; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA32
; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64

;; fptrunc
define void @fptrunc_v1f64_to_v1f32(ptr %res, ptr %a0) nounwind {
; CHECK-LABEL: fptrunc_v1f64_to_v1f32:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    fld.d $fa0, $a1, 0
; CHECK-NEXT:    fcvt.s.d $fa0, $fa0
; CHECK-NEXT:    fst.s $fa0, $a0, 0
; CHECK-NEXT:    ret
entry:
  %v0 = load <1 x double>, ptr %a0
  %trunc = fptrunc <1 x double> %v0 to <1 x float>
  store <1 x float> %trunc, ptr %res
  ret void
}

define void @fptrunc_v2f64_to_v2f32(ptr %res, ptr %a0) nounwind {
; LA32-LABEL: fptrunc_v2f64_to_v2f32:
; LA32:       # %bb.0: # %entry
; LA32-NEXT:    vld $vr0, $a1, 0
; LA32-NEXT:    vfcvt.s.d $vr0, $vr0, $vr0
; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
; LA32-NEXT:    ret
;
; LA64-LABEL: fptrunc_v2f64_to_v2f32:
; LA64:       # %bb.0: # %entry
; LA64-NEXT:    vld $vr0, $a1, 0
; LA64-NEXT:    vfcvt.s.d $vr0, $vr0, $vr0
; LA64-NEXT:    vstelm.d $vr0, $a0, 0, 0
; LA64-NEXT:    ret
entry:
  %v0 = load <2 x double>, ptr %a0
  %trunc = fptrunc <2 x double> %v0 to <2 x float>
  store <2 x float> %trunc, ptr %res
  ret void
}

define void @fptrunc_v4f64_to_v4f32(ptr %res, ptr %a0) nounwind {
; CHECK-LABEL: fptrunc_v4f64_to_v4f32:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vld $vr0, $a1, 0
; CHECK-NEXT:    vld $vr1, $a1, 16
; CHECK-NEXT:    vfcvt.s.d $vr0, $vr1, $vr0
; CHECK-NEXT:    vst $vr0, $a0, 0
; CHECK-NEXT:    ret
entry:
  %v0 = load <4 x double>, ptr %a0
  %trunc = fptrunc <4 x double> %v0 to <4 x float>
  store <4 x float> %trunc, ptr %res
  ret void
}

define void @fptrunc_concat_bitcast(ptr %res, ptr %a0) nounwind {
; CHECK-LABEL: fptrunc_concat_bitcast:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vld $vr0, $a1, 0
; CHECK-NEXT:    vld $vr1, $a1, 16
; CHECK-NEXT:    vfcvt.s.d $vr0, $vr1, $vr0
; CHECK-NEXT:    vst $vr0, $a0, 0
; CHECK-NEXT:    ret
entry:
  %x = load <4 x double>, ptr %a0

  ; split to 2 x <2 x double>
  %lo = shufflevector <4 x double> %x, <4 x double> poison,
                    <2 x i32> <i32 0, i32 1>
  %hi = shufflevector <4 x double> %x, <4 x double> poison,
                    <2 x i32> <i32 2, i32 3>

  %r0 = fptrunc <2 x double> %lo to <2 x float>
  %r1 = fptrunc <2 x double> %hi to <2 x float>

  %i0 = bitcast <2 x float> %r0 to i64
  %i1 = bitcast <2 x float> %r1 to i64

  ; concat as integer vector
  %cat0 = insertelement <2 x i64> poison, i64 %i0, i32 0
  %cat1 = insertelement <2 x i64> %cat0, i64 %i1, i32 1
  store <2 x i64> %cat1, ptr %res
  ret void
}