xref: /llvm-project/llvm/test/CodeGen/X86/shift-logic.ll (revision c3bf6d20ac306b829dc99939b3a8f9487f7f1c9a)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s
3
4define i8 @shl_and(i8 %x, i8 %y) nounwind {
5; CHECK-LABEL: shl_and:
6; CHECK:       # %bb.0:
7; CHECK-NEXT:    # kill: def $esi killed $esi def $rsi
8; CHECK-NEXT:    leal (,%rsi,4), %eax
9; CHECK-NEXT:    shlb $5, %dil
10; CHECK-NEXT:    andb %dil, %al
11; CHECK-NEXT:    # kill: def $al killed $al killed $eax
12; CHECK-NEXT:    retq
13  %sh0 = shl i8 %x, 3
14  %r = and i8 %sh0, %y
15  %sh1 = shl i8 %r, 2
16  ret i8 %sh1
17}
18
19define i16 @shl_or(i16 %x, i16 %y) nounwind {
20; CHECK-LABEL: shl_or:
21; CHECK:       # %bb.0:
22; CHECK-NEXT:    movl %edi, %eax
23; CHECK-NEXT:    shll $7, %esi
24; CHECK-NEXT:    shll $12, %eax
25; CHECK-NEXT:    orl %esi, %eax
26; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
27; CHECK-NEXT:    retq
28  %sh0 = shl i16 %x, 5
29  %r = or i16 %y, %sh0
30  %sh1 = shl i16 %r, 7
31  ret i16 %sh1
32}
33
34define i32 @shl_xor(i32 %x, i32 %y) nounwind {
35; CHECK-LABEL: shl_xor:
36; CHECK:       # %bb.0:
37; CHECK-NEXT:    movl %edi, %eax
38; CHECK-NEXT:    shll $7, %esi
39; CHECK-NEXT:    shll $12, %eax
40; CHECK-NEXT:    xorl %esi, %eax
41; CHECK-NEXT:    retq
42  %sh0 = shl i32 %x, 5
43  %r = xor i32 %sh0, %y
44  %sh1 = shl i32 %r, 7
45  ret i32 %sh1
46}
47
48define i64 @lshr_and(i64 %x, i64 %y) nounwind {
49; CHECK-LABEL: lshr_and:
50; CHECK:       # %bb.0:
51; CHECK-NEXT:    movq %rdi, %rax
52; CHECK-NEXT:    shrq $7, %rsi
53; CHECK-NEXT:    shrq $12, %rax
54; CHECK-NEXT:    andq %rsi, %rax
55; CHECK-NEXT:    retq
56  %sh0 = lshr i64 %x, 5
57  %r = and i64 %y, %sh0
58  %sh1 = lshr i64 %r, 7
59  ret i64 %sh1
60}
61
62define <4 x i32> @lshr_or(<4 x i32> %x, <4 x i32> %y) nounwind {
63; CHECK-LABEL: lshr_or:
64; CHECK:       # %bb.0:
65; CHECK-NEXT:    psrld $7, %xmm1
66; CHECK-NEXT:    psrld $12, %xmm0
67; CHECK-NEXT:    por %xmm1, %xmm0
68; CHECK-NEXT:    retq
69  %sh0 = lshr <4 x i32> %x, <i32 5, i32 5, i32 5, i32 5>
70  %r = or <4 x i32> %sh0, %y
71  %sh1 = lshr <4 x i32> %r, <i32 7, i32 7, i32 7, i32 7>
72  ret <4 x i32> %sh1
73}
74
75define <8 x i16> @lshr_xor(<8 x i16> %x, <8 x i16> %y) nounwind {
76; CHECK-LABEL: lshr_xor:
77; CHECK:       # %bb.0:
78; CHECK-NEXT:    psrlw $7, %xmm1
79; CHECK-NEXT:    psrlw $12, %xmm0
80; CHECK-NEXT:    pxor %xmm1, %xmm0
81; CHECK-NEXT:    retq
82  %sh0 = lshr <8 x i16> %x, <i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5>
83  %r = xor <8 x i16> %y, %sh0
84  %sh1 = lshr <8 x i16> %r, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
85  ret <8 x i16> %sh1
86}
87
88
89define <16 x i8> @ashr_and(<16 x i8> %x, <16 x i8> %y) nounwind {
90; CHECK-LABEL: ashr_and:
91; CHECK:       # %bb.0:
92; CHECK-NEXT:    psrlw $2, %xmm1
93; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
94; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
95; CHECK-NEXT:    pxor %xmm2, %xmm1
96; CHECK-NEXT:    psubb %xmm2, %xmm1
