2#include "arm_simd/shared/a32_float.hpp"
6#define nce [[gnu::always_inline]] inline
10template <
int n>[[gnu::always_inline]] nce float32x4_t convert(int32x4_t a) {
return vcvtq_n_f32_s32(a, n); }
11template <
int n>[[gnu::always_inline]] nce float32x4_t convert(uint32x4_t a) {
return vcvtq_n_f32_u32(a, n); }
12[[gnu::always_inline]] nce float16x8_t reverse_32bit(float16x8_t a) {
return vrev32q_f16(a); }
13[[gnu::always_inline]] nce float16x8_t reverse_64bit(float16x8_t a) {
return vrev64q_f16(a); }
14[[gnu::always_inline]] nce float16x8_t min_strict(float16x8_t a, float16x8_t b) {
return vminnmq_f16(a, b); }
15[[gnu::always_inline]] nce float16x8_t max_strict(float16x8_t a, float16x8_t b) {
return vmaxnmq_f16(a, b); }
16[[gnu::always_inline]] nce float16x8_t abs(float16x8_t a) {
return vabsq_f16(a); }
17[[gnu::always_inline]] nce float16x8_t bitwise_clear(float16x8_t a, float16x8_t b) {
return vbicq_f16(a, b); }
18[[gnu::always_inline]] nce float16x8_t multiply(float16x8_t a, float16_t b) {
return vmulq_n_f16(a, b); }
19[[gnu::always_inline]] nce float16x8_t bitwise_and(float16x8_t a, float16x8_t b) {
return vandq_f16(a, b); }
20[[gnu::always_inline]] nce float16x8_t bitwise_xor(float16x8_t a, float16x8_t b) {
return veorq_f16(a, b); }
21[[gnu::always_inline]] nce float16x8_t complex_multiply_add(float16x8_t a, float16x8_t b, float16x8_t c) {
return vcmlaq_f16(a, b, c); }
22[[gnu::always_inline]] nce float16x8_t complex_multiply_add_rotate_90(float16x8_t a, float16x8_t b, float16x8_t c) {
return vcmlaq_rot90_f16(a, b, c); }
23[[gnu::always_inline]] nce float16x8_t complex_multiply_add_rotate_180(float16x8_t a, float16x8_t b, float16x8_t c) {
return vcmlaq_rot180_f16(a, b, c); }
24[[gnu::always_inline]] nce float16x8_t complex_multiply_add_rotate_270(float16x8_t a, float16x8_t b, float16x8_t c) {
return vcmlaq_rot270_f16(a, b, c); }
25[[gnu::always_inline]] nce float32x4_t bitwise_clear(float32x4_t a, float32x4_t b) {
return vbicq_f32(a, b); }
26[[gnu::always_inline]] nce float32x4_t bitwise_and(float32x4_t a, float32x4_t b) {
return vandq_f32(a, b); }
27[[gnu::always_inline]] nce float32x4_t bitwise_xor(float32x4_t a, float32x4_t b) {
return veorq_f32(a, b); }
28[[gnu::always_inline]] nce float32x4_t complex_multiply_add(float32x4_t a, float32x4_t b, float32x4_t c) {
return vcmlaq_f32(a, b, c); }
29[[gnu::always_inline]] nce float32x4_t complex_multiply_add_rotate_90(float32x4_t a, float32x4_t b, float32x4_t c) {
return vcmlaq_rot90_f32(a, b, c); }
30[[gnu::always_inline]] nce float32x4_t complex_multiply_add_rotate_180(float32x4_t a, float32x4_t b, float32x4_t c) {
return vcmlaq_rot180_f32(a, b, c); }
31[[gnu::always_inline]] nce float32x4_t complex_multiply_add_rotate_270(float32x4_t a, float32x4_t b, float32x4_t c) {
return vcmlaq_rot270_f32(a, b, c); }
32template <
int n>[[gnu::always_inline]] nce int32x4_t convert_n_signed(float32x4_t a) {
return vcvtq_n_s32_f32(a, n); }
33template <
int n>[[gnu::always_inline]] nce uint32x4_t convert_n_unsigned(float32x4_t a) {
return vcvtq_n_u32_f32(a, n); }
34[[gnu::always_inline]] nce float16x8_t duplicate(float16_t a) {
return vdupq_n_f16(a); }
35[[gnu::always_inline]] nce float32x4_t duplicate(float32_t a) {
return vdupq_n_f32(a); }
36[[gnu::always_inline]]
inline float16x8_t load1(float16_t
const *ptr) {
return vld1q_f16(ptr); }
37[[gnu::always_inline]]
inline float32x4_t load1(float32_t
const *ptr) {
return vld1q_f32(ptr); }
38[[gnu::always_inline]]
inline void store1(float16_t *ptr, float16x8_t val) {
return vst1q_f16(ptr, val); }
1.14.0