[Ffmpeg-devel] I'm giving up
Panagiotis Issaris
takis.issaris
Fri Dec 8 23:00:48 CET 2006
Hi Michael,
On Thu, Dec 07, 2006 at 07:06:53PM +0100, Michael Niedermayer wrote:
> > >[...]
> > > > +static uint8_t *h264_write_nal_unit(int nal_ref_idc, int nal_unit_type, uint8_t *dest, int *destsize,
> > > > + PutBitContext *b2)
> > > > +{
> > > > + PutBitContext b;
> > > > + int i, destpos, rbsplen, escape_count;
> > > > + uint8_t *rbsp;
> > > > +
> > > > + // Align b2 on a byte boundary
> > > > +
> > > > + align_put_bits(b2);
> > >
> > > is the rbsp trailing stuff correct? shouldnt there be a put_bits(1,1) ?
> > Yes, but we had the put_bits(1,1) call in the calling context. For
> > end-of-stream NAL units, the rbsp should be empty, so in that case we
> > did not add the stopbit.
> >
> > How about something like this? (Tested and works, but the reference
> > decoder also decodes it if the extra bit is there.)
>
> iam fine with any solution as long as it conforms to the h.264 spec
Ok.
> > > [...]
> > > > + block[3][3] = pieces[3][0]-(pieces[3][1]<<1)+(pieces[3][2]<<1)-pieces[3][3];
> > >
> > > theres are alot of redundant operations in the above, these should be
> > > simplified
> > I haven't spotted the redundant operations yet, but I'll go over them
> > tomorrow (or this evening).
>
> pieces[0][0] = block[0][0]+block[1][0]+block[2][0]+block[3][0];
> pieces[1][0] = (block[0][0]<<1)+block[1][0]-block[2][0]-(block[3][0]<<1);
> pieces[2][0] = block[0][0]-block[1][0]-block[2][0]+block[3][0];
> pieces[3][0] = block[0][0]-(block[1][0]<<1)+(block[2][0]<<1)-block[3][0];
>
> vs .
>
> A = block[0][0]+block[3][0];
> C = block[0][0]-block[3][0];
> B = block[1][0]+block[2][0];
> D = block[1][0]-block[2][0];
> pieces[0][0] = A+B;
> pieces[2][0] = A-B;
> pieces[1][0] = (C<<1)+ D;
> pieces[3][0] = C -(D<<1);
Thanks! :)
Continuing using your suggestion and measuring using START|STOP_TIMER showed
that it brought down the decicycles from over 2700 down to 1911!
> > > [...]
> > > > diff --git a/libavcodec/Makefile b/libavcodec/Makefile
> > > > index 03c1ae4..ac369eb 100644
> > > > --- a/libavcodec/Makefile
> > > > +++ b/libavcodec/Makefile
> > > > @@ -86,6 +86,7 @@ OBJS-$(CONFIG_GIF_ENCODER) +
> > > > OBJS-$(CONFIG_H261_DECODER) += h261.o
> > > > OBJS-$(CONFIG_H261_ENCODER) += h261.o
> > > > OBJS-$(CONFIG_H264_DECODER) += h264.o
> > > > +OBJS-$(CONFIG_H264_ENCODER) += h264enc.o h264cavlc.o h264dsp.o
> > > > OBJS-$(CONFIG_HUFFYUV_DECODER) += huffyuv.o
> > > > OBJS-$(CONFIG_HUFFYUV_ENCODER) += huffyuv.o
> > > > OBJS-$(CONFIG_IDCIN_DECODER) += idcinvideo.o
> > >
> > > this can be commited if it does not break compilation
> > By adding a stub for h264cavlc.c or by disabling compilation by default?
>
> by waiting until h264cavlc.c is in svn IMHO
Ok. Or what would you think about adding the line with only the currently added
files?
