; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=SI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefixes=VI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s


define amdgpu_kernel void @s_usubo_i64_zext(ptr addrspace(1) %out, i64 %a, i64 %b) #0 {
; SI-LABEL: s_usubo_i64_zext:
; 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_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s0, s4
; SI-NEXT:    s_mov_b32 s1, s5
; SI-NEXT:    s_sub_u32 s4, s6, s8
; SI-NEXT:    v_mov_b32_e32 v0, s6
; SI-NEXT:    v_mov_b32_e32 v1, s7
; SI-NEXT:    s_subb_u32 s5, s7, s9
; SI-NEXT:    v_cmp_gt_u64_e32 vcc, s[4:5], v[0:1]
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
; SI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_usubo_i64_zext:
; 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:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    s_sub_u32 s0, s6, s0
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    s_subb_u32 s1, s7, s1
; VI-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[2:3]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_usubo_i64_zext:
; 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_mov_b32_e32 v2, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v0, s6
; GFX9-NEXT:    s_sub_u32 s0, s6, s2
; GFX9-NEXT:    v_mov_b32_e32 v1, s7
; GFX9-NEXT:    s_subb_u32 s1, s7, s3
; GFX9-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[0:1]
; GFX9-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
; GFX9-NEXT:    s_endpgm
  %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b) #0
  %val = extractvalue { i64, i1 } %usub, 0
  %carry = extractvalue { i64, i1 } %usub, 1
  %ext = zext i1 %carry to i64
  %add2 = add i64 %val, %ext
  store i64 %add2, ptr addrspace(1) %out, align 8
  ret void
}

; FIXME: Could do scalar
define amdgpu_kernel void @s_usubo_i32(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i32 %a, i32 %b) #0 {
; SI-LABEL: s_usubo_i32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0xd
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_mov_b32 s10, s2
; SI-NEXT:    s_mov_b32 s11, s3
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s0, s4
; SI-NEXT:    s_mov_b32 s1, s5
; SI-NEXT:    s_mov_b32 s8, s6
; SI-NEXT:    s_mov_b32 s9, s7
; SI-NEXT:    v_mov_b32_e32 v0, s13
; SI-NEXT:    v_sub_i32_e32 v0, vcc, s12, v0
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT:    buffer_store_byte v1, off, s[8:11], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_usubo_i32:
; 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:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v4, s1
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_sub_u32_e32 v4, vcc, s0, v4
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    v_cndmask_b32_e64 v5, 0, 1, vcc
; VI-NEXT:    flat_store_dword v[0:1], v4
; VI-NEXT:    flat_store_byte v[2:3], v5
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_usubo_i32:
; 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_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v1, s3
; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, s2, v1
; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; GFX9-NEXT:    global_store_dword v0, v1, s[4:5]
; GFX9-NEXT:    global_store_byte v0, v2, s[6:7]
; GFX9-NEXT:    s_endpgm
  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
  %val = extractvalue { i32, i1 } %usub, 0
  %carry = extractvalue { i32, i1 } %usub, 1
  store i32 %val, ptr addrspace(1) %out, align 4
  store i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @v_usubo_i32(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
; SI-LABEL: v_usubo_i32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s4
; SI-NEXT:    s_mov_b32 s13, s5
; SI-NEXT:    s_mov_b32 s16, s6
; SI-NEXT:    s_mov_b32 s17, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    buffer_load_dword v1, off, s[16:19], 0
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s4, s2
