; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=SSE2
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=SSE41
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=-prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=AVX --check-prefix=AVX512
; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=+prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefix=AVX --check-prefix=AVX2

target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"

@src64 = common global [8 x double] zeroinitializer, align 64
@dst64 = common global [8 x double] zeroinitializer, align 64
@src32 = common global [16 x float] zeroinitializer, align 64
@dst32 = common global [16 x float] zeroinitializer, align 64

declare double @llvm.ceil.f64(double %p)
declare double @llvm.floor.f64(double %p)
declare double @llvm.nearbyint.f64(double %p)
declare double @llvm.rint.f64(double %p)
declare double @llvm.trunc.f64(double %p)

declare float @llvm.ceil.f32(float %p)
declare float @llvm.floor.f32(float %p)
declare float @llvm.nearbyint.f32(float %p)
declare float @llvm.rint.f32(float %p)
declare float @llvm.trunc.f32(float %p)

define void @ceil_2f64() #0 {
; SSE2-LABEL: @ceil_2f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[CEIL0:%.*]] = call double @llvm.ceil.f64(double [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call double @llvm.ceil.f64(double [[LD1]])
; SSE2-NEXT:    store double [[CEIL0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[CEIL1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_2f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @ceil_2f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP1]])
; AVX-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ceil0 = call double @llvm.ceil.f64(double %ld0)
  %ceil1 = call double @llvm.ceil.f64(double %ld1)
  store double %ceil0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %ceil1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  ret void
}

define void @ceil_4f64() #0 {
; SSE2-LABEL: @ceil_4f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[CEIL0:%.*]] = call double @llvm.ceil.f64(double [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call double @llvm.ceil.f64(double [[LD1]])
; SSE2-NEXT:    [[CEIL2:%.*]] = call double @llvm.ceil.f64(double [[LD2]])
; SSE2-NEXT:    [[CEIL3:%.*]] = call double @llvm.ceil.f64(double [[LD3]])
; SSE2-NEXT:    store double [[CEIL0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[CEIL1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[CEIL2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[CEIL3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_4f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @ceil_4f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[TMP1]])
; AVX-NEXT:    store <4 x double> [[TMP2]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ceil0 = call double @llvm.ceil.f64(double %ld0)
  %ceil1 = call double @llvm.ceil.f64(double %ld1)
  %ceil2 = call double @llvm.ceil.f64(double %ld2)
  %ceil3 = call double @llvm.ceil.f64(double %ld3)
  store double %ceil0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %ceil1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %ceil2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %ceil3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  ret void
}

define void @ceil_8f64() #0 {
; SSE2-LABEL: @ceil_8f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[LD4:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
; SSE2-NEXT:    [[LD5:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
; SSE2-NEXT:    [[LD6:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
; SSE2-NEXT:    [[LD7:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
; SSE2-NEXT:    [[CEIL0:%.*]] = call double @llvm.ceil.f64(double [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call double @llvm.ceil.f64(double [[LD1]])
; SSE2-NEXT:    [[CEIL2:%.*]] = call double @llvm.ceil.f64(double [[LD2]])
; SSE2-NEXT:    [[CEIL3:%.*]] = call double @llvm.ceil.f64(double [[LD3]])
; SSE2-NEXT:    [[CEIL4:%.*]] = call double @llvm.ceil.f64(double [[LD4]])
; SSE2-NEXT:    [[CEIL5:%.*]] = call double @llvm.ceil.f64(double [[LD5]])
; SSE2-NEXT:    [[CEIL6:%.*]] = call double @llvm.ceil.f64(double [[LD6]])
; SSE2-NEXT:    [[CEIL7:%.*]] = call double @llvm.ceil.f64(double [[LD7]])
; SSE2-NEXT:    store double [[CEIL0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[CEIL1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[CEIL2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[CEIL3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    store double [[CEIL4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
; SSE2-NEXT:    store double [[CEIL5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
; SSE2-NEXT:    store double [[CEIL6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
; SSE2-NEXT:    store double [[CEIL7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_8f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[TMP4]])
; SSE41-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @ceil_8f64(
; AVX1-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[TMP2]])
; AVX1-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX1-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @ceil_8f64(
; AVX2-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[TMP2]])
; AVX2-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX2-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @ceil_8f64(
; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @src64 to <8 x double>*), align 8
; AVX512-NEXT:    [[TMP2:%.*]] = call <8 x double> @llvm.ceil.v8f64(<8 x double> [[TMP1]])
; AVX512-NEXT:    store <8 x double> [[TMP2]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 8
; AVX512-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ld4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
  %ld5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
  %ld6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
  %ld7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
  %ceil0 = call double @llvm.ceil.f64(double %ld0)
  %ceil1 = call double @llvm.ceil.f64(double %ld1)
  %ceil2 = call double @llvm.ceil.f64(double %ld2)
  %ceil3 = call double @llvm.ceil.f64(double %ld3)
  %ceil4 = call double @llvm.ceil.f64(double %ld4)
  %ceil5 = call double @llvm.ceil.f64(double %ld5)
  %ceil6 = call double @llvm.ceil.f64(double %ld6)
  %ceil7 = call double @llvm.ceil.f64(double %ld7)
  store double %ceil0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %ceil1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %ceil2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %ceil3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  store double %ceil4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
  store double %ceil5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
  store double %ceil6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
  store double %ceil7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
  ret void
}

define void @floor_2f64() #0 {
; SSE2-LABEL: @floor_2f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[FLOOR0:%.*]] = call double @llvm.floor.f64(double [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call double @llvm.floor.f64(double [[LD1]])
; SSE2-NEXT:    store double [[FLOOR0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[FLOOR1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_2f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @floor_2f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP1]])
; AVX-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %floor0 = call double @llvm.floor.f64(double %ld0)
  %floor1 = call double @llvm.floor.f64(double %ld1)
  store double %floor0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %floor1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  ret void
}

define void @floor_4f64() #0 {
; SSE2-LABEL: @floor_4f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[FLOOR0:%.*]] = call double @llvm.floor.f64(double [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call double @llvm.floor.f64(double [[LD1]])
; SSE2-NEXT:    [[FLOOR2:%.*]] = call double @llvm.floor.f64(double [[LD2]])
; SSE2-NEXT:    [[FLOOR3:%.*]] = call double @llvm.floor.f64(double [[LD3]])
; SSE2-NEXT:    store double [[FLOOR0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[FLOOR1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[FLOOR2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[FLOOR3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_4f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @floor_4f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[TMP1]])
; AVX-NEXT:    store <4 x double> [[TMP2]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %floor0 = call double @llvm.floor.f64(double %ld0)
  %floor1 = call double @llvm.floor.f64(double %ld1)
  %floor2 = call double @llvm.floor.f64(double %ld2)
  %floor3 = call double @llvm.floor.f64(double %ld3)
  store double %floor0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %floor1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %floor2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %floor3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  ret void
}

define void @floor_8f64() #0 {
; SSE2-LABEL: @floor_8f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[LD4:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
; SSE2-NEXT:    [[LD5:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
; SSE2-NEXT:    [[LD6:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
; SSE2-NEXT:    [[LD7:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
; SSE2-NEXT:    [[FLOOR0:%.*]] = call double @llvm.floor.f64(double [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call double @llvm.floor.f64(double [[LD1]])
; SSE2-NEXT:    [[FLOOR2:%.*]] = call double @llvm.floor.f64(double [[LD2]])
; SSE2-NEXT:    [[FLOOR3:%.*]] = call double @llvm.floor.f64(double [[LD3]])
; SSE2-NEXT:    [[FLOOR4:%.*]] = call double @llvm.floor.f64(double [[LD4]])
; SSE2-NEXT:    [[FLOOR5:%.*]] = call double @llvm.floor.f64(double [[LD5]])
; SSE2-NEXT:    [[FLOOR6:%.*]] = call double @llvm.floor.f64(double [[LD6]])
; SSE2-NEXT:    [[FLOOR7:%.*]] = call double @llvm.floor.f64(double [[LD7]])
; SSE2-NEXT:    store double [[FLOOR0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[FLOOR1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[FLOOR2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[FLOOR3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    store double [[FLOOR4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
; SSE2-NEXT:    store double [[FLOOR5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
; SSE2-NEXT:    store double [[FLOOR6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
; SSE2-NEXT:    store double [[FLOOR7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_8f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[TMP4]])
; SSE41-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @floor_8f64(
; AVX1-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[TMP2]])
; AVX1-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX1-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @floor_8f64(
; AVX2-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[TMP2]])
; AVX2-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX2-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @floor_8f64(
; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @src64 to <8 x double>*), align 8
; AVX512-NEXT:    [[TMP2:%.*]] = call <8 x double> @llvm.floor.v8f64(<8 x double> [[TMP1]])
; AVX512-NEXT:    store <8 x double> [[TMP2]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 8
; AVX512-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ld4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
  %ld5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
  %ld6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
  %ld7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
  %floor0 = call double @llvm.floor.f64(double %ld0)
  %floor1 = call double @llvm.floor.f64(double %ld1)
  %floor2 = call double @llvm.floor.f64(double %ld2)
  %floor3 = call double @llvm.floor.f64(double %ld3)
  %floor4 = call double @llvm.floor.f64(double %ld4)
  %floor5 = call double @llvm.floor.f64(double %ld5)
  %floor6 = call double @llvm.floor.f64(double %ld6)
  %floor7 = call double @llvm.floor.f64(double %ld7)
  store double %floor0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %floor1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %floor2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %floor3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  store double %floor4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
  store double %floor5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
  store double %floor6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
  store double %floor7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
  ret void
}

