| blob | a4fe112e1a79d623da7e3486b73bcc2d517b2385 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-apple-macosx -mattr=+sse2 -verify-machineinstrs | FileCheck %s
4 ; After tail duplication, two copies in an early exit BB can be cancelled out.
5 ; rdar://10640363
6 define i32 @t1(i32 %a, i32 %b) nounwind {
7 ; CHECK-LABEL: t1:
8 ; CHECK: ## BB#0: ## %entry
9 ; CHECK-NEXT: movl %esi, %edx
10 ; CHECK-NEXT: movl %edi, %eax
11 ; CHECK-NEXT: testl %edx, %edx
12 ; CHECK-NEXT: je LBB0_1
13 ; CHECK-NEXT: .p2align 4, 0x90
14 ; CHECK-NEXT: LBB0_2: ## %while.body
15 ; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1
16 ; CHECK-NEXT: movl %edx, %ecx
17 ; CHECK-NEXT: cltd
18 ; CHECK-NEXT: idivl %ecx
19 ; CHECK-NEXT: testl %edx, %edx
20 ; CHECK-NEXT: movl %ecx, %eax
21 ; CHECK-NEXT: jne LBB0_2
22 ; CHECK-NEXT: ## BB#3: ## %while.end
23 ; CHECK-NEXT: movl %ecx, %eax
24 ; CHECK-NEXT: retq
25 ; CHECK-NEXT: LBB0_1:
26 ; CHECK-NEXT: retq
27 entry:
28 %cmp1 = icmp eq i32 %b, 0
29 br i1 %cmp1, label %while.end, label %while.body
31 while.body: ; preds = %entry, %while.body
32 %a.addr.03 = phi i32 [ %b.addr.02, %while.body ], [ %a, %entry ]
33 %b.addr.02 = phi i32 [ %rem, %while.body ], [ %b, %entry ]
34 %rem = srem i32 %a.addr.03, %b.addr.02
35 %cmp = icmp eq i32 %rem, 0
36 br i1 %cmp, label %while.end, label %while.body
38 while.end: ; preds = %while.body, %entry
39 %a.addr.0.lcssa = phi i32 [ %a, %entry ], [ %b.addr.02, %while.body ]
40 ret i32 %a.addr.0.lcssa
41 }
43 ; Two movdqa (from phi-elimination) in the entry BB cancels out.
44 ; rdar://10428165
45 define <8 x i16> @t2(<8 x i16> %T0, <8 x i16> %T1) nounwind readnone {
46 ; CHECK-LABEL: t2:
47 ; CHECK: ## BB#0: ## %entry
48 ; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
49 ; CHECK-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]
50 ; CHECK-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
51 ; CHECK-NEXT: retq
52 entry:
53 %tmp8 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 undef, i32 undef, i32 7, i32 2, i32 8, i32 undef, i32 undef , i32 undef >
54 ret <8 x i16> %tmp8
55 }
57 define i32 @t3(i64 %a, i64 %b) nounwind {
58 ; CHECK-LABEL: t3:
59 ; CHECK: ## BB#0: ## %entry
60 ; CHECK-NEXT: movq %rsi, %rdx
61 ; CHECK-NEXT: movq %rdi, %rax
62 ; CHECK-NEXT: testq %rdx, %rdx
63 ; CHECK-NEXT: je LBB2_1
64 ; CHECK-NEXT: .p2align 4, 0x90
65 ; CHECK-NEXT: LBB2_2: ## %while.body
66 ; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1
67 ; CHECK-NEXT: movq %rdx, %rcx
68 ; CHECK-NEXT: cqto
69 ; CHECK-NEXT: idivq %rcx
70 ; CHECK-NEXT: testq %rdx, %rdx
71 ; CHECK-NEXT: movq %rcx, %rax
72 ; CHECK-NEXT: jne LBB2_2
73 ; CHECK-NEXT: ## BB#3: ## %while.end
74 ; CHECK-NEXT: movl %ecx, %eax
75 ; CHECK-NEXT: retq
76 ; CHECK-NEXT: LBB2_1:
77 ; CHECK-NEXT: retq
78 entry:
79 %cmp1 = icmp eq i64 %b, 0
80 br i1 %cmp1, label %while.end, label %while.body
82 while.body: ; preds = %entry, %while.body
83 %a.addr.03 = phi i64 [ %b.addr.02, %while.body ], [ %a, %entry ]
84 %b.addr.02 = phi i64 [ %rem, %while.body ], [ %b, %entry ]
85 %rem = srem i64 %a.addr.03, %b.addr.02
86 %cmp = icmp eq i64 %rem, 0
87 br i1 %cmp, label %while.end, label %while.body
89 while.end: ; preds = %while.body, %entry
90 %a.addr.0.lcssa = phi i64 [ %a, %entry ], [ %b.addr.02, %while.body ]
91 %t = trunc i64 %a.addr.0.lcssa to i32
92 ret i32 %t
93 }
95 ; Check that copy propagation does not kill thing like:
96 ; dst = copy src <-- do not kill that.
