Skip to content

Commit 286e98b

Browse files
fhahnfhahn
committed
[DSE] Add test cases with more complex redundant stores.
This patch adds more complex test cases with redundant stores of an existing memset, with other stores in between. It also makes a few of the existing tests more robust.
1 parent ff569ed commit 286e98b

File tree

3 files changed

+244
-29
lines changed

3 files changed

+244
-29
lines changed

‎llvm/test/Transforms/DeadStoreElimination/memset-unknown-sizes.ll

Lines changed: 21 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -6,17 +6,17 @@ declare i8* @_Znwm() local_unnamed_addr #0
66
; Function Attrs: argmemonly nounwind willreturn writeonly
77
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #1
88

9-
define void @test1(i1 %c) {
9+
define void @test1(i1 %c, i64 %N) {
1010
; CHECK-LABEL: @test1(
1111
; CHECK-NEXT: entry:
1212
; CHECK-NEXT: br i1 [[C:%.*]], label [[COND_TRUE_I_I_I:%.*]], label [[COND_END_I_I_I:%.*]]
1313
; CHECK: cond.true.i.i.i:
1414
; CHECK-NEXT: ret void
1515
; CHECK: cond.end.i.i.i:
16-
; CHECK-NEXT: [[CALL_I_I_I_I_I:%.*]] = tail call noalias nonnull i8* @_Znam()
17-
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i8* [[CALL_I_I_I_I_I]] to i64*
18-
; CHECK-NEXT: tail call void @llvm.memset.p0i8.i64(i8* nonnull align 8 [[CALL_I_I_I_I_I]], i8 0, i64 undef, i1 false)
19-
; CHECK-NEXT: store i64 0, i64* [[TMP0]], align 8
16+
; CHECK-NEXT: [[ALLOC:%.*]] = tail call noalias nonnull i8* @_Znam() #[[ATTR2:[0-9]+]]
17+
; CHECK-NEXT: [[ALLOC_BC:%.*]] = bitcast i8* [[ALLOC]] to i64*
18+
; CHECK-NEXT: tail call void @llvm.memset.p0i8.i64(i8* nonnull align 8 [[ALLOC]], i8 0, i64 [[N:%.*]], i1 false) #[[ATTR3:[0-9]+]]
19+
; CHECK-NEXT: store i64 0, i64* [[ALLOC_BC]], align 8
2020
; CHECK-NEXT: ret void
2121
;
2222
entry:
@@ -26,26 +26,26 @@ cond.true.i.i.i: ; preds = %entry
2626
ret void
2727

2828
cond.end.i.i.i: ; preds = %entry
29-
%call.i.i.i.i.i = tail call noalias nonnull i8* @_Znam() #2
30-
%0 = bitcast i8* %call.i.i.i.i.i to i64*
31-
tail call void @llvm.memset.p0i8.i64(i8* nonnull align 8 %call.i.i.i.i.i, i8 0, i64 undef, i1 false) #3
32-
store i64 0, i64* %0, align 8
29+
%alloc = tail call noalias nonnull i8* @_Znam() #2
30+
%alloc.bc = bitcast i8* %alloc to i64*
31+
tail call void @llvm.memset.p0i8.i64(i8* nonnull align 8 %alloc, i8 0, i64 %N, i1 false) #3
32+
store i64 0, i64* %alloc.bc, align 8
3333
ret void
3434
}
3535

