WiiFlow_Lite/source/gui/gcvid.cpp
fix94.1 971b5bdb52 -improved jpg support, now the cover width and height dont need
to be divisible with 4 anymore, you can use what you want
-moved sound buffers back in mem2, idk but for wiiflow it gives
a better result
-added mem2 memalign placeholder for new memory manager (after
a week still everything but finished)
-updated other small things, cleaned up source structure a bit
2012-08-05 13:48:15 +00:00

823 lines
18 KiB
C++

/***************************************************************************
* Copyright (C) 2010
* by thakis
*
* Modification and adjustment for the Wii by Dimok
*
* This software is provided 'as-is', without any express or implied
* warranty. In no event will the authors be held liable for any
* damages arising from the use of this software.
*
* Permission is granted to anyone to use this software for any
* purpose, including commercial applications, and to alter it and
* redistribute it freely, subject to the following restrictions:
*
* 1. The origin of this software must not be misrepresented; you
* must not claim that you wrote the original software. If you use
* this software in a product, an acknowledgment in the product
* documentation would be appreciated but is not required.
*
* 2. Altered source versions must be plainly marked as such, and
* must not be misrepresented as being the original software.
*
* 3. This notice may not be removed or altered from any source
* distribution.
*
* gcvid.cpp
***************************************************************************/
#include <cstdlib> //NULL
#include <cstring> //memcmp
#include <string>
#include <cassert>
#include "gcvid.h"
#include "loader/utils.h"
#include "memory/mem2.hpp"
using namespace std;
void readThpHeader(FILE* f, ThpHeader& h)
{
fread(&h, sizeof(h), 1, f);
}
void readThpComponents(FILE* f, ThpComponents& c)
{
fread(&c, sizeof(c), 1, f);
}
void readThpVideoInfo(FILE* f, ThpVideoInfo& i, bool isVersion11)
{
fread(&i, sizeof(i), 1, f);
if(!isVersion11)
{
i.unknown = 0;
fseek(f, -4, SEEK_CUR);
}
}
void readThpAudioInfo(FILE* f, ThpAudioInfo& i, bool isVersion11)
{
fread(&i, sizeof(i), 1, f);
if(!isVersion11)
{
i.numData = 1;
fseek(f, -4, SEEK_CUR);
}
}
void readMthHeader(FILE* f, MthHeader& h)
{
fread(&h, sizeof(h), 1, f);
}
struct DecStruct
{
const u8* currSrcByte;
u32 blockCount;
u8 index;
u8 shift;
};
void thpAudioInitialize(DecStruct& s, const u8* srcStart)
{
s.currSrcByte = srcStart;
s.blockCount = 2;
s.index = (*s.currSrcByte >> 4) & 0x7;
s.shift = *s.currSrcByte & 0xf;
++s.currSrcByte;
}
s32 thpAudioGetNewSample(DecStruct& s)
{
//the following if is executed all 14 calls
//to thpAudioGetNewSample() (once for each
//microblock) because mask & 0xf can contain
//16 different values and starts with 2
if((s.blockCount & 0xf) == 0)
{
s.index = (*s.currSrcByte >> 4) & 0x7;
s.shift = *s.currSrcByte & 0xf;
++s.currSrcByte;
s.blockCount += 2;
}
s32 ret;
if((s.blockCount & 1) != 0)
{
s32 t = (*s.currSrcByte << 28) & 0xf0000000;
ret = t >> 28; //this has to be an arithmetic shift
++s.currSrcByte;
}
else
{
s32 t = (*s.currSrcByte << 24) & 0xf0000000;
ret = t >> 28; //this has to be an arithmetic shift
}
++s.blockCount;
return ret;
}
int thpAudioDecode(s16 * destBuffer, const u8* srcBuffer, bool separateChannelsInOutput, bool isInputStereo)
{
if(destBuffer == NULL || srcBuffer == NULL)
return 0;
ThpAudioBlockHeader* head = (ThpAudioBlockHeader*)srcBuffer;
u32 channelInSize = head->channelSize;
u32 numSamples = head->numSamples;
const u8* srcChannel1 = srcBuffer + sizeof(ThpAudioBlockHeader);
const u8* srcChannel2 = srcChannel1 + channelInSize;
s16* table1 = head->table1;
s16* table2 = head->table2;
s16* destChannel1, * destChannel2;