; SI-NEXT:    s_mov_b32 s5, s3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    buffer_store_byte v1, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_i32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    flat_load_dword v4, v[0:1]
; VI-NEXT:    flat_load_dword v5, v[2:3]
; 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:    v_mov_b32_e32 v3, s3
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_sub_u32_e32 v4, vcc, v4, v5
; VI-NEXT:    v_cndmask_b32_e64 v5, 0, 1, vcc
; VI-NEXT:    flat_store_dword v[0:1], v4
; VI-NEXT:    flat_store_byte v[2:3], v5
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_i32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v1, v0, s[4:5]
; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, v1, v2
; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX9-NEXT:    global_store_byte v0, v2, s[2:3]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %tid.ext = sext i32 %tid to i64
  %a.gep = getelementptr inbounds i32, ptr addrspace(1) %a.ptr
  %b.gep = getelementptr inbounds i32, ptr addrspace(1) %b.ptr
  %a = load i32, ptr addrspace(1) %a.gep, align 4
  %b = load i32, ptr addrspace(1) %b.gep, align 4
  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
  %val = extractvalue { i32, i1 } %usub, 0
  %carry = extractvalue { i32, i1 } %usub, 1
  store i32 %val, ptr addrspace(1) %out, align 4
  store i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @v_usubo_i32_novcc(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
; SI-LABEL: v_usubo_i32_novcc:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s4
; SI-NEXT:    s_mov_b32 s13, s5
; SI-NEXT:    s_mov_b32 s16, s6
; SI-NEXT:    s_mov_b32 s17, s7
; SI-NEXT:    buffer_load_dword v0, off, s[12:15], 0
; SI-NEXT:    buffer_load_dword v1, off, s[16:19], 0
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s4, s2
; SI-NEXT:    s_mov_b32 s5, s3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    ;;#ASMSTART
; SI-NEXT:    ;;#ASMEND
; SI-NEXT:    buffer_store_byte v1, off, s[4:7], 0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_i32_novcc:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    flat_load_dword v4, v[0:1]
; VI-NEXT:    flat_load_dword v5, v[2:3]
; 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:    v_mov_b32_e32 v3, s3
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_sub_u32_e32 v4, vcc, v4, v5
; VI-NEXT:    v_cndmask_b32_e64 v5, 0, 1, vcc
; VI-NEXT:    flat_store_dword v[0:1], v4
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    ;;#ASMSTART
; VI-NEXT:    ;;#ASMEND
; VI-NEXT:    flat_store_byte v[2:3], v5
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_i32_novcc:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v1, v0, s[4:5]
; GFX9-NEXT:    global_load_dword v2, v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, v1, v2
; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    ;;#ASMSTART
; GFX9-NEXT:    ;;#ASMEND
; GFX9-NEXT:    global_store_byte v0, v2, s[2:3]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %tid.ext = sext i32 %tid to i64
  %a.gep = getelementptr inbounds i32, ptr addrspace(1) %a.ptr
  %b.gep = getelementptr inbounds i32, ptr addrspace(1) %b.ptr
  %a = load i32, ptr addrspace(1) %a.gep, align 4
  %b = load i32, ptr addrspace(1) %b.gep, align 4
  %uadd = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
  %val = extractvalue { i32, i1 } %uadd, 0
  %carry = extractvalue { i32, i1 } %uadd, 1
  store volatile i32 %val, ptr addrspace(1) %out, align 4
  call void asm sideeffect "", "~{vcc}"() #0
  store volatile i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @s_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i64 %a, i64 %b) #0 {
; SI-LABEL: s_usubo_i64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_sub_u32 s6, s4, s6
; SI-NEXT:    s_subb_u32 s7, s5, s7
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s12, s2
; SI-NEXT:    s_mov_b32 s13, s3
