llvm/test/CodeGen/AMDGPU/strict_fptrunc.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,SI %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx803 < %s | FileCheck -check-prefixes=GCN,GFX89,GFX8 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX89,GFX9 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX10 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX11-FAKE16 %s

define half @v_constrained_fptrunc_f32_to_f16_fpexcept_strict(float %arg) #0 {
; SI-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v0
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret half %val
}

define <2 x half> @v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict(<2 x float> %arg) #0 {
; SI-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT:    v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
; GFX8-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v1.h, v1
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v1.l, v0
; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v0, v1
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %val = call <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f32(<2 x float> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret <2 x half> %val
}

define <3 x half> @v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict(<3 x float> %arg) #0 {
; SI-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    v_cvt_f16_f32_sdwa v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT:    v_cvt_f16_f32_e32 v1, v2
; GFX8-NEXT:    v_or_b32_e32 v0, v0, v3
; GFX8-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT:    v_cvt_f16_f32_e32 v3, v1
; GFX9-NEXT:    v_cvt_f16_f32_e32 v1, v2
; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
; GFX9-NEXT:    v_perm_b32 v0, v3, v0, s4
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX10-NEXT:    v_cvt_f16_f32_e32 v1, v2
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v3.h, v1
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v3.l, v0
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v1.l, v2
; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v0, v3
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_v3f32_to_v3f16_fpexcept_strict:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v1, v2
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %val = call <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f32(<3 x float> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret <3 x half> %val
}

define float @v_constrained_fptrunc_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_f64_to_f32_fpexcept_strict:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret float %val
}

define <2 x float> @v_constrained_fptrunc_v2f64_to_v2f32_fpexcept_strict(<2 x double> %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_v2f64_to_v2f32_fpexcept_strict:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT:    v_cvt_f32_f64_e32 v1, v[2:3]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %val = call <2 x float> @llvm.experimental.constrained.fptrunc.v2f32.v2f64(<2 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret <2 x float> %val
}

define <3 x float> @v_constrained_fptrunc_v3f64_to_v3f32_fpexcept_strict(<3 x double> %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_v3f64_to_v3f32_fpexcept_strict:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT:    v_cvt_f32_f64_e32 v1, v[2:3]
; GCN-NEXT:    v_cvt_f32_f64_e32 v2, v[4:5]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %val = call <3 x float> @llvm.experimental.constrained.fptrunc.v3f32.v3f64(<3 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret <3 x float> %val
}

; FIXME:
; define half @v_constrained_fptrunc_f64_to_f16_fpexcept_strict(double %arg) #0 {
;   %val = call half @llvm.experimental.constrained.fptrunc.f16.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
;   ret half %val
; }

; define <2 x half> @v_constrained_fptrunc_v2f64_to_v2f16_fpexcept_strict(<2 x double> %arg) #0 {
;   %val = call <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f64(<2 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
;   ret <2 x half> %val
; }

; define <3 x half> @v_constrained_fptrunc_v3f64_to_v3f16_fpexcept_strict(<3 x double> %arg) #0 {
;   %val = call <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f64(<3 x double> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
;   ret <3 x half> %val
; }

define half @v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict(float %arg) #0 {
; SI-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_xor_b32_e32 v0, 0x80000000, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX89-NEXT:    v_xor_b32_e32 v0, 0x8000, v0
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    v_xor_b32_e32 v0, 0x8000, v0
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v0
; GFX11-TRUE16-NEXT:    v_xor_b16 v0.l, 0x8000, v0.l
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fneg_fptrunc_f32_to_f16_fpexcept_strict:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    v_xor_b32_e32 v0, 0x8000, v0
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  %neg.val = fneg half %val
  ret half %neg.val
}

define half @v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict(float %arg) #0 {
; SI-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e64 v0.l, -v0
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_fneg_f32_to_f16_fpexcept_strict:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %neg.arg = fneg float %arg
  %val = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %neg.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret half %val
}

define float @v_constrained_fneg_fptrunc_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fneg_fptrunc_f64_to_f32_fpexcept_strict:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_cvt_f32_f64_e32 v0, v[0:1]
; GCN-NEXT:    v_xor_b32_e32 v0, 0x80000000, v0
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  %neg.val = fneg float %val
  ret float %neg.val
}

