; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s
; RUN: llc -mtriple=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX7 %s
; RUN: llc -mtriple=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX8 %s

define float @lds_atomic_fadd_ret_f32(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_ret_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, 4.0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_add_rtn_f32 v0, v0, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_ret_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v1, 4.0
; GFX9-NEXT:    ds_add_rtn_f32 v0, v0, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_ret_f32:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v1, v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB0_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v2, v1
; GFX7-NEXT:    v_add_f32_e32 v1, 4.0, v2
; GFX7-NEXT:    ds_cmpst_rtn_b32 v1, v0, v2, v1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB0_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v0, v1
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_ret_f32:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v1, v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB0_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v2, v1
; GFX8-NEXT:    v_add_f32_e32 v1, 4.0, v2
; GFX8-NEXT:    ds_cmpst_rtn_b32 v1, v0, v2, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v2
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB0_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v0, v1
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, float 4.0 seq_cst
  ret float %result
}

define void @lds_atomic_fadd_noret_f32(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_noret_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, 4.0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_add_f32 v0, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_noret_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v1, 4.0
; GFX9-NEXT:    ds_add_f32 v0, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_noret_f32:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v1, v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB1_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_add_f32_e32 v2, 4.0, v1
; GFX7-NEXT:    ds_cmpst_rtn_b32 v2, v0, v1, v2
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v1, v2
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB1_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_noret_f32:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v1, v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB1_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_add_f32_e32 v2, 4.0, v1
; GFX8-NEXT:    ds_cmpst_rtn_b32 v2, v0, v1, v2
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v1
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v1, v2
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB1_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, float 4.0 seq_cst
  ret void
}

define amdgpu_kernel void @lds_ds_fadd(ptr addrspace(1) %out, ptr addrspace(3) %ptrf, i32 %idx) {
; VI-LABEL: lds_ds_fadd:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
; VI-NEXT:    s_mov_b64 s[4:5], exec
; VI-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_add_i32 s3, s3, 4
; VI-NEXT:    ; implicit-def: $vgpr1
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_and_saveexec_b64 s[6:7], vcc
; VI-NEXT:    s_cbranch_execz .LBB2_2
; VI-NEXT:  ; %bb.1:
; VI-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; VI-NEXT:    s_lshl_b32 s8, s3, 3
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, s4
; VI-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; VI-NEXT:    v_mov_b32_e32 v2, s8
; VI-NEXT:    ds_add_rtn_f32 v1, v2, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:  .LBB2_2:
; VI-NEXT:    s_or_b64 exec, exec, s[6:7]
; VI-NEXT:    s_mov_b64 s[6:7], exec
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_readfirstlane_b32 s8, v1
; VI-NEXT:    v_mbcnt_lo_u32_b32 v1, s6, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v1, s7, v1
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; VI-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT:    s_cbranch_execz .LBB2_4
; VI-NEXT:  ; %bb.3:
; VI-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, s6
; VI-NEXT:    s_lshl_b32 s3, s3, 4
; VI-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; VI-NEXT:    v_mov_b32_e32 v2, s3
; VI-NEXT:    ds_add_f32 v2, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:  .LBB2_4:
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; VI-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; VI-NEXT:    s_mov_b64 s[4:5], exec
; VI-NEXT:    v_add_f32_e32 v2, s8, v0
; VI-NEXT:    v_bfrev_b32_e32 v1, 1
; VI-NEXT:    ; implicit-def: $vgpr0
; VI-NEXT:  .LBB2_5: ; %ComputeLoop
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_ff1_i32_b64 s3, s[4:5]
; VI-NEXT:    s_lshl_b64 s[6:7], 1, s3
; VI-NEXT:    v_readfirstlane_b32 s8, v1
; VI-NEXT:    v_readlane_b32 s9, v2, s3
; VI-NEXT:    s_mov_b32 m0, s3
; VI-NEXT:    s_andn2_b64 s[4:5], s[4:5], s[6:7]
; VI-NEXT:    v_writelane_b32 v0, s8, m0
; VI-NEXT:    s_cmp_lg_u64 s[4:5], 0
; VI-NEXT:    v_add_f32_e32 v1, s9, v1
; VI-NEXT:    s_cbranch_scc1 .LBB2_5
; VI-NEXT:  ; %bb.6: ; %ComputeEnd
; VI-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v2
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
; VI-NEXT:    ; implicit-def: $vgpr2
; VI-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT:    s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT:    s_cbranch_execz .LBB2_8
; VI-NEXT:  ; %bb.7:
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    ds_add_rtn_f32 v2, v2, v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:  .LBB2_8:
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_readfirstlane_b32 s2, v2
; VI-NEXT:    v_add_f32_e32 v2, s2, v0
; 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: lds_ds_fadd:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_add_i32 s3, s3, 4
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[6:7], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB2_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_lshl_b32 s8, s3, 3
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, s4
; GFX9-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s8
