volk_32f_s32f_convert_8i.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_32f_s32f_convert_8i_u_H
#define INCLUDED_volk_32f_s32f_convert_8i_u_H
 
#include <inttypes.h>
 
static inline void volk_32f_s32f_convert_8i_single(int8_t* out, const float in)
{
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    if (in > max_val) {
        *out = (int8_t)(max_val);
    } else if (in < min_val) {
        *out = (int8_t)(min_val);
    } else {
        *out = (int8_t)(rintf(in));
    }
}
 
#ifdef LV_HAVE_GENERIC
 
static inline void volk_32f_s32f_convert_8i_generic(int8_t* outputVector,
                                                    const float* inputVector,
                                                    const float scalar,
                                                    unsigned int num_points)
{
    const float* inputVectorPtr = inputVector;
 
    for (unsigned int number = 0; number < num_points; number++) {
        const float r = *inputVectorPtr++ * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_GENERIC */
 
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32f_s32f_convert_8i_u_avx2(int8_t* outputVector,
                                                   const float* inputVector,
                                                   const float scalar,
                                                   unsigned int num_points)
{
    const unsigned int thirtysecondPoints = num_points / 32;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m256 vmin_val = _mm256_set1_ps(min_val);
    const __m256 vmax_val = _mm256_set1_ps(max_val);
 
    const __m256 vScalar = _mm256_set1_ps(scalar);
 
    for (unsigned int number = 0; number < thirtysecondPoints; number++) {
        __m256 inputVal1 = _mm256_loadu_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal2 = _mm256_loadu_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal3 = _mm256_loadu_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal4 = _mm256_loadu_ps(inputVectorPtr);
        inputVectorPtr += 8;
 
        inputVal1 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
        inputVal3 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal3, vScalar), vmax_val), vmin_val);
        inputVal4 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal4, vScalar), vmax_val), vmin_val);
 
        __m256i intInputVal1 = _mm256_cvtps_epi32(inputVal1);
        __m256i intInputVal2 = _mm256_cvtps_epi32(inputVal2);
        __m256i intInputVal3 = _mm256_cvtps_epi32(inputVal3);
        __m256i intInputVal4 = _mm256_cvtps_epi32(inputVal4);
 
        intInputVal1 = _mm256_packs_epi32(intInputVal1, intInputVal2);
        intInputVal1 = _mm256_permute4x64_epi64(intInputVal1, 0b11011000);
        intInputVal3 = _mm256_packs_epi32(intInputVal3, intInputVal4);
        intInputVal3 = _mm256_permute4x64_epi64(intInputVal3, 0b11011000);
 
        intInputVal1 = _mm256_packs_epi16(intInputVal1, intInputVal3);
        const __m256i intInputVal = _mm256_permute4x64_epi64(intInputVal1, 0b11011000);
 
        _mm256_storeu_si256((__m256i*)outputVectorPtr, intInputVal);
        outputVectorPtr += 32;
    }
 
    for (unsigned int number = thirtysecondPoints * 32; number < num_points; number++) {
        float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_AVX2 */
 
#ifdef LV_HAVE_AVX512F
#include <immintrin.h>
 
static inline void volk_32f_s32f_convert_8i_u_avx512(int8_t* outputVector,
                                                     const float* inputVector,
                                                     const float scalar,
                                                     unsigned int num_points)
{
    unsigned int number = 0;
 
    const unsigned int thirtysecondPoints = num_points / 32;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    float min_val = INT8_MIN;
    float max_val = INT8_MAX;
    float r;
 
    __m512 vScalar = _mm512_set1_ps(scalar);
    __m512 inputVal1, inputVal2;
    __m512i intInputVal1, intInputVal2;
    __m512 vmin_val = _mm512_set1_ps(min_val);
    __m512 vmax_val = _mm512_set1_ps(max_val);
    __m128i packed_result;
 
    for (; number < thirtysecondPoints; number++) {
        inputVal1 = _mm512_loadu_ps(inputVectorPtr);
        inputVectorPtr += 16;
        inputVal2 = _mm512_loadu_ps(inputVectorPtr);
        inputVectorPtr += 16;
 
        inputVal1 = _mm512_max_ps(
            _mm512_min_ps(_mm512_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 = _mm512_max_ps(
            _mm512_min_ps(_mm512_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
 
        intInputVal1 = _mm512_cvtps_epi32(inputVal1);
        intInputVal2 = _mm512_cvtps_epi32(inputVal2);
 
