* @author Nathan Caldwell
*/
+#include <inttypes.h>
+
#include "avcodec.h"
#include "get_bits.h"
#include "mathops.h"
-#include "dsputil.h"
+#include "huffyuvdsp.h"
#include "lagarithrac.h"
+#include "thread.h"
enum LagarithFrameType {
FRAME_RAW = 1, /**< uncompressed */
typedef struct LagarithContext {
AVCodecContext *avctx;
- AVFrame picture;
- DSPContext dsp;
+ HuffYUVDSPContext hdsp;
int zeros; /**< number of consecutive zero bytes encountered */
int zeros_rem; /**< number of zero bytes remaining to output */
uint8_t *rgb_planes;
+ int rgb_planes_allocated;
int rgb_stride;
} LagarithContext;
/**
- * Compute the 52bit mantissa of 1/(double)denom.
+ * Compute the 52-bit mantissa of 1/(double)denom.
* This crazy format uses floats in an entropy coder and we have to match x86
* rounding exactly, thus ordinary floats aren't portable enough.
* @param denom denominator
- * @return 52bit mantissa
+ * @return 52-bit mantissa
* @see softfloat_mul
*/
static uint64_t softfloat_reciprocal(uint32_t denom)
/**
* (uint32_t)(x*f), where f has the given mantissa, and exponent 0
* Used in combination with softfloat_reciprocal computes x/(double)denom.
- * @param x 32bit integer factor
+ * @param x 32-bit integer factor
* @param mantissa mantissa of f with exponent 0
- * @return 32bit integer value (x*f)
+ * @return 32-bit integer value (x*f)
* @see softfloat_reciprocal
*/
static uint32_t softfloat_mul(uint32_t x, uint64_t mantissa)
}
val = get_bits_long(gb, bits);
- val |= 1 << bits;
+ val |= 1U << bits;
*value = val - 1;
av_log(rac->avctx, AV_LOG_ERROR, "Invalid probability run encountered.\n");
return -1;
}
- if (prob > 257 - i)
- prob = 257 - i;
+ if (prob > 256 - i)
+ prob = 256 - i;
for (j = 0; j < prob; j++)
rac->prob[++i] = 0;
}
if (cumul_prob & (cumul_prob - 1)) {
uint64_t mul = softfloat_reciprocal(cumul_prob);
- for (i = 1; i < 257; i++) {
+ for (i = 1; i <= 128; i++) {
+ rac->prob[i] = softfloat_mul(rac->prob[i], mul);
+ scaled_cumul_prob += rac->prob[i];
+ }
+ if (scaled_cumul_prob <= 0) {
+ av_log(rac->avctx, AV_LOG_ERROR, "Scaled probabilities invalid\n");
+ return AVERROR_INVALIDDATA;
+ }
+ for (; i < 257; i++) {
rac->prob[i] = softfloat_mul(rac->prob[i], mul);
scaled_cumul_prob += rac->prob[i];
}
}
/* Comment from reference source:
* if (b & 0x80 == 0) { // order of operations is 'wrong'; it has been left this way
- * // since the compression change is negligable and fixing it
- * // breaks backwards compatibilty
+ * // since the compression change is negligible and fixing it
+ * // breaks backwards compatibility
* b =- (signed int)b;
* b &= 0xFF;
* } else {
uint8_t *diff, int w, int *left,
int *left_top)
{
- /* This is almost identical to add_hfyu_median_prediction in dsputil.h.
- * However the &0xFF on the gradient predictor yealds incorrect output
+ /* This is almost identical to add_hfyu_median_pred in huffyuvdsp.h.
+ * However the &0xFF on the gradient predictor yields incorrect output
* for lagarith.
