; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=tahiti < %s | FileCheck -enable-var-scope --check-prefixes=SI %s
; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -enable-var-scope --check-prefixes=VI %s
; RUN: llc -march=amdgcn -mcpu=gfx900 < %s | FileCheck -enable-var-scope --check-prefixes=GFX9 %s

declare half @llvm.copysign.f16(half, half) #0
declare float @llvm.copysign.f32(float, float) #0
declare double @llvm.copysign.f64(double, double) #0
declare <2 x half> @llvm.copysign.v2f16(<2 x half>, <2 x half>) #0
declare <3 x half> @llvm.copysign.v3f16(<3 x half>, <3 x half>) #0
declare <4 x half> @llvm.copysign.v4f16(<4 x half>, <4 x half>) #0
declare i32 @llvm.amdgcn.workitem.id.x() #0

define amdgpu_kernel void @s_copysign_f16(ptr addrspace(1) %arg_out, half %mag, half %sign) {
; SI-LABEL: s_copysign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_lshr_b32 s2, s2, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v1, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, v0, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_copysign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_movk_i32 s3, 0x7fff
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s4, s2, 16
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_bfi_b32 v2, s3, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_copysign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_lshr_b32 s1, s4, 16
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    v_mov_b32_e32 v2, s1
; GFX9-NEXT:    v_bfi_b32 v1, s0, v1, v2
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %out = call half @llvm.copysign.f16(half %mag, half %sign)
  store half %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_0(ptr addrspace(1) %out, half %mag) {
; SI-LABEL: s_test_copysign_f16_0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_and_b32 s4, s4, 0x7fff
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_and_b32 s2, s2, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_0:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_and_b32 s0, s4, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half %mag, half 0.0)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_1(ptr addrspace(1) %out, half %mag) {
; SI-LABEL: s_test_copysign_f16_1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_and_b32 s4, s4, 0x7fff
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_and_b32 s2, s2, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_1:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_and_b32 s0, s4, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half %mag, half 1.0)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_10.0(ptr addrspace(1) %out, half %mag) {
; SI-LABEL: s_test_copysign_f16_10.0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_and_b32 s4, s4, 0x7fff
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_10.0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_and_b32 s2, s2, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_10.0:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_and_b32 s0, s4, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half %mag, half 10.0)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_neg1(ptr addrspace(1) %out, half %mag) {
; SI-LABEL: s_test_copysign_f16_neg1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset1_b32 s4, 15
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_neg1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset1_b32 s2, 15
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_neg1:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_or_b32 s0, s4, 0x8000
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half %mag, half -1.0)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_neg10(ptr addrspace(1) %out, half %mag) {
; SI-LABEL: s_test_copysign_f16_neg10:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bitset1_b32 s4, 15
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_neg10:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bitset1_b32 s2, 15
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_neg10:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_or_b32 s0, s4, 0x8000
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half %mag, half -10.0)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_0_mag(ptr addrspace(1) %out, half %sign) {
; SI-LABEL: s_test_copysign_f16_0_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, 0, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_0_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v0, 0xffff8000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v2, s2, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_0_mag:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, s4, v1