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    v_mov_b32_e32 v1, s5
; SI-NEXT:    v_mov_b32_e32 v2, s6
; SI-NEXT:    v_mov_b32_e32 v3, s7
; SI-NEXT:    buffer_store_dwordx2 v[2:3], off, s[8:11], 0
; SI-NEXT:    v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1]
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; SI-NEXT:    buffer_store_byte v0, off, s[12:15], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_usubo_i64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    s_sub_u32 s0, s4, s6
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    s_subb_u32 s1, s5, s7
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    v_mov_b32_e32 v7, s1
; VI-NEXT:    v_cmp_gt_u64_e32 vcc, s[0:1], v[4:5]
; VI-NEXT:    v_mov_b32_e32 v6, s0
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    v_mov_b32_e32 v3, s3
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[6:7]
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; VI-NEXT:    flat_store_byte v[2:3], v0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_usubo_i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v4, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_sub_u32 s6, s4, s6
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NEXT:    s_subb_u32 s7, s5, s7
; GFX9-NEXT:    v_mov_b32_e32 v2, s6
; GFX9-NEXT:    v_cmp_gt_u64_e32 vcc, s[6:7], v[0:1]
; GFX9-NEXT:    v_mov_b32_e32 v3, s7
; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9-NEXT:    global_store_dwordx2 v4, v[2:3], s[0:1]
; GFX9-NEXT:    global_store_byte v4, v0, s[2:3]
; GFX9-NEXT:    s_endpgm
  %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
  %val = extractvalue { i64, i1 } %usub, 0
  %carry = extractvalue { i64, i1 } %usub, 1
  store i64 %val, ptr addrspace(1) %out, align 8
  store i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @v_usubo_i64(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
; SI-LABEL: v_usubo_i64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s4
; SI-NEXT:    s_mov_b32 s13, s5
; SI-NEXT:    s_mov_b32 s16, s6
; SI-NEXT:    s_mov_b32 s17, s7
; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s4, s2
; SI-NEXT:    s_mov_b32 s5, s3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e32 v2, vcc, v0, v2
; SI-NEXT:    v_subb_u32_e32 v3, vcc, v1, v3, vcc
; SI-NEXT:    buffer_store_dwordx2 v[2:3], off, s[8:11], 0
; SI-NEXT:    v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1]
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_i64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    flat_load_dwordx2 v[2:3], v[2:3]
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v6, s2
; VI-NEXT:    v_mov_b32_e32 v7, s3
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_sub_u32_e32 v2, vcc, v0, v2
; VI-NEXT:    v_subb_u32_e32 v3, vcc, v1, v3, vcc
; VI-NEXT:    v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1]
; VI-NEXT:    flat_store_dwordx2 v[4:5], v[2:3]
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; VI-NEXT:    flat_store_byte v[6:7], v0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v4, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v4, s[4:5]
; GFX9-NEXT:    global_load_dwordx2 v[2:3], v4, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_sub_co_u32_e32 v2, vcc, v0, v2
; GFX9-NEXT:    v_subb_co_u32_e32 v3, vcc, v1, v3, vcc
; GFX9-NEXT:    v_cmp_gt_u64_e32 vcc, v[2:3], v[0:1]
; GFX9-NEXT:    global_store_dwordx2 v4, v[2:3], s[0:1]
; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; GFX9-NEXT:    global_store_byte v4, v0, s[2:3]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %tid.ext = sext i32 %tid to i64
  %a.gep = getelementptr inbounds i64, ptr addrspace(1) %a.ptr
  %b.gep = getelementptr inbounds i64, ptr addrspace(1) %b.ptr
  %a = load i64, ptr addrspace(1) %a.gep
  %b = load i64, ptr addrspace(1) %b.gep
  %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
  %val = extractvalue { i64, i1 } %usub, 0
  %carry = extractvalue { i64, i1 } %usub, 1
