Skip to content

Commit 700b77b

Browse files
authored
[InstCombine] Don't sink if it would require dropping deref assumptions. (#166945)
Currently sinking assumes in instcombine drops assumes if they would prevent sinking. Removing dereferenceable assumptions earlier on can inhibit vectorization of early-exit loops in practice. Special-case deferenceable assumptions so that they block sinking. This can be combined with a separate change to drop dereferencebale assumptions after vectorization: https://clang.godbolt.org/z/jGqcx3sbs PR: #166945
1 parent 63fe60f commit 700b77b

File tree

3 files changed

+153
-64
lines changed

3 files changed

+153
-64
lines changed

llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

Lines changed: 9 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -5627,8 +5627,15 @@ bool InstCombinerImpl::run() {
56275627

56285628
for (Use &U : I->uses()) {
56295629
User *User = U.getUser();
5630-
if (User->isDroppable())
5631-
continue;
5630+
if (User->isDroppable()) {
5631+
// Do not sink if there are dereferenceable assumes that would be
5632+
// removed.
5633+
auto II = dyn_cast<IntrinsicInst>(User);
5634+
if (II->getIntrinsicID() != Intrinsic::assume ||
5635+
!II->getOperandBundle("dereferenceable"))
5636+
continue;
5637+
}
5638+
56325639
if (NumUsers > MaxSinkNumUsers)
56335640
return std::nullopt;
56345641

Lines changed: 98 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1,15 +1,44 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
22
; RUN: opt -p instcombine -S %s | FileCheck %s
33

4-
define i64 @test_sink_with_dereferenceable_assume(ptr %p, ptr %q, i1 %cond) {
5-
; CHECK-LABEL: define i64 @test_sink_with_dereferenceable_assume(
6-
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i1 [[COND:%.*]]) {
4+
define i64 @test_dereferenceable_assume(ptr %p, ptr %q, i1 %c.0) {
5+
; CHECK-LABEL: define i64 @test_dereferenceable_assume(
6+
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i1 [[C_0:%.*]]) {
77
; CHECK-NEXT: [[ENTRY:.*:]]
8-
; CHECK-NEXT: br i1 [[COND]], label %[[THEN:.*]], label %[[ELSE:.*]]
8+
; CHECK-NEXT: [[P_INT:%.*]] = ptrtoint ptr [[P]] to i64
9+
; CHECK-NEXT: [[Q_INT:%.*]] = ptrtoint ptr [[Q]] to i64
10+
; CHECK-NEXT: [[DIFF:%.*]] = sub i64 [[Q_INT]], [[P_INT]]
11+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[DIFF]]) ]
12+
; CHECK-NEXT: br i1 [[C_0]], label %[[THEN:.*]], label %[[ELSE:.*]]
13+
; CHECK: [[THEN]]:
14+
; CHECK-NEXT: ret i64 [[DIFF]]
15+
; CHECK: [[ELSE]]:
16+
; CHECK-NEXT: ret i64 0
17+
;
18+
entry:
19+
%p_int = ptrtoint ptr %p to i64
20+
%q_int = ptrtoint ptr %q to i64
21+
%diff = sub i64 %q_int, %p_int
22+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %diff) ]
23+
br i1 %c.0, label %then, label %else
24+
25+
then:
26+
ret i64 %diff
27+
28+
else:
29+
ret i64 0
30+
}
31+
32+
define i64 @test_sink_with_dereferenceable_assume_same_block_as_user(ptr %p, ptr %q, i1 %c.0) {
33+
; CHECK-LABEL: define i64 @test_sink_with_dereferenceable_assume_same_block_as_user(
34+
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i1 [[C_0:%.*]]) {
35+
; CHECK-NEXT: [[ENTRY:.*:]]
36+
; CHECK-NEXT: br i1 [[C_0]], label %[[THEN:.*]], label %[[ELSE:.*]]
937
; CHECK: [[THEN]]:
1038
; CHECK-NEXT: [[Q_INT:%.*]] = ptrtoint ptr [[Q]] to i64
1139
; CHECK-NEXT: [[P_INT:%.*]] = ptrtoint ptr [[P]] to i64
1240
; CHECK-NEXT: [[DIFF:%.*]] = sub i64 [[Q_INT]], [[P_INT]]
41+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[DIFF]]) ]
1342
; CHECK-NEXT: ret i64 [[DIFF]]
1443
; CHECK: [[ELSE]]:
1544
; CHECK-NEXT: ret i64 0
@@ -18,14 +47,77 @@ entry:
1847
%p_int = ptrtoint ptr %p to i64
1948
%q_int = ptrtoint ptr %q to i64
2049
%diff = sub i64 %q_int, %p_int
21-
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %diff) ]
22-
br i1 %cond, label %then, label %else
50+
br i1 %c.0, label %then, label %else
2351

