[FFmpeg-devel] [PATCHv2 1/4] lavc/h263dsp: add DCT dequantisation function

[FFmpeg-devel] [PATCHv2 1/4] lavc/h263dsp: add DCT dequantisation function

[Rémi Denis-Courmont]

To preserve the alignment promise of the block base address, we pass
the start offset separately (it is either 0 or 1).

Note that optimised implementations of this function will be taken into
actual use if MpegEncContext.dct_unquantize_h263_{inter,intra} are *not*
overloaded by existing optimisations.
---
 libavcodec/h263dsp.c | 17 +++++++++++++++++
 libavcodec/h263dsp.h |  2 ++
 2 files changed, 19 insertions(+)

diff --git a/libavcodec/h263dsp.c b/libavcodec/h263dsp.c
index 6a13353499..eb990f27bd 100644
--- a/libavcodec/h263dsp.c
+++ b/libavcodec/h263dsp.c
@@ -23,6 +23,22 @@
 #include "config.h"
 #include "h263dsp.h"
 
+static void h263_dct_unquantize_c(int16_t *block, size_t start, size_t end,
+                                  int qmul, int qadd)
+{
+    for (size_t i = start; i <= end; i++) {
+        int level = block[i];
+
+        if (level) {
+            if (level < 0)
+                level = level * qmul - qadd;
+            else
+                level = level * qmul + qadd;
+            block[i] = level;
+        }
+    }
+}
+
 const uint8_t ff_h263_loop_filter_strength[32] = {
     0, 1, 1, 2, 2, 3, 3,  4,  4,  4,  5,  5,  6,  6,  7, 7,
     7, 8, 8, 8, 9, 9, 9, 10, 10, 10, 11, 11, 11, 12, 12, 12
@@ -116,6 +132,7 @@ static void h263_v_loop_filter_c(uint8_t *src, int stride, int qscale)
 
 av_cold void ff_h263dsp_init(H263DSPContext *ctx)
 {
+    ctx->h263_dct_unquantize = h263_dct_unquantize_c;
     ctx->h263_h_loop_filter = h263_h_loop_filter_c;
     ctx->h263_v_loop_filter = h263_v_loop_filter_c;
 
diff --git a/libavcodec/h263dsp.h b/libavcodec/h263dsp.h
index 2dccd23392..93c128c9ac 100644
--- a/libavcodec/h263dsp.h
+++ b/libavcodec/h263dsp.h
@@ -24,6 +24,8 @@
 extern const uint8_t ff_h263_loop_filter_strength[32];
 
 typedef struct H263DSPContext {
+    void (*h263_dct_unquantize)(int16_t *block /* align 16 */, size_t offset,
+                                size_t len, int mul, int add);
     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
 } H263DSPContext;
-- 
2.45.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 2/4] lavc/mpegvideo: use H263DSP dequant function

[Rémi Denis-Courmont]

---
 libavcodec/mpegvideo.c | 30 +++++-------------------------
 1 file changed, 5 insertions(+), 25 deletions(-)

diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index 7af823b8bd..fa25d14970 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -201,7 +201,7 @@ static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
                                   int16_t *block, int n, int qscale)
 {
-    int i, level, qmul, qadd;
+    int qmul, qadd;
     int nCoeffs;
 
     av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
@@ -219,23 +219,13 @@ static void dct_unquantize_h263_intra_c(MpegEncContext *s,
     else
         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
 
-    for(i=1; i<=nCoeffs; i++) {
-        level = block[i];
-        if (level) {
-            if (level < 0) {
-                level = level * qmul - qadd;
-            } else {
-                level = level * qmul + qadd;
-            }
-            block[i] = level;
-        }
-    }
+    s->h263dsp.h263_dct_unquantize(block, 1, nCoeffs, qmul, qadd);
 }
 
