95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
1 ; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
121
|
2 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
120
|
3 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
5 declare i32 @llvm.r600.read.tidig.x() #0
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
6
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
7 ; FUNC-LABEL: {{^}}test2:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
8 ; EG: AND_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
9 ; EG: AND_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
10
|
121
|
11 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
|
12 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
13
|
121
|
14 define amdgpu_kernel void @test2(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
15 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
16 %a = load <2 x i32>, <2 x i32> addrspace(1) * %in
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
17 %b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
18 %result = and <2 x i32> %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
19 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
20 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
21 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
22
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
23 ; FUNC-LABEL: {{^}}test4:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
24 ; EG: AND_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
25 ; EG: AND_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
26 ; EG: AND_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
27 ; EG: AND_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
28
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
29
|
121
|
30 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
|
31 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
|
32 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
|
33 ; SI: s_and_b32 s{{[0-9]+, s[0-9]+, s[0-9]+}}
|
|
34
|
|
35 define amdgpu_kernel void @test4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
36 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
37 %a = load <4 x i32>, <4 x i32> addrspace(1) * %in
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
38 %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
39 %result = and <4 x i32> %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
40 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
41 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
42 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
43
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
44 ; FUNC-LABEL: {{^}}s_and_i32:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
45 ; SI: s_and_b32
|
121
|
46 define amdgpu_kernel void @s_and_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
47 %and = and i32 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
48 store i32 %and, i32 addrspace(1)* %out, align 4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
49 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
50 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
51
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
52 ; FUNC-LABEL: {{^}}s_and_constant_i32:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
53 ; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x12d687
|
121
|
54 define amdgpu_kernel void @s_and_constant_i32(i32 addrspace(1)* %out, i32 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
55 %and = and i32 %a, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
56 store i32 %and, i32 addrspace(1)* %out, align 4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
57 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
58 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
59
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
60 ; FIXME: We should really duplicate the constant so that the SALU use
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
61 ; can fold into the s_and_b32 and the VALU one is materialized
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
62 ; directly without copying from the SGPR.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
63
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
64 ; Second use is a VGPR use of the constant.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
65 ; FUNC-LABEL: {{^}}s_and_multi_use_constant_i32_0:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
66 ; SI: s_mov_b32 [[K:s[0-9]+]], 0x12d687
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
67 ; SI-DAG: s_and_b32 [[AND:s[0-9]+]], s{{[0-9]+}}, [[K]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
68 ; SI-DAG: v_mov_b32_e32 [[VK:v[0-9]+]], [[K]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
69 ; SI: buffer_store_dword [[VK]]
|
121
|
70 define amdgpu_kernel void @s_and_multi_use_constant_i32_0(i32 addrspace(1)* %out, i32 %a, i32 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