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half 0.0, half %sign)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}


define amdgpu_kernel void @s_test_copysign_f16_1_mag(ptr addrspace(1) %out, half %sign) {
; SI-LABEL: s_test_copysign_f16_1_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, 1.0, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_1_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v0, 0xffff8000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, s2, v0
; VI-NEXT:    v_or_b32_e32 v2, 0x3c00, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_1_mag:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, s4, v1
; GFX9-NEXT:    v_or_b32_e32 v1, 0x3c00, v1
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half 1.0, half %sign)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_10_mag(ptr addrspace(1) %out, half %sign) {
; SI-LABEL: s_test_copysign_f16_10_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    v_mov_b32_e32 v1, 0x41200000
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, v1, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_10_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v0, 0xffff8000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, s2, v0
; VI-NEXT:    v_or_b32_e32 v2, 0x4900, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_10_mag:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, s4, v1
; GFX9-NEXT:    v_or_b32_e32 v1, 0x4900, v1
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half 10.0, half %sign)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_neg1_mag(ptr addrspace(1) %out, half %sign) {
; SI-LABEL: s_test_copysign_f16_neg1_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, -1.0, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_neg1_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v0, 0xffff8000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, s2, v0
; VI-NEXT:    v_or_b32_e32 v2, 0x3c00, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_neg1_mag:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, s4, v1
; GFX9-NEXT:    v_or_b32_e32 v1, 0x3c00, v1
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half -1.0, half %sign)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define amdgpu_kernel void @s_test_copysign_f16_neg10_mag(ptr addrspace(1) %out, half %sign) {
; SI-LABEL: s_test_copysign_f16_neg10_mag:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[0:1], 0xb
; SI-NEXT:    v_mov_b32_e32 v1, 0xc1200000
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, v1, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_test_copysign_f16_neg10_mag:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    v_mov_b32_e32 v0, 0xffff8000
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, s2, v0
; VI-NEXT:    v_or_b32_e32 v2, 0x4900, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_test_copysign_f16_neg10_mag:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, s4, v1
; GFX9-NEXT:    v_or_b32_e32 v1, 0x4900, v1
; GFX9-NEXT:    global_store_short v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %result = call half @llvm.copysign.f16(half -10.0, half %sign)
  store half %result, ptr addrspace(1) %out, align 4
  ret void
}

define half @v_copysign_f16(half %mag, half %sign) {
; SI-LABEL: v_copysign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_brev_b32 s4, -2
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_bfi_b32 v0, s4, v0, v1
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_copysign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_movk_i32 s4, 0x7fff
; VI-NEXT:    v_bfi_b32 v0, s4, v0, v1
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_copysign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    s_movk_i32 s4, 0x7fff
; GFX9-NEXT:    v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half %sign)
  ret half %result
}

define half @v_test_copysign_f16_0(half %mag) {
; SI-LABEL: v_test_copysign_f16_0:
; 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_cvt_f32_f16_e64 v0, |v0|
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f16_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_copysign_f16_0:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half 0.0)
  ret half %result
}

define half @v_test_copysign_f16_1(half %mag) {
; SI-LABEL: v_test_copysign_f16_1:
; 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_cvt_f32_f16_e64 v0, |v0|
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f16_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_copysign_f16_1:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half 1.0)
  ret half %result
}

define half @v_test_copysign_f16_10(half %mag) {
; SI-LABEL: v_test_copysign_f16_10:
; 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_cvt_f32_f16_e64 v0, |v0|
