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.fp %s -verify-machineinstrs -o - | FileCheck %s
4 define void @arm_min_helium_f32(ptr %pSrc, i32 %blockSize, ptr nocapture %pResult, ptr nocapture %pIndex) {
5 ; CHECK-LABEL: arm_min_helium_f32:
6 ; CHECK: @ %bb.0: @ %entry
7 ; CHECK-NEXT: .save {r4, r5, r7, lr}
8 ; CHECK-NEXT: push {r4, r5, r7, lr}
9 ; CHECK-NEXT: movs r4, #0
10 ; CHECK-NEXT: mov.w r12, #4
11 ; CHECK-NEXT: vidup.u32 q2, r4, #1
12 ; CHECK-NEXT: movw r5, #54437
13 ; CHECK-NEXT: movt r5, #21352
14 ; CHECK-NEXT: vdup.32 q1, r5
15 ; CHECK-NEXT: vmov.i32 q0, #0x0
16 ; CHECK-NEXT: dlstp.32 lr, r1
17 ; CHECK-NEXT: .LBB0_1: @ %do.body
18 ; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
19 ; CHECK-NEXT: vldrw.u32 q3, [r0], #16
20 ; CHECK-NEXT: vptt.f32 ge, q1, q3
21 ; CHECK-NEXT: vmovt q0, q2
22 ; CHECK-NEXT: vmovt q1, q3
23 ; CHECK-NEXT: vadd.i32 q2, q2, r12
24 ; CHECK-NEXT: letp lr, .LBB0_1
25 ; CHECK-NEXT: @ %bb.2: @ %do.end
26 ; CHECK-NEXT: vldr s8, .LCPI0_0
27 ; CHECK-NEXT: vdup.32 q3, r1
28 ; CHECK-NEXT: vmov r0, s8
29 ; CHECK-NEXT: vminnmv.f32 r0, q1
30 ; CHECK-NEXT: vcmp.f32 le, q1, r0
31 ; CHECK-NEXT: vmov s8, r0
32 ; CHECK-NEXT: vpsel q0, q0, q3
33 ; CHECK-NEXT: vminv.u32 r1, q0
34 ; CHECK-NEXT: str r1, [r3]
35 ; CHECK-NEXT: vstr s8, [r2]
36 ; CHECK-NEXT: pop {r4, r5, r7, pc}
37 ; CHECK-NEXT: .p2align 2
38 ; CHECK-NEXT: @ %bb.3:
39 ; CHECK-NEXT: .LCPI0_0:
40 ; CHECK-NEXT: .long 0x5368d4a5 @ float 9.99999995E+11
42 %0 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32 0, i32 1)
43 %1 = extractvalue { <4 x i32>, i32 } %0, 0
46 do.body: ; preds = %do.body, %entry
47 %curExtremValVec.0 = phi <4 x float> [ <float 0x426D1A94A0000000, float 0x426D1A94A0000000, float 0x426D1A94A0000000, float 0x426D1A94A0000000>, %entry ], [ %8, %do.body ]
48 %indexVec.0 = phi <4 x i32> [ %1, %entry ], [ %11, %do.body ]
49 %2 = phi <4 x float> [ zeroinitializer, %entry ], [ %10, %do.body ]
50 %blkCnt.0 = phi i32 [ %blockSize, %entry ], [ %sub, %do.body ]
51 %pSrc.addr.0 = phi ptr [ %pSrc, %entry ], [ %add.ptr, %do.body ]
52 %3 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %blkCnt.0)
53 %4 = bitcast ptr %pSrc.addr.0 to ptr
54 %5 = tail call fast <4 x float> @llvm.masked.load.v4f32.p0(ptr %4, i32 4, <4 x i1> %3, <4 x float> zeroinitializer)
55 %6 = fcmp fast ole <4 x float> %5, %curExtremValVec.0
56 %7 = and <4 x i1> %6, %3
57 %8 = select fast <4 x i1> %7, <4 x float> %5, <4 x float> %curExtremValVec.0
58 %9 = bitcast <4 x i32> %indexVec.0 to <4 x float>
59 %10 = select fast <4 x i1> %7, <4 x float> %9, <4 x float> %2
60 %11 = add <4 x i32> %indexVec.0, <i32 4, i32 4, i32 4, i32 4>
61 %add.ptr = getelementptr inbounds float, ptr %pSrc.addr.0, i32 4
62 %sub = add nsw i32 %blkCnt.0, -4
63 %cmp = icmp sgt i32 %blkCnt.0, 4
64 br i1 %cmp, label %do.body, label %do.end
66 do.end: ; preds = %do.body
67 %12 = bitcast <4 x float> %10 to <4 x i32>
68 %13 = tail call fast float @llvm.arm.mve.minnmv.f32.v4f32(float 0x426D1A94A0000000, <4 x float> %8)
69 %.splatinsert = insertelement <4 x float> undef, float %13, i32 0
70 %.splat = shufflevector <4 x float> %.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
71 %14 = fcmp fast ole <4 x float> %8, %.splat
72 %.splatinsert1 = insertelement <4 x i32> undef, i32 %blockSize, i32 0
73 %.splat2 = shufflevector <4 x i32> %.splatinsert1, <4 x i32> undef, <4 x i32> zeroinitializer
74 %15 = select <4 x i1> %14, <4 x i32> %12, <4 x i32> %.splat2
75 %16 = tail call i32 @llvm.arm.mve.minv.v4i32(i32 %blockSize, <4 x i32> %15, i32 1)
76 store i32 %16, ptr %pIndex, align 4
77 store float %13, ptr %pResult, align 4
81 declare { <4 x i32>, i32 } @llvm.arm.mve.vidup.v4i32(i32, i32) #1
82 declare <4 x i1> @llvm.arm.mve.vctp32(i32) #1
83 declare <4 x float> @llvm.masked.load.v4f32.p0(ptr, i32 immarg, <4 x i1>, <4 x float>) #2
84 declare float @llvm.arm.mve.minnmv.f32.v4f32(float, <4 x float>) #1
85 declare i32 @llvm.arm.mve.minv.v4i32(i32, <4 x i32>, i32) #1