Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / 3ead7d73893dd6544651dc5cf6989abd7fe4f4bb / . / llvm / test / CodeGen / AMDGPU / build-vector-packed-partial-undef.ll
blob: b78dced5bcf2acb38f7b0f1436176b12bc34f87a [file] [log] [blame]
Matt Arsenault3ead7d72018-08-12 08:42:46 +0000[diff] [blame^]1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GFX9 %s
3; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GFX8 %s
4
5define void @undef_lo_v2i16(i16 %arg0) {
6; GFX9-LABEL: undef_lo_v2i16:
7; GFX9: ; %bb.0:
8; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
9; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
10; GFX9-NEXT: ;;#ASMSTART
11; GFX9-NEXT: ; use v0
12; GFX9-NEXT: ;;#ASMEND
13; GFX9-NEXT: s_setpc_b64 s[30:31]
14;
15; GFX8-LABEL: undef_lo_v2i16:
16; GFX8: ; %bb.0:
17; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
18; GFX8-NEXT: v_lshlrev_b32_e32 v0, 16, v0
19; GFX8-NEXT: ;;#ASMSTART
20; GFX8-NEXT: ; use v0
21; GFX8-NEXT: ;;#ASMEND
22; GFX8-NEXT: s_setpc_b64 s[30:31]
23 %undef.lo = insertelement <2 x i16> undef, i16 %arg0, i32 1
24 call void asm sideeffect "; use $0", "v"(<2 x i16> %undef.lo);
25 ret void
26}
27
28define void @undef_lo_v2f16(half %arg0) {
29; GFX9-LABEL: undef_lo_v2f16:
30; GFX9: ; %bb.0:
31; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
32; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
33; GFX9-NEXT: ;;#ASMSTART
34; GFX9-NEXT: ; use v0
35; GFX9-NEXT: ;;#ASMEND
36; GFX9-NEXT: s_setpc_b64 s[30:31]
37;
38; GFX8-LABEL: undef_lo_v2f16:
39; GFX8: ; %bb.0:
40; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
41; GFX8-NEXT: v_lshlrev_b32_e32 v0, 16, v0
42; GFX8-NEXT: ;;#ASMSTART
43; GFX8-NEXT: ; use v0
44; GFX8-NEXT: ;;#ASMEND
45; GFX8-NEXT: s_setpc_b64 s[30:31]
46 %undef.lo = insertelement <2 x half> undef, half %arg0, i32 1
47 call void asm sideeffect "; use $0", "v"(<2 x half> %undef.lo);
48 ret void
49}
50
51define void @undef_lo_op_v2f16(half %arg0) {
52; GFX9-LABEL: undef_lo_op_v2f16:
53; GFX9: ; %bb.0:
54; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
55; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
56; GFX9-NEXT: v_pk_add_f16 v0, v0, 1.0 op_sel_hi:[1,0]
57; GFX9-NEXT: ;;#ASMSTART
58; GFX9-NEXT: ; use v0
59; GFX9-NEXT: ;;#ASMEND
60; GFX9-NEXT: s_setpc_b64 s[30:31]
61;
62; GFX8-LABEL: undef_lo_op_v2f16:
63; GFX8: ; %bb.0:
64; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
65; GFX8-NEXT: v_mov_b32_e32 v1, 0x3c00
66; GFX8-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
67; GFX8-NEXT: v_or_b32_e32 v0, 0x7e00, v0
68; GFX8-NEXT: ;;#ASMSTART
69; GFX8-NEXT: ; use v0
70; GFX8-NEXT: ;;#ASMEND
71; GFX8-NEXT: s_setpc_b64 s[30:31]
72 %undef.lo = insertelement <2 x half> undef, half %arg0, i32 1
73 %op = fadd <2 x half> %undef.lo, <half 1.0, half 1.0>
74 call void asm sideeffect "; use $0", "v"(<2 x half> %op);
75 ret void
76}
77
78define void @undef_lo_op_v2i16(i16 %arg0) {
79; GFX9-LABEL: undef_lo_op_v2i16:
80; GFX9: ; %bb.0:
81; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
82; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
83; GFX9-NEXT: s_movk_i32 s6, 0x63
84; GFX9-NEXT: v_pk_add_u16 v0, v0, s6 op_sel_hi:[1,0]
85; GFX9-NEXT: ;;#ASMSTART
86; GFX9-NEXT: ; use v0
87; GFX9-NEXT: ;;#ASMEND
88; GFX9-NEXT: s_setpc_b64 s[30:31]
89;
90; GFX8-LABEL: undef_lo_op_v2i16:
91; GFX8: ; %bb.0:
92; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
93; GFX8-NEXT: v_mov_b32_e32 v1, 0x63
94; GFX8-NEXT: v_add_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
95; GFX8-NEXT: ;;#ASMSTART
96; GFX8-NEXT: ; use v0
97; GFX8-NEXT: ;;#ASMEND
98; GFX8-NEXT: s_setpc_b64 s[30:31]
99 %undef.lo = insertelement <2 x i16> undef, i16 %arg0, i32 1
100 %op = add <2 x i16> %undef.lo, <i16 99, i16 99>
101 call void asm sideeffect "; use $0", "v"(<2 x i16> %op);
102 ret void
103}
104
105define void @undef_lo3_v4i16(i16 %arg0) {
106; GFX9-LABEL: undef_lo3_v4i16:
107; GFX9: ; %bb.0:
108; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
109; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
110; GFX9-NEXT: ;;#ASMSTART
111; GFX9-NEXT: ; use v[0:1]
112; GFX9-NEXT: ;;#ASMEND
113; GFX9-NEXT: s_setpc_b64 s[30:31]
114;
115; GFX8-LABEL: undef_lo3_v4i16:
116; GFX8: ; %bb.0:
117; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
118; GFX8-NEXT: v_lshlrev_b32_e32 v0, 16, v0
119; GFX8-NEXT: ;;#ASMSTART
120; GFX8-NEXT: ; use v[0:1]
121; GFX8-NEXT: ;;#ASMEND
122; GFX8-NEXT: s_setpc_b64 s[30:31]
123 %undef.lo = insertelement <4 x i16> undef, i16 %arg0, i32 1
124 call void asm sideeffect "; use $0", "v"(<4 x i16> %undef.lo);
125 ret void
126}
127
128define void @undef_lo3_v4f16(half %arg0) {
129; GFX9-LABEL: undef_lo3_v4f16:
130; GFX9: ; %bb.0:
131; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
132; GFX9-NEXT: v_lshlrev_b32_e32 v0, 16, v0
133; GFX9-NEXT: ;;#ASMSTART
134; GFX9-NEXT: ; use v[0:1]
135; GFX9-NEXT: ;;#ASMEND
136; GFX9-NEXT: s_setpc_b64 s[30:31]
137;
138; GFX8-LABEL: undef_lo3_v4f16:
139; GFX8: ; %bb.0:
140; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
141; GFX8-NEXT: v_lshlrev_b32_e32 v0, 16, v0
142; GFX8-NEXT: ;;#ASMSTART
143; GFX8-NEXT: ; use v[0:1]
144; GFX8-NEXT: ;;#ASMEND
145; GFX8-NEXT: s_setpc_b64 s[30:31]
146 %undef.lo = insertelement <4 x half> undef, half %arg0, i32 1
147 call void asm sideeffect "; use $0", "v"(<4 x half> %undef.lo);
148 ret void
149}
150
151define void @undef_lo2_v4i16(<2 x i16> %arg0) {
152; GFX9-LABEL: undef_lo2_v4i16:
153; GFX9: ; %bb.0:
154; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
155; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v0
156; GFX9-NEXT: v_mov_b32_e32 v2, 0xffff0000
157; GFX9-NEXT: v_and_or_b32 v0, v0, v2, v1
158; GFX9-NEXT: ;;#ASMSTART
159; GFX9-NEXT: ; use v[0:1]
160; GFX9-NEXT: ;;#ASMEND
161; GFX9-NEXT: s_setpc_b64 s[30:31]
162;
163; GFX8-LABEL: undef_lo2_v4i16:
164; GFX8: ; %bb.0:
165; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
166; GFX8-NEXT: v_lshrrev_b32_e32 v0, 16, v0
167; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v0
168; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
169; GFX8-NEXT: ;;#ASMSTART
170; GFX8-NEXT: ; use v[0:1]
171; GFX8-NEXT: ;;#ASMEND
172; GFX8-NEXT: s_setpc_b64 s[30:31]
173 %undef.lo = shufflevector <2 x i16> %arg0, <2 x i16> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 3>
174 call void asm sideeffect "; use $0", "v"(<4 x i16> %undef.lo);
175 ret void
176}
177
178define void @undef_lo2_v4f16(<2 x half> %arg0) {
179; GFX9-LABEL: undef_lo2_v4f16:
180; GFX9: ; %bb.0:
181; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
182; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v0