*/
int i;
if (!line) {
/* Left prediction only for first line */
- L = l->dsp.add_hfyu_left_prediction(buf + 1, buf + 1,
- width - 1, buf[0]);
+ L = l->hdsp.add_hfyu_left_pred(buf, buf, width, 0);
+ } else {
+ /* Left pixel is actually prev_row[width] */
+ L = buf[width - stride - 1];
+
+ if (line == 1) {
+ /* Second line, left predict first pixel, the rest of the line is median predicted
+ * NOTE: In the case of RGB this pixel is top predicted */
+ TL = l->avctx->pix_fmt == AV_PIX_FMT_YUV420P ? buf[-stride] : L;
+ } else {
+ /* Top left is 2 rows back, last pixel */
+ TL = buf[width - (2 * stride) - 1];
+ }
+
+ add_lag_median_prediction(buf, buf - stride, buf,
+ width, &L, &TL);
+ }
+}
+
+static void lag_pred_line_yuy2(LagarithContext *l, uint8_t *buf,
+ int width, int stride, int line,
+ int is_luma)
+{
+ int L, TL;
+
+ if (!line) {
+ L= buf[0];
+ if (is_luma)
+ buf[0] = 0;
+ l->hdsp.add_hfyu_left_pred(buf, buf, width, 0);
+ if (is_luma)
+ buf[0] = L;
return;
}
- /* Left pixel is actually prev_row[width] */
- L = buf[width - stride - 1];
if (line == 1) {
- /* Second line, left predict first pixel, the rest of the line is median predicted
- * NOTE: In the case of RGB this pixel is top predicted */
- TL = l->avctx->pix_fmt == PIX_FMT_YUV420P ? buf[-stride] : L;
+ const int HEAD = is_luma ? 4 : 2;
+ int i;
+
+ L = buf[width - stride - 1];
+ TL = buf[HEAD - stride - 1];
+ for (i = 0; i < HEAD; i++) {
+ L += buf[i];
+ buf[i] = L;
+ }
+ for (; i < width; i++) {
+ L = mid_pred(L & 0xFF, buf[i - stride], (L + buf[i - stride] - TL) & 0xFF) + buf[i];
+ TL = buf[i - stride];
+ buf[i] = L;
+ }
} else {
- /* Top left is 2 rows back, last pixel */
TL = buf[width - (2 * stride) - 1];
+ L = buf[width - stride - 1];
+ l->hdsp.add_hfyu_median_pred(buf, buf - stride, buf, width, &L, &TL);
}
-
- add_lag_median_prediction(buf, buf - stride, buf,
- width, &L, &TL);
}
static int lag_decode_line(LagarithContext *l, lag_rac *rac,
}
static int lag_decode_zero_run_line(LagarithContext *l, uint8_t *dst,
- const uint8_t *src, const uint8_t *srcend, int width,
- int esc_count)
+ const uint8_t *src, const uint8_t *src_end,
+ int width, int esc_count)
{
int i = 0;
int count;
uint8_t zero_run = 0;
- const uint8_t *start = src;
+ const uint8_t *src_start = src;
uint8_t mask1 = -(esc_count < 2);
uint8_t mask2 = -(esc_count < 3);
uint8_t *end = dst + (width - 2);
+ avpriv_request_sample(l->avctx, "zero_run_line");
+
+ memset(dst, 0, width);
+
output_zeros:
if (l->zeros_rem) {
count = FFMIN(l->zeros_rem, width - i);
+ if (end - dst < count) {
+ av_log(l->avctx, AV_LOG_ERROR, "Too many zeros remaining.\n");
+ return AVERROR_INVALIDDATA;
+ }
+
memset(dst, 0, count);
l->zeros_rem -= count;
dst += count;
i = 0;
while (!zero_run && dst + i < end) {
i++;
- if (i+2 >= srcend - src)
- return src - start;
+ if (i+2 >= src_end - src)
+ return AVERROR_INVALIDDATA;
zero_run =
!(src[i] | (src[i + 1] & mask1) | (src[i + 2] & mask2));
}
} else {
memcpy(dst, src, i);
src += i;
+ dst += i;
}
}
- return src - start;
+ return src - src_start;
}
int esc_count;
GetBitContext gb;
lag_rac rac;
- const uint8_t *srcend = src + src_size;
+ const uint8_t *src_end = src + src_size;
+ int ret;
rac.avctx = l->avctx;
l->zeros = 0;
offset += 4;
}
- init_get_bits(&gb, src + offset, src_size * 8);
+ if ((ret = init_get_bits8(&gb, src + offset, src_size - offset)) < 0)
+ return ret;
if (lag_read_prob_header(&rac, &gb) < 0)
return -1;
if (read > length)
av_log(l->avctx, AV_LOG_WARNING,
- "Output more bytes than length (%d of %d)\n", read,
+ "Output more bytes than length (%d of %"PRIu32")\n", read,
length);
} else if (esc_count < 8) {
esc_count -= 4;
+ src ++;
+ src_size --;
if (esc_count > 0) {
/* Zero run coding only, no range coding. */
- for (i = 0; i < height; i++)
- src += lag_decode_zero_run_line(l, dst + (i * stride), src, srcend,
