Skip to content

Commit 91963df

Browse files
ritter-x2aritter-x2a
committed
Set GISelShouldIgnore for the 64-bit PTRADD patterns.
1 parent 010f5ed commit 91963df

23 files changed

+4014
-1281
lines changed

‎llvm/lib/Target/AMDGPU/SIInstructions.td

Lines changed: 6 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1380,9 +1380,15 @@ def : GCNPat <
13801380
// PTRADD Patterns
13811381
//===----------------------------------------------------------------------===//
13821382

1383+
// GlobalISel shouldn't generate 64-bit addition pseudos.
1384+
let GISelShouldIgnore = 1 in {
13831385
def : GCNPat<
13841386
(DivergentBinFrag<ptradd> i64:$src0, i64:$src1),
13851387
(V_ADD_U64_PSEUDO $src0, $src1)>;
1388+
def : GCNPat<
1389+
(UniformBinFrag<ptradd> i64:$src0, i64:$src1),
1390+
(S_ADD_U64_PSEUDO $src0, $src1)>;
1391+
}
13861392

13871393
def : GCNPat<
13881394
(DivergentBinFrag<ptradd> i32:$src0, i32:$src1),
@@ -1396,10 +1402,6 @@ def : GCNPat<
13961402
let SubtargetPredicate = NotHasAddNoCarryInsts;
13971403
}
13981404

1399-
def : GCNPat<
1400-
(UniformBinFrag<ptradd> i64:$src0, i64:$src1),
1401-
(S_ADD_U64_PSEUDO $src0, $src1)>;
1402-
14031405
// Whether we select S_ADD_I32 or S_ADD_U32 does not make much of a
14041406
// difference. Most notably, S_ADD_I32 instructions can be transformed
14051407
// to S_ADDK_I32, so we select that.

‎llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgpu-atomic-cmpxchg-flat.mir

Lines changed: 64 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -91,8 +91,14 @@ body: |
9191
; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
9292
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
9393
; GFX7-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 4, implicit $exec
94-
; GFX7-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc, implicit $exec
95-
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
94+
; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
95+
; GFX7-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
96+
; GFX7-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
97+
; GFX7-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
98+
; GFX7-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
99+
; GFX7-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
100+
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
101+
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
96102
; GFX7-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
97103
;
98104
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gep4
@@ -113,8 +119,14 @@ body: |
113119
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
114120
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
115121
; GFX10-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 4, implicit $exec
116-
; GFX10-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc_lo, implicit $exec
117-
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
122+
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
123+
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
124+
; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
125+
; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
126+
; GFX10-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
127+
; GFX10-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
128+
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
129+
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
118130
; GFX10-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
119131
;
120132
; GFX11-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gep4
@@ -231,8 +243,14 @@ body: |
231243
; GFX7-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
232244
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
233245
; GFX7-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 4, implicit $exec
234-
; GFX7-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc, implicit $exec
235-
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64))
246+
; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
247+
; GFX7-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
248+
; GFX7-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
249+
; GFX7-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
250+
; GFX7-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
251+
; GFX7-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
252+
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
253+
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64))
236254
; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
237255
;
238256
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_flat_gep4
@@ -253,8 +271,14 @@ body: |
253271
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
254272
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
255273
; GFX10-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 4, implicit $exec
256-
; GFX10-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc_lo, implicit $exec
257-
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64))
274+
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
275+
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
276+
; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
277+
; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
278+
; GFX10-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
279+
; GFX10-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
280+
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
281+
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s64))
258282
; GFX10-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
259283
;
260284
; GFX11-LABEL: name: amdgpu_atomic_cmpxchg_s64_flat_gep4
@@ -304,8 +328,14 @@ body: |
304328
; GFX7-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
305329
; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
306330
; GFX7-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO -4, implicit $exec
307-
; GFX7-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc, implicit $exec
308-
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
331+
; GFX7-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
332+
; GFX7-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
333+
; GFX7-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
334+
; GFX7-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
335+
; GFX7-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
336+
; GFX7-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
337+
; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
338+
; GFX7-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
309339
; GFX7-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
310340
;
311341
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gepm4
@@ -316,8 +346,14 @@ body: |
316346
; GFX9-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
317347
; GFX9-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
318348
; GFX9-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO -4, implicit $exec
319-
; GFX9-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc, implicit $exec
320-
; GFX9-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
349+
; GFX9-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
350+
; GFX9-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
351+
; GFX9-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
352+
; GFX9-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
353+
; GFX9-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
354+
; GFX9-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
355+
; GFX9-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
356+
; GFX9-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
321357
; GFX9-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
322358
;
323359
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gepm4
@@ -328,8 +364,14 @@ body: |
328364
; GFX10-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
329365
; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
330366
; GFX10-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO -4, implicit $exec
331-
; GFX10-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc_lo, implicit $exec
332-
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
367+
; GFX10-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
368+
; GFX10-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
369+
; GFX10-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
370+
; GFX10-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
371+
; GFX10-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
372+
; GFX10-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
373+
; GFX10-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
374+
; GFX10-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
333375
; GFX10-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
334376
;
335377
; GFX11-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gepm4
@@ -340,8 +382,14 @@ body: |
340382
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
341383
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
342384
; GFX11-NEXT: [[V_MOV_B:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO -4, implicit $exec
343-
; GFX11-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[COPY]], [[V_MOV_B]], implicit-def dead $vcc_lo, implicit $exec
344-
; GFX11-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[V_ADD_U]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
385+
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
386+
; GFX11-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub0
387+
; GFX11-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
388+
; GFX11-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[V_MOV_B]].sub1
389+
; GFX11-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
390+
; GFX11-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
391+
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, [[V_ADDC_U32_e64_]], %subreg.sub1
392+
; GFX11-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], 0, 1, implicit $exec, implicit $flat_scr :: (load store seq_cst (s32))
345393
; GFX11-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
346394
;
347395
; GFX12-LABEL: name: amdgpu_atomic_cmpxchg_s32_flat_gepm4

0 commit comments

Comments
 (0)