1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -amdgpu-sdwa-peephole=0 -mattr=-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=NOSDWA,GCN %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=fiji -amdgpu-sdwa-peephole -mattr=-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI,SDWA,GCN %s
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -amdgpu-sdwa-peephole -mattr=-fp64-fp16-denormals -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9,SDWA,GCN %s
5 ; GCN-LABEL: {{^}}add_shr_i32:
6 ; NOSDWA: v_lshrrev_b32_e32 v[[DST:[0-9]+]], 16, v{{[0-9]+}}
7 ; NOSDWA: v_add_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v[[DST]]
8 ; NOSDWA-NOT: v_add_{{(_co)?}}_u32_sdwa
10 ; VI: v_add_u32_sdwa v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
11 ; GFX9: v_add_u32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
13 define amdgpu_kernel void @add_shr_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
14 %a = load i32, i32 addrspace(1)* %in, align 4
15 %shr = lshr i32 %a, 16
16 %add = add i32 %a, %shr
17 store i32 %add, i32 addrspace(1)* %out, align 4
21 ; GCN-LABEL: {{^}}sub_shr_i32:
22 ; NOSDWA: v_lshrrev_b32_e32 v[[DST:[0-9]+]], 16, v{{[0-9]+}}
23 ; NOSDWA: v_subrev_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v[[DST]]
24 ; NOSDWA-NOT: v_subrev_{{(_co)?}}_u32_sdwa
26 ; VI: v_subrev_u32_sdwa v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
27 ; GFX9: v_sub_u32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
28 define amdgpu_kernel void @sub_shr_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
29 %a = load i32, i32 addrspace(1)* %in, align 4
30 %shr = lshr i32 %a, 16
31 %sub = sub i32 %shr, %a
32 store i32 %sub, i32 addrspace(1)* %out, align 4
36 ; GCN-LABEL: {{^}}mul_shr_i32:
37 ; NOSDWA: v_lshrrev_b32_e32 v[[DST0:[0-9]+]], 16, v{{[0-9]+}}
38 ; NOSDWA: v_lshrrev_b32_e32 v[[DST1:[0-9]+]], 16, v{{[0-9]+}}
39 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v[[DST0]], v[[DST1]]
40 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
42 ; SDWA: v_mul_u32_u24_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
44 define amdgpu_kernel void @mul_shr_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in1, i32 addrspace(1)* %in2) {
45 %a = load i32, i32 addrspace(1)* %in1, align 4
46 %b = load i32, i32 addrspace(1)* %in2, align 4
47 %shra = lshr i32 %a, 16
48 %shrb = lshr i32 %b, 16
49 %mul = mul i32 %shra, %shrb
50 store i32 %mul, i32 addrspace(1)* %out, align 4
54 ; GCN-LABEL: {{^}}mul_i16:
55 ; NOSDWA: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
56 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
57 ; SDWA: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
58 ; SDWA-NOT: v_mul_u32_u24_sdwa
60 define amdgpu_kernel void @mul_i16(i16 addrspace(1)* %out, i16 addrspace(1)* %ina, i16 addrspace(1)* %inb) {
62 %a = load i16, i16 addrspace(1)* %ina, align 4
63 %b = load i16, i16 addrspace(1)* %inb, align 4
65 store i16 %mul, i16 addrspace(1)* %out, align 4
69 ; GCN-LABEL: {{^}}mul_v2i16:
70 ; NOSDWA: v_lshrrev_b32_e32 v[[DST0:[0-9]+]], 16, v{{[0-9]+}}
71 ; NOSDWA: v_lshrrev_b32_e32 v[[DST1:[0-9]+]], 16, v{{[0-9]+}}
72 ; NOSDWA: v_mul_u32_u24_e32 v[[DST_MUL:[0-9]+]], v[[DST0]], v[[DST1]]
73 ; NOSDWA: v_lshlrev_b32_e32 v[[DST_SHL:[0-9]+]], 16, v[[DST_MUL]]
74 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[DST_SHL]]
75 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
77 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL_LO:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
78 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL_HI:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
79 ; VI: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL_LO]], v[[DST_MUL_HI]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
81 ; GFX9: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
83 define amdgpu_kernel void @mul_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %ina, <2 x i16> addrspace(1)* %inb) {
85 %a = load <2 x i16>, <2 x i16> addrspace(1)* %ina, align 4
86 %b = load <2 x i16>, <2 x i16> addrspace(1)* %inb, align 4
87 %mul = mul <2 x i16> %a, %b
