AMDGPU: Remove atomic inc/dec patterns

[arsenm]

Summary

There is no benefit to these since materializing the constant 1
requires the same number of instructions as materializing uint_max

[• tstellarAMD]

LGTM.

[arsenm]

r264215

lib/Target/AMDGPU/SIInstructions.td

(DS_WRITE2_B32 $ptr, (EXTRACT_SUBREG $value, sub0),
                        (i1 0))
 >;
 
 class DSAtomicRetPat<DS inst, ValueType vt, PatFrag frag> : Pat <
   (frag (DS1Addr1Offset i32:$ptr, i32:$offset), vt:$value),
   (inst $ptr, $value, (as_i16imm $offset), (i1 0))
 >;
 
-// Special case of DSAtomicRetPat for add / sub 1 -> inc / dec
-//
-// We need to use something for the data0, so we set a register to
-// -1. For the non-rtn variants, the manual says it does
-// DS[A] = (DS[A] >= D0) ? 0 : DS[A] + 1, and setting D0 to uint_max
-// will always do the increment so I'm assuming it's the same.
-class DSAtomicIncRetPat<DS inst, ValueType vt,
-                        Instruction LoadImm, PatFrag frag> : Pat <
-  (frag (DS1Addr1Offset i32:$ptr, i32:$offset), (vt 1)),
-  (inst $ptr, (LoadImm (vt -1)), (as_i16imm $offset), (i1 0))
->;
-
-
 class DSAtomicCmpXChg <DS inst, ValueType vt, PatFrag frag> : Pat <
   (frag (DS1Addr1Offset i32:$ptr, i32:$offset), vt:$cmp, vt:$swap),
   (inst $ptr, $cmp, $swap, (as_i16imm $offset), (i1 0))
 >;
 
 
 // 32-bit atomics.
-def : DSAtomicIncRetPat<DS_INC_RTN_U32, i32,
-                        V_MOV_B32_e32, si_atomic_load_add_local>;
-def : DSAtomicIncRetPat<DS_DEC_RTN_U32, i32,
-                        V_MOV_B32_e32, si_atomic_load_sub_local>;
-
 def : DSAtomicRetPat<DS_WRXCHG_RTN_B32, i32, si_atomic_swap_local>;
 def : DSAtomicRetPat<DS_ADD_RTN_U32, i32, si_atomic_load_add_local>;
 def : DSAtomicRetPat<DS_SUB_RTN_U32, i32, si_atomic_load_sub_local>;
 def : DSAtomicRetPat<DS_AND_RTN_B32, i32, si_atomic_load_and_local>;
 def : DSAtomicRetPat<DS_OR_RTN_B32, i32, si_atomic_load_or_local>;
 def : DSAtomicRetPat<DS_XOR_RTN_B32, i32, si_atomic_load_xor_local>;
 def : DSAtomicRetPat<DS_MIN_RTN_I32, i32, si_atomic_load_min_local>;
 def : DSAtomicRetPat<DS_MAX_RTN_I32, i32, si_atomic_load_max_local>;
 def : DSAtomicRetPat<DS_MIN_RTN_U32, i32, si_atomic_load_umin_local>;
 def : DSAtomicRetPat<DS_MAX_RTN_U32, i32, si_atomic_load_umax_local>;
 
 def : DSAtomicCmpXChg<DS_CMPST_RTN_B32, i32, si_atomic_cmp_swap_32_local>;
 