2452
then:
53+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %diff) ]
2554
ret i64 %diff
2655

2756
else:
2857
ret i64 0
2958
}
3059

60+
define i64 @test_sink_with_multiple_users_dominated_by_deref(ptr %p, ptr %q, i1 %c.0, i1 %c.1) {
61+
; CHECK-LABEL: define i64 @test_sink_with_multiple_users_dominated_by_deref(
62+
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i1 [[C_0:%.*]], i1 [[C_1:%.*]]) {
63+
; CHECK-NEXT: [[ENTRY:.*:]]
64+
; CHECK-NEXT: [[P_INT:%.*]] = ptrtoint ptr [[P]] to i64
65+
; CHECK-NEXT: [[Q_INT:%.*]] = ptrtoint ptr [[Q]] to i64
66+
; CHECK-NEXT: [[DIFF:%.*]] = sub i64 [[Q_INT]], [[P_INT]]
67+
; CHECK-NEXT: br i1 [[C_0]], label %[[THEN:.*]], label %[[ELSE:.*]]
68+
; CHECK: [[THEN]]:
69+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[DIFF]]) ]
70+
; CHECK-NEXT: br i1 [[C_1]], label %[[THEN_2:.*]], label %[[ELSE]]
71+
; CHECK: [[THEN_2]]:
72+
; CHECK-NEXT: [[DOUBLED:%.*]] = shl i64 [[DIFF]], 1
73+
; CHECK-NEXT: ret i64 [[DOUBLED]]
74+
; CHECK: [[ELSE]]:
75+
; CHECK-NEXT: ret i64 0
76+
;
77+
entry:
78+
%p_int = ptrtoint ptr %p to i64
79+
%q_int = ptrtoint ptr %q to i64
80+
%diff = sub i64 %q_int, %p_int
81+
br i1 %c.0, label %then, label %else
82+
83+
then:
84+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %diff) ]
85+
br i1 %c.1, label %then.2, label %else
86+
87+
then.2:
88+
%doubled = mul i64 %diff, 2
89+
ret i64 %doubled
90+
91+
else:
92+
ret i64 0
93+
}
94+
95+
define i64 @test_deref_user_does_not_dominate_other_user(ptr %p, ptr %q, i1 %c.0) {
96+
; CHECK-LABEL: define i64 @test_deref_user_does_not_dominate_other_user(
97+
; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i1 [[C_0:%.*]]) {
98+
; CHECK-NEXT: [[ENTRY:.*:]]
99+
; CHECK-NEXT: [[P_INT:%.*]] = ptrtoint ptr [[P]] to i64
100+
; CHECK-NEXT: [[Q_INT:%.*]] = ptrtoint ptr [[Q]] to i64
101+
; CHECK-NEXT: [[DIFF:%.*]] = sub i64 [[Q_INT]], [[P_INT]]
102+
; CHECK-NEXT: br i1 [[C_0]], label %[[MIDDLE:.*]], label %[[EXIT:.*]]
103+
; CHECK: [[MIDDLE]]:
104+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[DIFF]]) ]
105+
; CHECK-NEXT: br label %[[EXIT]]
106+
; CHECK: [[EXIT]]:
107+
; CHECK-NEXT: ret i64 [[DIFF]]
108+
;
109+
entry:
110+
%p_int = ptrtoint ptr %p to i64
111+
%q_int = ptrtoint ptr %q to i64
112+
%diff = sub i64 %q_int, %p_int
113+
br i1 %c.0, label %middle, label %exit
114+
115+
middle:
116+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %diff) ]
117+
br label %exit
118+
119+
exit:
120+
ret i64 %diff
121+
}
122+
31123
declare void @llvm.assume(i1 noundef)

