qdrawingprimitive_sse2_p.h

Go to the documentation of this file.

// Copyright (C) 2016 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only
 
#ifndef QDRAWINGPRIMITIVE_SSE2_P_H
#define QDRAWINGPRIMITIVE_SSE2_P_H
 
#include <QtGui/private/qtguiglobal_p.h>
#include <private/qsimd_p.h>
#include "qdrawhelper_x86_p.h"
#include "qrgba64_p.h"
 
#ifdef __SSE2__
 
//
//  W A R N I N G
//  -------------
//
// This file is not part of the Qt API.  It exists purely as an
// implementation detail.  This header file may change from version to
// version without notice, or even be removed.
//
// We mean it.
//
 
QT_BEGIN_NAMESPACE
 
/*
 * Multiply the components of pixelVector by alphaChannel
 * Each 32bits components of alphaChannel must be in the form 0x00AA00AA
 * colorMask must have 0x00ff00ff on each 32 bits component
 * half must have the value 128 (0x80) for each 32 bits component
 */
#define BYTE_MUL_SSE2(result, pixelVector, alphaChannel, colorMask, half) \
{
    /* 1. separate the colors in 2 vectors so each color is on 16 bits \
       (in order to be multiplied by the alpha \
       each 32 bit of dstVectorAG are in the form 0x00AA00GG \
       each 32 bit of dstVectorRB are in the form 0x00RR00BB */
    __m128i pixelVectorAG = _mm_srli_epi16(pixelVector, 8);
    __m128i pixelVectorRB = _mm_and_si128(pixelVector, colorMask);
 
    /* 2. multiply the vectors by the alpha channel */
    pixelVectorAG = _mm_mullo_epi16(pixelVectorAG, alphaChannel);
    pixelVectorRB = _mm_mullo_epi16(pixelVectorRB, alphaChannel);
 
    /* 3. divide by 255, that's the tricky part. \
       we do it like for BYTE_MUL(), with bit shift: X/255 ~= (X + X/256 + rounding)/256 */
    /** so first (X + X/256 + rounding) */
    pixelVectorRB = _mm_add_epi16(pixelVectorRB, _mm_srli_epi16(pixelVectorRB, 8));
    pixelVectorRB = _mm_add_epi16(pixelVectorRB, half);
    pixelVectorAG = _mm_add_epi16(pixelVectorAG, _mm_srli_epi16(pixelVectorAG, 8));
    pixelVectorAG = _mm_add_epi16(pixelVectorAG, half);
 
    /** second divide by 256 */
    pixelVectorRB = _mm_srli_epi16(pixelVectorRB, 8);
    /** for AG, we could >> 8 to divide followed by << 8 to put the \
        bytes in the correct position. By masking instead, we execute \
        only one instruction */
    pixelVectorAG = _mm_andnot_si128(colorMask, pixelVectorAG);
 
    /* 4. combine the 2 pairs of colors */
    result = _mm_or_si128(pixelVectorAG, pixelVectorRB); \
}
 
/*
 * Each 32bits components of alphaChannel must be in the form 0x00AA00AA
 * oneMinusAlphaChannel must be 255 - alpha for each 32 bits component
 * colorMask must have 0x00ff00ff on each 32 bits component
 * half must have the value 128 (0x80) for each 32 bits component
 */
#define INTERPOLATE_PIXEL_255_SSE2(result, srcVector, dstVector, alphaChannel, oneMinusAlphaChannel, colorMask, half) {
    /* interpolate AG */
    __m128i srcVectorAG = _mm_srli_epi16(srcVector, 8);
    __m128i dstVectorAG = _mm_srli_epi16(dstVector, 8);
    __m128i srcVectorAGalpha = _mm_mullo_epi16(srcVectorAG, alphaChannel);
    __m128i dstVectorAGoneMinusAlphalpha = _mm_mullo_epi16(dstVectorAG, oneMinusAlphaChannel);
    __m128i finalAG = _mm_add_epi16(srcVectorAGalpha, dstVectorAGoneMinusAlphalpha);
    finalAG = _mm_add_epi16(finalAG, _mm_srli_epi16(finalAG, 8));
    finalAG = _mm_add_epi16(finalAG, half);
    finalAG = _mm_andnot_si128(colorMask, finalAG);
 
