volk_16i_max_star_horizontal_16i.h

Go to the documentation of this file.

/* -*- c++ -*- */
/*
 * Copyright 2012, 2014 Free Software Foundation, Inc.
 *
 * This file is part of VOLK
 *
 * SPDX-License-Identifier: LGPL-3.0-or-later
 */
 
#ifndef INCLUDED_volk_16i_max_star_horizontal_16i_a_H
#define INCLUDED_volk_16i_max_star_horizontal_16i_a_H
 
#include <volk/volk_common.h>
 
#include <inttypes.h>
#include <stdio.h>
 
 
#ifdef LV_HAVE_SSSE3
 
#include <emmintrin.h>
#include <tmmintrin.h>
#include <xmmintrin.h>
 
static inline void volk_16i_max_star_horizontal_16i_a_ssse3(int16_t* target,
                                                            int16_t* src0,
                                                            unsigned int num_points)
{
    const unsigned int num_bytes = num_points * 2;
 
    static const uint8_t shufmask0[16] = {
        0x00, 0x01, 0x04, 0x05, 0x08, 0x09, 0x0c, 0x0d,
        0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
    };
    static const uint8_t shufmask1[16] = {
        0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
        0x00, 0x01, 0x04, 0x05, 0x08, 0x09, 0x0c, 0x0d
    };
    static const uint8_t andmask0[16] = {
        0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
    };
    static const uint8_t andmask1[16] = {
        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
        0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02
    };
 
    __m128i xmm0, xmm1, xmm2, xmm3, xmm4;
    __m128i xmm5, xmm6, xmm7, xmm8;
 
    xmm4 = _mm_load_si128((__m128i*)shufmask0);
    xmm5 = _mm_load_si128((__m128i*)shufmask1);
    xmm6 = _mm_load_si128((__m128i*)andmask0);
    xmm7 = _mm_load_si128((__m128i*)andmask1);
 
    __m128i *p_target, *p_src0;
 
    p_target = (__m128i*)target;
    p_src0 = (__m128i*)src0;
 
    int bound = num_bytes >> 5;
    int intermediate = (num_bytes >> 4) & 1;
    int leftovers = (num_bytes >> 1) & 7;
 
    int i = 0;
 
    for (i = 0; i < bound; ++i) {
        xmm0 = _mm_load_si128(p_src0);
        xmm1 = _mm_load_si128(&p_src0[1]);
 
        xmm2 = _mm_xor_si128(xmm2, xmm2);
        p_src0 += 2;
 
        xmm3 = _mm_hsub_epi16(xmm0, xmm1);
 
        xmm2 = _mm_cmpgt_epi16(xmm2, xmm3);
 
        xmm8 = _mm_and_si128(xmm2, xmm6);
        xmm3 = _mm_and_si128(xmm2, xmm7);
 
 
        xmm8 = _mm_add_epi8(xmm8, xmm4);
        xmm3 = _mm_add_epi8(xmm3, xmm5);
 
        xmm0 = _mm_shuffle_epi8(xmm0, xmm8);
        xmm1 = _mm_shuffle_epi8(xmm1, xmm3);
 
 
        xmm3 = _mm_add_epi16(xmm0, xmm1);
 
 
        _mm_store_si128(p_target, xmm3);
 
        p_target += 1;
    }
 
    if (intermediate) {
        xmm0 = _mm_load_si128(p_src0);
 
        xmm2 = _mm_xor_si128(xmm2, xmm2);
        p_src0 += 1;
 
        xmm3 = _mm_hsub_epi16(xmm0, xmm1);
        xmm2 = _mm_cmpgt_epi16(xmm2, xmm3);
 
        xmm8 = _mm_and_si128(xmm2, xmm6);
 
        xmm3 = _mm_add_epi8(xmm8, xmm4);
 
        xmm0 = _mm_shuffle_epi8(xmm0, xmm3);
 
        _mm_storel_pd((double*)p_target, bit128_p(&xmm0)->double_vec);
 
        p_target = (__m128i*)((int8_t*)p_target + 8);
    }
 
    for (i = (bound << 4) + (intermediate << 3);
         i < (bound << 4) + (intermediate << 3) + leftovers;
         i += 2) {
        target[i >> 1] = ((int16_t)(src0[i] - src0[i + 1]) > 0) ? src0[i] : src0[i + 1];
    }
}
 
#endif /*LV_HAVE_SSSE3*/
 
#ifdef LV_HAVE_NEON
 
#include <arm_neon.h>
static inline void volk_16i_max_star_horizontal_16i_neon(int16_t* target,
                                                         int16_t* src0,
                                                         unsigned int num_points)
{
    const unsigned int eighth_points = num_points / 16;
    unsigned number;
    int16x8x2_t input_vec;
    int16x8_t diff, max_vec, zeros;
    uint16x8_t comp1, comp2;
    zeros = vdupq_n_s16(0);
    for (number = 0; number < eighth_points; ++number) {
        input_vec = vld2q_s16(src0);
        //__VOLK_PREFETCH(src0+16);
        diff = vsubq_s16(input_vec.val[0], input_vec.val[1]);
        comp1 = vcgeq_s16(diff, zeros);
        comp2 = vcltq_s16(diff, zeros);
 
        input_vec.val[0] = vandq_s16(input_vec.val[0], (int16x8_t)comp1);
        input_vec.val[1] = vandq_s16(input_vec.val[1], (int16x8_t)comp2);
 
        max_vec = vaddq_s16(input_vec.val[0], input_vec.val[1]);
        vst1q_s16(target, max_vec);
        src0 += 16;
        target += 8;
    }
    for (number = 0; number < num_points % 16; number += 2) {
        target[number >> 1] = ((int16_t)(src0[number] - src0[number + 1]) > 0)
                                  ? src0[number]
                                  : src0[number + 1];
    }
}
#endif /* LV_HAVE_NEON */
 
#ifdef LV_HAVE_NEONV7
extern void volk_16i_max_star_horizontal_16i_a_neonasm(int16_t* target,
                                                       int16_t* src0,
                                                       unsigned int num_points);
#endif /* LV_HAVE_NEONV7 */
 
#ifdef LV_HAVE_GENERIC
static inline void volk_16i_max_star_horizontal_16i_generic(int16_t* target,
                                                            int16_t* src0,
                                                            unsigned int num_points)
{
    const unsigned int num_bytes = num_points * 2;
 
    int i = 0;
 
    int bound = num_bytes >> 1;
 
    for (i = 0; i < bound; i += 2) {
        target[i >> 1] = ((int16_t)(src0[i] - src0[i + 1]) > 0) ? src0[i] : src0[i + 1];
    }
}
 
#endif /*LV_HAVE_GENERIC*/
 
#endif /*INCLUDED_volk_16i_max_star_horizontal_16i_a_H*/