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f16_10:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_copysign_f16_10:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half 10.0)
  ret half %result
}

define half @v_test_copysign_f16_neg1(half %mag) {
; SI-LABEL: v_test_copysign_f16_neg1:
; 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_cvt_f32_f16_e64 v0, -|v0|
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f16_neg1:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_or_b32_e32 v0, 0x8000, v0
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_copysign_f16_neg1:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_or_b32_e32 v0, 0x8000, v0
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half -1.0)
  ret half %result
}

define half @v_test_copysign_f16_neg10(half %mag) {
; SI-LABEL: v_test_copysign_f16_neg10:
; 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_cvt_f32_f16_e64 v0, -|v0|
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: v_test_copysign_f16_neg10:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_or_b32_e32 v0, 0x8000, v0
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_copysign_f16_neg10:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_or_b32_e32 v0, 0x8000, v0
; GFX9-NEXT:    s_setpc_b64 s[30:31]
  %result = call half @llvm.copysign.f16(half %mag, half -10.0)
  ret half %result
}

define amdgpu_kernel void @v_copysign_out_f32_mag_f16_sign_f32(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f32_mag_f16_sign_f32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; SI-NEXT:    v_mov_b32_e32 v2, 0
; SI-NEXT:    buffer_load_ushort v3, v[1:2], s[12:15], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; SI-NEXT:    buffer_load_dword v0, v[1:2], s[0:3], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v0, s0, v1, v0
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f32_mag_f16_sign_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v2, s7
; VI-NEXT:    v_add_u32_e32 v1, vcc, s6, v1
; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
; VI-NEXT:    flat_load_ushort v2, v[1:2]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    s_brev_b32 s0, -2
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, s0, v2, v3
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f32_mag_f16_sign_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT:    s_brev_b32 s0, -2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_ushort v1, v1, s[6:7]
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    global_load_dword v0, v0, s[2:3]
; GFX9-NEXT:    s_waitcnt vmcnt(1)
; GFX9-NEXT:    v_cvt_f32_f16_e32 v1, v1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v0, s0, v1, v0
; GFX9-NEXT:    global_store_dword v2, v0, s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr half, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load half, ptr addrspace(1) %arg_mag_gep
  %mag.ext = fpext half %mag to float
  %arg_sign_gep = getelementptr float, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load float, ptr addrspace(1) %arg_sign_gep
  %out = call float @llvm.copysign.f32(float %mag.ext, float %sign)
  store float %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f64_mag_f16_sign_f64(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f64_mag_f16_sign_f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; SI-NEXT:    v_mov_b32_e32 v2, 0
; SI-NEXT:    buffer_load_ushort v3, v[1:2], s[12:15], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[1:2], s[0:3], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v3
; SI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; SI-NEXT:    v_bfi_b32 v3, s0, v3, v1
; SI-NEXT:    buffer_store_dwordx2 v[2:3], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f64_mag_f16_sign_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v2, s7
; VI-NEXT:    v_add_u32_e32 v1, vcc, s6, v1
; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
; VI-NEXT:    flat_load_ushort v2, v[1:2]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    s_brev_b32 s0, -2
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; VI-NEXT:    v_bfi_b32 v3, s0, v3, v1
; VI-NEXT:    flat_store_dwordx2 v[4:5], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f64_mag_f16_sign_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; GFX9-NEXT:    s_brev_b32 s0, -2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_ushort v2, v1, s[6:7]
; GFX9-NEXT:    s_nop 0
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_cvt_f32_f16_e32 v0, v2
; GFX9-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    v_bfi_b32 v3, s0, v3, v1
; GFX9-NEXT:    global_store_dwordx2 v0, v[2:3], s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr half, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load half, ptr addrspace(1) %arg_mag_gep
