xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/SystemZ/reductions-fmin-fmax.ll (revision 29441e4f5fa5f5c7709f7cf180815ba97f611297)

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z15 -passes=slp-vectorizer %s -S -o - \
; RUN:   | FileCheck %s

; Test vectorization and reassociation of fmin/fmax operations. Vectorization
; is more profitable if the loads are also vectorizable.

define double @fmin_double_4_nums_seq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define double @fmin_double_4_nums_seq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x double>, ptr [[X]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = call fast double @llvm.vector.reduce.fmin.v4f64(<4 x double> [[TMP1]])
; CHECK-NEXT:    ret double [[TMP2]]
;
  %g1 = getelementptr inbounds double, ptr %x, i64 1
  %g2 = getelementptr inbounds double, ptr %x, i64 2
  %g3 = getelementptr inbounds double, ptr %x, i64 3
  %t0 = load double, ptr %x, align 4
  %t1 = load double, ptr %g1, align 4
  %t2 = load double, ptr %g2, align 4
  %t3 = load double, ptr %g3, align 4
  %m1 = tail call fast double @llvm.minnum.f64(double %t1, double %t0)
  %m2 = tail call fast double @llvm.minnum.f64(double %t2, double %m1)
  %m3 = tail call fast double @llvm.minnum.f64(double %t3, double %m2)
  ret double %m3
}

define double @fmin_double_16_nums_nonseq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define double @fmin_double_16_nums_nonseq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    [[G1:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
; CHECK-NEXT:    [[G2:%.*]] = getelementptr inbounds double, ptr [[X]], i64 4
; CHECK-NEXT:    [[G3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 6
; CHECK-NEXT:    [[G4:%.*]] = getelementptr inbounds double, ptr [[X]], i64 8
; CHECK-NEXT:    [[G5:%.*]] = getelementptr inbounds double, ptr [[X]], i64 10
; CHECK-NEXT:    [[G6:%.*]] = getelementptr inbounds double, ptr [[X]], i64 12
; CHECK-NEXT:    [[G7:%.*]] = getelementptr inbounds double, ptr [[X]], i64 14
; CHECK-NEXT:    [[G8:%.*]] = getelementptr inbounds double, ptr [[X]], i64 16
; CHECK-NEXT:    [[G9:%.*]] = getelementptr inbounds double, ptr [[X]], i64 18
; CHECK-NEXT:    [[G10:%.*]] = getelementptr inbounds double, ptr [[X]], i64 20
; CHECK-NEXT:    [[G11:%.*]] = getelementptr inbounds double, ptr [[X]], i64 22
; CHECK-NEXT:    [[G12:%.*]] = getelementptr inbounds double, ptr [[X]], i64 24
; CHECK-NEXT:    [[G13:%.*]] = getelementptr inbounds double, ptr [[X]], i64 26
; CHECK-NEXT:    [[G14:%.*]] = getelementptr inbounds double, ptr [[X]], i64 28
; CHECK-NEXT:    [[G15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 30
; CHECK-NEXT:    [[T0:%.*]] = load double, ptr [[X]], align 4
; CHECK-NEXT:    [[T1:%.*]] = load double, ptr [[G1]], align 4
; CHECK-NEXT:    [[T2:%.*]] = load double, ptr [[G2]], align 4
; CHECK-NEXT:    [[T3:%.*]] = load double, ptr [[G3]], align 4
; CHECK-NEXT:    [[T4:%.*]] = load double, ptr [[G4]], align 4
; CHECK-NEXT:    [[T5:%.*]] = load double, ptr [[G5]], align 4
; CHECK-NEXT:    [[T6:%.*]] = load double, ptr [[G6]], align 4
; CHECK-NEXT:    [[T7:%.*]] = load double, ptr [[G7]], align 4
; CHECK-NEXT:    [[T8:%.*]] = load double, ptr [[G8]], align 4
; CHECK-NEXT:    [[T9:%.*]] = load double, ptr [[G9]], align 4
; CHECK-NEXT:    [[T10:%.*]] = load double, ptr [[G10]], align 4
; CHECK-NEXT:    [[T11:%.*]] = load double, ptr [[G11]], align 4
; CHECK-NEXT:    [[T12:%.*]] = load double, ptr [[G12]], align 4
; CHECK-NEXT:    [[T13:%.*]] = load double, ptr [[G13]], align 4
; CHECK-NEXT:    [[T14:%.*]] = load double, ptr [[G14]], align 4
