
+; SKX-NEXT: [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
+; SKX-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
+; SKX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
+; SKX-NEXT: [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
+; SKX-NEXT: [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
+; SKX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
+; SKX-NEXT: [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
+; SKX-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
+; SKX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
+; SKX-NEXT: [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
+; SKX-NEXT: [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
+; SKX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
+; SKX-NEXT: [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
+; SKX-NEXT: [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
+; SKX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
+; SKX-NEXT: [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
+; SKX-NEXT: [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
+; SKX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
+; SKX-NEXT: [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
+; SKX-NEXT: [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; SKX-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
+; SKX-NEXT: [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
+; SKX-NEXT: [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
+; SKX-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
+; SKX-NEXT: [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
+; SKX-NEXT: [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
+; SKX-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
+; SKX-NEXT: [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
+; SKX-NEXT: [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
+; SKX-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
+; SKX-NEXT: [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
+; SKX-NEXT: [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
+; SKX-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
+; SKX-NEXT: [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
+; SKX-NEXT: [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
+; SKX-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
+; SKX-NEXT: [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
+; SKX-NEXT: [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
+; SKX-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
+; SKX-NEXT: [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
+; SKX-NEXT: [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
+; SKX-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
+; SKX-NEXT: [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
+; SKX-NEXT: [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
+; SKX-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
+; SKX-NEXT: [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
+; SKX-NEXT: [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
+; SKX-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
+; SKX-NEXT: [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
+; SKX-NEXT: [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
+; SKX-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
+; SKX-NEXT: [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
+; SKX-NEXT: [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
+; SKX-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
+; SKX-NEXT: [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
+; SKX-NEXT: [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
+; SKX-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
+; SKX-NEXT: [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
+; SKX-NEXT: [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
+; SKX-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
+; SKX-NEXT: [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
+; SKX-NEXT: [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
+; SKX-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
+; SKX-NEXT: [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
+; SKX-NEXT: [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
+; SKX-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
+; SKX-NEXT: [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
+; SKX-NEXT: [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
+; SKX-NEXT: [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; SKX-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
+; SKX-NEXT: ret i32 [[TMP96]]
;
%2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
%3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
@@ -456,45 +1037,111 @@ define float @maxf8(float) {
;
; AVX-LABEL: @maxf8(
; AVX-NEXT: [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP24:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[BIN_RDX]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP25:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x float> [[BIN_RDX]], <8 x float> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x float> [[BIN_RDX2]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP26:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x float> [[BIN_RDX2]], <8 x float> [[RDX_SHUF3]]
-; AVX-NEXT: [[TMP27:%.*]] = extractelement <8 x float> [[BIN_RDX4]], i32 0
-; AVX: ret float [[TMP27]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
+; AVX-NEXT: [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX-NEXT: ret float [[TMP24]]
;
; AVX2-LABEL: @maxf8(
; AVX2-NEXT: [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP24:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[BIN_RDX]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP25:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x float> [[BIN_RDX]], <8 x float> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x float> [[BIN_RDX2]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP26:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x float> [[BIN_RDX2]], <8 x float> [[RDX_SHUF3]]
-; AVX2-NEXT: [[TMP27:%.*]] = extractelement <8 x float> [[BIN_RDX4]], i32 0
-; AVX2: ret float [[TMP27]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
+; AVX2-NEXT: [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX2-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX2-NEXT: ret float [[TMP24]]
;
; SKX-LABEL: @maxf8(
