volk_32fc_convert_16ic.h

Go to the documentation of this file.

/* -*- c++ -*- */
/*
 * Copyright 2016 Free Software Foundation, Inc.
 *
 * This file is part of VOLK
 *
 * SPDX-License-Identifier: LGPL-3.0-or-later
 */
 
#ifndef INCLUDED_volk_32fc_convert_16ic_a_H
#define INCLUDED_volk_32fc_convert_16ic_a_H
 
#include "volk/volk_complex.h"
#include <limits.h>
#include <math.h>
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32fc_convert_16ic_a_avx2(lv_16sc_t* outputVector,
                                                 const lv_32fc_t* inputVector,
                                                 unsigned int num_points)
{
    const unsigned int avx_iters = num_points / 8;
 
    float* inputVectorPtr = (float*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
    float aux;
 
    const float min_val = (float)SHRT_MIN;
    const float max_val = (float)SHRT_MAX;
 
    __m256 inputVal1, inputVal2;
    __m256i intInputVal1, intInputVal2;
    __m256 ret1, ret2;
    const __m256 vmin_val = _mm256_set1_ps(min_val);
    const __m256 vmax_val = _mm256_set1_ps(max_val);
    unsigned int i;
 
    for (i = 0; i < avx_iters; i++) {
        inputVal1 = _mm256_load_ps((float*)inputVectorPtr);
        inputVectorPtr += 8;
        inputVal2 = _mm256_load_ps((float*)inputVectorPtr);
        inputVectorPtr += 8;
        __VOLK_PREFETCH(inputVectorPtr + 16);
 
        // Clip
        ret1 = _mm256_max_ps(_mm256_min_ps(inputVal1, vmax_val), vmin_val);
        ret2 = _mm256_max_ps(_mm256_min_ps(inputVal2, vmax_val), vmin_val);
 
        intInputVal1 = _mm256_cvtps_epi32(ret1);
        intInputVal2 = _mm256_cvtps_epi32(ret2);
 
        intInputVal1 = _mm256_packs_epi32(intInputVal1, intInputVal2);
        intInputVal1 = _mm256_permute4x64_epi64(intInputVal1, 0xd8);
 
        _mm256_store_si256((__m256i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 16;
    }
 
    for (i = avx_iters * 16; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val)
            aux = max_val;
        else if (aux < min_val)
            aux = min_val;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_AVX2 */
 
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
 
static inline void volk_32fc_convert_16ic_a_sse2(lv_16sc_t* outputVector,
                                                 const lv_32fc_t* inputVector,
                                                 unsigned int num_points)
{
    const unsigned int sse_iters = num_points / 4;
 
    float* inputVectorPtr = (float*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
    float aux;
 
    const float min_val = (float)SHRT_MIN;
    const float max_val = (float)SHRT_MAX;
 
    __m128 inputVal1, inputVal2;
    __m128i intInputVal1, intInputVal2;
    __m128 ret1, ret2;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
    unsigned int i;
 
    for (i = 0; i < sse_iters; i++) {
        inputVal1 = _mm_load_ps((float*)inputVectorPtr);
        inputVectorPtr += 4;
        inputVal2 = _mm_load_ps((float*)inputVectorPtr);
        inputVectorPtr += 4;
        __VOLK_PREFETCH(inputVectorPtr + 8);
 
        // Clip
        ret1 = _mm_max_ps(_mm_min_ps(inputVal1, vmax_val), vmin_val);
        ret2 = _mm_max_ps(_mm_min_ps(inputVal2, vmax_val), vmin_val);
 
        intInputVal1 = _mm_cvtps_epi32(ret1);
        intInputVal2 = _mm_cvtps_epi32(ret2);
 
        intInputVal1 = _mm_packs_epi32(intInputVal1, intInputVal2);
 
        _mm_store_si128((__m128i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 8;
    }
 
    for (i = sse_iters * 8; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val)
            aux = max_val;
        else if (aux < min_val)
            aux = min_val;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_SSE2 */
 
