Skip to content

Commit 5178574

Browse files
amykhuangamykhuang
committed
Revert "Do actual DCE in LoopUnroll (try 3)"
This reverts commit b6320ee as it causes clang to assert; see https://reviews.llvm.org/rGb6320eeb8622f05e4a5d4c7f5420523357490fca.
1 parent 84a880e commit 5178574

11 files changed

+68
-78
lines changed

‎llvm/lib/Transforms/Utils/LoopUnroll.cpp

Lines changed: 15 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -213,27 +213,33 @@ void llvm::simplifyLoopAfterUnroll(Loop *L, bool SimplifyIVs, LoopInfo *LI,
213213

214214
// Aggressively clean up dead instructions that simplifyLoopIVs already
215215
// identified. Any remaining should be cleaned up below.
216-
RecursivelyDeleteTriviallyDeadInstructions(DeadInsts);
216+
while (!DeadInsts.empty()) {
217+
Value *V = DeadInsts.pop_back_val();
218+
if (Instruction *Inst = dyn_cast_or_null<Instruction>(V))
219+
RecursivelyDeleteTriviallyDeadInstructions(Inst);
220+
}
217221
}
218222

219-
// At this point, the code is well formed. Perform constprop, instsimplify,
220-
// and dce.
223+
// At this point, the code is well formed. We now do a quick sweep over the
224+
// inserted code, doing constant propagation and dead code elimination as we
225+
// go.
221226
const DataLayout &DL = L->getHeader()->getModule()->getDataLayout();
222-
SmallVector<WeakTrackingVH, 16> DeadInsts;
223227
for (BasicBlock *BB : L->getBlocks()) {
224228
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;) {
225229
Instruction *Inst = &*I++;
230+
226231
if (Value *V = SimplifyInstruction(Inst, {DL, nullptr, DT, AC}))
227232
if (LI->replacementPreservesLCSSAForm(Inst, V))
228233
Inst->replaceAllUsesWith(V);
229234
if (isInstructionTriviallyDead(Inst))
230-
DeadInsts.emplace_back(Inst);
235+
BB->getInstList().erase(Inst);
231236
}
232-
// We can't do recursive deletion until we're done iterating, as we might
233-
// have a phi which (potentially indirectly) uses instructions later in
234-
// the block we're iterating through.
235-
RecursivelyDeleteTriviallyDeadInstructions(DeadInsts);
236237
}
238+
239+
// TODO: after peeling or unrolling, previously loop variant conditions are
240+
// likely to fold to constants, eagerly propagating those here will require
241+
// fewer cleanup passes to be run. Alternatively, a LoopEarlyCSE might be
242+
// appropriate.
237243
}
238244

239245
/// Unroll the given loop by Count. The loop must be in LCSSA form. Unrolling

‎llvm/test/Transforms/LoopUnroll/AArch64/full-unroll-trip-count-upper-bound.ll

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -11,13 +11,12 @@
1111
; }
1212
;
1313
; This test is meant to check that this loop is unrolled into four iterations.
14-
; Note that the load on the last iteration is dead and thus doesn't appear in
15-
; the output.
1614

1715
; UNROLL-LABEL: @test
1816
; UNROLL: load i32, i32*
1917
; UNROLL: load i32, i32*
2018
; UNROLL: load i32, i32*
19+
; UNROLL: load i32, i32*
2120
; UNROLL-NOT: load i32, i32*
2221
; NOUNROLL-LABEL: @test
2322
; NOUNROLL: load i32, i32*

‎llvm/test/Transforms/LoopUnroll/dce.ll

Lines changed: 0 additions & 60 deletions
This file was deleted.