    /* interpolate RB */
    __m128i srcVectorRB = _mm_and_si128(srcVector, colorMask);
    __m128i dstVectorRB = _mm_and_si128(dstVector, colorMask);
    __m128i srcVectorRBalpha = _mm_mullo_epi16(srcVectorRB, alphaChannel);
    __m128i dstVectorRBoneMinusAlphalpha = _mm_mullo_epi16(dstVectorRB, oneMinusAlphaChannel);
    __m128i finalRB = _mm_add_epi16(srcVectorRBalpha, dstVectorRBoneMinusAlphalpha);
    finalRB = _mm_add_epi16(finalRB, _mm_srli_epi16(finalRB, 8));
    finalRB = _mm_add_epi16(finalRB, half);
    finalRB = _mm_srli_epi16(finalRB, 8);
 
    /* combine */
    result = _mm_or_si128(finalAG, finalRB); \
}
 
// same as BLEND_SOURCE_OVER_ARGB32_SSE2, but for one vector srcVector
#define BLEND_SOURCE_OVER_ARGB32_SSE2_helper(dst, srcVector, nullVector, half, one, colorMask, alphaMask) {
        const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask);
        if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, alphaMask)) == 0xffff) {
            /* all opaque */
            _mm_store_si128((__m128i *)&dst[x], srcVector);
        } else if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, nullVector)) != 0xffff) {
            /* not fully transparent */
            /* extract the alpha channel on 2 x 16 bits */
            /* so we have room for the multiplication */
            /* each 32 bits will be in the form 0x00AA00AA */
            /* with A being the 1 - alpha */
            __m128i alphaChannel = _mm_srli_epi32(srcVector, 24);
            alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16));
            alphaChannel = _mm_sub_epi16(one, alphaChannel);
 
            const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
            __m128i destMultipliedByOneMinusAlpha;
            BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half);
 
            /* result = s + d * (1-alpha) */
            const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha);
            _mm_store_si128((__m128i *)&dst[x], result);
        }
    }
 
 
// Basically blend src over dst with the const alpha defined as constAlphaVector.
// nullVector, half, one, colorMask are constant across the whole image/texture, and should be defined as:
//const __m128i nullVector = _mm_set1_epi32(0);
//const __m128i half = _mm_set1_epi16(0x80);
//const __m128i one = _mm_set1_epi16(0xff);
//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
//const __m128i alphaMask = _mm_set1_epi32(0xff000000);
//
// The computation being done is:
// result = s + d * (1-alpha)
// with shortcuts if fully opaque or fully transparent.
#define BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask) {
    int x = 0;
 
    /* First, get dst aligned. */
    ALIGNMENT_PROLOGUE_16BYTES(dst, x, length) {
        blend_pixel(dst[x], src[x]);
    }
 
    for (; x < length-3; x += 4) {
        const __m128i srcVector = _mm_loadu_si128((const __m128i *)&src[x]);
        BLEND_SOURCE_OVER_ARGB32_SSE2_helper(dst, srcVector, nullVector, half, one, colorMask, alphaMask)
    }
    SIMD_EPILOGUE(x, length, 3) {
        blend_pixel(dst[x], src[x]);
    } \
}
 
// Basically blend src over dst with the const alpha defined as constAlphaVector.
// nullVector, half, one, colorMask are constant across the whole image/texture, and should be defined as:
//const __m128i nullVector = _mm_set1_epi32(0);
//const __m128i half = _mm_set1_epi16(0x80);
//const __m128i one = _mm_set1_epi16(0xff);
//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
//
// The computation being done is:
// dest = (s + d * sia) * ca + d * cia
//      = s * ca + d * (sia * ca + cia)
//      = s * ca + d * (1 - sa*ca)
#define BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector) \
{
    int x = 0;
 
    ALIGNMENT_PROLOGUE_16BYTES(dst, x, length) {
        blend_pixel(dst[x], src[x], const_alpha);
    }
 
    for (; x < length-3; x += 4) {
        __m128i srcVector = _mm_loadu_si128((const __m128i *)&src[x]);
        if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) {
            BYTE_MUL_SSE2(srcVector, srcVector, constAlphaVector, colorMask, half);
 
            __m128i alphaChannel = _mm_srli_epi32(srcVector, 24);
            alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16));
            alphaChannel = _mm_sub_epi16(one, alphaChannel);
 
            const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
            __m128i destMultipliedByOneMinusAlpha;
            BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half);
 
            const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha);
            _mm_store_si128((__m128i *)&dst[x], result);
        }
    }
    SIMD_EPILOGUE(x, length, 3) {
        blend_pixel(dst[x], src[x], const_alpha);
    } \
}
 