 
#if LV_HAVE_NEONV7
#include <arm_neon.h>
 
#define VCVTRQ_S32_F32(result, value)                                       \
    __VOLK_ASM("VCVTR.S32.F32 %0, %1" : "=t"(result[0]) : "t"(value[0]) :); \
    __VOLK_ASM("VCVTR.S32.F32 %0, %1" : "=t"(result[1]) : "t"(value[1]) :); \
    __VOLK_ASM("VCVTR.S32.F32 %0, %1" : "=t"(result[2]) : "t"(value[2]) :); \
    __VOLK_ASM("VCVTR.S32.F32 %0, %1" : "=t"(result[3]) : "t"(value[3]) :);
 
static inline void volk_32fc_convert_16ic_neon(lv_16sc_t* outputVector,
                                               const lv_32fc_t* inputVector,
                                               unsigned int num_points)
{
 
    const unsigned int neon_iters = num_points / 4;
 
    float32_t* inputVectorPtr = (float32_t*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
 
    const float min_val_f = (float)SHRT_MIN;
    const float max_val_f = (float)SHRT_MAX;
    float32_t aux;
    unsigned int i;
 
    const float32x4_t min_val = vmovq_n_f32(min_val_f);
    const float32x4_t max_val = vmovq_n_f32(max_val_f);
    float32x4_t ret1, ret2, a, b;
 
    int32x4_t toint_a = { 0, 0, 0, 0 };
    int32x4_t toint_b = { 0, 0, 0, 0 };
    int16x4_t intInputVal1, intInputVal2;
    int16x8_t res;
 
    for (i = 0; i < neon_iters; i++) {
        a = vld1q_f32((const float32_t*)(inputVectorPtr));
        inputVectorPtr += 4;
        b = vld1q_f32((const float32_t*)(inputVectorPtr));
        inputVectorPtr += 4;
        __VOLK_PREFETCH(inputVectorPtr + 8);
 
        ret1 = vmaxq_f32(vminq_f32(a, max_val), min_val);
        ret2 = vmaxq_f32(vminq_f32(b, max_val), min_val);
 
        // vcvtr takes into account the current rounding mode (as does rintf)
        VCVTRQ_S32_F32(toint_a, ret1);
        VCVTRQ_S32_F32(toint_b, ret2);
 
        intInputVal1 = vqmovn_s32(toint_a);
        intInputVal2 = vqmovn_s32(toint_b);
 
        res = vcombine_s16(intInputVal1, intInputVal2);
        vst1q_s16((int16_t*)outputVectorPtr, res);
        outputVectorPtr += 8;
    }
 
    for (i = neon_iters * 8; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val_f)
            aux = max_val_f;
        else if (aux < min_val_f)
            aux = min_val_f;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
 
#undef VCVTRQ_S32_F32
#endif /* LV_HAVE_NEONV7 */
 
#if LV_HAVE_NEONV8
#include <arm_neon.h>
 
static inline void volk_32fc_convert_16ic_neonv8(lv_16sc_t* outputVector,
                                                 const lv_32fc_t* inputVector,
                                                 unsigned int num_points)
{
    const unsigned int neon_iters = num_points / 4;
 
    float32_t* inputVectorPtr = (float32_t*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
 
    const float min_val_f = (float)SHRT_MIN;
    const float max_val_f = (float)SHRT_MAX;
    float32_t aux;
    unsigned int i;
 
    const float32x4_t min_val = vmovq_n_f32(min_val_f);
    const float32x4_t max_val = vmovq_n_f32(max_val_f);
    float32x4_t ret1, ret2, a, b;
 
    int32x4_t toint_a = { 0, 0, 0, 0 }, toint_b = { 0, 0, 0, 0 };
    int16x4_t intInputVal1, intInputVal2;
    int16x8_t res;
 
    for (i = 0; i < neon_iters; i++) {
        a = vld1q_f32((const float32_t*)(inputVectorPtr));
        inputVectorPtr += 4;
        b = vld1q_f32((const float32_t*)(inputVectorPtr));
        inputVectorPtr += 4;
        __VOLK_PREFETCH(inputVectorPtr + 8);
 
        ret1 = vmaxq_f32(vminq_f32(a, max_val), min_val);
        ret2 = vmaxq_f32(vminq_f32(b, max_val), min_val);
 
        // vrndiq takes into account the current rounding mode (as does rintf)
        toint_a = vcvtq_s32_f32(vrndiq_f32(ret1));
        toint_b = vcvtq_s32_f32(vrndiq_f32(ret2));
 
        intInputVal1 = vqmovn_s32(toint_a);
        intInputVal2 = vqmovn_s32(toint_b);
 
        res = vcombine_s16(intInputVal1, intInputVal2);
        vst1q_s16((int16_t*)outputVectorPtr, res);
        outputVectorPtr += 8;
    }
 
    for (i = neon_iters * 8; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val_f)
            aux = max_val_f;
        else if (aux < min_val_f)
            aux = min_val_f;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_NEONV8 */
 