  %mag.ext = fpext half %mag to double
  %arg_sign_gep = getelementptr double, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load double, ptr addrspace(1) %arg_sign_gep
  %out = call double @llvm.copysign.f64(double %mag.ext, double %sign)
  store double %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f32_mag_f32_sign_f16(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f32_mag_f32_sign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; SI-NEXT:    v_mov_b32_e32 v2, 0
; SI-NEXT:    buffer_load_dword v3, v[1:2], s[12:15], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; SI-NEXT:    buffer_load_ushort v0, v[1:2], s[0:3], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_bfi_b32 v0, s0, v3, v0
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f32_mag_f32_sign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_ushort v4, v[0:1]
; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT:    flat_load_dword v2, v[0:1]
; VI-NEXT:    s_brev_b32 s0, -2
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshlrev_b32_e32 v3, 16, v4
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, s0, v2, v3
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f32_mag_f32_sign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT:    s_brev_b32 s0, -2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_ushort v1, v1, s[2:3]
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    global_load_dword v0, v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(1)
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v0, s0, v0, v1
; GFX9-NEXT:    global_store_dword v2, v0, s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr float, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load float, ptr addrspace(1) %arg_mag_gep
  %arg_sign_gep = getelementptr half, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load half, ptr addrspace(1) %arg_sign_gep
  %sign.ext = fpext half %sign to float
  %out = call float @llvm.copysign.f32(float %mag, float %sign.ext)
  store float %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f64_mag_f64_sign_f16(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f64_mag_f64_sign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
; SI-NEXT:    v_mov_b32_e32 v3, v1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    buffer_load_ushort v2, v[2:3], s[0:3], 0 addr64
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[12:15], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v1, s0, v1, v2
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f64_mag_f64_sign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_ushort v4, v[0:1]
; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    s_brev_b32 s0, -2
; VI-NEXT:    v_mov_b32_e32 v2, s4
; VI-NEXT:    v_mov_b32_e32 v3, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v1, s0, v1, v4
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f64_mag_f64_sign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; GFX9-NEXT:    s_brev_b32 s0, -2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_ushort v2, v1, s[2:3]
; GFX9-NEXT:    v_mov_b32_e32 v3, 0
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(1)
; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v1, s0, v1, v2
; GFX9-NEXT:    global_store_dwordx2 v3, v[0:1], s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr double, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load double, ptr addrspace(1) %arg_mag_gep
  %arg_sign_gep = getelementptr half, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load half, ptr addrspace(1) %arg_sign_gep
  %sign.ext = fpext half %sign to double
  %out = call double @llvm.copysign.f64(double %mag, double %sign.ext)
  store double %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f16_mag_f16_sign_f32(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f16_mag_f16_sign_f32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; SI-NEXT:    v_mov_b32_e32 v2, 0
; SI-NEXT:    buffer_load_ushort v3, v[1:2], s[12:15], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; SI-NEXT:    buffer_load_dword v0, v[1:2], s[0:3], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v0, s0, v1, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    buffer_store_short v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f16_mag_f16_sign_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v4, v[0:1]
; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT:    flat_load_ushort v2, v[0:1]
; VI-NEXT:    s_movk_i32 s0, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v4
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, s0, v2, v3
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f16_mag_f16_sign_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v1, v1, s[2:3]