 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
                                   int16_t *block, int n, int qscale)
 {
-    int i, level, qmul, qadd;
+    int qmul, qadd;
     int nCoeffs;
 
     av_assert2(s->block_last_index[n]>=0);
@@ -244,18 +234,7 @@ static void dct_unquantize_h263_inter_c(MpegEncContext *s,
     qmul = qscale << 1;
 
     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
-
-    for(i=0; i<=nCoeffs; i++) {
-        level = block[i];
-        if (level) {
-            if (level < 0) {
-                level = level * qmul - qadd;
-            } else {
-                level = level * qmul + qadd;
-            }
-            block[i] = level;
-        }
-    }
+    s->h263dsp.h263_dct_unquantize(block, 0, nCoeffs, qmul, qadd);
 }
 
 
@@ -275,6 +254,7 @@ static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 static av_cold int dct_init(MpegEncContext *s)
 {
     ff_blockdsp_init(&s->bdsp);
+    ff_h263dsp_init(&s->h263dsp);
     ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
     ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
 
-- 
2.45.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCHv2 3/4] checkasm/h263dsp: test dct_unquantize

[Rémi Denis-Courmont]

---
 tests/checkasm/h263dsp.c | 46 +++++++++++++++++++++++++++++++++++++++-
 1 file changed, 45 insertions(+), 1 deletion(-)

diff --git a/tests/checkasm/h263dsp.c b/tests/checkasm/h263dsp.c
index 2d0957a90b..546204cb28 100644
--- a/tests/checkasm/h263dsp.c
+++ b/tests/checkasm/h263dsp.c
@@ -18,13 +18,54 @@
  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
+#include <stdbool.h>
 #include <string.h>
 
 #include "checkasm.h"
 
-#include "libavcodec/h263dsp.h"
+#include "libavutil/avassert.h"
 #include "libavutil/mem.h"
 #include "libavutil/mem_internal.h"
+#include "libavcodec/h263dsp.h"
+#include "libavcodec/mpegvideodata.h"
+
+static uint_fast8_t mpeg_qscale_rnd(void)
+{
+    int n = rnd(), q = (n >> 1) & 31;
+
+    if (n & 1)
+        return ff_mpeg2_non_linear_qscale[q];
+    else
+        return q << 1;
+}
+
+static void check_dct_unquantize(H263DSPContext *ctx, bool intra)
+{
+#define LEN 64
+    LOCAL_ALIGNED_16(int16_t, block0, [LEN]);
+    LOCAL_ALIGNED_16(int16_t, block1, [LEN]);
+    size_t end = rnd() % LEN;
+    const int qscale = mpeg_qscale_rnd();
+    const int qmul = qscale << 1;
+    const int qadd = (rnd() & 1) ? (qscale - 1) | 1 : 0;
+
+    declare_func(void, int16_t *, size_t, size_t, int, int);
+
+    for (size_t i = 0; i < LEN; i++)
+        block1[i] = block0[i] = (rnd() & 1) ? rnd() : 0;
+
+    if (check_func(ctx->h263_dct_unquantize, "h263dsp.dct_unquantize_int%s",
+                   intra ? "ra" : "er")) {
+        av_assert0(end < LEN);
+        call_ref(block0, intra, end, qmul, qadd);
+        call_new(block1, intra, end, qmul, qadd);
+
+        if (memcmp(block0, block1, (end + 1) * sizeof (int16_t)))
+            fail();
+
+        bench_new(block1, intra, LEN, qmul, qadd);
+    }
+}
 
 typedef void (*filter)(uint8_t *src, int stride, int qscale);
 
@@ -56,6 +97,9 @@ void checkasm_check_h263dsp(void)
     H263DSPContext ctx;
 
     ff_h263dsp_init(&ctx);
+    check_dct_unquantize(&ctx, false);
+    check_dct_unquantize(&ctx, true);
+    report("dct_unquantize");
     check_loop_filter('h', ctx.h263_h_loop_filter);
     check_loop_filter('v', ctx.h263_v_loop_filter);
     report("loop_filter");
-- 
2.45.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 4/4] lavc/h263dsp: R-V V dct_unquantize

