clang/test/CodeGen/arm64_vdup.c

   1 // RUN: %clang_cc1 -triple arm64-apple-ios7 -target-feature +neon -ffreestanding -o - -emit-llvm %s | FileCheck %s
   2 // Test ARM64 SIMD duplicate lane and n intrinsics
   3
   4 // REQUIRES: aarch64-registered-target || arm-registered-target
   5
   6 #include <arm_neon.h>
   7
   8 void test_vdup_lane_s64(int64x1_t a1) {
   9   // CHECK-LABEL: test_vdup_lane_s64
  10   vdup_lane_s64(a1, 0);
  11   // CHECK: shufflevector
  12 }
  13
  14 void test_vdup_lane_u64(uint64x1_t a1) {
  15   // CHECK-LABEL: test_vdup_lane_u64
  16   vdup_lane_u64(a1, 0);
  17   // CHECK: shufflevector
  18 }
  19
  20 // uncomment out the following code once scalar_to_vector in the backend
  21 // works (for 64 bit?).  Change the "CHECK@" to "CHECK<colon>"
  22 /*
  23 float64x1_t test_vdup_n_f64(float64_t a1) {
  24   // CHECK-LABEL@ test_vdup_n_f64
  25   return vdup_n_f64(a1);
  26   // match that an element is inserted into part 0
  27   // CHECK@ insertelement {{.*, i32 0 *$}}
  28 }
  29 */
  30
  31 float16x8_t test_vdupq_n_f16(float16_t *a1) {
  32   // CHECK-LABEL: test_vdupq_n_f16
  33   return vdupq_n_f16(*a1);
  34   // match that an element is inserted into parts 0-7.  The backend better
  35   // turn that into a single dup instruction
  36   // CHECK: insertelement {{.*, i32 0 *$}}
  37   // CHECK: insertelement {{.*, i32 1 *$}}
  38   // CHECK: insertelement {{.*, i32 2 *$}}
  39   // CHECK: insertelement {{.*, i32 3 *$}}
  40   // CHECK: insertelement {{.*, i32 4 *$}}
  41   // CHECK: insertelement {{.*, i32 5 *$}}
  42   // CHECK: insertelement {{.*, i32 6 *$}}
  43   // CHECK: insertelement {{.*, i32 7 *$}}
  44 }