; CHECK-NEXT:    [[T15:%.*]] = load double, ptr [[G15]], align 4
; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <16 x double> poison, double [[T1]], i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <16 x double> [[TMP1]], double [[T0]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <16 x double> [[TMP2]], double [[T2]], i32 2
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <16 x double> [[TMP3]], double [[T3]], i32 3
; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <16 x double> [[TMP4]], double [[T4]], i32 4
; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <16 x double> [[TMP5]], double [[T5]], i32 5
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <16 x double> [[TMP6]], double [[T6]], i32 6
; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <16 x double> [[TMP7]], double [[T7]], i32 7
; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <16 x double> [[TMP8]], double [[T8]], i32 8
; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <16 x double> [[TMP9]], double [[T9]], i32 9
; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <16 x double> [[TMP10]], double [[T10]], i32 10
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <16 x double> [[TMP11]], double [[T11]], i32 11
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <16 x double> [[TMP12]], double [[T12]], i32 12
; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <16 x double> [[TMP13]], double [[T13]], i32 13
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <16 x double> [[TMP14]], double [[T14]], i32 14
; CHECK-NEXT:    [[TMP16:%.*]] = insertelement <16 x double> [[TMP15]], double [[T15]], i32 15
; CHECK-NEXT:    [[TMP17:%.*]] = call fast double @llvm.vector.reduce.fmin.v16f64(<16 x double> [[TMP16]])
; CHECK-NEXT:    ret double [[TMP17]]
;
  %g1 = getelementptr inbounds double, ptr %x, i64 2
  %g2 = getelementptr inbounds double, ptr %x, i64 4
  %g3 = getelementptr inbounds double, ptr %x, i64 6
  %g4 = getelementptr inbounds double, ptr %x, i64 8
  %g5 = getelementptr inbounds double, ptr %x, i64 10
  %g6 = getelementptr inbounds double, ptr %x, i64 12
  %g7 = getelementptr inbounds double, ptr %x, i64 14
  %g8 = getelementptr inbounds double, ptr %x, i64 16
  %g9 = getelementptr inbounds double, ptr %x, i64 18
  %g10 = getelementptr inbounds double, ptr %x, i64 20
  %g11 = getelementptr inbounds double, ptr %x, i64 22
  %g12 = getelementptr inbounds double, ptr %x, i64 24
  %g13 = getelementptr inbounds double, ptr %x, i64 26
  %g14 = getelementptr inbounds double, ptr %x, i64 28
  %g15 = getelementptr inbounds double, ptr %x, i64 30
  %t0 = load double, ptr %x, align 4
  %t1 = load double, ptr %g1, align 4
  %t2 = load double, ptr %g2, align 4
  %t3 = load double, ptr %g3, align 4
  %t4 = load double, ptr %g4, align 4
  %t5 = load double, ptr %g5, align 4
  %t6 = load double, ptr %g6, align 4
  %t7 = load double, ptr %g7, align 4
  %t8 = load double, ptr %g8, align 4
  %t9 = load double, ptr %g9, align 4
  %t10 = load double, ptr %g10, align 4
  %t11 = load double, ptr %g11, align 4
  %t12 = load double, ptr %g12, align 4
  %t13 = load double, ptr %g13, align 4
  %t14 = load double, ptr %g14, align 4
  %t15 = load double, ptr %g15, align 4
  %m1 = tail call fast double @llvm.minnum.f64(double %t1, double %t0)
  %m2 = tail call fast double @llvm.minnum.f64(double %t2, double %m1)
  %m3 = tail call fast double @llvm.minnum.f64(double %t3, double %m2)
  %m4 = tail call fast double @llvm.minnum.f64(double %t4, double %m3)
  %m5 = tail call fast double @llvm.minnum.f64(double %t5, double %m4)
  %m6 = tail call fast double @llvm.minnum.f64(double %t6, double %m5)
