path: root/src/gallium/drivers/swr/rasterizer/common/simdintrin.h

/****************************************************************************
* Copyright (C) 2014-2015 Intel Corporation.   All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
****************************************************************************/

#ifndef __SWR_SIMDINTRIN_H__
#define __SWR_SIMDINTRIN_H__

#include "os.h"

#include <cassert>

#include <emmintrin.h>
#include <immintrin.h>
#include <xmmintrin.h>

#if KNOB_SIMD_WIDTH == 8 
typedef __m256 simdscalar;
typedef __m256i simdscalari;
typedef uint8_t simdmask;
#else
#error Unsupported vector width
#endif

// simd vector
OSALIGNSIMD(union) simdvector
{
    simdscalar  v[4];
    struct
    {
        simdscalar x, y, z, w;
    };

    simdscalar& operator[] (const int i) { return v[i]; }
    const simdscalar& operator[] (const int i) const { return v[i]; }
};

#if KNOB_SIMD_WIDTH == 8
#define _simd128_maskstore_ps _mm_maskstore_ps
#define _simd_load_ps _mm256_load_ps
#define _simd_load1_ps _mm256_broadcast_ss
#define _simd_loadu_ps _mm256_loadu_ps
#define _simd_setzero_ps _mm256_setzero_ps
#define _simd_set1_ps   _mm256_set1_ps
#define _simd_blend_ps  _mm256_blend_ps
#define _simd_blendv_ps _mm256_blendv_ps
#define _simd_store_ps _mm256_store_ps
#define _simd_mul_ps _mm256_mul_ps
#define _simd_add_ps _mm256_add_ps
#define _simd_sub_ps _mm256_sub_ps
#define _simd_rsqrt_ps _mm256_rsqrt_ps
#define _simd_min_ps _mm256_min_ps
#define _simd_max_ps _mm256_max_ps
#define _simd_movemask_ps _mm256_movemask_ps
#define _simd_cvtps_epi32 _mm256_cvtps_epi32
#define _simd_cvttps_epi32 _mm256_cvttps_epi32
#define _simd_cvtepi32_ps _mm256_cvtepi32_ps
#define _simd_cmplt_ps(a, b) _mm256_cmp_ps(a, b, _CMP_LT_OQ)
#define _simd_cmpgt_ps(a, b) _mm256_cmp_ps(a, b, _CMP_GT_OQ)
#define _simd_cmpneq_ps(a, b) _mm256_cmp_ps(a, b, _CMP_NEQ_OQ)
#define _simd_cmpeq_ps(a, b) _mm256_cmp_ps(a, b, _CMP_EQ_OQ)
#define _simd_cmpge_ps(a, b) _mm256_cmp_ps(a, b, _CMP_GE_OQ)
#define _simd_cmple_ps(a, b) _mm256_cmp_ps(a, b, _CMP_LE_OQ)
#define _simd_cmp_ps(a, b, imm) _mm256_cmp_ps(a, b, imm)
#define _simd_and_ps _mm256_and_ps
#define _simd_or_ps _mm256_or_ps

#define _simd_rcp_ps _mm256_rcp_ps
#define _simd_div_ps _mm256_div_ps
#define _simd_castsi_ps _mm256_castsi256_ps
#define _simd_andnot_ps _mm256_andnot_ps
#define _simd_round_ps _mm256_round_ps
#define _simd_castpd_ps _mm256_castpd_ps
#define _simd_broadcast_ps(a) _mm256_broadcast_ps((const __m128*)(a))
#define _simd_stream_ps _mm256_stream_ps

#define _simd_load_sd _mm256_load_sd
#define _simd_movemask_pd _mm256_movemask_pd
#define _simd_castsi_pd _mm256_castsi256_pd

// emulated integer simd
#define SIMD_EMU_EPI(func, intrin) \
INLINE \
__m256i func(__m256i a, __m256i b)\
{\
    __m128i aHi = _mm256_extractf128_si256(a, 1);\
    __m128i bHi = _mm256_extractf128_si256(b, 1);\
    __m128i aLo = _mm256_castsi256_si128(a);\
    __m128i bLo = _mm256_castsi256_si128(b);\
\
    __m128i subLo = intrin(aLo, bLo);\
    __m128i subHi = intrin(aHi, bHi);\
\
    __m256i result = _mm256_castsi128_si256(subLo);\
            result = _mm256_insertf128_si256(result, subHi, 1);\
\
    return result;\
}

#if (KNOB_ARCH == KNOB_ARCH_AVX)
INLINE
__m256 _simdemu_permute_ps(__m256 a, __m256i b)
{
    __m128 aHi = _mm256_extractf128_ps(a, 1);
    __m128i bHi = _mm256_extractf128_si256(b, 1);
    __m128 aLo = _mm256_castps256_ps128(a);
    __m128i bLo = _mm256_castsi256_si128(b);

    __m128i indexHi = _mm_cmpgt_epi32(bLo, _mm_set1_epi32(3));
    __m128 resLow = _mm_permutevar_ps(aLo, _mm_and_si128(bLo, _mm_set1_epi32(0x3)));
    __m128 resHi = _mm_permutevar_ps(aHi, _mm_and_si128(bLo, _mm_set1_epi32(0x3)));
    __m128 blendLowRes = _mm_blendv_ps(resLow, resHi, _mm_castsi128_ps(indexHi));

    indexHi = _mm_cmpgt_epi32(bHi, _mm_set1_epi32(3));
    resLow = _mm_permutevar_ps(aLo, _mm_and_si128(bHi, _mm_set1_epi32(0x3)));
    resHi = _mm_permutevar_ps(aHi, _mm_and_si128(bHi, _mm_set1_epi32(0x3)));
    __m128 blendHiRes = _mm_blendv_ps(resLow, resHi, _mm_castsi128_ps(indexHi));

    __m256 result = _mm256_castps128_ps256(blendLowRes);
    result = _mm256_insertf128_ps(result, blendHiRes, 1);

    return result;
}

INLINE
__m256i _simdemu_permute_epi32(__m256i a, __m256i b)
{
    return _mm256_castps_si256(_simdemu_permute_ps(_mm256_castsi256_ps(a), b));
}

