clang/test/Sema/aarch64-neon-target.c

   1 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +neon -fsyntax-only -verify -emit-llvm -o - %s
   2 // REQUIRES: aarch64-registered-target
   3
   4 // Test that functions with the correct target attributes can use the correct NEON intrinsics.
   5
   6 #include <arm_neon.h>
   7
   8 __attribute__((target("dotprod")))
   9 void dotprod(uint32x2_t v2i32, uint8x16_t v16i8, uint8x8_t v8i8) {
  10   vdot_u32(v2i32, v8i8, v8i8);
  11   vdot_laneq_u32(v2i32, v8i8, v16i8, 1);
  12 }
  13
  14 __attribute__((target("fullfp16")))
  15 void fp16(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8, uint8x8_t v8i8, float32x2_t v2f32, float32x4_t v4f32, float16x4_t v4f16, bfloat16x4_t v4bf16) {
  16   vceqz_f16(v4f16);
  17   vrnd_f16(v4f16);
  18   vmaxnm_f16(v4f16, v4f16);
  19   vrndi_f16(v4f16);
  20 }
  21
  22 __attribute__((target("fp16fml")))
  23 void fp16fml(float32x2_t v2f32, float16x4_t v4f16) {
  24   vfmlal_low_f16(v2f32, v4f16, v4f16);
  25 }
  26
  27 __attribute__((target("i8mm")))
  28 void i8mm(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8, uint8x8_t v8i8, float32x2_t v2f32, float32x4_t v4f32, float16x4_t v4f16, bfloat16x4_t v4bf16) {
  29   vmmlaq_s32(v4i32, v8i16, v8i16);
  30   vusdot_laneq_s32(v2i32, v8i8, v8i16, 0);
  31 }
  32
  33 __attribute__((target("bf16")))
  34 void bf16(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8, uint8x8_t v8i8, float32x2_t v2f32, float32x4_t v4f32, float16x4_t v4f16, bfloat16x4_t v4bf16, __bf16 bf16) {
  35   vbfdot_f32(v2f32, v4bf16, v4bf16);
  36   vcreate_bf16(10);
  37   vdup_lane_bf16(v4bf16, 2);
  38   vdup_n_bf16(bf16);
  39   vld1_bf16(0);
  40   vcvt_f32_bf16(v4bf16);
  41   vcvt_bf16_f32(v4f32);
  42 }
  43
  44 __attribute__((target("arch=armv8.1-a")))
  45 void test_v81(int32x2_t d, int32x4_t v, int s) {
  46   vqrdmlahq_s32(v, v, v);
  47   vqrdmlah_laneq_s32(d, d, v, 1);
  48   vqrdmlahh_s16(1, 1, 1);
  49 }
  50
  51 __attribute__((target("arch=armv8.3-a+fp16")))
  52 void test_v83(float32x4_t v4f32, float16x4_t v4f16, float64x2_t v2f64) {
  53   vcaddq_rot90_f32(v4f32, v4f32);
  54   vcmla_rot90_f16(v4f16, v4f16, v4f16);
  55   vcmlaq_rot270_laneq_f64(v2f64, v2f64, v2f64, 1);
  56 }
  57
  58 __attribute__((target("arch=armv8.5-a")))
  59 void test_v85(float32x4_t v4f32) {
  60   vrnd32xq_f32(v4f32);
  61 }
  62
  63 void undefined(uint32x2_t v2i32, uint32x4_t v4i32, uint16x8_t v8i16, uint8x16_t v16i8, uint8x8_t v8i8, float32x2_t v2f32, float32x4_t v4f32, float16x4_t v4f16, float64x2_t v2f64, bfloat16x4_t v4bf16, __bf16 bf16, poly64_t poly64, poly64x2_t poly64x2) {
  64   // dotprod
  65   vdot_u32(v2i32, v8i8, v8i8); // expected-error {{always_inline function 'vdot_u32' requires target feature 'dotprod'}}
  66   vdot_laneq_u32(v2i32, v8i8, v16i8, 1); // expected-error {{always_inline function 'vdot_u32' requires target feature 'dotprod'}}
  67   // fp16
