#ifndef SCANN_UTILS_INTRINSICS_TRANSPOSE_NEON_H_
#define SCANN_UTILS_INTRINSICS_TRANSPOSE_NEON_H_
#ifdef __aarch64__
#include <arm_neon.h>
namespace research_scann {
namespace neon {
static inline void transpose_f32_4x4(const float32x4_t s0, const float32x4_t s1,
const float32x4_t s2, const float32x4_t s3,
float32x4_t* d0, float32x4_t* d1,
float32x4_t* d2, float32x4_t* d3) {
const float32x4_t a0 = vzip1q_f32(s0, s2);
const float32x4_t a1 = vzip2q_f32(s0, s2);
const float32x4_t a2 = vzip1q_f32(s1, s3);
const float32x4_t a3 = vzip2q_f32(s1, s3);
*d0 = vzip1q_f32(a0, a2);
*d1 = vzip2q_f32(a0, a2);
*d2 = vzip1q_f32(a1, a3);
*d3 = vzip2q_f32(a1, a3);
}
} }
#endif
#endif