97 ; ... = op1 dst<undef>
98 ; ... = op2 dst <-- this is used here.
99 define <16 x float> @foo(<16 x float> %x) {
100 ; CHECK-LABEL: foo:
101 ; CHECK: ## BB#0: ## %bb
102 ; CHECK-NEXT: movaps %xmm3, %xmm8
103 ; CHECK-NEXT: xorps %xmm3, %xmm3
104 ; CHECK-NEXT: pxor %xmm6, %xmm6
105 ; CHECK-NEXT: pcmpgtd %xmm0, %xmm6
106 ; CHECK-NEXT: movdqa {{.*#+}} xmm5 = [255,255,255,255]
107 ; CHECK-NEXT: pand %xmm6, %xmm5
108 ; CHECK-NEXT: packuswb %xmm5, %xmm5
109 ; CHECK-NEXT: packuswb %xmm5, %xmm5
110 ; CHECK-NEXT: cvttps2dq %xmm0, %xmm13
111 ; CHECK-NEXT: movdqa %xmm0, %xmm10
112 ; CHECK-NEXT: cmpltps %xmm3, %xmm10
113 ; CHECK-NEXT: movdqa %xmm6, %xmm9
114 ; CHECK-NEXT: pxor %xmm10, %xmm9
115 ; CHECK-NEXT: cvttps2dq %xmm1, %xmm14
116 ; CHECK-NEXT: movaps %xmm1, %xmm11
117 ; CHECK-NEXT: cmpltps %xmm3, %xmm11
118 ; CHECK-NEXT: movdqa %xmm6, %xmm7
119 ; CHECK-NEXT: pxor %xmm11, %xmm7
120 ; CHECK-NEXT: cvttps2dq %xmm2, %xmm1
121 ; CHECK-NEXT: cmpltps %xmm3, %xmm2
122 ; CHECK-NEXT: movdqa %xmm6, %xmm4
123 ; CHECK-NEXT: pxor %xmm2, %xmm4
124 ; CHECK-NEXT: cvttps2dq %xmm8, %xmm12
125 ; CHECK-NEXT: cmpltps %xmm3, %xmm8
126 ; CHECK-NEXT: pxor %xmm8, %xmm6
127 ; CHECK-NEXT: movdqa {{.*#+}} xmm0 = [1,1,1,1]
128 ; CHECK-NEXT: pand %xmm0, %xmm6
129 ; CHECK-NEXT: pand %xmm0, %xmm4
130 ; CHECK-NEXT: pand %xmm0, %xmm7
131 ; CHECK-NEXT: pand %xmm0, %xmm9
132 ; CHECK-NEXT: cvtdq2ps %xmm13, %xmm15
133 ; CHECK-NEXT: cvtdq2ps %xmm14, %xmm14
134 ; CHECK-NEXT: cvtdq2ps %xmm1, %xmm13
135 ; CHECK-NEXT: cvtdq2ps %xmm12, %xmm12
136 ; CHECK-NEXT: pxor %xmm0, %xmm0
137 ; CHECK-NEXT: cmpltps %xmm12, %xmm0
138 ; CHECK-NEXT: xorps %xmm1, %xmm1
139 ; CHECK-NEXT: cmpltps %xmm13, %xmm1
140 ; CHECK-NEXT: packssdw %xmm0, %xmm1
141 ; CHECK-NEXT: xorps %xmm0, %xmm0
142 ; CHECK-NEXT: cmpltps %xmm14, %xmm0
143 ; CHECK-NEXT: cmpltps %xmm15, %xmm3
144 ; CHECK-NEXT: packssdw %xmm0, %xmm3
145 ; CHECK-NEXT: packsswb %xmm1, %xmm3
146 ; CHECK-NEXT: pand %xmm5, %xmm3
147 ; CHECK-NEXT: movdqa %xmm3, %xmm1
148 ; CHECK-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
149 ; CHECK-NEXT: movdqa %xmm1, %xmm0
150 ; CHECK-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