71 %and = and i32 %a, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
72
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
73 ; Just to stop future replacement of copy to vgpr + store with VALU op.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
74 %foo = add i32 %and, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
75 store volatile i32 %foo, i32 addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
76 store volatile i32 1234567, i32 addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
77 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
78 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
79
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
80 ; Second use is another SGPR use of the constant.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
81 ; FUNC-LABEL: {{^}}s_and_multi_use_constant_i32_1:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
82 ; SI: s_mov_b32 [[K:s[0-9]+]], 0x12d687
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
83 ; SI: s_and_b32 [[AND:s[0-9]+]], s{{[0-9]+}}, [[K]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
84 ; SI: s_add_i32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
85 ; SI: s_add_i32 [[ADD:s[0-9]+]], s{{[0-9]+}}, [[K]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
86 ; SI: buffer_store_dword [[VK]]
|
121
|
87 define amdgpu_kernel void @s_and_multi_use_constant_i32_1(i32 addrspace(1)* %out, i32 %a, i32 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
88 %and = and i32 %a, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
89 %foo = add i32 %and, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
90 %bar = add i32 %foo, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
91 store volatile i32 %bar, i32 addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
92 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
93 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
94
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
95 ; FUNC-LABEL: {{^}}v_and_i32_vgpr_vgpr:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
96 ; SI: v_and_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
121
|
97 define amdgpu_kernel void @v_and_i32_vgpr_vgpr(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
98 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
99 %gep.a = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
100 %gep.b = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
101 %gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
102 %a = load i32, i32 addrspace(1)* %gep.a
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
103 %b = load i32, i32 addrspace(1)* %gep.b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
104 %and = and i32 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
105 store i32 %and, i32 addrspace(1)* %gep.out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
106 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
107 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
108
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
109 ; FUNC-LABEL: {{^}}v_and_i32_sgpr_vgpr:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
110 ; SI-DAG: s_load_dword [[SA:s[0-9]+]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
111 ; SI-DAG: {{buffer|flat}}_load_dword [[VB:v[0-9]+]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
112 ; SI: v_and_b32_e32 v{{[0-9]+}}, [[SA]], [[VB]]
|
121
|
113 define amdgpu_kernel void @v_and_i32_sgpr_vgpr(i32 addrspace(1)* %out, i32 %a, i32 addrspace(1)* %bptr) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
114 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
115 %gep.b = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
116 %gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
117 %b = load i32, i32 addrspace(1)* %gep.b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
118 %and = and i32 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
119 store i32 %and, i32 addrspace(1)* %gep.out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
120 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
121 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
122
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
123 ; FUNC-LABEL: {{^}}v_and_i32_vgpr_sgpr:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
124 ; SI-DAG: s_load_dword [[SA:s[0-9]+]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
125 ; SI-DAG: {{buffer|flat}}_load_dword [[VB:v[0-9]+]]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
126 ; SI: v_and_b32_e32 v{{[0-9]+}}, [[SA]], [[VB]]
|
121
|
127 define amdgpu_kernel void @v_and_i32_vgpr_sgpr(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
128 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
129 %gep.a = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
130 %gep.out = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
