fast copy from USWC using sse4.1. not supported by va

TODO: add memcpy like function. configurable cache size

QtAV.pro

@@ -32,6 +32,8 @@ OTHER_FILES += \
 
 EssentialDepends = avutil avcodec avformat swscale
 OptionalDepends = \
+    sse2 \
+    sse4_1 \
     swresample \
     avresample \
     gl

config.tests/sse2/sse2.cpp

@@ -0,0 +1,52 @@
+/****************************************************************************
+**
+** Copyright (C) 2013 Digia Plc and/or its subsidiary(-ies).
+** Contact: http://www.qt-project.org/legal
+**
+** This file is part of the config.tests of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** Commercial License Usage
+** Licensees holding valid commercial Qt licenses may use this file in
+** accordance with the commercial license agreement provided with the
+** Software or, alternatively, in accordance with the terms contained in
+** a written agreement between you and Digia.  For licensing terms and
+** conditions see http://qt.digia.com/licensing.  For further information
+** use the contact form at http://qt.digia.com/contact-us.
+**
+** 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, Digia gives you certain additional
+** rights.  These rights are described in the Digia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** GNU General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU
+** General Public License version 3.0 as published by the Free Software
+** Foundation and appearing in the file LICENSE.GPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU General Public License version 3.0 requirements will be
+** met: http://www.gnu.org/copyleft/gpl.html.
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <emmintrin.h>
+#if defined(__GNUC__) && __GNUC__ < 4 && __GNUC_MINOR__ < 3
+#error GCC < 3.2 is known to create internal compiler errors with our MMX code
+#endif
+
+int main(int, char**)
+{
+    __m128i a = _mm_setzero_si128();
+    _mm_maskmoveu_si128(a, _mm_setzero_si128(), 0);
+    return 0;
+}

config.tests/sse2/sse2.pro

@@ -0,0 +1,7 @@
+SOURCES = sse2.cpp
+CONFIG -= qt dylib release debug_and_release
+CONFIG += debug console
+#qt5 only has gcc, qcc, vc, linux icc. clang?
+win32-icc: QMAKE_CFLAGS_SSE2 *= -arch:SSE2
+isEmpty(QMAKE_CFLAGS_SSE2):error("This compiler does not support SSE2")
+else:QMAKE_CXXFLAGS += $$QMAKE_CFLAGS_SSE2

config.tests/sse4_1/sse4_1.cpp

@@ -0,0 +1,51 @@
+/****************************************************************************
+**
+** Copyright (C) 2013 Digia Plc and/or its subsidiary(-ies).
+** Contact: http://www.qt-project.org/legal
+**
+** This file is part of the config.tests of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** Commercial License Usage
+** Licensees holding valid commercial Qt licenses may use this file in
+** accordance with the commercial license agreement provided with the
+** Software or, alternatively, in accordance with the terms contained in
+** a written agreement between you and Digia.  For licensing terms and
+** conditions see http://qt.digia.com/licensing.  For further information
+** use the contact form at http://qt.digia.com/contact-us.
+**
+** 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, Digia gives you certain additional
+** rights.  These rights are described in the Digia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** GNU General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU
+** General Public License version 3.0 as published by the Free Software
+** Foundation and appearing in the file LICENSE.GPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU General Public License version 3.0 requirements will be
+** met: http://www.gnu.org/copyleft/gpl.html.
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <smmintrin.h>
+
+int main(int, char**)
+{
+    __m128 a = _mm_setzero_ps();
+    _mm_ceil_ps(a);
+    __m128i result = _mm_mullo_epi32(_mm_set1_epi32(42), _mm_set1_epi32(64));
+    (void)result;
+    return 0;
+}

config.tests/sse4_1/sse4_1.pro

@@ -0,0 +1,7 @@
+SOURCES = sse4_1.cpp
+CONFIG -= qt dylib release debug_and_release
+CONFIG += debug console
+#qt5 only has gcc, qcc, vc, linux icc. clang?
+win32-icc: QMAKE_CFLAGS_SSE4_1 *= -arch:SSE4.1
+isEmpty(QMAKE_CFLAGS_SSE4_1):error("This compiler does not support SSE4.1")
+else:QMAKE_CXXFLAGS += $$QMAKE_CFLAGS_SSE4_1