; SKX-NEXT: [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP24:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <8 x i1> [[TMP24]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[BIN_RDX]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP25:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <8 x i1> [[TMP25]], <8 x float> [[BIN_RDX]], <8 x float> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x float> [[BIN_RDX2]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP26:%.*]] = fcmp fast ogt <8 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <8 x i1> [[TMP26]], <8 x float> [[BIN_RDX2]], <8 x float> [[RDX_SHUF3]]
-; SKX-NEXT: [[TMP27:%.*]] = extractelement <8 x float> [[BIN_RDX4]], i32 0
-; SKX: ret float [[TMP27]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
+; SKX-NEXT: [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; SKX-NEXT: [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; SKX-NEXT: ret float [[TMP24]]
;
%2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
%3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
@@ -573,54 +1220,192 @@ define float @maxf16(float) {
;
; AVX-LABEL: @maxf16(
; AVX-NEXT: [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP48:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[BIN_RDX]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP49:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x float> [[BIN_RDX]], <16 x float> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x float> [[BIN_RDX2]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP50:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x float> [[BIN_RDX2]], <16 x float> [[RDX_SHUF3]]
-; AVX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x float> [[BIN_RDX4]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP51:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x float> [[BIN_RDX4]], <16 x float> [[RDX_SHUF5]]
-; AVX-NEXT: [[TMP52:%.*]] = extractelement <16 x float> [[BIN_RDX6]], i32 0
-; AVX: ret float [[TMP52]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX-NEXT: [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
+; AVX-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; AVX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; AVX-NEXT: [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
+; AVX-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; AVX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; AVX-NEXT: [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
+; AVX-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; AVX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; AVX-NEXT: [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
+; AVX-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; AVX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; AVX-NEXT: [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
+; AVX-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; AVX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; AVX-NEXT: [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
+; AVX-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; AVX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; AVX-NEXT: [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
+; AVX-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; AVX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; AVX-NEXT: [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
+; AVX-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
+; AVX-NEXT: [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; AVX-NEXT: ret float [[TMP48]]
;
; AVX2-LABEL: @maxf16(
; AVX2-NEXT: [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP48:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[BIN_RDX]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP49:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x float> [[BIN_RDX]], <16 x float> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x float> [[BIN_RDX2]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP50:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x float> [[BIN_RDX2]], <16 x float> [[RDX_SHUF3]]
-; AVX2-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x float> [[BIN_RDX4]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP51:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX2-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x float> [[BIN_RDX4]], <16 x float> [[RDX_SHUF5]]
-; AVX2-NEXT: [[TMP52:%.*]] = extractelement <16 x float> [[BIN_RDX6]], i32 0
-; AVX2: ret float [[TMP52]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX2-NEXT: [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
+; AVX2-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; AVX2-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; AVX2-NEXT: [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
+; AVX2-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; AVX2-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; AVX2-NEXT: [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
+; AVX2-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; AVX2-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; AVX2-NEXT: [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
+; AVX2-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; AVX2-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; AVX2-NEXT: [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
+; AVX2-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; AVX2-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; AVX2-NEXT: [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
+; AVX2-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; AVX2-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; AVX2-NEXT: [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
+; AVX2-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; AVX2-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; AVX2-NEXT: [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
+; AVX2-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
+; AVX2-NEXT: [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX2-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; AVX2-NEXT: ret float [[TMP48]]
;
; SKX-LABEL: @maxf16(
; SKX-NEXT: [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP48:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <16 x i1> [[TMP48]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[BIN_RDX]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP49:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <16 x i1> [[TMP49]], <16 x float> [[BIN_RDX]], <16 x float> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <16 x float> [[BIN_RDX2]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP50:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <16 x i1> [[TMP50]], <16 x float> [[BIN_RDX2]], <16 x float> [[RDX_SHUF3]]
-; SKX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <16 x float> [[BIN_RDX4]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP51:%.*]] = fcmp fast ogt <16 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; SKX-NEXT: [[BIN_RDX6:%.*]] = select <16 x i1> [[TMP51]], <16 x float> [[BIN_RDX4]], <16 x float> [[RDX_SHUF5]]
