OLD | NEW |
1 ; This file checks support for comparing vector values with the fcmp | 1 ; This file checks support for comparing vector values with the fcmp |
2 ; instruction. | 2 ; instruction. |
3 | 3 |
4 ; RUN: %p2i -i %s -a -O2 --verbose none \ | 4 ; RUN: %p2i -i %s --assemble --disassemble -a -O2 --verbose none \ |
5 ; RUN: | llvm-mc -triple=i686-none-nacl -filetype=obj \ | 5 ; RUN: | FileCheck %s |
6 ; RUN: | llvm-objdump -d --symbolize -x86-asm-syntax=intel - | FileCheck %s | 6 ; RUN: %p2i -i %s --assemble --disassemble -a -Om1 --verbose none \ |
7 ; RUN: %p2i -i %s -a -Om1 --verbose none \ | 7 ; RUN: | FileCheck %s |
8 ; RUN: | llvm-mc -triple=i686-none-nacl -filetype=obj \ | |
9 ; RUN: | llvm-objdump -d --symbolize -x86-asm-syntax=intel - | FileCheck %s | |
10 | 8 |
11 ; Check that sext elimination occurs when the result of the comparison | 9 ; Check that sext elimination occurs when the result of the comparison |
12 ; instruction is alrady sign extended. Sign extension to 4 x i32 uses | 10 ; instruction is alrady sign extended. Sign extension to 4 x i32 uses |
13 ; the pslld instruction. | 11 ; the pslld instruction. |
14 define <4 x i32> @sextElimination(<4 x float> %a, <4 x float> %b) { | 12 define <4 x i32> @sextElimination(<4 x float> %a, <4 x float> %b) { |
15 entry: | 13 entry: |
16 %res.trunc = fcmp oeq <4 x float> %a, %b | 14 %res.trunc = fcmp oeq <4 x float> %a, %b |
17 %res = sext <4 x i1> %res.trunc to <4 x i32> | 15 %res = sext <4 x i1> %res.trunc to <4 x i32> |
18 ret <4 x i32> %res | 16 ret <4 x i32> %res |
19 ; CHECK-LABEL: sextElimination: | 17 ; CHECK-LABEL: sextElimination |
20 ; CHECK: cmpeqps | 18 ; CHECK: cmpeqps |
21 ; CHECK-NOT: pslld | 19 ; CHECK-NOT: pslld |
22 } | 20 } |
23 | 21 |
24 define <4 x i32> @fcmpFalseVector(<4 x float> %a, <4 x float> %b) { | 22 define <4 x i32> @fcmpFalseVector(<4 x float> %a, <4 x float> %b) { |
25 entry: | 23 entry: |
26 %res.trunc = fcmp false <4 x float> %a, %b | 24 %res.trunc = fcmp false <4 x float> %a, %b |
27 %res = sext <4 x i1> %res.trunc to <4 x i32> | 25 %res = sext <4 x i1> %res.trunc to <4 x i32> |
28 ret <4 x i32> %res | 26 ret <4 x i32> %res |
29 ; CHECK-LABEL: fcmpFalseVector: | 27 ; CHECK-LABEL: fcmpFalseVector |
30 ; CHECK: pxor | 28 ; CHECK: pxor |
31 } | 29 } |
32 | 30 |
33 define <4 x i32> @fcmpOeqVector(<4 x float> %a, <4 x float> %b) { | 31 define <4 x i32> @fcmpOeqVector(<4 x float> %a, <4 x float> %b) { |
34 entry: | 32 entry: |
35 %res.trunc = fcmp oeq <4 x float> %a, %b | 33 %res.trunc = fcmp oeq <4 x float> %a, %b |
36 %res = sext <4 x i1> %res.trunc to <4 x i32> | 34 %res = sext <4 x i1> %res.trunc to <4 x i32> |
37 ret <4 x i32> %res | 35 ret <4 x i32> %res |