88 store <2 x i16> %mul, <2 x i16> addrspace(1)* %out, align 4
92 ; GCN-LABEL: {{^}}mul_v4i16:
93 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
94 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
95 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
96 ; NOSDWA: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
97 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
98 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
100 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL0:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
101 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL1:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
102 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL2:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
103 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL3:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
104 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL2]], v[[DST_MUL3]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
105 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL0]], v[[DST_MUL1]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
107 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
108 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
110 define amdgpu_kernel void @mul_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %ina, <4 x i16> addrspace(1)* %inb) {
112 %a = load <4 x i16>, <4 x i16> addrspace(1)* %ina, align 4
113 %b = load <4 x i16>, <4 x i16> addrspace(1)* %inb, align 4
114 %mul = mul <4 x i16> %a, %b
115 store <4 x i16> %mul, <4 x i16> addrspace(1)* %out, align 4
119 ; GCN-LABEL: {{^}}mul_v8i16:
120 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
121 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
122 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
123 ; NOSDWA: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
124 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
125 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
127 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL0:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
128 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL1:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
129 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL2:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
130 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL3:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
131 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL4:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
132 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL5:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
133 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL6:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
134 ; VI-DAG: v_mul_u32_u24_sdwa v[[DST_MUL7:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
135 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL6]], v[[DST_MUL7]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
136 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL4]], v[[DST_MUL5]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
137 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL2]], v[[DST_MUL3]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
138 ; VI-DAG: v_or_b32_sdwa v{{[0-9]+}}, v[[DST_MUL0]], v[[DST_MUL1]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
140 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
141 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
142 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
143 ; GFX9-DAG: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
145 define amdgpu_kernel void @mul_v8i16(<8 x i16> addrspace(1)* %out, <8 x i16> addrspace(1)* %ina, <8 x i16> addrspace(1)* %inb) {
147 %a = load <8 x i16>, <8 x i16> addrspace(1)* %ina, align 4
148 %b = load <8 x i16>, <8 x i16> addrspace(1)* %inb, align 4
149 %mul = mul <8 x i16> %a, %b
150 store <8 x i16> %mul, <8 x i16> addrspace(1)* %out, align 4
154 ; GCN-LABEL: {{^}}mul_half:
155 ; NOSDWA: v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