        // Pack int32 -> int16 -> int8
        packed_result = _mm512_cvtsepi32_epi8(intInputVal1);
        _mm_storeu_si128((__m128i*)outputVectorPtr, packed_result);
        outputVectorPtr += 16;
 
        packed_result = _mm512_cvtsepi32_epi8(intInputVal2);
        _mm_storeu_si128((__m128i*)outputVectorPtr, packed_result);
        outputVectorPtr += 16;
    }
 
    number = thirtysecondPoints * 32;
    for (; number < num_points; number++) {
        r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_AVX512F */
 
 
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
 
static inline void volk_32f_s32f_convert_8i_u_sse2(int8_t* outputVector,
                                                   const float* inputVector,
                                                   const float scalar,
                                                   unsigned int num_points)
{
    const unsigned int sixteenthPoints = num_points / 16;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
 
    const __m128 vScalar = _mm_set_ps1(scalar);
 
    for (unsigned int number = 0; number < sixteenthPoints; number++) {
        __m128 inputVal1 = _mm_loadu_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal2 = _mm_loadu_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal3 = _mm_loadu_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal4 = _mm_loadu_ps(inputVectorPtr);
        inputVectorPtr += 4;
 
        inputVal1 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
        inputVal3 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal3, vScalar), vmax_val), vmin_val);
        inputVal4 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal4, vScalar), vmax_val), vmin_val);
 
        __m128i intInputVal1 = _mm_cvtps_epi32(inputVal1);
        __m128i intInputVal2 = _mm_cvtps_epi32(inputVal2);
        __m128i intInputVal3 = _mm_cvtps_epi32(inputVal3);
        __m128i intInputVal4 = _mm_cvtps_epi32(inputVal4);
 
        intInputVal1 = _mm_packs_epi32(intInputVal1, intInputVal2);
        intInputVal3 = _mm_packs_epi32(intInputVal3, intInputVal4);
 
        intInputVal1 = _mm_packs_epi16(intInputVal1, intInputVal3);
 
        _mm_storeu_si128((__m128i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 16;
    }
 
    for (unsigned int number = sixteenthPoints * 16; number < num_points; number++) {
        const float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_SSE2 */
 
 
#ifdef LV_HAVE_SSE
#include <xmmintrin.h>
 
static inline void volk_32f_s32f_convert_8i_u_sse(int8_t* outputVector,
                                                  const float* inputVector,
                                                  const float scalar,
                                                  unsigned int num_points)
{
    const unsigned int quarterPoints = num_points / 4;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
 
    const __m128 vScalar = _mm_set_ps1(scalar);
 
    __VOLK_ATTR_ALIGNED(16) float outputFloatBuffer[4];
 
    for (unsigned int number = 0; number < quarterPoints; number++) {
        __m128 ret = _mm_loadu_ps(inputVectorPtr);
        inputVectorPtr += 4;
 
        ret = _mm_max_ps(_mm_min_ps(_mm_mul_ps(ret, vScalar), vmax_val), vmin_val);
 
        _mm_store_ps(outputFloatBuffer, ret);
        for (size_t inner_loop = 0; inner_loop < 4; inner_loop++) {
            *outputVectorPtr++ = (int8_t)(rintf(outputFloatBuffer[inner_loop]));
        }
    }
 
    for (unsigned int number = quarterPoints * 4; number < num_points; number++) {
        const float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_SSE */
 
 
#endif /* INCLUDED_volk_32f_s32f_convert_8i_u_H */
#ifndef INCLUDED_volk_32f_s32f_convert_8i_a_H
#define INCLUDED_volk_32f_s32f_convert_8i_a_H
 
