xref: /llvm-project/llvm/test/CodeGen/Thumb2/LowOverheadLoops/constbound.ll (revision b31fffbc7f1e0491bf599e82b7195e320d26e140) 
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 %s -o - | FileCheck %s
3
4define dso_local i32 @test_500_504(ptr nocapture readonly %x) {
5; CHECK-LABEL: test_500_504:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    .save {r7, lr}
8; CHECK-NEXT:    push {r7, lr}
9; CHECK-NEXT:    mov.w lr, #126
10; CHECK-NEXT:    adr r2, .LCPI0_0
11; CHECK-NEXT:    vldrw.u32 q0, [r2]
12; CHECK-NEXT:    mov.w r2, #500
13; CHECK-NEXT:    vdup.32 q1, r2
14; CHECK-NEXT:    movs r1, #0
15; CHECK-NEXT:    movs r2, #0
16; CHECK-NEXT:  .LBB0_1: @ %vector.body
17; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
18; CHECK-NEXT:    vqadd.u32 q2, q0, r1
19; CHECK-NEXT:    adds r1, #4
20; CHECK-NEXT:    vptt.u32 hi, q1, q2
21; CHECK-NEXT:    vldrwt.u32 q2, [r0], #16
22; CHECK-NEXT:    vaddvat.u32 r2, q2
23; CHECK-NEXT:    le lr, .LBB0_1
24; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
25; CHECK-NEXT:    mov r0, r2
26; CHECK-NEXT:    pop {r7, pc}
27; CHECK-NEXT:    .p2align 4
28; CHECK-NEXT:  @ %bb.3:
29; CHECK-NEXT:  .LCPI0_0:
30; CHECK-NEXT:    .long 0 @ 0x0
31; CHECK-NEXT:    .long 1 @ 0x1
32; CHECK-NEXT:    .long 2 @ 0x2
33; CHECK-NEXT:    .long 3 @ 0x3
34entry:
35  br label %vector.body
36
37vector.body:                                      ; preds = %vector.body, %entry
38  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
39  %vec.phi = phi i32 [ 0, %entry ], [ %4, %vector.body ]
40  %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 500)
41  %0 = getelementptr inbounds i32, ptr %x, i32 %index
42  %1 = bitcast ptr %0 to ptr
43  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
44  %2 = select <4 x i1> %active.lane.mask, <4 x i32> %wide.masked.load, <4 x i32> zeroinitializer
45  %3 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %2)
46  %4 = add i32 %3, %vec.phi
47  %index.next = add i32 %index, 4
48  %5 = icmp eq i32 %index.next, 504
49  br i1 %5, label %for.cond.cleanup, label %vector.body
50
51for.cond.cleanup:                                 ; preds = %vector.body
52  ret i32 %4
53}
54
55define dso_local i32 @test_501_504(ptr nocapture readonly %x) {
56; CHECK-LABEL: test_501_504:
57; CHECK:       @ %bb.0: @ %entry
58; CHECK-NEXT:    .save {r7, lr}
59; CHECK-NEXT:    push {r7, lr}
60; CHECK-NEXT:    movw r1, #501
61; CHECK-NEXT:    movs r2, #0
62; CHECK-NEXT:    dlstp.32 lr, r1
63; CHECK-NEXT:  .LBB1_1: @ %vector.body
64; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
65; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
66; CHECK-NEXT:    vaddva.u32 r2, q0
67; CHECK-NEXT:    letp lr, .LBB1_1
68; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
69; CHECK-NEXT:    mov r0, r2
70; CHECK-NEXT:    pop {r7, pc}
71entry:
72  br label %vector.body
73
74vector.body:                                      ; preds = %vector.body, %entry
75  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
76  %vec.phi = phi i32 [ 0, %entry ], [ %4, %vector.body ]
77  %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 501)
78  %0 = getelementptr inbounds i32, ptr %x, i32 %index
79  %1 = bitcast ptr %0 to ptr
80  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
81  %2 = select <4 x i1> %active.lane.mask, <4 x i32> %wide.masked.load, <4 x i32> zeroinitializer
82  %3 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %2)
83  %4 = add i32 %3, %vec.phi
84  %index.next = add i32 %index, 4
85  %5 = icmp eq i32 %index.next, 504
86  br i1 %5, label %for.cond.cleanup, label %vector.body
87
88for.cond.cleanup:                                 ; preds = %vector.body
89  ret i32 %4
90}
91
92define dso_local i32 @test_502_504(ptr nocapture readonly %x) {
93; CHECK-LABEL: test_502_504:
94; CHECK:       @ %bb.0: @ %entry
95; CHECK-NEXT:    .save {r7, lr}
96; CHECK-NEXT:    push {r7, lr}
97; CHECK-NEXT:    mov.w r1, #502
98; CHECK-NEXT:    movs r2, #0
99; CHECK-NEXT:    dlstp.32 lr, r1
100; CHECK-NEXT:  .LBB2_1: @ %vector.body
101; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
102; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
103; CHECK-NEXT:    vaddva.u32 r2, q0
104; CHECK-NEXT:    letp lr, .LBB2_1
105; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