156 ; NOSDWA-NOT: v_mul_f16_sdwa
157 ; SDWA: v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
158 ; SDWA-NOT: v_mul_f16_sdwa
160 define amdgpu_kernel void @mul_half(half addrspace(1)* %out, half addrspace(1)* %ina, half addrspace(1)* %inb) {
162 %a = load half, half addrspace(1)* %ina, align 4
163 %b = load half, half addrspace(1)* %inb, align 4
164 %mul = fmul half %a, %b
165 store half %mul, half addrspace(1)* %out, align 4
169 ; GCN-LABEL: {{^}}mul_v2half:
170 ; NOSDWA: v_lshrrev_b32_e32 v[[DST0:[0-9]+]], 16, v{{[0-9]+}}
171 ; NOSDWA: v_lshrrev_b32_e32 v[[DST1:[0-9]+]], 16, v{{[0-9]+}}
172 ; NOSDWA: v_mul_f16_e32 v[[DST_MUL:[0-9]+]], v[[DST0]], v[[DST1]]
173 ; NOSDWA: v_lshlrev_b32_e32 v[[DST_SHL:[0-9]+]], 16, v[[DST_MUL]]
174 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[DST_SHL]]
175 ; NOSDWA-NOT: v_mul_f16_sdwa
177 ; VI-DAG: v_mul_f16_sdwa v[[DST_MUL_HI:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
178 ; VI-DAG: v_mul_f16_e32 v[[DST_MUL_LO:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}}
179 ; VI: v_or_b32_e32 v{{[0-9]+}}, v[[DST_MUL_LO]], v[[DST_MUL_HI]]
181 ; GFX9: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
183 define amdgpu_kernel void @mul_v2half(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %ina, <2 x half> addrspace(1)* %inb) {
185 %a = load <2 x half>, <2 x half> addrspace(1)* %ina, align 4
186 %b = load <2 x half>, <2 x half> addrspace(1)* %inb, align 4
187 %mul = fmul <2 x half> %a, %b
188 store <2 x half> %mul, <2 x half> addrspace(1)* %out, align 4
192 ; GCN-LABEL: {{^}}mul_v4half:
193 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
194 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
195 ; NOSDWA: v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
196 ; NOSDWA: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
197 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
198 ; NOSDWA-NOT: v_mul_f16_sdwa
200 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
201 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
202 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
203 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
205 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
206 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
208 define amdgpu_kernel void @mul_v4half(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %ina, <4 x half> addrspace(1)* %inb) {
210 %a = load <4 x half>, <4 x half> addrspace(1)* %ina, align 4
211 %b = load <4 x half>, <4 x half> addrspace(1)* %inb, align 4
212 %mul = fmul <4 x half> %a, %b
213 store <4 x half> %mul, <4 x half> addrspace(1)* %out, align 4
217 ; GCN-LABEL: {{^}}mul_v8half:
218 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
219 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
220 ; NOSDWA: v_mul_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
221 ; NOSDWA: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
222 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
223 ; NOSDWA-NOT: v_mul_f16_sdwa
225 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
226 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
227 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
228 ; VI-DAG: v_mul_f16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
229 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
230 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
231 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
232 ; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
234 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
235 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
236 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
237 ; GFX9-DAG: v_pk_mul_f16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
239 define amdgpu_kernel void @mul_v8half(<8 x half> addrspace(1)* %out, <8 x half> addrspace(1)* %ina, <8 x half> addrspace(1)* %inb) {
241 %a = load <8 x half>, <8 x half> addrspace(1)* %ina, align 4
242 %b = load <8 x half>, <8 x half> addrspace(1)* %inb, align 4
243 %mul = fmul <8 x half> %a, %b
244 store <8 x half> %mul, <8 x half> addrspace(1)* %out, align 4
248 ; GCN-LABEL: {{^}}mul_i8:
249 ; NOSDWA: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
250 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
251 ; SDWA: v_mul_lo_i32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
252 ; SDWA-NOT: v_mul_u32_u24_sdwa
254 define amdgpu_kernel void @mul_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %ina, i8 addrspace(1)* %inb) {
256 %a = load i8, i8 addrspace(1)* %ina, align 4
257 %b = load i8, i8 addrspace(1)* %inb, align 4
259 store i8 %mul, i8 addrspace(1)* %out, align 4
263 ; GCN-LABEL: {{^}}mul_v2i8:
264 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
265 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
266 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
267 ; NOSDWA: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
268 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
269 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
271 ; VI: v_mul_u32_u24_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
273 ; GFX9-DAG: v_mul_lo_u16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
274 ; GFX9-DAG: v_mul_lo_u16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
275 ; GFX9: v_or_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
277 define amdgpu_kernel void @mul_v2i8(<2 x i8> addrspace(1)* %out, <2 x i8> addrspace(1)* %ina, <2 x i8> addrspace(1)* %inb) {
279 %a = load <2 x i8>, <2 x i8> addrspace(1)* %ina, align 4
280 %b = load <2 x i8>, <2 x i8> addrspace(1)* %inb, align 4
281 %mul = mul <2 x i8> %a, %b
282 store <2 x i8> %mul, <2 x i8> addrspace(1)* %out, align 4
286 ; GCN-LABEL: {{^}}mul_v4i8:
287 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
288 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
289 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
290 ; NOSDWA: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
291 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
292 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
294 ; VI-DAG: v_mul_u32_u24_sdwa
295 ; VI-DAG: v_mul_u32_u24_sdwa
296 ; VI-DAG: v_mul_u32_u24_sdwa
298 ; GFX9-DAG: v_mul_lo_u16_sdwa
299 ; GFX9-DAG: v_mul_lo_u16_sdwa
300 ; GFX9-DAG: v_mul_lo_u16_sdwa
302 define amdgpu_kernel void @mul_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> addrspace(1)* %ina, <4 x i8> addrspace(1)* %inb) {
304 %a = load <4 x i8>, <4 x i8> addrspace(1)* %ina, align 4
305 %b = load <4 x i8>, <4 x i8> addrspace(1)* %inb, align 4
306 %mul = mul <4 x i8> %a, %b
307 store <4 x i8> %mul, <4 x i8> addrspace(1)* %out, align 4
311 ; GCN-LABEL: {{^}}mul_v8i8:
312 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
313 ; NOSDWA: v_lshrrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
314 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
315 ; NOSDWA: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
316 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
317 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
319 ; VI-DAG: v_mul_u32_u24_sdwa
320 ; VI-DAG: v_mul_u32_u24_sdwa
321 ; VI-DAG: v_mul_u32_u24_sdwa
322 ; VI-DAG: v_mul_u32_u24_sdwa
323 ; VI-DAG: v_mul_u32_u24_sdwa
324 ; VI-DAG: v_mul_u32_u24_sdwa
326 ; GFX9-DAG: v_mul_lo_u16_sdwa
327 ; GFX9-DAG: v_mul_lo_u16_sdwa
328 ; GFX9-DAG: v_mul_lo_u16_sdwa
329 ; GFX9-DAG: v_mul_lo_u16_sdwa
330 ; GFX9-DAG: v_mul_lo_u16_sdwa
331 ; GFX9-DAG: v_mul_lo_u16_sdwa
333 define amdgpu_kernel void @mul_v8i8(<8 x i8> addrspace(1)* %out, <8 x i8> addrspace(1)* %ina, <8 x i8> addrspace(1)* %inb) {
335 %a = load <8 x i8>, <8 x i8> addrspace(1)* %ina, align 4
336 %b = load <8 x i8>, <8 x i8> addrspace(1)* %inb, align 4
337 %mul = mul <8 x i8> %a, %b
338 store <8 x i8> %mul, <8 x i8> addrspace(1)* %out, align 4
342 ; GCN-LABEL: {{^}}sitofp_v2i16_to_v2f16:
343 ; NOSDWA-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 0, 16
344 ; NOSDWA-DAG: v_ashrrev_i32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
345 ; NOSDWA-DAG: v_cvt_f32_i32_e32 v{{[0-9]+}}, v{{[0-9]+}}
346 ; NOSDWA-DAG: v_cvt_f32_i32_e32 v{{[0-9]+}}, v{{[0-9]+}}
347 ; NOSDWA-NOT: v_cvt_f32_i32_sdwa
349 ; SDWA-DAG: v_cvt_f32_i32_sdwa v{{[0-9]+}}, sext(v{{[0-9]+}}) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0