#include <inttypes.h>
 
#ifdef LV_HAVE_AVX2
#include <immintrin.h>
 
static inline void volk_32f_s32f_convert_8i_a_avx2(int8_t* outputVector,
                                                   const float* inputVector,
                                                   const float scalar,
                                                   unsigned int num_points)
{
    const unsigned int thirtysecondPoints = num_points / 32;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m256 vmin_val = _mm256_set1_ps(min_val);
    const __m256 vmax_val = _mm256_set1_ps(max_val);
 
    const __m256 vScalar = _mm256_set1_ps(scalar);
 
    for (unsigned int number = 0; number < thirtysecondPoints; number++) {
        __m256 inputVal1 = _mm256_load_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal2 = _mm256_load_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal3 = _mm256_load_ps(inputVectorPtr);
        inputVectorPtr += 8;
        __m256 inputVal4 = _mm256_load_ps(inputVectorPtr);
        inputVectorPtr += 8;
 
        inputVal1 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
        inputVal3 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal3, vScalar), vmax_val), vmin_val);
        inputVal4 = _mm256_max_ps(
            _mm256_min_ps(_mm256_mul_ps(inputVal4, vScalar), vmax_val), vmin_val);
 
        __m256i intInputVal1 = _mm256_cvtps_epi32(inputVal1);
        __m256i intInputVal2 = _mm256_cvtps_epi32(inputVal2);
        __m256i intInputVal3 = _mm256_cvtps_epi32(inputVal3);
        __m256i intInputVal4 = _mm256_cvtps_epi32(inputVal4);
 
        intInputVal1 = _mm256_packs_epi32(intInputVal1, intInputVal2);
        intInputVal1 = _mm256_permute4x64_epi64(intInputVal1, 0b11011000);
        intInputVal3 = _mm256_packs_epi32(intInputVal3, intInputVal4);
        intInputVal3 = _mm256_permute4x64_epi64(intInputVal3, 0b11011000);
 
        intInputVal1 = _mm256_packs_epi16(intInputVal1, intInputVal3);
        __m256i intInputVal = _mm256_permute4x64_epi64(intInputVal1, 0b11011000);
 
        _mm256_store_si256((__m256i*)outputVectorPtr, intInputVal);
        outputVectorPtr += 32;
    }
 
    for (unsigned int number = thirtysecondPoints * 32; number < num_points; number++) {
        const float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_AVX2 */
 
#ifdef LV_HAVE_AVX512F
#include <immintrin.h>
 
static inline void volk_32f_s32f_convert_8i_a_avx512(int8_t* outputVector,
                                                     const float* inputVector,
                                                     const float scalar,
                                                     unsigned int num_points)
{
    unsigned int number = 0;
 
    const unsigned int thirtysecondPoints = num_points / 32;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    float min_val = INT8_MIN;
    float max_val = INT8_MAX;
    float r;
 
    __m512 vScalar = _mm512_set1_ps(scalar);
    __m512 inputVal1, inputVal2;
    __m512i intInputVal1, intInputVal2;
    __m512 vmin_val = _mm512_set1_ps(min_val);
    __m512 vmax_val = _mm512_set1_ps(max_val);
    __m128i packed_result;
 
    for (; number < thirtysecondPoints; number++) {
        inputVal1 = _mm512_load_ps(inputVectorPtr);
        inputVectorPtr += 16;
        inputVal2 = _mm512_load_ps(inputVectorPtr);
        inputVectorPtr += 16;
 
        inputVal1 = _mm512_max_ps(
            _mm512_min_ps(_mm512_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 = _mm512_max_ps(
            _mm512_min_ps(_mm512_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
 
        intInputVal1 = _mm512_cvtps_epi32(inputVal1);
        intInputVal2 = _mm512_cvtps_epi32(inputVal2);
 
        // Pack int32 -> int16 -> int8
        packed_result = _mm512_cvtsepi32_epi8(intInputVal1);
        _mm_store_si128((__m128i*)outputVectorPtr, packed_result);
        outputVectorPtr += 16;
 
        packed_result = _mm512_cvtsepi32_epi8(intInputVal2);
        _mm_store_si128((__m128i*)outputVectorPtr, packed_result);
        outputVectorPtr += 16;
    }
 
    number = thirtysecondPoints * 32;
    for (; number < num_points; number++) {
        r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_AVX512F */
 