; GFX9-NEXT:    ds_add_rtn_f32 v1, v2, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:  .LBB2_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX9-NEXT:    s_mov_b64 s[6:7], exec
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v1, s6, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v1, s7, v1
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB2_4
; GFX9-NEXT:  ; %bb.3:
; GFX9-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, s6
; GFX9-NEXT:    s_lshl_b32 s3, s3, 4
; GFX9-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-NEXT:    ds_add_f32 v2, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:  .LBB2_4:
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX9-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_add_f32_e32 v2, s8, v0
; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
; GFX9-NEXT:    ; implicit-def: $vgpr0
; GFX9-NEXT:  .LBB2_5: ; %ComputeLoop
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_ff1_i32_b64 s3, s[4:5]
; GFX9-NEXT:    s_lshl_b64 s[6:7], 1, s3
; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
; GFX9-NEXT:    v_readlane_b32 s9, v2, s3
; GFX9-NEXT:    s_mov_b32 m0, s3
; GFX9-NEXT:    s_andn2_b64 s[4:5], s[4:5], s[6:7]
; GFX9-NEXT:    v_writelane_b32 v0, s8, m0
; GFX9-NEXT:    s_cmp_lg_u64 s[4:5], 0
; GFX9-NEXT:    v_add_f32_e32 v1, s9, v1
; GFX9-NEXT:    s_cbranch_scc1 .LBB2_5
; GFX9-NEXT:  ; %bb.6: ; %ComputeEnd
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v2
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
; GFX9-NEXT:    ; implicit-def: $vgpr2
; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT:    s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execz .LBB2_8
; GFX9-NEXT:  ; %bb.7:
; GFX9-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    ds_add_rtn_f32 v2, v2, v1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:  .LBB2_8:
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v2
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    v_add_f32_e32 v0, s2, v0
; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX7-LABEL: lds_ds_fadd:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
; GFX7-NEXT:    s_mov_b64 s[6:7], exec
; GFX7-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s6, 0
; GFX7-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s7, v0
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_add_i32 s3, s3, 4
; GFX7-NEXT:    ; implicit-def: $vgpr1
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX7-NEXT:    s_cbranch_execz .LBB2_4
; GFX7-NEXT:  ; %bb.1:
; GFX7-NEXT:    s_lshl_b32 s8, s3, 3
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    ds_read_b32 v1, v2
; GFX7-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v3, s6
; GFX7-NEXT:    v_mul_f32_e32 v3, 0x42280000, v3
; GFX7-NEXT:    s_mov_b64 s[6:7], 0
; GFX7-NEXT:  .LBB2_2: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v4, v1
; GFX7-NEXT:    v_add_f32_e32 v1, v4, v3
; GFX7-NEXT:    ds_cmpst_rtn_b32 v1, v2, v4, v1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v4
; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX7-NEXT:    s_cbranch_execnz .LBB2_2
; GFX7-NEXT:  ; %bb.3: ; %Flow15
; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX7-NEXT:  .LBB2_4: ; %Flow16
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_mov_b64 s[6:7], exec
; GFX7-NEXT:    v_readfirstlane_b32 s8, v1
; GFX7-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, s6, 0
; GFX7-NEXT:    v_mbcnt_hi_u32_b32_e32 v1, s7, v1
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX7-NEXT:    s_cbranch_execz .LBB2_7
; GFX7-NEXT:  ; %bb.5:
; GFX7-NEXT:    s_lshl_b32 s3, s3, 4
; GFX7-NEXT:    v_mov_b32_e32 v1, s3
; GFX7-NEXT:    ds_read_b32 v3, v1
; GFX7-NEXT:    s_bcnt1_i32_b64 s3, s[6:7]
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v2, s3
; GFX7-NEXT:    v_mul_f32_e32 v2, 0x42280000, v2
; GFX7-NEXT:    s_mov_b64 s[6:7], 0
; GFX7-NEXT:  .LBB2_6: ; %atomicrmw.start2
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_add_f32_e32 v4, v3, v2
; GFX7-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX7-NEXT:    v_mov_b32_e32 v3, v4
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX7-NEXT:    s_cbranch_execnz .LBB2_6
; GFX7-NEXT:  .LBB2_7: ; %Flow14
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v2, s2
; GFX7-NEXT:    ds_read_b32 v1, v2
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX7-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX7-NEXT:    v_add_f32_e32 v0, s8, v0
; GFX7-NEXT:    s_mov_b64 s[2:3], 0
; GFX7-NEXT:  .LBB2_8: ; %atomicrmw.start8
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v3, v1
; GFX7-NEXT:    v_add_f32_e32 v1, v3, v0
; GFX7-NEXT:    ds_cmpst_rtn_b32 v1, v2, v3, v1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v3
; GFX7-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GFX7-NEXT:    s_cbranch_execnz .LBB2_8
; GFX7-NEXT:  ; %bb.9: ; %atomicrmw.end7
; GFX7-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: lds_ds_fadd:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
; GFX8-NEXT:    s_mov_b64 s[6:7], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s6, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s7, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_add_i32 s3, s3, 4
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB2_4
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_lshl_b32 s8, s3, 3
; GFX8-NEXT:    v_mov_b32_e32 v2, s8
; GFX8-NEXT:    ds_read_b32 v1, v2
; GFX8-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v3, s6
; GFX8-NEXT:    v_mul_f32_e32 v3, 0x42280000, v3
; GFX8-NEXT:    s_mov_b64 s[6:7], 0
; GFX8-NEXT:  .LBB2_2: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v4, v1
; GFX8-NEXT:    v_add_f32_e32 v1, v4, v3
; GFX8-NEXT:    ds_cmpst_rtn_b32 v1, v2, v4, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v4
; GFX8-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX8-NEXT:    s_cbranch_execnz .LBB2_2
; GFX8-NEXT:  ; %bb.3: ; %Flow17
; GFX8-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX8-NEXT:  .LBB2_4: ; %Flow18
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_mov_b64 s[6:7], exec
; GFX8-NEXT:    v_readfirstlane_b32 s8, v1
; GFX8-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, s6, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32_e32 v1, s7, v1
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB2_7
; GFX8-NEXT:  ; %bb.5:
; GFX8-NEXT:    s_lshl_b32 s3, s3, 4
; GFX8-NEXT:    v_mov_b32_e32 v1, s3
; GFX8-NEXT:    ds_read_b32 v3, v1
; GFX8-NEXT:    s_bcnt1_i32_b64 s3, s[6:7]
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v2, s3
; GFX8-NEXT:    v_mul_f32_e32 v2, 0x42280000, v2
; GFX8-NEXT:    s_mov_b64 s[6:7], 0
; GFX8-NEXT:  .LBB2_6: ; %atomicrmw.start2
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_add_f32_e32 v4, v3, v2
; GFX8-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX8-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX8-NEXT:    v_mov_b32_e32 v3, v4
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX8-NEXT:    s_cbranch_execnz .LBB2_6
; GFX8-NEXT:  .LBB2_7: ; %Flow16
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v2, s2
; GFX8-NEXT:    ds_read_b32 v1, v2
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX8-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX8-NEXT:    v_add_f32_e32 v0, s8, v0
; GFX8-NEXT:    s_mov_b64 s[2:3], 0
; GFX8-NEXT:  .LBB2_8: ; %atomicrmw.start8
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v3, v1
; GFX8-NEXT:    v_add_f32_e32 v1, v3, v0
; GFX8-NEXT:    ds_cmpst_rtn_b32 v1, v2, v3, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v3
; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_cbranch_execnz .LBB2_8
; GFX8-NEXT:  ; %bb.9: ; %atomicrmw.end7
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX8-NEXT:    s_mov_b32 s3, 0xf000
; GFX8-NEXT:    s_mov_b32 s2, -1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX8-NEXT:    s_endpgm
  %idx.add = add nuw i32 %idx, 4
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  %a1 = atomicrmw fadd ptr addrspace(3) %ptr0, float 4.2e+1 seq_cst
  %a2 = atomicrmw fadd ptr addrspace(3) %ptr1, float 4.2e+1 seq_cst
  %a3 = atomicrmw fadd ptr addrspace(3) %ptrf, float %a1 seq_cst
  store float %a3, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @lds_ds_fadd_one_as(ptr addrspace(1) %out, ptr addrspace(3) %ptrf, i32 %idx) {
; VI-LABEL: lds_ds_fadd_one_as:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
; VI-NEXT:    s_mov_b64 s[4:5], exec
; VI-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_add_i32 s3, s3, 4
; VI-NEXT:    ; implicit-def: $vgpr1
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    s_and_saveexec_b64 s[6:7], vcc
; VI-NEXT:    s_cbranch_execz .LBB3_2
; VI-NEXT:  ; %bb.1:
; VI-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; VI-NEXT:    s_lshl_b32 s8, s3, 3
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, s4
; VI-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; VI-NEXT:    v_mov_b32_e32 v2, s8
; VI-NEXT:    ds_add_rtn_f32 v1, v2, v1
; VI-NEXT:  .LBB3_2:
; VI-NEXT:    s_or_b64 exec, exec, s[6:7]
; VI-NEXT:    s_mov_b64 s[6:7], exec
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_readfirstlane_b32 s8, v1
; VI-NEXT:    v_mbcnt_lo_u32_b32 v1, s6, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v1, s7, v1
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; VI-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT:    s_cbranch_execz .LBB3_4
; VI-NEXT:  ; %bb.3:
; VI-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, s6
; VI-NEXT:    s_lshl_b32 s3, s3, 4
; VI-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; VI-NEXT:    v_mov_b32_e32 v2, s3
; VI-NEXT:    ds_add_f32 v2, v1
; VI-NEXT:  .LBB3_4:
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; VI-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; VI-NEXT:    s_mov_b64 s[4:5], exec
; VI-NEXT:    v_add_f32_e32 v2, s8, v0
; VI-NEXT:    v_bfrev_b32_e32 v1, 1
; VI-NEXT:    ; implicit-def: $vgpr0
; VI-NEXT:  .LBB3_5: ; %ComputeLoop
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_ff1_i32_b64 s3, s[4:5]
; VI-NEXT:    s_lshl_b64 s[6:7], 1, s3
; VI-NEXT:    v_readfirstlane_b32 s8, v1
; VI-NEXT:    v_readlane_b32 s9, v2, s3
; VI-NEXT:    s_mov_b32 m0, s3
; VI-NEXT:    s_andn2_b64 s[4:5], s[4:5], s[6:7]
; VI-NEXT:    v_writelane_b32 v0, s8, m0
; VI-NEXT:    s_cmp_lg_u64 s[4:5], 0
; VI-NEXT:    v_add_f32_e32 v1, s9, v1
; VI-NEXT:    s_cbranch_scc1 .LBB3_5
; VI-NEXT:  ; %bb.6: ; %ComputeEnd
; VI-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
; VI-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v2
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
; VI-NEXT:    ; implicit-def: $vgpr2
; VI-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; VI-NEXT:    s_xor_b64 s[4:5], exec, s[4:5]
; VI-NEXT:    s_cbranch_execz .LBB3_8
; VI-NEXT:  ; %bb.7:
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_add_rtn_f32 v2, v2, v1
; VI-NEXT:  .LBB3_8:
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_readfirstlane_b32 s2, v2
; VI-NEXT:    v_add_f32_e32 v2, s2, v0
; 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: lds_ds_fadd_one_as:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_add_i32 s3, s3, 4
; GFX9-NEXT:    ; implicit-def: $vgpr1
; GFX9-NEXT:    s_and_saveexec_b64 s[6:7], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB3_2
; GFX9-NEXT:  ; %bb.1:
; GFX9-NEXT:    s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT:    s_lshl_b32 s8, s3, 3
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, s4
; GFX9-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s8
; GFX9-NEXT:    ds_add_rtn_f32 v1, v2, v1
; GFX9-NEXT:  .LBB3_2:
; GFX9-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX9-NEXT:    s_mov_b64 s[6:7], exec
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v1, s6, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v1, s7, v1
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT:    s_cbranch_execz .LBB3_4
; GFX9-NEXT:  ; %bb.3:
; GFX9-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, s6
; GFX9-NEXT:    s_lshl_b32 s3, s3, 4
; GFX9-NEXT:    v_mul_f32_e32 v1, 0x42280000, v1
; GFX9-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-NEXT:    ds_add_f32 v2, v1
; GFX9-NEXT:  .LBB3_4:
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX9-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX9-NEXT:    s_mov_b64 s[4:5], exec
; GFX9-NEXT:    v_add_f32_e32 v2, s8, v0
; GFX9-NEXT:    v_bfrev_b32_e32 v1, 1
; GFX9-NEXT:    ; implicit-def: $vgpr0
; GFX9-NEXT:  .LBB3_5: ; %ComputeLoop
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_ff1_i32_b64 s3, s[4:5]
; GFX9-NEXT:    s_lshl_b64 s[6:7], 1, s3
; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