36-
declare i8* @_Znam() local_unnamed_addr #0
36+
declare i8* @_Znam()
3737

3838

39-
define void @test2(i1 %c) {
39+
define void @test2(i1 %c, i64 %N) {
4040
; CHECK-LABEL: @test2(
4141
; CHECK-NEXT: entry:
4242
; CHECK-NEXT: br i1 [[C:%.*]], label [[CLEANUP_CONT104:%.*]], label [[IF_THEN:%.*]]
4343
; CHECK: if.then:
44-
; CHECK-NEXT: [[MUL_I_I_I_I:%.*]] = shl nuw nsw i64 undef, 3
45-
; CHECK-NEXT: [[CALL_I_I_I_I_I_I131:%.*]] = call noalias nonnull i8* @_Znwm()
46-
; CHECK-NEXT: [[DOTCAST_I_I:%.*]] = bitcast i8* [[CALL_I_I_I_I_I_I131]] to i64*
47-
; CHECK-NEXT: store i64 0, i64* [[DOTCAST_I_I]], align 8
48-
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* nonnull align 8 [[CALL_I_I_I_I_I_I131]], i8 0, i64 [[MUL_I_I_I_I]], i1 false)
44+
; CHECK-NEXT: [[MUL:%.*]] = shl nuw nsw i64 [[N:%.*]], 3
45+
; CHECK-NEXT: [[ALLOC:%.*]] = call noalias nonnull i8* @_Znwm() #[[ATTR2]]
46+
; CHECK-NEXT: [[ALLOC_BC:%.*]] = bitcast i8* [[ALLOC]] to i64*
47+
; CHECK-NEXT: store i64 0, i64* [[ALLOC_BC]], align 8
48+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* nonnull align 8 [[ALLOC]], i8 0, i64 [[MUL]], i1 false) #[[ATTR3]]
4949
; CHECK-NEXT: ret void
5050
; CHECK: cleanup.cont104:
5151
; CHECK-NEXT: ret void
@@ -54,11 +54,11 @@ entry:
5454
br i1 %c, label %cleanup.cont104, label %if.then
5555

5656
if.then: ; preds = %entry
57-
%mul.i.i.i.i = shl nuw nsw i64 undef, 3
58-
%call.i.i.i.i.i.i131 = call noalias nonnull i8* @_Znwm() #2
59-
%.cast.i.i = bitcast i8* %call.i.i.i.i.i.i131 to i64*
60-
store i64 0, i64* %.cast.i.i, align 8
61-
call void @llvm.memset.p0i8.i64(i8* nonnull align 8 %call.i.i.i.i.i.i131, i8 0, i64 %mul.i.i.i.i, i1 false) #3
57+
%mul = shl nuw nsw i64 %N, 3
58+
%alloc = call noalias nonnull i8* @_Znwm() #2
59+
%alloc.bc = bitcast i8* %alloc to i64*
60+
store i64 0, i64* %alloc.bc, align 8
61+
call void @llvm.memset.p0i8.i64(i8* nonnull align 8 %alloc, i8 0, i64 %mul, i1 false) #3
6262
ret void
6363

6464
cleanup.cont104: ; preds = %entry

‎llvm/test/Transforms/DeadStoreElimination/multiblock-throwing.ll

Lines changed: 31 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -4,17 +4,17 @@
44
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
55
declare void @unknown_func()
66

7-
define void @test6(i32* noalias %P) {
8-
; CHECK-LABEL: @test6(
9-
; CHECK-NEXT: store i32 0, i32* [[P:%.*]]
7+
define void @test6_store_same_value(i32* noalias %P) {
8+
; CHECK-LABEL: @test6_store_same_value(
9+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
1010
; CHECK-NEXT: br i1 true, label [[BB1:%.*]], label [[BB2:%.*]]
1111
; CHECK: bb1:
1212
; CHECK-NEXT: br label [[BB3:%.*]]
1313
; CHECK: bb2:
1414
; CHECK-NEXT: call void @unknown_func()
1515
; CHECK-NEXT: br label [[BB3]]
1616
; CHECK: bb3:
17-
; CHECK-NEXT: store i32 0, i32* [[P]]
17+
; CHECK-NEXT: store i32 0, i32* [[P]], align 4
1818
; CHECK-NEXT: ret void
1919
;
2020
store i32 0, i32* %P
@@ -29,6 +29,31 @@ bb3:
2929
ret void
3030
}
3131