 
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
 
static inline void volk_32f_s32f_convert_8i_a_sse2(int8_t* outputVector,
                                                   const float* inputVector,
                                                   const float scalar,
                                                   unsigned int num_points)
{
    const unsigned int sixteenthPoints = num_points / 16;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
 
    const __m128 vScalar = _mm_set_ps1(scalar);
 
    for (unsigned int number = 0; number < sixteenthPoints; number++) {
        __m128 inputVal1 = _mm_load_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal2 = _mm_load_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal3 = _mm_load_ps(inputVectorPtr);
        inputVectorPtr += 4;
        __m128 inputVal4 = _mm_load_ps(inputVectorPtr);
        inputVectorPtr += 4;
 
        inputVal1 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
        inputVal2 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
        inputVal3 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal3, vScalar), vmax_val), vmin_val);
        inputVal4 =
            _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal4, vScalar), vmax_val), vmin_val);
 
        __m128i intInputVal1 = _mm_cvtps_epi32(inputVal1);
        __m128i intInputVal2 = _mm_cvtps_epi32(inputVal2);
        __m128i intInputVal3 = _mm_cvtps_epi32(inputVal3);
        __m128i intInputVal4 = _mm_cvtps_epi32(inputVal4);
 
        intInputVal1 = _mm_packs_epi32(intInputVal1, intInputVal2);
        intInputVal3 = _mm_packs_epi32(intInputVal3, intInputVal4);
 
        intInputVal1 = _mm_packs_epi16(intInputVal1, intInputVal3);
 
        _mm_store_si128((__m128i*)outputVectorPtr, intInputVal1);
        outputVectorPtr += 16;
    }
 
    for (unsigned int number = sixteenthPoints * 16; number < num_points; number++) {
        const float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
#endif /* LV_HAVE_SSE2 */
 
 
#ifdef LV_HAVE_SSE
#include <xmmintrin.h>
 
static inline void volk_32f_s32f_convert_8i_a_sse(int8_t* outputVector,
                                                  const float* inputVector,
                                                  const float scalar,
                                                  unsigned int num_points)
{
    const unsigned int quarterPoints = num_points / 4;
 
    const float* inputVectorPtr = (const float*)inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
    const __m128 vmin_val = _mm_set_ps1(min_val);
    const __m128 vmax_val = _mm_set_ps1(max_val);
 
    const __m128 vScalar = _mm_set_ps1(scalar);
 
    __VOLK_ATTR_ALIGNED(16) float outputFloatBuffer[4];
 
    for (unsigned int number = 0; number < quarterPoints; number++) {
        __m128 ret = _mm_load_ps(inputVectorPtr);
        inputVectorPtr += 4;
 
        ret = _mm_max_ps(_mm_min_ps(_mm_mul_ps(ret, vScalar), vmax_val), vmin_val);
 
        _mm_store_ps(outputFloatBuffer, ret);
        for (size_t inner_loop = 0; inner_loop < 4; inner_loop++) {
            *outputVectorPtr++ = (int8_t)(rintf(outputFloatBuffer[inner_loop]));
        }
    }
 
    for (unsigned int number = quarterPoints * 4; number < num_points; number++) {
        const float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
 
#endif /* LV_HAVE_SSE */
 
 
#ifdef LV_HAVE_NEON
#include <arm_neon.h>
 
static inline void volk_32f_s32f_convert_8i_neon(int8_t* outputVector,
                                                 const float* inputVector,
                                                 const float scalar,
                                                 unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int sixteenthPoints = num_points / 16;
 
    const float* inputVectorPtr = inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
 
    float32x4_t vScalar = vdupq_n_f32(scalar);
    float32x4_t vmin_val = vdupq_n_f32(min_val);
    float32x4_t vmax_val = vdupq_n_f32(max_val);
    float32x4_t half = vdupq_n_f32(0.5f);
    float32x4_t neg_half = vdupq_n_f32(-0.5f);
    float32x4_t zero = vdupq_n_f32(0.0f);
 
    for (; number < sixteenthPoints; number++) {
        float32x4_t inputVal0 = vld1q_f32(inputVectorPtr);
        float32x4_t inputVal1 = vld1q_f32(inputVectorPtr + 4);
        float32x4_t inputVal2 = vld1q_f32(inputVectorPtr + 8);
        float32x4_t inputVal3 = vld1q_f32(inputVectorPtr + 12);
        inputVectorPtr += 16;
 