INLINE
__m256i _simdemu_srlv_epi32(__m256i vA, __m256i vCount)
{
    int32_t aHi, aLow, countHi, countLow;
    __m128i vAHi = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vA), 1));
    __m128i vALow = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vA), 0));
    __m128i vCountHi = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vCount), 1));
    __m128i vCountLow = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vCount), 0));

    aHi = _mm_extract_epi32(vAHi, 0);
    countHi = _mm_extract_epi32(vCountHi, 0);
    aHi >>= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 0);

    aLow = _mm_extract_epi32(vALow, 0);
    countLow = _mm_extract_epi32(vCountLow, 0);
    aLow >>= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 0);

    aHi = _mm_extract_epi32(vAHi, 1);
    countHi = _mm_extract_epi32(vCountHi, 1);
    aHi >>= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 1);

    aLow = _mm_extract_epi32(vALow, 1);
    countLow = _mm_extract_epi32(vCountLow, 1);
    aLow >>= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 1);

    aHi = _mm_extract_epi32(vAHi, 2);
    countHi = _mm_extract_epi32(vCountHi, 2);
    aHi >>= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 2);

    aLow = _mm_extract_epi32(vALow, 2);
    countLow = _mm_extract_epi32(vCountLow, 2);
    aLow >>= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 2);

    aHi = _mm_extract_epi32(vAHi, 3);
    countHi = _mm_extract_epi32(vCountHi, 3);
    aHi >>= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 3);

    aLow = _mm_extract_epi32(vALow, 3);
    countLow = _mm_extract_epi32(vCountLow, 3);
    aLow >>= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 3);

    __m256i ret = _mm256_set1_epi32(0);
    ret = _mm256_insertf128_si256(ret, vAHi, 1);
    ret = _mm256_insertf128_si256(ret, vALow, 0);
    return ret;
}


INLINE
__m256i _simdemu_sllv_epi32(__m256i vA, __m256i vCount)
{
    int32_t aHi, aLow, countHi, countLow;
    __m128i vAHi = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vA), 1));
    __m128i vALow = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vA), 0));
    __m128i vCountHi = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vCount), 1));
    __m128i vCountLow = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(vCount), 0));

    aHi = _mm_extract_epi32(vAHi, 0);
    countHi = _mm_extract_epi32(vCountHi, 0);
    aHi <<= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 0);

    aLow = _mm_extract_epi32(vALow, 0);
    countLow = _mm_extract_epi32(vCountLow, 0);
    aLow <<= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 0);

    aHi = _mm_extract_epi32(vAHi, 1);
    countHi = _mm_extract_epi32(vCountHi, 1);
    aHi <<= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 1);

    aLow = _mm_extract_epi32(vALow, 1);
    countLow = _mm_extract_epi32(vCountLow, 1);
    aLow <<= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 1);

    aHi = _mm_extract_epi32(vAHi, 2);
    countHi = _mm_extract_epi32(vCountHi, 2);
    aHi <<= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 2);

    aLow = _mm_extract_epi32(vALow, 2);
    countLow = _mm_extract_epi32(vCountLow, 2);
    aLow <<= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 2);

    aHi = _mm_extract_epi32(vAHi, 3);
    countHi = _mm_extract_epi32(vCountHi, 3);
    aHi <<= countHi;
    vAHi = _mm_insert_epi32(vAHi, aHi, 3);

    aLow = _mm_extract_epi32(vALow, 3);
    countLow = _mm_extract_epi32(vCountLow, 3);
    aLow <<= countLow;
    vALow = _mm_insert_epi32(vALow, aLow, 3);

    __m256i ret = _mm256_set1_epi32(0);
    ret = _mm256_insertf128_si256(ret, vAHi, 1);
    ret = _mm256_insertf128_si256(ret, vALow, 0);
    return ret;
}

#define _simd_mul_epi32 _simdemu_mul_epi32
#define _simd_mullo_epi32 _simdemu_mullo_epi32
#define _simd_sub_epi32 _simdemu_sub_epi32
#define _simd_sub_epi64 _simdemu_sub_epi64
#define _simd_min_epi32 _simdemu_min_epi32
#define _simd_min_epu32 _simdemu_min_epu32
#define _simd_max_epi32 _simdemu_max_epi32
#define _simd_max_epu32 _simdemu_max_epu32
#define _simd_add_epi32 _simdemu_add_epi32
#define _simd_and_si _simdemu_and_si
#define _simd_andnot_si _simdemu_andnot_si
#define _simd_cmpeq_epi32 _simdemu_cmpeq_epi32
#define _simd_cmplt_epi32 _simdemu_cmplt_epi32
#define _simd_cmpgt_epi32 _simdemu_cmpgt_epi32
#define _simd_or_si _simdemu_or_si
#define _simd_xor_si _simdemu_xor_si
#define _simd_castps_si _mm256_castps_si256
#define _simd_adds_epu8 _simdemu_adds_epu8
#define _simd_subs_epu8 _simdemu_subs_epu8
#define _simd_add_epi8 _simdemu_add_epi8
#define _simd_cmpeq_epi64 _simdemu_cmpeq_epi64
#define _simd_cmpgt_epi64 _simdemu_cmpgt_epi64
#define _simd_cmpgt_epi8 _simdemu_cmpgt_epi8
#define _simd_cmpeq_epi8 _simdemu_cmpeq_epi8
#define _simd_cmpgt_epi16 _simdemu_cmpgt_epi16
#define _simd_cmpeq_epi16 _simdemu_cmpeq_epi16
#define _simd_movemask_epi8 _simdemu_movemask_epi8
#define _simd_permute_ps _simdemu_permute_ps
#define _simd_permute_epi32 _simdemu_permute_epi32
#define _simd_srlv_epi32 _simdemu_srlv_epi32
#define _simd_sllv_epi32 _simdemu_sllv_epi32