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    global_load_ushort v0, v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(1)
; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v0, s0, v0, v1
; GFX9-NEXT:    global_store_short v2, v0, s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr half, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load half, ptr addrspace(1) %arg_mag_gep
  %arg_sign_gep = getelementptr float, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load float, ptr addrspace(1) %arg_sign_gep
  %sign.trunc = fptrunc float %sign to half
  %out = call half @llvm.copysign.f16(half %mag, half %sign.trunc)
  store half %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f16_mag_f16_sign_f64(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f16_mag_f16_sign_f64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_mov_b32 s14, s2
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s6
; SI-NEXT:    s_mov_b32 s13, s7
; SI-NEXT:    s_mov_b32 s15, s3
; SI-NEXT:    buffer_load_ushort v2, off, s[12:15], 0
; SI-NEXT:    s_mov_b32 s10, 0
; SI-NEXT:    s_mov_b32 s11, s3
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s1, s5
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v2
; SI-NEXT:    v_bfi_b32 v0, s0, v0, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s0, s4
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f16_mag_f16_sign_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s6
; VI-NEXT:    v_mov_b32_e32 v2, s1
; VI-NEXT:    v_add_u32_e32 v1, vcc, s0, v1
; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
; VI-NEXT:    flat_load_dwordx2 v[1:2], v[1:2]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s7
; VI-NEXT:    s_movk_i32 s0, 0x7fff
; VI-NEXT:    flat_load_ushort v3, v[0:1]
; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, s0, v3, v2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f16_mag_f16_sign_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    global_load_ushort v2, v0, s[6:7]
; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v1, s0, v2, v1
; GFX9-NEXT:    global_store_short v0, v1, s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr half, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load half, ptr addrspace(1) %arg_mag
  %arg_sign_gep = getelementptr double, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load double, ptr addrspace(1) %arg_sign_gep
  %sign.trunc = fptrunc double %sign to half
  %out = call half @llvm.copysign.f16(half %mag, half %sign.trunc)
  store half %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @v_copysign_out_f16_mag_f32_sign_f16(ptr addrspace(1) %arg_out, ptr addrspace(1) %arg_mag, ptr addrspace(1) %arg_sign) {
; SI-LABEL: v_copysign_out_f16_mag_f32_sign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s14, 0
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; SI-NEXT:    v_mov_b32_e32 v2, 0
; SI-NEXT:    buffer_load_dword v3, v[1:2], s[12:15], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v0
; SI-NEXT:    buffer_load_ushort v0, v[1:2], s[0:3], 0 addr64
; SI-NEXT:    s_brev_b32 s0, -2
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    v_bfi_b32 v0, s0, v1, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    buffer_store_short v0, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_copysign_out_f16_mag_f32_sign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; VI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v2, s7
; VI-NEXT:    v_add_u32_e32 v1, vcc, s6, v1
; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
; VI-NEXT:    flat_load_dword v2, v[1:2]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_ushort v3, v[0:1]
; VI-NEXT:    s_movk_i32 s0, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, s0, v2, v3
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_copysign_out_f16_mag_f32_sign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v1, v1, s[6:7]
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    global_load_ushort v0, v0, s[2:3]
; GFX9-NEXT:    s_waitcnt vmcnt(1)
; GFX9-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_bfi_b32 v0, s0, v1, v0
; GFX9-NEXT:    global_store_short v2, v0, s[4:5]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %arg_mag_gep = getelementptr float, ptr addrspace(1) %arg_mag, i32 %tid
  %mag = load float, ptr addrspace(1) %arg_mag_gep
  %mag.trunc = fptrunc float %mag to half
  %arg_sign_gep = getelementptr half, ptr addrspace(1) %arg_sign, i32 %tid
  %sign = load half, ptr addrspace(1) %arg_sign_gep
  %out = call half @llvm.copysign.f16(half %mag.trunc, half %sign)
  store half %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @s_copysign_out_f16_mag_f64_sign_f16(ptr addrspace(1) %arg_out, double %mag, half %sign) {
