150
|
1 // RUN: %clang_cc1 -triple arm64-apple-ios7 -target-feature +neon -ffreestanding -S -o - -emit-llvm %s | FileCheck %s
|
|
2 // Test ARM64 SIMD duplicate lane and n intrinsics
|
|
3
|
|
4 #include <arm_neon.h>
|
|
5
|
|
6 void test_vdup_lane_s64(int64x1_t a1) {
|
|
7 // CHECK-LABEL: test_vdup_lane_s64
|
|
8 vdup_lane_s64(a1, 0);
|
|
9 // CHECK: shufflevector
|
|
10 }
|
|
11
|
|
12 void test_vdup_lane_u64(uint64x1_t a1) {
|
|
13 // CHECK-LABEL: test_vdup_lane_u64
|
|
14 vdup_lane_u64(a1, 0);
|
|
15 // CHECK: shufflevector
|
|
16 }
|
|
17
|
|
18 // uncomment out the following code once scalar_to_vector in the backend
|
|
19 // works (for 64 bit?). Change the "CHECK@" to "CHECK<colon>"
|
|
20 /*
|
|
21 float64x1_t test_vdup_n_f64(float64_t a1) {
|
|
22 // CHECK-LABEL@ test_vdup_n_f64
|
|
23 return vdup_n_f64(a1);
|
|
24 // match that an element is inserted into part 0
|
|
25 // CHECK@ insertelement {{.*, i32 0 *$}}
|
|
26 }
|
|
27 */
|
|
28
|
|
29 float16x8_t test_vdupq_n_f16(float16_t *a1) {
|
|
30 // CHECK-LABEL: test_vdupq_n_f16
|
|
31 return vdupq_n_f16(*a1);
|
|
32 // match that an element is inserted into parts 0-7. The backend better
|
|
33 // turn that into a single dup instruction
|
|
34 // CHECK: insertelement {{.*, i32 0 *$}}
|
|
35 // CHECK: insertelement {{.*, i32 1 *$}}
|
|
36 // CHECK: insertelement {{.*, i32 2 *$}}
|
|
37 // CHECK: insertelement {{.*, i32 3 *$}}
|
|
38 // CHECK: insertelement {{.*, i32 4 *$}}
|
|
39 // CHECK: insertelement {{.*, i32 5 *$}}
|
|
40 // CHECK: insertelement {{.*, i32 6 *$}}
|
|
41 // CHECK: insertelement {{.*, i32 7 *$}}
|
|
42 }
|