131 %a = load i32, i32 addrspace(1)* %gep.a
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
132 %and = and i32 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
133 store i32 %and, i32 addrspace(1)* %gep.out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
134 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
135 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
136
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
137 ; FUNC-LABEL: {{^}}v_and_constant_i32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
138 ; SI: v_and_b32_e32 v{{[0-9]+}}, 0x12d687, v{{[0-9]+}}
|
121
|
139 define amdgpu_kernel void @v_and_constant_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr) {
|
|
140 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
141 %gep = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
|
|
142 %a = load i32, i32 addrspace(1)* %gep, align 4
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
143 %and = and i32 %a, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
144 store i32 %and, i32 addrspace(1)* %out, align 4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
145 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
146 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
147
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
148 ; FUNC-LABEL: {{^}}v_and_inline_imm_64_i32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
149 ; SI: v_and_b32_e32 v{{[0-9]+}}, 64, v{{[0-9]+}}
|
121
|
150 define amdgpu_kernel void @v_and_inline_imm_64_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr) {
|
|
151 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
152 %gep = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
|
|
153 %a = load i32, i32 addrspace(1)* %gep, align 4
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
154 %and = and i32 %a, 64
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
155 store i32 %and, i32 addrspace(1)* %out, align 4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
156 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
157 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
158
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
159 ; FUNC-LABEL: {{^}}v_and_inline_imm_neg_16_i32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
160 ; SI: v_and_b32_e32 v{{[0-9]+}}, -16, v{{[0-9]+}}
|
121
|
161 define amdgpu_kernel void @v_and_inline_imm_neg_16_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr) {
|
|
162 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
163 %gep = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
|
|
164 %a = load i32, i32 addrspace(1)* %gep, align 4
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
165 %and = and i32 %a, -16
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
166 store i32 %and, i32 addrspace(1)* %out, align 4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
167 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
168 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
169
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
170 ; FUNC-LABEL: {{^}}s_and_i64
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
171 ; SI: s_and_b64
|
121
|
172 define amdgpu_kernel void @s_and_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
173 %and = and i64 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
174 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
175 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
176 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
177
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
178 ; FIXME: Should use SGPRs
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
179 ; FUNC-LABEL: {{^}}s_and_i1:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
180 ; SI: v_and_b32
|
121
|
181 define amdgpu_kernel void @s_and_i1(i1 addrspace(1)* %out, i1 %a, i1 %b) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
182 %and = and i1 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
183 store i1 %and, i1 addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
184 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
185 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
186
|
100
|
187 ; FUNC-LABEL: {{^}}s_and_constant_i64:
|
|
188 ; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000{{$}}
|
|
189 ; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80{{$}}
|
|
190 ; SI: buffer_store_dwordx2
|
121
|
191 define amdgpu_kernel void @s_and_constant_i64(i64 addrspace(1)* %out, i64 %a) {
|
100
|
192 %and = and i64 %a, 549756338176
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
193 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
194 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
195 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
196
|
100
|
197 ; FUNC-LABEL: {{^}}s_and_multi_use_constant_i64:
|
|
198 ; XSI-DAG: s_mov_b32 s[[KLO:[0-9]+]], 0x80000{{$}}
|
|
199 ; XSI-DAG: s_mov_b32 s[[KHI:[0-9]+]], 0x80{{$}}
