1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; GCN-LABEL: {{^}}const_load_no_shrink_dword_to_unaligned_byte:
; GCN: s_load_dword s{{[0-9]+}}
; GCN: s_load_dword [[LD:s[0-9]+]],
; GCN: s_bfe_i32 s{{[0-9]+}}, [[LD]], 0x10013
define amdgpu_kernel void @const_load_no_shrink_dword_to_unaligned_byte(ptr addrspace(1) %out, ptr addrspace(4) %in, i32 %x) {
%ptr = getelementptr i32, ptr addrspace(4) %in, i32 %x
%load = load i32, ptr addrspace(4) %ptr, align 4
%and = and i32 %load, 524288
%cmp = icmp eq i32 %and, 0
%sel = select i1 %cmp, i32 0, i32 -1
store i32 %sel, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: const_load_no_shrink_dword_to_aligned_byte:
; GCN: s_load_dword s{{[0-9]+}}
; GCN: s_load_dword [[LD:s[0-9]+]],
; GCN: s_bfe_i32 s{{[0-9]+}}, [[LD]], 0x10003
define amdgpu_kernel void @const_load_no_shrink_dword_to_aligned_byte(ptr addrspace(1) %out, ptr addrspace(4) %in, i32 %x) {
%ptr = getelementptr i32, ptr addrspace(4) %in, i32 %x
%load = load i32, ptr addrspace(4) %ptr, align 4
%and = and i32 %load, 8
%cmp = icmp eq i32 %and, 0
%sel = select i1 %cmp, i32 0, i32 -1
store i32 %sel, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: global_load_no_shrink_dword_to_unaligned_byte:
; GCN: s_load_dword s{{[0-9]+}}
; GCN: s_load_dword [[LD:s[0-9]+]],
; GCN: s_bfe_i32 s{{[0-9]+}}, [[LD]], 0x10013
define amdgpu_kernel void @global_load_no_shrink_dword_to_unaligned_byte(ptr addrspace(1) noalias %out, ptr addrspace(1) noalias %in, i32 %x) {
%ptr = getelementptr i32, ptr addrspace(1) %in, i32 %x
%load = load i32, ptr addrspace(1) %ptr, align 4
%and = and i32 %load, 524288
%cmp = icmp eq i32 %and, 0
%sel = select i1 %cmp, i32 0, i32 -1
store i32 %sel, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: global_load_no_shrink_dword_to_aligned_byte:
; GCN: s_load_dword s{{[0-9]+}}
; GCN: s_load_dword [[LD:s[0-9]+]],
; GCN: s_bfe_i32 s{{[0-9]+}}, [[LD]], 0x10003
define amdgpu_kernel void @global_load_no_shrink_dword_to_aligned_byte(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %x) {
%ptr = getelementptr i32, ptr addrspace(1) %in, i32 %x
%load = load i32, ptr addrspace(1) %ptr, align 4
%and = and i32 %load, 8
%cmp = icmp eq i32 %and, 0
%sel = select i1 %cmp, i32 0, i32 -1
store i32 %sel, ptr addrspace(1) %out
ret void
}
; GCN-LABEL: const_load_shrink_dword_to_unaligned_byte:
; GCN: global_load_ushort
define amdgpu_kernel void @const_load_shrink_dword_to_unaligned_byte(ptr addrspace(1) %out, ptr addrspace(4) %in, i32 %x) {
%ptr = getelementptr i32, ptr addrspace(4) %in, i32 %x
%load = load i32, ptr addrspace(4) %ptr, align 2
%and = and i32 %load, 524288
%cmp = icmp eq i32 %and, 0
%sel = select i1 %cmp, i32 0, i32 -1
store i32 %sel, ptr addrspace(1) %out
ret void
}
|