define void @nearbyint_2f64() #0 {
; SSE2-LABEL: @nearbyint_2f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call double @llvm.nearbyint.f64(double [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call double @llvm.nearbyint.f64(double [[LD1]])
; SSE2-NEXT:    store double [[NEARBYINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[NEARBYINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_2f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @nearbyint_2f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP1]])
; AVX-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %nearbyint0 = call double @llvm.nearbyint.f64(double %ld0)
  %nearbyint1 = call double @llvm.nearbyint.f64(double %ld1)
  store double %nearbyint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %nearbyint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  ret void
}

define void @nearbyint_4f64() #0 {
; SSE2-LABEL: @nearbyint_4f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call double @llvm.nearbyint.f64(double [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call double @llvm.nearbyint.f64(double [[LD1]])
; SSE2-NEXT:    [[NEARBYINT2:%.*]] = call double @llvm.nearbyint.f64(double [[LD2]])
; SSE2-NEXT:    [[NEARBYINT3:%.*]] = call double @llvm.nearbyint.f64(double [[LD3]])
; SSE2-NEXT:    store double [[NEARBYINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[NEARBYINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[NEARBYINT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[NEARBYINT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_4f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @nearbyint_4f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> [[TMP1]])
; AVX-NEXT:    store <4 x double> [[TMP2]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %nearbyint0 = call double @llvm.nearbyint.f64(double %ld0)
  %nearbyint1 = call double @llvm.nearbyint.f64(double %ld1)
  %nearbyint2 = call double @llvm.nearbyint.f64(double %ld2)
  %nearbyint3 = call double @llvm.nearbyint.f64(double %ld3)
  store double %nearbyint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %nearbyint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %nearbyint2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %nearbyint3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  ret void
}

define void @nearbyint_8f64() #0 {
; SSE2-LABEL: @nearbyint_8f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[LD4:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
; SSE2-NEXT:    [[LD5:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
; SSE2-NEXT:    [[LD6:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
; SSE2-NEXT:    [[LD7:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call double @llvm.nearbyint.f64(double [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call double @llvm.nearbyint.f64(double [[LD1]])
; SSE2-NEXT:    [[NEARBYINT2:%.*]] = call double @llvm.nearbyint.f64(double [[LD2]])
; SSE2-NEXT:    [[NEARBYINT3:%.*]] = call double @llvm.nearbyint.f64(double [[LD3]])
; SSE2-NEXT:    [[NEARBYINT4:%.*]] = call double @llvm.nearbyint.f64(double [[LD4]])
; SSE2-NEXT:    [[NEARBYINT5:%.*]] = call double @llvm.nearbyint.f64(double [[LD5]])
; SSE2-NEXT:    [[NEARBYINT6:%.*]] = call double @llvm.nearbyint.f64(double [[LD6]])
; SSE2-NEXT:    [[NEARBYINT7:%.*]] = call double @llvm.nearbyint.f64(double [[LD7]])
; SSE2-NEXT:    store double [[NEARBYINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[NEARBYINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[NEARBYINT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[NEARBYINT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    store double [[NEARBYINT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
; SSE2-NEXT:    store double [[NEARBYINT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
; SSE2-NEXT:    store double [[NEARBYINT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
; SSE2-NEXT:    store double [[NEARBYINT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_8f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <2 x double> @llvm.nearbyint.v2f64(<2 x double> [[TMP4]])
; SSE41-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @nearbyint_8f64(
; AVX1-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> [[TMP2]])
; AVX1-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX1-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @nearbyint_8f64(
; AVX2-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.nearbyint.v4f64(<4 x double> [[TMP2]])
; AVX2-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX2-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @nearbyint_8f64(
; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @src64 to <8 x double>*), align 8
; AVX512-NEXT:    [[TMP2:%.*]] = call <8 x double> @llvm.nearbyint.v8f64(<8 x double> [[TMP1]])
; AVX512-NEXT:    store <8 x double> [[TMP2]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 8
; AVX512-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ld4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
  %ld5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
  %ld6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
  %ld7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
  %nearbyint0 = call double @llvm.nearbyint.f64(double %ld0)
  %nearbyint1 = call double @llvm.nearbyint.f64(double %ld1)
  %nearbyint2 = call double @llvm.nearbyint.f64(double %ld2)
  %nearbyint3 = call double @llvm.nearbyint.f64(double %ld3)
  %nearbyint4 = call double @llvm.nearbyint.f64(double %ld4)
  %nearbyint5 = call double @llvm.nearbyint.f64(double %ld5)
  %nearbyint6 = call double @llvm.nearbyint.f64(double %ld6)
  %nearbyint7 = call double @llvm.nearbyint.f64(double %ld7)
  store double %nearbyint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %nearbyint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %nearbyint2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %nearbyint3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  store double %nearbyint4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
  store double %nearbyint5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
  store double %nearbyint6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
  store double %nearbyint7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
  ret void
}

define void @rint_2f64() #0 {
; SSE2-LABEL: @rint_2f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[RINT0:%.*]] = call double @llvm.rint.f64(double [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call double @llvm.rint.f64(double [[LD1]])
; SSE2-NEXT:    store double [[RINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[RINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_2f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @rint_2f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP1]])
; AVX-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %rint0 = call double @llvm.rint.f64(double %ld0)
  %rint1 = call double @llvm.rint.f64(double %ld1)
  store double %rint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %rint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  ret void
}

define void @rint_4f64() #0 {
; SSE2-LABEL: @rint_4f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[RINT0:%.*]] = call double @llvm.rint.f64(double [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call double @llvm.rint.f64(double [[LD1]])
; SSE2-NEXT:    [[RINT2:%.*]] = call double @llvm.rint.f64(double [[LD2]])
; SSE2-NEXT:    [[RINT3:%.*]] = call double @llvm.rint.f64(double [[LD3]])
; SSE2-NEXT:    store double [[RINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[RINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[RINT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[RINT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_4f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @rint_4f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x double> @llvm.rint.v4f64(<4 x double> [[TMP1]])
; AVX-NEXT:    store <4 x double> [[TMP2]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %rint0 = call double @llvm.rint.f64(double %ld0)
  %rint1 = call double @llvm.rint.f64(double %ld1)
  %rint2 = call double @llvm.rint.f64(double %ld2)
  %rint3 = call double @llvm.rint.f64(double %ld3)
  store double %rint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %rint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %rint2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %rint3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  ret void
}

define void @rint_8f64() #0 {
; SSE2-LABEL: @rint_8f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[LD4:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
; SSE2-NEXT:    [[LD5:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
; SSE2-NEXT:    [[LD6:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
; SSE2-NEXT:    [[LD7:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
; SSE2-NEXT:    [[RINT0:%.*]] = call double @llvm.rint.f64(double [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call double @llvm.rint.f64(double [[LD1]])
; SSE2-NEXT:    [[RINT2:%.*]] = call double @llvm.rint.f64(double [[LD2]])
; SSE2-NEXT:    [[RINT3:%.*]] = call double @llvm.rint.f64(double [[LD3]])
; SSE2-NEXT:    [[RINT4:%.*]] = call double @llvm.rint.f64(double [[LD4]])
; SSE2-NEXT:    [[RINT5:%.*]] = call double @llvm.rint.f64(double [[LD5]])
; SSE2-NEXT:    [[RINT6:%.*]] = call double @llvm.rint.f64(double [[LD6]])
; SSE2-NEXT:    [[RINT7:%.*]] = call double @llvm.rint.f64(double [[LD7]])
; SSE2-NEXT:    store double [[RINT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[RINT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[RINT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[RINT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    store double [[RINT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
; SSE2-NEXT:    store double [[RINT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
; SSE2-NEXT:    store double [[RINT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
; SSE2-NEXT:    store double [[RINT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_8f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <2 x double> @llvm.rint.v2f64(<2 x double> [[TMP4]])
; SSE41-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @rint_8f64(
; AVX1-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.rint.v4f64(<4 x double> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.rint.v4f64(<4 x double> [[TMP2]])
; AVX1-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX1-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @rint_8f64(
; AVX2-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.rint.v4f64(<4 x double> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.rint.v4f64(<4 x double> [[TMP2]])
; AVX2-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX2-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @rint_8f64(
; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @src64 to <8 x double>*), align 8
; AVX512-NEXT:    [[TMP2:%.*]] = call <8 x double> @llvm.rint.v8f64(<8 x double> [[TMP1]])
; AVX512-NEXT:    store <8 x double> [[TMP2]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 8
; AVX512-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ld4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
  %ld5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
  %ld6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
  %ld7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
  %rint0 = call double @llvm.rint.f64(double %ld0)
  %rint1 = call double @llvm.rint.f64(double %ld1)
  %rint2 = call double @llvm.rint.f64(double %ld2)
  %rint3 = call double @llvm.rint.f64(double %ld3)
  %rint4 = call double @llvm.rint.f64(double %ld4)
  %rint5 = call double @llvm.rint.f64(double %ld5)
  %rint6 = call double @llvm.rint.f64(double %ld6)
  %rint7 = call double @llvm.rint.f64(double %ld7)
  store double %rint0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %rint1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %rint2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %rint3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  store double %rint4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
  store double %rint5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
  store double %rint6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
  store double %rint7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
  ret void
}

