// Point-sampling U/V sampler.
static int EmitSampledRGB(VP8Io* io, WebPDecParams* p) {
WebPDecBuffer* output = p.output;
WebPRGBABuffer* buf = &output.u.RGBA;
byte* dst = buf.rgba + io.mb_y * buf.stride;
byte* y_src = io.y;
byte* u_src = io.u;
byte* v_src = io.v;
WebPSampleLinePairFunc sample =
io.a ? WebPSamplersKeepAlpha[output.colorspace]
: WebPSamplers[output.colorspace];
int mb_w = io.mb_w;
int last = io.mb_h - 1;
int j;
for (j = 0; j < last; j += 2) {
sample(y_src, y_src + io.y_stride, u_src, v_src,
dst, dst + buf.stride, mb_w);
y_src += 2 * io.y_stride;
u_src += io.uv_stride;
v_src += io.uv_stride;
dst += 2 * buf.stride;
}
if (j == last) { // Just do the last line twice
sample(y_src, y_src, u_src, v_src, dst, dst, mb_w);
}
return io.mb_h;
}
int VP8InitIoInternal(VP8Io* io, int version) {
if (version != WEBP_DECODER_ABI_VERSION)
return 0; // mismatch error
if (io) {
memset(io, 0, sizeof(*io));
}
return 1;
}
//------------------------------------------------------------------------------
// Main YUV<.RGB conversion functions
static int EmitYUV(VP8Io* io, WebPDecParams* p) {
WebPDecBuffer* output = p.output;
WebPYUVABuffer* buf = &output.u.YUVA;
byte* y_dst = buf.y + io.mb_y * buf.y_stride;
byte* u_dst = buf.u + (io.mb_y >> 1) * buf.u_stride;
byte* v_dst = buf.v + (io.mb_y >> 1) * buf.v_stride;
int mb_w = io.mb_w;
int mb_h = io.mb_h;
int uv_w = (mb_w + 1) / 2;
int uv_h = (mb_h + 1) / 2;
int j;
for (j = 0; j < mb_h; ++j) {
memcpy(y_dst + j * buf.y_stride, io.y + j * io.y_stride, mb_w);
}
for (j = 0; j < uv_h; ++j) {
memcpy(u_dst + j * buf.u_stride, io.u + j * io.uv_stride, uv_w);
memcpy(v_dst + j * buf.v_stride, io.v + j * io.uv_stride, uv_w);
}
return io.mb_h;
}
//------------------------------------------------------------------------------
// YUV444 . RGB conversion
#if false // TODO(skal): this is for future rescaling.
static int EmitRGB(VP8Io* io, WebPDecParams* p) {
WebPDecBuffer* output = p.output;
WebPRGBABuffer* buf = &output.u.RGBA;
byte* dst = buf.rgba + io.mb_y * buf.stride;
byte* y_src = io.y;
byte* u_src = io.u;
byte* v_src = io.v;
WebPYUV444Converter convert = WebPYUV444Converters[output.colorspace];
int mb_w = io.mb_w;
int last = io.mb_h;
int j;
for (j = 0; j < last; ++j) {
convert(y_src, u_src, v_src, dst, mb_w);
y_src += io.y_stride;
u_src += io.uv_stride;
v_src += io.uv_stride;
dst += buf.stride;
}
return io.mb_h;
}
static int InitRGBRescaler(VP8Io* io, WebPDecParams* p) {
int has_alpha = IsAlphaMode(p.output.colorspace);
int out_width = io.scaled_width;
int out_height = io.scaled_height;
int uv_in_width = (io.mb_w + 1) >> 1;
int uv_in_height = (io.mb_h + 1) >> 1;
uint work_size = 2 * out_width; // scratch memory for one rescaler
int* work; // rescalers work area
byte* tmp; // tmp storage for scaled YUV444 samples before RGB conversion
uint tmp_size1, tmp_size2;
tmp_size1 = 3 * work_size;
tmp_size2 = 3 * out_width;
if (has_alpha) {
tmp_size1 += work_size;