97; CHECK-NEXT:    psrlw $5, %xmm0
98; CHECK-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
99; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
100; CHECK-NEXT:    pxor %xmm2, %xmm0
101; CHECK-NEXT:    psubb %xmm2, %xmm0
102; CHECK-NEXT:    pand %xmm1, %xmm0
103; CHECK-NEXT:    retq
104  %sh0 = ashr <16 x i8> %x, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
105  %r = and <16 x i8> %y, %sh0
106  %sh1 = ashr <16 x i8> %r, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
107  ret <16 x i8> %sh1
108}
109
110define <2 x i64> @ashr_or(<2 x i64> %x, <2 x i64> %y) nounwind {
111; CHECK-LABEL: ashr_or:
112; CHECK:       # %bb.0:
113; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
114; CHECK-NEXT:    psrad $7, %xmm2
115; CHECK-NEXT:    psrlq $7, %xmm1
116; CHECK-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
117; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
118; CHECK-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,3,2,3]
119; CHECK-NEXT:    psrad $12, %xmm2
120; CHECK-NEXT:    psrlq $12, %xmm0
121; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
122; CHECK-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
123; CHECK-NEXT:    por %xmm1, %xmm0
124; CHECK-NEXT:    retq
125  %sh0 = ashr <2 x i64> %x, <i64 5, i64 5>
126  %r = or <2 x i64> %sh0, %y
127  %sh1 = ashr <2 x i64> %r, <i64 7, i64 7>
128  ret <2 x i64> %sh1
129}
130
131define i32 @ashr_xor(i32 %x, i32 %y) nounwind {
132; CHECK-LABEL: ashr_xor:
133; CHECK:       # %bb.0:
134; CHECK-NEXT:    movl %edi, %eax
135; CHECK-NEXT:    sarl $7, %esi
136; CHECK-NEXT:    sarl $12, %eax
137; CHECK-NEXT:    xorl %esi, %eax
138; CHECK-NEXT:    retq
139  %sh0 = ashr i32 %x, 5
140  %r = xor i32 %y, %sh0
141  %sh1 = ashr i32 %r, 7
142  ret i32 %sh1
143}
144
145define i32 @shr_mismatch_xor(i32 %x, i32 %y) nounwind {
146; CHECK-LABEL: shr_mismatch_xor:
147; CHECK:       # %bb.0:
148; CHECK-NEXT:    movl %edi, %eax
149; CHECK-NEXT:    sarl $5, %eax
150; CHECK-NEXT:    xorl %esi, %eax
151; CHECK-NEXT:    shrl $7, %eax
152; CHECK-NEXT:    retq
153  %sh0 = ashr i32 %x, 5
154  %r = xor i32 %y, %sh0
155  %sh1 = lshr i32 %r, 7
156  ret i32 %sh1
157}
158
159define i32 @ashr_overshift_xor(i32 %x, i32 %y) nounwind {
160; CHECK-LABEL: ashr_overshift_xor:
161; CHECK:       # %bb.0:
162; CHECK-NEXT:    movl %edi, %eax
163; CHECK-NEXT:    sarl $15, %eax
164; CHECK-NEXT:    xorl %esi, %eax
165; CHECK-NEXT:    sarl $17, %eax
166; CHECK-NEXT:    retq
167  %sh0 = ashr i32 %x, 15
168  %r = xor i32 %y, %sh0
169  %sh1 = ashr i32 %r, 17
170  ret i32 %sh1
171}
172
173define i32 @lshr_or_extra_use(i32 %x, i32 %y, ptr %p) nounwind {
174; CHECK-LABEL: lshr_or_extra_use:
175; CHECK:       # %bb.0:
176; CHECK-NEXT:    movl %edi, %eax
177; CHECK-NEXT:    shrl $5, %eax
178; CHECK-NEXT:    orl %esi, %eax
179; CHECK-NEXT:    movl %eax, (%rdx)
180; CHECK-NEXT:    shrl $7, %eax
181; CHECK-NEXT:    retq
182  %sh0 = lshr i32 %x, 5
183  %r = or i32 %sh0, %y
184  store i32 %r, ptr %p
185  %sh1 = lshr i32 %r, 7
186  ret i32 %sh1
187}
188