[Rémi Denis-Courmont]

T-Head C908:
h263dsp.dct_unquantize_inter_c:       5.7
h263dsp.dct_unquantize_inter_rvv_i32: 2.7
h263dsp.dct_unquantize_intra_c:       5.7
h263dsp.dct_unquantize_intra_rvv_i32: 3.0

SpacemiT X60:
h263dsp.dct_unquantize_inter_c:       5.0
h263dsp.dct_unquantize_inter_rvv_i32: 1.2
h263dsp.dct_unquantize_intra_c:       5.0
h263dsp.dct_unquantize_intra_rvv_i32: 1.5
---
 libavcodec/riscv/h263dsp_init.c | 12 +++++++++---
 libavcodec/riscv/h263dsp_rvv.S  | 22 ++++++++++++++++++++++
 2 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/libavcodec/riscv/h263dsp_init.c b/libavcodec/riscv/h263dsp_init.c
index 21b536366c..5d73fde865 100644
--- a/libavcodec/riscv/h263dsp_init.c
+++ b/libavcodec/riscv/h263dsp_init.c
@@ -25,6 +25,7 @@
 #include "libavutil/riscv/cpu.h"
 #include "libavcodec/h263dsp.h"
 
+void ff_h263_dct_unquantize_rvv(int16_t *, size_t start, size_t end, int, int);
 void ff_h263_h_loop_filter_rvv(uint8_t *src, int stride, int q);
 void ff_h263_v_loop_filter_rvv(uint8_t *src, int stride, int q);
 
@@ -33,9 +34,14 @@ av_cold void ff_h263dsp_init_riscv(H263DSPContext *c)
 #if HAVE_RVV
     int flags = av_get_cpu_flags();
 
-    if ((flags & AV_CPU_FLAG_RVV_I32) && ff_rv_vlen_least(128)) {
-        c->h263_h_loop_filter = ff_h263_h_loop_filter_rvv;
-        c->h263_v_loop_filter = ff_h263_v_loop_filter_rvv;
+    if (flags & AV_CPU_FLAG_RVV_I32) {
+        if (flags & AV_CPU_FLAG_RVB_ADDR)
+            c->h263_dct_unquantize = ff_h263_dct_unquantize_rvv;
+
+        if (ff_rv_vlen_least(128)) {
+            c->h263_h_loop_filter = ff_h263_h_loop_filter_rvv;
+            c->h263_v_loop_filter = ff_h263_v_loop_filter_rvv;
+        }
     }
 #endif
 }
diff --git a/libavcodec/riscv/h263dsp_rvv.S b/libavcodec/riscv/h263dsp_rvv.S
index 97503d527c..319c51a0bb 100644
--- a/libavcodec/riscv/h263dsp_rvv.S
+++ b/libavcodec/riscv/h263dsp_rvv.S
@@ -20,6 +20,28 @@
 
 #include "libavutil/riscv/asm.S"
 
+func ff_h263_dct_unquantize_rvv, zve32x
+        sub     a2, a2, a1
+        sh1add  a0, a1, a0
+        addi    a2, a2, 1
+1:
+        vsetvli  t0, a2, e16, m4, ta, mu
+        vle16.v  v8, (a0)
+        sub      a2, a2, t0
+        vmv.v.x  v24, a4
+        vmslt.vi v0, v8, 0
+        vneg.v   v24, v24, v0.t
+        vmsne.vi v0, v8, 0
+        vwmul.vx v16, v8, a3
+        vwadd.wv v16, v16, v24, v0.t
+        vncvt.x.x.w v8, v16
+        vse16.v v8, (a0)
+        sh1add  a0, t0, a0
+        bnez    a2, 1b
+
+        ret
+endfunc
+
         .option push
         .option norelax
 func ff_h263_h_loop_filter_rvv, zve32x
-- 
2.45.1