llvm/test/Transforms/PhaseOrdering/AArch64/std-find.ll

Lines changed: 46 additions & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -133,75 +133,65 @@ define ptr @std_find_caller(ptr noundef %first, ptr noundef %last) {
133133
; CHECK-LABEL: define noundef ptr @std_find_caller(
134134
; CHECK-SAME: ptr noundef [[FIRST:%.*]], ptr noundef [[LAST:%.*]]) local_unnamed_addr #[[ATTR0]] {
135135
; CHECK-NEXT: [[ENTRY:.*]]:
136+
; CHECK-NEXT: [[FIRST3:%.*]] = ptrtoint ptr [[FIRST]] to i64
137+
; CHECK-NEXT: [[LAST_I64:%.*]] = ptrtoint ptr [[LAST]] to i64
138+
; CHECK-NEXT: [[PTR_SUB:%.*]] = sub i64 [[LAST_I64]], [[FIRST3]]
136139
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[FIRST]], i64 2) ]
137140
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[LAST]], i64 2) ]
141+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[FIRST]], i64 [[PTR_SUB]]) ]
138142
; CHECK-NEXT: [[PRE_I:%.*]] = icmp eq ptr [[FIRST]], [[LAST]]
139143
; CHECK-NEXT: br i1 [[PRE_I]], label %[[STD_FIND_GENERIC_IMPL_EXIT:.*]], label %[[LOOP_HEADER_I_PREHEADER:.*]]
140144
; CHECK: [[LOOP_HEADER_I_PREHEADER]]:
141-
; CHECK-NEXT: [[LAST2:%.*]] = ptrtoint ptr [[LAST]] to i64
142-
; CHECK-NEXT: [[FIRST3:%.*]] = ptrtoint ptr [[FIRST]] to i64
143-
; CHECK-NEXT: [[LAST_I64:%.*]] = ptrtoint ptr [[LAST]] to i64
144-
; CHECK-NEXT: [[FIRST1:%.*]] = ptrtoint ptr [[FIRST]] to i64
145-
; CHECK-NEXT: [[PTR_SUB:%.*]] = sub i64 [[LAST_I64]], [[FIRST1]]
146145
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[FIRST]], i64 [[PTR_SUB]]
147-
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[LAST2]], -2
146+
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[LAST_I64]], -2
148147
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[FIRST3]]
149148
; CHECK-NEXT: [[TMP2:%.*]] = lshr exact i64 [[TMP1]], 1
150149
; CHECK-NEXT: [[TMP3:%.*]] = add nuw i64 [[TMP2]], 1
151-
; CHECK-NEXT: [[XTRAITER:%.*]] = and i64 [[TMP3]], 3
152-
; CHECK-NEXT: [[TMP4:%.*]] = and i64 [[TMP1]], 6
153-
; CHECK-NEXT: [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[TMP4]], 6
154-
; CHECK-NEXT: br i1 [[LCMP_MOD_NOT]], label %[[LOOP_HEADER_I_PROL_LOOPEXIT:.*]], label %[[LOOP_HEADER_I_PROL:.*]]
155-
; CHECK: [[LOOP_HEADER_I_PROL]]:
156-
; CHECK-NEXT: [[PTR_IV_I_PROL:%.*]] = phi ptr [ [[PTR_IV_NEXT_I_PROL:%.*]], %[[LOOP_LATCH_I_PROL:.*]] ], [ [[FIRST]], %[[LOOP_HEADER_I_PREHEADER]] ]
157-
; CHECK-NEXT: [[PROL_ITER:%.*]] = phi i64 [ [[PROL_ITER_NEXT:%.*]], %[[LOOP_LATCH_I_PROL]] ], [ 0, %[[LOOP_HEADER_I_PREHEADER]] ]
158-
; CHECK-NEXT: [[L_I_PROL:%.*]] = load i16, ptr [[PTR_IV_I_PROL]], align 2
159-
; CHECK-NEXT: [[C_1_I_PROL:%.*]] = icmp eq i16 [[L_I_PROL]], 1
