xref: /llvm-project/llvm/test/CodeGen/Thumb2/mve-ucmp.ll (revision 4e3781607cd12eb337298ee6d16ebecde4ce5741) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp %s -o - | FileCheck %s
3
4define arm_aapcs_vfpcc <8 x i8> @u_v8i8(<8 x i8> %a, <8 x i8> %b) {
5; CHECK-LABEL: u_v8i8:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    vmovlb.u8 q1, q1
8; CHECK-NEXT:    vmovlb.u8 q0, q0
9; CHECK-NEXT:    vmov.i32 q2, #0x0
10; CHECK-NEXT:    vmov.i16 q3, #0x1
11; CHECK-NEXT:    vcmp.u16 hi, q0, q1
12; CHECK-NEXT:    vpsel q2, q3, q2
13; CHECK-NEXT:    vmov.i8 q3, #0xff
14; CHECK-NEXT:    vcmp.u16 hi, q1, q0
15; CHECK-NEXT:    vpsel q0, q3, q2
16; CHECK-NEXT:    bx lr
17entry:
18  %c = call <8 x i8> @llvm.ucmp(<8 x i8> %a, <8 x i8> %b)
19  ret <8 x i8> %c
20}
21
22define arm_aapcs_vfpcc <16 x i8> @u_v16i8(<16 x i8> %a, <16 x i8> %b) {
23; CHECK-LABEL: u_v16i8:
24; CHECK:       @ %bb.0: @ %entry
25; CHECK-NEXT:    vmov.i32 q2, #0x0
26; CHECK-NEXT:    vmov.i8 q3, #0x1
27; CHECK-NEXT:    vcmp.u8 hi, q0, q1
28; CHECK-NEXT:    vpsel q2, q3, q2
29; CHECK-NEXT:    vmov.i8 q3, #0xff
30; CHECK-NEXT:    vcmp.u8 hi, q1, q0
31; CHECK-NEXT:    vpsel q0, q3, q2
32; CHECK-NEXT:    bx lr
33entry:
34  %c = call <16 x i8> @llvm.ucmp(<16 x i8> %a, <16 x i8> %b)
35  ret <16 x i8> %c
36}
37
38define arm_aapcs_vfpcc <4 x i16> @u_v4i16(<4 x i16> %a, <4 x i16> %b) {
39; CHECK-LABEL: u_v4i16:
40; CHECK:       @ %bb.0: @ %entry
41; CHECK-NEXT:    vmovlb.u16 q1, q1
42; CHECK-NEXT:    vmovlb.u16 q0, q0
43; CHECK-NEXT:    vmov.i32 q2, #0x0
44; CHECK-NEXT:    vmov.i32 q3, #0x1
45; CHECK-NEXT:    vcmp.u32 hi, q0, q1
46; CHECK-NEXT:    vpsel q2, q3, q2
47; CHECK-NEXT:    vmov.i8 q3, #0xff
48; CHECK-NEXT:    vcmp.u32 hi, q1, q0
49; CHECK-NEXT:    vpsel q0, q3, q2
50; CHECK-NEXT:    bx lr
51entry:
52  %c = call <4 x i16> @llvm.ucmp(<4 x i16> %a, <4 x i16> %b)
53  ret <4 x i16> %c
54}
55
56define arm_aapcs_vfpcc <8 x i16> @u_v8i16(<8 x i16> %a, <8 x i16> %b) {
57; CHECK-LABEL: u_v8i16:
58; CHECK:       @ %bb.0: @ %entry
59; CHECK-NEXT:    vmov.i32 q2, #0x0
60; CHECK-NEXT:    vmov.i16 q3, #0x1
61; CHECK-NEXT:    vcmp.u16 hi, q0, q1
62; CHECK-NEXT:    vpsel q2, q3, q2
63; CHECK-NEXT:    vmov.i8 q3, #0xff
64; CHECK-NEXT:    vcmp.u16 hi, q1, q0
65; CHECK-NEXT:    vpsel q0, q3, q2
66; CHECK-NEXT:    bx lr
67entry:
68  %c = call <8 x i16> @llvm.ucmp(<8 x i16> %a, <8 x i16> %b)
69  ret <8 x i16> %c
70}
71
72define arm_aapcs_vfpcc <16 x i16> @u_v16i16(<16 x i16> %a, <16 x i16> %b) {
73; CHECK-LABEL: u_v16i16:
74; CHECK:       @ %bb.0: @ %entry
75; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
76; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
77; CHECK-NEXT:    vmov.i32 q4, #0x0
78; CHECK-NEXT:    vmov.i16 q5, #0x1
79; CHECK-NEXT:    vcmp.u16 hi, q0, q2
80; CHECK-NEXT:    vmov.i8 q7, #0xff
81; CHECK-NEXT:    vpsel q6, q5, q4
82; CHECK-NEXT:    vcmp.u16 hi, q2, q0
83; CHECK-NEXT:    vpsel q0, q7, q6
84; CHECK-NEXT:    vcmp.u16 hi, q1, q3
85; CHECK-NEXT:    vpsel q2, q5, q4
86; CHECK-NEXT:    vcmp.u16 hi, q3, q1
87; CHECK-NEXT:    vpsel q1, q7, q2
88; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
89; CHECK-NEXT:    bx lr
90entry:
91  %c = call <16 x i16> @llvm.ucmp(<16 x i16> %a, <16 x i16> %b)
92  ret <16 x i16> %c
93}
94
95define arm_aapcs_vfpcc <2 x i32> @u_v2i32(<2 x i32> %a, <2 x i32> %b) {
96; CHECK-LABEL: u_v2i32:
97; CHECK:       @ %bb.0: @ %entry
98; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, lr}
99; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, lr}
100; CHECK-NEXT:    vmov.i64 q2, #0xffffffff
101; CHECK-NEXT:    movs r4, #0
102; CHECK-NEXT:    vand q1, q1, q2
103; CHECK-NEXT:    vand q0, q0, q2
104; CHECK-NEXT:    vmov lr, r12, d0
105; CHECK-NEXT:    movs r0, #0
106; CHECK-NEXT:    vmov r3, r8, d2
107; CHECK-NEXT:    vmov r6, r7, d3
108; CHECK-NEXT:    vmov.i32 q1, #0x0
109; CHECK-NEXT:    subs.w r1, r3, lr
110; CHECK-NEXT:    sbcs.w r1, r8, r12
111; CHECK-NEXT:    csetm r1, lo
112; CHECK-NEXT:    bfi r4, r1, #0, #8
113; CHECK-NEXT:    vmov r1, r5, d1
114; CHECK-NEXT:    subs r2, r6, r1
115; CHECK-NEXT:    sbcs.w r2, r7, r5
116; CHECK-NEXT:    csetm r2, lo
117; CHECK-NEXT:    bfi r4, r2, #8, #8
118; CHECK-NEXT:    adr r2, .LCPI5_0
119; CHECK-NEXT:    vldrw.u32 q0, [r2]
120; CHECK-NEXT:    subs.w r2, lr, r3
121; CHECK-NEXT:    sbcs.w r2, r12, r8
122; CHECK-NEXT:    vmsr p0, r4
123; CHECK-NEXT:    csetm r2, lo
124; CHECK-NEXT:    subs r1, r1, r6
125; CHECK-NEXT:    sbcs.w r1, r5, r7
126; CHECK-NEXT:    bfi r0, r2, #0, #8
127; CHECK-NEXT:    csetm r1, lo
128; CHECK-NEXT:    vpsel q0, q0, q1
129; CHECK-NEXT:    bfi r0, r1, #8, #8
130; CHECK-NEXT:    vmov.i8 q1, #0xff
131; CHECK-NEXT:    vmsr p0, r0
132; CHECK-NEXT:    vpsel q0, q1, q0
133; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, pc}
134; CHECK-NEXT:    .p2align 4
135; CHECK-NEXT:  @ %bb.1:
136; CHECK-NEXT:  .LCPI5_0:
137; CHECK-NEXT:    .long 1 @ 0x1
138; CHECK-NEXT:    .long 0 @ 0x0
139; CHECK-NEXT:    .long 1 @ 0x1
140; CHECK-NEXT:    .long 0 @ 0x0
141entry:
142  %c = call <2 x i32> @llvm.ucmp(<2 x i32> %a, <2 x i32> %b)
143  ret <2 x i32> %c
144}
145
146define arm_aapcs_vfpcc <4 x i32> @u_v4i32(<4 x i32> %a, <4 x i32> %b) {
147; CHECK-LABEL: u_v4i32:
148; CHECK:       @ %bb.0: @ %entry
149; CHECK-NEXT:    vmov.i32 q2, #0x0
150; CHECK-NEXT:    vmov.i32 q3, #0x1
151; CHECK-NEXT:    vcmp.u32 hi, q0, q1
152; CHECK-NEXT:    vpsel q2, q3, q2
153; CHECK-NEXT:    vmov.i8 q3, #0xff
154; CHECK-NEXT:    vcmp.u32 hi, q1, q0
155; CHECK-NEXT:    vpsel q0, q3, q2
156; CHECK-NEXT:    bx lr
157entry:
158  %c = call <4 x i32> @llvm.ucmp(<4 x i32> %a, <4 x i32> %b)
159  ret <4 x i32> %c
160}
161
162define arm_aapcs_vfpcc <8 x i32> @u_v8i32(<8 x i32> %a, <8 x i32> %b) {
163; CHECK-LABEL: u_v8i32:
164; CHECK:       @ %bb.0: @ %entry
165; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
166; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
167; CHECK-NEXT:    vmov.i32 q4, #0x0
168; CHECK-NEXT:    vmov.i32 q5, #0x1
169; CHECK-NEXT:    vcmp.u32 hi, q0, q2
170; CHECK-NEXT:    vmov.i8 q7, #0xff
171; CHECK-NEXT:    vpsel q6, q5, q4
172; CHECK-NEXT:    vcmp.u32 hi, q2, q0
173; CHECK-NEXT:    vpsel q0, q7, q6
174; CHECK-NEXT:    vcmp.u32 hi, q1, q3
175; CHECK-NEXT:    vpsel q2, q5, q4
176; CHECK-NEXT:    vcmp.u32 hi, q3, q1
177; CHECK-NEXT:    vpsel q1, q7, q2
178; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
179; CHECK-NEXT:    bx lr
180entry:
181  %c = call <8 x i32> @llvm.ucmp(<8 x i32> %a, <8 x i32> %b)
182  ret <8 x i32> %c
183}
184
185define arm_aapcs_vfpcc <2 x i64> @u_v2i64(<2 x i64> %a, <2 x i64> %b) {
186; CHECK-LABEL: u_v2i64:
187; CHECK:       @ %bb.0: @ %entry
188; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, lr}
189; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, lr}
190; CHECK-NEXT:    vmov lr, r12, d0
191; CHECK-NEXT:    movs r4, #0
192; CHECK-NEXT:    vmov r3, r8, d2
193; CHECK-NEXT:    movs r0, #0
194; CHECK-NEXT:    vmov r6, r7, d3
195; CHECK-NEXT:    vmov.i32 q1, #0x0
196; CHECK-NEXT:    subs.w r1, r3, lr
197; CHECK-NEXT:    sbcs.w r1, r8, r12
198; CHECK-NEXT:    csetm r1, lo
199; CHECK-NEXT:    bfi r4, r1, #0, #8
200; CHECK-NEXT:    vmov r1, r5, d1
201; CHECK-NEXT:    subs r2, r6, r1
202; CHECK-NEXT:    sbcs.w r2, r7, r5
203; CHECK-NEXT:    csetm r2, lo
204; CHECK-NEXT:    bfi r4, r2, #8, #8
205; CHECK-NEXT:    adr r2, .LCPI8_0
206; CHECK-NEXT:    vldrw.u32 q0, [r2]
207; CHECK-NEXT:    subs.w r2, lr, r3
208; CHECK-NEXT:    sbcs.w r2, r12, r8
209; CHECK-NEXT:    vmsr p0, r4
210; CHECK-NEXT:    csetm r2, lo