38 ; CHECK-LABEL: fcmpOeqVector: | 36 ; CHECK-LABEL: fcmpOeqVector |
39 ; CHECK: cmpeqps | 37 ; CHECK: cmpeqps |
40 } | 38 } |
41 | 39 |
42 define <4 x i32> @fcmpOgeVector(<4 x float> %a, <4 x float> %b) { | 40 define <4 x i32> @fcmpOgeVector(<4 x float> %a, <4 x float> %b) { |
43 entry: | 41 entry: |
44 %res.trunc = fcmp oge <4 x float> %a, %b | 42 %res.trunc = fcmp oge <4 x float> %a, %b |
45 %res = sext <4 x i1> %res.trunc to <4 x i32> | 43 %res = sext <4 x i1> %res.trunc to <4 x i32> |
46 ret <4 x i32> %res | 44 ret <4 x i32> %res |
47 ; CHECK-LABEL: fcmpOgeVector: | 45 ; CHECK-LABEL: fcmpOgeVector |
48 ; CHECK: cmpleps | 46 ; CHECK: cmpleps |
49 } | 47 } |
50 | 48 |
51 define <4 x i32> @fcmpOgtVector(<4 x float> %a, <4 x float> %b) { | 49 define <4 x i32> @fcmpOgtVector(<4 x float> %a, <4 x float> %b) { |
52 entry: | 50 entry: |
53 %res.trunc = fcmp ogt <4 x float> %a, %b | 51 %res.trunc = fcmp ogt <4 x float> %a, %b |
54 %res = sext <4 x i1> %res.trunc to <4 x i32> | 52 %res = sext <4 x i1> %res.trunc to <4 x i32> |
55 ret <4 x i32> %res | 53 ret <4 x i32> %res |
56 ; CHECK-LABEL: fcmpOgtVector: | 54 ; CHECK-LABEL: fcmpOgtVector |
57 ; CHECK: cmpltps | 55 ; CHECK: cmpltps |
58 } | 56 } |
59 | 57 |
60 define <4 x i32> @fcmpOleVector(<4 x float> %a, <4 x float> %b) { | 58 define <4 x i32> @fcmpOleVector(<4 x float> %a, <4 x float> %b) { |
61 entry: | 59 entry: |
62 %res.trunc = fcmp ole <4 x float> %a, %b | 60 %res.trunc = fcmp ole <4 x float> %a, %b |
63 %res = sext <4 x i1> %res.trunc to <4 x i32> | 61 %res = sext <4 x i1> %res.trunc to <4 x i32> |
64 ret <4 x i32> %res | 62 ret <4 x i32> %res |
65 ; CHECK-LABEL: fcmpOleVector: | 63 ; CHECK-LABEL: fcmpOleVector |
66 ; CHECK: cmpleps | 64 ; CHECK: cmpleps |
67 } | 65 } |
68 | 66 |
69 define <4 x i32> @fcmpOltVector(<4 x float> %a, <4 x float> %b) { | 67 define <4 x i32> @fcmpOltVector(<4 x float> %a, <4 x float> %b) { |
70 entry: | 68 entry: |
71 %res.trunc = fcmp olt <4 x float> %a, %b | 69 %res.trunc = fcmp olt <4 x float> %a, %b |
72 %res = sext <4 x i1> %res.trunc to <4 x i32> | 70 %res = sext <4 x i1> %res.trunc to <4 x i32> |
73 ret <4 x i32> %res | 71 ret <4 x i32> %res |
74 ; CHECK-LABEL: fcmpOltVector: | 72 ; CHECK-LABEL: fcmpOltVector |
75 ; CHECK: cmpltps | 73 ; CHECK: cmpltps |
76 } | 74 } |
77 | 75 |
78 define <4 x i32> @fcmpOneVector(<4 x float> %a, <4 x float> %b) { | 76 define <4 x i32> @fcmpOneVector(<4 x float> %a, <4 x float> %b) { |
79 entry: | 77 entry: |
80 %res.trunc = fcmp one <4 x float> %a, %b | 78 %res.trunc = fcmp one <4 x float> %a, %b |
81 %res = sext <4 x i1> %res.trunc to <4 x i32> | 79 %res = sext <4 x i1> %res.trunc to <4 x i32> |