- width, esc_count);
+ for (i = 0; i < height; i++) {
+ int res = lag_decode_zero_run_line(l, dst + (i * stride), src,
+ src_end, width, esc_count);
+ if (res < 0)
+ return res;
+ src += res;
+ }
} else {
- if (src_size < height * width)
- return -1;
+ if (src_size < width * height)
+ return AVERROR_INVALIDDATA; // buffer not big enough
/* Plane is stored uncompressed */
for (i = 0; i < height; i++) {
memcpy(dst + (i * stride), src, width);
return -1;
}
- for (i = 0; i < height; i++) {
- lag_pred_line(l, dst, width, stride, i);
- dst += stride;
+ if (l->avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
+ for (i = 0; i < height; i++) {
+ lag_pred_line(l, dst, width, stride, i);
+ dst += stride;
+ }
+ } else {
+ for (i = 0; i < height; i++) {
+ lag_pred_line_yuy2(l, dst, width, stride, i,
+ width == l->avctx->width);
+ dst += stride;
+ }
}
return 0;
* @return number of consumed bytes on success or negative if decode fails
*/
static int lag_decode_frame(AVCodecContext *avctx,
- void *data, int *data_size, AVPacket *avpkt)
+ void *data, int *got_frame, AVPacket *avpkt)
{
const uint8_t *buf = avpkt->data;
unsigned int buf_size = avpkt->size;
LagarithContext *l = avctx->priv_data;
- AVFrame *const p = &l->picture;
+ ThreadFrame frame = { .f = data };
+ AVFrame *const p = data;
uint8_t frametype = 0;
uint32_t offset_gu = 0, offset_bv = 0, offset_ry = 9;
- int offs[4];
+ uint32_t offs[4];
uint8_t *srcs[4], *dst;
int i, j, planes = 3;
+ int ret;
- AVFrame *picture = data;
-
- if (p->data[0])
- avctx->release_buffer(avctx, p);
-
- p->reference = 0;
p->key_frame = 1;
frametype = buf[0];
switch (frametype) {
case FRAME_SOLID_RGBA:
- avctx->pix_fmt = PIX_FMT_RGB32;
+ avctx->pix_fmt = AV_PIX_FMT_RGB32;
+ case FRAME_SOLID_GRAY:
+ if (frametype == FRAME_SOLID_GRAY)
+ if (avctx->bits_per_coded_sample == 24) {
+ avctx->pix_fmt = AV_PIX_FMT_RGB24;
+ } else {
+ avctx->pix_fmt = AV_PIX_FMT_0RGB32;
+ planes = 4;
+ }
- if (avctx->get_buffer(avctx, p) < 0) {
- av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
- return -1;
- }
+ if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
+ return ret;
dst = p->data[0];
+ if (frametype == FRAME_SOLID_RGBA) {
for (j = 0; j < avctx->height; j++) {
for (i = 0; i < avctx->width; i++)
AV_WN32(dst + i * 4, offset_gu);
dst += p->linesize[0];
}
+ } else {
+ for (j = 0; j < avctx->height; j++) {
+ memset(dst, buf[1], avctx->width * planes);
+ dst += p->linesize[0];
+ }
+ }
+ break;
+ case FRAME_SOLID_COLOR:
+ if (avctx->bits_per_coded_sample == 24) {
+ avctx->pix_fmt = AV_PIX_FMT_RGB24;
+ } else {
+ avctx->pix_fmt = AV_PIX_FMT_RGB32;
+ offset_gu |= 0xFFU << 24;
+ }
+
+ if ((ret = ff_thread_get_buffer(avctx, &frame,0)) < 0)
+ return ret;
+
+ dst = p->data[0];
+ for (j = 0; j < avctx->height; j++) {
+ for (i = 0; i < avctx->width; i++)
+ if (avctx->bits_per_coded_sample == 24) {
+ AV_WB24(dst + i * 3, offset_gu);
+ } else {
+ AV_WN32(dst + i * 4, offset_gu);
+ }
+ dst += p->linesize[0];
+ }
break;
case FRAME_ARITH_RGBA:
- avctx->pix_fmt = PIX_FMT_RGB32;
+ avctx->pix_fmt = AV_PIX_FMT_RGB32;
planes = 4;
offset_ry += 4;
offs[3] = AV_RL32(buf + 9);
case FRAME_ARITH_RGB24:
- if (frametype == FRAME_ARITH_RGB24)
- avctx->pix_fmt = PIX_FMT_RGB24;
+ case FRAME_U_RGB24:
+ if (frametype == FRAME_ARITH_RGB24 || frametype == FRAME_U_RGB24)
+ avctx->pix_fmt = AV_PIX_FMT_RGB24;
- if (avctx->get_buffer(avctx, p) < 0) {
- av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
- return -1;
- }
+ if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
+ return ret;
offs[0] = offset_bv;
offs[1] = offset_gu;
offs[2] = offset_ry;
+ l->rgb_stride = FFALIGN(avctx->width, 16);
+ av_fast_malloc(&l->rgb_planes, &l->rgb_planes_allocated,
+ l->rgb_stride * avctx->height * planes + 1);
if (!l->rgb_planes) {
- l->rgb_stride = FFALIGN(avctx->width, 16);