|
|
200 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, s{{\[}}[[KLO]]:[[KHI]]{{\]}}
|
121
|
201 define amdgpu_kernel void @s_and_multi_use_constant_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) {
|
100
|
202 %and0 = and i64 %a, 549756338176
|
|
203 %and1 = and i64 %b, 549756338176
|
|
204 store volatile i64 %and0, i64 addrspace(1)* %out
|
|
205 store volatile i64 %and1, i64 addrspace(1)* %out
|
|
206 ret void
|
|
207 }
|
|
208
|
|
209 ; FUNC-LABEL: {{^}}s_and_32_bit_constant_i64:
|
|
210 ; SI: s_load_dwordx2
|
|
211 ; SI-NOT: and
|
|
212 ; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x12d687{{$}}
|
|
213 ; SI-NOT: and
|
|
214 ; SI: buffer_store_dwordx2
|
121
|
215 define amdgpu_kernel void @s_and_32_bit_constant_i64(i64 addrspace(1)* %out, i64 %a) {
|
100
|
216 %and = and i64 %a, 1234567
|
|
217 store i64 %and, i64 addrspace(1)* %out, align 8
|
|
218 ret void
|
|
219 }
|
|
220
|
|
221 ; FUNC-LABEL: {{^}}s_and_multi_use_inline_imm_i64:
|
121
|
222 ; SI: s_load_dwordx2
|
120
|
223 ; SI: s_load_dword [[A:s[0-9]+]]
|
|
224 ; SI: s_load_dword [[B:s[0-9]+]]
|
100
|
225 ; SI: s_load_dwordx2
|
|
226 ; SI-NOT: and
|
120
|
227 ; SI: s_lshl_b32 [[A]], [[A]], 1
|
|
228 ; SI: s_lshl_b32 [[B]], [[B]], 1
|
|
229 ; SI: s_and_b32 s{{[0-9]+}}, [[A]], 62
|
|
230 ; SI: s_and_b32 s{{[0-9]+}}, [[B]], 62
|
100
|
231 ; SI-NOT: and
|
|
232 ; SI: buffer_store_dwordx2
|
121
|
233 define amdgpu_kernel void @s_and_multi_use_inline_imm_i64(i64 addrspace(1)* %out, i64 %a, i64 %b, i64 %c) {
|
100
|
234 %shl.a = shl i64 %a, 1
|
|
235 %shl.b = shl i64 %b, 1
|
|
236 %and0 = and i64 %shl.a, 62
|
|
237 %and1 = and i64 %shl.b, 62
|
|
238 %add0 = add i64 %and0, %c
|
|
239 %add1 = add i64 %and1, %c
|
|
240 store volatile i64 %add0, i64 addrspace(1)* %out
|
|
241 store volatile i64 %add1, i64 addrspace(1)* %out
|
|
242 ret void
|
|
243 }
|
|
244
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
245 ; FUNC-LABEL: {{^}}v_and_i64:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
246 ; SI: v_and_b32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
247 ; SI: v_and_b32
|
121
|
248 define amdgpu_kernel void @v_and_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) {
|
|
249 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
250 %gep.a = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
|
|
251 %a = load i64, i64 addrspace(1)* %gep.a, align 8
|
|
252 %gep.b = getelementptr i64, i64 addrspace(1)* %bptr, i32 %tid
|
|
253 %b = load i64, i64 addrspace(1)* %gep.b, align 8
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
254 %and = and i64 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
255 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
256 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
257 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
258
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
259 ; FUNC-LABEL: {{^}}v_and_constant_i64:
|
100
|
260 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, 0xab19b207, {{v[0-9]+}}
|
|
261 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, 0x11e, {{v[0-9]+}}
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
262 ; SI: buffer_store_dwordx2
|
121
|
263 define amdgpu_kernel void @v_and_constant_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
|
264 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
265 %gep.a = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
|
|
266 %a = load i64, i64 addrspace(1)* %gep.a, align 8
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
267 %and = and i64 %a, 1231231234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
268 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
269 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
270 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
271
|
100
|
272 ; FUNC-LABEL: {{^}}v_and_multi_use_constant_i64:
|
120
|
273 ; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO0:[0-9]+]]:[[HI0:[0-9]+]]{{\]}}
|
|
274 ; SI-DAG: buffer_load_dwordx2 v{{\[}}[[LO1:[0-9]+]]:[[HI1:[0-9]+]]{{\]}}
|
|
275 ; SI-DAG: s_movk_i32 [[KHI:s[0-9]+]], 0x11e{{$}}
|
100
|
276 ; SI-DAG: s_mov_b32 [[KLO:s[0-9]+]], 0xab19b207{{$}}
|
|
277 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, [[KLO]], v[[LO0]]
|
|
278 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, [[KHI]], v[[HI0]]
|
|
279 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, [[KLO]], v[[LO1]]
|
|
280 ; SI-DAG: v_and_b32_e32 {{v[0-9]+}}, [[KHI]], v[[HI1]]
|
|
281 ; SI: buffer_store_dwordx2
|
|
282 ; SI: buffer_store_dwordx2
|
121
|
283 define amdgpu_kernel void @v_and_multi_use_constant_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
100
|
284 %a = load volatile i64, i64 addrspace(1)* %aptr
|
|
285 %b = load volatile i64, i64 addrspace(1)* %aptr
|
|
286 %and0 = and i64 %a, 1231231234567
|
|
287 %and1 = and i64 %b, 1231231234567
|
|
288 store volatile i64 %and0, i64 addrspace(1)* %out
