/****************************************************************************
**
** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
** All rights reserved.
** Contact: Nokia Corporation (qt-info@nokia.com)
**
** This file is part of the QtGui module of the Qt Toolkit.
**
** $QT_BEGIN_LICENSE:LGPL$
** No Commercial Usage
** This file contains pre-release code and may not be distributed.
** You may use this file in accordance with the terms and conditions
** contained in the Technology Preview License Agreement accompanying
** this package.
**
** GNU Lesser General Public License Usage
** Alternatively, this file may be used under the terms of the GNU Lesser
** General Public License version 2.1 as published by the Free Software
** Foundation and appearing in the file LICENSE.LGPL included in the
** packaging of this file. Please review the following information to
** ensure the GNU Lesser General Public License version 2.1 requirements
** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
**
** In addition, as a special exception, Nokia gives you certain additional
** rights. These rights are described in the Nokia Qt LGPL Exception
** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
**
** If you have questions regarding the use of this file, please contact
** Nokia at qt-info@nokia.com.
**
**
**
**
**
**
**
**
** $QT_END_LICENSE$
**
****************************************************************************/
#include <private/qdrawhelper_p.h>
#include <private/qblendfunctions_p.h>
#include <private/qmath_p.h>
#ifdef QT_HAVE_NEON
#include <private/qdrawhelper_neon_p.h>
#include <private/qpaintengine_raster_p.h>
#include <arm_neon.h>
QT_BEGIN_NAMESPACE
static inline uint16x8_t qvdiv_255_u16(uint16x8_t x, uint16x8_t half)
{
// result = (x + (x >> 8) + 0x80) >> 8
const uint16x8_t temp = vshrq_n_u16(x, 8); // x >> 8
const uint16x8_t sum_part = vaddq_u16(x, half); // x + 0x80
const uint16x8_t sum = vaddq_u16(temp, sum_part);
return vshrq_n_u16(sum, 8);
}
static inline uint16x8_t qvbyte_mul_u16(uint16x8_t x, uint16x8_t alpha, uint16x8_t half)
{
// t = qRound(x * alpha / 255.0)
const uint16x8_t t = vmulq_u16(x, alpha); // t
return qvdiv_255_u16(t, half);
}
static inline uint16x8_t qvinterpolate_pixel_255(uint16x8_t x, uint16x8_t a, uint16x8_t y, uint16x8_t b, uint16x8_t half)
{
// t = x * a + y * b
const uint16x8_t ta = vmulq_u16(x, a);
const uint16x8_t tb = vmulq_u16(y, b);
return qvdiv_255_u16(vaddq_u16(ta, tb), half);
}
static inline uint16x8_t qvsource_over_u16(uint16x8_t src16, uint16x8_t dst16, uint16x8_t half, uint16x8_t full)
{
const uint16x4_t alpha16_high = vdup_lane_u16(vget_high_u16(src16), 3);
const uint16x4_t alpha16_low = vdup_lane_u16(vget_low_u16(src16), 3);
const uint16x8_t alpha16 = vsubq_u16(full, vcombine_u16(alpha16_low, alpha16_high));
return vaddq_u16(src16, qvbyte_mul_u16(dst16, alpha16, half));
}
extern "C" void
pixman_composite_over_8888_0565_asm_neon (int32_t w,
int32_t h,
uint16_t *dst,
int32_t dst_stride,
uint32_t *src,
int32_t src_stride);
extern "C" void
pixman_composite_over_8888_8888_asm_neon (int32_t w,
int32_t h,
uint32_t *dst,
int32_t dst_stride,
uint32_t *src,
int32_t src_stride);
extern "C" void
pixman_composite_src_0565_8888_asm_neon (int32_t w,
int32_t h,
uint32_t *dst,
int32_t dst_stride,
uint16_t *src,
int32_t src_stride);
extern "C" void
pixman_composite_over_n_8_0565_asm_neon (int32_t w,
int32_t h,
uint16_t *dst,
int32_t dst_stride,
uint32_t src,
int32_t unused,
uint8_t *mask,
int32_t mask_stride);
extern "C" void
pixman_composite_scanline_over_asm_neon (int32_t w,
const uint32_t *dst,
const uint32_t *src);
// qblendfunctions.cpp
void qt_blend_argb32_on_rgb16_const_alpha(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha);