32+
define void @test6_store_other_value(i32* noalias %P) {
33+
; CHECK-LABEL: @test6_store_other_value(
34+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
35+
; CHECK-NEXT: br i1 true, label [[BB1:%.*]], label [[BB2:%.*]]
36+
; CHECK: bb1:
37+
; CHECK-NEXT: br label [[BB3:%.*]]
38+
; CHECK: bb2:
39+
; CHECK-NEXT: call void @unknown_func()
40+
; CHECK-NEXT: br label [[BB3]]
41+
; CHECK: bb3:
42+
; CHECK-NEXT: store i32 1, i32* [[P]], align 4
43+
; CHECK-NEXT: ret void
44+
;
45+
store i32 0, i32* %P
46+
br i1 true, label %bb1, label %bb2
47+
bb1:
48+
br label %bb3
49+
bb2:
50+
call void @unknown_func()
51+
br label %bb3
52+
bb3:
53+
store i32 1, i32* %P
54+
ret void
55+
}
56+
3257
define void @test23(i32* noalias %P) {
3358
; CHECK-LABEL: @test23(
3459
; CHECK-NEXT: br i1 true, label [[BB1:%.*]], label [[BB2:%.*]]
@@ -38,7 +63,7 @@ define void @test23(i32* noalias %P) {
3863
; CHECK-NEXT: call void @unknown_func()
3964
; CHECK-NEXT: br label [[BB3]]
4065
; CHECK: bb3:
41-
; CHECK-NEXT: store i32 0, i32* [[P:%.*]]
66+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
4267
; CHECK-NEXT: ret void
4368
;
4469
br i1 true, label %bb1, label %bb2
@@ -63,7 +88,7 @@ define void @test24(i32* noalias %P) {
6388
; CHECK-NEXT: call void @unknown_func()
6489
; CHECK-NEXT: br label [[BB3]]
6590
; CHECK: bb3:
66-
; CHECK-NEXT: store i32 0, i32* [[P:%.*]]
91+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
6792
; CHECK-NEXT: ret void
6893
;
6994
br i1 true, label %bb2, label %bb1

‎llvm/test/Transforms/DeadStoreElimination/stores-of-existing-values.ll

Lines changed: 192 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -235,7 +235,7 @@ bb3:
235235
}
236236

237237
; Make sure the store in %bb3 won't be eliminated because it may be clobbered before.
238-
define void @test8(i32* noalias %P) {
238+
define void @test8(i32* %P) {
239239
; CHECK-LABEL: @test8(
240240
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
241241
; CHECK-NEXT: br i1 true, label [[BB1:%.*]], label [[BB2:%.*]]
@@ -287,6 +287,197 @@ bb3:
287287
ret void
288288
}
289289