define void @trunc_2f64() #0 {
; SSE2-LABEL: @trunc_2f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
; SSE2-NEXT:    store double [[TRUNC0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[TRUNC1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_2f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @trunc_2f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
; AVX-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %trunc0 = call double @llvm.trunc.f64(double %ld0)
  %trunc1 = call double @llvm.trunc.f64(double %ld1)
  store double %trunc0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %trunc1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  ret void
}

define void @trunc_4f64() #0 {
; SSE2-LABEL: @trunc_4f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
; SSE2-NEXT:    [[TRUNC2:%.*]] = call double @llvm.trunc.f64(double [[LD2]])
; SSE2-NEXT:    [[TRUNC3:%.*]] = call double @llvm.trunc.f64(double [[LD3]])
; SSE2-NEXT:    store double [[TRUNC0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[TRUNC1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[TRUNC2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[TRUNC3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_4f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @trunc_4f64(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x double> @llvm.trunc.v4f64(<4 x double> [[TMP1]])
; AVX-NEXT:    store <4 x double> [[TMP2]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %trunc0 = call double @llvm.trunc.f64(double %ld0)
  %trunc1 = call double @llvm.trunc.f64(double %ld1)
  %trunc2 = call double @llvm.trunc.f64(double %ld2)
  %trunc3 = call double @llvm.trunc.f64(double %ld3)
  store double %trunc0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %trunc1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %trunc2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %trunc3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  ret void
}

define void @trunc_8f64() #0 {
; SSE2-LABEL: @trunc_8f64(
; SSE2-NEXT:    [[LD0:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
; SSE2-NEXT:    [[LD1:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
; SSE2-NEXT:    [[LD2:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
; SSE2-NEXT:    [[LD3:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
; SSE2-NEXT:    [[LD4:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
; SSE2-NEXT:    [[LD5:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
; SSE2-NEXT:    [[LD6:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
; SSE2-NEXT:    [[LD7:%.*]] = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
; SSE2-NEXT:    [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
; SSE2-NEXT:    [[TRUNC2:%.*]] = call double @llvm.trunc.f64(double [[LD2]])
; SSE2-NEXT:    [[TRUNC3:%.*]] = call double @llvm.trunc.f64(double [[LD3]])
; SSE2-NEXT:    [[TRUNC4:%.*]] = call double @llvm.trunc.f64(double [[LD4]])
; SSE2-NEXT:    [[TRUNC5:%.*]] = call double @llvm.trunc.f64(double [[LD5]])
; SSE2-NEXT:    [[TRUNC6:%.*]] = call double @llvm.trunc.f64(double [[LD6]])
; SSE2-NEXT:    [[TRUNC7:%.*]] = call double @llvm.trunc.f64(double [[LD7]])
; SSE2-NEXT:    store double [[TRUNC0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
; SSE2-NEXT:    store double [[TRUNC1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
; SSE2-NEXT:    store double [[TRUNC2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
; SSE2-NEXT:    store double [[TRUNC3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
; SSE2-NEXT:    store double [[TRUNC4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
; SSE2-NEXT:    store double [[TRUNC5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
; SSE2-NEXT:    store double [[TRUNC6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
; SSE2-NEXT:    store double [[TRUNC7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_8f64(
; SSE41-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast ([8 x double]* @src64 to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP3:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP4:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    [[TMP5:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP4]])
; SSE41-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 8
; SSE41-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 8
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @trunc_8f64(
; AVX1-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.trunc.v4f64(<4 x double> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.trunc.v4f64(<4 x double> [[TMP2]])
; AVX1-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX1-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @trunc_8f64(
; AVX2-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* bitcast ([8 x double]* @src64 to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x double>, <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.trunc.v4f64(<4 x double> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <4 x double> @llvm.trunc.v4f64(<4 x double> [[TMP2]])
; AVX2-NEXT:    store <4 x double> [[TMP3]], <4 x double>* bitcast ([8 x double]* @dst64 to <4 x double>*), align 8
; AVX2-NEXT:    store <4 x double> [[TMP4]], <4 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <4 x double>*), align 8
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @trunc_8f64(
; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, <8 x double>* bitcast ([8 x double]* @src64 to <8 x double>*), align 8
; AVX512-NEXT:    [[TMP2:%.*]] = call <8 x double> @llvm.trunc.v8f64(<8 x double> [[TMP1]])
; AVX512-NEXT:    store <8 x double> [[TMP2]], <8 x double>* bitcast ([8 x double]* @dst64 to <8 x double>*), align 8
; AVX512-NEXT:    ret void
;
  %ld0 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 0), align 8
  %ld1 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 1), align 8
  %ld2 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 2), align 8
  %ld3 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 3), align 8
  %ld4 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 4), align 8
  %ld5 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 5), align 8
  %ld6 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 6), align 8
  %ld7 = load double, double* getelementptr inbounds ([8 x double], [8 x double]* @src64, i32 0, i64 7), align 8
  %trunc0 = call double @llvm.trunc.f64(double %ld0)
  %trunc1 = call double @llvm.trunc.f64(double %ld1)
  %trunc2 = call double @llvm.trunc.f64(double %ld2)
  %trunc3 = call double @llvm.trunc.f64(double %ld3)
  %trunc4 = call double @llvm.trunc.f64(double %ld4)
  %trunc5 = call double @llvm.trunc.f64(double %ld5)
  %trunc6 = call double @llvm.trunc.f64(double %ld6)
  %trunc7 = call double @llvm.trunc.f64(double %ld7)
  store double %trunc0, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 8
  store double %trunc1, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
  store double %trunc2, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 8
  store double %trunc3, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
  store double %trunc4, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 8
  store double %trunc5, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
  store double %trunc6, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 8
  store double %trunc7, double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
  ret void
}

define void @ceil_4f32() #0 {
; SSE2-LABEL: @ceil_4f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[CEIL0:%.*]] = call float @llvm.ceil.f32(float [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call float @llvm.ceil.f32(float [[LD1]])
; SSE2-NEXT:    [[CEIL2:%.*]] = call float @llvm.ceil.f32(float [[LD2]])
; SSE2-NEXT:    [[CEIL3:%.*]] = call float @llvm.ceil.f32(float [[LD3]])
; SSE2-NEXT:    store float [[CEIL0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[CEIL1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[CEIL2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[CEIL3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_4f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @ceil_4f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP1]])
; AVX-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ceil0 = call float @llvm.ceil.f32(float %ld0)
  %ceil1 = call float @llvm.ceil.f32(float %ld1)
  %ceil2 = call float @llvm.ceil.f32(float %ld2)
  %ceil3 = call float @llvm.ceil.f32(float %ld3)
  store float %ceil0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %ceil1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %ceil2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %ceil3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  ret void
}

define void @ceil_8f32() #0 {
; SSE2-LABEL: @ceil_8f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[CEIL0:%.*]] = call float @llvm.ceil.f32(float [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call float @llvm.ceil.f32(float [[LD1]])
; SSE2-NEXT:    [[CEIL2:%.*]] = call float @llvm.ceil.f32(float [[LD2]])
; SSE2-NEXT:    [[CEIL3:%.*]] = call float @llvm.ceil.f32(float [[LD3]])
; SSE2-NEXT:    [[CEIL4:%.*]] = call float @llvm.ceil.f32(float [[LD4]])
; SSE2-NEXT:    [[CEIL5:%.*]] = call float @llvm.ceil.f32(float [[LD5]])
; SSE2-NEXT:    [[CEIL6:%.*]] = call float @llvm.ceil.f32(float [[LD6]])
; SSE2-NEXT:    [[CEIL7:%.*]] = call float @llvm.ceil.f32(float [[LD7]])
; SSE2-NEXT:    store float [[CEIL0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[CEIL1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[CEIL2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[CEIL3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[CEIL4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[CEIL5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[CEIL6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[CEIL7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_8f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP4]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @ceil_8f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[TMP1]])
; AVX-NEXT:    store <8 x float> [[TMP2]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ld4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
  %ld5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
  %ld6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
  %ld7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
  %ceil0 = call float @llvm.ceil.f32(float %ld0)
  %ceil1 = call float @llvm.ceil.f32(float %ld1)
  %ceil2 = call float @llvm.ceil.f32(float %ld2)
  %ceil3 = call float @llvm.ceil.f32(float %ld3)
  %ceil4 = call float @llvm.ceil.f32(float %ld4)
  %ceil5 = call float @llvm.ceil.f32(float %ld5)
  %ceil6 = call float @llvm.ceil.f32(float %ld6)
  %ceil7 = call float @llvm.ceil.f32(float %ld7)
  store float %ceil0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %ceil1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %ceil2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %ceil3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  store float %ceil4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
  store float %ceil5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
  store float %ceil6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
  store float %ceil7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
  ret void
}