void qt_blend_rgb16_on_argb32_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha)
{
dbpl /= 4;
sbpl /= 2;
quint32 *dst = (quint32 *) destPixels;
quint16 *src = (quint16 *) srcPixels;
if (const_alpha != 256) {
quint8 a = (255 * const_alpha) >> 8;
quint8 ia = 255 - a;
while (h--) {
for (int x=0; x<w; ++x)
dst[x] = INTERPOLATE_PIXEL_255(qt_colorConvert(src[x], dst[x]), a, dst[x], ia);
dst += dbpl;
src += sbpl;
}
return;
}
pixman_composite_src_0565_8888_asm_neon(w, h, dst, dbpl, src, sbpl);
}
extern "C" void blend_8_pixels_argb32_on_rgb16_neon(quint16 *dst, const quint32 *src, int const_alpha);
void qt_blend_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha)
{
quint16 *dst = (quint16 *) destPixels;
quint32 *src = (quint32 *) srcPixels;
if (const_alpha != 256) {
for (int y=0; y<h; ++y) {
int i = 0;
for (; i < w-7; i += 8)
blend_8_pixels_argb32_on_rgb16_neon(&dst[i], &src[i], const_alpha);
if (i < w) {
int tail = w - i;
quint16 dstBuffer[8];
quint32 srcBuffer[8];
for (int j = 0; j < tail; ++j) {
dstBuffer[j] = dst[i + j];
srcBuffer[j] = src[i + j];
}
blend_8_pixels_argb32_on_rgb16_neon(dstBuffer, srcBuffer, const_alpha);
for (int j = 0; j < tail; ++j) {
dst[i + j] = dstBuffer[j];
src[i + j] = srcBuffer[j];
}
}
dst = (quint16 *)(((uchar *) dst) + dbpl);
src = (quint32 *)(((uchar *) src) + sbpl);
}
return;
}
pixman_composite_over_8888_0565_asm_neon(w, h, dst, dbpl / 2, src, sbpl / 4);
}
void qt_blend_argb32_on_argb32_scanline_neon(uint *dest, const uint *src, int length, uint const_alpha)
{
if (const_alpha == 255) {
pixman_composite_scanline_over_asm_neon(length, dest, src);
} else {
qt_blend_argb32_on_argb32_neon((uchar *)dest, 4 * length, (uchar *)src, 4 * length, length, 1, (const_alpha * 256) / 255);
}
}
void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha)
{
const uint *src = (const uint *) srcPixels;
uint *dst = (uint *) destPixels;
uint16x8_t half = vdupq_n_u16(0x80);
uint16x8_t full = vdupq_n_u16(0xff);
if (const_alpha == 256) {
pixman_composite_over_8888_8888_asm_neon(w, h, (uint32_t *)destPixels, dbpl / 4, (uint32_t *)srcPixels, sbpl / 4);
} else if (const_alpha != 0) {
const_alpha = (const_alpha * 255) >> 8;
uint16x8_t const_alpha16 = vdupq_n_u16(const_alpha);
for (int y = 0; y < h; ++y) {
int x = 0;
for (; x < w-3; x += 4) {
if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]);
uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]);
const uint8x16_t src8 = vreinterpretq_u8_u32(src32);
const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32);
const uint8x8_t src8_low = vget_low_u8(src8);
const uint8x8_t dst8_low = vget_low_u8(dst8);
const uint8x8_t src8_high = vget_high_u8(src8);
const uint8x8_t dst8_high = vget_high_u8(dst8);
const uint16x8_t src16_low = vmovl_u8(src8_low);
const uint16x8_t dst16_low = vmovl_u8(dst8_low);
const uint16x8_t src16_high = vmovl_u8(src8_high);
const uint16x8_t dst16_high = vmovl_u8(dst8_high);
const uint16x8_t srcalpha16_low = qvbyte_mul_u16(src16_low, const_alpha16, half);
const uint16x8_t srcalpha16_high = qvbyte_mul_u16(src16_high, const_alpha16, half);
const uint16x8_t result16_low = qvsource_over_u16(srcalpha16_low, dst16_low, half, full);
const uint16x8_t result16_high = qvsource_over_u16(srcalpha16_high, dst16_high, half, full);
const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high));
}
}
for (; x<w; ++x) {
uint s = src[x];
if (s != 0) {
s = BYTE_MUL(s, const_alpha);
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
}
}
dst = (quint32 *)(((uchar *) dst) + dbpl);
src = (const quint32 *)(((const uchar *) src) + sbpl);
}
}
}
// qblendfunctions.cpp
void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha);
void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