tmp_size2 += out_width;
}
p.memory =
calloc(1, tmp_size1 * sizeof(*work) + tmp_size2 * sizeof(*tmp));
if (p.memory == null) {
return 0; // memory error
}
work = (int*)p.memory;
tmp = (byte*)(work + tmp_size1);
InitRescaler(&p.scaler_y, io.mb_w, io.mb_h,
tmp + 0 * out_width, out_width, out_height, 0,
io.mb_w, out_width, io.mb_h, out_height,
work + 0 * work_size);
InitRescaler(&p.scaler_u, uv_in_width, uv_in_height,
tmp + 1 * out_width, out_width, out_height, 0,
io.mb_w, 2 * out_width, io.mb_h, 2 * out_height,
work + 1 * work_size);
InitRescaler(&p.scaler_v, uv_in_width, uv_in_height,
tmp + 2 * out_width, out_width, out_height, 0,
io.mb_w, 2 * out_width, io.mb_h, 2 * out_height,
work + 2 * work_size);
p.emit = EmitRescaledRGB;
if (has_alpha) {
InitRescaler(&p.scaler_a, io.mb_w, io.mb_h,
tmp + 3 * out_width, out_width, out_height, 0,
io.mb_w, out_width, io.mb_h, out_height,
work + 3 * work_size);
p.emit_alpha = EmitRescaledAlphaRGB;
}
return 1;
}
static int EmitRescaledAlphaRGB(VP8Io* io, WebPDecParams* p) {
if (io.a != null) {
int (* output_func)(WebPDecParams* const, int) =
(p.output.colorspace == MODE_RGBA_4444) ? ExportAlphaRGBA4444
: ExportAlpha;
WebPRescaler* scaler = &p.scaler_a;
int j = 0, pos = 0;
while (j < io.mb_h) {
j += Import(io.a + j * io.width, io.width, io.mb_h - j, scaler);
pos += output_func(p, pos);
}
}
return 0;
}
static int EmitRescaledRGB(VP8Io* io, WebPDecParams* p) {
int mb_h = io.mb_h;
int uv_mb_h = (mb_h + 1) >> 1;
int j = 0, uv_j = 0;
int num_lines_out = 0;
while (j < mb_h) {
int y_lines_in = Import(io.y + j * io.y_stride, io.y_stride,
mb_h - j, &p.scaler_y);
int u_lines_in = Import(io.u + uv_j * io.uv_stride, io.uv_stride,
uv_mb_h - uv_j, &p.scaler_u);
int v_lines_in = Import(io.v + uv_j * io.uv_stride, io.uv_stride,
uv_mb_h - uv_j, &p.scaler_v);
(void)v_lines_in; // remove a gcc warning
assert(u_lines_in == v_lines_in);
j += y_lines_in;
uv_j += u_lines_in;
num_lines_out += ExportRGB(p, num_lines_out);
}
return num_lines_out;
}
int VP8ExitCritical(VP8Io io)
{
int ok = 1;
if (this.use_threads_)
{
ok = WebPWorkerSync(&this.worker_);
}
if (io.teardown)
{
io.teardown(io);
}
return ok;
}
// Topmost call
int VP8GetHeaders(VP8Decoder* dec, VP8Io* io) {
byte* buf;
uint buf_size;
byte* alpha_data_tmp;
uint alpha_size_tmp;
uint vp8_chunk_size;
uint bytes_skipped;
VP8FrameHeader* frm_hdr;
VP8PictureHeader* pic_hdr;
VP8BitReader* br;
VP8StatusCode status;
if (dec == null) {
return 0;
}
SetOk(dec);
if (io == null) {
return VP8SetError(dec, VP8_STATUS_INVALID_PARAM,
"null VP8Io passed to VP8GetHeaders()");
}
buf = io.data;
buf_size = io.data_size;
// Process Pre-VP8 chunks.
status = WebPParseHeaders(&buf, &buf_size, &vp8_chunk_size, &bytes_skipped,
&alpha_data_tmp, &alpha_size_tmp);
if (status != VP8_STATUS_OK) {
return VP8SetError(dec, status, "Incorrect/incomplete header.");
}
if (dec.alpha_data_ == null) {
assert(dec.alpha_data_size_ == 0);
// We have NOT set alpha data yet. Set it now.
