1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=arm-none-eabi -mattr=+neon | FileCheck %s --check-prefix=CHECK
4 declare half @llvm.vector.reduce.fmul.f16.v1f16(half, <1 x half>)
5 declare float @llvm.vector.reduce.fmul.f32.v1f32(float, <1 x float>)
6 declare double @llvm.vector.reduce.fmul.f64.v1f64(double, <1 x double>)
7 declare fp128 @llvm.vector.reduce.fmul.f128.v1f128(fp128, <1 x fp128>)
9 declare float @llvm.vector.reduce.fmul.f32.v3f32(float, <3 x float>)
10 declare fp128 @llvm.vector.reduce.fmul.f128.v2f128(fp128, <2 x fp128>)
11 declare float @llvm.vector.reduce.fmul.f32.v16f32(float, <16 x float>)
13 define half @test_v1f16(<1 x half> %a) nounwind {
14 ; CHECK-LABEL: test_v1f16:
16 ; CHECK-NEXT: .save {r11, lr}
17 ; CHECK-NEXT: push {r11, lr}
18 ; CHECK-NEXT: bl __aeabi_f2h
19 ; CHECK-NEXT: mov r1, #255
20 ; CHECK-NEXT: orr r1, r1, #65280
21 ; CHECK-NEXT: and r0, r0, r1
22 ; CHECK-NEXT: pop {r11, lr}
23 ; CHECK-NEXT: mov pc, lr
24 %b = call half @llvm.vector.reduce.fmul.f16.v1f16(half 1.0, <1 x half> %a)
28 define float @test_v1f32(<1 x float> %a) nounwind {
29 ; CHECK-LABEL: test_v1f32:
31 ; CHECK-NEXT: mov pc, lr
32 %b = call float @llvm.vector.reduce.fmul.f32.v1f32(float 1.0, <1 x float> %a)
36 define double @test_v1f64(<1 x double> %a) nounwind {
37 ; CHECK-LABEL: test_v1f64:
39 ; CHECK-NEXT: mov pc, lr
40 %b = call double @llvm.vector.reduce.fmul.f64.v1f64(double 1.0, <1 x double> %a)
44 define fp128 @test_v1f128(<1 x fp128> %a) nounwind {
45 ; CHECK-LABEL: test_v1f128:
47 ; CHECK-NEXT: mov pc, lr
48 %b = call fp128 @llvm.vector.reduce.fmul.f128.v1f128(fp128 0xL00000000000000003fff00000000000000, <1 x fp128> %a)
52 define float @test_v3f32(<3 x float> %a) nounwind {
53 ; CHECK-LABEL: test_v3f32:
55 ; CHECK-NEXT: vmov d1, r2, r3
56 ; CHECK-NEXT: vmov d0, r0, r1
57 ; CHECK-NEXT: vmul.f32 s4, s0, s1
58 ; CHECK-NEXT: vmul.f32 s0, s4, s2
59 ; CHECK-NEXT: vmov r0, s0
60 ; CHECK-NEXT: mov pc, lr
61 %b = call float @llvm.vector.reduce.fmul.f32.v3f32(float 1.0, <3 x float> %a)
65 define fp128 @test_v2f128(<2 x fp128> %a) nounwind {
66 ; CHECK-LABEL: test_v2f128:
68 ; CHECK-NEXT: .save {r4, r5, r11, lr}
69 ; CHECK-NEXT: push {r4, r5, r11, lr}
70 ; CHECK-NEXT: .pad #16
71 ; CHECK-NEXT: sub sp, sp, #16
72 ; CHECK-NEXT: ldr r12, [sp, #36]
73 ; CHECK-NEXT: ldr lr, [sp, #32]
74 ; CHECK-NEXT: ldr r4, [sp, #40]
75 ; CHECK-NEXT: ldr r5, [sp, #44]
76 ; CHECK-NEXT: str lr, [sp]
77 ; CHECK-NEXT: str r12, [sp, #4]
78 ; CHECK-NEXT: str r4, [sp, #8]
79 ; CHECK-NEXT: str r5, [sp, #12]
80 ; CHECK-NEXT: bl __multf3
81 ; CHECK-NEXT: add sp, sp, #16
82 ; CHECK-NEXT: pop {r4, r5, r11, lr}
83 ; CHECK-NEXT: mov pc, lr
84 %b = call fp128 @llvm.vector.reduce.fmul.f128.v2f128(fp128 0xL00000000000000003fff00000000000000, <2 x fp128> %a)
88 define float @test_v16f32(<16 x float> %a) nounwind {
89 ; CHECK-LABEL: test_v16f32:
91 ; CHECK-NEXT: vmov d1, r2, r3
92 ; CHECK-NEXT: vmov d0, r0, r1
93 ; CHECK-NEXT: mov r0, sp
94 ; CHECK-NEXT: vmul.f32 s4, s0, s1
95 ; CHECK-NEXT: vmul.f32 s4, s4, s2
96 ; CHECK-NEXT: vmul.f32 s0, s4, s3
97 ; CHECK-NEXT: vld1.64 {d2, d3}, [r0]
98 ; CHECK-NEXT: add r0, sp, #16
99 ; CHECK-NEXT: vmul.f32 s0, s0, s4
100 ; CHECK-NEXT: vmul.f32 s0, s0, s5
101 ; CHECK-NEXT: vmul.f32 s0, s0, s6
102 ; CHECK-NEXT: vmul.f32 s0, s0, s7
103 ; CHECK-NEXT: vld1.64 {d2, d3}, [r0]
104 ; CHECK-NEXT: add r0, sp, #32
105 ; CHECK-NEXT: vmul.f32 s0, s0, s4
106 ; CHECK-NEXT: vmul.f32 s0, s0, s5
107 ; CHECK-NEXT: vmul.f32 s0, s0, s6
108 ; CHECK-NEXT: vmul.f32 s0, s0, s7
109 ; CHECK-NEXT: vld1.64 {d2, d3}, [r0]
110 ; CHECK-NEXT: vmul.f32 s0, s0, s4
111 ; CHECK-NEXT: vmul.f32 s0, s0, s5
112 ; CHECK-NEXT: vmul.f32 s0, s0, s6
113 ; CHECK-NEXT: vmul.f32 s0, s0, s7
114 ; CHECK-NEXT: vmov r0, s0
115 ; CHECK-NEXT: mov pc, lr
116 %b = call float @llvm.vector.reduce.fmul.f32.v16f32(float 1.0, <16 x float> %a)