‎llvm/test/Transforms/LoopUnroll/full-unroll-invariant.ll

Lines changed: 29 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -34,7 +34,25 @@ define i32 @test2(i8 %a) {
3434
; CHECK-NEXT: entry:
3535
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
3636
; CHECK: for.body:
37-
; CHECK-NEXT: [[ZEXT_9:%.*]] = zext i8 [[A:%.*]] to i32
37+
; CHECK-NEXT: [[ZEXT:%.*]] = zext i8 [[A:%.*]] to i32
38+
; CHECK-NEXT: [[AND:%.*]] = and i32 [[ZEXT]], 31
39+
; CHECK-NEXT: [[ZEXT_1:%.*]] = zext i8 [[A]] to i32
40+
; CHECK-NEXT: [[AND_1:%.*]] = and i32 [[ZEXT_1]], 31
41+
; CHECK-NEXT: [[ZEXT_2:%.*]] = zext i8 [[A]] to i32
42+
; CHECK-NEXT: [[AND_2:%.*]] = and i32 [[ZEXT_2]], 31
43+
; CHECK-NEXT: [[ZEXT_3:%.*]] = zext i8 [[A]] to i32
44+
; CHECK-NEXT: [[AND_3:%.*]] = and i32 [[ZEXT_3]], 31
45+
; CHECK-NEXT: [[ZEXT_4:%.*]] = zext i8 [[A]] to i32
46+
; CHECK-NEXT: [[AND_4:%.*]] = and i32 [[ZEXT_4]], 31
47+
; CHECK-NEXT: [[ZEXT_5:%.*]] = zext i8 [[A]] to i32
48+
; CHECK-NEXT: [[AND_5:%.*]] = and i32 [[ZEXT_5]], 31
49+
; CHECK-NEXT: [[ZEXT_6:%.*]] = zext i8 [[A]] to i32
50+
; CHECK-NEXT: [[AND_6:%.*]] = and i32 [[ZEXT_6]], 31
51+
; CHECK-NEXT: [[ZEXT_7:%.*]] = zext i8 [[A]] to i32
52+
; CHECK-NEXT: [[AND_7:%.*]] = and i32 [[ZEXT_7]], 31
53+
; CHECK-NEXT: [[ZEXT_8:%.*]] = zext i8 [[A]] to i32
54+
; CHECK-NEXT: [[AND_8:%.*]] = and i32 [[ZEXT_8]], 31
55+
; CHECK-NEXT: [[ZEXT_9:%.*]] = zext i8 [[A]] to i32
3856
; CHECK-NEXT: [[AND_9:%.*]] = and i32 [[ZEXT_9]], 31
3957
; CHECK-NEXT: [[SHL_9:%.*]] = shl i32 [[AND_9]], 15
4058
; CHECK-NEXT: ret i32 [[SHL_9]]
@@ -61,7 +79,16 @@ define i32 @test3(i8 %a) {
6179
; CHECK-NEXT: entry:
6280
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
6381
; CHECK: for.body:
64-
; CHECK-NEXT: [[ZEXT_9:%.*]] = zext i8 [[A:%.*]] to i32
82+
; CHECK-NEXT: [[ZEXT:%.*]] = zext i8 [[A:%.*]] to i32
83+
; CHECK-NEXT: [[ZEXT_1:%.*]] = zext i8 [[A]] to i32
84+
; CHECK-NEXT: [[ZEXT_2:%.*]] = zext i8 [[A]] to i32
85+
; CHECK-NEXT: [[ZEXT_3:%.*]] = zext i8 [[A]] to i32
86+
; CHECK-NEXT: [[ZEXT_4:%.*]] = zext i8 [[A]] to i32
87+
; CHECK-NEXT: [[ZEXT_5:%.*]] = zext i8 [[A]] to i32
88+
; CHECK-NEXT: [[ZEXT_6:%.*]] = zext i8 [[A]] to i32
89+
; CHECK-NEXT: [[ZEXT_7:%.*]] = zext i8 [[A]] to i32
90+
; CHECK-NEXT: [[ZEXT_8:%.*]] = zext i8 [[A]] to i32
91+
; CHECK-NEXT: [[ZEXT_9:%.*]] = zext i8 [[A]] to i32
6592
; CHECK-NEXT: [[DIV_9:%.*]] = udiv i32 [[ZEXT_9]], 31
6693
; CHECK-NEXT: ret i32 [[DIV_9]]
6794
;