211; CHECK-NEXT:    subs r1, r1, r6
212; CHECK-NEXT:    sbcs.w r1, r5, r7
213; CHECK-NEXT:    bfi r0, r2, #0, #8
214; CHECK-NEXT:    csetm r1, lo
215; CHECK-NEXT:    vpsel q0, q0, q1
216; CHECK-NEXT:    bfi r0, r1, #8, #8
217; CHECK-NEXT:    vmov.i8 q1, #0xff
218; CHECK-NEXT:    vmsr p0, r0
219; CHECK-NEXT:    vpsel q0, q1, q0
220; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, pc}
221; CHECK-NEXT:    .p2align 4
222; CHECK-NEXT:  @ %bb.1:
223; CHECK-NEXT:  .LCPI8_0:
224; CHECK-NEXT:    .long 1 @ 0x1
225; CHECK-NEXT:    .long 0 @ 0x0
226; CHECK-NEXT:    .long 1 @ 0x1
227; CHECK-NEXT:    .long 0 @ 0x0
228entry:
229  %c = call <2 x i64> @llvm.ucmp(<2 x i64> %a, <2 x i64> %b)
230  ret <2 x i64> %c
231}
232
233define arm_aapcs_vfpcc <4 x i64> @u_v4i64(<4 x i64> %a, <4 x i64> %b) {
234; CHECK-LABEL: u_v4i64:
235; CHECK:       @ %bb.0: @ %entry
236; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, lr}
237; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, r9, lr}
238; CHECK-NEXT:    .pad #4
239; CHECK-NEXT:    sub sp, #4
240; CHECK-NEXT:    .vsave {d8, d9, d10, d11}
241; CHECK-NEXT:    vpush {d8, d9, d10, d11}
242; CHECK-NEXT:    vmov lr, r12, d0
243; CHECK-NEXT:    movs r4, #0
244; CHECK-NEXT:    vmov r3, r8, d4
245; CHECK-NEXT:    vmov.i32 q5, #0x0
246; CHECK-NEXT:    vmov r6, r7, d5
247; CHECK-NEXT:    mov.w r9, #0
248; CHECK-NEXT:    vmov.i8 q2, #0xff
249; CHECK-NEXT:    subs.w r1, r3, lr
250; CHECK-NEXT:    sbcs.w r1, r8, r12
251; CHECK-NEXT:    csetm r1, lo
252; CHECK-NEXT:    bfi r4, r1, #0, #8
253; CHECK-NEXT:    vmov r1, r5, d1
254; CHECK-NEXT:    subs r2, r6, r1
255; CHECK-NEXT:    sbcs.w r2, r7, r5
256; CHECK-NEXT:    csetm r2, lo
257; CHECK-NEXT:    bfi r4, r2, #8, #8
258; CHECK-NEXT:    adr r2, .LCPI9_0
259; CHECK-NEXT:    vldrw.u32 q4, [r2]
260; CHECK-NEXT:    subs.w r2, lr, r3
261; CHECK-NEXT:    sbcs.w r2, r12, r8
262; CHECK-NEXT:    mov.w r3, #0
263; CHECK-NEXT:    csetm r2, lo
264; CHECK-NEXT:    subs r1, r1, r6
265; CHECK-NEXT:    sbcs.w r1, r5, r7
266; CHECK-NEXT:    bfi r3, r2, #0, #8
267; CHECK-NEXT:    csetm r1, lo
268; CHECK-NEXT:    vmsr p0, r4
269; CHECK-NEXT:    bfi r3, r1, #8, #8
270; CHECK-NEXT:    vpsel q0, q4, q5
271; CHECK-NEXT:    vmsr p0, r3
272; CHECK-NEXT:    vmov lr, r12, d2
273; CHECK-NEXT:    vmov r3, r7, d6
274; CHECK-NEXT:    movs r5, #0
275; CHECK-NEXT:    vmov r2, r1, d7
276; CHECK-NEXT:    vpsel q0, q2, q0
277; CHECK-NEXT:    subs.w r6, r3, lr
278; CHECK-NEXT:    sbcs.w r6, r7, r12
279; CHECK-NEXT:    csetm r6, lo