  %m7 = tail call fast double @llvm.minnum.f64(double %t7, double %m6)
  %m8 = tail call fast double @llvm.minnum.f64(double %t8, double %m7)
  %m9 = tail call fast double @llvm.minnum.f64(double %t9, double %m8)
  %m10 = tail call fast double @llvm.minnum.f64(double %t10, double %m9)
  %m11 = tail call fast double @llvm.minnum.f64(double %t11, double %m10)
  %m12 = tail call fast double @llvm.minnum.f64(double %t12, double %m11)
  %m13 = tail call fast double @llvm.minnum.f64(double %t13, double %m12)
  %m14 = tail call fast double @llvm.minnum.f64(double %t14, double %m13)
  %m15 = tail call fast double @llvm.minnum.f64(double %t15, double %m14)
  ret double %m15
}

define float @fmin_float_12_nums_nonseq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define float @fmin_float_12_nums_nonseq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    [[G1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
; CHECK-NEXT:    [[G2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 4
; CHECK-NEXT:    [[G3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 6
; CHECK-NEXT:    [[G4:%.*]] = getelementptr inbounds float, ptr [[X]], i64 8
; CHECK-NEXT:    [[G5:%.*]] = getelementptr inbounds float, ptr [[X]], i64 10
; CHECK-NEXT:    [[G6:%.*]] = getelementptr inbounds float, ptr [[X]], i64 12
; CHECK-NEXT:    [[G7:%.*]] = getelementptr inbounds float, ptr [[X]], i64 14
; CHECK-NEXT:    [[G8:%.*]] = getelementptr inbounds float, ptr [[X]], i64 16
; CHECK-NEXT:    [[G9:%.*]] = getelementptr inbounds float, ptr [[X]], i64 18
; CHECK-NEXT:    [[G10:%.*]] = getelementptr inbounds float, ptr [[X]], i64 20
; CHECK-NEXT:    [[G11:%.*]] = getelementptr inbounds float, ptr [[X]], i64 22
; CHECK-NEXT:    [[T0:%.*]] = load float, ptr [[X]], align 4
; CHECK-NEXT:    [[T1:%.*]] = load float, ptr [[G1]], align 4
; CHECK-NEXT:    [[T2:%.*]] = load float, ptr [[G2]], align 4
; CHECK-NEXT:    [[T3:%.*]] = load float, ptr [[G3]], align 4
; CHECK-NEXT:    [[T4:%.*]] = load float, ptr [[G4]], align 4
; CHECK-NEXT:    [[T5:%.*]] = load float, ptr [[G5]], align 4
; CHECK-NEXT:    [[T6:%.*]] = load float, ptr [[G6]], align 4
; CHECK-NEXT:    [[T7:%.*]] = load float, ptr [[G7]], align 4
; CHECK-NEXT:    [[T8:%.*]] = load float, ptr [[G8]], align 4
; CHECK-NEXT:    [[T9:%.*]] = load float, ptr [[G9]], align 4
; CHECK-NEXT:    [[T10:%.*]] = load float, ptr [[G10]], align 4
; CHECK-NEXT:    [[T11:%.*]] = load float, ptr [[G11]], align 4
; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <12 x float> poison, float [[T1]], i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <12 x float> [[TMP1]], float [[T0]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <12 x float> [[TMP2]], float [[T2]], i32 2
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <12 x float> [[TMP3]], float [[T3]], i32 3
; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <12 x float> [[TMP4]], float [[T4]], i32 4
; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <12 x float> [[TMP5]], float [[T5]], i32 5
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <12 x float> [[TMP6]], float [[T6]], i32 6
; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <12 x float> [[TMP7]], float [[T7]], i32 7
; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <12 x float> [[TMP8]], float [[T8]], i32 8
; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <12 x float> [[TMP9]], float [[T9]], i32 9
; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <12 x float> [[TMP10]], float [[T10]], i32 10
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <12 x float> [[TMP11]], float [[T11]], i32 11
; CHECK-NEXT:    [[TMP13:%.*]] = call fast float @llvm.vector.reduce.fmin.v12f32(<12 x float> [[TMP12]])