SIMD_EMU_EPI(_simdemu_mul_epi32, _mm_mul_epi32)
SIMD_EMU_EPI(_simdemu_mullo_epi32, _mm_mullo_epi32)
SIMD_EMU_EPI(_simdemu_sub_epi32, _mm_sub_epi32)
SIMD_EMU_EPI(_simdemu_sub_epi64, _mm_sub_epi64)
SIMD_EMU_EPI(_simdemu_min_epi32, _mm_min_epi32)
SIMD_EMU_EPI(_simdemu_min_epu32, _mm_min_epu32)
SIMD_EMU_EPI(_simdemu_max_epi32, _mm_max_epi32)
SIMD_EMU_EPI(_simdemu_max_epu32, _mm_max_epu32)
SIMD_EMU_EPI(_simdemu_add_epi32, _mm_add_epi32)
SIMD_EMU_EPI(_simdemu_and_si, _mm_and_si128)
SIMD_EMU_EPI(_simdemu_andnot_si, _mm_andnot_si128)
SIMD_EMU_EPI(_simdemu_cmpeq_epi32, _mm_cmpeq_epi32)
SIMD_EMU_EPI(_simdemu_cmplt_epi32, _mm_cmplt_epi32)
SIMD_EMU_EPI(_simdemu_cmpgt_epi32, _mm_cmpgt_epi32)
SIMD_EMU_EPI(_simdemu_or_si, _mm_or_si128)
SIMD_EMU_EPI(_simdemu_xor_si, _mm_xor_si128)
SIMD_EMU_EPI(_simdemu_adds_epu8, _mm_adds_epu8)
SIMD_EMU_EPI(_simdemu_subs_epu8, _mm_subs_epu8)
SIMD_EMU_EPI(_simdemu_add_epi8, _mm_add_epi8)
SIMD_EMU_EPI(_simdemu_cmpeq_epi64, _mm_cmpeq_epi64)
SIMD_EMU_EPI(_simdemu_cmpgt_epi64, _mm_cmpgt_epi64)
SIMD_EMU_EPI(_simdemu_cmpgt_epi8, _mm_cmpgt_epi8)
SIMD_EMU_EPI(_simdemu_cmpeq_epi8, _mm_cmpeq_epi8)
SIMD_EMU_EPI(_simdemu_cmpgt_epi16, _mm_cmpgt_epi16)
SIMD_EMU_EPI(_simdemu_cmpeq_epi16, _mm_cmpeq_epi16)
SIMD_EMU_EPI(_simdemu_unpacklo_epi8, _mm_unpacklo_epi8)
SIMD_EMU_EPI(_simdemu_unpackhi_epi8, _mm_unpackhi_epi8)
SIMD_EMU_EPI(_simdemu_unpacklo_epi16, _mm_unpacklo_epi16)
SIMD_EMU_EPI(_simdemu_unpackhi_epi16, _mm_unpackhi_epi16)

#define _simd_unpacklo_epi8 _simdemu_unpacklo_epi8
#define _simd_unpackhi_epi8 _simdemu_unpackhi_epi8
#define _simd_unpacklo_epi16 _simdemu_unpacklo_epi16
#define _simd_unpackhi_epi16 _simdemu_unpackhi_epi16
#define _simd_unpacklo_epi32(a, b) _mm256_castps_si256(_mm256_unpacklo_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)))
#define _simd_unpackhi_epi32(a, b) _mm256_castps_si256(_mm256_unpackhi_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)))
#define _simd_unpacklo_epi64(a, b) _mm256_castpd_si256(_mm256_unpacklo_pd(_mm256_castsi256_pd(a), _mm256_castsi256_pd(b)))
#define _simd_unpackhi_epi64(a, b) _mm256_castpd_si256(_mm256_unpackhi_pd(_mm256_castsi256_pd(a), _mm256_castsi256_pd(b)))

#define _simd_slli_epi32(a,i) _simdemu_slli_epi32(a,i)
#define _simd_srai_epi32(a,i) _simdemu_srai_epi32(a,i)
#define _simd_srli_epi32(a,i) _simdemu_srli_epi32(a,i)
#define _simd_srlisi_ps(a,i) _mm256_castsi256_ps(_simdemu_srli_si128<i>(_mm256_castps_si256(a)))

#define _simd128_fmadd_ps _mm_fmaddemu_ps
#define _simd_fmadd_ps _mm_fmaddemu256_ps
#define _simd_fmsub_ps _mm_fmsubemu256_ps
#define _simd_shuffle_epi8 _simdemu_shuffle_epi8 
SIMD_EMU_EPI(_simdemu_shuffle_epi8, _mm_shuffle_epi8)

INLINE
__m128 _mm_fmaddemu_ps(__m128 a, __m128 b, __m128 c)
{
    __m128 res = _mm_mul_ps(a, b);
    res = _mm_add_ps(res, c);
    return res;
}

INLINE
__m256 _mm_fmaddemu256_ps(__m256 a, __m256 b, __m256 c)
{
    __m256 res = _mm256_mul_ps(a, b);
    res = _mm256_add_ps(res, c);
    return res;
}

INLINE
__m256 _mm_fmsubemu256_ps(__m256 a, __m256 b, __m256 c)
{
    __m256 res = _mm256_mul_ps(a, b);
    res = _mm256_sub_ps(res, c);
    return res;
}