183; GFX9-NEXT: v_and_b32_e32 v1, 0xffff, v0
184; GFX9-NEXT: v_lshl_or_b32 v0, v0, 16, v1
185; GFX9-NEXT: ;;#ASMSTART
186; GFX9-NEXT: ; use v[0:1]
187; GFX9-NEXT: ;;#ASMEND
188; GFX9-NEXT: s_setpc_b64 s[30:31]
189;
190; GFX8-LABEL: undef_lo2_v4f16:
191; GFX8: ; %bb.0:
192; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
193; GFX8-NEXT: v_lshrrev_b32_e32 v0, 16, v0
194; GFX8-NEXT: v_lshlrev_b32_e32 v1, 16, v0
195; GFX8-NEXT: v_or_b32_e32 v0, v0, v1
196; GFX8-NEXT: ;;#ASMSTART
197; GFX8-NEXT: ; use v[0:1]
198; GFX8-NEXT: ;;#ASMEND
199; GFX8-NEXT: s_setpc_b64 s[30:31]
200 %undef.lo = shufflevector <2 x half> %arg0, <2 x half> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 3>
201 call void asm sideeffect "; use $0", "v"(<4 x half> %undef.lo);
202 ret void
203}
204
205define void @undef_hi_v2i16(i16 %arg0) {
206; GFX9-LABEL: undef_hi_v2i16:
207; GFX9: ; %bb.0:
208; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
209; GFX9-NEXT: ;;#ASMSTART
210; GFX9-NEXT: ; use v0
211; GFX9-NEXT: ;;#ASMEND
212; GFX9-NEXT: s_setpc_b64 s[30:31]
213;
214; GFX8-LABEL: undef_hi_v2i16:
215; GFX8: ; %bb.0:
216; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
217; GFX8-NEXT: ;;#ASMSTART
218; GFX8-NEXT: ; use v0
219; GFX8-NEXT: ;;#ASMEND
220; GFX8-NEXT: s_setpc_b64 s[30:31]
221 %undef.hi = insertelement <2 x i16> undef, i16 %arg0, i32 0
222 call void asm sideeffect "; use $0", "v"(<2 x i16> %undef.hi);
223 ret void
224}
225
226define void @undef_hi_v2f16(half %arg0) {
227; GFX9-LABEL: undef_hi_v2f16:
228; GFX9: ; %bb.0:
229; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
230; GFX9-NEXT: ;;#ASMSTART
231; GFX9-NEXT: ; use v0
232; GFX9-NEXT: ;;#ASMEND
233; GFX9-NEXT: s_setpc_b64 s[30:31]
234;
235; GFX8-LABEL: undef_hi_v2f16:
236; GFX8: ; %bb.0:
237; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
238; GFX8-NEXT: ;;#ASMSTART
239; GFX8-NEXT: ; use v0
240; GFX8-NEXT: ;;#ASMEND
241; GFX8-NEXT: s_setpc_b64 s[30:31]
242 %undef.hi = insertelement <2 x half> undef, half %arg0, i32 0
243 call void asm sideeffect "; use $0", "v"(<2 x half> %undef.hi);
244 ret void
245}
246
247define void @undef_hi_op_v2f16(half %arg0) {
248; GFX9-LABEL: undef_hi_op_v2f16:
249; GFX9: ; %bb.0:
250; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
251; GFX9-NEXT: v_pk_add_f16 v0, v0, 1.0 op_sel_hi:[1,0]
252; GFX9-NEXT: ;;#ASMSTART
253; GFX9-NEXT: ; use v0
254; GFX9-NEXT: ;;#ASMEND
255; GFX9-NEXT: s_setpc_b64 s[30:31]
256;
257; GFX8-LABEL: undef_hi_op_v2f16:
258; GFX8: ; %bb.0:
259; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
260; GFX8-NEXT: v_add_f16_e32 v0, 1.0, v0
261; GFX8-NEXT: v_or_b32_e32 v0, 0x7e000000, v0
262; GFX8-NEXT: ;;#ASMSTART
263; GFX8-NEXT: ; use v0
264; GFX8-NEXT: ;;#ASMEND
265; GFX8-NEXT: s_setpc_b64 s[30:31]
266 %undef.hi = insertelement <2 x half> undef, half %arg0, i32 0
267 %op = fadd <2 x half> %undef.hi, <half 1.0, half 1.0>
268 call void asm sideeffect "; use $0", "v"(<2 x half> %op);
269 ret void
270}
271
272define void @undef_hi_op_v2i16(i16 %arg0) {
273; GFX9-LABEL: undef_hi_op_v2i16:
274; GFX9: ; %bb.0:
275; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
276; GFX9-NEXT: s_movk_i32 s6, 0x63
277; GFX9-NEXT: v_pk_add_u16 v0, v0, s6 op_sel_hi:[1,0]
278; GFX9-NEXT: ;;#ASMSTART
279; GFX9-NEXT: ; use v0
280; GFX9-NEXT: ;;#ASMEND