82 ret <4 x i32> %res | 80 ret <4 x i32> %res |
83 ; CHECK-LABEL: fcmpOneVector: | 81 ; CHECK-LABEL: fcmpOneVector |
84 ; CHECK: cmpneqps | 82 ; CHECK: cmpneqps |
85 ; CHECK: cmpordps | 83 ; CHECK: cmpordps |
86 ; CHECK: pand | 84 ; CHECK: pand |
87 } | 85 } |
88 | 86 |
89 define <4 x i32> @fcmpOrdVector(<4 x float> %a, <4 x float> %b) { | 87 define <4 x i32> @fcmpOrdVector(<4 x float> %a, <4 x float> %b) { |
90 entry: | 88 entry: |
91 %res.trunc = fcmp ord <4 x float> %a, %b | 89 %res.trunc = fcmp ord <4 x float> %a, %b |
92 %res = sext <4 x i1> %res.trunc to <4 x i32> | 90 %res = sext <4 x i1> %res.trunc to <4 x i32> |
93 ret <4 x i32> %res | 91 ret <4 x i32> %res |
94 ; CHECK-LABEL: fcmpOrdVector: | 92 ; CHECK-LABEL: fcmpOrdVector |
95 ; CHECK: cmpordps | 93 ; CHECK: cmpordps |
96 } | 94 } |
97 | 95 |
98 define <4 x i32> @fcmpTrueVector(<4 x float> %a, <4 x float> %b) { | 96 define <4 x i32> @fcmpTrueVector(<4 x float> %a, <4 x float> %b) { |
99 entry: | 97 entry: |
100 %res.trunc = fcmp true <4 x float> %a, %b | 98 %res.trunc = fcmp true <4 x float> %a, %b |
101 %res = sext <4 x i1> %res.trunc to <4 x i32> | 99 %res = sext <4 x i1> %res.trunc to <4 x i32> |
102 ret <4 x i32> %res | 100 ret <4 x i32> %res |
103 ; CHECK-LABEL: fcmpTrueVector: | 101 ; CHECK-LABEL: fcmpTrueVector |
104 ; CHECK: pcmpeqd | 102 ; CHECK: pcmpeqd |
105 } | 103 } |
106 | 104 |
107 define <4 x i32> @fcmpUeqVector(<4 x float> %a, <4 x float> %b) { | 105 define <4 x i32> @fcmpUeqVector(<4 x float> %a, <4 x float> %b) { |
108 entry: | 106 entry: |
109 %res.trunc = fcmp ueq <4 x float> %a, %b | 107 %res.trunc = fcmp ueq <4 x float> %a, %b |
110 %res = sext <4 x i1> %res.trunc to <4 x i32> | 108 %res = sext <4 x i1> %res.trunc to <4 x i32> |
111 ret <4 x i32> %res | 109 ret <4 x i32> %res |
112 ; CHECK-LABEL: fcmpUeqVector: | 110 ; CHECK-LABEL: fcmpUeqVector |
113 ; CHECK: cmpeqps | 111 ; CHECK: cmpeqps |
114 ; CHECK: cmpunordps | 112 ; CHECK: cmpunordps |
115 ; CHECK: por | 113 ; CHECK: por |
116 } | 114 } |
117 | 115 |
118 define <4 x i32> @fcmpUgeVector(<4 x float> %a, <4 x float> %b) { | 116 define <4 x i32> @fcmpUgeVector(<4 x float> %a, <4 x float> %b) { |
119 entry: | 117 entry: |
120 %res.trunc = fcmp uge <4 x float> %a, %b | 118 %res.trunc = fcmp uge <4 x float> %a, %b |
121 %res = sext <4 x i1> %res.trunc to <4 x i32> | 119 %res = sext <4 x i1> %res.trunc to <4 x i32> |
122 ret <4 x i32> %res | 120 ret <4 x i32> %res |
123 ; CHECK-LABEL: fcmpUgeVector: | 121 ; CHECK-LABEL: fcmpUgeVector |
124 ; CHECK: cmpnltps | 122 ; CHECK: cmpnltps |
125 } | 123 } |
126 | 124 |
127 define <4 x i32> @fcmpUgtVector(<4 x float> %a, <4 x float> %b) { | 125 define <4 x i32> @fcmpUgtVector(<4 x float> %a, <4 x float> %b) { |