// (This is to ensure that dec.alpha_data_ is NOT reset to null if
// WebPParseHeaders() is called more than once, as in incremental decoding
// case.)
dec.alpha_data_ = alpha_data_tmp;
dec.alpha_data_size_ = alpha_size_tmp;
}
// Process the VP8 frame header.
if (buf_size < 4) {
return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
"Truncated header.");
}
// Paragraph 9.1
{
uint bits = buf[0] | (buf[1] << 8) | (buf[2] << 16);
frm_hdr = &dec.frm_hdr_;
frm_hdr.key_frame_ = !(bits & 1);
frm_hdr.profile_ = (bits >> 1) & 7;
frm_hdr.show_ = (bits >> 4) & 1;
frm_hdr.partition_length_ = (bits >> 5);
if (frm_hdr.profile_ > 3)
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"Incorrect keyframe parameters.");
if (!frm_hdr.show_)
return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE,
"Frame not displayable.");
buf += 3;
buf_size -= 3;
}
pic_hdr = &dec.pic_hdr_;
if (frm_hdr.key_frame_) {
// Paragraph 9.2
if (buf_size < 7) {
return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
"cannot parse picture header");
}
if (buf[0] != 0x9d || buf[1] != 0x01 || buf[2] != 0x2a) {
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"Bad code word");
}
pic_hdr.width_ = ((buf[4] << 8) | buf[3]) & 0x3fff;
pic_hdr.xscale_ = buf[4] >> 6; // ratio: 1, 5/4 5/3 or 2
pic_hdr.height_ = ((buf[6] << 8) | buf[5]) & 0x3fff;
pic_hdr.yscale_ = buf[6] >> 6;
buf += 7;
buf_size -= 7;
dec.mb_w_ = (pic_hdr.width_ + 15) >> 4;
dec.mb_h_ = (pic_hdr.height_ + 15) >> 4;
// Setup default output area (can be later modified during io.setup())
io.width = pic_hdr.width_;
io.height = pic_hdr.height_;
io.use_scaling = 0;
io.use_cropping = 0;
io.crop_top = 0;
io.crop_left = 0;
io.crop_right = io.width;
io.crop_bottom = io.height;
io.mb_w = io.width; // sanity check
io.mb_h = io.height; // ditto
VP8ResetProba(&dec.proba_);
ResetSegmentHeader(&dec.segment_hdr_);
dec.segment_ = 0; // default for intra
}
// Check if we have all the partition #0 available, and initialize dec.br_
// to read this partition (and this partition only).
if (frm_hdr.partition_length_ > buf_size) {
return VP8SetError(dec, VP8_STATUS_NOT_ENOUGH_DATA,
"bad partition length");
}
br = &dec.br_;
VP8InitBitReader(br, buf, buf + frm_hdr.partition_length_);
buf += frm_hdr.partition_length_;
buf_size -= frm_hdr.partition_length_;
if (frm_hdr.key_frame_) {
pic_hdr.colorspace_ = VP8Get(br);
pic_hdr.clamp_type_ = VP8Get(br);
}
if (!ParseSegmentHeader(br, &dec.segment_hdr_, &dec.proba_)) {
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"cannot parse segment header");
}
// Filter specs
if (!ParseFilterHeader(br, dec)) {
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"cannot parse filter header");
}
status = ParsePartitions(dec, buf, buf_size);
if (status != VP8_STATUS_OK) {
return VP8SetError(dec, status, "cannot parse partitions");
}
// quantizer change
VP8ParseQuant(dec);
// Frame buffer marking
if (!frm_hdr.key_frame_) {
// Paragraph 9.7
#if !ONLY_KEYFRAME_CODE
dec.buffer_flags_ = VP8Get(br) << 0; // update golden
dec.buffer_flags_ |= VP8Get(br) << 1; // update alt ref
if (!(dec.buffer_flags_ & 1)) {
dec.buffer_flags_ |= VP8GetValue(br, 2) << 2;
}
if (!(dec.buffer_flags_ & 2)) {
dec.buffer_flags_ |= VP8GetValue(br, 2) << 4;
}
dec.buffer_flags_ |= VP8Get(br) << 6; // sign bias golden
dec.buffer_flags_ |= VP8Get(br) << 7; // sign bias alt ref
#else
return VP8SetError(dec, VP8_STATUS_UNSUPPORTED_FEATURE,
"Not a key frame.");
#endif
} else {