 // 64-bit atomics.
-def : DSAtomicIncRetPat<DS_INC_RTN_U64, i64,
-                        V_MOV_B64_PSEUDO, si_atomic_load_add_local>;
-def : DSAtomicIncRetPat<DS_DEC_RTN_U64, i64,
-                        V_MOV_B64_PSEUDO, si_atomic_load_sub_local>;
-
 def : DSAtomicRetPat<DS_WRXCHG_RTN_B64, i64, si_atomic_swap_local>;
 def : DSAtomicRetPat<DS_ADD_RTN_U64, i64, si_atomic_load_add_local>;
 def : DSAtomicRetPat<DS_SUB_RTN_U64, i64, si_atomic_load_sub_local>;
 def : DSAtomicRetPat<DS_AND_RTN_B64, i64, si_atomic_load_and_local>;
 def : DSAtomicRetPat<DS_OR_RTN_B64, i64, si_atomic_load_or_local>;
 def : DSAtomicRetPat<DS_XOR_RTN_B64, i64, si_atomic_load_xor_local>;
 def : DSAtomicRetPat<DS_MIN_RTN_I64, i64, si_atomic_load_min_local>;
 def : DSAtomicRetPat<DS_MAX_RTN_I64, i64, si_atomic_load_max_local>;

test/CodeGen/AMDGPU/local-atomics.ll

-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=CIVI -check-prefix=GCN -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=CIVI -check-prefix=GCN -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=redwood -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
 ; FUNC-LABEL: {{^}}lds_atomic_xchg_ret_i32:
 ; EG: LDS_WRXCHG_RET *
 ; GCN: v_mov_b32_e32 [[DATA:v[0-9]+]], 4
 ; GCN: s_load_dword [[SPTR:s[0-9]+]],
define void @lds_atomic_add_ret_i32_bad_si_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
   %sub = sub i32 %a, %b
   %add = add i32 %sub, 4
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 %add
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 4 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_ret_i32:
+; FUNC-LABEL: {{^}}lds_atomic_add1_ret_i32:
 ; EG: LDS_ADD_RET *
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_inc_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[NEGONE]]
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_add_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[ONE]]
 ; GCN: s_endpgm
-define void @lds_atomic_inc_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i32 addrspace(3)* %ptr, i32 1 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_ret_i32_offset:
+; FUNC-LABEL: {{^}}lds_atomic_add1_ret_i32_offset:
 ; EG: LDS_ADD_RET *
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_inc_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[NEGONE]] offset:16
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_add_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[ONE]] offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_inc_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 1 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_ret_i32_bad_si_offset:
+; FUNC-LABEL: {{^}}lds_atomic_add1_ret_i32_bad_si_offset:
 ; EG: LDS_ADD_RET *
-; SI: ds_inc_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
-; CIVI: ds_inc_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset:16
+; SI: ds_add_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; CIVI: ds_add_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_inc_ret_i32_bad_si_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
+define void @lds_atomic_add1_ret_i32_bad_si_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
   %sub = sub i32 %a, %b
   %add = add i32 %sub, 4
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 %add
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 1 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 

 ; GCN: s_endpgm
 define void @lds_atomic_sub_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i32 addrspace(3)* %gep, i32 4 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_ret_i32:
+; FUNC-LABEL: {{^}}lds_atomic_sub1_ret_i32:
 ; EG: LDS_SUB_RET *
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_dec_rtn_u32  v{{[0-9]+}}, v{{[0-9]+}}, [[NEGONE]]
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_sub_rtn_u32  v{{[0-9]+}}, v{{[0-9]+}}, [[ONE]]
 ; GCN: s_endpgm
-define void @lds_atomic_dec_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i32 addrspace(3)* %ptr, i32 1 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_ret_i32_offset:
+; FUNC-LABEL: {{^}}lds_atomic_sub1_ret_i32_offset:
 ; EG: LDS_SUB_RET *
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_dec_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[NEGONE]] offset:16
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_sub_rtn_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[ONE]] offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_dec_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i32 addrspace(3)* %gep, i32 1 seq_cst
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; FUNC-LABEL: {{^}}lds_atomic_and_ret_i32:
 ; EG: LDS_AND_RET *

 define void @lds_atomic_add_noret_i32_bad_si_offset(i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
   %sub = sub i32 %a, %b
   %add = add i32 %sub, 4
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 %add
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_noret_i32:
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_inc_u32 v{{[0-9]+}}, [[NEGONE]]
+; FUNC-LABEL: {{^}}lds_atomic_add1_noret_i32:
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_add_u32 v{{[0-9]+}}, [[ONE]]
 ; GCN: s_endpgm
-define void @lds_atomic_inc_noret_i32(i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_noret_i32(i32 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i32 addrspace(3)* %ptr, i32 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_noret_i32_offset:
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_inc_u32 v{{[0-9]+}}, [[NEGONE]] offset:16
+; FUNC-LABEL: {{^}}lds_atomic_add1_noret_i32_offset:
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_add_u32 v{{[0-9]+}}, [[ONE]] offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_inc_noret_i32_offset(i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_noret_i32_offset(i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_noret_i32_bad_si_offset:
-; SI: ds_inc_u32 v{{[0-9]+}}, v{{[0-9]+}}
-; CIVI: ds_inc_u32 v{{[0-9]+}}, v{{[0-9]+}} offset:16
+; FUNC-LABEL: {{^}}lds_atomic_add1_noret_i32_bad_si_offset:
+; SI: ds_add_u32 v{{[0-9]+}}, v{{[0-9]+}}
+; CIVI: ds_add_u32 v{{[0-9]+}}, v{{[0-9]+}} offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_inc_noret_i32_bad_si_offset(i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
+define void @lds_atomic_add1_noret_i32_bad_si_offset(i32 addrspace(3)* %ptr, i32 %a, i32 %b) nounwind {
   %sub = sub i32 %a, %b
   %add = add i32 %sub, 4
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 %add
   %result = atomicrmw add i32 addrspace(3)* %gep, i32 1 seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}lds_atomic_sub_noret_i32:
 ; GCN: ds_sub_u32
 ; GCN: s_endpgm
 define void @lds_atomic_sub_noret_i32(i32 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i32 addrspace(3)* %ptr, i32 4 seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}lds_atomic_sub_noret_i32_offset:
 ; GCN: ds_sub_u32 v{{[0-9]+}}, v{{[0-9]+}} offset:16
 ; GCN: s_endpgm
 define void @lds_atomic_sub_noret_i32_offset(i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i32 addrspace(3)* %gep, i32 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_noret_i32:
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_dec_u32  v{{[0-9]+}}, [[NEGONE]]
+; FUNC-LABEL: {{^}}lds_atomic_sub1_noret_i32:
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_sub_u32 v{{[0-9]+}}, [[ONE]]
 ; GCN: s_endpgm
-define void @lds_atomic_dec_noret_i32(i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_noret_i32(i32 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i32 addrspace(3)* %ptr, i32 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_noret_i32_offset:
-; GCN: v_mov_b32_e32 [[NEGONE:v[0-9]+]], -1
-; GCN: ds_dec_u32 v{{[0-9]+}}, [[NEGONE]] offset:16
+; FUNC-LABEL: {{^}}lds_atomic_sub1_noret_i32_offset:
+; GCN: v_mov_b32_e32 [[ONE:v[0-9]+]], 1{{$}}
+; GCN: ds_sub_u32 v{{[0-9]+}}, [[ONE]] offset:16
 ; GCN: s_endpgm
-define void @lds_atomic_dec_noret_i32_offset(i32 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_noret_i32_offset(i32 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i32, i32 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i32 addrspace(3)* %gep, i32 1 seq_cst
   ret void
 }
 
 ; FUNC-LABEL: {{^}}lds_atomic_and_noret_i32:
 ; GCN: ds_and_b32
 ; GCN: s_endpgm

test/CodeGen/AMDGPU/local-atomics64.ll

-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=SI -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=SI -check-prefix=GCN %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=VI -check-prefix=GCN %s
 
-; FUNC-LABEL: {{^}}lds_atomic_xchg_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_xchg_ret_i64:
 ; GCN: ds_wrxchg_rtn_b64
 ; GCN: s_endpgm
 define void @lds_atomic_xchg_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw xchg i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xchg_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_xchg_ret_i64_offset:
 ; GCN: ds_wrxchg_rtn_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_xchg_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw xchg i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_add_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_add_ret_i64:
 ; GCN: ds_add_rtn_u64
 ; GCN: s_endpgm
 define void @lds_atomic_add_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_add_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_add_ret_i64_offset:
 ; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 9
 ; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0
 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
 ; VI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
 ; GCN-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
 ; GCN: ds_add_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}} offset:32
 ; GCN: buffer_store_dwordx2 [[RESULT]],
 ; GCN: s_endpgm
 define void @lds_atomic_add_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i64 4
   %result = atomicrmw add i64 addrspace(3)* %gep, i64 9 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_ret_i64:
-; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], -1
-; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], -1
-; GCN: ds_inc_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
+; GCN-LABEL: {{^}}lds_atomic_add1_ret_i64:
+; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
+; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
+; GCN: ds_add_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
 ; GCN: buffer_store_dwordx2 [[RESULT]],
 ; GCN: s_endpgm
-define void @lds_atomic_inc_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i64 addrspace(3)* %ptr, i64 1 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_ret_i64_offset:
-; GCN: ds_inc_rtn_u64 {{.*}} offset:32
+; GCN-LABEL: {{^}}lds_atomic_add1_ret_i64_offset:
+; GCN: ds_add_rtn_u64 {{.*}} offset:32
 ; GCN: s_endpgm
-define void @lds_atomic_inc_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw add i64 addrspace(3)* %gep, i64 1 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_sub_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_sub_ret_i64:
 ; GCN: ds_sub_rtn_u64
 ; GCN: s_endpgm
 define void @lds_atomic_sub_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_sub_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_sub_ret_i64_offset:
 ; GCN: ds_sub_rtn_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_sub_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_ret_i64:
-; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], -1
-; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], -1
-; GCN: ds_dec_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
+; GCN-LABEL: {{^}}lds_atomic_sub1_ret_i64:
+; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
+; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
+; GCN: ds_sub_rtn_u64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
 ; GCN: buffer_store_dwordx2 [[RESULT]],
 ; GCN: s_endpgm
-define void @lds_atomic_dec_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i64 addrspace(3)* %ptr, i64 1 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_ret_i64_offset:
-; GCN: ds_dec_rtn_u64 {{.*}} offset:32
+; GCN-LABEL: {{^}}lds_atomic_sub1_ret_i64_offset:
+; GCN: ds_sub_rtn_u64 {{.*}} offset:32
 ; GCN: s_endpgm
-define void @lds_atomic_dec_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i64 addrspace(3)* %gep, i64 1 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_and_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_and_ret_i64:
 ; GCN: ds_and_rtn_b64
 ; GCN: s_endpgm
 define void @lds_atomic_and_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw and i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_and_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_and_ret_i64_offset:
 ; GCN: ds_and_rtn_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_and_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw and i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_or_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_or_ret_i64:
 ; GCN: ds_or_rtn_b64
 ; GCN: s_endpgm
 define void @lds_atomic_or_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw or i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_or_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_or_ret_i64_offset:
 ; GCN: ds_or_rtn_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_or_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw or i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xor_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_xor_ret_i64:
 ; GCN: ds_xor_rtn_b64
 ; GCN: s_endpgm
 define void @lds_atomic_xor_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw xor i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xor_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_xor_ret_i64_offset:
 ; GCN: ds_xor_rtn_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_xor_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw xor i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 ; FIXME: There is no atomic nand instr
-; XFUNC-LABEL: {{^}}lds_atomic_nand_ret_i64:uction, so we somehow need to expand this.
+; XGCN-LABEL: {{^}}lds_atomic_nand_ret_i64:uction, so we somehow need to expand this.
 ; define void @lds_atomic_nand_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
 ;   %result = atomicrmw nand i64 addrspace(3)* %ptr, i32 4 seq_cst
 ;   store i64 %result, i64 addrspace(1)* %out, align 8
 ;   ret void
 ; }
 
-; FUNC-LABEL: {{^}}lds_atomic_min_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_min_ret_i64:
 ; GCN: ds_min_rtn_i64
 ; GCN: s_endpgm
 define void @lds_atomic_min_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw min i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_min_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_min_ret_i64_offset:
 ; GCN: ds_min_rtn_i64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_min_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw min i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_max_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_max_ret_i64:
 ; GCN: ds_max_rtn_i64
 ; GCN: s_endpgm
 define void @lds_atomic_max_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw max i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_max_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_max_ret_i64_offset:
 ; GCN: ds_max_rtn_i64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_max_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw max i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umin_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_umin_ret_i64:
 ; GCN: ds_min_rtn_u64
 ; GCN: s_endpgm
 define void @lds_atomic_umin_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw umin i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umin_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_umin_ret_i64_offset:
 ; GCN: ds_min_rtn_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_umin_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw umin i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umax_ret_i64:
+; GCN-LABEL: {{^}}lds_atomic_umax_ret_i64:
 ; GCN: ds_max_rtn_u64
 ; GCN: s_endpgm
 define void @lds_atomic_umax_ret_i64(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw umax i64 addrspace(3)* %ptr, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umax_ret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_umax_ret_i64_offset:
 ; GCN: ds_max_rtn_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_umax_ret_i64_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw umax i64 addrspace(3)* %gep, i64 4 seq_cst
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xchg_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_xchg_noret_i64:
 ; GCN: ds_wrxchg_rtn_b64
 ; GCN: s_endpgm
 define void @lds_atomic_xchg_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw xchg i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xchg_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_xchg_noret_i64_offset:
 ; GCN: ds_wrxchg_rtn_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_xchg_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw xchg i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_add_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_add_noret_i64:
 ; GCN: ds_add_u64
 ; GCN: s_endpgm
 define void @lds_atomic_add_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_add_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_add_noret_i64_offset:
 ; SI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x9
 ; VI: s_load_dword [[PTR:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x24
 ; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 9
 ; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0
 ; GCN: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
 ; GCN: ds_add_u64 [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_add_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i64 4
   %result = atomicrmw add i64 addrspace(3)* %gep, i64 9 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_noret_i64:
-; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], -1
-; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], -1
-; GCN: ds_inc_u64 [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
+; GCN-LABEL: {{^}}lds_atomic_add1_noret_i64:
+; GCN-DAG: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
+; GCN-DAG: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
+; GCN: ds_add_u64 [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
 ; GCN: s_endpgm
-define void @lds_atomic_inc_noret_i64(i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw add i64 addrspace(3)* %ptr, i64 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_inc_noret_i64_offset:
-; GCN: ds_inc_u64 {{.*}} offset:32
+; GCN-LABEL: {{^}}lds_atomic_add1_noret_i64_offset:
+; GCN: ds_add_u64 {{.*}} offset:32
 ; GCN: s_endpgm
-define void @lds_atomic_inc_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_add1_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw add i64 addrspace(3)* %gep, i64 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_sub_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_sub_noret_i64:
 ; GCN: ds_sub_u64
 ; GCN: s_endpgm
 define void @lds_atomic_sub_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_sub_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_sub_noret_i64_offset:
 ; GCN: ds_sub_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_sub_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_noret_i64:
-; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], -1
-; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], -1
-; GCN: ds_dec_u64 [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
+; GCN-LABEL: {{^}}lds_atomic_sub1_noret_i64:
+; GCN: v_mov_b32_e32 v[[LOVDATA:[0-9]+]], 1{{$}}
+; GCN: v_mov_b32_e32 v[[HIVDATA:[0-9]+]], 0{{$}}
+; GCN: ds_sub_u64 [[VPTR]], v{{\[}}[[LOVDATA]]:[[HIVDATA]]{{\]}}
 ; GCN: s_endpgm
-define void @lds_atomic_dec_noret_i64(i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw sub i64 addrspace(3)* %ptr, i64 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_dec_noret_i64_offset:
-; GCN: ds_dec_u64 {{.*}} offset:32
+; GCN-LABEL: {{^}}lds_atomic_sub1_noret_i64_offset:
+; GCN: ds_sub_u64 {{.*}} offset:32
 ; GCN: s_endpgm
-define void @lds_atomic_dec_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
+define void @lds_atomic_sub1_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw sub i64 addrspace(3)* %gep, i64 1 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_and_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_and_noret_i64:
 ; GCN: ds_and_b64
 ; GCN: s_endpgm
 define void @lds_atomic_and_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw and i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_and_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_and_noret_i64_offset:
 ; GCN: ds_and_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_and_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw and i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_or_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_or_noret_i64:
 ; GCN: ds_or_b64
 ; GCN: s_endpgm
 define void @lds_atomic_or_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw or i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_or_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_or_noret_i64_offset:
 ; GCN: ds_or_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_or_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw or i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xor_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_xor_noret_i64:
 ; GCN: ds_xor_b64
 ; GCN: s_endpgm
 define void @lds_atomic_xor_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw xor i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_xor_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_xor_noret_i64_offset:
 ; GCN: ds_xor_b64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_xor_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw xor i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
 ; FIXME: There is no atomic nand instr
-; XFUNC-LABEL: {{^}}lds_atomic_nand_noret_i64:uction, so we somehow need to expand this.
+; XGCN-LABEL: {{^}}lds_atomic_nand_noret_i64:uction, so we somehow need to expand this.
 ; define void @lds_atomic_nand_noret_i64(i64 addrspace(3)* %ptr) nounwind {
 ;   %result = atomicrmw nand i64 addrspace(3)* %ptr, i32 4 seq_cst
 ;   ret void
 ; }
 
-; FUNC-LABEL: {{^}}lds_atomic_min_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_min_noret_i64:
 ; GCN: ds_min_i64
 ; GCN: s_endpgm
 define void @lds_atomic_min_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw min i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_min_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_min_noret_i64_offset:
 ; GCN: ds_min_i64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_min_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw min i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_max_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_max_noret_i64:
 ; GCN: ds_max_i64
 ; GCN: s_endpgm
 define void @lds_atomic_max_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw max i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_max_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_max_noret_i64_offset:
 ; GCN: ds_max_i64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_max_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw max i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umin_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_umin_noret_i64:
 ; GCN: ds_min_u64
 ; GCN: s_endpgm
 define void @lds_atomic_umin_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw umin i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umin_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_umin_noret_i64_offset:
 ; GCN: ds_min_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_umin_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw umin i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umax_noret_i64:
+; GCN-LABEL: {{^}}lds_atomic_umax_noret_i64:
 ; GCN: ds_max_u64
 ; GCN: s_endpgm
 define void @lds_atomic_umax_noret_i64(i64 addrspace(3)* %ptr) nounwind {
   %result = atomicrmw umax i64 addrspace(3)* %ptr, i64 4 seq_cst
   ret void
 }
 
-; FUNC-LABEL: {{^}}lds_atomic_umax_noret_i64_offset:
+; GCN-LABEL: {{^}}lds_atomic_umax_noret_i64_offset:
 ; GCN: ds_max_u64 {{.*}} offset:32
 ; GCN: s_endpgm
 define void @lds_atomic_umax_noret_i64_offset(i64 addrspace(3)* %ptr) nounwind {
   %gep = getelementptr i64, i64 addrspace(3)* %ptr, i32 4
   %result = atomicrmw umax i64 addrspace(3)* %gep, i64 4 seq_cst
   ret void
 }