; GFX9-NEXT:    v_readlane_b32 s9, v2, s3
; GFX9-NEXT:    s_mov_b32 m0, s3
; GFX9-NEXT:    s_andn2_b64 s[4:5], s[4:5], s[6:7]
; GFX9-NEXT:    v_writelane_b32 v0, s8, m0
; GFX9-NEXT:    s_cmp_lg_u64 s[4:5], 0
; GFX9-NEXT:    v_add_f32_e32 v1, s9, v1
; GFX9-NEXT:    s_cbranch_scc1 .LBB3_5
; GFX9-NEXT:  ; %bb.6: ; %ComputeEnd
; GFX9-NEXT:    v_mbcnt_lo_u32_b32 v2, exec_lo, 0
; GFX9-NEXT:    v_mbcnt_hi_u32_b32 v2, exec_hi, v2
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v2
; GFX9-NEXT:    ; implicit-def: $vgpr2
; GFX9-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX9-NEXT:    s_xor_b64 s[4:5], exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execz .LBB3_8
; GFX9-NEXT:  ; %bb.7:
; GFX9-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NEXT:    ds_add_rtn_f32 v2, v2, v1
; GFX9-NEXT:  .LBB3_8:
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_readfirstlane_b32 s2, v2
; GFX9-NEXT:    v_mov_b32_e32 v1, 0
; GFX9-NEXT:    v_add_f32_e32 v0, s2, v0
; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX9-NEXT:    s_endpgm
;
; GFX7-LABEL: lds_ds_fadd_one_as:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
; GFX7-NEXT:    s_mov_b64 s[6:7], exec
; GFX7-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s6, 0
; GFX7-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s7, v0
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_add_i32 s3, s3, 4
; GFX7-NEXT:    ; implicit-def: $vgpr1
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX7-NEXT:    s_cbranch_execz .LBB3_4
; GFX7-NEXT:  ; %bb.1:
; GFX7-NEXT:    s_lshl_b32 s8, s3, 3
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    ds_read_b32 v1, v2
; GFX7-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v3, s6
; GFX7-NEXT:    v_mul_f32_e32 v3, 0x42280000, v3
; GFX7-NEXT:    s_mov_b64 s[6:7], 0
; GFX7-NEXT:  .LBB3_2: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v4, v1
; GFX7-NEXT:    v_add_f32_e32 v1, v4, v3
; GFX7-NEXT:    ds_cmpst_rtn_b32 v1, v2, v4, v1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v4
; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX7-NEXT:    s_cbranch_execnz .LBB3_2
; GFX7-NEXT:  ; %bb.3: ; %Flow15
; GFX7-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX7-NEXT:  .LBB3_4: ; %Flow16
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_mov_b64 s[6:7], exec
; GFX7-NEXT:    v_readfirstlane_b32 s8, v1
; GFX7-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, s6, 0
; GFX7-NEXT:    v_mbcnt_hi_u32_b32_e32 v1, s7, v1
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX7-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX7-NEXT:    s_cbranch_execz .LBB3_7
; GFX7-NEXT:  ; %bb.5:
; GFX7-NEXT:    s_lshl_b32 s3, s3, 4
; GFX7-NEXT:    v_mov_b32_e32 v1, s3
; GFX7-NEXT:    ds_read_b32 v3, v1
; GFX7-NEXT:    s_bcnt1_i32_b64 s3, s[6:7]
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v2, s3
; GFX7-NEXT:    v_mul_f32_e32 v2, 0x42280000, v2
; GFX7-NEXT:    s_mov_b64 s[6:7], 0
; GFX7-NEXT:  .LBB3_6: ; %atomicrmw.start2
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_add_f32_e32 v4, v3, v2
; GFX7-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX7-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX7-NEXT:    v_mov_b32_e32 v3, v4
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX7-NEXT:    s_cbranch_execnz .LBB3_6
; GFX7-NEXT:  .LBB3_7: ; %Flow14
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v2, s2
; GFX7-NEXT:    ds_read_b32 v1, v2
; GFX7-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX7-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX7-NEXT:    v_add_f32_e32 v0, s8, v0
; GFX7-NEXT:    s_mov_b64 s[2:3], 0
; GFX7-NEXT:  .LBB3_8: ; %atomicrmw.start8
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v3, v1
; GFX7-NEXT:    v_add_f32_e32 v1, v3, v0
; GFX7-NEXT:    ds_cmpst_rtn_b32 v1, v2, v3, v1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v3
; GFX7-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GFX7-NEXT:    s_cbranch_execnz .LBB3_8
; GFX7-NEXT:  ; %bb.9: ; %atomicrmw.end7
; GFX7-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: lds_ds_fadd_one_as:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
; GFX8-NEXT:    s_mov_b64 s[6:7], exec
; GFX8-NEXT:    v_mbcnt_lo_u32_b32_e64 v0, s6, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32_e32 v0, s7, v0
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_add_i32 s3, s3, 4
; GFX8-NEXT:    ; implicit-def: $vgpr1
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB3_4
; GFX8-NEXT:  ; %bb.1:
; GFX8-NEXT:    s_lshl_b32 s8, s3, 3
; GFX8-NEXT:    v_mov_b32_e32 v2, s8
; GFX8-NEXT:    ds_read_b32 v1, v2
; GFX8-NEXT:    s_bcnt1_i32_b64 s6, s[6:7]
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v3, s6
; GFX8-NEXT:    v_mul_f32_e32 v3, 0x42280000, v3
; GFX8-NEXT:    s_mov_b64 s[6:7], 0
; GFX8-NEXT:  .LBB3_2: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v4, v1
; GFX8-NEXT:    v_add_f32_e32 v1, v4, v3
; GFX8-NEXT:    ds_cmpst_rtn_b32 v1, v2, v4, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v4
; GFX8-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX8-NEXT:    s_cbranch_execnz .LBB3_2
; GFX8-NEXT:  ; %bb.3: ; %Flow17
; GFX8-NEXT:    s_or_b64 exec, exec, s[6:7]
; GFX8-NEXT:  .LBB3_4: ; %Flow18
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_mov_b64 s[6:7], exec
; GFX8-NEXT:    v_readfirstlane_b32 s8, v1
; GFX8-NEXT:    v_mbcnt_lo_u32_b32_e64 v1, s6, 0
; GFX8-NEXT:    v_mbcnt_hi_u32_b32_e32 v1, s7, v1
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
; GFX8-NEXT:    s_and_saveexec_b64 s[4:5], vcc
; GFX8-NEXT:    s_cbranch_execz .LBB3_7
; GFX8-NEXT:  ; %bb.5:
; GFX8-NEXT:    s_lshl_b32 s3, s3, 4
; GFX8-NEXT:    v_mov_b32_e32 v1, s3
; GFX8-NEXT:    ds_read_b32 v3, v1
; GFX8-NEXT:    s_bcnt1_i32_b64 s3, s[6:7]
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v2, s3
; GFX8-NEXT:    v_mul_f32_e32 v2, 0x42280000, v2
; GFX8-NEXT:    s_mov_b64 s[6:7], 0
; GFX8-NEXT:  .LBB3_6: ; %atomicrmw.start2
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_add_f32_e32 v4, v3, v2
; GFX8-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX8-NEXT:    s_or_b64 s[6:7], vcc, s[6:7]
; GFX8-NEXT:    v_mov_b32_e32 v3, v4
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[6:7]
; GFX8-NEXT:    s_cbranch_execnz .LBB3_6
; GFX8-NEXT:  .LBB3_7: ; %Flow16
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v2, s2