  store i64 %val, ptr addrspace(1) %out, align 8
  store i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @v_usubo_i16(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
; SI-LABEL: v_usubo_i16:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s4
; SI-NEXT:    s_mov_b32 s13, s5
; SI-NEXT:    s_mov_b32 s16, s6
; SI-NEXT:    s_mov_b32 s17, s7
; SI-NEXT:    buffer_load_ushort v0, off, s[12:15], 0
; SI-NEXT:    buffer_load_ushort v1, off, s[16:19], 0
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s4, s2
; SI-NEXT:    s_mov_b32 s5, s3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v1
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v0
; SI-NEXT:    buffer_store_short v0, off, s[8:11], 0
; SI-NEXT:    v_cmp_ne_u32_e32 vcc, v1, v0
; SI-NEXT:    s_waitcnt expcnt(0)
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_i16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    flat_load_ushort v4, v[0:1]
; VI-NEXT:    flat_load_ushort v5, v[2:3]
; 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:    v_mov_b32_e32 v3, s3
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_sub_u16_e32 v5, v4, v5
; VI-NEXT:    v_cmp_gt_u16_e32 vcc, v5, v4
; VI-NEXT:    flat_store_short v[0:1], v5
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
; VI-NEXT:    flat_store_byte v[2:3], v0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_i16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_ushort v1, v0, s[4:5]
; GFX9-NEXT:    global_load_ushort v2, v0, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_sub_u16_e32 v2, v1, v2
; GFX9-NEXT:    v_cmp_gt_u16_e32 vcc, v2, v1
; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, 1, vcc
; GFX9-NEXT:    global_store_short v0, v2, s[0:1]
; GFX9-NEXT:    global_store_byte v0, v1, s[2:3]
; GFX9-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %tid.ext = sext i32 %tid to i64
  %a.gep = getelementptr inbounds i16, ptr addrspace(1) %a.ptr
  %b.gep = getelementptr inbounds i16, ptr addrspace(1) %b.ptr
  %a = load i16, ptr addrspace(1) %a.gep
  %b = load i16, ptr addrspace(1) %b.gep
  %usub = call { i16, i1 } @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
  %val = extractvalue { i16, i1 } %usub, 0
  %carry = extractvalue { i16, i1 } %usub, 1
  store i16 %val, ptr addrspace(1) %out
  store i1 %carry, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @v_usubo_v2i32(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %aptr, ptr addrspace(1) %bptr) nounwind {
; SI-LABEL: v_usubo_v2i32:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    s_mov_b32 s14, s10
; SI-NEXT:    s_mov_b32 s15, s11
; SI-NEXT:    s_mov_b32 s18, s10
; SI-NEXT:    s_mov_b32 s19, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s12, s4
; SI-NEXT:    s_mov_b32 s13, s5
; SI-NEXT:    s_mov_b32 s16, s6
; SI-NEXT:    s_mov_b32 s17, s7
; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[12:15], 0
; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[16:19], 0
; SI-NEXT:    s_mov_b32 s6, s10
; SI-NEXT:    s_mov_b32 s7, s11
; SI-NEXT:    s_mov_b32 s8, s0
; SI-NEXT:    s_mov_b32 s9, s1
; SI-NEXT:    s_mov_b32 s4, s2
; SI-NEXT:    s_mov_b32 s5, s3
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e32 v1, vcc, v1, v3
; SI-NEXT:    v_cndmask_b32_e64 v3, 0, 1, vcc
; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v2
; SI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; SI-NEXT:    buffer_store_dwordx2 v[2:3], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_v2i32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_mov_b32_e32 v2, s6
; VI-NEXT:    v_mov_b32_e32 v3, s7
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    flat_load_dwordx2 v[2:3], v[2:3]
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v6, s2
; VI-NEXT:    v_mov_b32_e32 v7, s3
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_sub_u32_e32 v1, vcc, v1, v3
; VI-NEXT:    v_cndmask_b32_e64 v3, 0, 1, vcc
; VI-NEXT:    v_sub_u32_e32 v0, vcc, v0, v2