-; SKX-NEXT: [[TMP52:%.*]] = extractelement <16 x float> [[BIN_RDX6]], i32 0
-; SKX: ret float [[TMP52]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; SKX-NEXT: [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
+; SKX-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; SKX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; SKX-NEXT: [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
+; SKX-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; SKX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; SKX-NEXT: [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
+; SKX-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; SKX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; SKX-NEXT: [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
+; SKX-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; SKX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; SKX-NEXT: [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
+; SKX-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; SKX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; SKX-NEXT: [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
+; SKX-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; SKX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; SKX-NEXT: [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
+; SKX-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; SKX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; SKX-NEXT: [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
+; SKX-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
+; SKX-NEXT: [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; SKX-NEXT: [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; SKX-NEXT: ret float [[TMP48]]
;
%2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
%3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
@@ -771,63 +1556,345 @@ define float @maxf32(float) {
;
; AVX-LABEL: @maxf32(
; AVX-NEXT: [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; AVX: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP96:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
-; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[BIN_RDX]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP97:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x float> [[BIN_RDX]], <32 x float> [[RDX_SHUF1]]
-; AVX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x float> [[BIN_RDX2]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP98:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x float> [[BIN_RDX2]], <32 x float> [[RDX_SHUF3]]
-; AVX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x float> [[BIN_RDX4]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP99:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x float> [[BIN_RDX4]], <32 x float> [[RDX_SHUF5]]
-; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[BIN_RDX6]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT: [[TMP100:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX6]], [[RDX_SHUF7]]
-; AVX-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x float> [[BIN_RDX6]], <32 x float> [[RDX_SHUF7]]
-; AVX-NEXT: [[TMP101:%.*]] = extractelement <32 x float> [[BIN_RDX8]], i32 0
-; AVX: ret float [[TMP101]]
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
+; AVX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
+; AVX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX-NEXT: [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
+; AVX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX-NEXT: [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
+; AVX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX-NEXT: [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
+; AVX-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; AVX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; AVX-NEXT: [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
+; AVX-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; AVX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; AVX-NEXT: [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
+; AVX-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; AVX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; AVX-NEXT: [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
+; AVX-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; AVX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; AVX-NEXT: [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
+; AVX-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; AVX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; AVX-NEXT: [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
+; AVX-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; AVX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; AVX-NEXT: [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
+; AVX-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; AVX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; AVX-NEXT: [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
+; AVX-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; AVX-NEXT: [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
+; AVX-NEXT: [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
+; AVX-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
+; AVX-NEXT: [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
+; AVX-NEXT: [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
+; AVX-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
+; AVX-NEXT: [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
+; AVX-NEXT: [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
+; AVX-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
+; AVX-NEXT: [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
+; AVX-NEXT: [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
+; AVX-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
+; AVX-NEXT: [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
+; AVX-NEXT: [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
+; AVX-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
+; AVX-NEXT: [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
+; AVX-NEXT: [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
+; AVX-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
+; AVX-NEXT: [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
+; AVX-NEXT: [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
+; AVX-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
+; AVX-NEXT: [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
+; AVX-NEXT: [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