290+
; The store in bb3 can be eliminated, because the store in bb1 cannot alias it.
291+
define void @test10(i32* noalias %P, i32* %Q, i1 %c) {
292+
; CHECK-LABEL: @test10(
293+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
294+
; CHECK-NEXT: br i1 [[C:%.*]], label [[BB1:%.*]], label [[BB2:%.*]]
295+
; CHECK: bb1:
296+
; CHECK-NEXT: store i32 10, i32* [[Q:%.*]], align 4
297+
; CHECK-NEXT: br label [[BB3:%.*]]
298+
; CHECK: bb2:
299+
; CHECK-NEXT: ret void
300+
; CHECK: bb3:
301+
; CHECK-NEXT: store i32 0, i32* [[P]], align 4
302+
; CHECK-NEXT: ret void
303+
;
304+
store i32 0, i32* %P
305+
br i1 %c, label %bb1, label %bb2
306+
307+
bb1:
308+
store i32 10, i32* %Q
309+
br label %bb3
310+
311+
bb2:
312+
ret void
313+
314+
bb3:
315+
store i32 0, i32* %P
316+
ret void
317+
}
318+
319+
define void @test11_smaller_later_store(i32* noalias %P, i32* %Q, i1 %c) {
320+
; CHECK-LABEL: @test11_smaller_later_store(
321+
; CHECK-NEXT: store i32 0, i32* [[P:%.*]], align 4
322+
; CHECK-NEXT: br i1 [[C:%.*]], label [[BB1:%.*]], label [[BB2:%.*]]
323+
; CHECK: bb1:
324+
; CHECK-NEXT: br label [[BB3:%.*]]
325+
; CHECK: bb2:
326+
; CHECK-NEXT: ret void
327+
; CHECK: bb3:
328+
; CHECK-NEXT: [[BC:%.*]] = bitcast i32* [[P]] to i8*
329+
; CHECK-NEXT: store i8 0, i8* [[BC]], align 1
330+
; CHECK-NEXT: ret void
331+
;
332+
store i32 0, i32* %P
333+
br i1 %c, label %bb1, label %bb2
334+
335+
bb1:
336+
br label %bb3
337+
338+
bb2:
339+
ret void
340+
341+
bb3:
342+
%bc = bitcast i32* %P to i8*
343+
store i8 0, i8* %bc
344+
ret void
345+
}
346+
347+
define void @test11_smaller_earlier_store(i32* noalias %P, i32* %Q, i1 %c) {
348+
; CHECK-LABEL: @test11_smaller_earlier_store(
349+
; CHECK-NEXT: [[BC:%.*]] = bitcast i32* [[P:%.*]] to i8*
350+
; CHECK-NEXT: store i8 0, i8* [[BC]], align 1
351+
; CHECK-NEXT: br i1 [[C:%.*]], label [[BB1:%.*]], label [[BB2:%.*]]
352+
; CHECK: bb1:
353+
; CHECK-NEXT: br label [[BB3:%.*]]
354+
; CHECK: bb2:
355+
; CHECK-NEXT: ret void
356+
; CHECK: bb3:
357+
; CHECK-NEXT: store i32 0, i32* [[P]], align 4
358+
; CHECK-NEXT: ret void
359+
;
360+
%bc = bitcast i32* %P to i8*
361+
store i8 0, i8* %bc
362+
br i1 %c, label %bb1, label %bb2
363+
364+
bb1:
365+
br label %bb3
366+
367+
bb2:
368+
ret void
369+
370+
bb3:
371+
store i32 0, i32* %P
372+
ret void
373+
}
374+
375+
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #1
376+
377+
define void @test12_memset_simple(i8* %ptr) {
378+
; CHECK-LABEL: @test12_memset_simple(
379+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[PTR:%.*]], i8 0, i64 10, i1 false)
380+
; CHECK-NEXT: [[PTR_5:%.*]] = getelementptr i8, i8* [[PTR]], i64 4
381+
; CHECK-NEXT: store i8 0, i8* [[PTR_5]], align 1
382+
; CHECK-NEXT: ret void
383+
;
384+
call void @llvm.memset.p0i8.i64(i8* %ptr, i8 0, i64 10, i1 false)
385+
%ptr.5 = getelementptr i8, i8* %ptr, i64 4
386+
store i8 0, i8* %ptr.5
387+
ret void
388+
}
389+
390+
define void @test12_memset_other_store_in_between(i8* %ptr) {
391+
; CHECK-LABEL: @test12_memset_other_store_in_between(
392+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[PTR:%.*]], i8 0, i64 10, i1 false)
393+
; CHECK-NEXT: [[PTR_4:%.*]] = getelementptr i8, i8* [[PTR]], i64 4
394+
; CHECK-NEXT: store i8 8, i8* [[PTR_4]], align 1
395+
; CHECK-NEXT: [[PTR_5:%.*]] = getelementptr i8, i8* [[PTR]], i64 5
396+
; CHECK-NEXT: store i8 0, i8* [[PTR_5]], align 1
397+
; CHECK-NEXT: ret void
398+
;
399+
call void @llvm.memset.p0i8.i64(i8* %ptr, i8 0, i64 10, i1 false)
400+
%ptr.4 = getelementptr i8, i8* %ptr, i64 4
401+
store i8 8, i8* %ptr.4
402+
%ptr.5 = getelementptr i8, i8* %ptr, i64 5
403+
store i8 0, i8* %ptr.5
404+
ret void
405+
}
406+
407+
define void @test12_memset_other_store_in_between_partial_overlap(i8* %ptr) {
408+
; CHECK-LABEL: @test12_memset_other_store_in_between_partial_overlap(