160-
; CHECK-NEXT: br i1 [[C_1_I_PROL]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_LATCH_I_PROL]]
161-
; CHECK: [[LOOP_LATCH_I_PROL]]:
162-
; CHECK-NEXT: [[PTR_IV_NEXT_I_PROL]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I_PROL]], i64 2
163-
; CHECK-NEXT: [[PROL_ITER_NEXT]] = add i64 [[PROL_ITER]], 1
150+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP1]], 158
151+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[LOOP_HEADER_I_PREHEADER2:.*]], label %[[VECTOR_PH:.*]]
152+
; CHECK: [[VECTOR_PH]]:
153+
; CHECK-NEXT: [[XTRAITER:%.*]] = and i64 [[TMP3]], -8
154+
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
155+
; CHECK: [[VECTOR_BODY]]:
156+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[PROL_ITER_NEXT:%.*]], %[[VECTOR_BODY]] ]
157+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
158+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[FIRST]], i64 [[OFFSET_IDX]]
159+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2
160+
; CHECK-NEXT: [[WIDE_LOAD_FR:%.*]] = freeze <8 x i16> [[WIDE_LOAD]]
161+
; CHECK-NEXT: [[TMP4:%.*]] = icmp eq <8 x i16> [[WIDE_LOAD_FR]], splat (i16 1)
162+
; CHECK-NEXT: [[PROL_ITER_NEXT]] = add nuw i64 [[INDEX]], 8
163+
; CHECK-NEXT: [[TMP5:%.*]] = bitcast <8 x i1> [[TMP4]] to i8
164+
; CHECK-NEXT: [[TMP6:%.*]] = icmp ne i8 [[TMP5]], 0
164165
; CHECK-NEXT: [[PROL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[PROL_ITER_NEXT]], [[XTRAITER]]
165-
; CHECK-NEXT: br i1 [[PROL_ITER_CMP_NOT]], label %[[LOOP_HEADER_I_PROL_LOOPEXIT]], label %[[LOOP_HEADER_I_PROL]], !llvm.loop [[LOOP3:![0-9]+]]
166-
; CHECK: [[LOOP_HEADER_I_PROL_LOOPEXIT]]:
167-
; CHECK-NEXT: [[PTR_IV_I_UNR:%.*]] = phi ptr [ [[FIRST]], %[[LOOP_HEADER_I_PREHEADER]] ], [ [[PTR_IV_NEXT_I_PROL]], %[[LOOP_LATCH_I_PROL]] ]
168-
; CHECK-NEXT: [[TMP5:%.*]] = icmp ult i64 [[TMP1]], 6
169-
; CHECK-NEXT: br i1 [[TMP5]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_HEADER_I:.*]]
166+
; CHECK-NEXT: [[TMP8:%.*]] = or i1 [[TMP6]], [[PROL_ITER_CMP_NOT]]
167+
; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_SPLIT:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
168+
; CHECK: [[MIDDLE_SPLIT]]:
169+
; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[XTRAITER]], 1
170+
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[FIRST]], i64 [[TMP9]]
171+
; CHECK-NEXT: br i1 [[TMP6]], label %[[VECTOR_EARLY_EXIT:.*]], label %[[MIDDLE_BLOCK:.*]]
172+
; CHECK: [[MIDDLE_BLOCK]]:
173+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[XTRAITER]]
174+
; CHECK-NEXT: br i1 [[CMP_N]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_HEADER_I_PREHEADER2]]