INLINE
__m256 _simd_i32gather_ps(const float* pBase, __m256i vOffsets, const int scale)
{
    uint32_t *pOffsets = (uint32_t*)&vOffsets;
    simdscalar vResult;
    float* pResult = (float*)&vResult;
    for (uint32_t i = 0; i < KNOB_SIMD_WIDTH; ++i)
    {
        uint32_t offset = pOffsets[i];
        offset = offset * scale;
        pResult[i] = *(float*)(((const uint8_t*)pBase + offset));
    }

    return vResult;
}

INLINE
__m256 _simd_mask_i32gather_ps(__m256 vSrc, const float* pBase, __m256i vOffsets, __m256 vMask, const int scale)
{
    uint32_t *pOffsets = (uint32_t*)&vOffsets;
    simdscalar vResult = vSrc;
    float* pResult = (float*)&vResult;
    DWORD index;
    uint32_t mask = _simd_movemask_ps(vMask);
    while (_BitScanForward(&index, mask))
    {
        mask &= ~(1 << index);
        uint32_t offset = pOffsets[index];
        offset = offset * scale;
        pResult[index] = *(float*)(((const uint8_t*)pBase + offset));
    }

    return vResult;
}

INLINE
__m256i _simd_abs_epi32(__m256i a)
{
        __m128i aHi = _mm256_extractf128_si256(a, 1);
        __m128i aLo = _mm256_castsi256_si128(a);
        __m128i absLo = _mm_abs_epi32(aLo);
        __m128i absHi = _mm_abs_epi32(aHi);
        __m256i result = _mm256_castsi128_si256(absLo);
        result = _mm256_insertf128_si256(result, absHi, 1);
        return result;
}

INLINE 
int _simdemu_movemask_epi8(__m256i a)
{
    __m128i aHi = _mm256_extractf128_si256(a, 1);
    __m128i aLo = _mm256_castsi256_si128(a);

    int resHi = _mm_movemask_epi8(aHi);
    int resLo = _mm_movemask_epi8(aLo);

    return (resHi << 16) | resLo;
}

INLINE
__m256i _simd_cvtepu8_epi16(__m128i a)
{
    __m128i resultlo = _mm_cvtepu8_epi16(a);
    __m128i resulthi = _mm_cvtepu8_epi16(_mm_srli_si128(a, 8));

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_cvtepu8_epi32(__m128i a)
{
    __m128i resultlo = _mm_cvtepu8_epi32(a);
    __m128i resulthi = _mm_cvtepu8_epi32(_mm_srli_si128(a, 4));

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_cvtepu16_epi32(__m128i a)
{
    __m128i resultlo = _mm_cvtepu16_epi32(a);
    __m128i resulthi = _mm_cvtepu16_epi32(_mm_srli_si128(a, 8));

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_cvtepu16_epi64(__m128i a)
{
    __m128i resultlo = _mm_cvtepu16_epi64(a);
    __m128i resulthi = _mm_cvtepu16_epi64(_mm_srli_si128(a, 4));

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_cvtepu32_epi64(__m128i a)
{
    __m128i resultlo = _mm_cvtepu32_epi64(a);
    __m128i resulthi = _mm_cvtepu32_epi64(_mm_srli_si128(a, 8));

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_packus_epi16(__m256i a, __m256i b)
{
    __m128i alo = _mm256_extractf128_si256(a, 0);
    __m128i ahi = _mm256_extractf128_si256(a, 1);

    __m128i blo = _mm256_extractf128_si256(b, 0);
    __m128i bhi = _mm256_extractf128_si256(b, 1);

    __m128i resultlo = _mm_packus_epi16(alo, blo);
    __m128i resulthi = _mm_packus_epi16(ahi, bhi);

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_packs_epi16(__m256i a, __m256i b)
{
    __m128i alo = _mm256_extractf128_si256(a, 0);
    __m128i ahi = _mm256_extractf128_si256(a, 1);

    __m128i blo = _mm256_extractf128_si256(b, 0);
    __m128i bhi = _mm256_extractf128_si256(b, 1);

    __m128i resultlo = _mm_packs_epi16(alo, blo);
    __m128i resulthi = _mm_packs_epi16(ahi, bhi);

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_packus_epi32(__m256i a, __m256i b)
{
    __m128i alo = _mm256_extractf128_si256(a, 0);
    __m128i ahi = _mm256_extractf128_si256(a, 1);

    __m128i blo = _mm256_extractf128_si256(b, 0);
    __m128i bhi = _mm256_extractf128_si256(b, 1);

    __m128i resultlo = _mm_packus_epi32(alo, blo);
    __m128i resulthi = _mm_packus_epi32(ahi, bhi);

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

INLINE
__m256i _simd_packs_epi32(__m256i a, __m256i b)
{
    __m128i alo = _mm256_extractf128_si256(a, 0);
    __m128i ahi = _mm256_extractf128_si256(a, 1);

    __m128i blo = _mm256_extractf128_si256(b, 0);
    __m128i bhi = _mm256_extractf128_si256(b, 1);

    __m128i resultlo = _mm_packs_epi32(alo, blo);
    __m128i resulthi = _mm_packs_epi32(ahi, bhi);

    __m256i result = _mm256_castsi128_si256(resultlo);

    return _mm256_insertf128_si256(result, resulthi, 1);
}

#else

#define _simd_mul_epi32 _mm256_mul_epi32
#define _simd_mullo_epi32 _mm256_mullo_epi32
#define _simd_sub_epi32 _mm256_sub_epi32
#define _simd_sub_epi64 _mm256_sub_epi64
#define _simd_min_epi32 _mm256_min_epi32
#define _simd_max_epi32 _mm256_max_epi32
#define _simd_min_epu32 _mm256_min_epu32
#define _simd_max_epu32 _mm256_max_epu32
#define _simd_add_epi32 _mm256_add_epi32
#define _simd_and_si _mm256_and_si256
#define _simd_andnot_si _mm256_andnot_si256
#define _simd_cmpeq_epi32 _mm256_cmpeq_epi32
#define _simd_cmplt_epi32(a,b) _mm256_cmpgt_epi32(b,a)
#define _simd_cmpgt_epi32(a,b) _mm256_cmpgt_epi32(a,b)
#define _simd_or_si _mm256_or_si256
#define _simd_xor_si _mm256_xor_si256
#define _simd_castps_si _mm256_castps_si256

#define _simd_unpacklo_epi8 _mm256_unpacklo_epi8
#define _simd_unpackhi_epi8 _mm256_unpackhi_epi8
#define _simd_unpacklo_epi16 _mm256_unpacklo_epi16
#define _simd_unpackhi_epi16 _mm256_unpackhi_epi16
#define _simd_unpacklo_epi32 _mm256_unpacklo_epi32
#define _simd_unpackhi_epi32 _mm256_unpackhi_epi32
#define _simd_unpacklo_epi64 _mm256_unpacklo_epi64
#define _simd_unpackhi_epi64 _mm256_unpackhi_epi64