src/libQtAV.pro

@@ -32,6 +32,15 @@ win32 {
 OTHER_FILES += $$RC_FILE
 TRANSLATIONS = $${PROJECTROOT}/i18n/QtAV_zh_CN.ts
 
+config_sse4_1|contains(TARGET_ARCH_SUB, sse4.1) {
+  DEFINES += QTAV_HAVE_SSE4_1=1
+  QMAKE_CXXFLAGS += $$QMAKE_CFLAGS_SSE4_1 #gcc -msse4.1
+}
+config_sse2|contains(TARGET_ARCH_SUB, sse2) {
+  DEFINES += QTAV_HAVE_SSE2=1
+  QMAKE_CXXFLAGS += $$QMAKE_CFLAGS_SSE2 #gcc -msse2
+}
+
 *msvc* {
 #link FFmpeg and portaudio which are built by gcc need /SAFESEH:NO
     QMAKE_LFLAGS += /SAFESEH:NO
@@ -166,6 +175,7 @@ config_libcedarv {
 SOURCES += \
     QtAV_Compat.cpp \
     QtAV_Global.cpp \
+    utils/GPUMemCopy.cpp \
     AudioThread.cpp \
     AVThread.cpp \
     AudioDecoder.cpp \
@@ -256,6 +266,7 @@ SDK_HEADERS *= \
 
 HEADERS *= \
     $$SDK_HEADERS \
+    utils/GPUMemCopy.h \
     QtAV/prepost.h \
     QtAV/AVDemuxThread.h \
     QtAV/AVThread.h \