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32fc_convert_16ic_generic(lv_16sc_t* outputVector,
                                                  const lv_32fc_t* inputVector,
                                                  unsigned int num_points)
{
    float* inputVectorPtr = (float*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
    const float min_val = (float)SHRT_MIN;
    const float max_val = (float)SHRT_MAX;
    float aux;
    unsigned int i;
    for (i = 0; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val)
            aux = max_val;
        else if (aux < min_val)
            aux = min_val;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_GENERIC */
 
#endif /* INCLUDED_volk_32fc_convert_16ic_a_H */
 
#ifndef INCLUDED_volk_32fc_convert_16ic_u_H
#define INCLUDED_volk_32fc_convert_16ic_u_H
 
#include "volk/volk_complex.h"
#include <limits.h>
#include <math.h>
 
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32fc_convert_16ic_u_avx2(lv_16sc_t* outputVector,
                                                 const lv_32fc_t* inputVector,
                                                 unsigned int num_points)
{
    const unsigned int avx_iters = num_points / 8;
 
    float* inputVectorPtr = (float*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
    float aux;
 
    const float min_val = (float)SHRT_MIN;
    const float max_val = (float)SHRT_MAX;
 
    __m256 inputVal1, inputVal2;
    __m256i intInputVal1, intInputVal2;
    __m256 ret1, ret2;
    const __m256 vmin_val = _mm256_set1_ps(min_val);
    const __m256 vmax_val = _mm256_set1_ps(max_val);
    unsigned int i;
 
    for (i = 0; i < avx_iters; i++) {
        inputVal1 = _mm256_loadu_ps((float*)inputVectorPtr);
        inputVectorPtr += 8;
        inputVal2 = _mm256_loadu_ps((float*)inputVectorPtr);
        inputVectorPtr += 8;
        __VOLK_PREFETCH(inputVectorPtr + 16);
 
        // Clip
        ret1 = _mm256_max_ps(_mm256_min_ps(inputVal1, vmax_val), vmin_val);
        ret2 = _mm256_max_ps(_mm256_min_ps(inputVal2, vmax_val), vmin_val);
 
        intInputVal1 = _mm256_cvtps_epi32(ret1);
        intInputVal2 = _mm256_cvtps_epi32(ret2);
 
        intInputVal1 = _mm256_packs_epi32(intInputVal1, intInputVal2);
        intInputVal1 = _mm256_permute4x64_epi64(intInputVal1, 0xd8);
 
        _mm256_storeu_si256((__m256i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 16;
    }
 
    for (i = avx_iters * 16; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val)
            aux = max_val;
        else if (aux < min_val)
            aux = min_val;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_AVX2 */
 
 
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
 
static inline void volk_32fc_convert_16ic_u_sse2(lv_16sc_t* outputVector,
                                                 const lv_32fc_t* inputVector,
                                                 unsigned int num_points)
{
    const unsigned int sse_iters = num_points / 4;
 
    float* inputVectorPtr = (float*)inputVector;
    int16_t* outputVectorPtr = (int16_t*)outputVector;
    float aux;
 
    const float min_val = (float)SHRT_MIN;
    const float max_val = (float)SHRT_MAX;
 
    __m128 inputVal1, inputVal2;
    __m128i intInputVal1, intInputVal2;
    __m128 ret1, ret2;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
 
    unsigned int i;
    for (i = 0; i < sse_iters; i++) {
        inputVal1 = _mm_loadu_ps((float*)inputVectorPtr);
        inputVectorPtr += 4;
        inputVal2 = _mm_loadu_ps((float*)inputVectorPtr);
        inputVectorPtr += 4;
        __VOLK_PREFETCH(inputVectorPtr + 8);
 
        // Clip
        ret1 = _mm_max_ps(_mm_min_ps(inputVal1, vmax_val), vmin_val);
        ret2 = _mm_max_ps(_mm_min_ps(inputVal2, vmax_val), vmin_val);
 
        intInputVal1 = _mm_cvtps_epi32(ret1);
        intInputVal2 = _mm_cvtps_epi32(ret2);
 
        intInputVal1 = _mm_packs_epi32(intInputVal1, intInputVal2);
 
        _mm_storeu_si128((__m128i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 8;
    }
 
    for (i = sse_iters * 8; i < num_points * 2; i++) {
        aux = *inputVectorPtr++;
        if (aux > max_val)
            aux = max_val;
        else if (aux < min_val)
            aux = min_val;
        *outputVectorPtr++ = (int16_t)rintf(aux);
    }
}
#endif /* LV_HAVE_SSE2 */
#endif /* INCLUDED_volk_32fc_convert_16ic_u_H */