+; AVX-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
+; AVX-NEXT: [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
+; AVX-NEXT: [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
+; AVX-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
+; AVX-NEXT: [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
+; AVX-NEXT: [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
+; AVX-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
+; AVX-NEXT: [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
+; AVX-NEXT: [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
+; AVX-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
+; AVX-NEXT: [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
+; AVX-NEXT: [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
+; AVX-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
+; AVX-NEXT: [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
+; AVX-NEXT: [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
+; AVX-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
+; AVX-NEXT: [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
+; AVX-NEXT: [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
+; AVX-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
+; AVX-NEXT: [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
+; AVX-NEXT: [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
+; AVX-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
+; AVX-NEXT: [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
+; AVX-NEXT: [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
+; AVX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
+; AVX-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
+; AVX-NEXT: [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
+; AVX-NEXT: ret float [[TMP96]]
;
; AVX2-LABEL: @maxf32(
; AVX2-NEXT: [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; AVX2: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP96:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
-; AVX2-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
-; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[BIN_RDX]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP97:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; AVX2-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x float> [[BIN_RDX]], <32 x float> [[RDX_SHUF1]]
-; AVX2-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x float> [[BIN_RDX2]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP98:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; AVX2-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x float> [[BIN_RDX2]], <32 x float> [[RDX_SHUF3]]
-; AVX2-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x float> [[BIN_RDX4]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP99:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; AVX2-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x float> [[BIN_RDX4]], <32 x float> [[RDX_SHUF5]]
-; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[BIN_RDX6]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX2-NEXT: [[TMP100:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX6]], [[RDX_SHUF7]]
-; AVX2-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x float> [[BIN_RDX6]], <32 x float> [[RDX_SHUF7]]
-; AVX2-NEXT: [[TMP101:%.*]] = extractelement <32 x float> [[BIN_RDX8]], i32 0
-; AVX2: ret float [[TMP101]]
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
+; AVX2-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
+; AVX2-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; AVX2-NEXT: [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
+; AVX2-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; AVX2-NEXT: [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
+; AVX2-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; AVX2-NEXT: [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
+; AVX2-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; AVX2-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; AVX2-NEXT: [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
+; AVX2-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; AVX2-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; AVX2-NEXT: [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
+; AVX2-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; AVX2-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; AVX2-NEXT: [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
+; AVX2-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; AVX2-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; AVX2-NEXT: [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
+; AVX2-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; AVX2-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; AVX2-NEXT: [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
+; AVX2-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; AVX2-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; AVX2-NEXT: [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
+; AVX2-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; AVX2-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; AVX2-NEXT: [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
+; AVX2-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX2-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; AVX2-NEXT: [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
+; AVX2-NEXT: [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
+; AVX2-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
+; AVX2-NEXT: [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
+; AVX2-NEXT: [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
+; AVX2-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
+; AVX2-NEXT: [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
+; AVX2-NEXT: [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
+; AVX2-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
+; AVX2-NEXT: [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
+; AVX2-NEXT: [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
+; AVX2-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
+; AVX2-NEXT: [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
+; AVX2-NEXT: [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
+; AVX2-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
+; AVX2-NEXT: [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
+; AVX2-NEXT: [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
+; AVX2-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
+; AVX2-NEXT: [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
+; AVX2-NEXT: [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