; CHECK-NEXT:    ret float [[TMP13]]
;
  %g1 = getelementptr inbounds float, ptr %x, i64 2
  %g2 = getelementptr inbounds float, ptr %x, i64 4
  %g3 = getelementptr inbounds float, ptr %x, i64 6
  %g4 = getelementptr inbounds float, ptr %x, i64 8
  %g5 = getelementptr inbounds float, ptr %x, i64 10
  %g6 = getelementptr inbounds float, ptr %x, i64 12
  %g7 = getelementptr inbounds float, ptr %x, i64 14
  %g8 = getelementptr inbounds float, ptr %x, i64 16
  %g9 = getelementptr inbounds float, ptr %x, i64 18
  %g10 = getelementptr inbounds float, ptr %x, i64 20
  %g11 = getelementptr inbounds float, ptr %x, i64 22
  %t0 = load float, ptr %x, align 4
  %t1 = load float, ptr %g1, align 4
  %t2 = load float, ptr %g2, align 4
  %t3 = load float, ptr %g3, align 4
  %t4 = load float, ptr %g4, align 4
  %t5 = load float, ptr %g5, align 4
  %t6 = load float, ptr %g6, align 4
  %t7 = load float, ptr %g7, align 4
  %t8 = load float, ptr %g8, align 4
  %t9 = load float, ptr %g9, align 4
  %t10 = load float, ptr %g10, align 4
  %t11 = load float, ptr %g11, align 4
  %m1 = tail call fast float @llvm.minnum.f32(float %t1, float %t0)
  %m2 = tail call fast float @llvm.minnum.f32(float %t2, float %m1)
  %m3 = tail call fast float @llvm.minnum.f32(float %t3, float %m2)
  %m4 = tail call fast float @llvm.minnum.f32(float %t4, float %m3)
  %m5 = tail call fast float @llvm.minnum.f32(float %t5, float %m4)
  %m6 = tail call fast float @llvm.minnum.f32(float %t6, float %m5)
  %m7 = tail call fast float @llvm.minnum.f32(float %t7, float %m6)
  %m8 = tail call fast float @llvm.minnum.f32(float %t8, float %m7)
  %m9 = tail call fast float @llvm.minnum.f32(float %t9, float %m8)
  %m10 = tail call fast float @llvm.minnum.f32(float %t10, float %m9)
  %m11 = tail call fast float @llvm.minnum.f32(float %t11, float %m10)
  ret float %m11
}

define double @fmax_double_4_nums_seq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define double @fmax_double_4_nums_seq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x double>, ptr [[X]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = call fast double @llvm.vector.reduce.fmax.v4f64(<4 x double> [[TMP1]])
; CHECK-NEXT:    ret double [[TMP2]]
;
  %g1 = getelementptr inbounds double, ptr %x, i64 1
  %g2 = getelementptr inbounds double, ptr %x, i64 2
  %g3 = getelementptr inbounds double, ptr %x, i64 3
  %t0 = load double, ptr %x, align 4
  %t1 = load double, ptr %g1, align 4
  %t2 = load double, ptr %g2, align 4
  %t3 = load double, ptr %g3, align 4
  %m1 = tail call fast double @llvm.maxnum.f64(double %t1, double %t0)
  %m2 = tail call fast double @llvm.maxnum.f64(double %t2, double %m1)
  %m3 = tail call fast double @llvm.maxnum.f64(double %t3, double %m2)
  ret double %m3
}

define double @fmax_double_16_nums_nonseq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define double @fmax_double_16_nums_nonseq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    [[G1:%.*]] = getelementptr inbounds double, ptr [[X]], i64 2
; CHECK-NEXT:    [[G2:%.*]] = getelementptr inbounds double, ptr [[X]], i64 4
; CHECK-NEXT:    [[G3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 6
; CHECK-NEXT:    [[G4:%.*]] = getelementptr inbounds double, ptr [[X]], i64 8
; CHECK-NEXT:    [[G5:%.*]] = getelementptr inbounds double, ptr [[X]], i64 10
; CHECK-NEXT:    [[G6:%.*]] = getelementptr inbounds double, ptr [[X]], i64 12
; CHECK-NEXT:    [[G7:%.*]] = getelementptr inbounds double, ptr [[X]], i64 14
; CHECK-NEXT:    [[G8:%.*]] = getelementptr inbounds double, ptr [[X]], i64 16
; CHECK-NEXT:    [[G9:%.*]] = getelementptr inbounds double, ptr [[X]], i64 18