175+
; CHECK: [[LOOP_HEADER_I_PREHEADER2]]:
176+
; CHECK-NEXT: [[PTR_IV_I_PH:%.*]] = phi ptr [ [[FIRST]], %[[LOOP_HEADER_I_PREHEADER]] ], [ [[TMP10]], %[[MIDDLE_BLOCK]] ]
177+
; CHECK-NEXT: br label %[[LOOP_HEADER_I:.*]]
178+
; CHECK: [[VECTOR_EARLY_EXIT]]:
179+
; CHECK-NEXT: [[TMP11:%.*]] = tail call i64 @llvm.experimental.cttz.elts.i64.v8i1(<8 x i1> [[TMP4]], i1 true)
180+
; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], [[TMP11]]
181+
; CHECK-NEXT: [[TMP13:%.*]] = shl i64 [[TMP12]], 1
182+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[FIRST]], i64 [[TMP13]]
183+
; CHECK-NEXT: br label %[[STD_FIND_GENERIC_IMPL_EXIT]]
170184
; CHECK: [[LOOP_HEADER_I]]:
171-
; CHECK-NEXT: [[PTR_IV_I:%.*]] = phi ptr [ [[PTR_IV_NEXT_I_3:%.*]], %[[LOOP_LATCH_I_3:.*]] ], [ [[PTR_IV_I_UNR]], %[[LOOP_HEADER_I_PROL_LOOPEXIT]] ]
185+
; CHECK-NEXT: [[PTR_IV_I:%.*]] = phi ptr [ [[PTR_IV_NEXT_I:%.*]], %[[LOOP_LATCH_I:.*]] ], [ [[PTR_IV_I_PH]], %[[LOOP_HEADER_I_PREHEADER2]] ]
172186
; CHECK-NEXT: [[L_I:%.*]] = load i16, ptr [[PTR_IV_I]], align 2
173187
; CHECK-NEXT: [[C_1_I:%.*]] = icmp eq i16 [[L_I]], 1
174-
; CHECK-NEXT: br i1 [[C_1_I]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_LATCH_I:.*]]
188+
; CHECK-NEXT: br i1 [[C_1_I]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_LATCH_I]]
175189
; CHECK: [[LOOP_LATCH_I]]:
176-
; CHECK-NEXT: [[PTR_IV_NEXT_I:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 2
177-
; CHECK-NEXT: [[L_I_1:%.*]] = load i16, ptr [[PTR_IV_NEXT_I]], align 2
178-
; CHECK-NEXT: [[C_1_I_1:%.*]] = icmp eq i16 [[L_I_1]], 1
179-
; CHECK-NEXT: br i1 [[C_1_I_1]], label %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT11:.*]], label %[[LOOP_LATCH_I_1:.*]]
180-
; CHECK: [[LOOP_LATCH_I_1]]:
181-
; CHECK-NEXT: [[PTR_IV_NEXT_I_1:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 4
182-
; CHECK-NEXT: [[L_I_2:%.*]] = load i16, ptr [[PTR_IV_NEXT_I_1]], align 2
183-
; CHECK-NEXT: [[C_1_I_2:%.*]] = icmp eq i16 [[L_I_2]], 1
184-
; CHECK-NEXT: br i1 [[C_1_I_2]], label %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT9:.*]], label %[[LOOP_LATCH_I_2:.*]]
185-
; CHECK: [[LOOP_LATCH_I_2]]:
186-
; CHECK-NEXT: [[PTR_IV_NEXT_I_2:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 6
187-
; CHECK-NEXT: [[L_I_3:%.*]] = load i16, ptr [[PTR_IV_NEXT_I_2]], align 2
188-
; CHECK-NEXT: [[C_1_I_3:%.*]] = icmp eq i16 [[L_I_3]], 1
189-
; CHECK-NEXT: br i1 [[C_1_I_3]], label %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT7:.*]], label %[[LOOP_LATCH_I_3]]
190-
; CHECK: [[LOOP_LATCH_I_3]]:
191-
; CHECK-NEXT: [[PTR_IV_NEXT_I_3]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 8