+; AVX2-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
+; AVX2-NEXT: [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
+; AVX2-NEXT: [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
+; AVX2-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
+; AVX2-NEXT: [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
+; AVX2-NEXT: [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
+; AVX2-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
+; AVX2-NEXT: [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
+; AVX2-NEXT: [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
+; AVX2-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
+; AVX2-NEXT: [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
+; AVX2-NEXT: [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
+; AVX2-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
+; AVX2-NEXT: [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
+; AVX2-NEXT: [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
+; AVX2-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
+; AVX2-NEXT: [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
+; AVX2-NEXT: [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
+; AVX2-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
+; AVX2-NEXT: [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
+; AVX2-NEXT: [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
+; AVX2-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
+; AVX2-NEXT: [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
+; AVX2-NEXT: [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
+; AVX2-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
+; AVX2-NEXT: [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
+; AVX2-NEXT: [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
+; AVX2-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
+; AVX2-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
+; AVX2-NEXT: [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX2-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
+; AVX2-NEXT: ret float [[TMP96]]
;
; SKX-LABEL: @maxf32(
; SKX-NEXT: [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; SKX: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP96:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
-; SKX-NEXT: [[BIN_RDX:%.*]] = select <32 x i1> [[TMP96]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
-; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[BIN_RDX]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP97:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX]], [[RDX_SHUF1]]
-; SKX-NEXT: [[BIN_RDX2:%.*]] = select <32 x i1> [[TMP97]], <32 x float> [[BIN_RDX]], <32 x float> [[RDX_SHUF1]]
-; SKX-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x float> [[BIN_RDX2]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP98:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX2]], [[RDX_SHUF3]]
-; SKX-NEXT: [[BIN_RDX4:%.*]] = select <32 x i1> [[TMP98]], <32 x float> [[BIN_RDX2]], <32 x float> [[RDX_SHUF3]]
-; SKX-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x float> [[BIN_RDX4]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP99:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX4]], [[RDX_SHUF5]]
-; SKX-NEXT: [[BIN_RDX6:%.*]] = select <32 x i1> [[TMP99]], <32 x float> [[BIN_RDX4]], <32 x float> [[RDX_SHUF5]]
-; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[BIN_RDX6]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SKX-NEXT: [[TMP100:%.*]] = fcmp fast ogt <32 x float> [[BIN_RDX6]], [[RDX_SHUF7]]
-; SKX-NEXT: [[BIN_RDX8:%.*]] = select <32 x i1> [[TMP100]], <32 x float> [[BIN_RDX6]], <32 x float> [[RDX_SHUF7]]
-; SKX-NEXT: [[TMP101:%.*]] = extractelement <32 x float> [[BIN_RDX8]], i32 0
-; SKX: ret float [[TMP101]]
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
+; SKX-NEXT: [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
+; SKX-NEXT: [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
+; SKX-NEXT: [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
+; SKX-NEXT: [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
+; SKX-NEXT: [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
+; SKX-NEXT: [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
+; SKX-NEXT: [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
+; SKX-NEXT: [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; SKX-NEXT: [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
+; SKX-NEXT: [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
+; SKX-NEXT: [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
+; SKX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
+; SKX-NEXT: [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
+; SKX-NEXT: [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
+; SKX-NEXT: [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
+; SKX-NEXT: [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
+; SKX-NEXT: [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
+; SKX-NEXT: [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
+; SKX-NEXT: [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
+; SKX-NEXT: [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
+; SKX-NEXT: [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
+; SKX-NEXT: [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
+; SKX-NEXT: [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
+; SKX-NEXT: [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
+; SKX-NEXT: [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
+; SKX-NEXT: [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
+; SKX-NEXT: [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
+; SKX-NEXT: [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
+; SKX-NEXT: [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
+; SKX-NEXT: [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
+; SKX-NEXT: [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
+; SKX-NEXT: [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; SKX-NEXT: [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
+; SKX-NEXT: [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
+; SKX-NEXT: [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
+; SKX-NEXT: [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
+; SKX-NEXT: [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
+; SKX-NEXT: [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
+; SKX-NEXT: [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
+; SKX-NEXT: [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
+; SKX-NEXT: [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
+; SKX-NEXT: [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
+; SKX-NEXT: [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