106; CHECK-NEXT:    mov r0, r2
107; CHECK-NEXT:    pop {r7, pc}
108entry:
109  br label %vector.body
110
111vector.body:                                      ; preds = %vector.body, %entry
112  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
113  %vec.phi = phi i32 [ 0, %entry ], [ %4, %vector.body ]
114  %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 502)
115  %0 = getelementptr inbounds i32, ptr %x, i32 %index
116  %1 = bitcast ptr %0 to ptr
117  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
118  %2 = select <4 x i1> %active.lane.mask, <4 x i32> %wide.masked.load, <4 x i32> zeroinitializer
119  %3 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %2)
120  %4 = add i32 %3, %vec.phi
121  %index.next = add i32 %index, 4
122  %5 = icmp eq i32 %index.next, 504
123  br i1 %5, label %for.cond.cleanup, label %vector.body
124
125for.cond.cleanup:                                 ; preds = %vector.body
126  ret i32 %4
127}
128
129define dso_local i32 @test_503_504(ptr nocapture readonly %x) {
130; CHECK-LABEL: test_503_504:
131; CHECK:       @ %bb.0: @ %entry
132; CHECK-NEXT:    .save {r7, lr}
133; CHECK-NEXT:    push {r7, lr}
134; CHECK-NEXT:    movw r1, #503
135; CHECK-NEXT:    movs r2, #0
136; CHECK-NEXT:    dlstp.32 lr, r1
137; CHECK-NEXT:  .LBB3_1: @ %vector.body
138; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
139; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
140; CHECK-NEXT:    vaddva.u32 r2, q0
141; CHECK-NEXT:    letp lr, .LBB3_1
142; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
143; CHECK-NEXT:    mov r0, r2
144; CHECK-NEXT:    pop {r7, pc}
145entry:
146  br label %vector.body
147
148vector.body:                                      ; preds = %vector.body, %entry
149  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
150  %vec.phi = phi i32 [ 0, %entry ], [ %4, %vector.body ]
151  %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 503)
152  %0 = getelementptr inbounds i32, ptr %x, i32 %index
153  %1 = bitcast ptr %0 to ptr
154  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
155  %2 = select <4 x i1> %active.lane.mask, <4 x i32> %wide.masked.load, <4 x i32> zeroinitializer
156  %3 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %2)
157  %4 = add i32 %3, %vec.phi
158  %index.next = add i32 %index, 4
159  %5 = icmp eq i32 %index.next, 504
160  br i1 %5, label %for.cond.cleanup, label %vector.body
161
162for.cond.cleanup:                                 ; preds = %vector.body
163  ret i32 %4
164}
165
166define dso_local i32 @test_504_504(ptr nocapture readonly %x) {
167; CHECK-LABEL: test_504_504:
168; CHECK:       @ %bb.0: @ %entry
169; CHECK-NEXT:    .save {r7, lr}
170; CHECK-NEXT:    push {r7, lr}
171; CHECK-NEXT:    mov.w r1, #504
172; CHECK-NEXT:    movs r2, #0
173; CHECK-NEXT:    dlstp.32 lr, r1
174; CHECK-NEXT:  .LBB4_1: @ %vector.body
175; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
176; CHECK-NEXT:    vldrw.u32 q0, [r0], #16
177; CHECK-NEXT:    vaddva.u32 r2, q0
178; CHECK-NEXT:    letp lr, .LBB4_1
179; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
180; CHECK-NEXT:    mov r0, r2
181; CHECK-NEXT:    pop {r7, pc}
182entry:
183  br label %vector.body
184
185vector.body:                                      ; preds = %vector.body, %entry
186  %index = phi i32 [ 0, %entry ], [ %index.next, %vector.body ]
187  %vec.phi = phi i32 [ 0, %entry ], [ %4, %vector.body ]
188  %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 504)
189  %0 = getelementptr inbounds i32, ptr %x, i32 %index
190  %1 = bitcast ptr %0 to ptr
191  %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)
192  %2 = select <4 x i1> %active.lane.mask, <4 x i32> %wide.masked.load, <4 x i32> zeroinitializer
193  %3 = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> %2)
194  %4 = add i32 %3, %vec.phi
195  %index.next = add i32 %index, 4
196  %5 = icmp eq i32 %index.next, 504
197  br i1 %5, label %for.cond.cleanup, label %vector.body
198
199for.cond.cleanup:                                 ; preds = %vector.body
200  ret i32 %4
201}
202
203declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)
204declare <4 x i32> @llvm.masked.load.v4i32.p0(ptr, i32 immarg, <4 x i1>, <4 x i32>)
205declare i32 @llvm.vector.reduce.add.v4i32(<4 x i32>)
206