|
|
289 store volatile i64 %and1, i64 addrspace(1)* %out
|
|
290 ret void
|
|
291 }
|
|
292
|
|
293 ; FUNC-LABEL: {{^}}v_and_multi_use_inline_imm_i64:
|
|
294 ; SI: buffer_load_dwordx2 v{{\[}}[[LO0:[0-9]+]]:[[HI0:[0-9]+]]{{\]}}
|
|
295 ; SI-NOT: and
|
|
296 ; SI: buffer_load_dwordx2 v{{\[}}[[LO1:[0-9]+]]:[[HI1:[0-9]+]]{{\]}}
|
|
297 ; SI-NOT: and
|
|
298 ; SI: v_and_b32_e32 v[[RESLO0:[0-9]+]], 63, v[[LO0]]
|
|
299 ; SI: v_and_b32_e32 v[[RESLO1:[0-9]+]], 63, v[[LO1]]
|
|
300 ; SI-NOT: and
|
120
|
301 ; SI: buffer_store_dwordx2 v{{\[}}[[RESLO0]]
|
|
302 ; SI: buffer_store_dwordx2 v{{\[}}[[RESLO1]]
|
121
|
303 define amdgpu_kernel void @v_and_multi_use_inline_imm_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
100
|
304 %a = load volatile i64, i64 addrspace(1)* %aptr
|
|
305 %b = load volatile i64, i64 addrspace(1)* %aptr
|
|
306 %and0 = and i64 %a, 63
|
|
307 %and1 = and i64 %b, 63
|
|
308 store volatile i64 %and0, i64 addrspace(1)* %out
|
|
309 store volatile i64 %and1, i64 addrspace(1)* %out
|
|
310 ret void
|
|
311 }
|
|
312
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
313 ; FUNC-LABEL: {{^}}v_and_i64_32_bit_constant:
|
121
|
314 ; SI: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
|
100
|
315 ; SI-NOT: and
|
|
316 ; SI: v_and_b32_e32 {{v[0-9]+}}, 0x12d687, [[VAL]]
|
|
317 ; SI-NOT: and
|
|
318 ; SI: buffer_store_dwordx2
|
121
|
319 define amdgpu_kernel void @v_and_i64_32_bit_constant(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
|
320 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
321 %gep.a = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
|
|
322 %a = load i64, i64 addrspace(1)* %gep.a, align 8
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
323 %and = and i64 %a, 1234567
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
324 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
325 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
326 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
327
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
328 ; FUNC-LABEL: {{^}}v_and_inline_imm_i64:
|
121
|
329 ; SI: {{buffer|flat}}_load_dword v{{[0-9]+}}
|
100
|
330 ; SI-NOT: and
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
331 ; SI: v_and_b32_e32 {{v[0-9]+}}, 64, {{v[0-9]+}}
|
100
|
332 ; SI-NOT: and
|
|
333 ; SI: buffer_store_dwordx2
|
121
|
334 define amdgpu_kernel void @v_and_inline_imm_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
|
335 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
336 %gep.a = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
|
|
337 %a = load i64, i64 addrspace(1)* %gep.a, align 8
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
338 %and = and i64 %a, 64
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
339 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
340 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
341 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
342
|
120
|
343 ; FIXME: Should be able to reduce load width
|
|
344 ; FUNC-LABEL: {{^}}v_and_inline_neg_imm_i64:
|
121
|
345 ; SI: {{buffer|flat}}_load_dwordx2 v{{\[}}[[VAL_LO:[0-9]+]]:[[VAL_HI:[0-9]+]]{{\]}}
|
120
|
346 ; SI-NOT: and
|
|
347 ; SI: v_and_b32_e32 v[[VAL_LO]], -8, v[[VAL_LO]]
|
|
348 ; SI-NOT: and
|
|
349 ; SI: buffer_store_dwordx2 v{{\[}}[[VAL_LO]]:[[VAL_HI]]{{\]}}
|
121
|
350 define amdgpu_kernel void @v_and_inline_neg_imm_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr) {
|
|
351 %tid = call i32 @llvm.r600.read.tidig.x() #0
|
|
352 %gep.a = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
|
|
353 %a = load i64, i64 addrspace(1)* %gep.a, align 8
|
120
|
354 %and = and i64 %a, -8
|
|
355 store i64 %and, i64 addrspace(1)* %out, align 8
|
|
356 ret void
|
|
357 }
|
|
358
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
359 ; FUNC-LABEL: {{^}}s_and_inline_imm_64_i64
|
100
|
360 ; SI: s_load_dword
|
|
361 ; SI-NOT: and
|
|
362 ; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 64
|
|
363 ; SI-NOT: and
|
|
364 ; SI: buffer_store_dword
|
121
|
365 define amdgpu_kernel void @s_and_inline_imm_64_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
366 %and = and i64 %a, 64
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
367 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
368 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
369 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
370
|
100
|
371 ; FUNC-LABEL: {{^}}s_and_inline_imm_64_i64_noshrink:
|
120
|
372 ; SI: s_load_dword [[A:s[0-9]+]]
|
|
373 ; SI: s_lshl_b32 [[A]], [[A]], 1{{$}}
|
100
|
374 ; SI-NOT: and
|
120
|
375 ; SI: s_and_b32 s{{[0-9]+}}, [[A]], 64
|
100
|
376 ; SI-NOT: and
|
|
377 ; SI: s_add_u32
|
|
378 ; SI-NEXT: s_addc_u32
|
121
|
379 define amdgpu_kernel void @s_and_inline_imm_64_i64_noshrink(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a, i64 %b) {
|
100
|
380 %shl = shl i64 %a, 1
|