+; SKX-NEXT: [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
+; SKX-NEXT: [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
+; SKX-NEXT: [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
+; SKX-NEXT: [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
+; SKX-NEXT: [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
+; SKX-NEXT: [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
+; SKX-NEXT: [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
+; SKX-NEXT: [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
+; SKX-NEXT: [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
+; SKX-NEXT: [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
+; SKX-NEXT: [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
+; SKX-NEXT: [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
+; SKX-NEXT: [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
+; SKX-NEXT: [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
+; SKX-NEXT: [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
+; SKX-NEXT: [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
+; SKX-NEXT: [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
+; SKX-NEXT: [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
+; SKX-NEXT: [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
+; SKX-NEXT: [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
+; SKX-NEXT: [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
+; SKX-NEXT: [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
+; SKX-NEXT: [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
+; SKX-NEXT: [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
+; SKX-NEXT: [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
+; SKX-NEXT: [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
+; SKX-NEXT: [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
+; SKX-NEXT: [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
+; SKX-NEXT: [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
+; SKX-NEXT: [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
+; SKX-NEXT: [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
+; SKX-NEXT: [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
+; SKX-NEXT: [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
+; SKX-NEXT: [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
+; SKX-NEXT: [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
+; SKX-NEXT: [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
+; SKX-NEXT: [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
+; SKX-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
+; SKX-NEXT: [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
+; SKX-NEXT: [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; SKX-NEXT: [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
+; SKX-NEXT: ret float [[TMP96]]
;
%2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
%3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
@@ -926,3 +1993,368 @@ define float @maxf32(float) {
ret float %95
}
+define i32 @maxi8_mutiple_uses(i32) {
+; CHECK-LABEL: @maxi8_mutiple_uses(
+; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
+; CHECK-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
+; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
+; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
+; CHECK-NEXT: [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
+; CHECK-NEXT: [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
+; CHECK-NEXT: [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; CHECK-NEXT: [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
+; CHECK-NEXT: [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
+; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; CHECK-NEXT: [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
+; CHECK-NEXT: [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
+; CHECK-NEXT: [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; CHECK-NEXT: [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
+; CHECK-NEXT: [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
+; CHECK-NEXT: [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; CHECK-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
+; CHECK-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
+; CHECK-NEXT: [[TMP24:%.*]] = select i1 [[TMP4]], i32 3, i32 4
+; CHECK-NEXT: store i32 [[TMP24]], i32* @var, align 8
+; CHECK-NEXT: ret i32 [[TMP23]]
+;
+; AVX-LABEL: @maxi8_mutiple_uses(
+; AVX-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
+; AVX-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
+; AVX-NEXT: [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
+; AVX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
+; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+; AVX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX-NEXT: [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; AVX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
+; AVX-NEXT: [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; AVX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
+; AVX-NEXT: [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; AVX-NEXT: [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
+; AVX-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
+; AVX-NEXT: [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
+; AVX-NEXT: [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
+; AVX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
+; AVX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
+; AVX-NEXT: [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
+; AVX-NEXT: [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX-NEXT: [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
+; AVX-NEXT: [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
+; AVX-NEXT: [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; AVX-NEXT: store i32 [[TMP32]], i32* @var, align 8
+; AVX-NEXT: ret i32 [[TMP31]]
+;
+; AVX2-LABEL: @maxi8_mutiple_uses(
+; AVX2-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
+; AVX2-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
+; AVX2-NEXT: [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
+; AVX2-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; AVX2-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
+; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+; AVX2-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX2-NEXT: [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; AVX2-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
+; AVX2-NEXT: [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; AVX2-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
