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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr="-wavefrontsize32,+wavefrontsize64" -o - < %s | FileCheck -check-prefixes=SDAG-W64 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr="-wavefrontsize32,+wavefrontsize64" -o - < %s | FileCheck -check-prefixes=GISEL-W64 %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr="+wavefrontsize32,-wavefrontsize64" -o - < %s | FileCheck -check-prefixes=SDAG-W32 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -mattr="+wavefrontsize32,-wavefrontsize64" -o - < %s | FileCheck -check-prefixes=GISEL-W32 %s
; Use ballot for easy access to lanemask
define amdgpu_ps i64 @test_nor(i64 inreg %a, i64 inreg %b) {
; SDAG-W64-LABEL: test_nor:
; SDAG-W64: ; %bb.0:
; SDAG-W64-NEXT: s_nor_b64 s[0:1], s[0:1], s[2:3]
; SDAG-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; SDAG-W64-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1]
; SDAG-W64-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v0
; SDAG-W64-NEXT: ; return to shader part epilog
;
; GISEL-W64-LABEL: test_nor:
; GISEL-W64: ; %bb.0:
; GISEL-W64-NEXT: s_nor_b64 s[0:1], s[0:1], s[2:3]
; GISEL-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GISEL-W64-NEXT: s_and_b64 s[0:1], s[0:1], exec
; GISEL-W64-NEXT: ; return to shader part epilog
;
; SDAG-W32-LABEL: test_nor:
; SDAG-W32: ; %bb.0:
; SDAG-W32-NEXT: s_nor_b32 s0, s0, s2
; SDAG-W32-NEXT: s_mov_b32 s1, 0
; SDAG-W32-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
; SDAG-W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; SDAG-W32-NEXT: v_cmp_ne_u32_e64 s0, 0, v0
; SDAG-W32-NEXT: ; return to shader part epilog
;
; GISEL-W32-LABEL: test_nor:
; GISEL-W32: ; %bb.0:
; GISEL-W32-NEXT: s_nor_b32 s0, s0, s2
; GISEL-W32-NEXT: s_mov_b32 s1, 0
; GISEL-W32-NEXT: s_and_b32 s0, s0, exec_lo
; GISEL-W32-NEXT: ; return to shader part epilog
%a.lanemask = call i1 @llvm.amdgcn.inverse.ballot.i64(i64 %a)
%b.lanemask = call i1 @llvm.amdgcn.inverse.ballot.i64(i64 %b)
%or = or i1 %a.lanemask, %b.lanemask
%xor = xor i1 %or, true
%r = call i64 @llvm.amdgcn.ballot.i64(i1 %xor)
ret i64 %r
}
define amdgpu_ps i64 @test_or_two_uses(i64 inreg %a, i64 inreg %b) {
; SDAG-W64-LABEL: test_or_two_uses:
; SDAG-W64: ; %bb.0:
; SDAG-W64-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3]
; SDAG-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_2)
; SDAG-W64-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[0:1]
; SDAG-W64-NEXT: s_xor_b64 s[0:1], s[0:1], -1
; SDAG-W64-NEXT: s_waitcnt_depctr 0xfffe
; SDAG-W64-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[0:1]
; SDAG-W64-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
; SDAG-W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; SDAG-W64-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v1
; SDAG-W64-NEXT: s_and_b64 s[0:1], s[0:1], vcc
; SDAG-W64-NEXT: s_waitcnt_depctr 0xfffe
; SDAG-W64-NEXT: ; return to shader part epilog
;
; GISEL-W64-LABEL: test_or_two_uses:
; GISEL-W64: ; %bb.0:
; GISEL-W64-NEXT: s_or_b64 s[0:1], s[0:1], s[2:3]
; GISEL-W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
; GISEL-W64-NEXT: s_xor_b64 s[2:3], s[0:1], -1
; GISEL-W64-NEXT: s_and_b64 s[0:1], s[0:1], exec
; GISEL-W64-NEXT: s_and_b64 s[2:3], s[2:3], exec
; GISEL-W64-NEXT: s_and_b64 s[0:1], s[2:3], s[0:1]
; GISEL-W64-NEXT: ; return to shader part epilog
;
; SDAG-W32-LABEL: test_or_two_uses:
; SDAG-W32: ; %bb.0:
; SDAG-W32-NEXT: s_or_b32 s0, s0, s2
; SDAG-W32-NEXT: s_mov_b32 s3, 0
; SDAG-W32-NEXT: v_cndmask_b32_e64 v0, 0, 1, s0
; SDAG-W32-NEXT: s_xor_b32 s0, s0, -1
; SDAG-W32-NEXT: s_mov_b32 s1, s3
; SDAG-W32-NEXT: v_cndmask_b32_e64 v1, 0, 1, s0
; SDAG-W32-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; SDAG-W32-NEXT: v_cmp_ne_u32_e64 s0, 0, v0
; SDAG-W32-NEXT: v_cmp_ne_u32_e64 s2, 0, v1
; SDAG-W32-NEXT: s_and_b64 s[0:1], s[2:3], s[0:1]
; SDAG-W32-NEXT: ; return to shader part epilog
;
; GISEL-W32-LABEL: test_or_two_uses:
; GISEL-W32: ; %bb.0:
; GISEL-W32-NEXT: s_or_b32 s0, s0, s2
; GISEL-W32-NEXT: s_mov_b32 s1, 0
; GISEL-W32-NEXT: s_xor_b32 s4, s0, -1
; GISEL-W32-NEXT: s_and_b32 s2, s0, exec_lo
; GISEL-W32-NEXT: s_mov_b32 s3, s1
; GISEL-W32-NEXT: s_and_b32 s0, s4, exec_lo
; GISEL-W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GISEL-W32-NEXT: s_and_b64 s[0:1], s[0:1], s[2:3]
; GISEL-W32-NEXT: ; return to shader part epilog
%a.lanemask = call i1 @llvm.amdgcn.inverse.ballot.i64(i64 %a)
%b.lanemask = call i1 @llvm.amdgcn.inverse.ballot.i64(i64 %b)
%or = or i1 %a.lanemask, %b.lanemask
%xor = xor i1 %or, true
%r0 = call i64 @llvm.amdgcn.ballot.i64(i1 %xor)
%r1 = call i64 @llvm.amdgcn.ballot.i64(i1 %or)
%r = and i64 %r0, %r1
ret i64 %r
}
|
