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
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX12 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250 %s
define amdgpu_cs float @test_cvt_f32_bf8_byte0(i32 %a) {
; GFX12-LABEL: test_cvt_f32_bf8_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_bf8_dpp v0, v0 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_cvt_f32_bf8_dpp v0, v0 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: ; return to shader part epilog
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %tmp0, i32 0)
ret float %ret
}
define amdgpu_cs float @test_cvt_f32_bf8_byte1(i32 %a) {
; GFX12-LABEL: test_cvt_f32_bf8_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_bf8_e64_dpp v0, v0 byte_sel:1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_cvt_f32_bf8_e64_dpp v0, v0 byte_sel:1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: ; return to shader part epilog
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %tmp0, i32 1)
ret float %ret
}
define amdgpu_cs float @test_cvt_f32_bf8_byte2(i32 %a) {
; GFX12-LABEL: test_cvt_f32_bf8_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_bf8_e64_dpp v0, v0 byte_sel:2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_bf8_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_cvt_f32_bf8_e64_dpp v0, v0 byte_sel:2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: ; return to shader part epilog
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%ret = tail call float @llvm.amdgcn.cvt.f32.bf8(i32 %tmp0, i32 2)
ret float %ret
}
define amdgpu_cs float @test_cvt_f32_fp8_byte3(i32 %a) {
; GFX12-LABEL: test_cvt_f32_fp8_byte3:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_f32_fp8_e64_dpp v0, v0 byte_sel:3 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: ; return to shader part epilog
;
; GFX1250-LABEL: test_cvt_f32_fp8_byte3:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_cvt_f32_fp8_e64_dpp v0, v0 byte_sel:3 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: ; return to shader part epilog
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%ret = tail call float @llvm.amdgcn.cvt.f32.fp8(i32 %tmp0, i32 3)
ret float %ret
}
define amdgpu_cs void @test_cvt_pk_bf8_f32_word0(i32 %a, float %y, i32 %old, ptr addrspace(1) %out) {
; GFX12-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_pk_bf8_f32_e64_dpp v2, v0, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: global_store_b32 v[3:4], v2, off
; GFX12-NEXT: s_endpgm
;
; GFX1250-LABEL: test_cvt_pk_bf8_f32_word0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_dual_mov_b32 v5, v4 :: v_dual_mov_b32 v4, v3
; GFX1250-NEXT: v_cvt_pk_bf8_f32_e64_dpp v2, v0, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: global_store_b32 v[4:5], v2, off
; GFX1250-NEXT: s_endpgm
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%tmp1 = bitcast i32 %tmp0 to float
%ret = tail call i32 @llvm.amdgcn.cvt.pk.bf8.f32(float %tmp1, float %y, i32 %old, i1 false)
store i32 %ret, ptr addrspace(1) %out
ret void
}
define amdgpu_cs void @test_cvt_pk_fp8_f32_word1(i32 %a, float %y, i32 %old, ptr addrspace(1) %out) {
; GFX12-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_mov_b32_dpp v0, v0 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX12-NEXT: global_store_b32 v[3:4], v2, off
; GFX12-NEXT: s_endpgm
;
; GFX1250-LABEL: test_cvt_pk_fp8_f32_word1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_mov_b32_dpp v0, v0 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: v_dual_mov_b32 v5, v4 :: v_dual_mov_b32 v4, v3
; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX1250-NEXT: v_cvt_pk_fp8_f32 v2, v0, v1 op_sel:[0,0,1]
; GFX1250-NEXT: global_store_b32 v[4:5], v2, off
; GFX1250-NEXT: s_endpgm
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%tmp1 = bitcast i32 %tmp0 to float
%ret = tail call i32 @llvm.amdgcn.cvt.pk.fp8.f32(float %tmp1, float %y, i32 %old, i1 true)
store i32 %ret, ptr addrspace(1) %out
ret void
}
define amdgpu_cs void @test_cvt_sr_bf8_f32_byte0(i32 %a, i32 %r, i32 %old, ptr addrspace(1) %out) {
; GFX12-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_sr_bf8_f32_e64_dpp v2, v0, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: global_store_b32 v[3:4], v2, off
; GFX12-NEXT: s_endpgm
;
; GFX1250-LABEL: test_cvt_sr_bf8_f32_byte0:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_dual_mov_b32 v5, v4 :: v_dual_mov_b32 v4, v3
; GFX1250-NEXT: v_cvt_sr_bf8_f32_e64_dpp v2, v0, v1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: global_store_b32 v[4:5], v2, off
; GFX1250-NEXT: s_endpgm
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%tmp1 = bitcast i32 %tmp0 to float
%ret = tail call i32 @llvm.amdgcn.cvt.sr.bf8.f32(float %tmp1, i32 %r, i32 %old, i32 0)
store i32 %ret, ptr addrspace(1) %out
ret void
}
define amdgpu_cs void @test_cvt_sr_fp8_f32_byte1(i32 %a, i32 %r, i32 %old, ptr addrspace(1) %out) {
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_sr_fp8_f32_e64_dpp v2, v0, v1 byte_sel:1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: global_store_b32 v[3:4], v2, off
; GFX12-NEXT: s_endpgm
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte1:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_dual_mov_b32 v5, v4 :: v_dual_mov_b32 v4, v3
; GFX1250-NEXT: v_cvt_sr_fp8_f32_e64_dpp v2, v0, v1 byte_sel:1 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: global_store_b32 v[4:5], v2, off
; GFX1250-NEXT: s_endpgm
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%tmp1 = bitcast i32 %tmp0 to float
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %tmp1, i32 %r, i32 %old, i32 1)
store i32 %ret, ptr addrspace(1) %out
ret void
}
define amdgpu_cs void @test_cvt_sr_fp8_f32_byte2(i32 %a, i32 %r, i32 %old, ptr addrspace(1) %out) {
; GFX12-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_cvt_sr_fp8_f32_e64_dpp v2, v0, v1 byte_sel:2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX12-NEXT: global_store_b32 v[3:4], v2, off
; GFX12-NEXT: s_endpgm
;
; GFX1250-LABEL: test_cvt_sr_fp8_f32_byte2:
; GFX1250: ; %bb.0:
; GFX1250-NEXT: v_dual_mov_b32 v5, v4 :: v_dual_mov_b32 v4, v3
; GFX1250-NEXT: v_cvt_sr_fp8_f32_e64_dpp v2, v0, v1 byte_sel:2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX1250-NEXT: global_store_b32 v[4:5], v2, off
; GFX1250-NEXT: s_endpgm
%tmp0 = call i32 @llvm.amdgcn.mov.dpp.i32(i32 %a, i32 228, i32 15, i32 15, i1 1)
%tmp1 = bitcast i32 %tmp0 to float
%ret = tail call i32 @llvm.amdgcn.cvt.sr.fp8.f32(float %tmp1, i32 %r, i32 %old, i32 2)
store i32 %ret, ptr addrspace(1) %out
ret void
}
declare float @llvm.amdgcn.cvt.f32.bf8(i32, i32)
declare float @llvm.amdgcn.cvt.f32.fp8(i32, i32)
declare i32 @llvm.amdgcn.cvt.pk.bf8.f32(float, float, i32, i1)
declare i32 @llvm.amdgcn.cvt.pk.fp8.f32(float, float, i32, i1)
declare i32 @llvm.amdgcn.cvt.sr.bf8.f32(float, i32, i32, i32)
declare i32 @llvm.amdgcn.cvt.sr.fp8.f32(float, i32, i32, i32)
declare i32 @llvm.amdgcn.mov.dpp.i32(i32, i32, i32, i32, i1) #1
declare i32 @llvm.amdgcn.mov.dpp8.i32(i32, i32) #1
attributes #0 = { nounwind convergent }
attributes #1 = { nounwind readnone convergent }
|
