1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -dse -S | FileCheck %s
4 @BUFFER = external local_unnamed_addr global [0 x i8], align 1
6 define void @ArrayTestFullyOverlapping(i64 %0) {
8 ; The DSE pass will try to kill the store of size i32 using the store of
9 ; size i64 because they fully overlap, in fact:
11 ; - they use the same base pointer (in SCEV style '@BUFFER + %0')
12 ; - the offset between the two stores is 32 bits
13 ; - the size of the earlier store is 32 bits
14 ; - the size of the later store is 64 bits
16 ; CHECK-LABEL: @ArrayTestFullyOverlapping(
17 ; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP0:%.*]], -8
18 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 [[TMP2]]
19 ; CHECK-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to i64*
20 ; CHECK-NEXT: store i64 0, i64* [[TMP4]], align 4
21 ; CHECK-NEXT: ret void
24 %3 = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 %2
25 %4 = bitcast i8* %3 to i64*
27 %6 = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 %5
28 %7 = bitcast i8* %6 to i32*
34 define void @VectorTestFullyOverlapping(float* %arg, i32 %i) {
35 ; CHECK-LABEL: @VectorTestFullyOverlapping(
37 ; CHECK-NEXT: [[I2:%.*]] = zext i32 [[I:%.*]] to i64
38 ; CHECK-NEXT: [[I3:%.*]] = getelementptr inbounds float, float* [[ARG:%.*]], i64 [[I2]]
39 ; CHECK-NEXT: [[I4:%.*]] = bitcast float* [[I3]] to <2 x float>*
40 ; CHECK-NEXT: store <2 x float> zeroinitializer, <2 x float>* [[I4]], align 16
41 ; CHECK-NEXT: ret void
44 %i7 = add nuw nsw i32 %i, 1
45 %i8 = zext i32 %i7 to i64
46 %i9 = getelementptr inbounds float, float* %arg, i64 %i8
47 store float 0.0, float* %i9, align 4
48 %i2 = zext i32 %i to i64
49 %i3 = getelementptr inbounds float, float* %arg, i64 %i2
50 %i4 = bitcast float* %i3 to <2 x float>*
51 store <2 x float> <float 0.0, float 0.0>, <2 x float>* %i4, align 16
55 define void @ArrayTestPartiallyOverlapping(i64 %0) {
57 ; The DSE pass will not kill the store because the overlap is partial
58 ; and won't fully clobber the i32 store.
60 ; CHECK-LABEL: @ArrayTestPartiallyOverlapping(
61 ; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP0:%.*]], 10
62 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 [[TMP2]]
63 ; CHECK-NEXT: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to i64*
64 ; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP0]], 15
65 ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 [[TMP5]]
66 ; CHECK-NEXT: [[TMP7:%.*]] = bitcast i8* [[TMP6]] to i32*
67 ; CHECK-NEXT: store i32 1, i32* [[TMP7]], align 4
68 ; CHECK-NEXT: store i64 0, i64* [[TMP4]], align 4
69 ; CHECK-NEXT: ret void
72 %3 = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 %2
73 %4 = bitcast i8* %3 to i64*
75 %6 = getelementptr inbounds [0 x i8], [0 x i8]* @BUFFER, i64 0, i64 %5
76 %7 = bitcast i8* %6 to i32*
82 define void @VectorTestPartiallyOverlapping(float* %arg, i32 %i) {
84 ; The DSE pass will not kill the store because the overlap is partial
85 ; and won't fully clobber the original store.
87 ; CHECK-LABEL: @VectorTestPartiallyOverlapping(
89 ; CHECK-NEXT: [[I2:%.*]] = zext i32 [[I:%.*]] to i64
90 ; CHECK-NEXT: [[I3:%.*]] = getelementptr inbounds float, float* [[ARG:%.*]], i64 [[I2]]
91 ; CHECK-NEXT: [[I4:%.*]] = bitcast float* [[I3]] to <2 x float>*
92 ; CHECK-NEXT: store <2 x float> <float 1.000000e+00, float 1.000000e+00>, <2 x float>* [[I4]], align 16
93 ; CHECK-NEXT: [[I5:%.*]] = add nuw nsw i32 [[I]], 1
94 ; CHECK-NEXT: [[I6:%.*]] = zext i32 [[I5]] to i64
95 ; CHECK-NEXT: [[I7:%.*]] = getelementptr inbounds float, float* [[ARG]], i64 [[I6]]
96 ; CHECK-NEXT: [[I8:%.*]] = bitcast float* [[I7]] to <2 x float>*
97 ; CHECK-NEXT: store <2 x float> zeroinitializer, <2 x float>* [[I8]], align 16
98 ; CHECK-NEXT: ret void
101 %i2 = zext i32 %i to i64
102 %i3 = getelementptr inbounds float, float* %arg, i64 %i2
103 %i4 = bitcast float* %i3 to <2 x float>*
104 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, <2 x float>* %i4, align 16
105 %i5 = add nuw nsw i32 %i, 1
106 %i6 = zext i32 %i5 to i64
107 %i7 = getelementptr inbounds float, float* %arg, i64 %i6
108 %i8 = bitcast float* %i7 to <2 x float>*
109 store <2 x float> <float 0.0, float 0.0>, <2 x float>* %i8, align 16