u32 delta;
if(separateChannelsInOutput)
{
//separated channels in output
destChannel1 = destBuffer;
destChannel2 = destBuffer + numSamples;
delta = 1;
}
else
{
//interleaved channels in output
destChannel1 = destBuffer;
destChannel2 = destBuffer + 1;
delta = 2;
}
DecStruct s;
if(!isInputStereo)
{
//mono channel in input
thpAudioInitialize(s, srcChannel1);
s16 prev1 = *(s16*)(srcBuffer + 72);
s16 prev2 = *(s16*)(srcBuffer + 74);
for(u32 i = 0; i < numSamples; ++i)
{
s64 res = (s64)thpAudioGetNewSample(s);
res = ((res << s.shift) << 11); //convert to 53.11 fixedpoint
//these values are 53.11 fixed point numbers
s64 val1 = table1[2*s.index];
s64 val2 = table1[2*s.index + 1];
//convert to 48.16 fixed point
res = (val1*prev1 + val2*prev2 + res) << 5;
//rounding:
u16 decimalPlaces = res & 0xffff;
if(decimalPlaces > 0x8000) //i.e. > 0.5
//round up
++res;
else if(decimalPlaces == 0x8000) //i.e. == 0.5
if((res & 0x10000) != 0)
//round up every other number
++res;
//get nonfractional parts of number, clamp to [-32768, 32767]
s32 final = (res >> 16);
if(final > 32767) final = 32767;
else if(final < -32768) final = -32768;
prev2 = prev1;
prev1 = final;
*destChannel1 = (s16)final;
*destChannel2 = (s16)final;
destChannel1 += delta;
destChannel2 += delta;
}
}
else
{
//two channels in input - nearly the same as for one channel,
//so no comments here (different lines are marked with XXX)
thpAudioInitialize(s, srcChannel1);
s16 prev1 = *(s16*)(srcBuffer + 72);
s16 prev2 = *(s16*)(srcBuffer + 74);
for(u32 i = 0; i < numSamples; ++i)
{
s64 res = (s64)thpAudioGetNewSample(s);
res = ((res << s.shift) << 11);
s64 val1 = table1[2*s.index];
s64 val2 = table1[2*s.index + 1];
res = (val1*prev1 + val2*prev2 + res) << 5;
u16 decimalPlaces = res & 0xffff;
if(decimalPlaces > 0x8000)
++res;
else if(decimalPlaces == 0x8000)
if((res & 0x10000) != 0)
++res;
s32 final = (res >> 16);
if(final > 32767) final = 32767;
else if(final < -32768) final = -32768;
prev2 = prev1;
prev1 = final;
*destChannel1 = (s16)final;
destChannel1 += delta;
}
thpAudioInitialize(s, srcChannel2);//XXX
prev1 = *(s16*)(srcBuffer + 76);//XXX
prev2 = *(s16*)(srcBuffer + 78);//XXX
for(u32 j = 0; j < numSamples; ++j)
{
s64 res = (s64)thpAudioGetNewSample(s);
res = ((res << s.shift) << 11);
s64 val1 = table2[2*s.index];//XXX
s64 val2 = table2[2*s.index + 1];//XXX
res = (val1*prev1 + val2*prev2 + res) << 5;
u16 decimalPlaces = res & 0xffff;
if(decimalPlaces > 0x8000)
++res;
else if(decimalPlaces == 0x8000)
if((res & 0x10000) != 0)
++res;
s32 final = (res >> 16);
if(final > 32767) final = 32767;
else if(final < -32768) final = -32768;
prev2 = prev1;
prev1 = final;
*destChannel2 = (s16)final;
destChannel2 += delta;
}
}
return numSamples;
}
VideoFrame::VideoFrame()
: _data(NULL), _w(0), _h(0), _p(0)
{}
VideoFrame::~VideoFrame()
{ dealloc(); }
void VideoFrame::resize(int width, int height)
{
if(width == _w && height == _h)
return;
dealloc();
_w = width;
_h = height;
//24 bpp, 4 byte padding
_p = 3*width;
_p += (4 - _p%4)%4;
_data = (u8 *)malloc(_p * _h);
}
int VideoFrame::getWidth() const
{ return _w; }
int VideoFrame::getHeight() const
{ return _h; }
int VideoFrame::getPitch() const
{ return _p; }
u8* VideoFrame::getData()
{ return _data; }
const u8* VideoFrame::getData() const
{ return _data; }
void VideoFrame::dealloc()
{
if(_data != NULL)
free(_data);
_w = _h = _p = 0;
}
//swaps red and blue channel of a video frame
void swapRB(VideoFrame& f)
{
u8* currLine = f.getData();
int hyt = f.getHeight();
int pitch = f.getPitch();
for(int y = 0; y < hyt; ++y)
{
for(int x = 0, x2 = 2; x < pitch; x += 3, x2 += 3)
{
u8 t = currLine[x];
currLine[x] = currLine[x2];
currLine[x2] = t;
}
currLine += pitch;
}
}
enum FILETYPE
{