define float @v_constrained_fptrunc_fneg_f64_to_f32_fpexcept_strict(double %arg) #0 {
; GCN-LABEL: v_constrained_fptrunc_fneg_f64_to_f32_fpexcept_strict:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_cvt_f32_f64_e64 v0, -v[0:1]
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %neg.arg = fneg double %arg
  %val = call float @llvm.experimental.constrained.fptrunc.f32.f64(double %neg.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret float %val
}

define void @v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi(float %arg, ptr addrspace(1) %ptr) #0 {
; SI-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v0
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret void
}

define void @v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi(<2 x float> %arg, ptr addrspace(1) %ptr) #0 {
; SI-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_mov_b32 s4, s6
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    s_mov_b32 s5, s6
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_or_b32_e32 v0, v0, v1
; SI-NEXT:    buffer_store_dword v0, v[2:3], s[4:7], 0 addr64
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT:    v_cvt_f16_f32_sdwa v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
; GFX8-NEXT:    flat_store_dword v[2:3], v0
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT:    s_mov_b32 s4, 0x5040100
; GFX9-NEXT:    v_perm_b32 v0, v1, v0, s4
; GFX9-NEXT:    global_store_dword v[2:3], v0, off
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX10-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX10-NEXT:    global_store_dword v[2:3], v0, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v1.h, v1
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v1.l, v0
; GFX11-TRUE16-NEXT:    global_store_b32 v[2:3], v1, off
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_v2f32_to_v2f16_fpexcept_strict_noabi:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-FAKE16-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
; GFX11-FAKE16-NEXT:    global_store_b32 v[2:3], v0, off
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %result = call <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f32(<2 x float> %arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  store <2 x half> %result, ptr addrspace(1) %ptr
  ret void
}

define void @v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg(float %arg, ptr addrspace(1) %ptr) #0 {
; SI-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e64 v0.l, -v0
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fneg:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e64 v0, -v0
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %neg.arg = fneg float %arg
  %result = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %neg.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret void
}

define void @v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs(float %arg, ptr addrspace(1) %ptr) #0 {
; SI-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e64 v0, |v0|
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs:
; GFX89:       ; %bb.0:
; GFX89-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX89-NEXT:    v_cvt_f16_f32_e64 v0, |v0|
; GFX89-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_cvt_f16_f32_e64 v0, |v0|
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-TRUE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs:
; GFX11-TRUE16:       ; %bb.0:
; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e64 v0.l, |v0|
; GFX11-TRUE16-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: v_constrained_fptrunc_f32_to_f16_fpexcept_strict_noabi_fabs:
; GFX11-FAKE16:       ; %bb.0:
; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT:    v_cvt_f16_f32_e64 v0, |v0|
; GFX11-FAKE16-NEXT:    s_setpc_b64 s[30:31]
  %abs.arg = call float @llvm.fabs.f32(float %arg) #0
  %result = call half @llvm.experimental.constrained.fptrunc.f16.f32(float %abs.arg, metadata !"round.tonearest", metadata !"fpexcept.strict")
  ret void
}

declare half @llvm.experimental.constrained.fptrunc.f16.f32(float, metadata, metadata)
declare <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f32(<2 x float>, metadata, metadata)
declare <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f32(<3 x float>, metadata, metadata)

declare float @llvm.experimental.constrained.fptrunc.f32.f64(double, metadata, metadata)
declare <2 x float> @llvm.experimental.constrained.fptrunc.v2f32.v2f64(<2 x double>, metadata, metadata)
declare <3 x float> @llvm.experimental.constrained.fptrunc.v3f32.v3f64(<3 x double>, metadata, metadata)

declare half @llvm.experimental.constrained.fptrunc.f16.f64(double, metadata, metadata)
declare <2 x half> @llvm.experimental.constrained.fptrunc.v2f16.v2f64(<2 x double>, metadata, metadata)
declare <3 x half> @llvm.experimental.constrained.fptrunc.v3f16.v3f64(<3 x double>, metadata, metadata)

declare float @llvm.fabs.f32(float)

attributes #0 = { strictfp }