With friendly regards,
Takis
-------------- next part --------------
Index: libavcodec/dsputil.c
===================================================================
--- libavcodec/dsputil.c (revision 7262)
+++ libavcodec/dsputil.c (working copy)
@@ -2549,6 +2549,11 @@
}
#endif /* CONFIG_VC1_DECODER||CONFIG_WMV3_DECODER */
+#if defined(CONFIG_H264_ENCODER)
+/* H264 specific */
+void ff_h264dsp_init(DSPContext* c, AVCodecContext *avctx);
+#endif /* CONFIG_H264_ENCODER */
+
static void wmv2_mspel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int w){
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
int i;
@@ -4027,6 +4032,9 @@
#if defined(CONFIG_VC1_DECODER) || defined(CONFIG_WMV3_DECODER)
ff_vc1dsp_init(c,avctx);
#endif
+#if defined(CONFIG_H264_ENCODER)
+ ff_h264dsp_init(c,avctx);
+#endif
c->put_mspel_pixels_tab[0]= put_mspel8_mc00_c;
c->put_mspel_pixels_tab[1]= put_mspel8_mc10_c;
Index: libavcodec/h264dsp.c
===================================================================
--- libavcodec/h264dsp.c (revision 0)
+++ libavcodec/h264dsp.c (revision 0)
@@ -0,0 +1,122 @@
+/*
+ * H.264/MPEG-4 Part 10 (Base profile) encoder.
+ *
+ * DSP functions
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file h264dsp.c
+ * H.264 encoder related DSP utils
+ *
+ */
+
+#include "dsputil.h"
+
+extern const uint8_t ff_div6[52];
+extern const uint8_t ff_rem6[52];
+
+/**
+ * Transform the provided matrix using the H.264 modified DCT.
+ * @note
+ * we'll always work with transposed input blocks, to avoid having to make a
+ * distinction between C and mmx implementations.
+ *
+ * @param block transposed input block
+ */
+static void h264_dct_c(DCTELEM block[4][4])
+{
+ DCTELEM pieces[4][4];
+ DCTELEM a, b, c, d;
+
+ a = block[0][0]+block[3][0];
+ c = block[0][0]-block[3][0];
+ b = block[1][0]+block[2][0];
+ d = block[1][0]-block[2][0];
+ pieces[0][0] = a+b;
+ pieces[2][0] = a-b;
+ pieces[1][0] = (c<<1)+d;
+ pieces[3][0] = c-(d<<1);
+
+ a = block[0][1]+block[3][1];
+ c = block[0][1]-block[3][1];
+ b = block[1][1]+block[2][1];
+ d = block[1][1]-block[2][1];
+ pieces[0][1] = a+b;
+ pieces[2][1] = a-b;
+ pieces[1][1] = (c<<1)+d;
+ pieces[3][1] = c-(d<<1);
+
+ a = block[0][2]+block[3][2];
+ b = block[1][2]+block[2][2];
+ c = block[0][2]-block[3][2];
+ d = block[1][2]-block[2][2];
+ pieces[0][2] = a+b;
+ pieces[2][2] = a-b;
+ pieces[1][2] = (c<<1)+d;
+ pieces[3][2] = c-(d<<1);
+
+ a = block[0][3]+block[3][3];
+ b = block[1][3]+block[2][3];
+ c = block[0][3]-block[3][3];
+ d = block[1][3]-block[2][3];
+ pieces[0][3] = a+b;
+ pieces[1][3] = (c<<1)+d;
+ pieces[2][3] = a-b;
+ pieces[3][3] = c-(d<<1);
+
+ a = pieces[0][0]+pieces[0][3];
+ c = pieces[0][0]-pieces[0][3];
+ b = pieces[0][1]+pieces[0][2];
+ d = pieces[0][1]-pieces[0][2];
+ block[0][0] = a+b;
+ block[2][0] = a-b;
+ block[1][0] = (c<<1)+d;
+ block[3][0] = c-(d<<1);
+
+ a = pieces[1][0]+pieces[1][3];
+ c = pieces[1][0]-pieces[1][3];
+ b = pieces[1][1]+pieces[1][2];
+ d = pieces[1][1]-pieces[1][2];
+ block[0][1] = a+b;
+ block[2][1] = a-b;
+ block[1][1] = (c<<1)+d;
+ block[3][1] = c-(d<<1);
+
+ a = pieces[2][0]+pieces[2][3];
+ c = pieces[2][0]-pieces[2][3];
+ b = pieces[2][1]+pieces[2][2];
+ d = pieces[2][1]-pieces[2][2];
+ block[0][2] = a+b;
+ block[2][2] = a-b;