350 ; SDWA-DAG: v_cvt_f32_i32_sdwa v{{[0-9]+}}, sext(v{{[0-9]+}}) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
352 define amdgpu_kernel void @sitofp_v2i16_to_v2f16(
353 <2 x half> addrspace(1)* %r,
354 <2 x i16> addrspace(1)* %a) {
356 %a.val = load <2 x i16>, <2 x i16> addrspace(1)* %a
357 %r.val = sitofp <2 x i16> %a.val to <2 x half>
358 store <2 x half> %r.val, <2 x half> addrspace(1)* %r
363 ; GCN-LABEL: {{^}}mac_v2half:
364 ; NOSDWA: v_lshrrev_b32_e32 v[[DST0:[0-9]+]], 16, v{{[0-9]+}}
365 ; NOSDWA: v_lshrrev_b32_e32 v[[DST1:[0-9]+]], 16, v{{[0-9]+}}
366 ; NOSDWA: v_mac_f16_e32 v[[DST_MAC:[0-9]+]], v[[DST0]], v[[DST1]]
367 ; NOSDWA: v_lshlrev_b32_e32 v[[DST_SHL:[0-9]+]], 16, v[[DST_MAC]]
368 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[DST_SHL]]
369 ; NOSDWA-NOT: v_mac_f16_sdwa
371 ; VI: v_mac_f16_sdwa v[[DST_MAC:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
372 ; VI: v_lshlrev_b32_e32 v[[DST_SHL:[0-9]+]], 16, v[[DST_MAC]]
374 ; GFX9: v_pk_mul_f16 v[[DST_MUL:[0-9]+]], v{{[0-9]+}}, v[[SRC:[0-9]+]]
375 ; GFX9: v_pk_add_f16 v{{[0-9]+}}, v[[DST_MUL]], v[[SRC]]
377 define amdgpu_kernel void @mac_v2half(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %ina, <2 x half> addrspace(1)* %inb) {
379 %a = load <2 x half>, <2 x half> addrspace(1)* %ina, align 4
380 %b = load <2 x half>, <2 x half> addrspace(1)* %inb, align 4
381 %mul = fmul <2 x half> %a, %b
382 %mac = fadd <2 x half> %mul, %b
383 store <2 x half> %mac, <2 x half> addrspace(1)* %out, align 4
387 ; GCN-LABEL: {{^}}immediate_mul_v2i16:
388 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
389 ; VI-DAG: v_mov_b32_e32 v[[M321:[0-9]+]], 0x141
390 ; VI-DAG: v_mov_b32_e32 v[[M123:[0-9]+]], 0x7b
391 ; VI-DAG: v_mul_u32_u24_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v[[M123]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
392 ; VI-DAG: v_mul_u32_u24_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v[[M321]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
394 ; GFX9: s_mov_b32 s[[IMM:[0-9]+]], 0x141007b
395 ; GFX9: v_pk_mul_lo_u16 v{{[0-9]+}}, v{{[0-9]+}}, s[[IMM]]
397 define amdgpu_kernel void @immediate_mul_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) {
399 %a = load <2 x i16>, <2 x i16> addrspace(1)* %in, align 4
400 %mul = mul <2 x i16> %a, <i16 123, i16 321>
401 store <2 x i16> %mul, <2 x i16> addrspace(1)* %out, align 4
405 ; Double use of same src - should not convert it
406 ; GCN-LABEL: {{^}}mulmul_v2i16:
407 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
408 ; NOSDWA: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
409 ; NOSDWA: v_mul_u32_u24_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
410 ; NOSDWA: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
411 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
412 ; NOSDWA-NOT: v_mul_u32_u24_sdwa
414 ; VI: v_mul_u32_u24_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
416 ; GFX9: v_pk_mul_lo_u16 v[[DST1:[0-9]+]], v{{[0-9]+}}, v{{[0-9]+}}
417 ; GFX9: v_pk_mul_lo_u16 v{{[0-9]+}}, v[[DST1]], v{{[0-9]+}}
419 define amdgpu_kernel void @mulmul_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %ina, <2 x i16> addrspace(1)* %inb) {
421 %a = load <2 x i16>, <2 x i16> addrspace(1)* %ina, align 4
422 %b = load <2 x i16>, <2 x i16> addrspace(1)* %inb, align 4
423 %mul = mul <2 x i16> %a, %b
424 %mul2 = mul <2 x i16> %mul, %b
425 store <2 x i16> %mul2, <2 x i16> addrspace(1)* %out, align 4
429 ; GCN-LABEL: {{^}}add_bb_v2i16:
430 ; NOSDWA-NOT: v_add_{{(_co)?}}_u32_sdwa
432 ; VI: v_add_u32_sdwa v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
434 ; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
436 define amdgpu_kernel void @add_bb_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %ina, <2 x i16> addrspace(1)* %inb) {
438 %a = load <2 x i16>, <2 x i16> addrspace(1)* %ina, align 4
439 %b = load <2 x i16>, <2 x i16> addrspace(1)* %inb, align 4
442 %add = add <2 x i16> %a, %b
443 br label %store_label
445 store <2 x i16> %add, <2 x i16> addrspace(1)* %out, align 4
450 ; Check that "pulling out" SDWA operands works correctly.