‎llvm/test/Transforms/LoopUnroll/nonlatchcondbr.ll

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -40,6 +40,7 @@ define void @test1(i32* noalias %A) {
4040
; CHECK: for.body.3:
4141
; CHECK-NEXT: br i1 false, label [[FOR_BODY_FOR_BODY_CRIT_EDGE_3:%.*]], label [[FOR_END:%.*]]
4242
; CHECK: for.body.for.body_crit_edge.3:
43+
; CHECK-NEXT: [[ARRAYIDX_PHI_TRANS_INSERT_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 4
4344
; CHECK-NEXT: unreachable
4445
;
4546
entry:
@@ -123,7 +124,7 @@ define void @test2(i32* noalias %A) {
123124
; CHECK: for.body.for.body_crit_edge.3:
124125
; CHECK-NEXT: [[ARRAYIDX_PHI_TRANS_INSERT_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INC_3]]
125126
; CHECK-NEXT: [[DOTPRE_3]] = load i32, i32* [[ARRAYIDX_PHI_TRANS_INSERT_3]], align 4
126-
; CHECK-NEXT: br label [[FOR_HEADER]], !llvm.loop [[LOOP0:![0-9]+]]
127+
; CHECK-NEXT: br label [[FOR_HEADER]], !llvm.loop !0
127128
;
128129
entry:
129130
br i1 true, label %for.preheader, label %for.end
@@ -201,7 +202,7 @@ define void @test3(i32* noalias %A, i1 %cond) {
201202
; CHECK: for.body.for.body_crit_edge.3:
202203
; CHECK-NEXT: [[ARRAYIDX_PHI_TRANS_INSERT_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INC_3]]
203204
; CHECK-NEXT: [[DOTPRE_3]] = load i32, i32* [[ARRAYIDX_PHI_TRANS_INSERT_3]], align 4
204-
; CHECK-NEXT: br label [[FOR_HEADER]], !llvm.loop [[LOOP2:![0-9]+]]
205+
; CHECK-NEXT: br label [[FOR_HEADER]], !llvm.loop !2
205206
;
206207
entry:
207208
%0 = load i32, i32* %A, align 4

‎llvm/test/Transforms/LoopUnroll/optsize-loop-size.ll

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -12,6 +12,7 @@ define i32 @test(i32 %a, i32 %b, i32 %c) optsize {
1212
; CHECK-NEXT: store i32 [[B:%.*]], i32* [[ARRAYINIT_ELEMENT]], align 4
1313
; CHECK-NEXT: [[ARRAYINIT_ELEMENT1:%.*]] = getelementptr inbounds [3 x i32], [3 x i32]* [[REF_TMP]], i64 0, i64 2
1414
; CHECK-NEXT: store i32 [[C:%.*]], i32* [[ARRAYINIT_ELEMENT1]], align 4
15+
; CHECK-NEXT: [[ADD_PTR_I_I:%.*]] = getelementptr inbounds [3 x i32], [3 x i32]* [[REF_TMP]], i64 0, i64 3
1516
; CHECK-NEXT: [[CMP_I_I_I3:%.*]] = icmp slt i32 [[A]], [[B]]
1617
; CHECK-NEXT: [[SPEC_SELECT_I_I4:%.*]] = select i1 [[CMP_I_I_I3]], i32* [[ARRAYINIT_ELEMENT]], i32* [[ARRAYINIT_BEGIN]]
1718
; CHECK-NEXT: [[INCDEC_PTR_I_I5:%.*]] = getelementptr inbounds [3 x i32], [3 x i32]* [[REF_TMP]], i64 0, i64 2
@@ -21,6 +22,7 @@ define i32 @test(i32 %a, i32 %b, i32 %c) optsize {
2122
; CHECK-NEXT: [[DOTPRE2:%.*]] = load i32, i32* [[INCDEC_PTR_I_I5]], align 4
2223
; CHECK-NEXT: [[CMP_I_I_I:%.*]] = icmp slt i32 [[DOTPRE]], [[DOTPRE2]]
2324
; CHECK-NEXT: [[SPEC_SELECT_I_I:%.*]] = select i1 [[CMP_I_I_I]], i32* [[INCDEC_PTR_I_I5]], i32* [[SPEC_SELECT_I_I4]]
25+
; CHECK-NEXT: [[INCDEC_PTR_I_I:%.*]] = getelementptr inbounds i32, i32* [[INCDEC_PTR_I_I5]], i64 1
2426
; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[SPEC_SELECT_I_I]], align 4
2527
; CHECK-NEXT: ret i32 [[TMP1]]
2628
;