define void @ceil_16f32() #0 {
; SSE2-LABEL: @ceil_16f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[LD8:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8), align 4
; SSE2-NEXT:    [[LD9:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9), align 4
; SSE2-NEXT:    [[LD10:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
; SSE2-NEXT:    [[LD11:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
; SSE2-NEXT:    [[LD12:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
; SSE2-NEXT:    [[LD13:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
; SSE2-NEXT:    [[LD14:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
; SSE2-NEXT:    [[LD15:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
; SSE2-NEXT:    [[CEIL0:%.*]] = call float @llvm.ceil.f32(float [[LD0]])
; SSE2-NEXT:    [[CEIL1:%.*]] = call float @llvm.ceil.f32(float [[LD1]])
; SSE2-NEXT:    [[CEIL2:%.*]] = call float @llvm.ceil.f32(float [[LD2]])
; SSE2-NEXT:    [[CEIL3:%.*]] = call float @llvm.ceil.f32(float [[LD3]])
; SSE2-NEXT:    [[CEIL4:%.*]] = call float @llvm.ceil.f32(float [[LD4]])
; SSE2-NEXT:    [[CEIL5:%.*]] = call float @llvm.ceil.f32(float [[LD5]])
; SSE2-NEXT:    [[CEIL6:%.*]] = call float @llvm.ceil.f32(float [[LD6]])
; SSE2-NEXT:    [[CEIL7:%.*]] = call float @llvm.ceil.f32(float [[LD7]])
; SSE2-NEXT:    [[CEIL8:%.*]] = call float @llvm.ceil.f32(float [[LD8]])
; SSE2-NEXT:    [[CEIL9:%.*]] = call float @llvm.ceil.f32(float [[LD9]])
; SSE2-NEXT:    [[CEIL10:%.*]] = call float @llvm.ceil.f32(float [[LD10]])
; SSE2-NEXT:    [[CEIL11:%.*]] = call float @llvm.ceil.f32(float [[LD11]])
; SSE2-NEXT:    [[CEIL12:%.*]] = call float @llvm.ceil.f32(float [[LD12]])
; SSE2-NEXT:    [[CEIL13:%.*]] = call float @llvm.ceil.f32(float [[LD13]])
; SSE2-NEXT:    [[CEIL14:%.*]] = call float @llvm.ceil.f32(float [[LD14]])
; SSE2-NEXT:    [[CEIL15:%.*]] = call float @llvm.ceil.f32(float [[LD15]])
; SSE2-NEXT:    store float [[CEIL0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[CEIL1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[CEIL2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[CEIL3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[CEIL4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[CEIL5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[CEIL6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[CEIL7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    store float [[CEIL8]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
; SSE2-NEXT:    store float [[CEIL9]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
; SSE2-NEXT:    store float [[CEIL10]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
; SSE2-NEXT:    store float [[CEIL11]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
; SSE2-NEXT:    store float [[CEIL12]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
; SSE2-NEXT:    store float [[CEIL13]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
; SSE2-NEXT:    store float [[CEIL14]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
; SSE2-NEXT:    store float [[CEIL15]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @ceil_16f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[TMP4]])
; SSE41-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP7]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @ceil_16f32(
; AVX1-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[TMP2]])
; AVX1-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX1-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @ceil_16f32(
; AVX2-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[TMP2]])
; AVX2-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX2-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @ceil_16f32(
; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @src32 to <16 x float>*), align 4
; AVX512-NEXT:    [[TMP2:%.*]] = call <16 x float> @llvm.ceil.v16f32(<16 x float> [[TMP1]])
; AVX512-NEXT:    store <16 x float> [[TMP2]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
; AVX512-NEXT:    ret void
;
  %ld0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0 ), align 4
  %ld1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1 ), align 4
  %ld2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2 ), align 4
  %ld3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3 ), align 4
  %ld4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4 ), align 4
  %ld5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5 ), align 4
  %ld6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6 ), align 4
  %ld7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7 ), align 4
  %ld8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8 ), align 4
  %ld9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9 ), align 4
  %ld10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
  %ld11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
  %ld12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
  %ld13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
  %ld14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
  %ld15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
  %ceil0  = call float @llvm.ceil.f32(float %ld0 )
  %ceil1  = call float @llvm.ceil.f32(float %ld1 )
  %ceil2  = call float @llvm.ceil.f32(float %ld2 )
  %ceil3  = call float @llvm.ceil.f32(float %ld3 )
  %ceil4  = call float @llvm.ceil.f32(float %ld4 )
  %ceil5  = call float @llvm.ceil.f32(float %ld5 )
  %ceil6  = call float @llvm.ceil.f32(float %ld6 )
  %ceil7  = call float @llvm.ceil.f32(float %ld7 )
  %ceil8  = call float @llvm.ceil.f32(float %ld8 )
  %ceil9  = call float @llvm.ceil.f32(float %ld9 )
  %ceil10 = call float @llvm.ceil.f32(float %ld10)
  %ceil11 = call float @llvm.ceil.f32(float %ld11)
  %ceil12 = call float @llvm.ceil.f32(float %ld12)
  %ceil13 = call float @llvm.ceil.f32(float %ld13)
  %ceil14 = call float @llvm.ceil.f32(float %ld14)
  %ceil15 = call float @llvm.ceil.f32(float %ld15)
  store float %ceil0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0 ), align 4
  store float %ceil1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1 ), align 4
  store float %ceil2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2 ), align 4
  store float %ceil3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3 ), align 4
  store float %ceil4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4 ), align 4
  store float %ceil5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5 ), align 4
  store float %ceil6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6 ), align 4
  store float %ceil7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7 ), align 4
  store float %ceil8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8 ), align 4
  store float %ceil9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9 ), align 4
  store float %ceil10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
  store float %ceil11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
  store float %ceil12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
  store float %ceil13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
  store float %ceil14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
  store float %ceil15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
  ret void
}

define void @floor_4f32() #0 {
; SSE2-LABEL: @floor_4f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[FLOOR0:%.*]] = call float @llvm.floor.f32(float [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call float @llvm.floor.f32(float [[LD1]])
; SSE2-NEXT:    [[FLOOR2:%.*]] = call float @llvm.floor.f32(float [[LD2]])
; SSE2-NEXT:    [[FLOOR3:%.*]] = call float @llvm.floor.f32(float [[LD3]])
; SSE2-NEXT:    store float [[FLOOR0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[FLOOR1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[FLOOR2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[FLOOR3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_4f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @floor_4f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP1]])
; AVX-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %floor0 = call float @llvm.floor.f32(float %ld0)
  %floor1 = call float @llvm.floor.f32(float %ld1)
  %floor2 = call float @llvm.floor.f32(float %ld2)
  %floor3 = call float @llvm.floor.f32(float %ld3)
  store float %floor0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %floor1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %floor2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %floor3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  ret void
}

define void @floor_8f32() #0 {
; SSE2-LABEL: @floor_8f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[FLOOR0:%.*]] = call float @llvm.floor.f32(float [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call float @llvm.floor.f32(float [[LD1]])
; SSE2-NEXT:    [[FLOOR2:%.*]] = call float @llvm.floor.f32(float [[LD2]])
; SSE2-NEXT:    [[FLOOR3:%.*]] = call float @llvm.floor.f32(float [[LD3]])
; SSE2-NEXT:    [[FLOOR4:%.*]] = call float @llvm.floor.f32(float [[LD4]])
; SSE2-NEXT:    [[FLOOR5:%.*]] = call float @llvm.floor.f32(float [[LD5]])
; SSE2-NEXT:    [[FLOOR6:%.*]] = call float @llvm.floor.f32(float [[LD6]])
; SSE2-NEXT:    [[FLOOR7:%.*]] = call float @llvm.floor.f32(float [[LD7]])
; SSE2-NEXT:    store float [[FLOOR0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[FLOOR1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[FLOOR2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[FLOOR3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[FLOOR4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[FLOOR5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[FLOOR6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[FLOOR7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_8f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP4]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @floor_8f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[TMP1]])
; AVX-NEXT:    store <8 x float> [[TMP2]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ld4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
  %ld5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
  %ld6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
  %ld7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
  %floor0 = call float @llvm.floor.f32(float %ld0)
  %floor1 = call float @llvm.floor.f32(float %ld1)
  %floor2 = call float @llvm.floor.f32(float %ld2)
  %floor3 = call float @llvm.floor.f32(float %ld3)
  %floor4 = call float @llvm.floor.f32(float %ld4)
  %floor5 = call float @llvm.floor.f32(float %ld5)
  %floor6 = call float @llvm.floor.f32(float %ld6)
  %floor7 = call float @llvm.floor.f32(float %ld7)
  store float %floor0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %floor1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %floor2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %floor3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  store float %floor4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
  store float %floor5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
  store float %floor6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
  store float %floor7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
  ret void
}