; SI-LABEL: s_copysign_out_f16_mag_f64_sign_f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s4, s[0:1], 0xd
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s4
; SI-NEXT:    s_lshr_b32 s4, s3, 8
; SI-NEXT:    s_and_b32 s5, s3, 0x1ff
; SI-NEXT:    s_and_b32 s6, s4, 0xffe
; SI-NEXT:    s_or_b32 s2, s5, s2
; SI-NEXT:    s_cmp_lg_u32 s2, 0
; SI-NEXT:    s_cselect_b64 s[4:5], -1, 0
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[4:5]
; SI-NEXT:    v_readfirstlane_b32 s2, v1
; SI-NEXT:    s_bfe_u32 s5, s3, 0xb0014
; SI-NEXT:    s_or_b32 s2, s6, s2
; SI-NEXT:    s_sub_i32 s6, 0x3f1, s5
; SI-NEXT:    v_med3_i32 v1, s6, 0, 13
; SI-NEXT:    s_or_b32 s4, s2, 0x1000
; SI-NEXT:    v_readfirstlane_b32 s6, v1
; SI-NEXT:    s_lshr_b32 s6, s4, s6
; SI-NEXT:    v_lshl_b32_e32 v1, s6, v1
; SI-NEXT:    v_cmp_ne_u32_e32 vcc, s4, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; SI-NEXT:    s_add_i32 s8, s5, 0xfffffc10
; SI-NEXT:    v_readfirstlane_b32 s4, v1
; SI-NEXT:    s_lshl_b32 s5, s8, 12
; SI-NEXT:    s_or_b32 s4, s6, s4
; SI-NEXT:    s_or_b32 s5, s2, s5
; SI-NEXT:    s_cmp_lt_i32 s8, 1
; SI-NEXT:    s_cselect_b32 s9, s4, s5
; SI-NEXT:    s_and_b32 s6, s9, 7
; SI-NEXT:    s_cmp_gt_i32 s6, 5
; SI-NEXT:    s_cselect_b64 s[4:5], -1, 0
; SI-NEXT:    s_cmp_eq_u32 s6, 3
; SI-NEXT:    s_cselect_b64 s[6:7], -1, 0
; SI-NEXT:    s_or_b64 s[4:5], s[6:7], s[4:5]
; SI-NEXT:    s_lshr_b32 s6, s9, 2
; SI-NEXT:    s_or_b32 s4, s4, s5
; SI-NEXT:    s_cmp_lg_u32 s4, 0
; SI-NEXT:    s_addc_u32 s4, s6, 0
; SI-NEXT:    s_cmp_lt_i32 s8, 31
; SI-NEXT:    s_cselect_b32 s6, s4, 0x7c00
; SI-NEXT:    s_cmp_lg_u32 s2, 0
; SI-NEXT:    s_cselect_b64 s[4:5], -1, 0
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[4:5]
; SI-NEXT:    v_lshlrev_b32_e32 v1, 9, v1
; SI-NEXT:    s_cmpk_eq_i32 s8, 0x40f
; SI-NEXT:    v_or_b32_e32 v1, 0x7c00, v1
; SI-NEXT:    v_mov_b32_e32 v2, s6
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_lshr_b32 s2, s3, 16
; SI-NEXT:    v_cndmask_b32_e32 v1, v2, v1, vcc
; SI-NEXT:    s_and_b32 s2, s2, 0x8000
; SI-NEXT:    v_or_b32_e32 v1, s2, v1
; SI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    v_bfi_b32 v0, s2, v1, v0
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_copysign_out_f16_mag_f64_sign_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    s_load_dword s8, s[0:1], 0x34
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s0, s7, 8
; VI-NEXT:    s_and_b32 s1, s7, 0x1ff
; VI-NEXT:    s_and_b32 s2, s0, 0xffe
; VI-NEXT:    s_or_b32 s0, s1, s6
; VI-NEXT:    s_cmp_lg_u32 s0, 0
; VI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; VI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s[0:1]
; VI-NEXT:    v_readfirstlane_b32 s0, v2
; VI-NEXT:    s_bfe_u32 s1, s7, 0xb0014
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    s_or_b32 s4, s2, s0
; VI-NEXT:    s_sub_i32 s2, 0x3f1, s1
; VI-NEXT:    v_med3_i32 v2, s2, 0, 13
; VI-NEXT:    s_or_b32 s0, s4, 0x1000
; VI-NEXT:    v_readfirstlane_b32 s2, v2
; VI-NEXT:    s_lshr_b32 s2, s0, s2
; VI-NEXT:    v_lshlrev_b32_e64 v2, v2, s2
; VI-NEXT:    v_cmp_ne_u32_e32 vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; VI-NEXT:    s_add_i32 s5, s1, 0xfffffc10
; VI-NEXT:    v_readfirstlane_b32 s0, v2
; VI-NEXT:    s_lshl_b32 s1, s5, 12
; VI-NEXT:    s_or_b32 s0, s2, s0
; VI-NEXT:    s_or_b32 s1, s4, s1
; VI-NEXT:    s_cmp_lt_i32 s5, 1
; VI-NEXT:    s_cselect_b32 s6, s0, s1
; VI-NEXT:    s_and_b32 s2, s6, 7
; VI-NEXT:    s_cmp_gt_i32 s2, 5
; VI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; VI-NEXT:    s_cmp_eq_u32 s2, 3
; VI-NEXT:    s_cselect_b64 s[2:3], -1, 0
; VI-NEXT:    s_or_b64 s[0:1], s[2:3], s[0:1]
; VI-NEXT:    s_lshr_b32 s2, s6, 2
; VI-NEXT:    s_cmp_lg_u64 s[0:1], 0
; VI-NEXT:    s_addc_u32 s0, s2, 0
; VI-NEXT:    s_cmp_lt_i32 s5, 31
; VI-NEXT:    s_cselect_b32 s2, s0, 0x7c00
; VI-NEXT:    s_cmp_lg_u32 s4, 0
; VI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; VI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s[0:1]
; VI-NEXT:    v_lshlrev_b32_e32 v2, 9, v2
; VI-NEXT:    s_cmpk_eq_i32 s5, 0x40f
; VI-NEXT:    v_or_b32_e32 v2, 0x7c00, v2
; VI-NEXT:    v_mov_b32_e32 v3, s2
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_lshr_b32 s0, s7, 16
; VI-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
; VI-NEXT:    s_and_b32 s0, s0, 0x8000
; VI-NEXT:    v_or_b32_e32 v2, s0, v2
; VI-NEXT:    s_movk_i32 s0, 0x7fff
; VI-NEXT:    v_mov_b32_e32 v3, s8
; VI-NEXT:    v_bfi_b32 v2, s0, v2, v3
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_copysign_out_f16_mag_f64_sign_f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    s_load_dword s8, s[0:1], 0x34
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_lshr_b32 s0, s7, 8
; GFX9-NEXT:    s_and_b32 s1, s7, 0x1ff
