[FFmpeg-devel] [PATCH v4 3/4] libavcodec/vulkan: Add modifications to common shader for VC2 vulkan encoder
IndecisiveTurtle
geoster3d at gmail.com
Mon May 19 20:02:21 EEST 2025
> This differs quite a lot from the software implementation: It does not
> presume that the PutBitContext is flushed and instead of simply skipping
> over the buffer it actually fills the buffer with n 0xFF bytes,
> effectively adding the memset used in the VC2 slice writing code to
> skip_put_bytes(). But this file is (if I am not mistaken) supposed to be
> generic, not vc2 specific, so this feels very wrong.
Would it be enough to move it to vc2_encode.comp or should I also
rename the function?
Στις Δευ 19 Μαΐ 2025 στις 7:46 μ.μ., ο/η Andreas Rheinhardt
<andreas.rheinhardt at outlook.com> έγραψε:
>
> IndecisiveTurtle:
> > From: IndecisiveTurtle <geoster3d at gmail.com>
> >
> > ---
> > libavcodec/vulkan/common.comp | 54 ++++++++++++++++++++++++++++-------
> > 1 file changed, 44 insertions(+), 10 deletions(-)
> >
> > diff --git a/libavcodec/vulkan/common.comp b/libavcodec/vulkan/common.comp
> > index 10af9c0623..db216a2ac6 100644
> > --- a/libavcodec/vulkan/common.comp
> > +++ b/libavcodec/vulkan/common.comp
> > @@ -18,6 +18,9 @@
> > * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> > */
> >
> > +#extension GL_EXT_buffer_reference : require
> > +#extension GL_EXT_buffer_reference2 : require
> > +
> > layout(buffer_reference, buffer_reference_align = 1) buffer u8buf {
> > uint8_t v;
> > };
> > @@ -61,22 +64,20 @@ layout(buffer_reference, buffer_reference_align = 8) buffer u64buf {
> > #define mid_pred(a, b, c) \
> > max(min((a), (b)), min(max((a), (b)), (c)))
> >
> > -/* TODO: optimize */
> > +
> > uint align(uint src, uint a)
> > {
> > - uint res = src % a;
> > - if (res == 0)
> > - return src;
> > - return src + a - res;
> > + return (src + a - 1) & ~(a - 1);
> > +}
> > +
> > +int align(int src, int a)
> > +{
> > + return (src + a - 1) & ~(a - 1);
> > }
> >
> > -/* TODO: optimize */
> > uint64_t align64(uint64_t src, uint64_t a)
> > {
> > - uint64_t res = src % a;
> > - if (res == 0)
> > - return src;
> > - return src + a - res;
> > + return (src + a - 1) & ~(a - 1);
> > }
> >
> > #define reverse4(src) \
> > @@ -167,6 +168,39 @@ uint32_t flush_put_bits(inout PutBitContext pb)
> > return uint32_t(pb.buf - pb.buf_start);
> > }
> >
> > +void skip_put_bytes(inout PutBitContext pb, int n)
> > +{
> > + int bytes_left = pb.bit_left >> 3;
> > + if (n < bytes_left)
> > + {
> > + int n_bits = n << 3;
> > + int mask = (1 << n_bits) - 1;
> > + pb.bit_buf <<= n_bits;
> > + pb.bit_buf |= mask;
> > + pb.bit_left -= uint8_t(n_bits);
> > + return;
> > + }
> > + if (pb.bit_left < BUF_BITS)
> > + {
> > + int mask = (1 << pb.bit_left) - 1;
> > + pb.bit_buf <<= pb.bit_left;
> > + pb.bit_buf |= mask;
> > + u32vec2buf(pb.buf).v = BUF_REVERSE(pb.bit_buf);
> > + pb.buf += BUF_BYTES;
> > + n -= pb.bit_left >> 3;
> > + }
> > + int skip_dwords = n >> 2;
> > + while (skip_dwords > 0)
> > + {
> > + u8vec4buf(pb.buf).v = u8vec4(0xFF);
> > + pb.buf += 4;
> > + skip_dwords--;
> > + }
> > + int skip_bits = (n & 3) << 3;
> > + pb.bit_buf = (1 << skip_bits) - 1;
> > + pb.bit_left = uint8_t(BUF_BITS - skip_bits);
> > +}
>
> This differs quite a lot from the software implementation: It does not
> presume that the PutBitContext is flushed and instead of simply skipping
> over the buffer it actually fills the buffer with n 0xFF bytes,
> effectively adding the memset used in the VC2 slice writing code to
> skip_put_bytes(). But this file is (if I am not mistaken) supposed to be
> generic, not vc2 specific, so this feels very wrong.
>
> > +
> > void init_put_bits(out PutBitContext pb, u8buf data, uint64_t len)
> > {
> > pb.buf_start = uint64_t(data);
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
More information about the ffmpeg-devel
mailing list