; VI-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; VI-NEXT:    flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT:    flat_store_dwordx2 v[6:7], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_v2i32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v4, 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dwordx2 v[0:1], v4, s[4:5]
; GFX9-NEXT:    global_load_dwordx2 v[2:3], v4, s[6:7]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, v1, v3
; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, 1, vcc
; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, v0, v2
; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
; GFX9-NEXT:    global_store_dwordx2 v4, v[2:3], s[2:3]
; GFX9-NEXT:    s_endpgm
  %a = load <2 x i32>, ptr addrspace(1) %aptr, align 4
  %b = load <2 x i32>, ptr addrspace(1) %bptr, align 4
  %sadd = call { <2 x i32>, <2 x i1> } @llvm.usub.with.overflow.v2i32(<2 x i32> %a, <2 x i32> %b) nounwind
  %val = extractvalue { <2 x i32>, <2 x i1> } %sadd, 0
  %carry = extractvalue { <2 x i32>, <2 x i1> } %sadd, 1
  store <2 x i32> %val, ptr addrspace(1) %out, align 4
  %carry.ext = zext <2 x i1> %carry to <2 x i32>
  store <2 x i32> %carry.ext, ptr addrspace(1) %carryout
  ret void
}

define amdgpu_kernel void @s_usubo_clamp_bit(ptr addrspace(1) %out, ptr addrspace(1) %carryout, i32 %a, i32 %b) #0 {
; SI-LABEL: s_usubo_clamp_bit:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_mov_b32_e32 v0, s3
; SI-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
; SI-NEXT:    s_cmp_eq_u32 s2, s3
; SI-NEXT:    s_mov_b64 s[2:3], 0
; SI-NEXT:    s_cbranch_scc1 .LBB8_2
; SI-NEXT:  ; %bb.1: ; %if
; SI-NEXT:    s_xor_b64 s[2:3], vcc, -1
; SI-NEXT:  .LBB8_2: ; %exit
; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s11, 0xf000
; SI-NEXT:    s_mov_b32 s10, -1
; SI-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[2:3]
; SI-NEXT:    s_mov_b32 s2, s10
; SI-NEXT:    s_mov_b32 s3, s11
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s8, s4
; SI-NEXT:    s_mov_b32 s9, s5
; SI-NEXT:    s_mov_b32 s0, s6
; SI-NEXT:    s_mov_b32 s1, s7
; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
; SI-NEXT:    buffer_store_byte v1, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_usubo_clamp_bit:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    s_cmp_eq_u32 s2, s3
; VI-NEXT:    v_sub_u32_e32 v0, vcc, s2, v0
; VI-NEXT:    s_mov_b64 s[2:3], 0
; VI-NEXT:    s_cbranch_scc1 .LBB8_2
; VI-NEXT:  ; %bb.1: ; %if
; VI-NEXT:    s_xor_b64 s[2:3], vcc, -1
; VI-NEXT:  .LBB8_2: ; %exit
; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; VI-NEXT:    v_cndmask_b32_e64 v5, 0, 1, s[2:3]
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_mov_b32_e32 v3, s6
; VI-NEXT:    v_mov_b32_e32 v4, s7
; VI-NEXT:    flat_store_dword v[1:2], v0
; VI-NEXT:    flat_store_byte v[3:4], v5
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: s_usubo_clamp_bit:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v0, s3
; GFX9-NEXT:    s_cmp_eq_u32 s2, s3
; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s2, v0
; GFX9-NEXT:    s_mov_b64 s[2:3], 0
; GFX9-NEXT:    s_cbranch_scc1 .LBB8_2
; GFX9-NEXT:  ; %bb.1: ; %if
; GFX9-NEXT:    s_xor_b64 s[2:3], vcc, -1
; GFX9-NEXT:  .LBB8_2: ; %exit
; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, 1, s[2:3]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_store_dword v1, v0, s[4:5]
; GFX9-NEXT:    global_store_byte v1, v2, s[6:7]
; GFX9-NEXT:    s_endpgm
entry:
  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
  %val = extractvalue { i32, i1 } %usub, 0
  %carry = extractvalue { i32, i1 } %usub, 1
  %c2 = icmp eq i1 %carry, false
  %cc = icmp eq i32 %a, %b
  br i1 %cc, label %exit, label %if

if:
  br label %exit

exit:
  %cout = phi i1 [false, %entry], [%c2, %if]
  store i32 %val, ptr addrspace(1) %out, align 4
  store i1 %cout, ptr addrspace(1) %carryout
  ret void
}