409+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[PTR:%.*]], i8 0, i64 10, i1 false)
410+
; CHECK-NEXT: [[PTR_4:%.*]] = getelementptr i8, i8* [[PTR]], i64 4
411+
; CHECK-NEXT: [[BC_4:%.*]] = bitcast i8* [[PTR_4]] to i16*
412+
; CHECK-NEXT: store i16 8, i16* [[BC_4]], align 2
413+
; CHECK-NEXT: [[PTR_5:%.*]] = getelementptr i8, i8* [[PTR]], i64 5
414+
; CHECK-NEXT: [[BC_5:%.*]] = bitcast i8* [[PTR_5]] to i16*
415+
; CHECK-NEXT: store i16 0, i16* [[BC_5]], align 2
416+
; CHECK-NEXT: ret void
417+
;
418+
call void @llvm.memset.p0i8.i64(i8* %ptr, i8 0, i64 10, i1 false)
419+
%ptr.4 = getelementptr i8, i8* %ptr, i64 4
420+
%bc.4 = bitcast i8* %ptr.4 to i16*
421+
store i16 8, i16* %bc.4
422+
%ptr.5 = getelementptr i8, i8* %ptr, i64 5
423+
%bc.5 = bitcast i8* %ptr.5 to i16*
424+
store i16 0, i16* %bc.5
425+
ret void
426+
}
427+
428+
define void @test12_memset_later_store_exceeds_memset(i8* %ptr) {
429+
; CHECK-LABEL: @test12_memset_later_store_exceeds_memset(
430+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 1 [[PTR:%.*]], i8 0, i64 8, i1 false)
431+
; CHECK-NEXT: [[PTR_4:%.*]] = getelementptr i8, i8* [[PTR]], i64 4
432+
; CHECK-NEXT: store i8 8, i8* [[PTR_4]], align 1
433+
; CHECK-NEXT: [[PTR_5:%.*]] = getelementptr i8, i8* [[PTR]], i64 8
434+
; CHECK-NEXT: [[BC:%.*]] = bitcast i8* [[PTR_5]] to i64*
435+
; CHECK-NEXT: store i64 0, i64* [[BC]], align 8
436+
; CHECK-NEXT: ret void
437+
;
438+
call void @llvm.memset.p0i8.i64(i8* %ptr, i8 0, i64 10, i1 false)
439+
%ptr.4 = getelementptr i8, i8* %ptr, i64 4
440+
store i8 8, i8* %ptr.4
441+
%ptr.5 = getelementptr i8, i8* %ptr, i64 8
442+
%bc = bitcast i8* %ptr.5 to i64*
443+
store i64 0, i64* %bc
444+
ret void
445+
}
446+
447+
define void @test12_memset_later_store_before_memset(i8* %ptr) {
448+
; CHECK-LABEL: @test12_memset_later_store_before_memset(
449+
; CHECK-NEXT: [[PTR_1:%.*]] = getelementptr i8, i8* [[PTR:%.*]], i64 1
450+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, i8* [[PTR_1]], i64 7
451+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 1 [[TMP1]], i8 0, i64 3, i1 false)
452+
; CHECK-NEXT: [[BC:%.*]] = bitcast i8* [[PTR]] to i64*
453+
; CHECK-NEXT: store i64 0, i64* [[BC]], align 8
454+
; CHECK-NEXT: ret void
455+
;
456+
%ptr.1 = getelementptr i8, i8* %ptr, i64 1
457+
call void @llvm.memset.p0i8.i64(i8* %ptr.1, i8 0, i64 10, i1 false)
458+
%ptr.4 = getelementptr i8, i8* %ptr, i64 4
459+
store i8 8, i8* %ptr.4
460+
%bc = bitcast i8* %ptr to i64*
461+
store i64 0, i64* %bc
462+
ret void
463+
}
464+
465+
; The memset will be shortened and the store will not be redundant afterwards.
466+
; It cannot be eliminated.
467+
define void @test13_memset_shortened(i64* %ptr) {
468+
; CHECK-LABEL: @test13_memset_shortened(
469+
; CHECK-NEXT: [[PTR_I8:%.*]] = bitcast i64* [[PTR:%.*]] to i8*
470+
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, i8* [[PTR_I8]], i64 8
471+
; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 1 [[TMP1]], i8 0, i64 16, i1 false)
472+
; CHECK-NEXT: store i64 0, i64* [[PTR]], align 8
473+
; CHECK-NEXT: ret void
474+
;
475+
%ptr.i8 = bitcast i64* %ptr to i8*
476+
call void @llvm.memset.p0i8.i64(i8* %ptr.i8, i8 0, i64 24, i1 false)
477+
store i64 0, i64* %ptr
478+
ret void
479+
}
480+
290481
define void @pr49927(i32* %q, i32* %p) {
291482
; CHECK-LABEL: @pr49927(
292483
; CHECK-NEXT: [[V:%.*]] = load i32, i32* [[P:%.*]], align 4
@@ -301,7 +492,6 @@ define void @pr49927(i32* %q, i32* %p) {
301492
ret void
302493
}
303494

304-
305495
define void @pr50339(i8* nocapture readonly %0) {
306496
; CHECK-LABEL: @pr50339(
307497
; CHECK-NEXT: tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* noundef nonnull align 16 dereferenceable(16) getelementptr inbounds ([32 x i8], [32 x i8]* @a, i64 0, i64 0), i8* noundef nonnull align 1 dereferenceable(16) [[TMP0:%.*]], i64 16, i1 false)

0 commit comments

Comments
 (0)