Revision control
Copy as Markdown
// 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
#include "qvideoframeconversionhelper_p.h"
#ifdef QT_COMPILER_SUPPORTS_AVX2
QT_BEGIN_NAMESPACE
namespace {
template<int a, int r, int g, int b>
void convert_to_ARGB32_avx2(const QVideoFrame &frame, uchar *output)
{
FETCH_INFO_PACKED(frame)
MERGE_LOOPS(width, height, stride, 4)
quint32 *argb = reinterpret_cast<quint32*>(output);
#if Q_BYTE_ORDER == Q_LITTLE_ENDIAN
__m256i shuffleMask = _mm256_set_epi8(12 + a, 12 + r, 12 + g, 12 + b,
8 + a, 8 + r, 8 + g, 8 + b,
4 + a, 4 + r, 4 + g, 4 + b,
0 + a, 0 + r, 0 + g, 0 + b,
12 + a, 12 + r, 12 + g, 12 + b,
8 + a, 8 + r, 8 + g, 8 + b,
4 + a, 4 + r, 4 + g, 4 + b,
0 + a, 0 + r, 0 + g, 0 + b);
#else
__m256i shuffleMask = _mm256_set_epi8(15 - a, 15 - r, 15 - g, 15 - b,
11 - a, 11 - r, 11 - g, 11 - b,
7 - a, 7 - r, 7 - g, 7 - b,
3 - a, 3 - r, 3 - g, 3 - b,
15 - a, 15 - r, 15 - g, 15 - b,
11 - a, 11 - r, 11 - g, 11 - b,
7 - a, 7 - r, 7 - g, 7 - b,
3 - a, 3 - r, 3 - g, 3 - b);
#endif
using Pixel = const ArgbPixel<a, r, g, b>;
for (int y = 0; y < height; ++y) {
auto *pixel = reinterpret_cast<const Pixel *>(src);
int x = 0;
QT_MEDIA_ALIGN(32, argb, x, width) {
*argb = pixel->convert();
++pixel;
++argb;
}
for (; x < width - 15; x += 16) {
__m256i pixelData = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(pixel));
__m256i pixelData2 = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(pixel + 8));
pixel += 16;
pixelData = _mm256_shuffle_epi8(pixelData, shuffleMask);
pixelData2 = _mm256_shuffle_epi8(pixelData2, shuffleMask);
_mm256_store_si256(reinterpret_cast<__m256i*>(argb), pixelData);
_mm256_store_si256(reinterpret_cast<__m256i*>(argb + 8), pixelData2);
argb += 16;
}
// leftovers
for (; x < width; ++x) {
*argb = pixel->convert();
++pixel;
++argb;
}
src += stride;
}
}
}
void QT_FASTCALL qt_convert_ARGB8888_to_ARGB32_avx2(const QVideoFrame &frame, uchar *output)
{
convert_to_ARGB32_avx2<0, 1, 2, 3>(frame, output);
}
void QT_FASTCALL qt_convert_ABGR8888_to_ARGB32_avx2(const QVideoFrame &frame, uchar *output)
{
convert_to_ARGB32_avx2<0, 3, 2, 1>(frame, output);
}
void QT_FASTCALL qt_convert_RGBA8888_to_ARGB32_avx2(const QVideoFrame &frame, uchar *output)
{
convert_to_ARGB32_avx2<3, 0, 1, 2>(frame, output);
}
void QT_FASTCALL qt_convert_BGRA8888_to_ARGB32_avx2(const QVideoFrame &frame, uchar *output)
{
convert_to_ARGB32_avx2<3, 2, 1, 0>(frame, output);
}
void QT_FASTCALL qt_copy_pixels_with_mask_avx2(uint32_t *dst, const uint32_t *src, size_t size, uint32_t mask)
{
const auto mask256 = _mm256_set_epi32(mask, mask, mask, mask, mask, mask, mask, mask);
size_t x = 0;
QT_MEDIA_ALIGN(32, dst, x, size)
*(dst++) = *(src++) | mask;
for (; x + 8 * 4 <= size; x += 8 * 4) {
const auto srcData1 = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src));
const auto srcData2 = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src += 8));
const auto srcData3 = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src += 8));
const auto srcData4 = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src += 8));
_mm256_store_si256(reinterpret_cast<__m256i *>(dst), _mm256_or_si256(srcData1, mask256));
_mm256_store_si256(reinterpret_cast<__m256i *>(dst += 8), _mm256_or_si256(srcData2, mask256));
_mm256_store_si256(reinterpret_cast<__m256i *>(dst += 8), _mm256_or_si256(srcData3, mask256));
_mm256_store_si256(reinterpret_cast<__m256i *>(dst += 8), _mm256_or_si256(srcData4, mask256));
src += 8;
dst += 8;
}
// leftovers
for (; x + 8 <= size; x += 8) {
const auto srcData = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(src));
_mm256_store_si256(reinterpret_cast<__m256i *>(dst), _mm256_or_si256(srcData, mask256));
src += 8;
dst += 8;
}
for (; x < size; ++x)
*(dst++) = *(src++) | mask;
}
QT_END_NAMESPACE
#endif