280; CHECK-NEXT:    bfi r5, r6, #0, #8
281; CHECK-NEXT:    vmov r6, r4, d3
282; CHECK-NEXT:    subs r0, r2, r6
283; CHECK-NEXT:    sbcs.w r0, r1, r4
284; CHECK-NEXT:    csetm r0, lo
285; CHECK-NEXT:    bfi r5, r0, #8, #8
286; CHECK-NEXT:    subs.w r0, lr, r3
287; CHECK-NEXT:    sbcs.w r0, r12, r7
288; CHECK-NEXT:    vmsr p0, r5
289; CHECK-NEXT:    csetm r0, lo
290; CHECK-NEXT:    vpsel q1, q4, q5
291; CHECK-NEXT:    bfi r9, r0, #0, #8
292; CHECK-NEXT:    subs r0, r6, r2
293; CHECK-NEXT:    sbcs.w r0, r4, r1
294; CHECK-NEXT:    csetm r0, lo
295; CHECK-NEXT:    bfi r9, r0, #8, #8
296; CHECK-NEXT:    vmsr p0, r9
297; CHECK-NEXT:    vpsel q1, q2, q1
298; CHECK-NEXT:    vpop {d8, d9, d10, d11}
299; CHECK-NEXT:    add sp, #4
300; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, pc}
301; CHECK-NEXT:    .p2align 4
302; CHECK-NEXT:  @ %bb.1:
303; CHECK-NEXT:  .LCPI9_0:
304; CHECK-NEXT:    .long 1 @ 0x1
305; CHECK-NEXT:    .long 0 @ 0x0
306; CHECK-NEXT:    .long 1 @ 0x1
307; CHECK-NEXT:    .long 0 @ 0x0
308entry:
309  %c = call <4 x i64> @llvm.ucmp(<4 x i64> %a, <4 x i64> %b)
310  ret <4 x i64> %c
311}
312
313define arm_aapcs_vfpcc <16 x i8> @signOf_neon(<8 x i16> %s0_lo, <8 x i16> %s0_hi, <8 x i16> %s1_lo, <8 x i16> %s1_hi) {
314; CHECK-LABEL: signOf_neon:
315; CHECK:       @ %bb.0: @ %entry
316; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
317; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}
318; CHECK-NEXT:    .pad #16
319; CHECK-NEXT:    sub sp, #16
320; CHECK-NEXT:    vmov.i32 q4, #0x0
321; CHECK-NEXT:    vmov.i16 q5, #0x1
322; CHECK-NEXT:    vcmp.u16 hi, q1, q3
323; CHECK-NEXT:    vmov.i8 q7, #0xff
324; CHECK-NEXT:    vpsel q6, q5, q4
325; CHECK-NEXT:    vcmp.u16 hi, q3, q1
326; CHECK-NEXT:    vpsel q1, q7, q6
327; CHECK-NEXT:    vcmp.u16 hi, q0, q2
328; CHECK-NEXT:    vpsel q3, q5, q4
329; CHECK-NEXT:    vcmp.u16 hi, q2, q0
330; CHECK-NEXT:    mov r0, sp
331; CHECK-NEXT:    vpsel q0, q7, q3
332; CHECK-NEXT:    vstrb.16 q1, [r0, #8]
333; CHECK-NEXT:    vstrb.16 q0, [r0]
334; CHECK-NEXT:    vldrw.u32 q0, [r0]
335; CHECK-NEXT:    add sp, #16
336; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}
337; CHECK-NEXT:    bx lr
338entry:
339  %0 = shufflevector <8 x i16> %s0_lo, <8 x i16> %s0_hi, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
340  %1 = shufflevector <8 x i16> %s1_lo, <8 x i16> %s1_hi, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
341  %or.i = tail call <16 x i8> @llvm.ucmp.v16i8.v16i16(<16 x i16> %0, <16 x i16> %1)
342  ret <16 x i8> %or.i
343}
344