src/utils/GPUMemCopy.cpp

@@ -0,0 +1,224 @@
+#include "GPUMemCopy.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <stddef.h>
+
+#include <algorithm>
+
+#include "QtAV/QtAV_Compat.h"
+
+// for mingw gcc
+#if QTAV_HAVE(SSE4_1)
+#include <smmintrin.h> //stream load
+#endif
+
+
+/* Branch prediction */
+#ifdef __GNUC__
+#   define likely(p)   __builtin_expect(!!(p), 1)
+#   define unlikely(p) __builtin_expect(!!(p), 0)
+#else
+#   define likely(p)   (!!(p))
+#   define unlikely(p) (!!(p))
+#endif
+
+namespace QtAV {
+
+#if QTAV_HAVE(SSE2) //FIXME
+// from vlc_common.h begin
+#ifdef __MINGW32__
+# define Memalign(align, size) (__mingw_aligned_malloc(size, align))
+# define Free(base)            (__mingw_aligned_free(base))
+#elif defined(_MSC_VER)
+# define Memalign(align, size) (_aligned_malloc(size, align))
+# define Free(base)            (_aligned_free(base))
+#elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
+static inline void *Memalign(size_t align, size_t size)
+{
+    long diff;
+    void *ptr;
+
+    ptr = malloc(size+align);
+    if(!ptr)
+        return ptr;
+    diff = ((-(long)ptr - 1)&(align-1)) + 1;
+    ptr  = (char*)ptr + diff;
+    ((char*)ptr)[-1]= diff;
+    return ptr;
+}
+
+static void Free(void *ptr)
+{
+    if (ptr)
+        free((char*)ptr - ((char*)ptr)[-1]);
+}
+#else
+static inline void *Memalign(size_t align, size_t size)
+{
+    void *base;
+    if (unlikely(posix_memalign(&base, align, size)))
+        base = NULL;
+    return base;
+}
+# define Free(base) free(base)
+#endif
+#endif //QTAV_HAVE(SSE2)
+
+// from vlc_common.h end
+
+// from https://software.intel.com/en-us/articles/copying-accelerated-video-decode-frame-buffers
+/*
+ * 1. Fill a 4K byte cached (WB) memory buffer from the USWC video frame
+ * 2. Copy the 4K byte cache contents to the destination WB frame
+ * 3. Repeat steps 1 and 2 until the whole frame buffer has been copied.
+ *
+ * _mm_store_si128 and _mm_load_si128 intrinsics will compile to the MOVDQA instruction, _mm_stream_load_si128 and _mm_stream_si128 intrinsics compile to the MOVNTDQA and MOVNTDQ instructions
+ *
+ *  using the same pitch (which is assumed to be a multiple of 64 bytes), and expecting 64 byte alignment of every row of the source, cached 4K buffer and destination buffers.
+ * The MOVNTDQA streaming load instruction and the MOVNTDQ streaming store instruction require at least 16 byte alignment in their memory addresses.
+ */
+//  CopyFrame( )
+//
+//  COPIES VIDEO FRAMES FROM USWC MEMORY TO WB SYSTEM MEMORY VIA CACHED BUFFER
+//    ASSUMES PITCH IS A MULTIPLE OF 64B CACHE LINE SIZE, WIDTH MAY NOT BE
+
+#define CACHED_BUFFER_SIZE 4096
+typedef unsigned int UINT;
+void CopyGPUFrame_SSE4_1(void *pSrc, void *pDest, void * pCacheBlock, UINT width, UINT height, UINT pitch);
+
+bool GPUMemCopy::isAvailable()
+{
+#ifdef __SSE4_1__
+    return true;
+#endif
+#if QTAV_HAVE(SSE4_1)
+    return true;
+#endif
+    static bool is_sse41 = !!(av_get_cpu_flags() & AV_CPU_FLAG_SSE4);
+    return is_sse41;
+}
+
+GPUMemCopy::GPUMemCopy()
+{
+#if QTAV_HAVE(SSE2)
+    mCache.buffer = 0;
+    mCache.size = 0;
+#endif
+}
+
+GPUMemCopy::~GPUMemCopy()
+{
+    cleanCache();
+}
+
+bool GPUMemCopy::initCache(unsigned width)
+{
+#if QTAV_HAVE(SSE2)
+    mCache.size = std::max<size_t>((width + 0x0f) & ~ 0x0f, CACHED_BUFFER_SIZE);
+    mCache.buffer = (unsigned char*)Memalign(16, mCache.size);
+    return !!mCache.buffer;
+#endif
+    return false;
+}
+
+void GPUMemCopy::cleanCache()
+{
+#if QTAV_HAVE(SSE2)
+    if (mCache.buffer) {
+        Free(mCache.buffer);
+    }
+    mCache.buffer = 0;
+    mCache.size = 0;
+#endif
+}
+
+void GPUMemCopy::copyFrame(void *pSrc, void *pDest, unsigned width, unsigned height, unsigned pitch)
+{
+    CopyGPUFrame_SSE4_1(pSrc, pDest, mCache.buffer, width, height, pitch);
+}
+
+void CopyGPUFrame_SSE4_1(void *pSrc, void *pDest, void *pCacheBlock, UINT width, UINT height, UINT pitch)
+{
+#if QTAV_HAVE(SSE4_1)
+    __m128i		x0, x1, x2, x3;
+    __m128i		*pLoad;
+    __m128i		*pStore;
+    __m128i		*pCache;
+    UINT		x, y, yLoad, yStore;
+    UINT		rowsPerBlock;
+    UINT		width64;
+    UINT		extraPitch;
+
+    rowsPerBlock = CACHED_BUFFER_SIZE / pitch;
+    width64 = (width + 63) & ~0x03f;
+    extraPitch = (pitch - width64) / 16;
+
+    pLoad  = (__m128i *)pSrc;
+    pStore = (__m128i *)pDest;
+
+    //  COPY THROUGH 4KB CACHED BUFFER
+    for (y = 0; y < height; y += rowsPerBlock) {
+        //  ROWS LEFT TO COPY AT END
+        if (y + rowsPerBlock > height)
+            rowsPerBlock = height - y;
+
+        pCache = (__m128i *)pCacheBlock;
+
+        _mm_mfence();
+
+        // LOAD ROWS OF PITCH WIDTH INTO CACHED BLOCK
+        for (yLoad = 0; yLoad < rowsPerBlock; yLoad++) {
+            // COPY A ROW, CACHE LINE AT A TIME
+            for (x = 0; x < pitch; x +=64) {
+                x0 = _mm_stream_load_si128( pLoad +0 );
+                x1 = _mm_stream_load_si128( pLoad +1 );
+                x2 = _mm_stream_load_si128( pLoad +2 );
+                x3 = _mm_stream_load_si128( pLoad +3 );
+
+                _mm_store_si128( pCache +0,	x0 );
+                _mm_store_si128( pCache +1, x1 );
+                _mm_store_si128( pCache +2, x2 );
+                _mm_store_si128( pCache +3, x3 );
+
+                pCache += 4;
+                pLoad += 4;
+            }
+        }
+
+        _mm_mfence();
+
+        pCache = (__m128i *)pCacheBlock;
+
+        // STORE ROWS OF FRAME WIDTH FROM CACHED BLOCK
+        for (yStore = 0; yStore < rowsPerBlock; yStore++) {
+            // copy a row, cache line at a time
+            for (x = 0; x < width64; x +=64) {
+                x0 = _mm_load_si128( pCache );
+                x1 = _mm_load_si128( pCache +1 );
+                x2 = _mm_load_si128( pCache +2 );
+                x3 = _mm_load_si128( pCache +3 );
+
+                _mm_stream_si128( pStore,	x0 );
+                _mm_stream_si128( pStore +1, x1 );
+                _mm_stream_si128( pStore +2, x2 );
+                _mm_stream_si128( pStore +3, x3 );
+
+                pCache += 4;
+                pStore += 4;
+            }
+            pCache += extraPitch;
+            pStore += extraPitch;
+        }
+    }
+#else
+    Q_UNUSED(pSrc);
+    Q_UNUSED(pDest);
+    Q_UNUSED(pCacheBlock);
+    Q_UNUSED(width);
+    Q_UNUSED(height);
+    Q_UNUSED(pitch);
+#endif //QTAV_HAVE(SSE4_1)
+}
+
+} //namespace QtAV