dec.buffer_flags_ = 0x003 | 0x100;
}
// Paragraph 9.8
#if !ONLY_KEYFRAME_CODE
dec.update_proba_ = VP8Get(br);
if (!dec.update_proba_) { // save for later restore
dec.proba_saved_ = dec.proba_;
}
dec.buffer_flags_ &= 1 << 8;
dec.buffer_flags_ |=
(frm_hdr.key_frame_ || VP8Get(br)) << 8; // refresh last frame
#else
VP8Get(br); // just ignore the value of update_proba_
#endif
VP8ParseProba(br, dec);
#if WEBP_EXPERIMENTAL_FEATURES
// Extensions
if (dec.pic_hdr_.colorspace_) {
uint kTrailerSize = 8;
byte kTrailerMarker = 0x01;
byte* ext_buf = buf - kTrailerSize;
uint size;
if (frm_hdr.partition_length_ < kTrailerSize ||
ext_buf[kTrailerSize - 1] != kTrailerMarker) {
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"RIFF: Inconsistent extra information.");
}
// Layer
size = (ext_buf[0] << 0) | (ext_buf[1] << 8) | (ext_buf[2] << 16);
dec.layer_data_size_ = size;
dec.layer_data_ = null; // will be set later
dec.layer_colorspace_ = ext_buf[3];
}
#endif
// sanitized state
dec.ready_ = 1;
return 1;
}
//------------------------------------------------------------------------------
static void CustomTeardown(VP8Io* io) {
WebPDecParams* p = (WebPDecParams*)io.opaque;
free(p.memory);
p.memory = null;
}
static int EmitAlphaRGBA4444(VP8Io* io, WebPDecParams* p) {
int mb_w = io.mb_w;
int mb_h = io.mb_h;
int i, j;
WebPRGBABuffer* buf = &p.output.u.RGBA;
byte* dst = buf.rgba + io.mb_y * buf.stride + 1;
byte* alpha = io.a;
if (alpha) {
for (j = 0; j < mb_h; ++j) {
for (i = 0; i < mb_w; ++i) {
// Fill in the alpha value (converted to 4 bits).
uint alpha_val = clip((alpha[i] + 8) >> 4, 15);
dst[2 * i] = (dst[2 * i] & 0xf0) | alpha_val;
}
alpha += io.width;
dst += buf.stride;
}
}
return 0;
}
static int EmitAlphaRGB(VP8Io* io, WebPDecParams* p) {
int mb_w = io.mb_w;
int mb_h = io.mb_h;
int i, j;
WebPRGBABuffer* buf = &p.output.u.RGBA;
byte* dst = buf.rgba + io.mb_y * buf.stride +
(p.output.colorspace == MODE_ARGB ? 0 : 3);
byte* alpha = io.a;
if (alpha) {
for (j = 0; j < mb_h; ++j) {
for (i = 0; i < mb_w; ++i) {
dst[4 * i] = alpha[i];
}
alpha += io.width;
dst += buf.stride;
}
}
return 0;
}
// FANCY_UPSAMPLING
//------------------------------------------------------------------------------
static int EmitAlphaYUV(VP8Io* io, WebPDecParams* p) {
byte* alpha = io.a;
if (alpha != null) {
int j;
int mb_w = io.mb_w;
int mb_h = io.mb_h;
WebPYUVABuffer* buf = &p.output.u.YUVA;
byte* dst = buf.a + io.mb_y * buf.a_stride;
for (j = 0; j < mb_h; ++j) {
memcpy(dst, alpha, mb_w * sizeof(*dst));
alpha += io.width;
dst += buf.a_stride;
}
}
return 0;
}
int VP8ProcessRow(VP8Io io)
{
int ok = 1;
VP8ThreadContext* ctx = &this.thread_ctx_;
if (!this.use_threads_)
{
// ctx.id_ and ctx.f_info_ are already set
ctx.mb_y_ = this.mb_y_;
ctx.filter_row_ = this.filter_row_;
ok = FinishRow(dec, io);
}
else
{
WebPWorker* worker = &this.worker_;
// Finish previous job *before* updating context
ok &= WebPWorkerSync(worker);
assert(worker.status_ == OK);
if (ok)
{ // spawn a new deblocking/output job
ctx.io_ = *io;
ctx.id_ = this.cache_id_;
ctx.mb_y_ = this.mb_y_;
ctx.filter_row_ = this.filter_row_;
if (ctx.filter_row_)
{ // just swap filter info
VP8FInfo* tmp = ctx.f_info_;
ctx.f_info_ = this.f_info_;
this.f_info_ = tmp;
}
WebPWorkerLaunch(worker);
if (++this.cache_id_ == this.num_caches_)
{
this.cache_id_ = 0;
}
}
}
return ok;
}
int VP8InitFrame(VP8Io io)
{
if (!InitThreadContext(dec)) return 0; // call first. Sets this.num_caches_.