; GFX8-NEXT:    ds_read_b32 v1, v2
; GFX8-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
; GFX8-NEXT:    v_mul_f32_e32 v0, 0x42280000, v0
; GFX8-NEXT:    v_add_f32_e32 v0, s8, v0
; GFX8-NEXT:    s_mov_b64 s[2:3], 0
; GFX8-NEXT:  .LBB3_8: ; %atomicrmw.start8
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v3, v1
; GFX8-NEXT:    v_add_f32_e32 v1, v3, v0
; GFX8-NEXT:    ds_cmpst_rtn_b32 v1, v2, v3, v1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v1, v3
; GFX8-NEXT:    s_or_b64 s[2:3], vcc, s[2:3]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_cbranch_execnz .LBB3_8
; GFX8-NEXT:  ; %bb.9: ; %atomicrmw.end7
; GFX8-NEXT:    s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX8-NEXT:    s_mov_b32 s3, 0xf000
; GFX8-NEXT:    s_mov_b32 s2, -1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    buffer_store_dword v1, off, s[0:3], 0
; GFX8-NEXT:    s_endpgm
  %idx.add = add nuw i32 %idx, 4
  %shl0 = shl i32 %idx.add, 3
  %shl1 = shl i32 %idx.add, 4
  %ptr0 = inttoptr i32 %shl0 to ptr addrspace(3)
  %ptr1 = inttoptr i32 %shl1 to ptr addrspace(3)
  %a1 = atomicrmw fadd ptr addrspace(3) %ptr0, float 4.2e+1 syncscope("one-as") seq_cst
  %a2 = atomicrmw fadd ptr addrspace(3) %ptr1, float 4.2e+1 syncscope("one-as") seq_cst
  %a3 = atomicrmw fadd ptr addrspace(3) %ptrf, float %a1 syncscope("one-as") seq_cst
  store float %a3, ptr addrspace(1) %out
  ret void
}

define double @lds_atomic_fadd_ret_f64(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_ret_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    v_mov_b32_e32 v2, v0
; VI-NEXT:    ds_read_b64 v[0:1], v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB4_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v4, v1
; VI-NEXT:    v_mov_b32_e32 v3, v0
; VI-NEXT:    v_add_f64 v[0:1], v[3:4], 4.0
; VI-NEXT:    ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1]
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4]
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB4_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_ret_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v2, v0
; GFX9-NEXT:    ds_read_b64 v[0:1], v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB4_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v4, v1
; GFX9-NEXT:    v_mov_b32_e32 v3, v0
; GFX9-NEXT:    v_add_f64 v[0:1], v[3:4], 4.0
; GFX9-NEXT:    ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4]
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB4_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_ret_f64:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    v_mov_b32_e32 v2, v0
; GFX7-NEXT:    ds_read_b64 v[0:1], v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB4_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v4, v1
; GFX7-NEXT:    v_mov_b32_e32 v3, v0
; GFX7-NEXT:    v_add_f64 v[0:1], v[3:4], 4.0
; GFX7-NEXT:    ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1]
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4]
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB4_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_ret_f64:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    v_mov_b32_e32 v2, v0
; GFX8-NEXT:    ds_read_b64 v[0:1], v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB4_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v4, v1
; GFX8-NEXT:    v_mov_b32_e32 v3, v0
; GFX8-NEXT:    v_add_f64 v[0:1], v[3:4], 4.0
; GFX8-NEXT:    ds_cmpst_rtn_b64 v[0:1], v2, v[3:4], v[0:1]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u64_e32 vcc, v[0:1], v[3:4]
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB4_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
  ret double %result
}

define void @lds_atomic_fadd_noret_f64(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_noret_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b64 v[1:2], v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB5_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_add_f64 v[3:4], v[1:2], 4.0
; VI-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4]
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2]
; VI-NEXT:    v_mov_b32_e32 v1, v3
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v2, v4
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB5_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_noret_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    ds_read_b64 v[1:2], v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB5_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_add_f64 v[3:4], v[1:2], 4.0
; GFX9-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2]
; GFX9-NEXT:    v_mov_b32_e32 v1, v3
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v2, v4
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB5_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_noret_f64:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b64 v[1:2], v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB5_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_add_f64 v[3:4], v[1:2], 4.0
; GFX7-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4]
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2]
; GFX7-NEXT:    v_mov_b32_e32 v1, v3
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v2, v4
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB5_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_noret_f64:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b64 v[1:2], v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB5_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_add_f64 v[3:4], v[1:2], 4.0
; GFX8-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[1:2], v[3:4]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[1:2]
; GFX8-NEXT:    v_mov_b32_e32 v1, v3
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v2, v4
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB5_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, double 4.0 seq_cst
  ret void
}

define float @lds_atomic_fsub_ret_f32(ptr addrspace(3) %ptr, float %val) nounwind {
; VI-LABEL: lds_atomic_fsub_ret_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v2, v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB6_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, v2
; VI-NEXT:    v_sub_f32_e32 v2, v3, v1
; VI-NEXT:    ds_cmpst_rtn_b32 v2, v0, v3, v2
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v3