; CHECK-NEXT:    [[G10:%.*]] = getelementptr inbounds double, ptr [[X]], i64 20
; CHECK-NEXT:    [[G11:%.*]] = getelementptr inbounds double, ptr [[X]], i64 22
; CHECK-NEXT:    [[G12:%.*]] = getelementptr inbounds double, ptr [[X]], i64 24
; CHECK-NEXT:    [[G13:%.*]] = getelementptr inbounds double, ptr [[X]], i64 26
; CHECK-NEXT:    [[G14:%.*]] = getelementptr inbounds double, ptr [[X]], i64 28
; CHECK-NEXT:    [[G15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 30
; CHECK-NEXT:    [[T0:%.*]] = load double, ptr [[X]], align 4
; CHECK-NEXT:    [[T1:%.*]] = load double, ptr [[G1]], align 4
; CHECK-NEXT:    [[T2:%.*]] = load double, ptr [[G2]], align 4
; CHECK-NEXT:    [[T3:%.*]] = load double, ptr [[G3]], align 4
; CHECK-NEXT:    [[T4:%.*]] = load double, ptr [[G4]], align 4
; CHECK-NEXT:    [[T5:%.*]] = load double, ptr [[G5]], align 4
; CHECK-NEXT:    [[T6:%.*]] = load double, ptr [[G6]], align 4
; CHECK-NEXT:    [[T7:%.*]] = load double, ptr [[G7]], align 4
; CHECK-NEXT:    [[T8:%.*]] = load double, ptr [[G8]], align 4
; CHECK-NEXT:    [[T9:%.*]] = load double, ptr [[G9]], align 4
; CHECK-NEXT:    [[T10:%.*]] = load double, ptr [[G10]], align 4
; CHECK-NEXT:    [[T11:%.*]] = load double, ptr [[G11]], align 4
; CHECK-NEXT:    [[T12:%.*]] = load double, ptr [[G12]], align 4
; CHECK-NEXT:    [[T13:%.*]] = load double, ptr [[G13]], align 4
; CHECK-NEXT:    [[T14:%.*]] = load double, ptr [[G14]], align 4
; CHECK-NEXT:    [[T15:%.*]] = load double, ptr [[G15]], align 4
; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <16 x double> poison, double [[T1]], i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <16 x double> [[TMP1]], double [[T0]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <16 x double> [[TMP2]], double [[T2]], i32 2
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <16 x double> [[TMP3]], double [[T3]], i32 3
; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <16 x double> [[TMP4]], double [[T4]], i32 4
; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <16 x double> [[TMP5]], double [[T5]], i32 5
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <16 x double> [[TMP6]], double [[T6]], i32 6
; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <16 x double> [[TMP7]], double [[T7]], i32 7
; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <16 x double> [[TMP8]], double [[T8]], i32 8
; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <16 x double> [[TMP9]], double [[T9]], i32 9
; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <16 x double> [[TMP10]], double [[T10]], i32 10
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <16 x double> [[TMP11]], double [[T11]], i32 11
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <16 x double> [[TMP12]], double [[T12]], i32 12
; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <16 x double> [[TMP13]], double [[T13]], i32 13
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <16 x double> [[TMP14]], double [[T14]], i32 14
; CHECK-NEXT:    [[TMP16:%.*]] = insertelement <16 x double> [[TMP15]], double [[T15]], i32 15
; CHECK-NEXT:    [[TMP17:%.*]] = call fast double @llvm.vector.reduce.fmax.v16f64(<16 x double> [[TMP16]])
; CHECK-NEXT:    ret double [[TMP17]]
;
  %g1 = getelementptr inbounds double, ptr %x, i64 2
  %g2 = getelementptr inbounds double, ptr %x, i64 4
  %g3 = getelementptr inbounds double, ptr %x, i64 6
  %g4 = getelementptr inbounds double, ptr %x, i64 8
  %g5 = getelementptr inbounds double, ptr %x, i64 10
  %g6 = getelementptr inbounds double, ptr %x, i64 12
  %g7 = getelementptr inbounds double, ptr %x, i64 14