if (!AllocateMemory(dec)) return 0;
InitIo(dec, io);
VP8DspInit(); // Init critical function pointers and look-up tables.
return 1;
}
//------------------------------------------------------------------------------
// Default custom functions
// Setup crop_xxx fields, mb_w and mb_h
static int InitFromOptions(WebPDecoderOptions* options,
VP8Io* io) {
int W = io.width;
int H = io.height;
int x = 0, y = 0, w = W, h = H;
// Cropping
io.use_cropping = (options != null) && (options.use_cropping > 0);
if (io.use_cropping) {
w = options.crop_width;
h = options.crop_height;
// TODO(skal): take colorspace into account. Don't assume YUV420.
x = options.crop_left & ~1;
y = options.crop_top & ~1;
if (x < 0 || y < 0 || w <= 0 || h <= 0 || x + w > W || y + h > H) {
return 0; // out of frame boundary error
}
}
io.crop_left = x;
io.crop_top = y;
io.crop_right = x + w;
io.crop_bottom = y + h;
io.mb_w = w;
io.mb_h = h;
// Scaling
io.use_scaling = (options != null) && (options.use_scaling > 0);
if (io.use_scaling) {
if (options.scaled_width <= 0 || options.scaled_height <= 0) {
return 0;
}
io.scaled_width = options.scaled_width;
io.scaled_height = options.scaled_height;
}
// Filter
io.bypass_filtering = options && options.bypass_filtering;
// Fancy upsampler
#if FANCY_UPSAMPLING
io.fancy_upsampling = (options == null) || (!options.no_fancy_upsampling);
#endif
if (io.use_scaling) {
// disable filter (only for large downscaling ratio).
io.bypass_filtering = (io.scaled_width < W * 3 / 4) &&
(io.scaled_height < H * 3 / 4);
io.fancy_upsampling = 0;
}
return 1;
}
static int EmitRescaledYUV(VP8Io* io, WebPDecParams* p) {
int mb_h = io.mb_h;
int uv_mb_h = (mb_h + 1) >> 1;
int num_lines_out = Rescale(io.y, io.y_stride, mb_h, &p.scaler_y);
Rescale(io.u, io.uv_stride, uv_mb_h, &p.scaler_u);
Rescale(io.v, io.uv_stride, uv_mb_h, &p.scaler_v);
return num_lines_out;
}
static int CustomSetup(VP8Io* io) {
WebPDecParams* p = (WebPDecParams*)io.opaque;
int is_rgb = (p.output.colorspace < MODE_YUV);
p.memory = null;
p.emit = null;
p.emit_alpha = null;
if (!InitFromOptions(p.options, io)) {
return 0;
}
if (io.use_scaling) {
int ok = is_rgb ? InitRGBRescaler(io, p) : InitYUVRescaler(io, p);
if (!ok) {
return 0; // memory error
}
} else {
if (is_rgb) {
p.emit = EmitSampledRGB; // default
#if FANCY_UPSAMPLING
if (io.fancy_upsampling) {
int uv_width = (io.mb_w + 1) >> 1;
p.memory = malloc(io.mb_w + 2 * uv_width);
if (p.memory == null) {
return 0; // memory error.