THP, MTH, JPG,
UNKNOWN = -1
};
FILETYPE getFiletype(FILE* f)
{
long t = ftell(f);
fseek(f, 0, SEEK_SET);
u8 buff[4];
fread(buff, 1, 4, f);
FILETYPE ret = UNKNOWN;
if(memcmp("THP\0", buff, 4) == 0)
ret = THP;
else if(memcmp("MTHP", buff, 4) == 0)
ret = MTH;
else if(buff[0] == 0xff && buff[1] == 0xd8)
ret = JPG;
fseek(f, t, SEEK_SET);
return ret;
}
long getFilesize(FILE* f)
{
long t = ftell(f);
fseek(f, 0, SEEK_END);
long ret = ftell(f);
fseek(f, t, SEEK_SET);
return ret;
}
void decodeJpeg(const u8* data, int size, VideoFrame& dest);
VideoFile::VideoFile(FILE* f)
: loop(true), _f(f)
{}
VideoFile::~VideoFile()
{
fclose(_f);
}
int VideoFile::getWidth() const
{ return 0; }
int VideoFile::getHeight() const
{ return 0; }
float VideoFile::getFps() const
{ return 0.f; }
int VideoFile::getFrameCount() const
{ return 0; }
int VideoFile::getCurrentFrameNr() const
{ return 0; }
bool VideoFile::hasSound() const
{ return false; }
int VideoFile::getNumChannels() const
{ return 0; }
int VideoFile::getFrequency() const
{ return 0; }
int VideoFile::getMaxAudioSamples() const
{ return 0; }
int VideoFile::getCurrentBuffer(s16*) const
{ return 0; }
void VideoFile::loadFrame(VideoFrame& frame, const u8* data, int size) const
{
decodeJpeg(data, size, frame);
}
ThpVideoFile::ThpVideoFile(FILE* f)
: VideoFile(f)
{
readThpHeader(f, _head);
//this is just to find files that have this field != 0, i
//have no such a file
assert(_head.offsetsDataOffset == 0);
readThpComponents(f, _components);
for(u32 i = 0; i < _components.numComponents; ++i)
{
if(_components.componentTypes[i] == 0) //video
readThpVideoInfo(_f, _videoInfo, _head.version == 0x00011000);
else if(_components.componentTypes[i] == 1) //audio
{
readThpAudioInfo(_f, _audioInfo, _head.version == 0x00011000);
assert(_head.maxAudioSamples != 0);
}
}
_numInts = 3;
if(_head.maxAudioSamples != 0)
_numInts = 4;
_currFrameNr = -1;
_nextFrameOffset = _head.firstFrameOffset;
_nextFrameSize = _head.firstFrameSize;
_currFrameData.resize(_head.maxBufferSize); //include some padding
loadNextFrame();
}
int ThpVideoFile::getWidth() const
{ return _videoInfo.width; }
int ThpVideoFile::getHeight() const
{ return _videoInfo.height; }
float ThpVideoFile::getFps() const
{ return _head.fps; }
int ThpVideoFile::getFrameCount() const
{ return _head.numFrames; }
int ThpVideoFile::getCurrentFrameNr() const
{ return _currFrameNr; }
bool ThpVideoFile::loadNextFrame()
{
++_currFrameNr;
if(_currFrameNr >= (int) _head.numFrames)
{
if (!loop)
return false;
_currFrameNr = 0;
_nextFrameOffset = _head.firstFrameOffset;
_nextFrameSize = _head.firstFrameSize;
}
fseek(_f, _nextFrameOffset, SEEK_SET);
fread(&_currFrameData[0], 1, _nextFrameSize, _f);
_nextFrameOffset += _nextFrameSize;
_nextFrameSize = *(u32*)&_currFrameData[0];
return true;
}
void ThpVideoFile::getCurrentFrame(VideoFrame& f) const
{
int size = *(u32*)(&_currFrameData[0] + 8);
loadFrame(f, &_currFrameData[0] + 4*_numInts, size);
}
bool ThpVideoFile::hasSound() const
{ return _head.maxAudioSamples != 0; }
int ThpVideoFile::getNumChannels() const
{
if(hasSound())
return _audioInfo.numChannels;
else
return 0;
}
int ThpVideoFile::getFrequency() const
{
if(hasSound())
return _audioInfo.frequency;
else
return 0;
}
int ThpVideoFile::getMaxAudioSamples() const
{ return _head.maxAudioSamples; }
int ThpVideoFile::getCurrentBuffer(s16* data) const
{
if(!hasSound())
return 0;
int jpegSize = *(u32*)(&_currFrameData[0] + 8);
const u8* src = &_currFrameData[0] + _numInts*4 + jpegSize;
return thpAudioDecode(data, src, false, _audioInfo.numChannels == 2);
}
MthVideoFile::MthVideoFile(FILE* f)
: VideoFile(f)
{
readMthHeader(f, _head);
_currFrameNr = -1;
_nextFrameOffset = _head.offset;
_nextFrameSize = _head.firstFrameSize;