int w, int h,
int const_alpha)
{
if (const_alpha != 256) {
if (const_alpha != 0) {
const uint *src = (const uint *) srcPixels;
uint *dst = (uint *) destPixels;
uint16x8_t half = vdupq_n_u16(0x80);
const_alpha = (const_alpha * 255) >> 8;
int one_minus_const_alpha = 255 - const_alpha;
uint16x8_t const_alpha16 = vdupq_n_u16(const_alpha);
uint16x8_t one_minus_const_alpha16 = vdupq_n_u16(255 - const_alpha);
for (int y = 0; y < h; ++y) {
int x = 0;
for (; x < w-3; x += 4) {
uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]);
uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]);
const uint8x16_t src8 = vreinterpretq_u8_u32(src32);
const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32);
const uint8x8_t src8_low = vget_low_u8(src8);
const uint8x8_t dst8_low = vget_low_u8(dst8);
const uint8x8_t src8_high = vget_high_u8(src8);
const uint8x8_t dst8_high = vget_high_u8(dst8);
const uint16x8_t src16_low = vmovl_u8(src8_low);
const uint16x8_t dst16_low = vmovl_u8(dst8_low);
const uint16x8_t src16_high = vmovl_u8(src8_high);
const uint16x8_t dst16_high = vmovl_u8(dst8_high);
const uint16x8_t result16_low = qvinterpolate_pixel_255(src16_low, const_alpha16, dst16_low, one_minus_const_alpha16, half);
const uint16x8_t result16_high = qvinterpolate_pixel_255(src16_high, const_alpha16, dst16_high, one_minus_const_alpha16, half);
const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high));
}
for (; x<w; ++x) {
uint s = src[x];
s = BYTE_MUL(s, const_alpha);
dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], one_minus_const_alpha);
}
dst = (quint32 *)(((uchar *) dst) + dbpl);
src = (const quint32 *)(((const uchar *) src) + sbpl);
}
}
} else {
qt_blend_rgb32_on_rgb32(destPixels, dbpl, srcPixels, sbpl, w, h, const_alpha);
}
}
void qt_alphamapblit_quint16_neon(QRasterBuffer *rasterBuffer,
int x, int y, quint32 color,
const uchar *bitmap,
int mapWidth, int mapHeight, int mapStride,
const QClipData *)
{
quint16 *dest = reinterpret_cast<quint16*>(rasterBuffer->scanLine(y)) + x;
const int destStride = rasterBuffer->bytesPerLine() / sizeof(quint16);
uchar *mask = const_cast<uchar *>(bitmap);
pixman_composite_over_n_8_0565_asm_neon(mapWidth, mapHeight, dest, destStride, color, 0, mask, mapStride);
}
extern "C" void blend_8_pixels_rgb16_on_rgb16_neon(quint16 *dst, const quint16 *src, int const_alpha);
template <typename SRC, typename BlendFunc>
struct Blend_on_RGB16_SourceAndConstAlpha_Neon {
Blend_on_RGB16_SourceAndConstAlpha_Neon(BlendFunc blender, int const_alpha)
: m_index(0)
, m_blender(blender)
, m_const_alpha(const_alpha)
{
}
inline void write(quint16 *dst, quint32 src)
{
srcBuffer[m_index++] = src;
if (m_index == 8) {
m_blender(dst - 7, srcBuffer, m_const_alpha);
m_index = 0;
}
}
inline void flush(quint16 *dst)
{
if (m_index > 0) {
quint16 dstBuffer[8];
for (int i = 0; i < m_index; ++i)
dstBuffer[i] = dst[i - m_index];
m_blender(dstBuffer, srcBuffer, m_const_alpha);
for (int i = 0; i < m_index; ++i)
dst[i - m_index] = dstBuffer[i];
m_index = 0;
}
}
SRC srcBuffer[8];
int m_index;
BlendFunc m_blender;
int m_const_alpha;
};
template <typename SRC, typename BlendFunc>
Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc>
Blend_on_RGB16_SourceAndConstAlpha_Neon_create(BlendFunc blender, int const_alpha)
{
return Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc>(blender, const_alpha);
}
void qt_scale_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
int const_alpha)
{
if (const_alpha == 0)
return;
qt_scale_image_16bit<quint32>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip,
Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha));
}
void qt_scale_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