; GFX9-NEXT:    s_and_b32 s2, s0, 0xffe
; GFX9-NEXT:    s_or_b32 s0, s1, s6
; GFX9-NEXT:    s_cmp_lg_u32 s0, 0
; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[0:1]
; GFX9-NEXT:    v_readfirstlane_b32 s0, v1
; GFX9-NEXT:    s_bfe_u32 s1, s7, 0xb0014
; GFX9-NEXT:    s_or_b32 s6, s2, s0
; GFX9-NEXT:    s_sub_i32 s2, 0x3f1, s1
; GFX9-NEXT:    v_med3_i32 v1, s2, 0, 13
; GFX9-NEXT:    s_or_b32 s0, s6, 0x1000
; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
; GFX9-NEXT:    s_lshr_b32 s2, s0, s2
; GFX9-NEXT:    v_lshlrev_b32_e64 v1, v1, s2
; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, s0, v1
; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; GFX9-NEXT:    s_add_i32 s9, s1, 0xfffffc10
; GFX9-NEXT:    v_readfirstlane_b32 s0, v1
; GFX9-NEXT:    s_lshl_b32 s1, s9, 12
; GFX9-NEXT:    s_or_b32 s0, s2, s0
; GFX9-NEXT:    s_or_b32 s1, s6, s1
; GFX9-NEXT:    s_cmp_lt_i32 s9, 1
; GFX9-NEXT:    s_cselect_b32 s10, s0, s1
; GFX9-NEXT:    s_and_b32 s2, s10, 7
; GFX9-NEXT:    s_cmp_gt_i32 s2, 5
; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX9-NEXT:    s_cmp_eq_u32 s2, 3
; GFX9-NEXT:    s_cselect_b64 s[2:3], -1, 0
; GFX9-NEXT:    s_or_b64 s[0:1], s[2:3], s[0:1]
; GFX9-NEXT:    s_lshr_b32 s2, s10, 2
; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
; GFX9-NEXT:    s_addc_u32 s0, s2, 0
; GFX9-NEXT:    s_cmp_lt_i32 s9, 31
; GFX9-NEXT:    s_cselect_b32 s2, s0, 0x7c00
; GFX9-NEXT:    s_cmp_lg_u32 s6, 0
; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[0:1]
; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 9, v1
; GFX9-NEXT:    s_cmpk_eq_i32 s9, 0x40f
; GFX9-NEXT:    v_or_b32_e32 v1, 0x7c00, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
; GFX9-NEXT:    s_lshr_b32 s0, s7, 16
; GFX9-NEXT:    v_cndmask_b32_e32 v1, v2, v1, vcc
; GFX9-NEXT:    s_and_b32 s0, s0, 0x8000
; GFX9-NEXT:    v_or_b32_e32 v1, s0, v1
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v2, s8
; GFX9-NEXT:    v_bfi_b32 v1, s0, v1, v2
; GFX9-NEXT:    global_store_short v0, v1, s[4:5]
; GFX9-NEXT:    s_endpgm
  %mag.trunc = fptrunc double %mag to half
  %result = call half @llvm.copysign.f16(half %mag.trunc, half %sign)
  store half %result, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @s_copysign_v2f16(ptr addrspace(1) %arg_out, <2 x half> %arg_mag, <2 x half> %arg_sign) {
; SI-LABEL: s_copysign_v2f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshr_b32 s4, s2, 16
; SI-NEXT:    s_lshr_b32 s5, s3, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s4
; SI-NEXT:    v_cvt_f32_f16_e32 v1, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v2, s2
; SI-NEXT:    v_cvt_f32_f16_e32 v3, s3
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v0, s2, v0, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_bfi_b32 v1, s2, v2, v3
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_or_b32_e32 v0, v1, v0
; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_copysign_v2f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; VI-NEXT:    s_movk_i32 s4, 0x7fff
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    s_lshr_b32 s3, s3, 16
; VI-NEXT:    s_lshr_b32 s2, s2, 16
; VI-NEXT:    v_bfi_b32 v0, s4, v0, v1
; VI-NEXT:    v_mov_b32_e32 v1, s2
; VI-NEXT:    v_mov_b32_e32 v2, s3
; VI-NEXT:    v_bfi_b32 v1, s4, v1, v2
; VI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; VI-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_copysign_v2f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
; GFX9-NEXT:    s_movk_i32 s4, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v1, s2
; GFX9-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-NEXT:    s_lshr_b32 s3, s3, 16
; GFX9-NEXT:    s_lshr_b32 s2, s2, 16
; GFX9-NEXT:    v_bfi_b32 v1, s4, v1, v2
; GFX9-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-NEXT:    v_bfi_b32 v2, s4, v2, v3
; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX9-NEXT:    s_endpgm
  %out = call <2 x half> @llvm.copysign.v2f16(<2 x half> %arg_mag, <2 x half> %arg_sign)
  store <2 x half> %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @s_copysign_v3f16(ptr addrspace(1) %arg_out, <3 x half> %arg_mag, <3 x half> %arg_sign) {
; SI-LABEL: s_copysign_v3f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshr_b32 s2, s4, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v2, s2
; SI-NEXT:    s_lshr_b32 s2, s6, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v3, s2
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v1, s4
; SI-NEXT:    v_cvt_f32_f16_e32 v4, s7
; SI-NEXT:    v_cvt_f32_f16_e32 v5, s6
; SI-NEXT:    s_brev_b32 s2, -2
; SI-NEXT:    v_bfi_b32 v2, s2, v2, v3
; SI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; SI-NEXT:    v_bfi_b32 v1, s2, v1, v5
; SI-NEXT:    v_bfi_b32 v0, s2, v0, v4
