xref: /llvm-project/llvm/test/CodeGen/Thumb2/mve-vmull-loop.ll (revision c4a60c9d34375e73fc2da5e02215eabe4bc90e8f) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK
3
4define arm_aapcs_vfpcc void @test32(ptr noalias nocapture readonly %x, ptr noalias nocapture readonly %y, ptr nocapture %z, i32 %n) {
5; CHECK-LABEL: test32:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    cmp r3, #1
8; CHECK-NEXT:    it lt
9; CHECK-NEXT:    bxlt lr
10; CHECK-NEXT:  .LBB0_1: @ %vector.body.preheader
11; CHECK-NEXT:    .save {r4, r5, r7, lr}
12; CHECK-NEXT:    push {r4, r5, r7, lr}
13; CHECK-NEXT:  .LBB0_2: @ %vector.body
14; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
15; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
16; CHECK-NEXT:    vldrw.u32 q1, [r1], #16
17; CHECK-NEXT:    subs r3, #4
18; CHECK-NEXT:    vmullb.s32 q2, q1, q0
19; CHECK-NEXT:    vmullt.s32 q3, q1, q0
20; CHECK-NEXT:    vmov r12, r5, d5
21; CHECK-NEXT:    lsrl r12, r5, #31
22; CHECK-NEXT:    vmov r4, r5, d4
23; CHECK-NEXT:    lsrl r4, r5, #31
24; CHECK-NEXT:    vmov q2[2], q2[0], r4, r12
25; CHECK-NEXT:    vmov r12, r5, d7
26; CHECK-NEXT:    lsrl r12, r5, #31
27; CHECK-NEXT:    vmov r4, r5, d6
28; CHECK-NEXT:    lsrl r4, r5, #31
29; CHECK-NEXT:    vmov q2[3], q2[1], r4, r12
30; CHECK-NEXT:    vstrb.8 q2, [r2], #16
31; CHECK-NEXT:    bne .LBB0_2
32; CHECK-NEXT:  @ %bb.3:
33; CHECK-NEXT:    pop.w {r4, r5, r7, lr}
34; CHECK-NEXT:    bx lr
35entry:
36  %0 = and i32 %n, 3
37  %cmp = icmp eq i32 %0, 0
38  %cmp113 = icmp sgt i32 %n, 0
39  br i1 %cmp113, label %vector.body, label %for.cond.cleanup
40
41vector.body:                                      ; preds = %vector.body, %entry
42  %index = phi i32 [ %index.next, %vector.body ], [ 0, %entry ]
43  %1 = getelementptr inbounds i32, ptr %x, i32 %index
44  %wide.load = load <4 x i32>, ptr %1, align 4
45  %2 = shufflevector <4 x i32> %wide.load, <4 x i32> %wide.load, <2 x i32> <i32 0, i32 2>
46  %3 = shufflevector <4 x i32> %wide.load, <4 x i32> %wide.load, <2 x i32> <i32 1, i32 3>
47  %4 = sext <2 x i32> %2 to <2 x i64>
48  %5 = sext <2 x i32> %3 to <2 x i64>
49  %6 = getelementptr inbounds i32, ptr %y, i32 %index
50  %wide.load15 = load <4 x i32>, ptr %6, align 4
51  %7 = shufflevector <4 x i32> %wide.load15, <4 x i32> %wide.load15, <2 x i32> <i32 0, i32 2>
52  %8 = shufflevector <4 x i32> %wide.load15, <4 x i32> %wide.load15, <2 x i32> <i32 1, i32 3>
53  %9 = sext <2 x i32> %7 to <2 x i64>
54  %10 = sext <2 x i32> %8 to <2 x i64>
55  %11 = mul <2 x i64> %9, %4
56  %12 = mul <2 x i64> %10, %5
57  %13 = lshr <2 x i64> %11, <i64 31, i64 31>
58  %14 = lshr <2 x i64> %12, <i64 31, i64 31>
59  %15 = shufflevector <2 x i64> %13, <2 x i64> %14, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
60  %16 = trunc <4 x i64> %15 to <4 x i32>
61  %17 = getelementptr inbounds i32, ptr %z, i32 %index
62  store <4 x i32> %16, ptr %17, align 4
63  %index.next = add i32 %index, 4
64  %18 = icmp eq i32 %index.next, %n
65  br i1 %18, label %for.cond.cleanup, label %vector.body
66
67for.cond.cleanup:                                 ; preds = %vector.body, %entry
68  ret void
69}
70
71define arm_aapcs_vfpcc void @test16(ptr noalias nocapture readonly %x, ptr noalias nocapture readonly %y, ptr nocapture %z, i32 %n) {
72; CHECK-LABEL: test16:
73; CHECK:       @ %bb.0: @ %entry
74; CHECK-NEXT:    cmp r3, #1
75; CHECK-NEXT:    it lt
76; CHECK-NEXT:    bxlt lr
77; CHECK-NEXT:  .LBB1_1: @ %vector.body
78; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
79; CHECK-NEXT:    vldrh.u16 q0, [r0], #16
80; CHECK-NEXT:    vldrh.u16 q1, [r1], #16
81; CHECK-NEXT:    subs r3, #8
82; CHECK-NEXT:    vmullt.s16 q2, q1, q0
83; CHECK-NEXT:    vmullb.s16 q0, q1, q0
84; CHECK-NEXT:    vshr.u32 q2, q2, #15
85; CHECK-NEXT:    vshr.u32 q0, q0, #15
86; CHECK-NEXT:    vmovnt.i32 q0, q2
87; CHECK-NEXT:    vstrb.8 q0, [r2], #16
88; CHECK-NEXT:    bne .LBB1_1
89; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
90; CHECK-NEXT:    bx lr
91entry:
92  %0 = and i32 %n, 7
93  %cmp = icmp eq i32 %0, 0
94  %cmp113 = icmp sgt i32 %n, 0
95  br i1 %cmp113, label %vector.body, label %for.cond.cleanup
96
97vector.body:                                      ; preds = %vector.body, %entry
98  %index = phi i32 [ %index.next, %vector.body ], [ 0, %entry ]