#define _simd_srli_si(a,i) _simdemu_srli_si128<i>(a)
#define _simd_slli_epi32 _mm256_slli_epi32
#define _simd_srai_epi32 _mm256_srai_epi32
#define _simd_srli_epi32 _mm256_srli_epi32
#define _simd_srlisi_ps(a,i) _mm256_castsi256_ps(_simdemu_srli_si128<i>(_mm256_castps_si256(a)))
#define _simd128_fmadd_ps _mm_fmadd_ps
#define _simd_fmadd_ps _mm256_fmadd_ps
#define _simd_fmsub_ps _mm256_fmsub_ps
#define _simd_shuffle_epi8 _mm256_shuffle_epi8 
#define _simd_adds_epu8 _mm256_adds_epu8
#define _simd_subs_epu8 _mm256_subs_epu8
#define _simd_add_epi8 _mm256_add_epi8
#define _simd_i32gather_ps _mm256_i32gather_ps
#define _simd_mask_i32gather_ps _mm256_mask_i32gather_ps
#define _simd_abs_epi32 _mm256_abs_epi32

#define _simd_cmpeq_epi64 _mm256_cmpeq_epi64
#define _simd_cmpgt_epi64 _mm256_cmpgt_epi64
#define _simd_cmpgt_epi8  _mm256_cmpgt_epi8
#define _simd_cmpeq_epi8  _mm256_cmpeq_epi8
#define _simd_cmpgt_epi16  _mm256_cmpgt_epi16
#define _simd_cmpeq_epi16  _mm256_cmpeq_epi16
#define _simd_movemask_epi8 _mm256_movemask_epi8
#define _simd_permute_ps _mm256_permutevar8x32_ps
#define _simd_permute_epi32 _mm256_permutevar8x32_epi32
#define _simd_srlv_epi32 _mm256_srlv_epi32
#define _simd_sllv_epi32 _mm256_sllv_epi32
#define _simd_cvtepu8_epi16 _mm256_cvtepu8_epi16
#define _simd_cvtepu8_epi32 _mm256_cvtepu8_epi32
#define _simd_cvtepu16_epi32 _mm256_cvtepu16_epi32
#define _simd_cvtepu16_epi64 _mm256_cvtepu16_epi64
#define _simd_cvtepu32_epi64 _mm256_cvtepu32_epi64
#define _simd_packus_epi16 _mm256_packus_epi16
#define _simd_packs_epi16 _mm256_packs_epi16
#define _simd_packus_epi32 _mm256_packus_epi32
#define _simd_packs_epi32 _mm256_packs_epi32

#endif

#define _simd_unpacklo_ps _mm256_unpacklo_ps
#define _simd_unpackhi_ps _mm256_unpackhi_ps
#define _simd_unpacklo_pd _mm256_unpacklo_pd
#define _simd_unpackhi_pd _mm256_unpackhi_pd
#define _simd_insertf128_ps _mm256_insertf128_ps
#define _simd_insertf128_pd _mm256_insertf128_pd
#define _simd_insertf128_si _mm256_insertf128_si256
#define _simd_extractf128_ps _mm256_extractf128_ps
#define _simd_extractf128_pd _mm256_extractf128_pd
#define _simd_extractf128_si _mm256_extractf128_si256
#define _simd_permute2f128_ps _mm256_permute2f128_ps
#define _simd_permute2f128_pd _mm256_permute2f128_pd
#define _simd_permute2f128_si _mm256_permute2f128_si256
#define _simd_shuffle_ps _mm256_shuffle_ps
#define _simd_shuffle_pd _mm256_shuffle_pd
#define _simd_shuffle_epi32(a, b, imm8) _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), imm8))
#define _simd_shuffle_epi64(a, b, imm8) _mm256_castps_si256(_mm256_shuffle_pd(_mm256_castsi256_pd(a), _mm256_castsi256_pd(b), imm8))
#define _simd_set1_epi32 _mm256_set1_epi32
#define _simd_set_epi32 _mm256_set_epi32
#define _simd_set1_epi8 _mm256_set1_epi8
#define _simd_setzero_si _mm256_setzero_si256
#define _simd_cvttps_epi32 _mm256_cvttps_epi32
#define _simd_store_si _mm256_store_si256
#define _simd_broadcast_ss _mm256_broadcast_ss
#define _simd_maskstore_ps _mm256_maskstore_ps
#define _simd_load_si _mm256_load_si256
#define _simd_loadu_si _mm256_loadu_si256
#define _simd_sub_ps _mm256_sub_ps
#define _simd_testz_ps _mm256_testz_ps
#define _simd_testz_si _mm256_testz_si256
#define _simd_xor_ps _mm256_xor_ps

INLINE
simdscalari _simd_loadu2_si(const __m128i *hiaddr, const __m128i *loaddr)
{
    __m128i lo = _mm_loadu_si128(loaddr);
    __m128i hi = _mm_loadu_si128(hiaddr);

    return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1);
}

INLINE
void _simd_storeu2_si(__m128i *hiaddr, __m128i *loaddr, simdscalari a)
{
    _mm_storeu_si128(loaddr, _mm256_castsi256_si128(a));
    _mm_storeu_si128(hiaddr, _mm256_extractf128_si256(a, 1));
}

INLINE
simdscalari _simd_blendv_epi32(simdscalari a, simdscalari b, simdscalar mask)
{
    return _simd_castps_si(_simd_blendv_ps(_simd_castsi_ps(a), _simd_castsi_ps(b), mask));
}

INLINE
simdscalari _simd_blendv_epi32(simdscalari a, simdscalari b, simdscalari mask)
{
    return _simd_castps_si(_simd_blendv_ps(_simd_castsi_ps(a), _simd_castsi_ps(b), _simd_castsi_ps(mask)));
}

template<int mask>
INLINE
__m128i _simd_blend4_epi32(__m128i a, __m128i b)
{
    return _mm_castps_si128(_mm_blend_ps(_mm_castsi128_ps(a), _mm_castsi128_ps(b), mask));
}