define void @floor_16f32() #0 {
; SSE2-LABEL: @floor_16f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[LD8:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8), align 4
; SSE2-NEXT:    [[LD9:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9), align 4
; SSE2-NEXT:    [[LD10:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
; SSE2-NEXT:    [[LD11:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
; SSE2-NEXT:    [[LD12:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
; SSE2-NEXT:    [[LD13:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
; SSE2-NEXT:    [[LD14:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
; SSE2-NEXT:    [[LD15:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
; SSE2-NEXT:    [[FLOOR0:%.*]] = call float @llvm.floor.f32(float [[LD0]])
; SSE2-NEXT:    [[FLOOR1:%.*]] = call float @llvm.floor.f32(float [[LD1]])
; SSE2-NEXT:    [[FLOOR2:%.*]] = call float @llvm.floor.f32(float [[LD2]])
; SSE2-NEXT:    [[FLOOR3:%.*]] = call float @llvm.floor.f32(float [[LD3]])
; SSE2-NEXT:    [[FLOOR4:%.*]] = call float @llvm.floor.f32(float [[LD4]])
; SSE2-NEXT:    [[FLOOR5:%.*]] = call float @llvm.floor.f32(float [[LD5]])
; SSE2-NEXT:    [[FLOOR6:%.*]] = call float @llvm.floor.f32(float [[LD6]])
; SSE2-NEXT:    [[FLOOR7:%.*]] = call float @llvm.floor.f32(float [[LD7]])
; SSE2-NEXT:    [[FLOOR8:%.*]] = call float @llvm.floor.f32(float [[LD8]])
; SSE2-NEXT:    [[FLOOR9:%.*]] = call float @llvm.floor.f32(float [[LD9]])
; SSE2-NEXT:    [[FLOOR10:%.*]] = call float @llvm.floor.f32(float [[LD10]])
; SSE2-NEXT:    [[FLOOR11:%.*]] = call float @llvm.floor.f32(float [[LD11]])
; SSE2-NEXT:    [[FLOOR12:%.*]] = call float @llvm.floor.f32(float [[LD12]])
; SSE2-NEXT:    [[FLOOR13:%.*]] = call float @llvm.floor.f32(float [[LD13]])
; SSE2-NEXT:    [[FLOOR14:%.*]] = call float @llvm.floor.f32(float [[LD14]])
; SSE2-NEXT:    [[FLOOR15:%.*]] = call float @llvm.floor.f32(float [[LD15]])
; SSE2-NEXT:    store float [[FLOOR0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[FLOOR1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[FLOOR2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[FLOOR3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[FLOOR4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[FLOOR5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[FLOOR6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[FLOOR7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    store float [[FLOOR8]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
; SSE2-NEXT:    store float [[FLOOR9]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
; SSE2-NEXT:    store float [[FLOOR10]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
; SSE2-NEXT:    store float [[FLOOR11]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
; SSE2-NEXT:    store float [[FLOOR12]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
; SSE2-NEXT:    store float [[FLOOR13]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
; SSE2-NEXT:    store float [[FLOOR14]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
; SSE2-NEXT:    store float [[FLOOR15]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @floor_16f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[TMP4]])
; SSE41-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP7]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @floor_16f32(
; AVX1-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[TMP2]])
; AVX1-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX1-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @floor_16f32(
; AVX2-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[TMP2]])
; AVX2-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX2-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @floor_16f32(
; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @src32 to <16 x float>*), align 4
; AVX512-NEXT:    [[TMP2:%.*]] = call <16 x float> @llvm.floor.v16f32(<16 x float> [[TMP1]])
; AVX512-NEXT:    store <16 x float> [[TMP2]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
; AVX512-NEXT:    ret void
;
  %ld0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0 ), align 4
  %ld1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1 ), align 4
  %ld2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2 ), align 4
  %ld3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3 ), align 4
  %ld4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4 ), align 4
  %ld5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5 ), align 4
  %ld6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6 ), align 4
  %ld7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7 ), align 4
  %ld8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8 ), align 4
  %ld9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9 ), align 4
  %ld10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
  %ld11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
  %ld12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
  %ld13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
  %ld14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
  %ld15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
  %floor0  = call float @llvm.floor.f32(float %ld0 )
  %floor1  = call float @llvm.floor.f32(float %ld1 )
  %floor2  = call float @llvm.floor.f32(float %ld2 )
  %floor3  = call float @llvm.floor.f32(float %ld3 )
  %floor4  = call float @llvm.floor.f32(float %ld4 )
  %floor5  = call float @llvm.floor.f32(float %ld5 )
  %floor6  = call float @llvm.floor.f32(float %ld6 )
  %floor7  = call float @llvm.floor.f32(float %ld7 )
  %floor8  = call float @llvm.floor.f32(float %ld8 )
  %floor9  = call float @llvm.floor.f32(float %ld9 )
  %floor10 = call float @llvm.floor.f32(float %ld10)
  %floor11 = call float @llvm.floor.f32(float %ld11)
  %floor12 = call float @llvm.floor.f32(float %ld12)
  %floor13 = call float @llvm.floor.f32(float %ld13)
  %floor14 = call float @llvm.floor.f32(float %ld14)
  %floor15 = call float @llvm.floor.f32(float %ld15)
  store float %floor0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0 ), align 4
  store float %floor1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1 ), align 4
  store float %floor2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2 ), align 4
  store float %floor3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3 ), align 4
  store float %floor4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4 ), align 4
  store float %floor5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5 ), align 4
  store float %floor6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6 ), align 4
  store float %floor7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7 ), align 4
  store float %floor8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8 ), align 4
  store float %floor9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9 ), align 4
  store float %floor10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
  store float %floor11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
  store float %floor12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
  store float %floor13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
  store float %floor14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
  store float %floor15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
  ret void
}

define void @nearbyint_4f32() #0 {
; SSE2-LABEL: @nearbyint_4f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call float @llvm.nearbyint.f32(float [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call float @llvm.nearbyint.f32(float [[LD1]])
; SSE2-NEXT:    [[NEARBYINT2:%.*]] = call float @llvm.nearbyint.f32(float [[LD2]])
; SSE2-NEXT:    [[NEARBYINT3:%.*]] = call float @llvm.nearbyint.f32(float [[LD3]])
; SSE2-NEXT:    store float [[NEARBYINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[NEARBYINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[NEARBYINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[NEARBYINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_4f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @nearbyint_4f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP1]])
; AVX-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %nearbyint0 = call float @llvm.nearbyint.f32(float %ld0)
  %nearbyint1 = call float @llvm.nearbyint.f32(float %ld1)
  %nearbyint2 = call float @llvm.nearbyint.f32(float %ld2)
  %nearbyint3 = call float @llvm.nearbyint.f32(float %ld3)
  store float %nearbyint0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %nearbyint1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %nearbyint2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %nearbyint3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  ret void
}

define void @nearbyint_8f32() #0 {
; SSE2-LABEL: @nearbyint_8f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call float @llvm.nearbyint.f32(float [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call float @llvm.nearbyint.f32(float [[LD1]])
; SSE2-NEXT:    [[NEARBYINT2:%.*]] = call float @llvm.nearbyint.f32(float [[LD2]])
; SSE2-NEXT:    [[NEARBYINT3:%.*]] = call float @llvm.nearbyint.f32(float [[LD3]])
; SSE2-NEXT:    [[NEARBYINT4:%.*]] = call float @llvm.nearbyint.f32(float [[LD4]])
; SSE2-NEXT:    [[NEARBYINT5:%.*]] = call float @llvm.nearbyint.f32(float [[LD5]])
; SSE2-NEXT:    [[NEARBYINT6:%.*]] = call float @llvm.nearbyint.f32(float [[LD6]])
; SSE2-NEXT:    [[NEARBYINT7:%.*]] = call float @llvm.nearbyint.f32(float [[LD7]])
; SSE2-NEXT:    store float [[NEARBYINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[NEARBYINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[NEARBYINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[NEARBYINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[NEARBYINT4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[NEARBYINT5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[NEARBYINT6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[NEARBYINT7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_8f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP4]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @nearbyint_8f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> [[TMP1]])
; AVX-NEXT:    store <8 x float> [[TMP2]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ld4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
  %ld5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
  %ld6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
  %ld7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
  %nearbyint0 = call float @llvm.nearbyint.f32(float %ld0)
  %nearbyint1 = call float @llvm.nearbyint.f32(float %ld1)
  %nearbyint2 = call float @llvm.nearbyint.f32(float %ld2)
  %nearbyint3 = call float @llvm.nearbyint.f32(float %ld3)
  %nearbyint4 = call float @llvm.nearbyint.f32(float %ld4)
  %nearbyint5 = call float @llvm.nearbyint.f32(float %ld5)
  %nearbyint6 = call float @llvm.nearbyint.f32(float %ld6)
  %nearbyint7 = call float @llvm.nearbyint.f32(float %ld7)
  store float %nearbyint0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %nearbyint1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %nearbyint2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %nearbyint3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  store float %nearbyint4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
  store float %nearbyint5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
  store float %nearbyint6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
  store float %nearbyint7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
  ret void
}