+ block[1][2] = (c<<1)+d;
+ block[3][2] = c-(d<<1);
+
+ a = pieces[3][0]+pieces[3][3];
+ c = pieces[3][0]-pieces[3][3];
+ b = pieces[3][1]+pieces[3][2];
+ d = pieces[3][1]-pieces[3][2];
+ block[0][3] = a+b;
+ block[2][3] = a-b;
+ block[1][3] = (c<<1)+d;
+ block[3][3] = c-(d<<1);
+}
+
+void ff_h264dsp_init(DSPContext* c, AVCodecContext *avctx)
+{
+ c->h264_dct = h264_dct_c;
+}
+
Index: libavcodec/Makefile
===================================================================
--- libavcodec/Makefile (revision 7262)
+++ libavcodec/Makefile (working copy)
@@ -86,6 +86,7 @@
OBJS-$(CONFIG_H261_DECODER) += h261.o
OBJS-$(CONFIG_H261_ENCODER) += h261.o
OBJS-$(CONFIG_H264_DECODER) += h264.o
+OBJS-$(CONFIG_H264_ENCODER) += h264enc.o h264dsp.o
OBJS-$(CONFIG_HUFFYUV_DECODER) += huffyuv.o
OBJS-$(CONFIG_HUFFYUV_ENCODER) += huffyuv.o
OBJS-$(CONFIG_IDCIN_DECODER) += idcinvideo.o
Index: libavcodec/h264enc.c
===================================================================
--- libavcodec/h264enc.c (revision 0)
+++ libavcodec/h264enc.c (revision 0)
@@ -0,0 +1,107 @@
+/*
+ * H.264 encoder
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+
+#include "common.h"
+#include "bitstream.h"
+#include "mpegvideo.h"
+#include "h264data.h"
+
+/**
+ * Write out the provided data into a NAL unit.
+ * @param nal_ref_idc NAL reference IDC
+ * @param nal_unit_type NAL unit payload type
+ * @param dest the target buffer, dst+1 == src is allowed as a special case
+ * @param destsize the length of the dst array
+ * @param b2 the data which should be escaped
+ * @returns pointer to current position in the output buffer or NULL if an error occured
+ */
+static uint8_t *h264_write_nal_unit(int nal_ref_idc, int nal_unit_type, uint8_t *dest, int *destsize,
+ PutBitContext *b2)
+{
+ PutBitContext b;
+ int i, destpos, rbsplen, escape_count;
+ uint8_t *rbsp;
+
+ if (nal_unit_type != NAL_END_STREAM)
+ put_bits(b2,1,1); // rbsp_stop_bit
+
+ // Align b2 on a byte boundary
+ align_put_bits(b2);
+ rbsplen = put_bits_count(b2)/8;
+ flush_put_bits(b2);
+ rbsp = b2->buf;
+
+ init_put_bits(&b,dest,*destsize);
+
+ put_bits(&b,16,0);
+ put_bits(&b,16,0x01);
+
+ put_bits(&b,1,0); // forbidden zero bit
+ put_bits(&b,2,nal_ref_idc); // nal_ref_idc
+ put_bits(&b,5,nal_unit_type); // nal_unit_type
+
+ flush_put_bits(&b);
+
+ destpos = 5;
+ escape_count= 0;
+
+ for (i=0; i<rbsplen; i+=2)
+ {
+ if (rbsp[i]) continue;
+ if (i>0 && rbsp[i-1]==0)
+ i--;
+ if (i+2<rbsplen && rbsp[i+1]==0 && rbsp[i+2]<=3)
+ {
+ escape_count++;
+ i+=2;
+ }
+ }
+
+ if(escape_count==0)
+ {
+ if(dest+destpos != rbsp)
+ {
+ memcpy(dest+destpos, rbsp, rbsplen);
+ *destsize -= (rbsplen+destpos);
+ }
+ return dest+rbsplen+destpos;
+ }
+
+ if(rbsplen + escape_count + 1> *destsize)
+ {
+ av_log(NULL, AV_LOG_ERROR, "Destination buffer too small!\n");
+ return NULL;
+ }
+
+ // this should be damn rare (hopefully)
+ for (i = 0 ; i < rbsplen ; i++)
+ {
+ if (i + 2 < rbsplen && (rbsp[i] == 0 && rbsp[i+1] == 0 && rbsp[i+2] < 4))
+ {
+ dest[destpos++] = rbsp[i++];
+ dest[destpos++] = rbsp[i];
+ dest[destpos++] = 0x03; // emulation prevention byte
+ }
+ else
+ dest[destpos++] = rbsp[i];
+ }
+ *destsize -= destpos;
+ return dest+destpos;
+}
+
More information about the ffmpeg-devel
mailing list