451 ; GCN-LABEL: {{^}}pulled_out_test:
452 ; NOSDWA-DAG: v_and_b32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
453 ; NOSDWA-DAG: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
454 ; NOSDWA-DAG: v_and_b32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
455 ; NOSDWA-DAG: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
456 ; NOSDWA: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
457 ; NOSDWA-NOT: v_and_b32_sdwa
458 ; NOSDWA-NOT: v_or_b32_sdwa
460 ; VI-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
461 ; GFX9-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
462 ; SDWA-DAG: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
463 ; VI-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
464 ; GFX9-DAG: v_and_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
465 ; SDWA-DAG: v_lshlrev_b16_e32 v{{[0-9]+}}, 8, v{{[0-9]+}}
466 ; SDWA: v_or_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
468 define amdgpu_kernel void @pulled_out_test(<8 x i8> addrspace(1)* %sourceA, <8 x i8> addrspace(1)* %destValues) {
470 %idxprom = ashr exact i64 15, 32
471 %arrayidx = getelementptr inbounds <8 x i8>, <8 x i8> addrspace(1)* %sourceA, i64 %idxprom
472 %tmp = load <8 x i8>, <8 x i8> addrspace(1)* %arrayidx, align 8
474 %tmp1 = extractelement <8 x i8> %tmp, i32 0
475 %tmp2 = extractelement <8 x i8> %tmp, i32 1
476 %tmp3 = extractelement <8 x i8> %tmp, i32 2
477 %tmp4 = extractelement <8 x i8> %tmp, i32 3
478 %tmp5 = extractelement <8 x i8> %tmp, i32 4
479 %tmp6 = extractelement <8 x i8> %tmp, i32 5
480 %tmp7 = extractelement <8 x i8> %tmp, i32 6
481 %tmp8 = extractelement <8 x i8> %tmp, i32 7
483 %tmp9 = insertelement <2 x i8> undef, i8 %tmp1, i32 0
484 %tmp10 = insertelement <2 x i8> %tmp9, i8 %tmp2, i32 1
485 %tmp11 = insertelement <2 x i8> undef, i8 %tmp3, i32 0
486 %tmp12 = insertelement <2 x i8> %tmp11, i8 %tmp4, i32 1
487 %tmp13 = insertelement <2 x i8> undef, i8 %tmp5, i32 0
488 %tmp14 = insertelement <2 x i8> %tmp13, i8 %tmp6, i32 1
489 %tmp15 = insertelement <2 x i8> undef, i8 %tmp7, i32 0
490 %tmp16 = insertelement <2 x i8> %tmp15, i8 %tmp8, i32 1
492 %tmp17 = shufflevector <2 x i8> %tmp10, <2 x i8> %tmp12, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
493 %tmp18 = shufflevector <2 x i8> %tmp14, <2 x i8> %tmp16, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
494 %tmp19 = shufflevector <4 x i8> %tmp17, <4 x i8> %tmp18, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
496 %arrayidx5 = getelementptr inbounds <8 x i8>, <8 x i8> addrspace(1)* %destValues, i64 %idxprom
497 store <8 x i8> %tmp19, <8 x i8> addrspace(1)* %arrayidx5, align 8
501 ; GCN-LABEL: {{^}}sdwa_crash_inlineasm_def:
502 ; GCN: s_mov_b32 s{{[0-9]+}}, 0xffff
503 ; GCN: v_and_b32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
504 ; GCN: v_or_b32_e32 v{{[0-9]+}}, 0x10000,
505 define amdgpu_kernel void @sdwa_crash_inlineasm_def() #0 {
509 bb1: ; preds = %bb11, %bb
510 %tmp = phi <2 x i32> [ %tmp12, %bb11 ], [ undef, %bb ]
511 br i1 true, label %bb2, label %bb11
514 %tmp3 = call i32 asm "v_and_b32_e32 $0, $1, $2", "=v,s,v"(i32 65535, i32 undef) #1
515 %tmp5 = or i32 %tmp3, 65536
516 %tmp6 = insertelement <2 x i32> %tmp, i32 %tmp5, i64 0
519 bb11: ; preds = %bb10, %bb2
520 %tmp12 = phi <2 x i32> [ %tmp6, %bb2 ], [ %tmp, %bb1 ]