  68   vceqz_f16(v4f16); // expected-error {{always_inline function 'vceqz_f16' requires target feature 'fullfp16'}}
  69   vrnd_f16(v4f16); // expected-error {{always_inline function 'vrnd_f16' requires target feature 'fullfp16'}}
  70   vmaxnm_f16(v4f16, v4f16); // expected-error {{always_inline function 'vmaxnm_f16' requires target feature 'fullfp16'}}
  71   vrndi_f16(v4f16); // expected-error {{always_inline function 'vrndi_f16' requires target feature 'fullfp16'}}
  72   // fp16fml
  73   vfmlal_low_f16(v2f32, v4f16, v4f16); // expected-error {{always_inline function 'vfmlal_low_f16' requires target feature 'fp16fml'}}
  74   // i8mm
  75   vmmlaq_s32(v4i32, v8i16, v8i16); // expected-error {{always_inline function 'vmmlaq_s32' requires target feature 'i8mm'}}
  76   vusdot_laneq_s32(v2i32, v8i8, v8i16, 0); // expected-error {{always_inline function 'vusdot_s32' requires target feature 'i8mm'}}
  77   // bf16
  78   vbfdot_f32(v2f32, v4bf16, v4bf16); // expected-error {{always_inline function 'vbfdot_f32' requires target feature 'bf16'}}
  79   vcreate_bf16(10);
  80   vdup_lane_bf16(v4bf16, 2); // expected-error {{'__builtin_neon_splat_lane_bf16' needs target feature bf16}}
  81   vdup_n_bf16(bf16); // expected-error {{always_inline function 'vdup_n_bf16' requires target feature 'bf16'}}
  82   vld1_bf16(0); // expected-error {{'__builtin_neon_vld1_bf16' needs target feature bf16}}
  83   vcvt_f32_bf16(v4bf16); // expected-error {{always_inline function 'vcvt_f32_bf16' requires target feature 'bf16'}}
  84   vcvt_bf16_f32(v4f32); // expected-error {{always_inline function 'vcvt_bf16_f32' requires target feature 'bf16'}}
  85   // v8.1 - qrdmla
  86   vqrdmlahq_s32(v4i32, v4i32, v4i32); // expected-error {{always_inline function 'vqrdmlahq_s32' requires target feature 'v8.1a'}}
  87   vqrdmlah_laneq_s32(v2i32, v2i32, v4i32, 1); // expected-error {{always_inline function 'vqrdmlah_s32' requires target feature 'v8.1a'}}
  88   vqrdmlahh_s16(1, 1, 1); // expected-error {{always_inline function 'vqrdmlahh_s16' requires target feature 'v8.1a'}}
  89   // 8.3 - complex
  90   vcaddq_rot90_f32(v4f32, v4f32); // expected-error {{always_inline function 'vcaddq_rot90_f32' requires target feature 'v8.3a'}}
  91   vcmla_rot90_f16(v4f16, v4f16, v4f16); // expected-error {{always_inline function 'vcmla_rot90_f16' requires target feature 'v8.3a'}}
  92   vcmlaq_rot270_laneq_f64(v2f64, v2f64, v2f64, 1); // expected-error {{always_inline function 'vcmlaq_rot270_f64' requires target feature 'v8.3a'}}
  93   // 8.5 - frint
  94   vrnd32xq_f32(v4f32); // expected-error {{always_inline function 'vrnd32xq_f32' requires target feature 'v8.5a'}}
  95
  96   vmull_p64(poly64, poly64);  // expected-error {{always_inline function 'vmull_p64' requires target feature 'aes'}}
  97   vmull_high_p64(poly64x2, poly64x2);  // expected-error {{always_inline function 'vmull_high_p64' requires target feature 'aes'}}
  98
  99 }