// convert bitmask to vector mask
INLINE
simdscalar vMask(int32_t mask)
{
    __m256i vec = _mm256_set1_epi32(mask);
    const __m256i bit = _mm256_set_epi32(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01);
    vec = _simd_and_si(vec, bit);
    vec = _simd_cmplt_epi32(_mm256_setzero_si256(), vec);
    return _simd_castsi_ps(vec);
}

INLINE
simdscalari vMaski(int32_t mask)
{
    __m256i vec = _mm256_set1_epi32(mask);
    const __m256i bit = _mm256_set_epi32(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01);
    vec = _simd_and_si(vec, bit);
    return _simd_cmplt_epi32(_mm256_setzero_si256(), vec);
}

INLINE
void _simd_mov(simdscalar &r, unsigned int rlane, simdscalar& s, unsigned int slane)
{
    OSALIGNSIMD(float) rArray[KNOB_SIMD_WIDTH], sArray[KNOB_SIMD_WIDTH];
    _mm256_store_ps(rArray, r);
    _mm256_store_ps(sArray, s);
    rArray[rlane] = sArray[slane];
    r = _mm256_load_ps(rArray);
}

INLINE __m256i _simdemu_slli_epi32(__m256i a, uint32_t i)
{
    __m128i aHi = _mm256_extractf128_si256(a, 1);
    __m128i aLo = _mm256_castsi256_si128(a);

    __m128i resHi = _mm_slli_epi32(aHi, i);
    __m128i resLo = _mm_slli_epi32(aLo, i);

    __m256i result = _mm256_castsi128_si256(resLo);
            result = _mm256_insertf128_si256(result, resHi, 1);

    return result;
}

INLINE __m256i _simdemu_srai_epi32(__m256i a, uint32_t i)
{
    __m128i aHi = _mm256_extractf128_si256(a, 1);
    __m128i aLo = _mm256_castsi256_si128(a);

    __m128i resHi = _mm_srai_epi32(aHi, i);
    __m128i resLo = _mm_srai_epi32(aLo, i);

    __m256i result = _mm256_castsi128_si256(resLo);
            result = _mm256_insertf128_si256(result, resHi, 1);

    return result;
}

INLINE __m256i _simdemu_srli_epi32(__m256i a, uint32_t i)
{
    __m128i aHi = _mm256_extractf128_si256(a, 1);
    __m128i aLo = _mm256_castsi256_si128(a);

    __m128i resHi = _mm_srli_epi32(aHi, i);
    __m128i resLo = _mm_srli_epi32(aLo, i);

    __m256i result = _mm256_castsi128_si256(resLo);
    result = _mm256_insertf128_si256(result, resHi, 1);

    return result;
}

INLINE
void _simdvec_transpose(simdvector &v)
{
    SWR_INVALID("Need to implement 8 wide version");
}

#else
#error Unsupported vector width
#endif

// Populates a simdvector from a vector. So p = xyzw becomes xxxx yyyy zzzz wwww.
INLINE
void _simdvec_load_ps(simdvector& r, const float *p)
{
    r[0] = _simd_set1_ps(p[0]);
    r[1] = _simd_set1_ps(p[1]);
    r[2] = _simd_set1_ps(p[2]);
    r[3] = _simd_set1_ps(p[3]);
}

INLINE
void _simdvec_mov(simdvector& r, const simdscalar& s)
{
    r[0] = s;
    r[1] = s;
    r[2] = s;
    r[3] = s;
}

INLINE
void _simdvec_mov(simdvector& r, const simdvector& v)
{
    r[0] = v[0];
    r[1] = v[1];
    r[2] = v[2];
    r[3] = v[3];
}

#if 0
// just move a lane from the source simdvector to dest simdvector
INLINE
void _simdvec_mov(simdvector &r, unsigned int rlane, simdvector& s, unsigned int slane)
{
    _simd_mov(r[0], rlane, s[0], slane);
    _simd_mov(r[1], rlane, s[1], slane);
    _simd_mov(r[2], rlane, s[2], slane);
    _simd_mov(r[3], rlane, s[3], slane);
}

#endif
INLINE
void _simdvec_dp3_ps(simdscalar& r, const simdvector& v0, const simdvector& v1)
{
    simdscalar tmp;
    r   = _simd_mul_ps(v0[0], v1[0]);   // (v0.x*v1.x)

    tmp = _simd_mul_ps(v0[1], v1[1]);       // (v0.y*v1.y)
    r   = _simd_add_ps(r, tmp);         // (v0.x*v1.x) + (v0.y*v1.y)

    tmp = _simd_mul_ps(v0[2], v1[2]);   // (v0.z*v1.z)
    r   = _simd_add_ps(r, tmp);         // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)
}

INLINE
void _simdvec_dp4_ps(simdscalar& r, const simdvector& v0, const simdvector& v1)
{
    simdscalar tmp;
    r   = _simd_mul_ps(v0[0], v1[0]);   // (v0.x*v1.x)

    tmp = _simd_mul_ps(v0[1], v1[1]);       // (v0.y*v1.y)
    r   = _simd_add_ps(r, tmp);         // (v0.x*v1.x) + (v0.y*v1.y)

    tmp = _simd_mul_ps(v0[2], v1[2]);   // (v0.z*v1.z)
    r   = _simd_add_ps(r, tmp);         // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)

    tmp = _simd_mul_ps(v0[3], v1[3]);   // (v0.w*v1.w)
    r   = _simd_add_ps(r, tmp);         // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)
}

INLINE
simdscalar _simdvec_rcp_length_ps(const simdvector& v)
{
    simdscalar length;
    _simdvec_dp4_ps(length, v, v);
    return _simd_rsqrt_ps(length);
}

INLINE
void _simdvec_normalize_ps(simdvector& r, const simdvector& v)
{
    simdscalar vecLength;
    vecLength = _simdvec_rcp_length_ps(v);

    r[0] = _simd_mul_ps(v[0], vecLength);
    r[1] = _simd_mul_ps(v[1], vecLength);
    r[2] = _simd_mul_ps(v[2], vecLength);
    r[3] = _simd_mul_ps(v[3], vecLength);
}