QT_END_NAMESPACE
 
#endif // __SSE2__
 
QT_BEGIN_NAMESPACE
#if QT_COMPILER_SUPPORTS_HERE(SSE4_1)
QT_FUNCTION_TARGET(SSE2)
static inline void Q_DECL_VECTORCALL reciprocal_mul_ss(__m128 &ia, const __m128 a, float mul)
{
    ia = _mm_rcp_ss(a); // Approximate 1/a
    // Improve precision of ia using Newton-Raphson
    ia = _mm_sub_ss(_mm_add_ss(ia, ia), _mm_mul_ss(ia, _mm_mul_ss(ia, a)));
    ia = _mm_mul_ss(ia, _mm_set_ss(mul));
    ia = _mm_shuffle_ps(ia, ia, _MM_SHUFFLE(0,0,0,0));
}
 
QT_FUNCTION_TARGET(SSE4_1)
static inline QRgb qUnpremultiply_sse4(QRgb p)
{
    const uint alpha = qAlpha(p);
    if (alpha == 255)
        return p;
    if (alpha == 0)
        return 0;
    const __m128 va = _mm_set1_ps(alpha);
    __m128 via;
    reciprocal_mul_ss(via, va, 255.0f); // Approximate 1/a
    __m128i vl = _mm_cvtepu8_epi32(_mm_cvtsi32_si128(p));
    vl = _mm_cvtps_epi32(_mm_mul_ps(_mm_cvtepi32_ps(vl), via));
    vl = _mm_packus_epi32(vl, vl);
    vl = _mm_insert_epi16(vl, alpha, 3);
    vl = _mm_packus_epi16(vl, vl);
    return _mm_cvtsi128_si32(vl);
}
 
template<enum QtPixelOrder PixelOrder>
QT_FUNCTION_TARGET(SSE4_1)
static inline uint qConvertArgb32ToA2rgb30_sse4(QRgb p)
{
    const uint alpha = qAlpha(p);
    if (alpha == 255)
        return qConvertRgb32ToRgb30<PixelOrder>(p);
    if (alpha == 0)
        return 0;
    constexpr float mult = 1023.0f / (255 >> 6);
    const uint newalpha = (alpha >> 6);
    const __m128 va = _mm_set1_ps(alpha);
    __m128 via;
    reciprocal_mul_ss(via, va, mult * newalpha);
    __m128i vl = _mm_cvtsi32_si128(p);
    vl = _mm_cvtepu8_epi32(vl);
    vl = _mm_cvtps_epi32(_mm_mul_ps(_mm_cvtepi32_ps(vl), via));
    vl = _mm_packus_epi32(vl, vl);
    uint rgb30 = (newalpha << 30);
    rgb30 |= ((uint)_mm_extract_epi16(vl, 1)) << 10;
    if (PixelOrder == PixelOrderRGB) {
        rgb30 |= ((uint)_mm_extract_epi16(vl, 2)) << 20;
        rgb30 |= ((uint)_mm_extract_epi16(vl, 0));
    } else {
        rgb30 |= ((uint)_mm_extract_epi16(vl, 0)) << 20;
        rgb30 |= ((uint)_mm_extract_epi16(vl, 2));
    }
    return rgb30;
}
 
template<enum QtPixelOrder PixelOrder>
QT_FUNCTION_TARGET(SSE4_1)
static inline uint qConvertRgba64ToRgb32_sse4(QRgba64 p)
{
    if (p.isTransparent())
        return 0;
    __m128i vl = _mm_loadl_epi64(reinterpret_cast<const __m128i *>(&p));
    if (!p.isOpaque()) {
        const __m128 va = _mm_set1_ps(p.alpha());
        __m128 via;
        reciprocal_mul_ss(via, va, 65535.0f);
        vl = _mm_unpacklo_epi16(vl, _mm_setzero_si128());
        vl = _mm_cvtps_epi32(_mm_mul_ps(_mm_cvtepi32_ps(vl) , via));
        vl = _mm_packus_epi32(vl, vl);
        vl = _mm_insert_epi16(vl, p.alpha(), 3);
    }
    if (PixelOrder == PixelOrderBGR)
        vl = _mm_shufflelo_epi16(vl, _MM_SHUFFLE(3, 0, 1, 2));
    return toArgb32(vl);
}
#endif
QT_END_NAMESPACE
 
#endif // QDRAWINGPRIMITIVE_SSE2_P_H