|
381 %and = and i64 %shl, 64
|
|
382 %add = add i64 %and, %b
|
|
383 store i64 %add, i64 addrspace(1)* %out, align 8
|
|
384 ret void
|
|
385 }
|
|
386
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
387 ; FUNC-LABEL: {{^}}s_and_inline_imm_1_i64
|
100
|
388 ; SI: s_load_dwordx2
|
|
389 ; SI-NOT: and
|
|
390 ; SI: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 1
|
|
391 ; SI-NOT: and
|
|
392 ; SI: buffer_store_dwordx2
|
121
|
393 define amdgpu_kernel void @s_and_inline_imm_1_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
394 %and = and i64 %a, 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
395 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
396 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
397 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
398
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
399 ; FUNC-LABEL: {{^}}s_and_inline_imm_1.0_i64
|
100
|
400 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 1.0
|
|
401
|
|
402 ; SI: s_load_dwordx2
|
|
403 ; SI: s_load_dwordx2
|
|
404 ; SI-NOT: and
|
|
405 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0x3ff00000
|
|
406 ; SI-NOT: and
|
|
407 ; SI: buffer_store_dwordx2
|
121
|
408 define amdgpu_kernel void @s_and_inline_imm_1.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
409 %and = and i64 %a, 4607182418800017408
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
410 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
411 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
412 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
413
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
414 ; FUNC-LABEL: {{^}}s_and_inline_imm_neg_1.0_i64
|
100
|
415 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, -1.0
|
|
416
|
|
417 ; SI: s_load_dwordx2
|
|
418 ; SI: s_load_dwordx2
|
|
419 ; SI-NOT: and
|
|
420 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0xbff00000
|
|
421 ; SI-NOT: and
|
|
422 ; SI: buffer_store_dwordx2
|
121
|
423 define amdgpu_kernel void @s_and_inline_imm_neg_1.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
424 %and = and i64 %a, 13830554455654793216
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
425 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
426 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
427 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
428
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
429 ; FUNC-LABEL: {{^}}s_and_inline_imm_0.5_i64
|
100
|
430 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0.5
|
|
431
|
|
432 ; SI: s_load_dwordx2
|
|
433 ; SI: s_load_dwordx2
|
|
434 ; SI-NOT: and
|
|
435 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0x3fe00000
|
|
436 ; SI-NOT: and
|
|
437 ; SI: buffer_store_dwordx2
|
121
|
438 define amdgpu_kernel void @s_and_inline_imm_0.5_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
439 %and = and i64 %a, 4602678819172646912
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
440 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
441 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
442 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
443
|
100
|
444 ; FUNC-LABEL: {{^}}s_and_inline_imm_neg_0.5_i64:
|
|
445 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, -0.5
|
|
446
|
|
447 ; SI: s_load_dwordx2
|
|
448 ; SI: s_load_dwordx2
|
|
449 ; SI-NOT: and
|
|
450 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0xbfe00000
|
|
451 ; SI-NOT: and
|
|
452 ; SI: buffer_store_dwordx2
|
121
|
453 define amdgpu_kernel void @s_and_inline_imm_neg_0.5_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
454 %and = and i64 %a, 13826050856027422720
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
455 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
456 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
457 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
458
|
100
|
459 ; FUNC-LABEL: {{^}}s_and_inline_imm_2.0_i64:
|
|
460 ; SI: s_load_dwordx2
|
|
461 ; SI: s_load_dwordx2
|
|
462 ; SI-NOT: and
|
|
463 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 2.0
|
|
464 ; SI-NOT: and
|
|
465 ; SI: buffer_store_dwordx2
|
121
|
466 define amdgpu_kernel void @s_and_inline_imm_2.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
467 %and = and i64 %a, 4611686018427387904
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
468 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
469 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
470 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
471
|
100
|
472 ; FUNC-LABEL: {{^}}s_and_inline_imm_neg_2.0_i64:
|
|
473 ; SI: s_load_dwordx2
|
|
474 ; SI: s_load_dwordx2
|
|
475 ; SI-NOT: and
|
|
476 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, -2.0
|
|
477 ; SI-NOT: and
|
|
478 ; SI: buffer_store_dwordx2
|
121
|
479 define amdgpu_kernel void @s_and_inline_imm_neg_2.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