151 ; CHECK-NEXT: pslld $31, %xmm0
152 ; CHECK-NEXT: psrad $31, %xmm0
153 ; CHECK-NEXT: punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
154 ; CHECK-NEXT: pslld $31, %xmm1
155 ; CHECK-NEXT: psrad $31, %xmm1
156 ; CHECK-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
157 ; CHECK-NEXT: movdqa %xmm3, %xmm5
158 ; CHECK-NEXT: punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
159 ; CHECK-NEXT: pslld $31, %xmm5
160 ; CHECK-NEXT: psrad $31, %xmm5
161 ; CHECK-NEXT: punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
162 ; CHECK-NEXT: pslld $31, %xmm3
163 ; CHECK-NEXT: psrad $31, %xmm3
164 ; CHECK-NEXT: pxor %xmm9, %xmm0
165 ; CHECK-NEXT: pxor %xmm15, %xmm0
166 ; CHECK-NEXT: pxor %xmm7, %xmm1
167 ; CHECK-NEXT: pxor %xmm14, %xmm1
168 ; CHECK-NEXT: pxor %xmm4, %xmm5
169 ; CHECK-NEXT: pxor %xmm13, %xmm5
170 ; CHECK-NEXT: pxor %xmm6, %xmm3
171 ; CHECK-NEXT: pxor %xmm12, %xmm3
172 ; CHECK-NEXT: pand %xmm8, %xmm3
173 ; CHECK-NEXT: pand %xmm2, %xmm5
174 ; CHECK-NEXT: pand %xmm11, %xmm1
175 ; CHECK-NEXT: pand %xmm10, %xmm0
176 ; CHECK-NEXT: pxor %xmm9, %xmm0
177 ; CHECK-NEXT: pxor %xmm7, %xmm1
178 ; CHECK-NEXT: pxor %xmm4, %xmm5
179 ; CHECK-NEXT: pxor %xmm6, %xmm3
180 ; CHECK-NEXT: movdqa %xmm5, %xmm2
181 ; CHECK-NEXT: retq
182 bb:
183 %v3 = icmp slt <16 x i32> undef, zeroinitializer
184 %v14 = zext <16 x i1> %v3 to <16 x i32>
185 %v16 = fcmp olt <16 x float> %x, zeroinitializer
186 %v17 = sext <16 x i1> %v16 to <16 x i32>
187 %v18 = zext <16 x i1> %v16 to <16 x i32>
188 %v19 = xor <16 x i32> %v14, %v18
189 %v20 = or <16 x i32> %v17, undef
190 %v21 = fptosi <16 x float> %x to <16 x i32>
191 %v22 = sitofp <16 x i32> %v21 to <16 x float>
192 %v69 = fcmp ogt <16 x float> %v22, zeroinitializer
193 %v75 = and <16 x i1> %v69, %v3
194 %v77 = bitcast <16 x float> %v22 to <16 x i32>
195 %v79 = sext <16 x i1> %v75 to <16 x i32>
196 %v80 = and <16 x i32> undef, %v79
197 %v81 = xor <16 x i32> %v77, %v80
198 %v82 = and <16 x i32> undef, %v81
199 %v83 = xor <16 x i32> %v19, %v82
200 %v84 = and <16 x i32> %v83, %v20
201 %v85 = xor <16 x i32> %v19, %v84
202 %v86 = bitcast <16 x i32> %v85 to <16 x float>
203 ret <16 x float> %v86
204 }