281; GFX9-NEXT: s_setpc_b64 s[30:31]
282;
283; GFX8-LABEL: undef_hi_op_v2i16:
284; GFX8: ; %bb.0:
285; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
286; GFX8-NEXT: v_add_u16_e32 v0, 0x63, v0
287; GFX8-NEXT: ;;#ASMSTART
288; GFX8-NEXT: ; use v0
289; GFX8-NEXT: ;;#ASMEND
290; GFX8-NEXT: s_setpc_b64 s[30:31]
291 %undef.hi = insertelement <2 x i16> undef, i16 %arg0, i32 0
292 %op = add <2 x i16> %undef.hi, <i16 99, i16 99>
293 call void asm sideeffect "; use $0", "v"(<2 x i16> %op);
294 ret void
295}
296
297define void @undef_hi3_v4i16(i16 %arg0) {
298; GFX9-LABEL: undef_hi3_v4i16:
299; GFX9: ; %bb.0:
300; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
301; GFX9-NEXT: ;;#ASMSTART
302; GFX9-NEXT: ; use v[0:1]
303; GFX9-NEXT: ;;#ASMEND
304; GFX9-NEXT: s_setpc_b64 s[30:31]
305;
306; GFX8-LABEL: undef_hi3_v4i16:
307; GFX8: ; %bb.0:
308; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
309; GFX8-NEXT: ;;#ASMSTART
310; GFX8-NEXT: ; use v[0:1]
311; GFX8-NEXT: ;;#ASMEND
312; GFX8-NEXT: s_setpc_b64 s[30:31]
313 %undef.hi = insertelement <4 x i16> undef, i16 %arg0, i32 0
314 call void asm sideeffect "; use $0", "v"(<4 x i16> %undef.hi);
315 ret void
316}
317
318define void @undef_hi3_v4f16(half %arg0) {
319; GFX9-LABEL: undef_hi3_v4f16:
320; GFX9: ; %bb.0:
321; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
322; GFX9-NEXT: ;;#ASMSTART
323; GFX9-NEXT: ; use v[0:1]
324; GFX9-NEXT: ;;#ASMEND
325; GFX9-NEXT: s_setpc_b64 s[30:31]
326;
327; GFX8-LABEL: undef_hi3_v4f16:
328; GFX8: ; %bb.0:
329; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
330; GFX8-NEXT: ;;#ASMSTART
331; GFX8-NEXT: ; use v[0:1]
332; GFX8-NEXT: ;;#ASMEND
333; GFX8-NEXT: s_setpc_b64 s[30:31]
334 %undef.hi = insertelement <4 x half> undef, half %arg0, i32 0
335 call void asm sideeffect "; use $0", "v"(<4 x half> %undef.hi);
336 ret void
337}
338
339define void @undef_hi2_v4i16(<2 x i16> %arg0) {
340; GFX9-LABEL: undef_hi2_v4i16:
341; GFX9: ; %bb.0:
342; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
343; GFX9-NEXT: ;;#ASMSTART
344; GFX9-NEXT: ; use v[0:1]
345; GFX9-NEXT: ;;#ASMEND
346; GFX9-NEXT: s_setpc_b64 s[30:31]
347;
348; GFX8-LABEL: undef_hi2_v4i16:
349; GFX8: ; %bb.0:
350; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
351; GFX8-NEXT: ;;#ASMSTART
352; GFX8-NEXT: ; use v[0:1]
353; GFX8-NEXT: ;;#ASMEND
354; GFX8-NEXT: s_setpc_b64 s[30:31]
355 %undef.hi = shufflevector <2 x i16> %arg0, <2 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
356 call void asm sideeffect "; use $0", "v"(<4 x i16> %undef.hi);
357 ret void
358}
359
360define void @undef_hi2_v4f16(<2 x half> %arg0) {
361; GFX9-LABEL: undef_hi2_v4f16:
362; GFX9: ; %bb.0:
363; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
364; GFX9-NEXT: ;;#ASMSTART
365; GFX9-NEXT: ; use v[0:1]
366; GFX9-NEXT: ;;#ASMEND
367; GFX9-NEXT: s_setpc_b64 s[30:31]
368;
369; GFX8-LABEL: undef_hi2_v4f16:
370; GFX8: ; %bb.0:
371; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
372; GFX8-NEXT: ;;#ASMSTART
373; GFX8-NEXT: ; use v[0:1]
374; GFX8-NEXT: ;;#ASMEND
375; GFX8-NEXT: s_setpc_b64 s[30:31]
376 %undef.hi = shufflevector <2 x half> %arg0, <2 x half> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
377 call void asm sideeffect "; use $0", "v"(<4 x half> %undef.hi);
378 ret void
379}
380