99  %1 = getelementptr inbounds i16, ptr %x, i32 %index
100  %wide.load = load <8 x i16>, ptr %1, align 2
101  %2 = shufflevector <8 x i16> %wide.load, <8 x i16> %wide.load, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
102  %3 = shufflevector <8 x i16> %wide.load, <8 x i16> %wide.load, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
103  %4 = sext <4 x i16> %2 to <4 x i32>
104  %5 = sext <4 x i16> %3 to <4 x i32>
105  %6 = getelementptr inbounds i16, ptr %y, i32 %index
106  %wide.load15 = load <8 x i16>, ptr %6, align 2
107  %7 = shufflevector <8 x i16> %wide.load15, <8 x i16> %wide.load15, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
108  %8 = shufflevector <8 x i16> %wide.load15, <8 x i16> %wide.load15, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
109  %9 = sext <4 x i16> %7 to <4 x i32>
110  %10 = sext <4 x i16> %8 to <4 x i32>
111  %11 = mul <4 x i32> %9, %4
112  %12 = mul <4 x i32> %10, %5
113  %13 = lshr <4 x i32> %11, <i32 15, i32 15, i32 15, i32 15>
114  %14 = lshr <4 x i32> %12, <i32 15, i32 15, i32 15, i32 15>
115  %15 = shufflevector <4 x i32> %13, <4 x i32> %14, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
116  %16 = trunc <8 x i32> %15 to <8 x i16>
117  %17 = getelementptr inbounds i16, ptr %z, i32 %index
118  store <8 x i16> %16, ptr %17, align 2
119  %index.next = add i32 %index, 8
120  %18 = icmp eq i32 %index.next, %n
121  br i1 %18, label %for.cond.cleanup, label %vector.body
122
123for.cond.cleanup:                                 ; preds = %vector.body, %entry
124  ret void
125}
126
127define arm_aapcs_vfpcc void @test8(ptr noalias nocapture readonly %x, ptr noalias nocapture readonly %y, ptr nocapture %z, i32 %n) {
128; CHECK-LABEL: test8:
129; CHECK:       @ %bb.0: @ %entry
130; CHECK-NEXT:    cmp r3, #1
131; CHECK-NEXT:    it lt
132; CHECK-NEXT:    bxlt lr
133; CHECK-NEXT:  .LBB2_1: @ %vector.body
134; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
135; CHECK-NEXT:    vldrb.u8 q0, [r0], #16
136; CHECK-NEXT:    vldrb.u8 q1, [r1], #16
137; CHECK-NEXT:    subs r3, #16
138; CHECK-NEXT:    vmullt.u8 q2, q1, q0
139; CHECK-NEXT:    vmullb.u8 q0, q1, q0
140; CHECK-NEXT:    vshr.u16 q2, q2, #7
141; CHECK-NEXT:    vshr.u16 q0, q0, #7
142; CHECK-NEXT:    vmovnt.i16 q0, q2
143; CHECK-NEXT:    vstrb.8 q0, [r2], #16
144; CHECK-NEXT:    bne .LBB2_1
145; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
146; CHECK-NEXT:    bx lr
147entry:
148  %0 = and i32 %n, 15
149  %cmp = icmp eq i32 %0, 0
150  %cmp117 = icmp sgt i32 %n, 0
151  br i1 %cmp117, label %vector.body, label %for.cond.cleanup
152
153vector.body:                                      ; preds = %vector.body, %entry
154  %index = phi i32 [ %index.next, %vector.body ], [ 0, %entry ]
155  %1 = getelementptr inbounds i8, ptr %x, i32 %index
156  %wide.load = load <16 x i8>, ptr %1, align 1
157  %2 = shufflevector <16 x i8> %wide.load, <16 x i8> %wide.load, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
158  %3 = shufflevector <16 x i8> %wide.load, <16 x i8> %wide.load, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
159  %4 = zext <8 x i8> %2 to <8 x i16>
160  %5 = zext <8 x i8> %3 to <8 x i16>
161  %6 = getelementptr inbounds i8, ptr %y, i32 %index
162  %wide.load19 = load <16 x i8>, ptr %6, align 1
163  %7 = shufflevector <16 x i8> %wide.load19, <16 x i8> %wide.load19, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
164  %8 = shufflevector <16 x i8> %wide.load19, <16 x i8> %wide.load19, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
165  %9 = zext <8 x i8> %7 to <8 x i16>
166  %10 = zext <8 x i8> %8 to <8 x i16>
167  %11 = mul <8 x i16> %9, %4
168  %12 = mul <8 x i16> %10, %5
169  %13 = lshr <8 x i16> %11, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
170  %14 = lshr <8 x i16> %12, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
171  %15 = shufflevector <8 x i16> %13, <8 x i16> %14, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
172  %16 = trunc <16 x i16> %15 to <16 x i8>
173  %17 = getelementptr inbounds i8, ptr %z, i32 %index
174  store <16 x i8> %16, ptr %17, align 1
175  %index.next = add i32 %index, 16
176  %18 = icmp eq i32 %index.next, %n
177  br i1 %18, label %for.cond.cleanup, label %vector.body
178
179for.cond.cleanup:                                 ; preds = %vector.body, %entry
180  ret void
181}
182