}
p.tmp_y = (byte*)p.memory;
p.tmp_u = p.tmp_y + io.mb_w;
p.tmp_v = p.tmp_u + uv_width;
p.emit = EmitFancyRGB;
WebPInitUpsamplers();
}
#endif
} else {
p.emit = EmitYUV;
}
if (IsAlphaMode(p.output.colorspace)) {
// We need transparency output
p.emit_alpha =
is_rgb ? (p.output.colorspace == MODE_RGBA_4444 ?
EmitAlphaRGBA4444 : EmitAlphaRGB) : EmitAlphaYUV;
}
}
if (is_rgb) {
VP8YUVInit();
}
return 1;
}
static int EmitRescaledAlphaYUV(VP8Io* io, WebPDecParams* p) {
if (io.a != null) {
Rescale(io.a, io.width, io.mb_h, &p.scaler_a);
}
return 0;
}
//------------------------------------------------------------------------------
static int CustomPut(VP8Io* io) {
WebPDecParams* p = (WebPDecParams*)io.opaque;
int mb_w = io.mb_w;
int mb_h = io.mb_h;
int num_lines_out;
assert(!(io.mb_y & 1));
if (mb_w <= 0 || mb_h <= 0) {
return 0;
}
num_lines_out = p.emit(io, p);
if (p.emit_alpha) {
p.emit_alpha(io, p);
}
p.last_y += num_lines_out;
return 1;
}
void InitIo(VP8Io io)
{
// prepare 'io'
io.mb_y = 0;
io.y = this.cache_y_;
io.u = this.cache_u_;
io.v = this.cache_v_;
io.y_stride = this.cache_y_stride_;
io.uv_stride = this.cache_uv_stride_;
io.fancy_upsampling = 0; // default
io.a = null;
}
//------------------------------------------------------------------------------
// Fancy upsampling
#if FANCY_UPSAMPLING
static int EmitFancyRGB(VP8Io* io, WebPDecParams* p) {
int num_lines_out = io.mb_h; // a priori guess
WebPRGBABuffer* buf = &p.output.u.RGBA;
byte* dst = buf.rgba + io.mb_y * buf.stride;
WebPUpsampleLinePairFunc upsample =
io.a ? WebPUpsamplersKeepAlpha[p.output.colorspace]
: WebPUpsamplers[p.output.colorspace];
byte* cur_y = io.y;
byte* cur_u = io.u;
byte* cur_v = io.v;
byte* top_u = p.tmp_u;
byte* top_v = p.tmp_v;
int y = io.mb_y;
int y_end = io.mb_y + io.mb_h;
int mb_w = io.mb_w;
int uv_w = (mb_w + 1) / 2;
if (y == 0) {
// First line is special cased. We mirror the u/v samples at boundary.
upsample(null, cur_y, cur_u, cur_v, cur_u, cur_v, null, dst, mb_w);
} else {
// We can finish the left-over line from previous call.
// Warning! Don't overwrite the alpha values (if any), as they
// are not lagging one line behind but are already written.
upsample(p.tmp_y, cur_y, top_u, top_v, cur_u, cur_v,
dst - buf.stride, dst, mb_w);
++num_lines_out;
}
// Loop over each output pairs of row.
for (; y + 2 < y_end; y += 2) {
top_u = cur_u;
top_v = cur_v;
cur_u += io.uv_stride;
cur_v += io.uv_stride;
dst += 2 * buf.stride;
cur_y += 2 * io.y_stride;
upsample(cur_y - io.y_stride, cur_y,
top_u, top_v, cur_u, cur_v,
dst - buf.stride, dst, mb_w);
}
// move to last row
cur_y += io.y_stride;
if (io.crop_top + y_end < io.crop_bottom) {
// Save the unfinished samples for next call (as we're not done yet).