_thisFrameSize = 0;
_currFrameData.resize(_head.maxFrameSize);
loadNextFrame();
}
int MthVideoFile::getWidth() const
{ return _head.width; }
int MthVideoFile::getHeight() const
{ return _head.height; }
float MthVideoFile::getFps() const
{
return (float) 1.0f*_head.fps; //TODO: This has to be in there somewhere
}
int MthVideoFile::getFrameCount() const
{
return _head.numFrames;
}
int MthVideoFile::getCurrentFrameNr() const
{ return _currFrameNr; }
bool MthVideoFile::loadNextFrame()
{
++_currFrameNr;
if(_currFrameNr >= (int) _head.numFrames)
{
if (!loop)
return false;
_currFrameNr = 0;
_nextFrameOffset = _head.offset;
_nextFrameSize = _head.firstFrameSize;
}
fseek(_f, _nextFrameOffset, SEEK_SET);
_currFrameData.resize(_nextFrameSize);
fread(&_currFrameData[0], 1, _nextFrameSize, _f);
_thisFrameSize = _nextFrameSize;
u32 nextSize;
nextSize = *(u32*)(&_currFrameData[0]);
_nextFrameOffset += _nextFrameSize;
_nextFrameSize = nextSize;
return true;
}
void MthVideoFile::getCurrentFrame(VideoFrame& f) const
{
int size = _thisFrameSize;
loadFrame(f, &_currFrameData[0] + 4, size - 4);
}
JpgVideoFile::JpgVideoFile(FILE* f)
: VideoFile(f)
{
vector<u8> data(getFilesize(f));
fread(&data[0], 1, getFilesize(f), f);
loadFrame(_currFrame, &data[0], getFilesize(f));
}
int JpgVideoFile::getWidth() const
{ return _currFrame.getWidth(); }
int JpgVideoFile::getHeight() const
{ return _currFrame.getHeight(); }
int JpgVideoFile::getFrameCount() const
{ return 1; }
void JpgVideoFile::getCurrentFrame(VideoFrame& f) const
{
f.resize(_currFrame.getWidth(), _currFrame.getHeight());
memcpy(f.getData(), _currFrame.getData(),f.getPitch()*f.getHeight());
}
VideoFile* openVideo(const string& fileName)
{
FILE* f = fopen(fileName.c_str(), "rb");
if(f == NULL)
return NULL;
FILETYPE type = getFiletype(f);
switch(type)
{
case THP:
return new ThpVideoFile(f);
case MTH:
return new MthVideoFile(f);
case JPG:
return new JpgVideoFile(f);
default:
fclose(f);
return NULL;
}
}
void closeVideo(VideoFile*& vf)
{
if(vf != NULL)
delete vf;
vf = NULL;
}
//as mentioned above, we have to convert 0xff to 0xff 0x00
//after the image date has begun (ie, after the 0xff 0xda marker)
//but we must not convert the end-of-image-marker (0xff 0xd9)
//this way. There may be 0xff 0xd9 bytes embedded in the image
//data though, so I add 4 bytes to the input buffer
//and fill them with zeroes and check for 0xff 0xd9 0 0
//as end-of-image marker. this is not correct, but works
//and is easier to code... ;-)
//a better solution would be to patch jpeglib so that this conversion
//is not neccessary
u8 endBytesThp[] = { 0xff, 0xd9, 0, 0 }; //used in thp files
u8 endBytesMth[] = { 0xff, 0xd9, 0xff, 0 }; //used in mth files
int countRequiredSize(const u8* data, int size, int& start, int& end)
{
start = 2*size;
end = size;
int count = 0;
int j;
for(j = size - 1; data[j] == 0; --j)
; //search end of data
if(data[j] == 0xd9) //thp file
end = j - 1;
else if(data[j] == 0xff) //mth file
end = j - 2;
for(int i = 0; i < end; ++i)
{
if(data[i] == 0xff)
{
//if i == srcSize - 1, then this would normally overrun src - that's why 4 padding
//bytes are included at the end of src
if(data[i + 1] == 0xda && start == 2*size)
start = i;
if(i > start)
++count;
}
}
return size + count;
}
void convertToRealJpeg(u8* dest, const u8* src, int srcSize, int start, int end)
{
int di = 0;
for(int i = 0; i < srcSize; ++i, ++di)
{
dest[di] = src[i];
//if i == srcSize - 1, then this would normally overrun src - that's why 4 padding
//bytes are included at the end of src
if(src[i] == 0xff && i > start && i < end)
{
++di;
dest[di] = 0;
}
}
}
void decodeJpeg(const u8* data, int size, VideoFrame& dest)
{
//convert format so jpeglib understands it...