‎llvm/test/Transforms/LoopUnroll/pr45939-peel-count-and-complete-unroll.ll

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -163,6 +163,7 @@ define void @test1() {
163163
; PEEL8-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds [8 x i32], [8 x i32]* @a, i64 0, i64 [[INDVARS_IV_NEXT_6]]
164164
; PEEL8-NEXT: [[TMP15:%.*]] = trunc i64 [[INDVARS_IV_NEXT_6]] to i32
165165
; PEEL8-NEXT: store i32 [[TMP15]], i32* [[ARRAYIDX_7]], align 4
166+
; PEEL8-NEXT: [[INDVARS_IV_NEXT_7:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_6]], 1
166167
; PEEL8-NEXT: br label [[FOR_EXIT]]
167168
; PEEL8: for.exit:
168169
; PEEL8-NEXT: ret void
@@ -205,7 +206,7 @@ define void @test1() {
205206
; PEEL2UNROLL2-NEXT: store i32 [[TMP3]], i32* [[ARRAYIDX_1]], align 4
206207
; PEEL2UNROLL2-NEXT: [[INDVARS_IV_NEXT_1]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
207208
; PEEL2UNROLL2-NEXT: [[EXITCOND_1:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT_1]], 8
208-
; PEEL2UNROLL2-NEXT: br i1 [[EXITCOND_1]], label [[FOR_BODY]], label [[FOR_EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
209+
; PEEL2UNROLL2-NEXT: br i1 [[EXITCOND_1]], label [[FOR_BODY]], label [[FOR_EXIT_LOOPEXIT:%.*]], !llvm.loop !0
209210
; PEEL2UNROLL2: for.exit.loopexit:
210211
; PEEL2UNROLL2-NEXT: br label [[FOR_EXIT]]
211212
; PEEL2UNROLL2: for.exit:

‎llvm/test/Transforms/LoopUnroll/scevunroll.ll

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -10,6 +10,7 @@
1010
define i32 @sansCanonical(i32* %base) nounwind {
1111
; CHECK-LABEL: @sansCanonical(
1212
; CHECK-NEXT: entry:
13+
; CHECK-NEXT: [[ZEXT:%.*]] = zext i32 0 to i64
1314
; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
1415
; CHECK: while.body:
1516
; CHECK-NEXT: [[ADR:%.*]] = getelementptr inbounds i32, i32* [[BASE:%.*]], i64 9
@@ -38,6 +39,7 @@ define i32 @sansCanonical(i32* %base) nounwind {
3839
; CHECK-NEXT: [[ADR_8:%.*]] = getelementptr inbounds i32, i32* [[BASE]], i64 1
3940
; CHECK-NEXT: [[TMP_8:%.*]] = load i32, i32* [[ADR_8]], align 8
4041
; CHECK-NEXT: [[SUM_NEXT_8:%.*]] = add i32 [[SUM_NEXT_7]], [[TMP_8]]
42+
; CHECK-NEXT: [[TMP_9:%.*]] = load i32, i32* [[BASE]], align 8
4143
; CHECK-NEXT: ret i32 [[SUM_NEXT_8]]
4244
;
4345
entry:

‎llvm/test/Transforms/LoopUnroll/unroll-header-exiting-with-phis.ll

Lines changed: 11 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -10,18 +10,23 @@ define i16 @full_unroll(i16* %A) {
1010
; CHECK-NEXT: entry:
1111
; CHECK-NEXT: br label [[FOR_COND:%.*]]
1212
; CHECK: for.cond:
13+
; CHECK-NEXT: [[TMP2:%.*]] = load i16, i16* [[A:%.*]], align 2
1314
; CHECK-NEXT: br label [[FOR_COND_CLEANUP3:%.*]]
1415
; CHECK: for.cond.cleanup:
1516
; CHECK-NEXT: [[DOTLCSSA10_LCSSA:%.*]] = phi i16 [ [[TMP2_2:%.*]], [[FOR_COND_CLEANUP3_2:%.*]] ]
1617
; CHECK-NEXT: [[TMP3:%.*]] = call i16 @func(i16 [[DOTLCSSA10_LCSSA]])
1718
; CHECK-NEXT: ret i16 0
1819
; CHECK: for.cond.cleanup3:
20+
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i16, i16* [[A]], i64 1
21+
; CHECK-NEXT: [[TMP2_1:%.*]] = load i16, i16* [[PTR_1]], align 2
1922
; CHECK-NEXT: br label [[FOR_COND_CLEANUP3_1:%.*]]
2023
; CHECK: for.cond.cleanup3.1:
21-
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds i16, i16* [[A:%.*]], i64 2
24+
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds i16, i16* [[A]], i64 2
2225
; CHECK-NEXT: [[TMP2_2]] = load i16, i16* [[PTR_2]], align 2
2326
; CHECK-NEXT: br label [[FOR_COND_CLEANUP3_2]]
2427
; CHECK: for.cond.cleanup3.2:
28+
; CHECK-NEXT: [[PTR_3:%.*]] = getelementptr inbounds i16, i16* [[A]], i64 3
29+
; CHECK-NEXT: [[TMP2_3:%.*]] = load i16, i16* [[PTR_3]], align 2
2530
; CHECK-NEXT: br i1 false, label [[FOR_COND_CLEANUP3_3:%.*]], label [[FOR_COND_CLEANUP:%.*]]
2631
; CHECK: for.cond.cleanup3.3:
2732
; CHECK-NEXT: unreachable
@@ -54,18 +59,22 @@ define i16 @partial_unroll(i16* %A) {
5459
; CHECK-NEXT: br label [[FOR_COND:%.*]]
5560
; CHECK: for.cond:
5661
; CHECK-NEXT: [[I_0:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC9_2:%.*]], [[FOR_COND_CLEANUP3_2:%.*]] ]
62+
; CHECK-NEXT: [[PTR:%.*]] = getelementptr inbounds i16, i16* [[A:%.*]], i64 [[I_0]]
63+
; CHECK-NEXT: [[TMP2:%.*]] = load i16, i16* [[PTR]], align 2
5764
; CHECK-NEXT: br label [[FOR_COND_CLEANUP3:%.*]]
5865
; CHECK: for.cond.cleanup:
5966
; CHECK-NEXT: [[DOTLCSSA10_LCSSA:%.*]] = phi i16 [ [[TMP2_1:%.*]], [[FOR_COND_CLEANUP3_1:%.*]] ]
6067
; CHECK-NEXT: [[TMP3:%.*]] = call i16 @func(i16 [[DOTLCSSA10_LCSSA]])
6168
; CHECK-NEXT: ret i16 0
6269
; CHECK: for.cond.cleanup3:
6370
; CHECK-NEXT: [[INC9:%.*]] = add nuw nsw i64 [[I_0]], 1
64-
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i16, i16* [[A:%.*]], i64 [[INC9]]
71+
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds i16, i16* [[A]], i64 [[INC9]]
6572
; CHECK-NEXT: [[TMP2_1]] = load i16, i16* [[PTR_1]], align 2
6673
; CHECK-NEXT: br label [[FOR_COND_CLEANUP3_1]]
6774
; CHECK: for.cond.cleanup3.1:
6875
; CHECK-NEXT: [[INC9_1:%.*]] = add nuw nsw i64 [[INC9]], 1
76+
; CHECK-NEXT: [[PTR_2:%.*]] = getelementptr inbounds i16, i16* [[A]], i64 [[INC9_1]]
77+
; CHECK-NEXT: [[TMP2_2:%.*]] = load i16, i16* [[PTR_2]], align 2
6978
; CHECK-NEXT: [[CMP_2:%.*]] = icmp ult i64 [[INC9_1]], 200
7079
; CHECK-NEXT: br i1 [[CMP_2]], label [[FOR_COND_CLEANUP3_2]], label [[FOR_COND_CLEANUP:%.*]]
7180
; CHECK: for.cond.cleanup3.2:

‎llvm/test/Transforms/LoopUnroll/unroll-unconditional-latch.ll

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -68,6 +68,7 @@ define double @test_with_lcssa(double %arg1, double* %arg2) {
6868
; CHECK-NEXT: [[RES_LCSSA:%.*]] = phi double [ [[RES_1]], [[LOOP_LATCH]] ]
6969
; CHECK-NEXT: ret double [[RES_LCSSA]]
7070
; CHECK: loop.latch.1:
71+
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr inbounds double, double* [[ARG2]], i64 2
7172
; CHECK-NEXT: unreachable
7273
;
7374

‎llvm/test/Transforms/LoopUnrollAndJam/unroll-and-jam.ll

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -141,6 +141,8 @@ target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
141141
; CHECK-NEXT: [[ADD_LCSSA_EPIL_2:%.*]] = phi i32 [ [[ADD_EPIL_2]], [[FOR_INNER_EPIL_2]] ]
142142
; CHECK-NEXT: [[ARRAYIDX6_EPIL_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i32 [[ADD8_EPIL_1]]
143143
; CHECK-NEXT: store i32 [[ADD_LCSSA_EPIL_2]], i32* [[ARRAYIDX6_EPIL_2]], align 4, !tbaa !0
144+
; CHECK-NEXT: [[ADD8_EPIL_2:%.*]] = add nuw i32 [[ADD8_EPIL_1]], 1
145+
; CHECK-NEXT: [[EPIL_ITER_SUB_2:%.*]] = sub i32 [[EPIL_ITER_SUB_1]], 1
144146
; CHECK-NEXT: br label [[FOR_END_LOOPEXIT_EPILOG_LCSSA]]
145147
define void @test1(i32 %I, i32 %J, i32* noalias nocapture %A, i32* noalias nocapture readonly %B) #0 {
146148
entry:

0 commit comments

Comments
 (0)