; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512fp16 -mattr=+avx512vl -O3 | FileCheck %s --check-prefixes=X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16 -mattr=+avx512vl -O3 | FileCheck %s --check-prefixes=X64

declare half @llvm.experimental.constrained.fadd.f16(half, half, metadata, metadata)
declare half @llvm.experimental.constrained.fsub.f16(half, half, metadata, metadata)
declare half @llvm.experimental.constrained.fmul.f16(half, half, metadata, metadata)
declare half @llvm.experimental.constrained.fdiv.f16(half, half, metadata, metadata)
declare float @llvm.experimental.constrained.fpext.f32.f16(half, metadata)
declare double @llvm.experimental.constrained.fpext.f64.f16(half, metadata)
declare half @llvm.experimental.constrained.fptrunc.f16.f32(float, metadata, metadata)
declare half @llvm.experimental.constrained.fptrunc.f16.f64(double, metadata, metadata)
declare half @llvm.experimental.constrained.sqrt.f16(half, metadata, metadata)
declare half @llvm.experimental.constrained.fma.f16(half, half, half, metadata, metadata)

define half @fadd_f16(half %a, half %b) nounwind strictfp {
; X86-LABEL: fadd_f16:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vaddsh {{[0-9]+}}(%esp), %xmm0, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: fadd_f16:
; X64:       # %bb.0:
; X64-NEXT:    vaddsh %xmm1, %xmm0, %xmm0
; X64-NEXT:    retq
  %ret = call half @llvm.experimental.constrained.fadd.f16(half %a, half %b,
                                                           metadata !"round.dynamic",
                                                           metadata !"fpexcept.strict") #0
  ret half %ret
}

define half @fsub_f16(half %a, half %b) nounwind strictfp {
; X86-LABEL: fsub_f16:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vsubsh {{[0-9]+}}(%esp), %xmm0, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: fsub_f16:
; X64:       # %bb.0:
; X64-NEXT:    vsubsh %xmm1, %xmm0, %xmm0
; X64-NEXT:    retq
  %ret = call half @llvm.experimental.constrained.fsub.f16(half %a, half %b,
                                                           metadata !"round.dynamic",
                                                           metadata !"fpexcept.strict") #0
  ret half %ret
}

define half @fmul_f16(half %a, half %b) nounwind strictfp {
; X86-LABEL: fmul_f16:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vmulsh {{[0-9]+}}(%esp), %xmm0, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: fmul_f16:
; X64:       # %bb.0:
; X64-NEXT:    vmulsh %xmm1, %xmm0, %xmm0
; X64-NEXT:    retq
  %ret = call half @llvm.experimental.constrained.fmul.f16(half %a, half %b,
                                                           metadata !"round.dynamic",
                                                           metadata !"fpexcept.strict") #0
  ret half %ret
}

define half @fdiv_f16(half %a, half %b) nounwind strictfp {
; X86-LABEL: fdiv_f16:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vdivsh {{[0-9]+}}(%esp), %xmm0, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: fdiv_f16:
; X64:       # %bb.0:
; X64-NEXT:    vdivsh %xmm1, %xmm0, %xmm0
; X64-NEXT:    retq
  %ret = call half @llvm.experimental.constrained.fdiv.f16(half %a, half %b,
                                                           metadata !"round.dynamic",
                                                           metadata !"fpexcept.strict") #0
  ret half %ret
}

define void @fpext_f16_to_f32(half* %val, float* %ret) nounwind strictfp {
; X86-LABEL: fpext_f16_to_f32:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    vmovsh (%ecx), %xmm0
; X86-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovss %xmm0, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: fpext_f16_to_f32:
; X64:       # %bb.0:
; X64-NEXT:    vmovsh (%rdi), %xmm0
; X64-NEXT:    vcvtsh2ss %xmm0, %xmm0, %xmm0
; X64-NEXT:    vmovss %xmm0, (%rsi)
; X64-NEXT:    retq
  %1 = load half, half* %val, align 4
  %res = call float @llvm.experimental.constrained.fpext.f32.f16(half %1,
                                                                 metadata !"fpexcept.strict") #0
  store float %res, float* %ret, align 8
  ret void
}