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB6_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v0, v2
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fsub_ret_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    ds_read_b32 v2, v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB6_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v3, v2
; GFX9-NEXT:    v_sub_f32_e32 v2, v3, v1
; GFX9-NEXT:    ds_cmpst_rtn_b32 v2, v0, v3, v2
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v3
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB6_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v0, v2
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fsub_ret_f32:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v2, v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB6_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v3, v2
; GFX7-NEXT:    v_sub_f32_e32 v2, v3, v1
; GFX7-NEXT:    ds_cmpst_rtn_b32 v2, v0, v3, v2
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v3
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB6_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v0, v2
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fsub_ret_f32:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v2, v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB6_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v3, v2
; GFX8-NEXT:    v_sub_f32_e32 v2, v3, v1
; GFX8-NEXT:    ds_cmpst_rtn_b32 v2, v0, v3, v2
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v2, v3
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB6_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v0, v2
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fsub ptr addrspace(3) %ptr, float %val seq_cst
  ret float %result
}

define void @lds_atomic_fsub_noret_f32(ptr addrspace(3) %ptr, float %val) nounwind {
; VI-LABEL: lds_atomic_fsub_noret_f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v2, v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB7_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_sub_f32_e32 v3, v2, v1
; VI-NEXT:    ds_cmpst_rtn_b32 v3, v0, v2, v3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v2
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v2, v3
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB7_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fsub_noret_f32:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    ds_read_b32 v2, v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB7_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_sub_f32_e32 v3, v2, v1
; GFX9-NEXT:    ds_cmpst_rtn_b32 v3, v0, v2, v3
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v2
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v2, v3
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB7_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fsub_noret_f32:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v2, v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB7_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_sub_f32_e32 v3, v2, v1
; GFX7-NEXT:    ds_cmpst_rtn_b32 v3, v0, v2, v3
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v2
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v2, v3
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB7_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fsub_noret_f32:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v2, v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB7_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_sub_f32_e32 v3, v2, v1
; GFX8-NEXT:    ds_cmpst_rtn_b32 v3, v0, v2, v3
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v2
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v2, v3
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB7_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fsub ptr addrspace(3) %ptr, float %val seq_cst
  ret void
}

define double @lds_atomic_fsub_ret_f64(ptr addrspace(3) %ptr, double %val) nounwind {
; VI-LABEL: lds_atomic_fsub_ret_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b64 v[3:4], v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB8_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v6, v4
; VI-NEXT:    v_mov_b32_e32 v5, v3
; VI-NEXT:    v_add_f64 v[3:4], v[5:6], -v[1:2]
; VI-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[5:6], v[3:4]
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[5:6]
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB8_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v0, v3
; VI-NEXT:    v_mov_b32_e32 v1, v4
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fsub_ret_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    ds_read_b64 v[3:4], v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB8_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v6, v4
; GFX9-NEXT:    v_mov_b32_e32 v5, v3
; GFX9-NEXT:    v_add_f64 v[3:4], v[5:6], -v[1:2]
; GFX9-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[5:6], v[3:4]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[5:6]
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB8_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v0, v3
; GFX9-NEXT:    v_mov_b32_e32 v1, v4
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fsub_ret_f64:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b64 v[3:4], v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB8_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v6, v4
; GFX7-NEXT:    v_mov_b32_e32 v5, v3
; GFX7-NEXT:    v_add_f64 v[3:4], v[5:6], -v[1:2]
; GFX7-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[5:6], v[3:4]
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[5:6]
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB8_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v0, v3
; GFX7-NEXT:    v_mov_b32_e32 v1, v4
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fsub_ret_f64:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b64 v[3:4], v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB8_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v6, v4
; GFX8-NEXT:    v_mov_b32_e32 v5, v3
; GFX8-NEXT:    v_add_f64 v[3:4], v[5:6], -v[1:2]
; GFX8-NEXT:    ds_cmpst_rtn_b64 v[3:4], v0, v[5:6], v[3:4]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u64_e32 vcc, v[3:4], v[5:6]
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB8_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v0, v3