  %g8 = getelementptr inbounds double, ptr %x, i64 16
  %g9 = getelementptr inbounds double, ptr %x, i64 18
  %g10 = getelementptr inbounds double, ptr %x, i64 20
  %g11 = getelementptr inbounds double, ptr %x, i64 22
  %g12 = getelementptr inbounds double, ptr %x, i64 24
  %g13 = getelementptr inbounds double, ptr %x, i64 26
  %g14 = getelementptr inbounds double, ptr %x, i64 28
  %g15 = getelementptr inbounds double, ptr %x, i64 30
  %t0 = load double, ptr %x, align 4
  %t1 = load double, ptr %g1, align 4
  %t2 = load double, ptr %g2, align 4
  %t3 = load double, ptr %g3, align 4
  %t4 = load double, ptr %g4, align 4
  %t5 = load double, ptr %g5, align 4
  %t6 = load double, ptr %g6, align 4
  %t7 = load double, ptr %g7, align 4
  %t8 = load double, ptr %g8, align 4
  %t9 = load double, ptr %g9, align 4
  %t10 = load double, ptr %g10, align 4
  %t11 = load double, ptr %g11, align 4
  %t12 = load double, ptr %g12, align 4
  %t13 = load double, ptr %g13, align 4
  %t14 = load double, ptr %g14, align 4
  %t15 = load double, ptr %g15, align 4
  %m1 = tail call fast double @llvm.maxnum.f64(double %t1, double %t0)
  %m2 = tail call fast double @llvm.maxnum.f64(double %t2, double %m1)
  %m3 = tail call fast double @llvm.maxnum.f64(double %t3, double %m2)
  %m4 = tail call fast double @llvm.maxnum.f64(double %t4, double %m3)
  %m5 = tail call fast double @llvm.maxnum.f64(double %t5, double %m4)
  %m6 = tail call fast double @llvm.maxnum.f64(double %t6, double %m5)
  %m7 = tail call fast double @llvm.maxnum.f64(double %t7, double %m6)
  %m8 = tail call fast double @llvm.maxnum.f64(double %t8, double %m7)
  %m9 = tail call fast double @llvm.maxnum.f64(double %t9, double %m8)
  %m10 = tail call fast double @llvm.maxnum.f64(double %t10, double %m9)
  %m11 = tail call fast double @llvm.maxnum.f64(double %t11, double %m10)
  %m12 = tail call fast double @llvm.maxnum.f64(double %t12, double %m11)
  %m13 = tail call fast double @llvm.maxnum.f64(double %t13, double %m12)
  %m14 = tail call fast double @llvm.maxnum.f64(double %t14, double %m13)
  %m15 = tail call fast double @llvm.maxnum.f64(double %t15, double %m14)
  ret double %m15
}

define float @fmax_float_12_nums_nonseq(ptr nocapture noundef readonly %x) {
; CHECK-LABEL: define float @fmax_float_12_nums_nonseq(
; CHECK-SAME: ptr noundef readonly captures(none) [[X:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    [[G1:%.*]] = getelementptr inbounds float, ptr [[X]], i64 2
; CHECK-NEXT:    [[G2:%.*]] = getelementptr inbounds float, ptr [[X]], i64 4
; CHECK-NEXT:    [[G3:%.*]] = getelementptr inbounds float, ptr [[X]], i64 6
; CHECK-NEXT:    [[G4:%.*]] = getelementptr inbounds float, ptr [[X]], i64 8
; CHECK-NEXT:    [[G5:%.*]] = getelementptr inbounds float, ptr [[X]], i64 10
; CHECK-NEXT:    [[G6:%.*]] = getelementptr inbounds float, ptr [[X]], i64 12
; CHECK-NEXT:    [[G7:%.*]] = getelementptr inbounds float, ptr [[X]], i64 14
; CHECK-NEXT:    [[G8:%.*]] = getelementptr inbounds float, ptr [[X]], i64 16
; CHECK-NEXT:    [[G9:%.*]] = getelementptr inbounds float, ptr [[X]], i64 18
; CHECK-NEXT:    [[G10:%.*]] = getelementptr inbounds float, ptr [[X]], i64 20
; CHECK-NEXT:    [[G11:%.*]] = getelementptr inbounds float, ptr [[X]], i64 22
; CHECK-NEXT:    [[T0:%.*]] = load float, ptr [[X]], align 4
; CHECK-NEXT:    [[T1:%.*]] = load float, ptr [[G1]], align 4
; CHECK-NEXT:    [[T2:%.*]] = load float, ptr [[G2]], align 4
; CHECK-NEXT:    [[T3:%.*]] = load float, ptr [[G3]], align 4
; CHECK-NEXT:    [[T4:%.*]] = load float, ptr [[G4]], align 4
; CHECK-NEXT:    [[T5:%.*]] = load float, ptr [[G5]], align 4