- l->rgb_planes = av_malloc(l->rgb_stride * avctx->height * planes + 16);
- if (!l->rgb_planes) {
- av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
- return AVERROR(ENOMEM);
- }
+ av_log(avctx, AV_LOG_ERROR, "cannot allocate temporary buffer\n");
+ return AVERROR(ENOMEM);
}
for (i = 0; i < planes; i++)
srcs[i] = l->rgb_planes + (i + 1) * l->rgb_stride * avctx->height - l->rgb_stride;
- for (i = 0; i < planes; i++) {
+ for (i = 0; i < planes; i++)
if (buf_size <= offs[i]) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Invalid frame offsets\n");
return AVERROR_INVALIDDATA;
}
+
+ for (i = 0; i < planes; i++)
lag_decode_arith_plane(l, srcs[i],
avctx->width, avctx->height,
-l->rgb_stride, buf + offs[i],
buf_size - offs[i]);
- }
dst = p->data[0];
for (i = 0; i < planes; i++)
srcs[i] = l->rgb_planes + i * l->rgb_stride * avctx->height;
srcs[i] += l->rgb_stride;
}
break;
- case FRAME_ARITH_YV12:
- avctx->pix_fmt = PIX_FMT_YUV420P;
+ case FRAME_ARITH_YUY2:
+ avctx->pix_fmt = AV_PIX_FMT_YUV422P;
- if (avctx->get_buffer(avctx, p) < 0) {
- av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
- return -1;
+ if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
+ return ret;
+
+ if (offset_ry >= buf_size ||
+ offset_gu >= buf_size ||
+ offset_bv >= buf_size) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Invalid frame offsets\n");
+ return AVERROR_INVALIDDATA;
}
+
+ lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
+ p->linesize[0], buf + offset_ry,
+ buf_size - offset_ry);
+ lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
+ avctx->height, p->linesize[1],
+ buf + offset_gu, buf_size - offset_gu);
+ lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
+ avctx->height, p->linesize[2],
+ buf + offset_bv, buf_size - offset_bv);
+ break;
+ case FRAME_ARITH_YV12:
+ avctx->pix_fmt = AV_PIX_FMT_YUV420P;
+
+ if ((ret = ff_thread_get_buffer(avctx, &frame, 0)) < 0)
+ return ret;
if (buf_size <= offset_ry || buf_size <= offset_gu || buf_size <= offset_bv) {
return AVERROR_INVALIDDATA;
}
+ if (offset_ry >= buf_size ||
+ offset_gu >= buf_size ||
+ offset_bv >= buf_size) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Invalid frame offsets\n");
+ return AVERROR_INVALIDDATA;
+ }
+
lag_decode_arith_plane(l, p->data[0], avctx->width, avctx->height,
p->linesize[0], buf + offset_ry,
buf_size - offset_ry);
- lag_decode_arith_plane(l, p->data[2], avctx->width / 2,
- avctx->height / 2, p->linesize[2],
+ lag_decode_arith_plane(l, p->data[2], (avctx->width + 1) / 2,
+ (avctx->height + 1) / 2, p->linesize[2],
buf + offset_gu, buf_size - offset_gu);
- lag_decode_arith_plane(l, p->data[1], avctx->width / 2,
- avctx->height / 2, p->linesize[1],
+ lag_decode_arith_plane(l, p->data[1], (avctx->width + 1) / 2,
+ (avctx->height + 1) / 2, p->linesize[1],
buf + offset_bv, buf_size - offset_bv);
break;
default:
av_log(avctx, AV_LOG_ERROR,
- "Unsupported Lagarith frame type: %#x\n", frametype);
- return -1;
+ "Unsupported Lagarith frame type: %#"PRIx8"\n", frametype);
+ return AVERROR_PATCHWELCOME;
}
- *picture = *p;
- *data_size = sizeof(AVFrame);
+ *got_frame = 1;
return buf_size;
}
LagarithContext *l = avctx->priv_data;
l->avctx = avctx;
- ff_dsputil_init(&l->dsp, avctx);
+ ff_huffyuvdsp_init(&l->hdsp);
return 0;
}
{
LagarithContext *l = avctx->priv_data;
- if (l->picture.data[0])
- avctx->release_buffer(avctx, &l->picture);
av_freep(&l->rgb_planes);
return 0;
AVCodec ff_lagarith_decoder = {
.name = "lagarith",
+ .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
.type = AVMEDIA_TYPE_VIDEO,
- .id = CODEC_ID_LAGARITH,
+ .id = AV_CODEC_ID_LAGARITH,
.priv_data_size = sizeof(LagarithContext),
.init = lag_decode_init,
.close = lag_decode_end,
.decode = lag_decode_frame,
- .capabilities = CODEC_CAP_DR1,
- .long_name = NULL_IF_CONFIG_SMALL("Lagarith lossless"),
+ .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
};