nuudlman/llvm: llvm/test/CodeGen/X86/avx512-intrinsics-x86

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=+avx512f | FileCheck %s


define i64 @test_x86_sse2_cvtsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_sse2_cvtsd2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsd2si %xmm0, %rax
; CHECK-NEXT:    retq
  %res = call i64 @llvm.x86.sse2.cvtsd2si64(<2 x double> %a0) ; <i64> [#uses=1]
  ret i64 %res
}
declare i64 @llvm.x86.sse2.cvtsd2si64(<2 x double>) nounwind readnone

define <2 x double> @test_x86_sse2_cvtsi642sd(<2 x double> %a0, i64 %a1) {
; CHECK-LABEL: test_x86_sse2_cvtsi642sd:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsi2sd %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double> %a0, i64 %a1) ; <<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double>, i64) nounwind readnone

define i64 @test_x86_avx512_cvttsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvttsd2si %xmm0, %rcx
; CHECK-NEXT:    vcvttsd2si {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttsd2si64(<2 x double> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttsd2si64(<2 x double> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttsd2si64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_avx512_cvttsd2usi64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2usi64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvttsd2usi %xmm0, %rcx
; CHECK-NEXT:    vcvttsd2usi {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_sse_cvtss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_sse_cvtss2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtss2si %xmm0, %rax
; CHECK-NEXT:    retq
  %res = call i64 @llvm.x86.sse.cvtss2si64(<4 x float> %a0) ; <i64> [#uses=1]
  ret i64 %res
}
declare i64 @llvm.x86.sse.cvtss2si64(<4 x float>) nounwind readnone


define <4 x float> @test_x86_sse_cvtsi642ss(<4 x float> %a0, i64 %a1) {
; CHECK-LABEL: test_x86_sse_cvtsi642ss:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsi2ss %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float> %a0, i64 %a1) ; <<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float>, i64) nounwind readnone


define i64 @test_x86_avx512_cvttss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvttss2si %xmm0, %rcx
; CHECK-NEXT:    vcvttss2si {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttss2si64(<4 x float> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttss2si64(<4 x float> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttss2si64(<4 x float>, i32) nounwind readnone

define i32 @test_x86_avx512_cvttss2usi(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2usi:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvttss2usi {sae}, %xmm0, %ecx
; CHECK-NEXT:    vcvttss2usi %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq
  %res0 = call i32 @llvm.x86.avx512.cvttss2usi(<4 x float> %a0, i32 8) ;
  %res1 = call i32 @llvm.x86.avx512.cvttss2usi(<4 x float> %a0, i32 4) ;
  %res2 = add i32 %res0, %res1
  ret i32 %res2
}
declare i32 @llvm.x86.avx512.cvttss2usi(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvttss2usi64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2usi64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvttss2usi %xmm0, %rcx
; CHECK-NEXT:    vcvttss2usi {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttss2usi64(<4 x float> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttss2usi64(<4 x float> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttss2usi64(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtsd2usi64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2usi64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsd2usi %xmm0, %rax
; CHECK-NEXT:    vcvtsd2usi {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtsd2usi {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 11)
  %res2 = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 9)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsd2si %xmm0, %rax
; CHECK-NEXT:    vcvtsd2si {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtsd2si {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 11)
  %res2 = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 9)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtss2usi64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2usi64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtss2usi %xmm0, %rax
; CHECK-NEXT:    vcvtss2usi {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtss2usi {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 11)
  %res2 = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 9)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2si64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtss2si %xmm0, %rax
; CHECK-NEXT:    vcvtss2si {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtss2si {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 11)
  %res2 = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 9)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtss2si64(<4 x float>, i32) nounwind readnone

define <2 x double> @test_x86_avx512_cvtsi2sd64(<2 x double> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512_cvtsi2sd64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsi2sd %rdi, {rz-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.cvtsi2sd64(<2 x double> %a, i64 %b, i32 11) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.avx512.cvtsi2sd64(<2 x double>, i64, i32) nounwind readnone

define <4 x float> @test_x86_avx512_cvtsi2ss64(<4 x float> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512_cvtsi2ss64:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtsi2ss %rdi, {rz-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.cvtsi2ss64(<4 x float> %a, i64 %b, i32 11) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtsi2ss64(<4 x float>, i64, i32) nounwind readnone

define <4 x float> @_mm_cvt_roundu64_ss (<4 x float> %a, i64 %b) {
; CHECK-LABEL: _mm_cvt_roundu64_ss:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtusi2ss %rdi, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float> %a, i64 %b, i32 9) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}

define <4 x float> @_mm_cvtu64_ss(<4 x float> %a, i64 %b) {
; CHECK-LABEL: _mm_cvtu64_ss:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtusi2ss %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float> %a, i64 %b, i32 4) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float>, i64, i32) nounwind readnone

define <2 x double> @test_x86_avx512_mm_cvtu64_sd(<2 x double> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512_mm_cvtu64_sd:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtusi2sd %rdi, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double> %a, i64 %b, i32 9) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}

define <2 x double> @test_x86_avx512__mm_cvt_roundu64_sd(<2 x double> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512__mm_cvt_roundu64_sd:
; CHECK:       ## %bb.0:
; CHECK-NEXT:    vcvtusi2sd %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double> %a, i64 %b, i32 4) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double>, i64, i32) nounwind readnone