int const_alpha);
void qt_scale_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
int const_alpha)
{
if (const_alpha == 0)
return;
if (const_alpha == 256) {
qt_scale_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, const_alpha);
return;
}
qt_scale_image_16bit<quint16>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip,
Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha));
}
extern void qt_transform_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
const QTransform &targetRectTransform,
int const_alpha);
void qt_transform_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
const QTransform &targetRectTransform,
int const_alpha)
{
if (const_alpha == 0)
return;
if (const_alpha == 256) {
qt_transform_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, targetRectTransform, const_alpha);
return;
}
qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl,
reinterpret_cast<const quint16 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform,
Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha));
}
void qt_transform_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
const uchar *srcPixels, int sbpl,
const QRectF &targetRect,
const QRectF &sourceRect,
const QRect &clip,
const QTransform &targetRectTransform,
int const_alpha)
{
if (const_alpha == 0)
return;
qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl,
reinterpret_cast<const quint32 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform,
Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha));
}
static inline void convert_8_pixels_rgb16_to_argb32(quint32 *dst, const quint16 *src)
{
asm volatile (
"vld1.16 { d0, d1 }, [%[SRC]]\n\t"
/* convert 8 r5g6b5 pixel data from {d0, d1} to planar 8-bit format
and put data into d4 - red, d3 - green, d2 - blue */
"vshrn.u16 d4, q0, #8\n\t"
"vshrn.u16 d3, q0, #3\n\t"
"vsli.u16 q0, q0, #5\n\t"
"vsri.u8 d4, d4, #5\n\t"
"vsri.u8 d3, d3, #6\n\t"
"vshrn.u16 d2, q0, #2\n\t"
/* fill d5 - alpha with 0xff */
"mov r2, #255\n\t"
"vdup.8 d5, r2\n\t"
"vst4.8 { d2, d3, d4, d5 }, [%[DST]]"
: : [DST]"r" (dst), [SRC]"r" (src)
: "memory", "r2", "d0", "d1", "d2", "d3", "d4", "d5"
);
}
uint * QT_FASTCALL qt_destFetchRGB16_neon(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length)
{
const ushort *data = (const ushort *)rasterBuffer->scanLine(y) + x;
int i = 0;
for (; i < length - 7; i += 8)
convert_8_pixels_rgb16_to_argb32(&buffer[i], &data[i]);
if (i < length) {
quint16 srcBuffer[8];
quint32 dstBuffer[8];
int tail = length - i;
for (int j = 0; j < tail; ++j)
srcBuffer[j] = data[i + j];
convert_8_pixels_rgb16_to_argb32(dstBuffer, srcBuffer);
for (int j = 0; j < tail; ++j)
buffer[i + j] = dstBuffer[j];
}
return buffer;
}
static inline void convert_8_pixels_argb32_to_rgb16(quint16 *dst, const quint32 *src)
{
asm volatile (
"vld4.8 { d0, d1, d2, d3 }, [%[SRC]]\n\t"
/* convert to r5g6b5 and store it into {d28, d29} */
"vshll.u8 q14, d2, #8\n\t"
"vshll.u8 q8, d1, #8\n\t"
"vshll.u8 q9, d0, #8\n\t"
"vsri.u16 q14, q8, #5\n\t"
"vsri.u16 q14, q9, #11\n\t"
"vst1.16 { d28, d29 }, [%[DST]]"
: : [DST]"r" (dst), [SRC]"r" (src)
: "memory", "d0", "d1", "d2", "d3", "d16", "d17", "d18", "d19", "d28", "d29"
);
}
void QT_FASTCALL qt_destStoreRGB16_neon(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length)
{
quint16 *data = (quint16*)rasterBuffer->scanLine(y) + x;
int i = 0;
for (; i < length - 7; i += 8)
convert_8_pixels_argb32_to_rgb16(&data[i], &buffer[i]);
if (i < length) {
quint32 srcBuffer[8];
quint16 dstBuffer[8];
int tail = length - i;
for (int j = 0; j < tail; ++j)
srcBuffer[j] = buffer[i + j];
convert_8_pixels_argb32_to_rgb16(dstBuffer, srcBuffer);
for (int j = 0; j < tail; ++j)
data[i + j] = dstBuffer[j];
}
}
QT_END_NAMESPACE
#endif // QT_HAVE_NEON