; CHECK-NEXT:    [[T6:%.*]] = load float, ptr [[G6]], align 4
; CHECK-NEXT:    [[T7:%.*]] = load float, ptr [[G7]], align 4
; CHECK-NEXT:    [[T8:%.*]] = load float, ptr [[G8]], align 4
; CHECK-NEXT:    [[T9:%.*]] = load float, ptr [[G9]], align 4
; CHECK-NEXT:    [[T10:%.*]] = load float, ptr [[G10]], align 4
; CHECK-NEXT:    [[T11:%.*]] = load float, ptr [[G11]], align 4
; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <12 x float> poison, float [[T1]], i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <12 x float> [[TMP1]], float [[T0]], i32 1
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <12 x float> [[TMP2]], float [[T2]], i32 2
; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <12 x float> [[TMP3]], float [[T3]], i32 3
; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <12 x float> [[TMP4]], float [[T4]], i32 4
; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <12 x float> [[TMP5]], float [[T5]], i32 5
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <12 x float> [[TMP6]], float [[T6]], i32 6
; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <12 x float> [[TMP7]], float [[T7]], i32 7
; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <12 x float> [[TMP8]], float [[T8]], i32 8
; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <12 x float> [[TMP9]], float [[T9]], i32 9
; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <12 x float> [[TMP10]], float [[T10]], i32 10
; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <12 x float> [[TMP11]], float [[T11]], i32 11
; CHECK-NEXT:    [[TMP13:%.*]] = call fast float @llvm.vector.reduce.fmax.v12f32(<12 x float> [[TMP12]])
; CHECK-NEXT:    ret float [[TMP13]]
;
  %g1 = getelementptr inbounds float, ptr %x, i64 2
  %g2 = getelementptr inbounds float, ptr %x, i64 4
  %g3 = getelementptr inbounds float, ptr %x, i64 6
  %g4 = getelementptr inbounds float, ptr %x, i64 8
  %g5 = getelementptr inbounds float, ptr %x, i64 10
  %g6 = getelementptr inbounds float, ptr %x, i64 12
  %g7 = getelementptr inbounds float, ptr %x, i64 14
  %g8 = getelementptr inbounds float, ptr %x, i64 16
  %g9 = getelementptr inbounds float, ptr %x, i64 18
  %g10 = getelementptr inbounds float, ptr %x, i64 20
  %g11 = getelementptr inbounds float, ptr %x, i64 22
  %t0 = load float, ptr %x, align 4
  %t1 = load float, ptr %g1, align 4
  %t2 = load float, ptr %g2, align 4
  %t3 = load float, ptr %g3, align 4
  %t4 = load float, ptr %g4, align 4
  %t5 = load float, ptr %g5, align 4
  %t6 = load float, ptr %g6, align 4
  %t7 = load float, ptr %g7, align 4
  %t8 = load float, ptr %g8, align 4
  %t9 = load float, ptr %g9, align 4
  %t10 = load float, ptr %g10, align 4
  %t11 = load float, ptr %g11, align 4
  %m1 = tail call fast float @llvm.maxnum.f32(float %t1, float %t0)
  %m2 = tail call fast float @llvm.maxnum.f32(float %t2, float %m1)
  %m3 = tail call fast float @llvm.maxnum.f32(float %t3, float %m2)
  %m4 = tail call fast float @llvm.maxnum.f32(float %t4, float %m3)
  %m5 = tail call fast float @llvm.maxnum.f32(float %t5, float %m4)
  %m6 = tail call fast float @llvm.maxnum.f32(float %t6, float %m5)
  %m7 = tail call fast float @llvm.maxnum.f32(float %t7, float %m6)
  %m8 = tail call fast float @llvm.maxnum.f32(float %t8, float %m7)
  %m9 = tail call fast float @llvm.maxnum.f32(float %t9, float %m8)
  %m10 = tail call fast float @llvm.maxnum.f32(float %t10, float %m9)
  %m11 = tail call fast float @llvm.maxnum.f32(float %t11, float %m10)
  ret float %m11
}

declare float @llvm.minnum.f32(float, float)
declare double @llvm.minnum.f64(double, double)
declare float @llvm.maxnum.f32(float, float)
declare double @llvm.maxnum.f64(double, double)