memcpy(p.tmp_y, cur_y, mb_w * sizeof(*p.tmp_y));
memcpy(p.tmp_u, cur_u, uv_w * sizeof(*p.tmp_u));
memcpy(p.tmp_v, cur_v, uv_w * sizeof(*p.tmp_v));
// The fancy upsampler leaves a row unfinished behind
// (except for the very last row)
num_lines_out--;
} else {
// Process the very last row of even-sized picture
if (!(y_end & 1)) {
upsample(cur_y, null, cur_u, cur_v, cur_u, cur_v,
dst + buf.stride, null, mb_w);
}
}
return num_lines_out;
}
static int InitYUVRescaler(VP8Io* io, WebPDecParams* p) {
int has_alpha = IsAlphaMode(p.output.colorspace);
WebPYUVABuffer* buf = &p.output.u.YUVA;
int out_width = io.scaled_width;
int out_height = io.scaled_height;
int uv_out_width = (out_width + 1) >> 1;
int uv_out_height = (out_height + 1) >> 1;
int uv_in_width = (io.mb_w + 1) >> 1;
int uv_in_height = (io.mb_h + 1) >> 1;
uint work_size = 2 * out_width; // scratch memory for luma rescaler
uint uv_work_size = 2 * uv_out_width; // and for each u/v ones
uint tmp_size;
int* work;
tmp_size = work_size + 2 * uv_work_size;
if (has_alpha) {
tmp_size += work_size;
}
p.memory = calloc(1, tmp_size * sizeof(*work));
if (p.memory == null) {
return 0; // memory error
}
work = (int*)p.memory;
InitRescaler(&p.scaler_y, io.mb_w, io.mb_h,
buf.y, out_width, out_height, buf.y_stride,
io.mb_w, out_width, io.mb_h, out_height,
work);
InitRescaler(&p.scaler_u, uv_in_width, uv_in_height,
buf.u, uv_out_width, uv_out_height, buf.u_stride,
uv_in_width, uv_out_width,
uv_in_height, uv_out_height,
work + work_size);
InitRescaler(&p.scaler_v, uv_in_width, uv_in_height,
buf.v, uv_out_width, uv_out_height, buf.v_stride,
uv_in_width, uv_out_width,
uv_in_height, uv_out_height,
work + work_size + uv_work_size);
p.emit = EmitRescaledYUV;
if (has_alpha) {
InitRescaler(&p.scaler_a, io.mb_w, io.mb_h,
buf.a, out_width, out_height, buf.a_stride,
io.mb_w, out_width, io.mb_h, out_height,
work + work_size + 2 * uv_work_size);
p.emit_alpha = EmitRescaledAlphaYUV;
}
return 1;
}
/// <summary>
/// Finalize and transmit a complete row. Return false in case of user-abort.
/// </summary>
/// <param name="dec"></param>
/// <param name="io"></param>
/// <returns></returns>
int FinishRow(VP8Io io)
{
int ok = 1;
VP8ThreadContext ctx = this.thread_ctx_;
int extra_y_rows = kFilterExtraRows[this.filter_type_];
int ysize = extra_y_rows * this.cache_y_stride_;
int uvsize = (extra_y_rows / 2) * this.cache_uv_stride_;
int y_offset = ctx.id_ * 16 * this.cache_y_stride_;
int uv_offset = ctx.id_ * 8 * this.cache_uv_stride_;
byte* ydst = this.cache_y_ - ysize + y_offset;
byte* udst = this.cache_u_ - uvsize + uv_offset;
byte* vdst = this.cache_v_ - uvsize + uv_offset;
bool first_row = (ctx.mb_y_ == 0);
bool last_row = (ctx.mb_y_ >= this.br_mb_y_ - 1);
int y_start = MACROBLOCK_VPOS(ctx.mb_y_);
int y_end = MACROBLOCK_VPOS(ctx.mb_y_ + 1);
if (ctx.filter_row_)
{
FilterRow(dec);
}
if (io.put)
{
if (!first_row)
{
y_start -= extra_y_rows;
io.y = ydst;
io.u = udst;
io.v = vdst;
}
else
{
io.y = this.cache_y_ + y_offset;
io.u = this.cache_u_ + uv_offset;
io.v = this.cache_v_ + uv_offset;
}
if (!last_row)
{
y_end -= extra_y_rows;
}
if (y_end > io.crop_bottom)
{
y_end = io.crop_bottom; // make sure we don't overflow on last row.