define void @nearbyint_16f32() #0 {
; SSE2-LABEL: @nearbyint_16f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[LD8:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8), align 4
; SSE2-NEXT:    [[LD9:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9), align 4
; SSE2-NEXT:    [[LD10:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
; SSE2-NEXT:    [[LD11:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
; SSE2-NEXT:    [[LD12:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
; SSE2-NEXT:    [[LD13:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
; SSE2-NEXT:    [[LD14:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
; SSE2-NEXT:    [[LD15:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
; SSE2-NEXT:    [[NEARBYINT0:%.*]] = call float @llvm.nearbyint.f32(float [[LD0]])
; SSE2-NEXT:    [[NEARBYINT1:%.*]] = call float @llvm.nearbyint.f32(float [[LD1]])
; SSE2-NEXT:    [[NEARBYINT2:%.*]] = call float @llvm.nearbyint.f32(float [[LD2]])
; SSE2-NEXT:    [[NEARBYINT3:%.*]] = call float @llvm.nearbyint.f32(float [[LD3]])
; SSE2-NEXT:    [[NEARBYINT4:%.*]] = call float @llvm.nearbyint.f32(float [[LD4]])
; SSE2-NEXT:    [[NEARBYINT5:%.*]] = call float @llvm.nearbyint.f32(float [[LD5]])
; SSE2-NEXT:    [[NEARBYINT6:%.*]] = call float @llvm.nearbyint.f32(float [[LD6]])
; SSE2-NEXT:    [[NEARBYINT7:%.*]] = call float @llvm.nearbyint.f32(float [[LD7]])
; SSE2-NEXT:    [[NEARBYINT8:%.*]] = call float @llvm.nearbyint.f32(float [[LD8]])
; SSE2-NEXT:    [[NEARBYINT9:%.*]] = call float @llvm.nearbyint.f32(float [[LD9]])
; SSE2-NEXT:    [[NEARBYINT10:%.*]] = call float @llvm.nearbyint.f32(float [[LD10]])
; SSE2-NEXT:    [[NEARBYINT11:%.*]] = call float @llvm.nearbyint.f32(float [[LD11]])
; SSE2-NEXT:    [[NEARBYINT12:%.*]] = call float @llvm.nearbyint.f32(float [[LD12]])
; SSE2-NEXT:    [[NEARBYINT13:%.*]] = call float @llvm.nearbyint.f32(float [[LD13]])
; SSE2-NEXT:    [[NEARBYINT14:%.*]] = call float @llvm.nearbyint.f32(float [[LD14]])
; SSE2-NEXT:    [[NEARBYINT15:%.*]] = call float @llvm.nearbyint.f32(float [[LD15]])
; SSE2-NEXT:    store float [[NEARBYINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[NEARBYINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[NEARBYINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[NEARBYINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[NEARBYINT4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[NEARBYINT5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[NEARBYINT6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[NEARBYINT7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    store float [[NEARBYINT8]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
; SSE2-NEXT:    store float [[NEARBYINT9]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
; SSE2-NEXT:    store float [[NEARBYINT10]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
; SSE2-NEXT:    store float [[NEARBYINT11]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
; SSE2-NEXT:    store float [[NEARBYINT12]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
; SSE2-NEXT:    store float [[NEARBYINT13]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
; SSE2-NEXT:    store float [[NEARBYINT14]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
; SSE2-NEXT:    store float [[NEARBYINT15]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @nearbyint_16f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <4 x float> @llvm.nearbyint.v4f32(<4 x float> [[TMP4]])
; SSE41-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP7]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @nearbyint_16f32(
; AVX1-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> [[TMP2]])
; AVX1-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX1-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @nearbyint_16f32(
; AVX2-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.nearbyint.v8f32(<8 x float> [[TMP2]])
; AVX2-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX2-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @nearbyint_16f32(
; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @src32 to <16 x float>*), align 4
; AVX512-NEXT:    [[TMP2:%.*]] = call <16 x float> @llvm.nearbyint.v16f32(<16 x float> [[TMP1]])
; AVX512-NEXT:    store <16 x float> [[TMP2]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
; AVX512-NEXT:    ret void
;
  %ld0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0 ), align 4
  %ld1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1 ), align 4
  %ld2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2 ), align 4
  %ld3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3 ), align 4
  %ld4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4 ), align 4
  %ld5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5 ), align 4
  %ld6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6 ), align 4
  %ld7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7 ), align 4
  %ld8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8 ), align 4
  %ld9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9 ), align 4
  %ld10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
  %ld11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
  %ld12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
  %ld13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
  %ld14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
  %ld15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
  %nearbyint0  = call float @llvm.nearbyint.f32(float %ld0 )
  %nearbyint1  = call float @llvm.nearbyint.f32(float %ld1 )
  %nearbyint2  = call float @llvm.nearbyint.f32(float %ld2 )
  %nearbyint3  = call float @llvm.nearbyint.f32(float %ld3 )
  %nearbyint4  = call float @llvm.nearbyint.f32(float %ld4 )
  %nearbyint5  = call float @llvm.nearbyint.f32(float %ld5 )
  %nearbyint6  = call float @llvm.nearbyint.f32(float %ld6 )
  %nearbyint7  = call float @llvm.nearbyint.f32(float %ld7 )
  %nearbyint8  = call float @llvm.nearbyint.f32(float %ld8 )
  %nearbyint9  = call float @llvm.nearbyint.f32(float %ld9 )
  %nearbyint10 = call float @llvm.nearbyint.f32(float %ld10)
  %nearbyint11 = call float @llvm.nearbyint.f32(float %ld11)
  %nearbyint12 = call float @llvm.nearbyint.f32(float %ld12)
  %nearbyint13 = call float @llvm.nearbyint.f32(float %ld13)
  %nearbyint14 = call float @llvm.nearbyint.f32(float %ld14)
  %nearbyint15 = call float @llvm.nearbyint.f32(float %ld15)
  store float %nearbyint0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0 ), align 4
  store float %nearbyint1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1 ), align 4
  store float %nearbyint2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2 ), align 4
  store float %nearbyint3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3 ), align 4
  store float %nearbyint4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4 ), align 4
  store float %nearbyint5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5 ), align 4
  store float %nearbyint6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6 ), align 4
  store float %nearbyint7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7 ), align 4
  store float %nearbyint8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8 ), align 4
  store float %nearbyint9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9 ), align 4
  store float %nearbyint10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
  store float %nearbyint11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
  store float %nearbyint12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
  store float %nearbyint13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
  store float %nearbyint14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
  store float %nearbyint15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
  ret void
}

define void @rint_4f32() #0 {
; SSE2-LABEL: @rint_4f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[RINT0:%.*]] = call float @llvm.rint.f32(float [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call float @llvm.rint.f32(float [[LD1]])
; SSE2-NEXT:    [[RINT2:%.*]] = call float @llvm.rint.f32(float [[LD2]])
; SSE2-NEXT:    [[RINT3:%.*]] = call float @llvm.rint.f32(float [[LD3]])
; SSE2-NEXT:    store float [[RINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[RINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[RINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[RINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_4f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @rint_4f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP1]])
; AVX-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %rint0 = call float @llvm.rint.f32(float %ld0)
  %rint1 = call float @llvm.rint.f32(float %ld1)
  %rint2 = call float @llvm.rint.f32(float %ld2)
  %rint3 = call float @llvm.rint.f32(float %ld3)
  store float %rint0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %rint1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %rint2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %rint3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  ret void
}

define void @rint_8f32() #0 {
; SSE2-LABEL: @rint_8f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[RINT0:%.*]] = call float @llvm.rint.f32(float [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call float @llvm.rint.f32(float [[LD1]])
; SSE2-NEXT:    [[RINT2:%.*]] = call float @llvm.rint.f32(float [[LD2]])
; SSE2-NEXT:    [[RINT3:%.*]] = call float @llvm.rint.f32(float [[LD3]])
; SSE2-NEXT:    [[RINT4:%.*]] = call float @llvm.rint.f32(float [[LD4]])
; SSE2-NEXT:    [[RINT5:%.*]] = call float @llvm.rint.f32(float [[LD5]])
; SSE2-NEXT:    [[RINT6:%.*]] = call float @llvm.rint.f32(float [[LD6]])
; SSE2-NEXT:    [[RINT7:%.*]] = call float @llvm.rint.f32(float [[LD7]])
; SSE2-NEXT:    store float [[RINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[RINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[RINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[RINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[RINT4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[RINT5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[RINT6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[RINT7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_8f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP4]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @rint_8f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <8 x float> @llvm.rint.v8f32(<8 x float> [[TMP1]])
; AVX-NEXT:    store <8 x float> [[TMP2]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ld4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
  %ld5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
  %ld6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
  %ld7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
  %rint0 = call float @llvm.rint.f32(float %ld0)
  %rint1 = call float @llvm.rint.f32(float %ld1)
  %rint2 = call float @llvm.rint.f32(float %ld2)
  %rint3 = call float @llvm.rint.f32(float %ld3)
  %rint4 = call float @llvm.rint.f32(float %ld4)
  %rint5 = call float @llvm.rint.f32(float %ld5)
  %rint6 = call float @llvm.rint.f32(float %ld6)
  %rint7 = call float @llvm.rint.f32(float %ld7)
  store float %rint0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %rint1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %rint2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %rint3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  store float %rint4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
  store float %rint5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
  store float %rint6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
  store float %rint7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
  ret void
}