128 entry: | 126 entry: |
129 %res.trunc = fcmp ugt <4 x float> %a, %b | 127 %res.trunc = fcmp ugt <4 x float> %a, %b |
130 %res = sext <4 x i1> %res.trunc to <4 x i32> | 128 %res = sext <4 x i1> %res.trunc to <4 x i32> |
131 ret <4 x i32> %res | 129 ret <4 x i32> %res |
132 ; CHECK-LABEL: fcmpUgtVector: | 130 ; CHECK-LABEL: fcmpUgtVector |
133 ; CHECK: cmpnleps | 131 ; CHECK: cmpnleps |
134 } | 132 } |
135 | 133 |
136 define <4 x i32> @fcmpUleVector(<4 x float> %a, <4 x float> %b) { | 134 define <4 x i32> @fcmpUleVector(<4 x float> %a, <4 x float> %b) { |
137 entry: | 135 entry: |
138 %res.trunc = fcmp ule <4 x float> %a, %b | 136 %res.trunc = fcmp ule <4 x float> %a, %b |
139 %res = sext <4 x i1> %res.trunc to <4 x i32> | 137 %res = sext <4 x i1> %res.trunc to <4 x i32> |
140 ret <4 x i32> %res | 138 ret <4 x i32> %res |
141 ; CHECK-LABEL: fcmpUleVector: | 139 ; CHECK-LABEL: fcmpUleVector |
142 ; CHECK: cmpnltps | 140 ; CHECK: cmpnltps |
143 } | 141 } |
144 | 142 |
145 define <4 x i32> @fcmpUltVector(<4 x float> %a, <4 x float> %b) { | 143 define <4 x i32> @fcmpUltVector(<4 x float> %a, <4 x float> %b) { |
146 entry: | 144 entry: |
147 %res.trunc = fcmp ult <4 x float> %a, %b | 145 %res.trunc = fcmp ult <4 x float> %a, %b |
148 %res = sext <4 x i1> %res.trunc to <4 x i32> | 146 %res = sext <4 x i1> %res.trunc to <4 x i32> |
149 ret <4 x i32> %res | 147 ret <4 x i32> %res |
150 ; CHECK-LABEL: fcmpUltVector: | 148 ; CHECK-LABEL: fcmpUltVector |
151 ; CHECK: cmpnleps | 149 ; CHECK: cmpnleps |
152 } | 150 } |
153 | 151 |
154 define <4 x i32> @fcmpUneVector(<4 x float> %a, <4 x float> %b) { | 152 define <4 x i32> @fcmpUneVector(<4 x float> %a, <4 x float> %b) { |
155 entry: | 153 entry: |
156 %res.trunc = fcmp une <4 x float> %a, %b | 154 %res.trunc = fcmp une <4 x float> %a, %b |
157 %res = sext <4 x i1> %res.trunc to <4 x i32> | 155 %res = sext <4 x i1> %res.trunc to <4 x i32> |
158 ret <4 x i32> %res | 156 ret <4 x i32> %res |
159 ; CHECK-LABEL: fcmpUneVector: | 157 ; CHECK-LABEL: fcmpUneVector |
160 ; CHECK: cmpneqps | 158 ; CHECK: cmpneqps |
161 } | 159 } |
162 | 160 |
163 define <4 x i32> @fcmpUnoVector(<4 x float> %a, <4 x float> %b) { | 161 define <4 x i32> @fcmpUnoVector(<4 x float> %a, <4 x float> %b) { |
164 entry: | 162 entry: |
165 %res.trunc = fcmp uno <4 x float> %a, %b | 163 %res.trunc = fcmp uno <4 x float> %a, %b |
166 %res = sext <4 x i1> %res.trunc to <4 x i32> | 164 %res = sext <4 x i1> %res.trunc to <4 x i32> |
167 ret <4 x i32> %res | 165 ret <4 x i32> %res |
168 ; CHECK-LABEL: fcmpUnoVector: | 166 ; CHECK-LABEL: fcmpUnoVector |
169 ; CHECK: cmpunordps | 167 ; CHECK: cmpunordps |
170 } | 168 } |
OLD | NEW |