        // Scale and clip
        float32x4_t ret0 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal0, vScalar), vmax_val), vmin_val);
        float32x4_t ret1 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal1, vScalar), vmax_val), vmin_val);
        float32x4_t ret2 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal2, vScalar), vmax_val), vmin_val);
        float32x4_t ret3 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal3, vScalar), vmax_val), vmin_val);
 
        // Round to nearest: add copysign(0.5, x) before truncating
        uint32x4_t neg0 = vcltq_f32(ret0, zero);
        uint32x4_t neg1 = vcltq_f32(ret1, zero);
        uint32x4_t neg2 = vcltq_f32(ret2, zero);
        uint32x4_t neg3 = vcltq_f32(ret3, zero);
        ret0 = vaddq_f32(ret0, vbslq_f32(neg0, neg_half, half));
        ret1 = vaddq_f32(ret1, vbslq_f32(neg1, neg_half, half));
        ret2 = vaddq_f32(ret2, vbslq_f32(neg2, neg_half, half));
        ret3 = vaddq_f32(ret3, vbslq_f32(neg3, neg_half, half));
 
        // Convert to int32 (truncates towards zero, but we pre-rounded)
        int32x4_t intVal0 = vcvtq_s32_f32(ret0);
        int32x4_t intVal1 = vcvtq_s32_f32(ret1);
        int32x4_t intVal2 = vcvtq_s32_f32(ret2);
        int32x4_t intVal3 = vcvtq_s32_f32(ret3);
 
        // Narrow to int16 with saturation
        int16x4_t narrow16_0 = vqmovn_s32(intVal0);
        int16x4_t narrow16_1 = vqmovn_s32(intVal1);
        int16x4_t narrow16_2 = vqmovn_s32(intVal2);
        int16x4_t narrow16_3 = vqmovn_s32(intVal3);
        int16x8_t wide16_0 = vcombine_s16(narrow16_0, narrow16_1);
        int16x8_t wide16_1 = vcombine_s16(narrow16_2, narrow16_3);
 
        // Narrow to int8 with saturation
        int8x8_t narrow8_0 = vqmovn_s16(wide16_0);
        int8x8_t narrow8_1 = vqmovn_s16(wide16_1);
        int8x16_t result = vcombine_s8(narrow8_0, narrow8_1);
 
        vst1q_s8(outputVectorPtr, result);
        outputVectorPtr += 16;
    }
 
    number = sixteenthPoints * 16;
    for (; number < num_points; number++) {
        float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
#endif /* LV_HAVE_NEON */
 
 
#ifdef LV_HAVE_NEONV8
#include <arm_neon.h>
 
static inline void volk_32f_s32f_convert_8i_neonv8(int8_t* outputVector,
                                                   const float* inputVector,
                                                   const float scalar,
                                                   unsigned int num_points)
{
    unsigned int number = 0;
    const unsigned int thirtysecondPoints = num_points / 32;
 
    const float* inputVectorPtr = inputVector;
    int8_t* outputVectorPtr = outputVector;
 
    const float min_val = INT8_MIN;
    const float max_val = INT8_MAX;
 
    float32x4_t vScalar = vdupq_n_f32(scalar);
    float32x4_t vmin_val = vdupq_n_f32(min_val);
    float32x4_t vmax_val = vdupq_n_f32(max_val);
 
    for (; number < thirtysecondPoints; number++) {
        float32x4_t inputVal0 = vld1q_f32(inputVectorPtr);
        float32x4_t inputVal1 = vld1q_f32(inputVectorPtr + 4);
        float32x4_t inputVal2 = vld1q_f32(inputVectorPtr + 8);
        float32x4_t inputVal3 = vld1q_f32(inputVectorPtr + 12);
        float32x4_t inputVal4 = vld1q_f32(inputVectorPtr + 16);
        float32x4_t inputVal5 = vld1q_f32(inputVectorPtr + 20);
        float32x4_t inputVal6 = vld1q_f32(inputVectorPtr + 24);
        float32x4_t inputVal7 = vld1q_f32(inputVectorPtr + 28);
        __VOLK_PREFETCH(inputVectorPtr + 32);
        inputVectorPtr += 32;
 