; GFX8-NEXT:    v_mov_b32_e32 v1, v4
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fsub ptr addrspace(3) %ptr, double %val seq_cst
  ret double %result
}

define void @lds_atomic_fsub_noret_f64(ptr addrspace(3) %ptr, double %val) nounwind {
; VI-LABEL: lds_atomic_fsub_noret_f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b64 v[3:4], v0
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB9_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_add_f64 v[5:6], v[3:4], -v[1:2]
; VI-NEXT:    ds_cmpst_rtn_b64 v[5:6], v0, v[3:4], v[5:6]
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u64_e32 vcc, v[5:6], v[3:4]
; VI-NEXT:    v_mov_b32_e32 v3, v5
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v4, v6
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB9_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fsub_noret_f64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    ds_read_b64 v[3:4], v0
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:  .LBB9_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_add_f64 v[5:6], v[3:4], -v[1:2]
; GFX9-NEXT:    ds_cmpst_rtn_b64 v[5:6], v0, v[3:4], v[5:6]
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u64_e32 vcc, v[5:6], v[3:4]
; GFX9-NEXT:    v_mov_b32_e32 v3, v5
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v4, v6
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB9_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fsub_noret_f64:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b64 v[3:4], v0
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB9_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_add_f64 v[5:6], v[3:4], -v[1:2]
; GFX7-NEXT:    ds_cmpst_rtn_b64 v[5:6], v0, v[3:4], v[5:6]
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u64_e32 vcc, v[5:6], v[3:4]
; GFX7-NEXT:    v_mov_b32_e32 v3, v5
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v4, v6
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB9_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fsub_noret_f64:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b64 v[3:4], v0
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB9_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_add_f64 v[5:6], v[3:4], -v[1:2]
; GFX8-NEXT:    ds_cmpst_rtn_b64 v[5:6], v0, v[3:4], v[5:6]
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u64_e32 vcc, v[5:6], v[3:4]
; GFX8-NEXT:    v_mov_b32_e32 v3, v5
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v4, v6
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB9_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fsub ptr addrspace(3) %ptr, double %val seq_cst
  ret void
}

define bfloat @lds_atomic_fadd_ret_bf16(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_ret_bf16:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v1, -4, v0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v3, v1
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_mov_b32 s4, 0xffff
; VI-NEXT:    v_and_b32_e32 v0, 24, v2
; VI-NEXT:    v_lshlrev_b32_e64 v2, v2, s4
; VI-NEXT:    v_not_b32_e32 v2, v2
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB10_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v4, v3
; VI-NEXT:    v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_add_f32_e32 v3, 4.0, v3
; VI-NEXT:    v_bfe_u32 v6, v3, 16, 1
; VI-NEXT:    v_add_u32_e32 v6, vcc, v6, v3
; VI-NEXT:    v_add_u32_e32 v6, vcc, 0x7fff, v6
; VI-NEXT:    v_or_b32_e32 v7, 0x400000, v3
; VI-NEXT:    v_cmp_u_f32_e32 vcc, v3, v3
; VI-NEXT:    v_cndmask_b32_e32 v3, v6, v7, vcc
; VI-NEXT:    v_and_b32_e32 v5, v4, v2
; VI-NEXT:    v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; VI-NEXT:    v_or_b32_e32 v3, v5, v3
; VI-NEXT:    ds_cmpst_rtn_b32 v3, v1, v4, v3
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v4
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB10_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    v_lshrrev_b32_e32 v0, v0, v3
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_ret_bf16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX9-NEXT:    ds_read_b32 v3, v1
; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX9-NEXT:    s_mov_b32 s4, 0xffff
; GFX9-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s4
; GFX9-NEXT:    v_not_b32_e32 v2, v2
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:    s_movk_i32 s6, 0x7fff
; GFX9-NEXT:  .LBB10_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_mov_b32_e32 v4, v3
; GFX9-NEXT:    v_lshrrev_b32_sdwa v3, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX9-NEXT:    v_add_f32_e32 v3, 4.0, v3
; GFX9-NEXT:    v_bfe_u32 v5, v3, 16, 1
; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v3
; GFX9-NEXT:    v_add3_u32 v5, v5, v3, s6
; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v3, v3
; GFX9-NEXT:    v_cndmask_b32_e32 v3, v5, v6, vcc
; GFX9-NEXT:    v_lshlrev_b32_sdwa v3, v0, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT:    v_and_or_b32 v3, v4, v2, v3
; GFX9-NEXT:    ds_cmpst_rtn_b32 v3, v1, v4, v3
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v4
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB10_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    v_lshrrev_b32_e32 v0, v0, v3
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_ret_bf16:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v3, v1
; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX7-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX7-NEXT:    v_lshl_b32_e32 v2, 0xffff, v2
; GFX7-NEXT:    v_not_b32_e32 v2, v2
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB10_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_mov_b32_e32 v4, v3
; GFX7-NEXT:    v_lshrrev_b32_e32 v3, v0, v4
; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; GFX7-NEXT:    v_add_f32_e32 v3, 4.0, v3
; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; GFX7-NEXT:    v_and_b32_e32 v5, v4, v2
; GFX7-NEXT:    v_lshlrev_b32_e32 v3, v0, v3
; GFX7-NEXT:    v_or_b32_e32 v3, v5, v3
; GFX7-NEXT:    ds_cmpst_rtn_b32 v3, v1, v4, v3
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v4
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB10_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    v_lshrrev_b32_e32 v0, v0, v3
; GFX7-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_ret_bf16:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v3, v1
; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX8-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX8-NEXT:    v_lshl_b32_e32 v2, 0xffff, v2
; GFX8-NEXT:    v_not_b32_e32 v2, v2
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB10_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v4, v3
; GFX8-NEXT:    v_lshrrev_b32_e32 v3, v0, v4
; GFX8-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; GFX8-NEXT:    v_add_f32_e32 v3, 4.0, v3
; GFX8-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; GFX8-NEXT:    v_and_b32_e32 v5, v4, v2
; GFX8-NEXT:    v_lshlrev_b32_e32 v3, v0, v3
; GFX8-NEXT:    v_or_b32_e32 v3, v5, v3
; GFX8-NEXT:    ds_cmpst_rtn_b32 v3, v1, v4, v3
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v4
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB10_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    v_lshrrev_b32_e32 v0, v0, v3
; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, bfloat 4.0 seq_cst
  ret bfloat %result
}

define void @lds_atomic_fadd_noret_bf16(ptr addrspace(3) %ptr) nounwind {
; VI-LABEL: lds_atomic_fadd_noret_bf16:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_and_b32_e32 v1, -4, v0
; VI-NEXT:    s_mov_b32 m0, -1
; VI-NEXT:    ds_read_b32 v3, v1
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_mov_b32 s4, 0xffff
; VI-NEXT:    v_and_b32_e32 v0, 24, v2
; VI-NEXT:    v_lshlrev_b32_e64 v2, v2, s4
; VI-NEXT:    v_not_b32_e32 v2, v2
; VI-NEXT:    s_mov_b64 s[4:5], 0
; VI-NEXT:  .LBB11_1: ; %atomicrmw.start
; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_add_f32_e32 v4, 4.0, v4
; VI-NEXT:    v_bfe_u32 v6, v4, 16, 1
; VI-NEXT:    v_add_u32_e32 v6, vcc, v6, v4
; VI-NEXT:    v_add_u32_e32 v6, vcc, 0x7fff, v6
; VI-NEXT:    v_or_b32_e32 v7, 0x400000, v4
; VI-NEXT:    v_cmp_u_f32_e32 vcc, v4, v4
; VI-NEXT:    v_cndmask_b32_e32 v4, v6, v7, vcc
; VI-NEXT:    v_and_b32_e32 v5, v3, v2
; VI-NEXT:    v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; VI-NEXT:    v_or_b32_e32 v4, v5, v4
; VI-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; VI-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; VI-NEXT:    v_mov_b32_e32 v3, v4
; VI-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; VI-NEXT:    s_cbranch_execnz .LBB11_1
; VI-NEXT:  ; %bb.2: ; %atomicrmw.end
; VI-NEXT:    s_or_b64 exec, exec, s[4:5]
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: lds_atomic_fadd_noret_bf16:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX9-NEXT:    ds_read_b32 v3, v1
; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX9-NEXT:    s_mov_b32 s4, 0xffff
; GFX9-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX9-NEXT:    v_lshlrev_b32_e64 v2, v2, s4
; GFX9-NEXT:    v_not_b32_e32 v2, v2
; GFX9-NEXT:    s_mov_b64 s[4:5], 0
; GFX9-NEXT:    s_movk_i32 s6, 0x7fff
; GFX9-NEXT:  .LBB11_1: ; %atomicrmw.start
; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_lshrrev_b32_sdwa v4, v0, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX9-NEXT:    v_add_f32_e32 v4, 4.0, v4
; GFX9-NEXT:    v_bfe_u32 v5, v4, 16, 1
; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v4
; GFX9-NEXT:    v_add3_u32 v5, v5, v4, s6
; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v4, v4
; GFX9-NEXT:    v_cndmask_b32_e32 v4, v5, v6, vcc
; GFX9-NEXT:    v_lshlrev_b32_sdwa v4, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NEXT:    v_and_or_b32 v4, v3, v2, v4
; GFX9-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX9-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX9-NEXT:    v_mov_b32_e32 v3, v4
; GFX9-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_cbranch_execnz .LBB11_1
; GFX9-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX9-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; GFX7-LABEL: lds_atomic_fadd_noret_bf16:
; GFX7:       ; %bb.0:
; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX7-NEXT:    s_mov_b32 m0, -1
; GFX7-NEXT:    ds_read_b32 v3, v1
; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX7-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX7-NEXT:    v_lshl_b32_e32 v2, 0xffff, v2
; GFX7-NEXT:    v_not_b32_e32 v2, v2
; GFX7-NEXT:    s_mov_b64 s[4:5], 0
; GFX7-NEXT:  .LBB11_1: ; %atomicrmw.start
; GFX7-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_lshrrev_b32_e32 v4, v0, v3
; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
; GFX7-NEXT:    v_add_f32_e32 v4, 4.0, v4
; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v4
; GFX7-NEXT:    v_and_b32_e32 v5, v3, v2
; GFX7-NEXT:    v_lshlrev_b32_e32 v4, v0, v4
; GFX7-NEXT:    v_or_b32_e32 v4, v5, v4
; GFX7-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX7-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT:    v_mov_b32_e32 v3, v4
; GFX7-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_cbranch_execnz .LBB11_1
; GFX7-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX7-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX7-NEXT:    s_setpc_b64 s[30:31]
;
; GFX8-LABEL: lds_atomic_fadd_noret_bf16:
; GFX8:       ; %bb.0:
; GFX8-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT:    v_and_b32_e32 v1, -4, v0
; GFX8-NEXT:    s_mov_b32 m0, -1
; GFX8-NEXT:    ds_read_b32 v3, v1
; GFX8-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX8-NEXT:    v_and_b32_e32 v0, 24, v2
; GFX8-NEXT:    v_lshl_b32_e32 v2, 0xffff, v2
; GFX8-NEXT:    v_not_b32_e32 v2, v2
; GFX8-NEXT:    s_mov_b64 s[4:5], 0
; GFX8-NEXT:  .LBB11_1: ; %atomicrmw.start
; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_lshrrev_b32_e32 v4, v0, v3
; GFX8-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
; GFX8-NEXT:    v_add_f32_e32 v4, 4.0, v4
; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 16, v4
; GFX8-NEXT:    v_and_b32_e32 v5, v3, v2
; GFX8-NEXT:    v_lshlrev_b32_e32 v4, v0, v4
; GFX8-NEXT:    v_or_b32_e32 v4, v5, v4
; GFX8-NEXT:    ds_cmpst_rtn_b32 v4, v1, v3, v4
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_cmp_eq_u32_e32 vcc, v4, v3
; GFX8-NEXT:    s_or_b64 s[4:5], vcc, s[4:5]
; GFX8-NEXT:    v_mov_b32_e32 v3, v4
; GFX8-NEXT:    s_andn2_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_cbranch_execnz .LBB11_1
; GFX8-NEXT:  ; %bb.2: ; %atomicrmw.end
; GFX8-NEXT:    s_or_b64 exec, exec, s[4:5]
; GFX8-NEXT:    s_setpc_b64 s[30:31]
  %result = atomicrmw fadd ptr addrspace(3) %ptr, bfloat 4.0 seq_cst
  ret void
}