INLINE
void _simdvec_mul_ps(simdvector& r, const simdvector& v, const simdscalar& s)
{
    r[0] = _simd_mul_ps(v[0], s);
    r[1] = _simd_mul_ps(v[1], s);
    r[2] = _simd_mul_ps(v[2], s);
    r[3] = _simd_mul_ps(v[3], s);
}

INLINE
void _simdvec_mul_ps(simdvector& r, const simdvector& v0, const simdvector& v1)
{
    r[0] = _simd_mul_ps(v0[0], v1[0]);
    r[1] = _simd_mul_ps(v0[1], v1[1]);
    r[2] = _simd_mul_ps(v0[2], v1[2]);
    r[3] = _simd_mul_ps(v0[3], v1[3]);
}

INLINE
void _simdvec_add_ps(simdvector& r, const simdvector& v0, const simdvector& v1)
{
    r[0] = _simd_add_ps(v0[0], v1[0]);
    r[1] = _simd_add_ps(v0[1], v1[1]);
    r[2] = _simd_add_ps(v0[2], v1[2]);
    r[3] = _simd_add_ps(v0[3], v1[3]);
}

INLINE
void _simdvec_min_ps(simdvector& r, const simdvector& v0, const simdscalar& s)
{
    r[0] = _simd_min_ps(v0[0], s);
    r[1] = _simd_min_ps(v0[1], s);
    r[2] = _simd_min_ps(v0[2], s);
    r[3] = _simd_min_ps(v0[3], s);
}

INLINE
void _simdvec_max_ps(simdvector& r, const simdvector& v0, const simdscalar& s)
{
    r[0] = _simd_max_ps(v0[0], s);
    r[1] = _simd_max_ps(v0[1], s);
    r[2] = _simd_max_ps(v0[2], s);
    r[3] = _simd_max_ps(v0[3], s);
}

// Matrix4x4 * Vector4
//   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * v.w)
//   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * v.w)
//   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * v.w)
//   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * v.w)
INLINE
void _simd_mat4x4_vec4_multiply(
    simdvector& result,
    const float *pMatrix,
    const simdvector& v)
{
    simdscalar m;
    simdscalar r0;
    simdscalar r1;

    m   = _simd_load1_ps(pMatrix + 0*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 0*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 0*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 0*4 + 3);    // m[row][3]
    r1  = _simd_mul_ps(m, v[3]);                // (m3 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
    result[0] = r0;

    m   = _simd_load1_ps(pMatrix + 1*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 1*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 1*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 1*4 + 3);    // m[row][3]
    r1  = _simd_mul_ps(m, v[3]);                // (m3 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
    result[1] = r0;

    m   = _simd_load1_ps(pMatrix + 2*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 2*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 2*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 2*4 + 3);    // m[row][3]
    r1  = _simd_mul_ps(m, v[3]);                // (m3 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
    result[2] = r0;

    m   = _simd_load1_ps(pMatrix + 3*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 3*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 3*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 3*4 + 3);    // m[row][3]
    r1  = _simd_mul_ps(m, v[3]);                // (m3 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
    result[3] = r0;
}

// Matrix4x4 * Vector3 - Direction Vector where w = 0.
//   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 0)
//   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 0)
//   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 0)
//   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 0)
INLINE
void _simd_mat3x3_vec3_w0_multiply(
    simdvector& result,
    const float *pMatrix,
    const simdvector& v)
{
    simdscalar m;
    simdscalar r0;
    simdscalar r1;

    m   = _simd_load1_ps(pMatrix + 0*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 0*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 0*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    result[0] = r0;

    m   = _simd_load1_ps(pMatrix + 1*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 1*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 1*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    result[1] = r0;

    m   = _simd_load1_ps(pMatrix + 2*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 2*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 2*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    result[2] = r0;

    result[3] = _simd_setzero_ps();
}

// Matrix4x4 * Vector3 - Position vector where w = 1.
//   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 1)
//   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 1)
//   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 1)
//   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 1)
INLINE
void _simd_mat4x4_vec3_w1_multiply(
    simdvector& result,
    const float *pMatrix,
    const simdvector& v)
{
    simdscalar m;
    simdscalar r0;
    simdscalar r1;

    m   = _simd_load1_ps(pMatrix + 0*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 0*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 0*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 0*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[0] = r0;

    m   = _simd_load1_ps(pMatrix + 1*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 1*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 1*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 1*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[1] = r0;

    m   = _simd_load1_ps(pMatrix + 2*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 2*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 2*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 2*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[2] = r0;

    m   = _simd_load1_ps(pMatrix + 3*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 3*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 3*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 3*4 + 3);    // m[row][3]
    result[3]   = _simd_add_ps(r0, m);          // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
}

INLINE
void _simd_mat4x3_vec3_w1_multiply(
    simdvector& result,
    const float *pMatrix,
    const simdvector& v)
{
    simdscalar m;
    simdscalar r0;
    simdscalar r1;

    m   = _simd_load1_ps(pMatrix + 0*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 0*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 0*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 0*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[0] = r0;

    m   = _simd_load1_ps(pMatrix + 1*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 1*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 1*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 1*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[1] = r0;

    m   = _simd_load1_ps(pMatrix + 2*4 + 0);    // m[row][0]
    r0  = _simd_mul_ps(m, v[0]);                // (m00 * v.x)
    m   = _simd_load1_ps(pMatrix + 2*4 + 1);    // m[row][1]
    r1  = _simd_mul_ps(m, v[1]);                // (m1 * v.y)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y)
    m   = _simd_load1_ps(pMatrix + 2*4 + 2);    // m[row][2]
    r1  = _simd_mul_ps(m, v[2]);                // (m2 * v.z)
    r0  = _simd_add_ps(r0, r1);                 // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
    m   = _simd_load1_ps(pMatrix + 2*4 + 3);    // m[row][3]
    r0  = _simd_add_ps(r0, m);                  // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
    result[2] = r0;
    result[3] = _simd_set1_ps(1.0f);
}