        // Scale and clip
        float32x4_t ret0 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal0, vScalar), vmax_val), vmin_val);
        float32x4_t ret1 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal1, vScalar), vmax_val), vmin_val);
        float32x4_t ret2 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal2, vScalar), vmax_val), vmin_val);
        float32x4_t ret3 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal3, vScalar), vmax_val), vmin_val);
        float32x4_t ret4 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal4, vScalar), vmax_val), vmin_val);
        float32x4_t ret5 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal5, vScalar), vmax_val), vmin_val);
        float32x4_t ret6 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal6, vScalar), vmax_val), vmin_val);
        float32x4_t ret7 =
            vmaxq_f32(vminq_f32(vmulq_f32(inputVal7, vScalar), vmax_val), vmin_val);
 
        // Convert to int32 using round-to-nearest (ARMv8)
        int32x4_t intVal0 = vcvtnq_s32_f32(ret0);
        int32x4_t intVal1 = vcvtnq_s32_f32(ret1);
        int32x4_t intVal2 = vcvtnq_s32_f32(ret2);
        int32x4_t intVal3 = vcvtnq_s32_f32(ret3);
        int32x4_t intVal4 = vcvtnq_s32_f32(ret4);
        int32x4_t intVal5 = vcvtnq_s32_f32(ret5);
        int32x4_t intVal6 = vcvtnq_s32_f32(ret6);
        int32x4_t intVal7 = vcvtnq_s32_f32(ret7);
 
        // Narrow to int16 with saturation
        int16x4_t narrow16_0 = vqmovn_s32(intVal0);
        int16x4_t narrow16_1 = vqmovn_s32(intVal1);
        int16x4_t narrow16_2 = vqmovn_s32(intVal2);
        int16x4_t narrow16_3 = vqmovn_s32(intVal3);
        int16x4_t narrow16_4 = vqmovn_s32(intVal4);
        int16x4_t narrow16_5 = vqmovn_s32(intVal5);
        int16x4_t narrow16_6 = vqmovn_s32(intVal6);
        int16x4_t narrow16_7 = vqmovn_s32(intVal7);
 
        int16x8_t wide16_0 = vcombine_s16(narrow16_0, narrow16_1);
        int16x8_t wide16_1 = vcombine_s16(narrow16_2, narrow16_3);
        int16x8_t wide16_2 = vcombine_s16(narrow16_4, narrow16_5);
        int16x8_t wide16_3 = vcombine_s16(narrow16_6, narrow16_7);
 
        // Narrow to int8 with saturation
        int8x8_t narrow8_0 = vqmovn_s16(wide16_0);
        int8x8_t narrow8_1 = vqmovn_s16(wide16_1);
        int8x8_t narrow8_2 = vqmovn_s16(wide16_2);
        int8x8_t narrow8_3 = vqmovn_s16(wide16_3);
 
        int8x16_t result0 = vcombine_s8(narrow8_0, narrow8_1);
        int8x16_t result1 = vcombine_s8(narrow8_2, narrow8_3);
 
        vst1q_s8(outputVectorPtr, result0);
        vst1q_s8(outputVectorPtr + 16, result1);
        outputVectorPtr += 32;
    }
 
    number = thirtysecondPoints * 32;
    for (; number < num_points; number++) {
        float r = inputVector[number] * scalar;
        volk_32f_s32f_convert_8i_single(&outputVector[number], r);
    }
}
#endif /* LV_HAVE_NEONV8 */
 
 
#ifdef LV_HAVE_RVV
#include <riscv_vector.h>
 
static inline void volk_32f_s32f_convert_8i_rvv(int8_t* outputVector,
                                                const float* inputVector,
                                                const float scalar,
                                                unsigned int num_points)
{
    size_t n = num_points;
    for (size_t vl; n > 0; n -= vl, inputVector += vl, outputVector += vl) {
        vl = __riscv_vsetvl_e32m8(n);
        vfloat32m8_t v = __riscv_vle32_v_f32m8(inputVector, vl);
        vint16m4_t vi = __riscv_vfncvt_x(__riscv_vfmul(v, scalar, vl), vl);
        __riscv_vse8(outputVector, __riscv_vnclip(vi, 0, 0, vl), vl);
    }
}
#endif /*LV_HAVE_RVV*/
 
#endif /* INCLUDED_volk_32f_s32f_convert_8i_a_H */