int start, end;
int newSize = countRequiredSize(data, size, start, end);
u8* buff = new u8[newSize];
convertToRealJpeg(buff, data, size, start, end);
//...and feed it to jpeglib
decodeRealJpeg(buff, newSize, dest);
delete [] buff;
}
extern "C"
{
#include "jpeglib.h"
#include <setjmp.h>
}
//the following functions are needed to let
//libjpeg read from memory instead of from a file...
//it's a little clumsy to do :-|
const u8* g_jpegBuffer;
int g_jpegSize;
bool g_isLoading = false;
void jpegInitSource(j_decompress_ptr)
{}
boolean jpegFillInputBuffer(j_decompress_ptr cinfo)
{
cinfo->src->next_input_byte = g_jpegBuffer;
cinfo->src->bytes_in_buffer = g_jpegSize;
return TRUE;
}
void jpegSkipInputData(j_decompress_ptr cinfo, long num_bytes)
{
cinfo->src->next_input_byte += num_bytes;
cinfo->src->bytes_in_buffer -= num_bytes;
}
boolean jpegResyncToRestart(j_decompress_ptr cinfo, int desired)
{
jpeg_resync_to_restart(cinfo, desired);
return TRUE;
}
void jpegTermSource(j_decompress_ptr)
{}
void jpegErrorHandler(j_common_ptr cinfo)
{
char buff[1024];
(*cinfo->err->format_message)(cinfo, buff);
//MessageBox(g_hWnd, buff, "JpegLib error:", MB_OK);
}
void decodeRealJpeg(const u8* data, int size, VideoFrame& dest)
{
if(g_isLoading)
return;
g_isLoading = true;
//decompressor state
jpeg_decompress_struct cinfo;
jpeg_error_mgr errorMgr;
//read from memory manager
jpeg_source_mgr sourceMgr;
cinfo.err = jpeg_std_error(&errorMgr);
errorMgr.error_exit = jpegErrorHandler;
jpeg_create_decompress(&cinfo);
//setup read-from-memory
g_jpegBuffer = data;
g_jpegSize = size;
sourceMgr.bytes_in_buffer = size;
sourceMgr.next_input_byte = data;
sourceMgr.init_source = jpegInitSource;
sourceMgr.fill_input_buffer = jpegFillInputBuffer;
sourceMgr.skip_input_data = jpegSkipInputData;
sourceMgr.resync_to_restart = jpegResyncToRestart;
sourceMgr.term_source = jpegTermSource;
cinfo.src = &sourceMgr;
jpeg_read_header(&cinfo, TRUE);
cinfo.do_fancy_upsampling = TRUE;
cinfo.do_block_smoothing = TRUE;
cinfo.dct_method = JDCT_ISLOW;
jpeg_start_decompress(&cinfo);
dest.resize(ALIGN(4, cinfo.output_width), ALIGN(4, cinfo.output_height));
if(cinfo.num_components == 3)
{
int y = 0;
while(cinfo.output_scanline < cinfo.output_height)
{
//invert image because windows wants it downside up
u8* destBuffer = &dest.getData()[(dest.getHeight() - y - 1)*dest.getPitch()];
//NO idea why jpeglib wants a pointer to a pointer
jpeg_read_scanlines(&cinfo, &destBuffer, 1);
++y;
}
//jpeglib gives an error in jpeg_finish_decompress() if no all
//scanlines are read by the application... :-|
//(but because we read all scanlines, it's not really needed)
cinfo.output_scanline = cinfo.output_height;
}
else
{
//MessageBox(g_hWnd, "Only RGB videos are currently supported.", "oops?", MB_OK);
}
jpeg_finish_decompress(&cinfo);
jpeg_destroy_decompress(&cinfo);
g_isLoading = false;
}