; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse4a | FileCheck %s --check-prefix=X32
; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse4a,+avx | FileCheck %s --check-prefix=X32
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4a | FileCheck %s --check-prefix=X64
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4a,+avx | FileCheck %s --check-prefix=X64

define void @test_movntss(i8* %p, <4 x float> %a) nounwind optsize ssp {
; X32-LABEL: test_movntss:
; X32:       # BB#0:
; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X32-NEXT:    movntss %xmm0, (%eax)
; X32-NEXT:    retl
;
; X64-LABEL: test_movntss:
; X64:       # BB#0:
; X64-NEXT:    movntss %xmm0, (%rdi)
; X64-NEXT:    retq
  tail call void @llvm.x86.sse4a.movnt.ss(i8* %p, <4 x float> %a) nounwind
  ret void
}

declare void @llvm.x86.sse4a.movnt.ss(i8*, <4 x float>)

define void @test_movntsd(i8* %p, <2 x double> %a) nounwind optsize ssp {
; X32-LABEL: test_movntsd:
; X32:       # BB#0:
; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X32-NEXT:    movntsd %xmm0, (%eax)
; X32-NEXT:    retl
;
; X64-LABEL: test_movntsd:
; X64:       # BB#0:
; X64-NEXT:    movntsd %xmm0, (%rdi)
; X64-NEXT:    retq
  tail call void @llvm.x86.sse4a.movnt.sd(i8* %p, <2 x double> %a) nounwind
  ret void
}

declare void @llvm.x86.sse4a.movnt.sd(i8*, <2 x double>)

define <2 x i64> @test_extrqi(<2 x i64> %x) nounwind uwtable ssp {
; X32-LABEL: test_extrqi:
; X32:       # BB#0:
; X32-NEXT:    extrq $2, $3, %xmm0
; X32-NEXT:    retl
;
; X64-LABEL: test_extrqi:
; X64:       # BB#0:
; X64-NEXT:    extrq $2, $3, %xmm0
; X64-NEXT:    retq
  %1 = tail call <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64> %x, i8 3, i8 2)
  ret <2 x i64> %1
}

declare <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64>, i8, i8) nounwind

define <2 x i64> @test_extrq(<2 x i64> %x, <2 x i64> %y) nounwind uwtable ssp {
; X32-LABEL: test_extrq:
; X32:       # BB#0:
; X32-NEXT:    extrq %xmm1, %xmm0
; X32-NEXT:    retl
;
; X64-LABEL: test_extrq:
; X64:       # BB#0:
; X64-NEXT:    extrq %xmm1, %xmm0
; X64-NEXT:    retq
  %1 = bitcast <2 x i64> %y to <16 x i8>
  %2 = tail call <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64> %x, <16 x i8> %1) nounwind
  ret <2 x i64> %2
}

declare <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64>, <16 x i8>) nounwind

define <2 x i64> @test_insertqi(<2 x i64> %x, <2 x i64> %y) nounwind uwtable ssp {
; X32-LABEL: test_insertqi:
; X32:       # BB#0:
; X32-NEXT:    insertq $6, $5, %xmm1, %xmm0
; X32-NEXT:    retl
;
; X64-LABEL: test_insertqi:
; X64:       # BB#0:
; X64-NEXT:    insertq $6, $5, %xmm1, %xmm0
; X64-NEXT:    retq
  %1 = tail call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> %y, i8 5, i8 6)
  ret <2 x i64> %1
}

declare <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64>, <2 x i64>, i8, i8) nounwind

define <2 x i64> @test_insertq(<2 x i64> %x, <2 x i64> %y) nounwind uwtable ssp {
; X32-LABEL: test_insertq:
; X32:       # BB#0:
; X32-NEXT:    insertq %xmm1, %xmm0
; X32-NEXT:    retl
;
; X64-LABEL: test_insertq:
; X64:       # BB#0:
; X64-NEXT:    insertq %xmm1, %xmm0
; X64-NEXT:    retq
  %1 = tail call <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64> %x, <2 x i64> %y) nounwind
  ret <2 x i64> %1
}

declare <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64>, <2 x i64>) nounwind