1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=instcombine -S | FileCheck %s
4 ; If we have some pattern that leaves only some low bits set, and then performs
5 ; left-shift of those bits, we can combine those two shifts into a shift+mask.
7 ; There are many variants to this pattern:
8 ; e) ((x << maskNbits) l>> maskNbits) << shiftNbits
10 ; (x << shiftNbits) & (-1 >> ((-(maskNbits+shiftNbits))+32))
14 declare void @use32(i32)
16 define i32 @t0_basic(i32 %x, i32 %nbits) {
17 ; CHECK-LABEL: @t0_basic(
18 ; CHECK-NEXT: [[T0:%.*]] = shl i32 [[X:%.*]], [[NBITS:%.*]]
19 ; CHECK-NEXT: [[T2:%.*]] = add i32 [[NBITS]], -1
20 ; CHECK-NEXT: call void @use32(i32 [[T0]])
21 ; CHECK-NEXT: call void @use32(i32 [[T2]])
22 ; CHECK-NEXT: [[TMP1:%.*]] = shl i32 [[X]], [[T2]]
23 ; CHECK-NEXT: [[T3:%.*]] = and i32 [[TMP1]], 2147483647
24 ; CHECK-NEXT: ret i32 [[T3]]
26 %t0 = shl i32 %x, %nbits
27 %t1 = lshr i32 %t0, %nbits
28 %t2 = add i32 %nbits, -1
29 call void @use32(i32 %t0)
30 call void @use32(i32 %t2)
31 %t3 = shl i32 %t1, %t2 ; shift is smaller than mask
37 declare void @use8xi32(<8 x i32>)
39 define <8 x i32> @t1_vec_splat(<8 x i32> %x, <8 x i32> %nbits) {
40 ; CHECK-LABEL: @t1_vec_splat(
41 ; CHECK-NEXT: [[T0:%.*]] = shl <8 x i32> [[X:%.*]], [[NBITS:%.*]]
42 ; CHECK-NEXT: [[T2:%.*]] = add <8 x i32> [[NBITS]], <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
43 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T0]])
44 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T2]])
45 ; CHECK-NEXT: [[TMP1:%.*]] = shl <8 x i32> [[X]], [[T2]]
46 ; CHECK-NEXT: [[T3:%.*]] = and <8 x i32> [[TMP1]], <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
47 ; CHECK-NEXT: ret <8 x i32> [[T3]]
49 %t0 = shl <8 x i32> %x, %nbits
50 %t1 = lshr <8 x i32> %t0, %nbits
51 %t2 = add <8 x i32> %nbits, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
52 call void @use8xi32(<8 x i32> %t0)
53 call void @use8xi32(<8 x i32> %t2)
54 %t3 = shl <8 x i32> %t1, %t2 ; shift is smaller than mask
58 define <8 x i32> @t1_vec_splat_undef(<8 x i32> %x, <8 x i32> %nbits) {
59 ; CHECK-LABEL: @t1_vec_splat_undef(
60 ; CHECK-NEXT: [[T0:%.*]] = shl <8 x i32> [[X:%.*]], [[NBITS:%.*]]
61 ; CHECK-NEXT: [[T2:%.*]] = add <8 x i32> [[NBITS]], <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 undef, i32 -1>
62 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T0]])
63 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T2]])
64 ; CHECK-NEXT: [[TMP1:%.*]] = shl <8 x i32> [[X]], [[T2]]
65 ; CHECK-NEXT: [[T3:%.*]] = and <8 x i32> [[TMP1]], <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 poison, i32 2147483647>
66 ; CHECK-NEXT: ret <8 x i32> [[T3]]
68 %t0 = shl <8 x i32> %x, %nbits
69 %t1 = lshr <8 x i32> %t0, %nbits
70 %t2 = add <8 x i32> %nbits, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 undef, i32 -1>
71 call void @use8xi32(<8 x i32> %t0)
72 call void @use8xi32(<8 x i32> %t2)
73 %t3 = shl <8 x i32> %t1, %t2 ; shift is smaller than mask
77 define <8 x i32> @t1_vec_nonsplat(<8 x i32> %x, <8 x i32> %nbits) {
78 ; CHECK-LABEL: @t1_vec_nonsplat(
79 ; CHECK-NEXT: [[T0:%.*]] = shl <8 x i32> [[X:%.*]], [[NBITS:%.*]]
80 ; CHECK-NEXT: [[T2:%.*]] = add <8 x i32> [[NBITS]], <i32 -32, i32 -31, i32 -1, i32 0, i32 1, i32 31, i32 32, i32 33>
81 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T0]])
82 ; CHECK-NEXT: call void @use8xi32(<8 x i32> [[T2]])
83 ; CHECK-NEXT: [[TMP1:%.*]] = shl <8 x i32> [[X]], [[T2]]
84 ; CHECK-NEXT: [[T3:%.*]] = and <8 x i32> [[TMP1]], <i32 poison, i32 1, i32 2147483647, i32 -1, i32 -1, i32 -1, i32 -1, i32 poison>
85 ; CHECK-NEXT: ret <8 x i32> [[T3]]
87 %t0 = shl <8 x i32> %x, %nbits
88 %t1 = lshr <8 x i32> %t0, %nbits
89 %t2 = add <8 x i32> %nbits, <i32 -32, i32 -31, i32 -1, i32 0, i32 1, i32 31, i32 32, i32 33>
90 call void @use8xi32(<8 x i32> %t0)
91 call void @use8xi32(<8 x i32> %t2)
92 %t3 = shl <8 x i32> %t1, %t2 ; shift is smaller than mask
98 define i32 @n3_extrause(i32 %x, i32 %nbits) {
99 ; CHECK-LABEL: @n3_extrause(
100 ; CHECK-NEXT: [[T0:%.*]] = shl i32 [[X:%.*]], [[NBITS:%.*]]
101 ; CHECK-NEXT: [[T1:%.*]] = lshr exact i32 [[T0]], [[NBITS]]
102 ; CHECK-NEXT: [[T2:%.*]] = add i32 [[NBITS]], -1
103 ; CHECK-NEXT: call void @use32(i32 [[T0]])
104 ; CHECK-NEXT: call void @use32(i32 [[T1]])
105 ; CHECK-NEXT: call void @use32(i32 [[T2]])
106 ; CHECK-NEXT: [[T3:%.*]] = shl i32 [[T1]], [[T2]]
107 ; CHECK-NEXT: ret i32 [[T3]]
109 %t0 = shl i32 %x, %nbits
110 %t1 = lshr i32 %t0, %nbits
111 %t2 = add i32 %nbits, -1
112 call void @use32(i32 %t0)
113 call void @use32(i32 %t1) ; BAD
114 call void @use32(i32 %t2)
115 %t3 = shl i32 %t1, %t2 ; shift is smaller than mask