define void @rint_16f32() #0 {
; SSE2-LABEL: @rint_16f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[LD8:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8), align 4
; SSE2-NEXT:    [[LD9:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9), align 4
; SSE2-NEXT:    [[LD10:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
; SSE2-NEXT:    [[LD11:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
; SSE2-NEXT:    [[LD12:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
; SSE2-NEXT:    [[LD13:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
; SSE2-NEXT:    [[LD14:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
; SSE2-NEXT:    [[LD15:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
; SSE2-NEXT:    [[RINT0:%.*]] = call float @llvm.rint.f32(float [[LD0]])
; SSE2-NEXT:    [[RINT1:%.*]] = call float @llvm.rint.f32(float [[LD1]])
; SSE2-NEXT:    [[RINT2:%.*]] = call float @llvm.rint.f32(float [[LD2]])
; SSE2-NEXT:    [[RINT3:%.*]] = call float @llvm.rint.f32(float [[LD3]])
; SSE2-NEXT:    [[RINT4:%.*]] = call float @llvm.rint.f32(float [[LD4]])
; SSE2-NEXT:    [[RINT5:%.*]] = call float @llvm.rint.f32(float [[LD5]])
; SSE2-NEXT:    [[RINT6:%.*]] = call float @llvm.rint.f32(float [[LD6]])
; SSE2-NEXT:    [[RINT7:%.*]] = call float @llvm.rint.f32(float [[LD7]])
; SSE2-NEXT:    [[RINT8:%.*]] = call float @llvm.rint.f32(float [[LD8]])
; SSE2-NEXT:    [[RINT9:%.*]] = call float @llvm.rint.f32(float [[LD9]])
; SSE2-NEXT:    [[RINT10:%.*]] = call float @llvm.rint.f32(float [[LD10]])
; SSE2-NEXT:    [[RINT11:%.*]] = call float @llvm.rint.f32(float [[LD11]])
; SSE2-NEXT:    [[RINT12:%.*]] = call float @llvm.rint.f32(float [[LD12]])
; SSE2-NEXT:    [[RINT13:%.*]] = call float @llvm.rint.f32(float [[LD13]])
; SSE2-NEXT:    [[RINT14:%.*]] = call float @llvm.rint.f32(float [[LD14]])
; SSE2-NEXT:    [[RINT15:%.*]] = call float @llvm.rint.f32(float [[LD15]])
; SSE2-NEXT:    store float [[RINT0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[RINT1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[RINT2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[RINT3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[RINT4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[RINT5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[RINT6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[RINT7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    store float [[RINT8]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
; SSE2-NEXT:    store float [[RINT9]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
; SSE2-NEXT:    store float [[RINT10]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
; SSE2-NEXT:    store float [[RINT11]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
; SSE2-NEXT:    store float [[RINT12]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
; SSE2-NEXT:    store float [[RINT13]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
; SSE2-NEXT:    store float [[RINT14]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
; SSE2-NEXT:    store float [[RINT15]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @rint_16f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <4 x float> @llvm.rint.v4f32(<4 x float> [[TMP4]])
; SSE41-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP7]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @rint_16f32(
; AVX1-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.rint.v8f32(<8 x float> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.rint.v8f32(<8 x float> [[TMP2]])
; AVX1-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX1-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @rint_16f32(
; AVX2-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.rint.v8f32(<8 x float> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.rint.v8f32(<8 x float> [[TMP2]])
; AVX2-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX2-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @rint_16f32(
; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @src32 to <16 x float>*), align 4
; AVX512-NEXT:    [[TMP2:%.*]] = call <16 x float> @llvm.rint.v16f32(<16 x float> [[TMP1]])
; AVX512-NEXT:    store <16 x float> [[TMP2]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
; AVX512-NEXT:    ret void
;
  %ld0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0 ), align 4
  %ld1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1 ), align 4
  %ld2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2 ), align 4
  %ld3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3 ), align 4
  %ld4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4 ), align 4
  %ld5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5 ), align 4
  %ld6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6 ), align 4
  %ld7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7 ), align 4
  %ld8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8 ), align 4
  %ld9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9 ), align 4
  %ld10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
  %ld11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
  %ld12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
  %ld13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
  %ld14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
  %ld15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
  %rint0  = call float @llvm.rint.f32(float %ld0 )
  %rint1  = call float @llvm.rint.f32(float %ld1 )
  %rint2  = call float @llvm.rint.f32(float %ld2 )
  %rint3  = call float @llvm.rint.f32(float %ld3 )
  %rint4  = call float @llvm.rint.f32(float %ld4 )
  %rint5  = call float @llvm.rint.f32(float %ld5 )
  %rint6  = call float @llvm.rint.f32(float %ld6 )
  %rint7  = call float @llvm.rint.f32(float %ld7 )
  %rint8  = call float @llvm.rint.f32(float %ld8 )
  %rint9  = call float @llvm.rint.f32(float %ld9 )
  %rint10 = call float @llvm.rint.f32(float %ld10)
  %rint11 = call float @llvm.rint.f32(float %ld11)
  %rint12 = call float @llvm.rint.f32(float %ld12)
  %rint13 = call float @llvm.rint.f32(float %ld13)
  %rint14 = call float @llvm.rint.f32(float %ld14)
  %rint15 = call float @llvm.rint.f32(float %ld15)
  store float %rint0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0 ), align 4
  store float %rint1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1 ), align 4
  store float %rint2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2 ), align 4
  store float %rint3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3 ), align 4
  store float %rint4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4 ), align 4
  store float %rint5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5 ), align 4
  store float %rint6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6 ), align 4
  store float %rint7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7 ), align 4
  store float %rint8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8 ), align 4
  store float %rint9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9 ), align 4
  store float %rint10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
  store float %rint11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
  store float %rint12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
  store float %rint13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
  store float %rint14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
  store float %rint15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
  ret void
}

define void @trunc_4f32() #0 {
; SSE2-LABEL: @trunc_4f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
; SSE2-NEXT:    [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
; SSE2-NEXT:    [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
; SSE2-NEXT:    store float [[TRUNC0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[TRUNC1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[TRUNC2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[TRUNC3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_4f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @trunc_4f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
; AVX-NEXT:    store <4 x float> [[TMP2]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %trunc0 = call float @llvm.trunc.f32(float %ld0)
  %trunc1 = call float @llvm.trunc.f32(float %ld1)
  %trunc2 = call float @llvm.trunc.f32(float %ld2)
  %trunc3 = call float @llvm.trunc.f32(float %ld3)
  store float %trunc0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %trunc1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %trunc2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %trunc3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  ret void
}

define void @trunc_8f32() #0 {
; SSE2-LABEL: @trunc_8f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
; SSE2-NEXT:    [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
; SSE2-NEXT:    [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
; SSE2-NEXT:    [[TRUNC4:%.*]] = call float @llvm.trunc.f32(float [[LD4]])
; SSE2-NEXT:    [[TRUNC5:%.*]] = call float @llvm.trunc.f32(float [[LD5]])
; SSE2-NEXT:    [[TRUNC6:%.*]] = call float @llvm.trunc.f32(float [[LD6]])
; SSE2-NEXT:    [[TRUNC7:%.*]] = call float @llvm.trunc.f32(float [[LD7]])
; SSE2-NEXT:    store float [[TRUNC0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[TRUNC1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[TRUNC2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[TRUNC3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[TRUNC4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[TRUNC5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[TRUNC6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[TRUNC7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_8f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP4:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    store <4 x float> [[TMP3]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP4]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX-LABEL: @trunc_8f32(
; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX-NEXT:    [[TMP2:%.*]] = call <8 x float> @llvm.trunc.v8f32(<8 x float> [[TMP1]])
; AVX-NEXT:    store <8 x float> [[TMP2]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX-NEXT:    ret void
;
  %ld0 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
  %ld1 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
  %ld2 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
  %ld3 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
  %ld4 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
  %ld5 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
  %ld6 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
  %ld7 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
  %trunc0 = call float @llvm.trunc.f32(float %ld0)
  %trunc1 = call float @llvm.trunc.f32(float %ld1)
  %trunc2 = call float @llvm.trunc.f32(float %ld2)
  %trunc3 = call float @llvm.trunc.f32(float %ld3)
  %trunc4 = call float @llvm.trunc.f32(float %ld4)
  %trunc5 = call float @llvm.trunc.f32(float %ld5)
  %trunc6 = call float @llvm.trunc.f32(float %ld6)
  %trunc7 = call float @llvm.trunc.f32(float %ld7)
  store float %trunc0, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
  store float %trunc1, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
  store float %trunc2, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
  store float %trunc3, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
  store float %trunc4, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
  store float %trunc5, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
  store float %trunc6, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
  store float %trunc7, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
  ret void
}