//////////////////////////////////////////////////////////////////////////
/// @brief Compute plane equation vA * vX + vB * vY + vC
INLINE simdscalar vplaneps(simdscalar vA, simdscalar vB, simdscalar vC, simdscalar &vX, simdscalar &vY)
{
    simdscalar vOut = _simd_fmadd_ps(vA, vX, vC);
    vOut = _simd_fmadd_ps(vB, vY, vOut);
    return vOut;
}

//////////////////////////////////////////////////////////////////////////
/// @brief Compute plane equation vA * vX + vB * vY + vC
INLINE __m128 vplaneps128(__m128 vA, __m128 vB, __m128 vC, __m128 &vX, __m128 &vY)
{
    __m128 vOut = _simd128_fmadd_ps(vA, vX, vC);
    vOut = _simd128_fmadd_ps(vB, vY, vOut);
    return vOut;
}

//////////////////////////////////////////////////////////////////////////
/// @brief Interpolates a single component.
/// @param vI - barycentric I
/// @param vJ - barycentric J
/// @param pInterpBuffer - pointer to attribute barycentric coeffs
template<UINT Attrib, UINT Comp, UINT numComponents = 4>
static INLINE simdscalar InterpolateComponent(simdscalar vI, simdscalar vJ, const float *pInterpBuffer)
{
    const float *pInterpA = &pInterpBuffer[Attrib * 3 * numComponents + 0 + Comp];
    const float *pInterpB = &pInterpBuffer[Attrib * 3 * numComponents + numComponents + Comp];
    const float *pInterpC = &pInterpBuffer[Attrib * 3 * numComponents + numComponents * 2 + Comp];

    simdscalar vA = _simd_broadcast_ss(pInterpA);
    simdscalar vB = _simd_broadcast_ss(pInterpB);
    simdscalar vC = _simd_broadcast_ss(pInterpC);

    simdscalar vk = _simd_sub_ps(_simd_sub_ps(_simd_set1_ps(1.0f), vI), vJ);
    vC = _simd_mul_ps(vk, vC);
    
    return vplaneps(vA, vB, vC, vI, vJ);
}

//////////////////////////////////////////////////////////////////////////
/// @brief Interpolates a single component (flat shade).
/// @param pInterpBuffer - pointer to attribute barycentric coeffs
template<UINT Attrib, UINT Comp, UINT numComponents = 4>
static INLINE simdscalar InterpolateComponentFlat(const float *pInterpBuffer)
{
    const float *pInterpA = &pInterpBuffer[Attrib * 3 * numComponents + 0 + Comp];

    simdscalar vA = _simd_broadcast_ss(pInterpA);

    return vA;
}

//////////////////////////////////////////////////////////////////////////
/// @brief Interpolates a single component.
/// @param vI - barycentric I
/// @param vJ - barycentric J
/// @param pInterpBuffer - pointer to attribute barycentric coeffs
template<UINT Attrib, UINT Comp, UINT numComponents = 4>
static INLINE __m128 InterpolateComponent(__m128 vI, __m128 vJ, const float *pInterpBuffer)
{
    const float *pInterpA = &pInterpBuffer[Attrib * 3 * numComponents + 0 + Comp];
    const float *pInterpB = &pInterpBuffer[Attrib * 3 * numComponents + numComponents + Comp];
    const float *pInterpC = &pInterpBuffer[Attrib * 3 * numComponents + numComponents * 2 + Comp];

    __m128 vA = _mm_broadcast_ss(pInterpA);
    __m128 vB = _mm_broadcast_ss(pInterpB);
    __m128 vC = _mm_broadcast_ss(pInterpC);

    __m128 vk = _mm_sub_ps(_mm_sub_ps(_mm_set1_ps(1.0f), vI), vJ);
    vC = _mm_mul_ps(vk, vC);

    return vplaneps128(vA, vB, vC, vI, vJ);
}

static INLINE __m128 _simd128_abs_ps(__m128 a)
{
    __m128i ai = _mm_castps_si128(a);
    return _mm_castsi128_ps(_mm_and_si128(ai, _mm_set1_epi32(0x7fffffff)));
}

static INLINE simdscalar _simd_abs_ps(simdscalar a)
{
    simdscalari ai = _simd_castps_si(a);
    return _simd_castsi_ps(_simd_and_si(ai, _simd_set1_epi32(0x7fffffff)));
}

INLINE
UINT pdep_u32(UINT a, UINT mask)
{
#if KNOB_ARCH >= KNOB_ARCH_AVX2
    return _pdep_u32(a, mask);
#else
    UINT result = 0;

    // copied from http://wm.ite.pl/articles/pdep-soft-emu.html 
    // using bsf instead of funky loop
    DWORD maskIndex;
    while (_BitScanForward(&maskIndex, mask))
    {
        // 1. isolate lowest set bit of mask
        const UINT lowest = 1 << maskIndex;

        // 2. populate LSB from src
        const UINT LSB = (UINT)((int)(a << 31) >> 31);

        // 3. copy bit from mask
        result |= LSB & lowest;

        // 4. clear lowest bit
        mask &= ~lowest;

        // 5. prepare for next iteration
        a >>= 1;
    }

    return result;
#endif
}

INLINE
UINT pext_u32(UINT a, UINT mask)
{
#if KNOB_ARCH >= KNOB_ARCH_AVX2
    return _pext_u32(a, mask);
#else
    UINT result = 0;
    DWORD maskIndex;
    uint32_t currentBit = 0;
    while (_BitScanForward(&maskIndex, mask))
    {
        // 1. isolate lowest set bit of mask
        const UINT lowest = 1 << maskIndex;

        // 2. copy bit from mask
        result |= ((a & lowest) > 0) << currentBit++;

        // 3. clear lowest bit
        mask &= ~lowest;
    }
    return result;
#endif
}

#if ENABLE_AVX512_SIMD16
#include "simd16intrin.h"
#endif//ENABLE_AVX512_SIMD16

#endif//__SWR_SIMDINTRIN_H__