480 %and = and i64 %a, 13835058055282163712
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
481 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
482 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
483 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
484
|
100
|
485 ; FUNC-LABEL: {{^}}s_and_inline_imm_4.0_i64:
|
|
486 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 4.0
|
|
487
|
|
488 ; SI: s_load_dwordx2
|
|
489 ; SI: s_load_dwordx2
|
|
490 ; SI-NOT: and
|
|
491 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0x40100000
|
|
492 ; SI-NOT: and
|
|
493 ; SI: buffer_store_dwordx2
|
121
|
494 define amdgpu_kernel void @s_and_inline_imm_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
495 %and = and i64 %a, 4616189618054758400
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
496 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
497 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
498 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
499
|
100
|
500 ; FUNC-LABEL: {{^}}s_and_inline_imm_neg_4.0_i64:
|
|
501 ; XSI: s_and_b64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, -4.0
|
|
502
|
|
503 ; SI: s_load_dwordx2
|
|
504 ; SI: s_load_dwordx2
|
|
505 ; SI-NOT: and
|
|
506 ; SI: s_and_b32 {{s[0-9]+}}, {{s[0-9]+}}, 0xc0100000
|
|
507 ; SI-NOT: and
|
|
508 ; SI: buffer_store_dwordx2
|
121
|
509 define amdgpu_kernel void @s_and_inline_imm_neg_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
510 %and = and i64 %a, 13839561654909534208
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
511 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
512 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
513 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
514
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
515
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
516 ; Test with the 64-bit integer bitpattern for a 32-bit float in the
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
517 ; low 32-bits, which is not a valid 64-bit inline immmediate.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
518
|
100
|
519 ; FUNC-LABEL: {{^}}s_and_inline_imm_f32_4.0_i64:
|
|
520 ; SI: s_load_dwordx2
|
|
521 ; SI: s_load_dword s
|
|
522 ; SI-NOT: and
|
|
523 ; SI: s_and_b32 s[[K_HI:[0-9]+]], s{{[0-9]+}}, 4.0
|
|
524 ; SI-NOT: and
|
|
525 ; SI: buffer_store_dwordx2
|
121
|
526 define amdgpu_kernel void @s_and_inline_imm_f32_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
527 %and = and i64 %a, 1082130432
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
528 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
529 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
530 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
531
|
100
|
532 ; FUNC-LABEL: {{^}}s_and_inline_imm_f32_neg_4.0_i64:
|
|
533 ; SI: s_load_dwordx2
|
|
534 ; SI: s_load_dwordx2
|
|
535 ; SI-NOT: and
|
|
536 ; SI: s_and_b32 s[[K_HI:[0-9]+]], s{{[0-9]+}}, -4.0
|
|
537 ; SI-NOT: and
|
|
538 ; SI: buffer_store_dwordx2
|
121
|
539 define amdgpu_kernel void @s_and_inline_imm_f32_neg_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
540 %and = and i64 %a, -1065353216
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
541 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
542 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
543 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
544
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
545 ; Shift into upper 32-bits
|
100
|
546 ; SI: s_load_dwordx2
|
|
547 ; SI: s_load_dwordx2
|
|
548 ; SI-NOT: and
|
|
549 ; SI: s_and_b32 s[[K_HI:[0-9]+]], s{{[0-9]+}}, 4.0
|
|
550 ; SI-NOT: and
|
|
551 ; SI: buffer_store_dwordx2
|
121
|
552 define amdgpu_kernel void @s_and_inline_high_imm_f32_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
553 %and = and i64 %a, 4647714815446351872
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
554 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
555 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
556 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
557
|
100
|
558 ; FUNC-LABEL: {{^}}s_and_inline_high_imm_f32_neg_4.0_i64:
|
|
559 ; SI: s_load_dwordx2
|
|
560 ; SI: s_load_dwordx2
|
|
561 ; SI-NOT: and
|
|
562 ; SI: s_and_b32 s[[K_HI:[0-9]+]], s{{[0-9]+}}, -4.0
|
|
563 ; SI-NOT: and
|
|
564 ; SI: buffer_store_dwordx2
|
121
|
565 define amdgpu_kernel void @s_and_inline_high_imm_f32_neg_4.0_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 %a) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
566 %and = and i64 %a, 13871086852301127680
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
567 store i64 %and, i64 addrspace(1)* %out, align 8
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
568 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
569 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
570 attributes #0 = { nounwind readnone }
|