define void @trunc_16f32() #0 {
; SSE2-LABEL: @trunc_16f32(
; SSE2-NEXT:    [[LD0:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0), align 4
; SSE2-NEXT:    [[LD1:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1), align 4
; SSE2-NEXT:    [[LD2:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2), align 4
; SSE2-NEXT:    [[LD3:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3), align 4
; SSE2-NEXT:    [[LD4:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4), align 4
; SSE2-NEXT:    [[LD5:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5), align 4
; SSE2-NEXT:    [[LD6:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6), align 4
; SSE2-NEXT:    [[LD7:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7), align 4
; SSE2-NEXT:    [[LD8:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8), align 4
; SSE2-NEXT:    [[LD9:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9), align 4
; SSE2-NEXT:    [[LD10:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
; SSE2-NEXT:    [[LD11:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
; SSE2-NEXT:    [[LD12:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
; SSE2-NEXT:    [[LD13:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
; SSE2-NEXT:    [[LD14:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
; SSE2-NEXT:    [[LD15:%.*]] = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
; SSE2-NEXT:    [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
; SSE2-NEXT:    [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
; SSE2-NEXT:    [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
; SSE2-NEXT:    [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
; SSE2-NEXT:    [[TRUNC4:%.*]] = call float @llvm.trunc.f32(float [[LD4]])
; SSE2-NEXT:    [[TRUNC5:%.*]] = call float @llvm.trunc.f32(float [[LD5]])
; SSE2-NEXT:    [[TRUNC6:%.*]] = call float @llvm.trunc.f32(float [[LD6]])
; SSE2-NEXT:    [[TRUNC7:%.*]] = call float @llvm.trunc.f32(float [[LD7]])
; SSE2-NEXT:    [[TRUNC8:%.*]] = call float @llvm.trunc.f32(float [[LD8]])
; SSE2-NEXT:    [[TRUNC9:%.*]] = call float @llvm.trunc.f32(float [[LD9]])
; SSE2-NEXT:    [[TRUNC10:%.*]] = call float @llvm.trunc.f32(float [[LD10]])
; SSE2-NEXT:    [[TRUNC11:%.*]] = call float @llvm.trunc.f32(float [[LD11]])
; SSE2-NEXT:    [[TRUNC12:%.*]] = call float @llvm.trunc.f32(float [[LD12]])
; SSE2-NEXT:    [[TRUNC13:%.*]] = call float @llvm.trunc.f32(float [[LD13]])
; SSE2-NEXT:    [[TRUNC14:%.*]] = call float @llvm.trunc.f32(float [[LD14]])
; SSE2-NEXT:    [[TRUNC15:%.*]] = call float @llvm.trunc.f32(float [[LD15]])
; SSE2-NEXT:    store float [[TRUNC0]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0), align 4
; SSE2-NEXT:    store float [[TRUNC1]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1), align 4
; SSE2-NEXT:    store float [[TRUNC2]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2), align 4
; SSE2-NEXT:    store float [[TRUNC3]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3), align 4
; SSE2-NEXT:    store float [[TRUNC4]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4), align 4
; SSE2-NEXT:    store float [[TRUNC5]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5), align 4
; SSE2-NEXT:    store float [[TRUNC6]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6), align 4
; SSE2-NEXT:    store float [[TRUNC7]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7), align 4
; SSE2-NEXT:    store float [[TRUNC8]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8), align 4
; SSE2-NEXT:    store float [[TRUNC9]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9), align 4
; SSE2-NEXT:    store float [[TRUNC10]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
; SSE2-NEXT:    store float [[TRUNC11]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
; SSE2-NEXT:    store float [[TRUNC12]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
; SSE2-NEXT:    store float [[TRUNC13]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
; SSE2-NEXT:    store float [[TRUNC14]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
; SSE2-NEXT:    store float [[TRUNC15]], float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
; SSE2-NEXT:    ret void
;
; SSE41-LABEL: @trunc_16f32(
; SSE41-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([16 x float]* @src32 to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP2:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP4:%.*]] = load <4 x float>, <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    [[TMP5:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
; SSE41-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP2]])
; SSE41-NEXT:    [[TMP7:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP3]])
; SSE41-NEXT:    [[TMP8:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP4]])
; SSE41-NEXT:    store <4 x float> [[TMP5]], <4 x float>* bitcast ([16 x float]* @dst32 to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP6]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP7]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <4 x float>*), align 4
; SSE41-NEXT:    store <4 x float> [[TMP8]], <4 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12) to <4 x float>*), align 4
; SSE41-NEXT:    ret void
;
; AVX1-LABEL: @trunc_16f32(
; AVX1-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.trunc.v8f32(<8 x float> [[TMP1]])
; AVX1-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.trunc.v8f32(<8 x float> [[TMP2]])
; AVX1-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX1-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX1-NEXT:    ret void
;
; AVX2-LABEL: @trunc_16f32(
; AVX2-NEXT:    [[TMP1:%.*]] = load <8 x float>, <8 x float>* bitcast ([16 x float]* @src32 to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.trunc.v8f32(<8 x float> [[TMP1]])
; AVX2-NEXT:    [[TMP4:%.*]] = call <8 x float> @llvm.trunc.v8f32(<8 x float> [[TMP2]])
; AVX2-NEXT:    store <8 x float> [[TMP3]], <8 x float>* bitcast ([16 x float]* @dst32 to <8 x float>*), align 4
; AVX2-NEXT:    store <8 x float> [[TMP4]], <8 x float>* bitcast (float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8) to <8 x float>*), align 4
; AVX2-NEXT:    ret void
;
; AVX512-LABEL: @trunc_16f32(
; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, <16 x float>* bitcast ([16 x float]* @src32 to <16 x float>*), align 4
; AVX512-NEXT:    [[TMP2:%.*]] = call <16 x float> @llvm.trunc.v16f32(<16 x float> [[TMP1]])
; AVX512-NEXT:    store <16 x float> [[TMP2]], <16 x float>* bitcast ([16 x float]* @dst32 to <16 x float>*), align 4
; AVX512-NEXT:    ret void
;
  %ld0  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 0 ), align 4
  %ld1  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 1 ), align 4
  %ld2  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 2 ), align 4
  %ld3  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 3 ), align 4
  %ld4  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 4 ), align 4
  %ld5  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 5 ), align 4
  %ld6  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 6 ), align 4
  %ld7  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 7 ), align 4
  %ld8  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 8 ), align 4
  %ld9  = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 9 ), align 4
  %ld10 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 10), align 4
  %ld11 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 11), align 4
  %ld12 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 12), align 4
  %ld13 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 13), align 4
  %ld14 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 14), align 4
  %ld15 = load float, float* getelementptr inbounds ([16 x float], [16 x float]* @src32, i32 0, i64 15), align 4
  %trunc0  = call float @llvm.trunc.f32(float %ld0 )
  %trunc1  = call float @llvm.trunc.f32(float %ld1 )
  %trunc2  = call float @llvm.trunc.f32(float %ld2 )
  %trunc3  = call float @llvm.trunc.f32(float %ld3 )
  %trunc4  = call float @llvm.trunc.f32(float %ld4 )
  %trunc5  = call float @llvm.trunc.f32(float %ld5 )
  %trunc6  = call float @llvm.trunc.f32(float %ld6 )
  %trunc7  = call float @llvm.trunc.f32(float %ld7 )
  %trunc8  = call float @llvm.trunc.f32(float %ld8 )
  %trunc9  = call float @llvm.trunc.f32(float %ld9 )
  %trunc10 = call float @llvm.trunc.f32(float %ld10)
  %trunc11 = call float @llvm.trunc.f32(float %ld11)
  %trunc12 = call float @llvm.trunc.f32(float %ld12)
  %trunc13 = call float @llvm.trunc.f32(float %ld13)
  %trunc14 = call float @llvm.trunc.f32(float %ld14)
  %trunc15 = call float @llvm.trunc.f32(float %ld15)
  store float %trunc0 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 0 ), align 4
  store float %trunc1 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 1 ), align 4
  store float %trunc2 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 2 ), align 4
  store float %trunc3 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 3 ), align 4
  store float %trunc4 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 4 ), align 4
  store float %trunc5 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 5 ), align 4
  store float %trunc6 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 6 ), align 4
  store float %trunc7 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 7 ), align 4
  store float %trunc8 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 8 ), align 4
  store float %trunc9 , float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 9 ), align 4
  store float %trunc10, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 10), align 4
  store float %trunc11, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 11), align 4
  store float %trunc12, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 12), align 4
  store float %trunc13, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 13), align 4
  store float %trunc14, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 14), align 4
  store float %trunc15, float* getelementptr inbounds ([16 x float], [16 x float]* @dst32, i32 0, i64 15), align 4
  ret void
}

attributes #0 = { nounwind }