1
; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3
declare i32 @llvm.r600.read.tidig.x() nounwind readnone
5
; FUNC-LABEL: @v_test_imin_sle_i32
7
define void @v_test_imin_sle_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
8
%tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
9
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
10
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
11
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
12
%a = load i32, i32 addrspace(1)* %gep0, align 4
13
%b = load i32, i32 addrspace(1)* %gep1, align 4
14
%cmp = icmp sle i32 %a, %b
15
%val = select i1 %cmp, i32 %a, i32 %b
16
store i32 %val, i32 addrspace(1)* %outgep, align 4
20
; FUNC-LABEL: @s_test_imin_sle_i32
22
define void @s_test_imin_sle_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
23
%cmp = icmp sle i32 %a, %b
24
%val = select i1 %cmp, i32 %a, i32 %b
25
store i32 %val, i32 addrspace(1)* %out, align 4
29
; FUNC-LABEL: @v_test_imin_slt_i32
31
define void @v_test_imin_slt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
32
%tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
33
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
34
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
35
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
36
%a = load i32, i32 addrspace(1)* %gep0, align 4
37
%b = load i32, i32 addrspace(1)* %gep1, align 4
38
%cmp = icmp slt i32 %a, %b
39
%val = select i1 %cmp, i32 %a, i32 %b
40
store i32 %val, i32 addrspace(1)* %outgep, align 4
44
; FUNC-LABEL: @s_test_imin_slt_i32
46
define void @s_test_imin_slt_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
47
%cmp = icmp slt i32 %a, %b
48
%val = select i1 %cmp, i32 %a, i32 %b
49
store i32 %val, i32 addrspace(1)* %out, align 4
53
; FUNC-LABEL: {{^}}s_test_imin_slt_imm_i32:
54
; SI: s_min_i32 {{s[0-9]+}}, {{s[0-9]+}}, 8
55
define void @s_test_imin_slt_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
56
%cmp = icmp slt i32 %a, 8
57
%val = select i1 %cmp, i32 %a, i32 8
58
store i32 %val, i32 addrspace(1)* %out, align 4
62
; FUNC-LABEL: {{^}}s_test_imin_sle_imm_i32:
63
; SI: s_min_i32 {{s[0-9]+}}, {{s[0-9]+}}, 8
64
define void @s_test_imin_sle_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
65
%cmp = icmp sle i32 %a, 8
66
%val = select i1 %cmp, i32 %a, i32 8
67
store i32 %val, i32 addrspace(1)* %out, align 4
71
; FUNC-LABEL: @v_test_umin_ule_i32
73
define void @v_test_umin_ule_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
74
%tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
75
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
76
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
77
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
78
%a = load i32, i32 addrspace(1)* %gep0, align 4
79
%b = load i32, i32 addrspace(1)* %gep1, align 4
80
%cmp = icmp ule i32 %a, %b
81
%val = select i1 %cmp, i32 %a, i32 %b
82
store i32 %val, i32 addrspace(1)* %outgep, align 4
86
; FUNC-LABEL: @s_test_umin_ule_i32
88
define void @s_test_umin_ule_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
89
%cmp = icmp ule i32 %a, %b
90
%val = select i1 %cmp, i32 %a, i32 %b
91
store i32 %val, i32 addrspace(1)* %out, align 4
95
; FUNC-LABEL: @v_test_umin_ult_i32
97
define void @v_test_umin_ult_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
98
%tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
99
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
100
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
101
%outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
102
%a = load i32, i32 addrspace(1)* %gep0, align 4
103
%b = load i32, i32 addrspace(1)* %gep1, align 4
104
%cmp = icmp ult i32 %a, %b
105
%val = select i1 %cmp, i32 %a, i32 %b
106
store i32 %val, i32 addrspace(1)* %outgep, align 4
110
; FUNC-LABEL: @s_test_umin_ult_i32
112
define void @s_test_umin_ult_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
113
%cmp = icmp ult i32 %a, %b
114
%val = select i1 %cmp, i32 %a, i32 %b
115
store i32 %val, i32 addrspace(1)* %out, align 4
119
; FUNC-LABEL: @v_test_umin_ult_i32_multi_use
122
; SI-NEXT: v_cndmask_b32
125
define void @v_test_umin_ult_i32_multi_use(i32 addrspace(1)* %out0, i1 addrspace(1)* %out1, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
126
%tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
127
%gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
128
%gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
129
%outgep0 = getelementptr i32, i32 addrspace(1)* %out0, i32 %tid
130
%outgep1 = getelementptr i1, i1 addrspace(1)* %out1, i32 %tid
131
%a = load i32, i32 addrspace(1)* %gep0, align 4
132
%b = load i32, i32 addrspace(1)* %gep1, align 4
133
%cmp = icmp ult i32 %a, %b
134
%val = select i1 %cmp, i32 %a, i32 %b
135
store i32 %val, i32 addrspace(1)* %outgep0, align 4
136
store i1 %cmp, i1 addrspace(1)* %outgep1
140
; Make sure redundant and removed
141
; FUNC-LABEL: {{^}}simplify_demanded_bits_test_umin_ult_i16:
142
; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
143
; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
144
; SI: s_min_u32 [[MIN:s[0-9]+]], [[A]], [[B]]
145
; SI-NEXT: v_mov_b32_e32 [[VMIN:v[0-9]+]], [[MIN]]
146
; SI-NEXT: buffer_store_dword [[VMIN]]
147
define void @simplify_demanded_bits_test_umin_ult_i16(i32 addrspace(1)* %out, i16 zeroext %a, i16 zeroext %b) nounwind {
148
%a.ext = zext i16 %a to i32
149
%b.ext = zext i16 %b to i32
150
%cmp = icmp ult i32 %a.ext, %b.ext
151
%val = select i1 %cmp, i32 %a.ext, i32 %b.ext
152
%mask = and i32 %val, 65535
153
store i32 %mask, i32 addrspace(1)* %out
157
; Make sure redundant sign_extend_inreg removed.
159
; FUNC-LABEL: {{^}}simplify_demanded_bits_test_min_slt_i16:
160
; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
161
; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
162
; SI: s_min_i32 [[MIN:s[0-9]+]], [[A]], [[B]]
163
; SI-NEXT: v_mov_b32_e32 [[VMIN:v[0-9]+]], [[MIN]]
164
; SI-NEXT: buffer_store_dword [[VMIN]]
165
define void @simplify_demanded_bits_test_min_slt_i16(i32 addrspace(1)* %out, i16 signext %a, i16 signext %b) nounwind {
166
%a.ext = sext i16 %a to i32
167
%b.ext = sext i16 %b to i32
168
%cmp = icmp slt i32 %a.ext, %b.ext
169
%val = select i1 %cmp, i32 %a.ext, i32 %b.ext
170
%shl = shl i32 %val, 16
171
%sextinreg = ashr i32 %shl, 16
172
store i32 %sextinreg, i32 addrspace(1)* %out
176
; FIXME: Should get match min/max through extends inserted by
179
; FUNC-LABEL: {{^}}s_test_imin_sle_i16:
182
; SI: v_cmp_le_i32_e32
184
define void @s_test_imin_sle_i16(i16 addrspace(1)* %out, i16 %a, i16 %b) nounwind {
185
%cmp = icmp sle i16 %a, %b
186
%val = select i1 %cmp, i16 %a, i16 %b
187
store i16 %val, i16 addrspace(1)* %out