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCHv2 1/4] lavc/h263dsp: add DCT dequantisation function

[Rémi Denis-Courmont]

Le sunnuntaina 9. kesäkuuta 2024, 12.27.06 EEST Rémi Denis-Courmont a écrit :
> To preserve the alignment promise of the block base address, we pass
> the start offset separately (it is either 0 or 1).
> 
> Note that optimised implementations of this function will be taken into
> actual use if MpegEncContext.dct_unquantize_h263_{inter,intra} are *not*
> overloaded by existing optimisations.
> ---
>  libavcodec/h263dsp.c | 17 +++++++++++++++++
>  libavcodec/h263dsp.h |  2 ++
>  2 files changed, 19 insertions(+)
> 
> diff --git a/libavcodec/h263dsp.c b/libavcodec/h263dsp.c
> index 6a13353499..eb990f27bd 100644
> --- a/libavcodec/h263dsp.c
> +++ b/libavcodec/h263dsp.c
> @@ -23,6 +23,22 @@
>  #include "config.h"
>  #include "h263dsp.h"
> 
> +static void h263_dct_unquantize_c(int16_t *block, size_t start, size_t end,
> +                                  int qmul, int qadd)
> +{
> +    for (size_t i = start; i <= end; i++) {

C and R-V V don't care, but for the sake of more constrained SIMD extensions, 
it should be easier to keep intra and inter DCT functions separate.

-- 
Rémi Denis-Courmont
http://www.remlab.net/



_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/4] lavc/mpegvideo: use H263DSP dequant function

[Andreas Rheinhardt]

Rémi Denis-Courmont:
> ---
>  libavcodec/mpegvideo.c | 30 +++++-------------------------
>  1 file changed, 5 insertions(+), 25 deletions(-)
> 
> diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
> index 7af823b8bd..fa25d14970 100644
> --- a/libavcodec/mpegvideo.c
> +++ b/libavcodec/mpegvideo.c
> @@ -201,7 +201,7 @@ static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
>  static void dct_unquantize_h263_intra_c(MpegEncContext *s,
>                                    int16_t *block, int n, int qscale)
>  {
> -    int i, level, qmul, qadd;
> +    int qmul, qadd;
>      int nCoeffs;
>  
>      av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
> @@ -219,23 +219,13 @@ static void dct_unquantize_h263_intra_c(MpegEncContext *s,
>      else
>          nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
>  
> -    for(i=1; i<=nCoeffs; i++) {
> -        level = block[i];
> -        if (level) {
> -            if (level < 0) {
> -                level = level * qmul - qadd;
> -            } else {
> -                level = level * qmul + qadd;
> -            }
> -            block[i] = level;
> -        }
> -    }
> +    s->h263dsp.h263_dct_unquantize(block, 1, nCoeffs, qmul, qadd);
>  }
>  
>  static void dct_unquantize_h263_inter_c(MpegEncContext *s,
>                                    int16_t *block, int n, int qscale)
>  {
> -    int i, level, qmul, qadd;
> +    int qmul, qadd;
>      int nCoeffs;
>  
>      av_assert2(s->block_last_index[n]>=0);
> @@ -244,18 +234,7 @@ static void dct_unquantize_h263_inter_c(MpegEncContext *s,
>      qmul = qscale << 1;
>  
>      nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
> -
> -    for(i=0; i<=nCoeffs; i++) {
> -        level = block[i];
> -        if (level) {
> -            if (level < 0) {
> -                level = level * qmul - qadd;
> -            } else {
> -                level = level * qmul + qadd;
> -            }
> -            block[i] = level;
> -        }
> -    }
> +    s->h263dsp.h263_dct_unquantize(block, 0, nCoeffs, qmul, qadd);
>  }
>  
>  
> @@ -275,6 +254,7 @@ static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
>  static av_cold int dct_init(MpegEncContext *s)
>  {
>      ff_blockdsp_init(&s->bdsp);
> +    ff_h263dsp_init(&s->h263dsp);
>      ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
>      ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
>  

This approach will make H.261 use a h263dsp function which is a misnomer
and will lead to undefined references if no H.263 decoder or encoder is
enabled.