+; AVX2-NEXT: [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX2-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; AVX2-NEXT: [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX2-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
+; AVX2-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
+; AVX2-NEXT: [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
+; AVX2-NEXT: [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
+; AVX2-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
+; AVX2-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
+; AVX2-NEXT: [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
+; AVX2-NEXT: [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX2-NEXT: [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
+; AVX2-NEXT: [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
+; AVX2-NEXT: [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; AVX2-NEXT: store i32 [[TMP32]], i32* @var, align 8
+; AVX2-NEXT: ret i32 [[TMP31]]
+;
+; SKX-LABEL: @maxi8_mutiple_uses(
+; SKX-NEXT: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; SKX-NEXT: [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
+; SKX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
+; SKX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+; SKX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; SKX-NEXT: [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; SKX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; SKX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
+; SKX-NEXT: [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; SKX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; SKX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
+; SKX-NEXT: [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; SKX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; SKX-NEXT: [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; SKX-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
+; SKX-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
+; SKX-NEXT: [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
+; SKX-NEXT: [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
+; SKX-NEXT: [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
+; SKX-NEXT: [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
+; SKX-NEXT: [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
+; SKX-NEXT: [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; SKX-NEXT: [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
+; SKX-NEXT: [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
+; SKX-NEXT: [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; SKX-NEXT: store i32 [[TMP32]], i32* @var, align 8
+; SKX-NEXT: ret i32 [[TMP31]]
+;
+ %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+ %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+ %4 = icmp sgt i32 %2, %3
+ %5 = select i1 %4, i32 %2, i32 %3
+ %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
+ %7 = icmp sgt i32 %5, %6
+ %8 = select i1 %7, i32 %5, i32 %6
+ %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
+ %10 = icmp sgt i32 %8, %9
+ %11 = select i1 %10, i32 %8, i32 %9
+ %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+ %13 = icmp sgt i32 %11, %12
+ %14 = select i1 %13, i32 %11, i32 %12
+ %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+ %16 = icmp sgt i32 %14, %15
+ %17 = select i1 %16, i32 %14, i32 %15
+ %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+ %19 = icmp sgt i32 %17, %18
+ %20 = select i1 %19, i32 %17, i32 %18
+ %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+ %22 = icmp sgt i32 %20, %21
+ %23 = select i1 %22, i32 %20, i32 %21
+ %24 = select i1 %4, i32 3, i32 4
+ store i32 %24, i32* @var, align 8
+ ret i32 %23
+}
+
+define i32 @maxi8_wrong_parent(i32) {
+; CHECK-LABEL: @maxi8_wrong_parent(
+; CHECK-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
+; CHECK-NEXT: br label [[PP:%.*]]
+; CHECK: pp:
+; CHECK-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
+; CHECK-NEXT: [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
+; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
+; CHECK-NEXT: [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
+; CHECK-NEXT: [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
+; CHECK-NEXT: [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; CHECK-NEXT: [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
+; CHECK-NEXT: [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
+; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; CHECK-NEXT: [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
+; CHECK-NEXT: [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
+; CHECK-NEXT: [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; CHECK-NEXT: [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
+; CHECK-NEXT: [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
+; CHECK-NEXT: [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; CHECK-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
+; CHECK-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
+; CHECK-NEXT: ret i32 [[TMP23]]
+;
+; AVX-LABEL: @maxi8_wrong_parent(
+; AVX-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+; AVX-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+; AVX-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
+; AVX-NEXT: br label [[PP:%.*]]
+; AVX: pp:
+; AVX-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
+; AVX-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
+; AVX-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
+; AVX-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP5]], [[TMP7]]
+; AVX-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP5]], i32 undef
+; AVX-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
+; AVX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
+; AVX-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX-NEXT: [[TMP16:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
+; AVX-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX-NEXT: [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
+; AVX-NEXT: [[TMP22:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
+; AVX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; AVX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; AVX-NEXT: [[TMP24:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX-NEXT: [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP19]]
+; AVX-NEXT: [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP19]]
+; AVX-NEXT: [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], [[TMP22]]
+; AVX-NEXT: [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 [[TMP22]]
+; AVX-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], [[TMP5]]
+; AVX-NEXT: [[OP_EXTRA:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 [[TMP5]]