}
io.a = null;
if (this.alpha_data_ && y_start < y_end)
{
io.a = VP8DecompressAlphaRows(dec, y_start, y_end - y_start);
if (io.a == null)
{
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
"Could not decode alpha data.");
}
}
if (y_start < io.crop_top)
{
int delta_y = io.crop_top - y_start;
y_start = io.crop_top;
assert(!(delta_y & 1));
io.y += this.cache_y_stride_ * delta_y;
io.u += this.cache_uv_stride_ * (delta_y >> 1);
io.v += this.cache_uv_stride_ * (delta_y >> 1);
if (io.a)
{
io.a += io.width * delta_y;
}
}
if (y_start < y_end)
{
io.y += io.crop_left;
io.u += io.crop_left >> 1;
io.v += io.crop_left >> 1;
if (io.a)
{
io.a += io.crop_left;
}
io.mb_y = y_start - io.crop_top;
io.mb_w = io.crop_right - io.crop_left;
io.mb_h = y_end - y_start;
ok = io.put(io);
}
}
// rotate top samples if needed
if (ctx.id_ + 1 == this.num_caches_)
{
if (!last_row)
{
memcpy(this.cache_y_ - ysize, ydst + 16 * this.cache_y_stride_, ysize);
memcpy(this.cache_u_ - uvsize, udst + 8 * this.cache_uv_stride_, uvsize);
memcpy(this.cache_v_ - uvsize, vdst + 8 * this.cache_uv_stride_, uvsize);
}
}
return ok;
}
/// <summary>
/// Finish setting up the decoding parameter once user's setup() is called.
/// </summary>
/// <param name="io"></param>
/// <returns></returns>
VP8StatusCode VP8EnterCritical(VP8Io io)
{
// Call setup() first. This may trigger additional decoding features on 'io'.
// Note: Afterward, we must call teardown() not matter what.
if (io.setup && !io.setup(io))
{
VP8SetError(dec, VP8_STATUS_USER_ABORT, "Frame setup failed");
return this.status_;
}
// Disable filtering per user request
if (io.bypass_filtering)
{
this.filter_type_ = 0;
}
// TODO(skal): filter type / strength / sharpness forcing
// Define the area where we can skip in-loop filtering, in case of cropping.
//
// 'Simple' filter reads two luma samples outside of the macroblock and
// and filters one. It doesn't filter the chroma samples. Hence, we can
// avoid doing the in-loop filtering before crop_top/crop_left position.
// For the 'Complex' filter, 3 samples are read and up to 3 are filtered.
// Means: there's a dependency chain that goes all the way up to the
// top-left corner of the picture (MB #0). We must filter all the previous
// macroblocks.
// TODO(skal): add an 'approximate_decoding' option, that won't produce
// a 1:1 bit-exactness for complex filtering?
{
int extra_pixels = kFilterExtraRows[this.filter_type_];
if (this.filter_type_ == 2)
{
// For complex filter, we need to preserve the dependency chain.
this.tl_mb_x_ = 0;
this.tl_mb_y_ = 0;
}
else
{
// For simple filter, we can filter only the cropped region.
// We include 'extra_pixels' on the other side of the boundary, since
// vertical or horizontal filtering of the previous macroblock can
// modify some abutting pixels.
this.tl_mb_x_ = (io.crop_left - extra_pixels) >> 4;
this.tl_mb_y_ = (io.crop_top - extra_pixels) >> 4;
if (this.tl_mb_x_ < 0) this.tl_mb_x_ = 0;
if (this.tl_mb_y_ < 0) this.tl_mb_y_ = 0;
}
// We need some 'extra' pixels on the right/bottom.
this.br_mb_y_ = (io.crop_bottom + 15 + extra_pixels) >> 4;
this.br_mb_x_ = (io.crop_right + 15 + extra_pixels) >> 4;
if (this.br_mb_x_ > this.mb_w_)
{
this.br_mb_x_ = this.mb_w_;
}
if (this.br_mb_y_ > this.mb_h_)
{
this.br_mb_y_ = this.mb_h_;
}
}
return VP8StatusCode.VP8_STATUS_OK;
}