- Andreas

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/4] lavc/mpegvideo: use H263DSP dequant function

[Rémi Denis-Courmont]

Le sunnuntaina 9. kesäkuuta 2024, 19.13.54 EEST Andreas Rheinhardt a écrit :
> Rémi Denis-Courmont:
> > ---
> > 
> >  libavcodec/mpegvideo.c | 30 +++++-------------------------
> >  1 file changed, 5 insertions(+), 25 deletions(-)
> > 
> > diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
> > index 7af823b8bd..fa25d14970 100644
> > --- a/libavcodec/mpegvideo.c
> > +++ b/libavcodec/mpegvideo.c
> > @@ -201,7 +201,7 @@ static void
> > dct_unquantize_mpeg2_inter_c(MpegEncContext *s,> 
> >  static void dct_unquantize_h263_intra_c(MpegEncContext *s,
> >  
> >                                    int16_t *block, int n, int qscale)
> >  
> >  {
> > 
> > -    int i, level, qmul, qadd;
> > +    int qmul, qadd;
> > 
> >      int nCoeffs;
> >      
> >      av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
> > 
> > @@ -219,23 +219,13 @@ static void
> > dct_unquantize_h263_intra_c(MpegEncContext *s,> 
> >      else
> >      
> >          nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
> > 
> > -    for(i=1; i<=nCoeffs; i++) {
> > -        level = block[i];
> > -        if (level) {
> > -            if (level < 0) {
> > -                level = level * qmul - qadd;
> > -            } else {
> > -                level = level * qmul + qadd;
> > -            }
> > -            block[i] = level;
> > -        }
> > -    }
> > +    s->h263dsp.h263_dct_unquantize(block, 1, nCoeffs, qmul, qadd);
> > 
> >  }
> >  
> >  static void dct_unquantize_h263_inter_c(MpegEncContext *s,
> >  
> >                                    int16_t *block, int n, int qscale)
> >  
> >  {
> > 
> > -    int i, level, qmul, qadd;
> > +    int qmul, qadd;
> > 
> >      int nCoeffs;
> >      
> >      av_assert2(s->block_last_index[n]>=0);
> > 
> > @@ -244,18 +234,7 @@ static void
> > dct_unquantize_h263_inter_c(MpegEncContext *s,> 
> >      qmul = qscale << 1;
> >      
> >      nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
> > 
> > -
> > -    for(i=0; i<=nCoeffs; i++) {
> > -        level = block[i];
> > -        if (level) {
> > -            if (level < 0) {
> > -                level = level * qmul - qadd;
> > -            } else {
> > -                level = level * qmul + qadd;
> > -            }
> > -            block[i] = level;
> > -        }
> > -    }
> > +    s->h263dsp.h263_dct_unquantize(block, 0, nCoeffs, qmul, qadd);
> > 
> >  }
> > 
> > @@ -275,6 +254,7 @@ static void gray8(uint8_t *dst, const uint8_t *src,
> > ptrdiff_t linesize, int h)> 
> >  static av_cold int dct_init(MpegEncContext *s)
> >  {
> >  
> >      ff_blockdsp_init(&s->bdsp);
> > 
> > +    ff_h263dsp_init(&s->h263dsp);
> > 
> >      ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
> >      ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
> 
> This approach will make H.261 use a h263dsp function which is a misnomer

In reality, that is an existing problem: the *existing* pair of DCT functions 
for H.263 is provisioned regardless of the codec. Fixing the god object and 
spaghetti anti-patterns in MpegEncContext is way outside the scope of this MR.

> and will lead to undefined references if no H.263 decoder or encoder is
> enabled.

Fair enough, *that* is easy to fix.

-- 
Rémi Denis-Courmont
http://www.remlab.net/



_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".