define amdgpu_kernel void @v_usubo_clamp_bit(ptr addrspace(1) %out, ptr addrspace(1) %carryout, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
; SI-LABEL: v_usubo_clamp_bit:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_mov_b32 s14, s2
; SI-NEXT:    s_mov_b32 s15, s3
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s0, s8
; SI-NEXT:    s_mov_b32 s1, s9
; SI-NEXT:    s_mov_b32 s12, s10
; SI-NEXT:    s_mov_b32 s13, s11
; SI-NEXT:    buffer_load_dword v1, off, s[0:3], 0
; SI-NEXT:    buffer_load_dword v2, off, s[12:15], 0
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_sub_i32_e64 v0, s[0:1], v1, v2
; SI-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
; SI-NEXT:    s_mov_b64 s[8:9], 0
; SI-NEXT:    s_cbranch_vccnz .LBB9_2
; SI-NEXT:  ; %bb.1: ; %if
; SI-NEXT:    s_xor_b64 s[8:9], s[0:1], -1
; SI-NEXT:  .LBB9_2: ; %exit
; SI-NEXT:    s_mov_b32 s0, s4
; SI-NEXT:    s_mov_b32 s1, s5
; SI-NEXT:    s_mov_b32 s4, s6
; SI-NEXT:    s_mov_b32 s5, s7
; SI-NEXT:    s_mov_b32 s6, s2
; SI-NEXT:    s_mov_b32 s7, s3
; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT:    s_waitcnt expcnt(0)
; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[8:9]
; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_usubo_clamp_bit:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
; VI-NEXT:    s_mov_b64 s[2:3], 0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s8
; VI-NEXT:    v_mov_b32_e32 v1, s9
; VI-NEXT:    v_mov_b32_e32 v2, s10
; VI-NEXT:    v_mov_b32_e32 v3, s11
; VI-NEXT:    flat_load_dword v1, v[0:1]
; VI-NEXT:    flat_load_dword v2, v[2:3]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
; VI-NEXT:    v_sub_u32_e64 v0, s[0:1], v1, v2
; VI-NEXT:    s_cbranch_vccnz .LBB9_2
; VI-NEXT:  ; %bb.1: ; %if
; VI-NEXT:    s_xor_b64 s[2:3], s[0:1], -1
; VI-NEXT:  .LBB9_2: ; %exit
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_mov_b32_e32 v3, s6
; VI-NEXT:    v_mov_b32_e32 v4, s7
; VI-NEXT:    flat_store_dword v[1:2], v0
; VI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[2:3]
; VI-NEXT:    flat_store_byte v[3:4], v0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: v_usubo_clamp_bit:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x24
; GFX9-NEXT:    v_mov_b32_e32 v0, 0
; GFX9-NEXT:    s_mov_b64 s[2:3], 0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    global_load_dword v2, v0, s[8:9]
; GFX9-NEXT:    global_load_dword v3, v0, s[10:11]
; GFX9-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v3
; GFX9-NEXT:    v_sub_co_u32_e64 v1, s[0:1], v2, v3
; GFX9-NEXT:    s_cbranch_vccnz .LBB9_2
; GFX9-NEXT:  ; %bb.1: ; %if
; GFX9-NEXT:    s_xor_b64 s[2:3], s[0:1], -1
; GFX9-NEXT:  .LBB9_2: ; %exit
; GFX9-NEXT:    global_store_dword v0, v1, s[4:5]
; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s[2:3]
; GFX9-NEXT:    global_store_byte v0, v1, s[6:7]
; GFX9-NEXT:    s_endpgm
entry:
  %tid = call i32 @llvm.amdgcn.workitem.id.x()
  %tid.ext = sext i32 %tid to i64
  %a.gep = getelementptr inbounds i32, ptr addrspace(1) %a.ptr
  %b.gep = getelementptr inbounds i32, ptr addrspace(1) %b.ptr
  %a = load i32, ptr addrspace(1) %a.gep, align 4
  %b = load i32, ptr addrspace(1) %b.gep, align 4
  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
  %val = extractvalue { i32, i1 } %usub, 0
  %carry = extractvalue { i32, i1 } %usub, 1
  %c2 = icmp eq i1 %carry, false
  %cc = icmp eq i32 %a, %b
  br i1 %cc, label %exit, label %if

if:
  br label %exit

exit:
  %cout = phi i1 [false, %entry], [%c2, %if]
  store i32 %val, ptr addrspace(1) %out, align 4
  store i1 %cout, ptr addrspace(1) %carryout
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x() #1
declare { i16, i1 } @llvm.usub.with.overflow.i16(i16, i16) #1
declare { i32, i1 } @llvm.usub.with.overflow.i32(i32, i32) #1
declare { i64, i1 } @llvm.usub.with.overflow.i64(i64, i64) #1
declare { <2 x i32>, <2 x i1> } @llvm.usub.with.overflow.v2i32(<2 x i32>, <2 x i32>) nounwind readnone

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }