Skip to content

Commit 8ea018c

Browse files
Pierre-vhPierre-vh
authored
[DAGISel] Fix MMRA Handling in copyExtraInfo (#124730)
#78569 did not implement this correctly and an edge case breaks it by triggering `Assertion `!Leafs.empty()' failed.` Fixes SWDEV-507698
1 parent 8353aa2 commit 8ea018c

File tree

2 files changed

+103
-4
lines changed

2 files changed

+103
-4
lines changed

‎llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -13635,7 +13635,7 @@ void SelectionDAG::copyExtraInfo(SDNode *From, SDNode *To) {
1363513635
// Use of operator[] on the DenseMap may cause an insertion, which invalidates
1363613636
// the iterator, hence the need to make a copy to prevent a use-after-free.
1363713637
NodeExtraInfo NEI = I->second;
13638-
if (LLVM_LIKELY(!NEI.PCSections) && LLVM_LIKELY(!NEI.MMRA)) {
13638+
if (LLVM_LIKELY(!NEI.PCSections)) {
1363913639
// No deep copy required for the types of extra info set.
1364013640
//
1364113641
// FIXME: Investigate if other types of extra info also need deep copy. This

‎llvm/test/CodeGen/AMDGPU/mmra.ll

Lines changed: 102 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -17,7 +17,7 @@ define void @fence_loads(ptr %ptr) {
1717
; CHECK-NEXT: ATOMIC_FENCE 5, 1, mmra !0
1818
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]], mmra !1
1919
; CHECK-NEXT: [[FLAT_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = FLAT_LOAD_UBYTE [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr, mmra !1 :: (load acquire (s8) from %ir.ptr, align 4)
20-
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1, implicit $exec, mmra !2
20+
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 1, implicit $exec
2121
; CHECK-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]], mmra !2
2222
; CHECK-NEXT: FLAT_STORE_BYTE [[COPY3]], killed [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr, mmra !2 :: (store release (s8) into %ir.ptr, align 4)
2323
; CHECK-NEXT: SI_RETURN
@@ -82,7 +82,7 @@ define void @atomicrmw_rel(ptr %ptr) {
8282
; CHECK-NEXT: [[V_AND_B32_e64_2:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[PHI1]], killed [[V_OR_B32_e64_]], implicit $exec
8383
; CHECK-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
8484
; CHECK-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
85-
; CHECK-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_AND_B32_e64_2]], %subreg.sub0, [[PHI1]], %subreg.sub1, mmra !2
85+
; CHECK-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_AND_B32_e64_2]], %subreg.sub0, [[PHI1]], %subreg.sub1
8686
; CHECK-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE2]], mmra !2
8787
; CHECK-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY4]], killed [[COPY6]], 0, 1, implicit $exec, implicit $flat_scr, mmra !2 :: (load store release monotonic (s32) on %ir.AlignedAddr)
8888
; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[FLAT_ATOMIC_CMPSWAP_RTN]], [[PHI1]], implicit $exec, mmra !2
@@ -140,7 +140,7 @@ define void @cmpxchg(ptr %ptr) {
140140
; CHECK-NEXT: [[V_OR_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[PHI2]], [[V_LSHLREV_B32_e64_2]], implicit $exec
141141
; CHECK-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
142142
; CHECK-NEXT: [[DEF6:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
143-
; CHECK-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_OR_B32_e64_]], %subreg.sub0, [[PHI2]], %subreg.sub1, mmra !1
143+
; CHECK-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_OR_B32_e64_]], %subreg.sub0, [[PHI2]], %subreg.sub1
144144
; CHECK-NEXT: [[COPY6:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE2]], mmra !1
145145
; CHECK-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY4]], killed [[COPY6]], 0, 1, implicit $exec, implicit $flat_scr, mmra !1 :: (load store acquire acquire (s32) on %ir.AlignedAddr)
146146
; CHECK-NEXT: [[V_CMP_NE_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_NE_U32_e64 [[FLAT_ATOMIC_CMPSWAP_RTN]], [[PHI2]], implicit $exec
@@ -180,6 +180,105 @@ define void @cmpxchg(ptr %ptr) {
180180
ret void
181181
}
182182

183+
declare <32 x half> @f()
184+
185+
; Variant of atomicrmw_rel that provoked a crash in SelectionDAG::copyExtraInfo
186+
define void @atomicrmw_rel_deepcopy(ptr %ptr) {
187+
; CHECK-LABEL: name: atomicrmw_rel_deepcopy
188+
; CHECK: bb.0 (%ir-block.0):
189+
; CHECK-NEXT: successors: %bb.1(0x80000000)
190+
; CHECK-NEXT: liveins: $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9, $sgpr10_sgpr11, $sgpr12, $sgpr13, $sgpr14, $sgpr15, $vgpr0, $vgpr1, $vgpr31
191+
; CHECK-NEXT: {{ $}}
192+
; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr31
193+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
194+
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
195+
; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr15
196+
; CHECK-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr14
197+
; CHECK-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr13
198+
; CHECK-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr12
199+
; CHECK-NEXT: [[COPY7:%[0-9]+]]:sgpr_64 = COPY $sgpr10_sgpr11
200+
; CHECK-NEXT: [[COPY8:%[0-9]+]]:sgpr_64 = COPY $sgpr8_sgpr9
201+
; CHECK-NEXT: [[COPY9:%[0-9]+]]:sgpr_64 = COPY $sgpr6_sgpr7
202+
; CHECK-NEXT: [[COPY10:%[0-9]+]]:sgpr_64 = COPY $sgpr4_sgpr5
203+
; CHECK-NEXT: [[DEF:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
204+
; CHECK-NEXT: [[DEF1:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
205+
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
206+
; CHECK-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
207+
; CHECK-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
208+
; CHECK-NEXT: [[SI_PC_ADD_REL_OFFSET:%[0-9]+]]:sreg_64 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-gotprel32-lo) @f, target-flags(amdgpu-gotprel32-hi) @f, implicit-def dead $scc
209+
; CHECK-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
210+
; CHECK-NEXT: [[COPY12:%[0-9]+]]:sgpr_128 = COPY $sgpr0_sgpr1_sgpr2_sgpr3
211+
; CHECK-NEXT: $sgpr4_sgpr5 = COPY [[COPY10]]
212+
; CHECK-NEXT: $sgpr6_sgpr7 = COPY [[COPY9]]
213+
; CHECK-NEXT: $sgpr8_sgpr9 = COPY [[COPY8]]
214+
; CHECK-NEXT: $sgpr10_sgpr11 = COPY [[COPY7]]
215+
; CHECK-NEXT: $sgpr12 = COPY [[COPY6]]
216+
; CHECK-NEXT: $sgpr13 = COPY [[COPY5]]
217+
; CHECK-NEXT: $sgpr14 = COPY [[COPY4]]
218+
; CHECK-NEXT: $sgpr15 = COPY [[COPY3]]
219+
; CHECK-NEXT: $vgpr31 = COPY [[COPY]]
220+
; CHECK-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY12]]
221+
; CHECK-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @f, csr_amdgpu, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit $sgpr15, implicit $vgpr31, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit-def $vgpr0, implicit-def $vgpr1, implicit-def $vgpr2, implicit-def $vgpr3, implicit-def $vgpr4, implicit-def $vgpr5, implicit-def $vgpr6, implicit-def $vgpr7, implicit-def $vgpr8, implicit-def $vgpr9, implicit-def $vgpr10, implicit-def $vgpr11, implicit-def $vgpr12, implicit-def $vgpr13, implicit-def $vgpr14, implicit-def $vgpr15
222+
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
223+
; CHECK-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY $vgpr0
224+
; CHECK-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY $vgpr1
225+
; CHECK-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY $vgpr2
226+
; CHECK-NEXT: [[COPY16:%[0-9]+]]:vgpr_32 = COPY $vgpr3
227+
; CHECK-NEXT: [[COPY17:%[0-9]+]]:vgpr_32 = COPY $vgpr4
228+
; CHECK-NEXT: [[COPY18:%[0-9]+]]:vgpr_32 = COPY $vgpr5
229+
; CHECK-NEXT: [[COPY19:%[0-9]+]]:vgpr_32 = COPY $vgpr6
230+
; CHECK-NEXT: [[COPY20:%[0-9]+]]:vgpr_32 = COPY $vgpr7
231+
; CHECK-NEXT: [[COPY21:%[0-9]+]]:vgpr_32 = COPY $vgpr8
232+
; CHECK-NEXT: [[COPY22:%[0-9]+]]:vgpr_32 = COPY $vgpr9
233+
; CHECK-NEXT: [[COPY23:%[0-9]+]]:vgpr_32 = COPY $vgpr10
234+
; CHECK-NEXT: [[COPY24:%[0-9]+]]:vgpr_32 = COPY $vgpr11
235+
; CHECK-NEXT: [[COPY25:%[0-9]+]]:vgpr_32 = COPY $vgpr12
236+
; CHECK-NEXT: [[COPY26:%[0-9]+]]:vgpr_32 = COPY $vgpr13
237+
; CHECK-NEXT: [[COPY27:%[0-9]+]]:vgpr_32 = COPY $vgpr14
238+
; CHECK-NEXT: [[COPY28:%[0-9]+]]:vgpr_32 = COPY $vgpr15
239+
; CHECK-NEXT: [[COPY29:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
240+
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -4
241+
; CHECK-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY29]], killed [[S_MOV_B32_]], implicit $exec
242+
; CHECK-NEXT: [[DEF2:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
243+
; CHECK-NEXT: [[DEF3:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
244+
; CHECK-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_AND_B32_e64_]], %subreg.sub0, [[COPY11]], %subreg.sub1
245+
; CHECK-NEXT: [[COPY30:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
246+
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 3
247+
; CHECK-NEXT: [[V_AND_B32_e64_1:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY29]], [[S_MOV_B32_1]], implicit $exec
248+
; CHECK-NEXT: [[V_LSHLREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 [[S_MOV_B32_1]], killed [[V_AND_B32_e64_1]], implicit $exec
249+
; CHECK-NEXT: [[S_MOV_B32_2:%[0-9]+]]:sreg_32 = S_MOV_B32 255
250+
; CHECK-NEXT: [[V_LSHLREV_B32_e64_1:%[0-9]+]]:vgpr_32 = V_LSHLREV_B32_e64 killed [[V_LSHLREV_B32_e64_]], killed [[S_MOV_B32_2]], implicit $exec
251+
; CHECK-NEXT: [[V_NOT_B32_e32_:%[0-9]+]]:vgpr_32 = V_NOT_B32_e32 [[V_LSHLREV_B32_e64_1]], implicit $exec
252+
; CHECK-NEXT: [[COPY31:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]], mmra !0
253+
; CHECK-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY31]], 0, 0, implicit $exec, implicit $flat_scr, mmra !0 :: (load (s32) from %ir.AlignedAddr)
254+
; CHECK-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
255+
; CHECK-NEXT: {{ $}}
256+
; CHECK-NEXT: bb.1.atomicrmw.start:
257+
; CHECK-NEXT: successors: %bb.2(0x04000000), %bb.1(0x7c000000)
258+
; CHECK-NEXT: {{ $}}
259+
; CHECK-NEXT: [[PHI:%[0-9]+]]:sreg_64 = PHI [[S_MOV_B64_]], %bb.0, %7, %bb.1
260+
; CHECK-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[FLAT_LOAD_DWORD]], %bb.0, %6, %bb.1
261+
; CHECK-NEXT: [[V_OR_B32_e64_:%[0-9]+]]:vgpr_32 = V_OR_B32_e64 [[V_NOT_B32_e32_]], [[V_LSHLREV_B32_e64_1]], implicit $exec
262+
; CHECK-NEXT: [[V_AND_B32_e64_2:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[PHI1]], killed [[V_OR_B32_e64_]], implicit $exec
263+
; CHECK-NEXT: [[DEF4:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
264+
; CHECK-NEXT: [[DEF5:%[0-9]+]]:sgpr_32 = IMPLICIT_DEF
265+
; CHECK-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_AND_B32_e64_2]], %subreg.sub0, [[PHI1]], %subreg.sub1
266+
; CHECK-NEXT: [[COPY32:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE2]], mmra !0
267+
; CHECK-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY30]], killed [[COPY32]], 0, 1, implicit $exec, implicit $flat_scr, mmra !0 :: (load store release monotonic (s32) on %ir.AlignedAddr)
268+
; CHECK-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_EQ_U32_e64 [[FLAT_ATOMIC_CMPSWAP_RTN]], [[PHI1]], implicit $exec, mmra !0
269+
; CHECK-NEXT: [[SI_IF_BREAK:%[0-9]+]]:sreg_64 = SI_IF_BREAK killed [[V_CMP_EQ_U32_e64_]], [[PHI]], implicit-def dead $scc
270+
; CHECK-NEXT: SI_LOOP [[SI_IF_BREAK]], %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
271+
; CHECK-NEXT: S_BRANCH %bb.2
272+
; CHECK-NEXT: {{ $}}
273+
; CHECK-NEXT: bb.2.atomicrmw.end:
274+
; CHECK-NEXT: [[PHI2:%[0-9]+]]:sreg_64 = PHI [[SI_IF_BREAK]], %bb.1
275+
; CHECK-NEXT: SI_END_CF [[PHI2]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
276+
; CHECK-NEXT: SI_RETURN
277+
%C = call <32 x half> @f()
278+
%old.2 = atomicrmw add ptr %ptr, i8 0 release, align 1, !mmra !0
279+
ret void
280+
}
281+
183282
attributes #0 = { memory(read) }
184283
attributes #1 = { memory(write) }
185284

0 commit comments

Comments
 (0)