define void @fpext_f16_to_f64(half* %val, double* %ret) nounwind strictfp {
; X86-LABEL: fpext_f16_to_f64:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    vmovsh (%ecx), %xmm0
; X86-NEXT:    vcvtsh2sd %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsd %xmm0, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: fpext_f16_to_f64:
; X64:       # %bb.0:
; X64-NEXT:    vmovsh (%rdi), %xmm0
; X64-NEXT:    vcvtsh2sd %xmm0, %xmm0, %xmm0
; X64-NEXT:    vmovsd %xmm0, (%rsi)
; X64-NEXT:    retq
  %1 = load half, half* %val, align 4
  %res = call double @llvm.experimental.constrained.fpext.f64.f16(half %1,
                                                                  metadata !"fpexcept.strict") #0
  store double %res, double* %ret, align 8
  ret void
}

define void @fptrunc_float_to_f16(float* %val, half *%ret) nounwind strictfp {
; X86-LABEL: fptrunc_float_to_f16:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsh %xmm0, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: fptrunc_float_to_f16:
; X64:       # %bb.0:
; X64-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X64-NEXT:    vcvtss2sh %xmm0, %xmm0, %xmm0
; X64-NEXT:    vmovsh %xmm0, (%rsi)
; X64-NEXT:    retq
  %1 = load float, float* %val, align 8
  %res = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %1,
                                                                  metadata !"round.dynamic",
                                                                  metadata !"fpexcept.strict") #0
  store half %res, half* %ret, align 4
  ret void
}

define void @fptrunc_double_to_f16(double* %val, half *%ret) nounwind strictfp {
; X86-LABEL: fptrunc_double_to_f16:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X86-NEXT:    vcvtsd2sh %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsh %xmm0, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: fptrunc_double_to_f16:
; X64:       # %bb.0:
; X64-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; X64-NEXT:    vcvtsd2sh %xmm0, %xmm0, %xmm0
; X64-NEXT:    vmovsh %xmm0, (%rsi)
; X64-NEXT:    retq
  %1 = load double, double* %val, align 8
  %res = call half @llvm.experimental.constrained.fptrunc.f16.f64(double %1,
                                                                  metadata !"round.dynamic",
                                                                  metadata !"fpexcept.strict") #0
  store half %res, half* %ret, align 4
  ret void
}

define void @fsqrt_f16(half* %a) nounwind strictfp {
; X86-LABEL: fsqrt_f16:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vmovsh (%eax), %xmm0
; X86-NEXT:    vsqrtsh %xmm0, %xmm0, %xmm0
; X86-NEXT:    vmovsh %xmm0, (%eax)
; X86-NEXT:    retl
;
; X64-LABEL: fsqrt_f16:
; X64:       # %bb.0:
; X64-NEXT:    vmovsh (%rdi), %xmm0
; X64-NEXT:    vsqrtsh %xmm0, %xmm0, %xmm0
; X64-NEXT:    vmovsh %xmm0, (%rdi)
; X64-NEXT:    retq
  %1 = load half, half* %a, align 4
  %res = call half @llvm.experimental.constrained.sqrt.f16(half %1,
                                                           metadata !"round.dynamic",
                                                           metadata !"fpexcept.strict") #0
  store half %res, half* %a, align 4
  ret void
}

define half @fma_f16(half %a, half %b, half %c) nounwind strictfp {
; X86-LABEL: fma_f16:
; X86:       # %bb.0:
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm1
; X86-NEXT:    vmovsh {{[0-9]+}}(%esp), %xmm0
; X86-NEXT:    vfmadd213sh {{[0-9]+}}(%esp), %xmm1, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: fma_f16:
; X64:       # %bb.0:
; X64-NEXT:    vfmadd213sh %xmm2, %xmm1, %xmm0
; X64-NEXT:    retq
  %res = call half @llvm.experimental.constrained.fma.f16(half %a, half %b, half %c,
                                                          metadata !"round.dynamic",
                                                          metadata !"fpexcept.strict") #0
  ret half %res
}

attributes #0 = { strictfp }