+; AVX-NEXT: [[TMP30:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 [[TMP22]]
+; AVX-NEXT: ret i32 [[OP_EXTRA]]
+;
+; AVX2-LABEL: @maxi8_wrong_parent(
+; AVX2-NEXT: [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+; AVX2-NEXT: [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+; AVX2-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
+; AVX2-NEXT: br label [[PP:%.*]]
+; AVX2: pp:
+; AVX2-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
+; 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
+; AVX2-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
+; AVX2-NEXT: [[TMP8:%.*]] = icmp sgt i32 [[TMP5]], [[TMP7]]
+; AVX2-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP5]], i32 undef
+; AVX2-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
+; AVX2-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
+; AVX2-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
+; AVX2-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
+; AVX2-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
+; AVX2-NEXT: [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT: [[TMP16:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
+; AVX2-NEXT: [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
+; AVX2-NEXT: [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
+; AVX2-NEXT: [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX2-NEXT: [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT: [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
+; AVX2-NEXT: [[TMP22:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX2-NEXT: [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
+; AVX2-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; AVX2-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; AVX2-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; AVX2-NEXT: [[TMP24:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX2-NEXT: [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP19]]
+; AVX2-NEXT: [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP19]]
+; AVX2-NEXT: [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], [[TMP22]]
+; AVX2-NEXT: [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 [[TMP22]]
+; AVX2-NEXT: [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], [[TMP5]]
+; AVX2-NEXT: [[OP_EXTRA:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 [[TMP5]]
+; AVX2-NEXT: [[TMP30:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 [[TMP22]]
+; AVX2-NEXT: ret i32 [[OP_EXTRA]]
+;
+; SKX-LABEL: @maxi8_wrong_parent(
+; SKX-NEXT: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
+; SKX-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
+; SKX-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
+; SKX-NEXT: [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
+; SKX-NEXT: br label [[PP:%.*]]
+; SKX: pp:
+; SKX-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
+; SKX-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
+; SKX-NEXT: [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; SKX-NEXT: [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; SKX-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
+; SKX-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
+; SKX-NEXT: [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; SKX-NEXT: [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
+; SKX-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; SKX-NEXT: [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], [[TMP8]]
+; SKX-NEXT: [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 [[TMP8]]
+; SKX-NEXT: [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], [[TMP9]]
+; SKX-NEXT: [[TMP14:%.*]] = insertelement <2 x i1> undef, i1 [[TMP13]], i32 0
+; SKX-NEXT: [[TMP15:%.*]] = insertelement <2 x i1> [[TMP14]], i1 [[TMP5]], i32 1
+; SKX-NEXT: [[TMP16:%.*]] = insertelement <2 x i32> undef, i32 [[TMP12]], i32 0
+; SKX-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> [[TMP16]], i32 [[TMP3]], i32 1
+; SKX-NEXT: [[TMP18:%.*]] = insertelement <2 x i32> undef, i32 [[TMP9]], i32 0
+; SKX-NEXT: [[TMP19:%.*]] = insertelement <2 x i32> [[TMP18]], i32 [[TMP4]], i32 1
+; SKX-NEXT: [[TMP20:%.*]] = select <2 x i1> [[TMP15]], <2 x i32> [[TMP17]], <2 x i32> [[TMP19]]
+; SKX-NEXT: [[TMP21:%.*]] = extractelement <2 x i32> [[TMP20]], i32 1
+; SKX-NEXT: [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP7]]
+; SKX-NEXT: [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 undef
+; SKX-NEXT: [[TMP24:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
+; SKX-NEXT: [[TMP25:%.*]] = icmp sgt i32 [[TMP23]], [[TMP24]]
+; SKX-NEXT: [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP23]], i32 undef
+; SKX-NEXT: [[TMP27:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
+; SKX-NEXT: [[TMP28:%.*]] = icmp sgt i32 [[TMP26]], [[TMP27]]
+; SKX-NEXT: [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP26]], i32 undef
+; SKX-NEXT: [[TMP30:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
+; SKX-NEXT: [[TMP31:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
+; SKX-NEXT: [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP29]], i32 undef
+; SKX-NEXT: [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], [[TMP8]]
+; SKX-NEXT: [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 [[TMP8]]
+; SKX-NEXT: [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], [[TMP9]]
+; SKX-NEXT: [[TMP36:%.*]] = extractelement <2 x i32> [[TMP20]], i32 0
+; SKX-NEXT: [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], [[TMP21]]
+; SKX-NEXT: [[OP_EXTRA:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 [[TMP21]]
+; SKX-NEXT: [[TMP38:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 [[TMP9]]
+; SKX-NEXT: ret i32 [[OP_EXTRA]]
+;
+ %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
+ %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
+ %4 = icmp sgt i32 %2, %3
+ br label %pp
+
+pp:
+ %5 = select i1 %4, i32 %2, i32 %3
+ %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
+ %7 = icmp sgt i32 %5, %6
+ %8 = select i1 %7, i32 %5, i32 %6
+ %9 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
+ %10 = icmp sgt i32 %8, %9
+ %11 = select i1 %10, i32 %8, i32 %9
+ %12 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+ %13 = icmp sgt i32 %11, %12
+ %14 = select i1 %13, i32 %11, i32 %12
+ %15 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+ %16 = icmp sgt i32 %14, %15
+ %17 = select i1 %16, i32 %14, i32 %15
+ %18 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+ %19 = icmp sgt i32 %17, %18
+ %20 = select i1 %19, i32 %17, i32 %18
+ %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+ %22 = icmp sgt i32 %20, %21
+ %23 = select i1 %22, i32 %20, i32 %21
+ ret i32 %23
+}
+
More information about the llvm-commits
mailing list