192-
; CHECK-NEXT: [[EC_I_3:%.*]] = icmp eq ptr [[PTR_IV_NEXT_I_3]], [[LAST]]
193-
; CHECK-NEXT: br i1 [[EC_I_3]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_HEADER_I]]
194-
; CHECK: [[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT7]]:
195-
; CHECK-NEXT: [[PTR_IV_NEXT_I_2_LE:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 6
196-
; CHECK-NEXT: br label %[[STD_FIND_GENERIC_IMPL_EXIT]]
197-
; CHECK: [[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT9]]:
198-
; CHECK-NEXT: [[PTR_IV_NEXT_I_1_LE:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 4
199-
; CHECK-NEXT: br label %[[STD_FIND_GENERIC_IMPL_EXIT]]
200-
; CHECK: [[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT11]]:
201-
; CHECK-NEXT: [[PTR_IV_NEXT_I_LE:%.*]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 2
202-
; CHECK-NEXT: br label %[[STD_FIND_GENERIC_IMPL_EXIT]]
190+
; CHECK-NEXT: [[PTR_IV_NEXT_I]] = getelementptr inbounds nuw i8, ptr [[PTR_IV_I]], i64 2
191+
; CHECK-NEXT: [[EC_I:%.*]] = icmp eq ptr [[PTR_IV_NEXT_I]], [[LAST]]
192+
; CHECK-NEXT: br i1 [[EC_I]], label %[[STD_FIND_GENERIC_IMPL_EXIT]], label %[[LOOP_HEADER_I]], !llvm.loop [[LOOP4:![0-9]+]]
203193
; CHECK: [[STD_FIND_GENERIC_IMPL_EXIT]]:
204-
; CHECK-NEXT: [[RES_I:%.*]] = phi ptr [ [[FIRST]], %[[ENTRY]] ], [ [[SCEVGEP]], %[[LOOP_HEADER_I_PROL_LOOPEXIT]] ], [ [[PTR_IV_NEXT_I_2_LE]], %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT7]] ], [ [[PTR_IV_NEXT_I_1_LE]], %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT9]] ], [ [[PTR_IV_NEXT_I_LE]], %[[STD_FIND_GENERIC_IMPL_EXIT_LOOPEXIT_UNR_LCSSA_LOOPEXIT_SPLIT_LOOP_EXIT11]] ], [ [[SCEVGEP]], %[[LOOP_LATCH_I_3]] ], [ [[PTR_IV_I]], %[[LOOP_HEADER_I]] ], [ [[PTR_IV_I_PROL]], %[[LOOP_HEADER_I_PROL]] ]
194+
; CHECK-NEXT: [[RES_I:%.*]] = phi ptr [ [[FIRST]], %[[ENTRY]] ], [ [[SCEVGEP]], %[[MIDDLE_BLOCK]] ], [ [[TMP14]], %[[VECTOR_EARLY_EXIT]] ], [ [[SCEVGEP]], %[[LOOP_LATCH_I]] ], [ [[PTR_IV_I]], %[[LOOP_HEADER_I]] ]
205195
; CHECK-NEXT: ret ptr [[RES_I]]
206196
;
207197
entry:
@@ -241,6 +231,6 @@ declare void @llvm.assume(i1 noundef)
241231
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
242232
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
243233
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
244-
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META4:![0-9]+]]}
245-
; CHECK: [[META4]] = !{!"llvm.loop.unroll.disable"}
234+
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]], [[META2]]}
235+
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META2]], [[META1]]}
246236
;.

0 commit comments

Comments
 (0)