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; SI-NEXT:    v_or_b32_e32 v1, v1, v2
; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:4
; SI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_copysign_v3f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_movk_i32 s2, 0x7fff
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s6
; VI-NEXT:    s_lshr_b32 s3, s6, 16
; VI-NEXT:    s_lshr_b32 s4, s4, 16
; VI-NEXT:    v_bfi_b32 v0, s2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_mov_b32_e32 v2, s3
; VI-NEXT:    v_bfi_b32 v1, s2, v1, v2
; VI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; VI-NEXT:    v_or_b32_sdwa v2, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v0, s5
; VI-NEXT:    v_mov_b32_e32 v1, s7
; VI-NEXT:    v_bfi_b32 v3, s2, v0, v1
; VI-NEXT:    s_add_u32 s2, s0, 4
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_store_short v[0:1], v3
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_copysign_v3f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NEXT:    v_mov_b32_e32 v2, s6
; GFX9-NEXT:    s_lshr_b32 s1, s6, 16
; GFX9-NEXT:    s_lshr_b32 s4, s4, 16
; GFX9-NEXT:    v_bfi_b32 v1, s0, v1, v2
; GFX9-NEXT:    v_mov_b32_e32 v2, s4
; GFX9-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NEXT:    v_bfi_b32 v2, s0, v2, v3
; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s5
; GFX9-NEXT:    v_mov_b32_e32 v3, s7
; GFX9-NEXT:    v_bfi_b32 v2, s0, v2, v3
; GFX9-NEXT:    global_store_short v0, v2, s[2:3] offset:4
; GFX9-NEXT:    global_store_dword v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %out = call <3 x half> @llvm.copysign.v3f16(<3 x half> %arg_mag, <3 x half> %arg_sign)
  store <3 x half> %out, ptr addrspace(1) %arg_out
  ret void
}

define amdgpu_kernel void @s_copysign_v4f16(ptr addrspace(1) %arg_out, <4 x half> %arg_mag, <4 x half> %arg_sign) {
; SI-LABEL: s_copysign_v4f16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshr_b32 s8, s4, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v2, s4
; SI-NEXT:    s_lshr_b32 s4, s6, 16
; SI-NEXT:    s_lshr_b32 s9, s5, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v4, s4
; SI-NEXT:    s_lshr_b32 s4, s7, 16
; SI-NEXT:    v_cvt_f32_f16_e32 v0, s8
; SI-NEXT:    v_cvt_f32_f16_e32 v1, s9
; SI-NEXT:    v_cvt_f32_f16_e32 v5, s4
; SI-NEXT:    v_cvt_f32_f16_e32 v3, s5
; SI-NEXT:    v_cvt_f32_f16_e32 v6, s6
; SI-NEXT:    v_cvt_f32_f16_e32 v7, s7
; SI-NEXT:    s_brev_b32 s4, -2
; SI-NEXT:    v_bfi_b32 v1, s4, v1, v5
; SI-NEXT:    v_bfi_b32 v0, s4, v0, v4
; SI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; SI-NEXT:    v_bfi_b32 v3, s4, v3, v7
; SI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; SI-NEXT:    v_bfi_b32 v2, s4, v2, v6
; SI-NEXT:    v_cvt_f16_f32_e32 v3, v3
; SI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; SI-NEXT:    v_or_b32_e32 v1, v3, v1
; SI-NEXT:    v_or_b32_e32 v0, v2, v0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_copysign_v4f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_movk_i32 s2, 0x7fff
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s5
; VI-NEXT:    v_mov_b32_e32 v1, s7
; VI-NEXT:    s_lshr_b32 s3, s7, 16
; VI-NEXT:    s_lshr_b32 s5, s5, 16
; VI-NEXT:    v_bfi_b32 v0, s2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s3
; VI-NEXT:    v_bfi_b32 v1, s2, v1, v2
; VI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; VI-NEXT:    v_or_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    s_lshr_b32 s3, s6, 16
; VI-NEXT:    s_lshr_b32 s4, s4, 16
; VI-NEXT:    v_bfi_b32 v0, s2, v0, v2
; VI-NEXT:    v_mov_b32_e32 v2, s4
; VI-NEXT:    v_mov_b32_e32 v3, s3
; VI-NEXT:    v_bfi_b32 v2, s2, v2, v3
; VI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; VI-NEXT:    v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_copysign_v4f16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
; GFX9-NEXT:    v_mov_b32_e32 v2, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v0, s5
; GFX9-NEXT:    v_mov_b32_e32 v1, s7
; GFX9-NEXT:    s_lshr_b32 s1, s7, 16
; GFX9-NEXT:    s_lshr_b32 s5, s5, 16
; GFX9-NEXT:    v_bfi_b32 v0, s0, v0, v1
; GFX9-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NEXT:    v_bfi_b32 v1, s0, v1, v3
; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    v_mov_b32_e32 v3, s6
; GFX9-NEXT:    s_lshr_b32 s1, s6, 16
; GFX9-NEXT:    s_lshr_b32 s4, s4, 16
; GFX9-NEXT:    v_bfi_b32 v0, s0, v0, v3
; GFX9-NEXT:    v_mov_b32_e32 v3, s4
; GFX9-NEXT:    v_mov_b32_e32 v4, s1
; GFX9-NEXT:    v_bfi_b32 v3, s0, v3, v4
; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX9-NEXT:    s_endpgm
  %out = call <4 x half> @llvm.copysign.v4f16(<4 x half> %arg_mag, <4 x half> %arg_sign)
  store <4 x half> %out, ptr addrspace(1) %arg_out
  ret void
}

attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }