Re: [libav-devel] [PATCH 8/8] hevcdsp: add x86 SIMD for MC
Quoting James Almer (2015-08-20 00:34:58) > On 19/08/15 4:43 PM, Anton Khirnov wrote: > > --- > > libavcodec/hevc.c | 6 +- > > libavcodec/hevc.h | 2 +- > > libavcodec/hevcdsp.c | 24 +- > > libavcodec/hevcdsp.h | 5 +- > > libavcodec/hevcdsp_template.c | 8 +- > > libavcodec/x86/Makefile | 3 +- > > libavcodec/x86/hevc_mc.asm| 816 > > ++ > > libavcodec/x86/hevcdsp_init.c | 405 + > > 8 files changed, 1258 insertions(+), 11 deletions(-) > > create mode 100644 libavcodec/x86/hevc_mc.asm > > I'm getting segmentation faults with quite a few of samples. > For example http://www.elecard.com/assets/files/other/clips/bbb_1080p_c.ts Cannot reproduce here. Can you give me more details (system, where exactly does it crash, etc.)? -- Anton Khirnov ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 8/8] hevcdsp: add x86 SIMD for MC
On 19/08/15 8:23 PM, Ronald S. Bultje wrote: > Hi, > > On Wed, Aug 19, 2015 at 6:34 PM, James Almer wrote: > >> On 19/08/15 4:43 PM, Anton Khirnov wrote: >>> --- >>> libavcodec/hevc.c | 6 +- >>> libavcodec/hevc.h | 2 +- >>> libavcodec/hevcdsp.c | 24 +- >>> libavcodec/hevcdsp.h | 5 +- >>> libavcodec/hevcdsp_template.c | 8 +- >>> libavcodec/x86/Makefile | 3 +- >>> libavcodec/x86/hevc_mc.asm| 816 >> ++ >>> libavcodec/x86/hevcdsp_init.c | 405 + >>> 8 files changed, 1258 insertions(+), 11 deletions(-) >>> create mode 100644 libavcodec/x86/hevc_mc.asm >> >> I'm getting segmentation faults with quite a few of samples. >> For example http://www.elecard.com/assets/files/other/clips/bbb_1080p_c.ts > > > So, at the risk of godwin, why was this reimplemented from scratch, rather > than basing it on what ffmpeg has? How could this possibly be an advantage > to our users? Or OpenHEVC for that matter, which is the source of almost every hevc asm optimization, x86 or otherwise, and a project that afaik branched off libav. > > Ronald ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 8/8] hevcdsp: add x86 SIMD for MC
Hi, On Wed, Aug 19, 2015 at 6:34 PM, James Almer wrote: > On 19/08/15 4:43 PM, Anton Khirnov wrote: > > --- > > libavcodec/hevc.c | 6 +- > > libavcodec/hevc.h | 2 +- > > libavcodec/hevcdsp.c | 24 +- > > libavcodec/hevcdsp.h | 5 +- > > libavcodec/hevcdsp_template.c | 8 +- > > libavcodec/x86/Makefile | 3 +- > > libavcodec/x86/hevc_mc.asm| 816 > ++ > > libavcodec/x86/hevcdsp_init.c | 405 + > > 8 files changed, 1258 insertions(+), 11 deletions(-) > > create mode 100644 libavcodec/x86/hevc_mc.asm > > I'm getting segmentation faults with quite a few of samples. > For example http://www.elecard.com/assets/files/other/clips/bbb_1080p_c.ts So, at the risk of godwin, why was this reimplemented from scratch, rather than basing it on what ffmpeg has? How could this possibly be an advantage to our users? Ronald ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 8/8] hevcdsp: add x86 SIMD for MC
On 19/08/15 4:43 PM, Anton Khirnov wrote: > --- > libavcodec/hevc.c | 6 +- > libavcodec/hevc.h | 2 +- > libavcodec/hevcdsp.c | 24 +- > libavcodec/hevcdsp.h | 5 +- > libavcodec/hevcdsp_template.c | 8 +- > libavcodec/x86/Makefile | 3 +- > libavcodec/x86/hevc_mc.asm| 816 > ++ > libavcodec/x86/hevcdsp_init.c | 405 + > 8 files changed, 1258 insertions(+), 11 deletions(-) > create mode 100644 libavcodec/x86/hevc_mc.asm I'm getting segmentation faults with quite a few of samples. For example http://www.elecard.com/assets/files/other/clips/bbb_1080p_c.ts ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 2/2] imgutils: Fix a typo in avcodec_get_pix_fmt_loss
Quoting Luca Barbato (2015-08-18 16:22:20) > If the candidate does not have alpha and the source does have alpha > report the loss of alpha. > > CC: libav-sta...@libav.org > --- > libavcodec/imgconvert.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libavcodec/imgconvert.c b/libavcodec/imgconvert.c > index 1f6d587..6b389ac 100644 > --- a/libavcodec/imgconvert.c > +++ b/libavcodec/imgconvert.c > @@ -76,7 +76,7 @@ int avcodec_get_pix_fmt_loss(enum AVPixelFormat dst_pix_fmt, > loss |= FF_LOSS_COLORSPACE; > > if (has_alpha && !(dst_desc->flags & AV_PIX_FMT_FLAG_ALPHA) && > - (dst_desc->flags & AV_PIX_FMT_FLAG_ALPHA)) > + (src_desc->flags & AV_PIX_FMT_FLAG_ALPHA)) > loss |= FF_LOSS_ALPHA; > > if (dst_pix_fmt == AV_PIX_FMT_PAL8 && !is_gray(src_desc)) > -- > 2.5.0 > Ok -- Anton Khirnov ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 1/2] imgutils: Use av_pix_fmt_get_chroma_sub_sample in avcodec_get_chroma_sub_sample
Quoting Luca Barbato (2015-08-18 16:22:19)
> Avoid a NULL dereference.
>
> CC: libav-sta...@libav.org
> ---
> libavcodec/imgconvert.c | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/libavcodec/imgconvert.c b/libavcodec/imgconvert.c
> index 2d32602..1f6d587 100644
> --- a/libavcodec/imgconvert.c
> +++ b/libavcodec/imgconvert.c
> @@ -41,9 +41,7 @@
>
> void avcodec_get_chroma_sub_sample(enum AVPixelFormat pix_fmt, int *h_shift,
> int *v_shift)
> {
> -const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
> -*h_shift = desc->log2_chroma_w;
> -*v_shift = desc->log2_chroma_h;
> +av_pix_fmt_get_chroma_sub_sample(pix_fmt, h_shift, v_shift);
> }
Since this function has no way to signal errors, I would argue that a
segfault would be better than leaving random possibly uninitialized data
around.
--
Anton Khirnov
___
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
[libav-devel] [PATCH 2/8] hevcdsp: fix a function name
put_weighted_pred_avg should be put_unweighted_pred_avg, there is no
weighting there.
---
libavcodec/hevc.c | 8
libavcodec/hevcdsp.c | 2 +-
libavcodec/hevcdsp.h | 6 +++---
libavcodec/hevcdsp_template.c | 8
4 files changed, 12 insertions(+), 12 deletions(-)
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 6395563..f17c313 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1806,8 +1806,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
dst0, s->frame->linesize[0],
tmp, tmp2, tmpstride, nPbW, nPbH);
} else {
-s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
- tmp, tmp2, tmpstride, nPbW, nPbH);
+s->hevcdsp.put_unweighted_pred_avg(dst0, s->frame->linesize[0],
+ tmp, tmp2, tmpstride, nPbW,
nPbH);
}
chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
@@ -1832,8 +1832,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
dst2, s->frame->linesize[2], tmp2,
tmp4,
tmpstride, nPbW / 2, nPbH / 2);
} else {
-s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp,
tmp3, tmpstride, nPbW/2, nPbH/2);
-s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2],
tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
+s->hevcdsp.put_unweighted_pred_avg(dst1, s->frame->linesize[1],
tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
+s->hevcdsp.put_unweighted_pred_avg(dst2, s->frame->linesize[2],
tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
}
}
}
diff --git a/libavcodec/hevcdsp.c b/libavcodec/hevcdsp.c
index 0abee9b..216101a 100644
--- a/libavcodec/hevcdsp.c
+++ b/libavcodec/hevcdsp.c
@@ -162,7 +162,7 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int
bit_depth)
hevcdsp->put_hevc_epel[1][1] = FUNC(put_hevc_epel_hv, depth); \
\
hevcdsp->put_unweighted_pred = FUNC(put_unweighted_pred, depth); \
-hevcdsp->put_weighted_pred_avg = FUNC(put_weighted_pred_avg, depth);\
+hevcdsp->put_unweighted_pred_avg = FUNC(put_unweighted_pred_avg, depth);
\
\
hevcdsp->weighted_pred = FUNC(weighted_pred, depth);\
hevcdsp->weighted_pred_avg = FUNC(weighted_pred_avg, depth);\
diff --git a/libavcodec/hevcdsp.h b/libavcodec/hevcdsp.h
index aad96db..7278464 100644
--- a/libavcodec/hevcdsp.h
+++ b/libavcodec/hevcdsp.h
@@ -67,9 +67,9 @@ typedef struct HEVCDSPContext {
void (*put_unweighted_pred)(uint8_t *dst, ptrdiff_t dststride, int16_t
*src,
ptrdiff_t srcstride, int width, int height);
-void (*put_weighted_pred_avg)(uint8_t *dst, ptrdiff_t dststride,
- int16_t *src1, int16_t *src2,
- ptrdiff_t srcstride, int width, int height);
+void (*put_unweighted_pred_avg)(uint8_t *dst, ptrdiff_t dststride,
+int16_t *src1, int16_t *src2,
+ptrdiff_t srcstride, int width, int
height);
void (*weighted_pred)(uint8_t denom, int16_t wlxFlag, int16_t olxFlag,
uint8_t *dst, ptrdiff_t dststride, int16_t *src,
ptrdiff_t srcstride, int width, int height);
diff --git a/libavcodec/hevcdsp_template.c b/libavcodec/hevcdsp_template.c
index ae7e021..390f683 100644
--- a/libavcodec/hevcdsp_template.c
+++ b/libavcodec/hevcdsp_template.c
@@ -1033,10 +1033,10 @@ static void FUNC(put_unweighted_pred)(uint8_t *_dst,
ptrdiff_t _dststride,
}
}
-static void FUNC(put_weighted_pred_avg)(uint8_t *_dst, ptrdiff_t _dststride,
-int16_t *src1, int16_t *src2,
-ptrdiff_t srcstride,
-int width, int height)
+static void FUNC(put_unweighted_pred_avg)(uint8_t *_dst, ptrdiff_t _dststride,
+ int16_t *src1, int16_t *src2,
+ ptrdiff_t srcstride,
+ int width, int height)
{
int x, y;
pixel *dst = (pixel *)_dst;
--
2.0.0
___
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
[libav-devel] [PATCH 5/8] hevcdsp: split the pred functions by width
This should allow for more efficient SIMD.
---
libavcodec/hevc.c | 118 +-
libavcodec/hevcdsp.c | 33 ++--
libavcodec/hevcdsp.h | 36 -
libavcodec/hevcdsp_template.c | 81 ++---
4 files changed, 174 insertions(+), 94 deletions(-)
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 3dc510d..5da8249 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1725,32 +1725,32 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
-s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
-
s->sh.luma_weight_l0[current_mv.ref_idx[0]],
-
s->sh.luma_offset_l0[current_mv.ref_idx[0]],
- dst0, s->frame->linesize[0], tmp,
- tmpstride, nPbW, nPbH);
+s->hevcdsp.weighted_pred[pred_idx](s->sh.luma_log2_weight_denom,
+
s->sh.luma_weight_l0[current_mv.ref_idx[0]],
+
s->sh.luma_offset_l0[current_mv.ref_idx[0]],
+ dst0, s->frame->linesize[0],
tmp,
+ tmpstride, nPbH);
} else {
-s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp,
tmpstride, nPbW, nPbH);
+s->hevcdsp.put_unweighted_pred[pred_idx](dst0,
s->frame->linesize[0], tmp, tmpstride, nPbH);
}
chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2,
pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
-s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
-
s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
-
s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
- dst1, s->frame->linesize[1], tmp,
tmpstride,
- nPbW / 2, nPbH / 2);
-s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
-
s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
-
s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
- dst2, s->frame->linesize[2], tmp2,
tmpstride,
- nPbW / 2, nPbH / 2);
+
s->hevcdsp.weighted_pred_chroma[pred_idx](s->sh.chroma_log2_weight_denom,
+
s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
+
s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
+ dst1,
s->frame->linesize[1], tmp, tmpstride,
+ nPbH / 2);
+
s->hevcdsp.weighted_pred_chroma[pred_idx](s->sh.chroma_log2_weight_denom,
+
s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
+
s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
+ dst2,
s->frame->linesize[2], tmp2, tmpstride,
+ nPbH / 2);
} else {
-s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp,
tmpstride, nPbW/2, nPbH/2);
-s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2,
tmpstride, nPbW/2, nPbH/2);
+s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst1,
s->frame->linesize[1], tmp, tmpstride, nPbH/2);
+s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst2,
s->frame->linesize[2], tmp2, tmpstride, nPbH/2);
}
} else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
@@ -1761,13 +1761,13 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
-s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
-
s->sh.luma_weight_l1[current_mv.ref_idx[1]],
-
s->sh.luma_offset_l1[current_mv.ref_idx[1]],
- dst0, s->frame->linesize[0], tmp,
tmpstride,
-
[libav-devel] [PATCH 7/8] checkasm: add HEVC MC tests
---
tests/checkasm/Makefile | 1 +
tests/checkasm/checkasm.c | 3 +
tests/checkasm/checkasm.h | 1 +
tests/checkasm/hevc_mc.c | 294 ++
4 files changed, 299 insertions(+)
create mode 100644 tests/checkasm/hevc_mc.c
diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index 9498ebf..4024137 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -2,6 +2,7 @@
AVCODECOBJS-$(CONFIG_BSWAPDSP) += bswapdsp.o
AVCODECOBJS-$(CONFIG_H264PRED) += h264pred.o
AVCODECOBJS-$(CONFIG_H264QPEL) += h264qpel.o
+AVCODECOBJS-$(CONFIG_HEVC_DECODER) += hevc_mc.o
CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index b564e7e..1984bca 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -66,6 +66,9 @@ static const struct {
#if CONFIG_H264QPEL
{ "h264qpel", checkasm_check_h264qpel },
#endif
+#if CONFIG_HEVC_DECODER
+{ "hevc_mc", checkasm_check_hevc_mc },
+#endif
{ NULL }
};
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index 443546a..8ef250e 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -32,6 +32,7 @@
void checkasm_check_bswapdsp(void);
void checkasm_check_h264pred(void);
void checkasm_check_h264qpel(void);
+void checkasm_check_hevc_mc(void);
intptr_t (*checkasm_check_func(intptr_t (*func)(), const char *name, ...))()
av_printf_format(2, 3);
int checkasm_bench_func(void);
diff --git a/tests/checkasm/hevc_mc.c b/tests/checkasm/hevc_mc.c
new file mode 100644
index 000..b1b1dd6
--- /dev/null
+++ b/tests/checkasm/hevc_mc.c
@@ -0,0 +1,294 @@
+/*
+ * Copyright (c) 2015 Anton Khirnov
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * Libav 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with Libav; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include
+
+#include "checkasm.h"
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/hevcdsp.h"
+
+#include "libavutil/common.h"
+#include "libavutil/intreadwrite.h"
+
+// max PU size + interpolation stencil
+#define BUF_SIZE (FFALIGN(64 + 7, 16) * (64 + 7) * 2)
+
+#define PIXEL_SIZE(depth) ((depth + 7) / 8)
+
+#define randomize_buffers(buf, size, depth) \
+do {\
+uint32_t mask = pixel_mask[depth - 8]; \
+int i; \
+for (i = 0; i < size; i += 4) { \
+uint32_t r = rnd() & mask; \
+AV_WN32A(buf + i, r); \
+} \
+} while (0)
+
+static const uint32_t pixel_mask[3] = { 0x, 0x01ff01ff, 0x03ff03ff };
+
+static const int pred_heights[][7] = {
+[2] = { 8, 4, 2, 0 },
+[4] = { 16, 8, 4, 2, 0 },
+[6] = { 8, 0 },
+[8] = { 32, 16, 8, 4, 2, 0 },
+[12] = { 16, 0 },
+[16] = { 64, 32, 16, 12, 8, 4, 0 },
+[24] = { 32, 0 },
+[32] = { 64, 32, 24, 16, 8, 0 },
+[48] = { 64, 0 },
+[64] = { 64, 48, 32, 16, 0 },
+};
+
+static const int pred_widths[] = { 4, 8, 12, 16, 24, 32, 48, 64 };
+
+static const char *interp_names[2][2] = { { "pixels", "h" }, { "v", "hv" } };
+
+static void unweighted_pred(uint8_t *dst0, uint8_t *dst1, int16_t *src0,
int16_t *src1,
+int width, int bit_depth)
+{
+const int srcstride = FFALIGN(width, 16) * sizeof(*src0);
+const int dststride = FFALIGN(width, 16) * PIXEL_SIZE(bit_depth);
+int i;
+
+randomize_buffers(src0, BUF_SIZE * sizeof(*src0), 8);
+if (src1)
+randomize_buffers(src1, BUF_SIZE * sizeof(*src1), 8);
+
+memset(dst0, 0, BUF_SIZE * sizeof(*dst0));
+memset(dst1, 0, BUF_SIZE * sizeof(*dst1));
+
+for (i = 0; i < FF_ARRAY_ELEMS(pred_heights[i]); i++) {
+int height = pred_heights[width][i];
+
+if (!height)
+break;
+
+if (!src1) {
+call_ref(dst0, dststride, src0, srcstride, height);
+call_new(dst1, dststride, src0, srcstride, height);
+} else {
+call_ref(dst0, dststride, src0, src1, srcstride, height);
+call_new(dst1, dststride, src0, src1, srcstride, height);
+}
+if (memcmp(dst0, dst1, dststride * height))
+fail();
+
+if (!src1)
+bench_new(dst1, dststride, src0, srcstride, h
[libav-devel] [PATCH 1/8] hevc: avoid invalid shifts of negative values
--- libavcodec/hevc.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c index 0dfe7a2..6395563 100644 --- a/libavcodec/hevc.c +++ b/libavcodec/hevc.c @@ -1494,7 +1494,7 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride, x_off += mv->x >> 2; y_off += mv->y >> 2; -src += y_off * srcstride + (x_off << s->ps.sps->pixel_shift); +src += y_off * srcstride + (x_off * (1 << s->ps.sps->pixel_shift)); if (x_off < extra_left || y_off < extra_top || x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] || @@ -1548,8 +1548,8 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2, x_off += mv->x >> 3; y_off += mv->y >> 3; -src1 += y_off * src1stride + (x_off << s->ps.sps->pixel_shift); -src2 += y_off * src2stride + (x_off << s->ps.sps->pixel_shift); +src1 += y_off * src1stride + (x_off * (1 << s->ps.sps->pixel_shift)); +src2 += y_off * src2stride + (x_off * (1 << s->ps.sps->pixel_shift)); if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER || x_off >= pic_width - block_w - EPEL_EXTRA_AFTER || -- 2.0.0 ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
[libav-devel] [PATCH 4/8] hevcdsp: split the epel functions by width
This should allow for more efficient SIMD.
---
libavcodec/hevc.c | 29 --
libavcodec/hevcdsp.c | 18 ---
libavcodec/hevcdsp.h | 6 ++--
libavcodec/hevcdsp_template.c | 71 +--
4 files changed, 85 insertions(+), 39 deletions(-)
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 7070106..3dc510d 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1533,7 +1533,7 @@ static void luma_mc(HEVCContext *s, int16_t *dst,
ptrdiff_t dststride,
*/
static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
- int x_off, int y_off, int block_w, int block_h)
+ int x_off, int y_off, int block_w, int block_h, int
pred_idx)
{
HEVCLocalContext *lc = &s->HEVClc;
uint8_t *src1= ref->data[1];
@@ -1571,8 +1571,8 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1,
int16_t *dst2,
src1 = lc->edge_emu_buffer + buf_offset1;
src1stride = edge_emu_stride;
-s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
- block_w, block_h, mx, my,
lc->mc_buffer);
+s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst1, dststride, src1,
src1stride,
+ block_h, mx, my,
lc->mc_buffer);
s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
edge_emu_stride, src2stride,
@@ -1583,16 +1583,13 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1,
int16_t *dst2,
src2 = lc->edge_emu_buffer + buf_offset2;
src2stride = edge_emu_stride;
-s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
- block_w, block_h, mx, my,
- lc->mc_buffer);
+s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst2, dststride, src2,
src2stride,
+ block_h, mx, my,
lc->mc_buffer);
} else {
-s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
- block_w, block_h, mx, my,
- lc->mc_buffer);
-s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
- block_w, block_h, mx, my,
- lc->mc_buffer);
+s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst1, dststride, src1,
src1stride,
+ block_h, mx, my,
lc->mc_buffer);
+s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst2, dststride, src2,
src2stride,
+ block_h, mx, my,
lc->mc_buffer);
}
}
@@ -1737,7 +1734,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp,
tmpstride, nPbW, nPbH);
}
chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
- ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
+ ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2,
pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
@@ -1774,7 +1771,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
}
chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
- ¤t_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
+ ¤t_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2, pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
@@ -1816,9 +1813,9 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
}
chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
- ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
+ ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2,
pred_idx);
chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
- ¤t_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
+ ¤t_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2,
pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
diff --git a/libavcodec/hevcdsp.c b/libavcodec/hevcdsp.c
index 4e311a6..7cb273b 100644
--- a/libavcodec/hevcdsp.c
+++ b/libavcodec/hevcdsp.c
@@ -122,6 +122,12 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int
bit_depth)
hevcdsp->put_hevc_qpel[1][0][i] = FUNC(put_hevc_qpel_v_ ## wi
[libav-devel] [PATCH 8/8] hevcdsp: add x86 SIMD for MC
---
libavcodec/hevc.c | 6 +-
libavcodec/hevc.h | 2 +-
libavcodec/hevcdsp.c | 24 +-
libavcodec/hevcdsp.h | 5 +-
libavcodec/hevcdsp_template.c | 8 +-
libavcodec/x86/Makefile | 3 +-
libavcodec/x86/hevc_mc.asm| 816 ++
libavcodec/x86/hevcdsp_init.c | 405 +
8 files changed, 1258 insertions(+), 11 deletions(-)
create mode 100644 libavcodec/x86/hevc_mc.asm
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index dd54525..9cae92c 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -38,9 +38,9 @@
#include "golomb.h"
#include "hevc.h"
-const uint8_t ff_hevc_qpel_extra_before[4] = { 0, 3, 3, 2 };
-const uint8_t ff_hevc_qpel_extra_after[4] = { 0, 3, 4, 4 };
-const uint8_t ff_hevc_qpel_extra[4]= { 0, 6, 7, 6 };
+const uint8_t ff_hevc_qpel_extra_before[4] = { 0, 3, 3, 3 };
+const uint8_t ff_hevc_qpel_extra_after[4] = { 0, 4, 4, 4 };
+const uint8_t ff_hevc_qpel_extra[4]= { 0, 7, 7, 7 };
static const uint8_t scan_1x1[1] = { 0 };
diff --git a/libavcodec/hevc.h b/libavcodec/hevc.h
index c6e05bc..7c87a55 100644
--- a/libavcodec/hevc.h
+++ b/libavcodec/hevc.h
@@ -740,7 +740,7 @@ typedef struct HEVCPredContext {
} HEVCPredContext;
typedef struct HEVCLocalContext {
-DECLARE_ALIGNED(16, int16_t, mc_buffer[(MAX_PB_SIZE + 7) * MAX_PB_SIZE]);
+DECLARE_ALIGNED(16, int16_t, mc_buffer[(MAX_PB_SIZE + 24) * MAX_PB_SIZE]);
uint8_t cabac_state[HEVC_CONTEXTS];
uint8_t first_qp_group;
diff --git a/libavcodec/hevcdsp.c b/libavcodec/hevcdsp.c
index ab9ba3b..2b29b19 100644
--- a/libavcodec/hevcdsp.c
+++ b/libavcodec/hevcdsp.c
@@ -89,7 +89,7 @@ static const int8_t transform[32][32] = {
90, -90, 88, -85, 82, -78, 73, -67, 61, -54, 46, -38, 31, -22,
13, -4 },
};
-DECLARE_ALIGNED(16, const int8_t, ff_hevc_epel_filters[7][16]) = {
+DECLARE_ALIGNED(16, const int16_t, ff_hevc_epel_coeffs[7][16]) = {
{ -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2 },
{ -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2 },
{ -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4 },
@@ -99,6 +99,28 @@ DECLARE_ALIGNED(16, const int8_t,
ff_hevc_epel_filters[7][16]) = {
{ -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2 },
};
+DECLARE_ALIGNED(16, const int8_t, ff_hevc_epel_coeffs8[7][16]) = {
+{ -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2, -2, 58, 10, -2 },
+{ -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2, -4, 54, 16, -2 },
+{ -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4, -6, 46, 28, -4 },
+{ -4, 36, 36, -4, -4, 36, 36, -4, -4, 36, 36, -4, -4, 36, 36, -4 },
+{ -4, 28, 46, -6, -4, 28, 46, -6, -4, 28, 46, -6, -4, 28, 46, -6 },
+{ -2, 16, 54, -4, -2, 16, 54, -4, -2, 16, 54, -4, -2, 16, 54, -4 },
+{ -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2, -2, 10, 58, -2 },
+};
+
+DECLARE_ALIGNED(16, const int16_t, ff_hevc_qpel_coeffs[3][8]) = {
+{ -1, 4, -10, 58, 17, -5, 1, 0 },
+{ -1, 4, -11, 40, 40, -11, 4, -1 },
+{ 0, 1, -5, 17, 58, -10, 4, -1 },
+};
+
+DECLARE_ALIGNED(16, const int8_t, ff_hevc_qpel_coeffs8[3][16]) = {
+{ -1, 4, -10, 58, 17, -5, 1, 0, -1, 4, -10, 58, 17, -5, 1, 0 },
+{ -1, 4, -11, 40, 40, -11, 4, -1, -1, 4, -11, 40, 40, -11, 4, -1 },
+{ 0, 1, -5, 17, 58, -10, 4, -1, 0, 1, -5, 17, 58, -10, 4, -1 },
+};
+
#define BIT_DEPTH 8
#include "hevcdsp_template.c"
#undef BIT_DEPTH
diff --git a/libavcodec/hevcdsp.h b/libavcodec/hevcdsp.h
index ee3aa70..4daa2e5 100644
--- a/libavcodec/hevcdsp.h
+++ b/libavcodec/hevcdsp.h
@@ -118,6 +118,9 @@ void ff_hevc_dsp_init(HEVCDSPContext *hpc, int bit_depth);
void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth);
-extern const int8_t ff_hevc_epel_filters[7][16];
+extern const int16_t ff_hevc_epel_coeffs[7][16];
+extern const int8_t ff_hevc_epel_coeffs8[7][16];
+extern const int16_t ff_hevc_qpel_coeffs[3][8];
+extern const int8_t ff_hevc_qpel_coeffs8[3][16];
#endif /* AVCODEC_HEVCDSP_H */
diff --git a/libavcodec/hevcdsp_template.c b/libavcodec/hevcdsp_template.c
index 8bb0a57..1f9adee 100644
--- a/libavcodec/hevcdsp_template.c
+++ b/libavcodec/hevcdsp_template.c
@@ -999,7 +999,7 @@ static inline void FUNC(put_hevc_epel_h)(int16_t *dst,
ptrdiff_t dststride,
int x, y;
pixel *src = (pixel *)_src;
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
-const int8_t *filter = ff_hevc_epel_filters[mx - 1];
+const int16_t *filter = ff_hevc_epel_coeffs[mx - 1];
int8_t filter_0 = filter[0];
int8_t filter_1 = filter[1];
int8_t filter_2 = filter[2];
@@ -1021,7 +1021,7 @@ static inline void FUNC(put_hevc_epel_v)(int16_t *dst,
ptrdiff_t dststride,
int x, y;
pixel *src = (pixel *)_src;
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
-const int8_t *filter = ff_hevc_epel_filters[my - 1];
+const int16_t *filter = ff_hevc_epel_co
[libav-devel] [PATCH 6/8] hevc: change the stride of the MC buffer to be in bytes instead of elements
Currently, the frame stride is passed in bytes, while the MC buffer size
is in int16_t elements, This can be confusing, so pass both strides in
bytes.
---
libavcodec/hevc.c | 2 +-
libavcodec/hevcdsp_template.c | 12
2 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index 5da8249..dd54525 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1666,7 +1666,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
RefPicList *refPicList = s->ref->refPicList;
HEVCFrame *ref0, *ref1;
-int tmpstride = MAX_PB_SIZE;
+int tmpstride = MAX_PB_SIZE * sizeof(int16_t);
uint8_t *dst0 = POS(0, x0, y0);
uint8_t *dst1 = POS(1, x0, y0);
diff --git a/libavcodec/hevcdsp_template.c b/libavcodec/hevcdsp_template.c
index df570f3..8bb0a57 100644
--- a/libavcodec/hevcdsp_template.c
+++ b/libavcodec/hevcdsp_template.c
@@ -785,6 +785,7 @@ FUNC(put_hevc_qpel_pixels)(int16_t *dst, ptrdiff_t
dststride,
pixel *src = (pixel *)_src;
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
+dststride /= sizeof(*dst);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
dst[x] = src[x] << (14 - BIT_DEPTH);
@@ -832,6 +833,7 @@ static void FUNC(put_hevc_qpel_h ## H)(int16_t *dst,
ptrdiff_t dststride, \
pixel *src = (pixel*)_src;
\
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
\
\
+dststride /= sizeof(*dst);
\
for (y = 0; y < height; y++) {
\
for (x = 0; x < width; x++)
\
dst[x] = QPEL_FILTER_ ## H(src, 1) >> (BIT_DEPTH - 8);
\
@@ -850,6 +852,7 @@ static void FUNC(put_hevc_qpel_v ## V)(int16_t *dst,
ptrdiff_t dststride, \
pixel *src = (pixel*)_src;
\
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
\
\
+dststride /= sizeof(*dst);
\
for (y = 0; y < height; y++) {
\
for (x = 0; x < width; x++)
\
dst[x] = QPEL_FILTER_ ## V(src, srcstride) >> (BIT_DEPTH - 8);
\
@@ -873,6 +876,7 @@ static void FUNC(put_hevc_qpel_h ## H ## v ## V)(int16_t
*dst, \
int16_t tmp_array[(MAX_PB_SIZE + 7) * MAX_PB_SIZE];
\
int16_t *tmp = tmp_array;
\
\
+dststride /= sizeof(*dst);
\
src -= ff_hevc_qpel_extra_before[V] * srcstride;
\
\
for (y = 0; y < height + ff_hevc_qpel_extra[V]; y++) {
\
@@ -972,6 +976,7 @@ static inline void FUNC(put_hevc_epel_pixels)(int16_t *dst,
ptrdiff_t dststride,
pixel *src = (pixel *)_src;
ptrdiff_t srcstride = _srcstride / sizeof(pixel);
+dststride /= sizeof(*dst);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
dst[x] = src[x] << (14 - BIT_DEPTH);
@@ -999,6 +1004,7 @@ static inline void FUNC(put_hevc_epel_h)(int16_t *dst,
ptrdiff_t dststride,
int8_t filter_1 = filter[1];
int8_t filter_2 = filter[2];
int8_t filter_3 = filter[3];
+dststride /= sizeof(*dst);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
dst[x] = EPEL_FILTER(src, 1) >> (BIT_DEPTH - 8);
@@ -1021,6 +1027,7 @@ static inline void FUNC(put_hevc_epel_v)(int16_t *dst,
ptrdiff_t dststride,
int8_t filter_2 = filter[2];
int8_t filter_3 = filter[3];
+dststride /= sizeof(*dst);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
dst[x] = EPEL_FILTER(src, srcstride) >> (BIT_DEPTH - 8);
@@ -1046,6 +1053,7 @@ static inline void FUNC(put_hevc_epel_hv)(int16_t *dst,
ptrdiff_t dststride,
int16_t tmp_array[(MAX_PB_SIZE + 3) * MAX_PB_SIZE];
int16_t *tmp = tmp_array;
+dststride /= sizeof(*dst);
src -= EPEL_EXTRA_BEFORE * srcstride;
for (y = 0; y < height + EPEL_EXTRA; y++) {
@@ -1122,6 +1130,7 @@ FUNC(put_unweighted_pred)(uint8_t *_dst, ptrdiff_t
_dststride,
#else
int offset = 0;
#endif
+srcstride /= sizeof(*src);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
dst[x] = av_clip_pixel((src[x] + offset) >> shift);
@@ -114
[libav-devel] [PATCH 3/8] hevcdsp: split the qpel functions by width instead of by the subpixel fraction
This should allow for more efficient SIMD.
Keep the C versions as they are now, to allow the compiler to inline the
interpolation coefficients.
---
libavcodec/hevc.c | 19 -
libavcodec/hevcdsp.c | 30 ++---
libavcodec/hevcdsp.h | 6 ++---
libavcodec/hevcdsp_template.c | 63 ---
4 files changed, 89 insertions(+), 29 deletions(-)
diff --git a/libavcodec/hevc.c b/libavcodec/hevc.c
index f17c313..7070106 100644
--- a/libavcodec/hevc.c
+++ b/libavcodec/hevc.c
@@ -1479,7 +1479,7 @@ static void hls_mvd_coding(HEVCContext *s, int x0, int
y0, int log2_cb_size)
*/
static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
AVFrame *ref, const Mv *mv, int x_off, int y_off,
-int block_w, int block_h)
+int block_w, int block_h, int pred_idx)
{
HEVCLocalContext *lc = &s->HEVClc;
uint8_t *src = ref->data[0];
@@ -1513,8 +1513,8 @@ static void luma_mc(HEVCContext *s, int16_t *dst,
ptrdiff_t dststride,
src = lc->edge_emu_buffer + buf_offset;
srcstride = edge_emu_stride;
}
-s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
- block_h, lc->mc_buffer);
+s->hevcdsp.put_hevc_qpel[!!my][!!mx][pred_idx](dst, dststride, src,
srcstride,
+ block_h, mx, my,
lc->mc_buffer);
}
/**
@@ -1651,6 +1651,11 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
int nPbW, int nPbH,
int log2_cb_size, int partIdx)
{
+static const int pred_indices[] = {
+[4] = 0, [8] = 1, [12] = 2, [16] = 3, [24] = 4, [32] = 5, [48] = 6,
[64] = 7,
+};
+const int pred_idx = pred_indices[nPbW];
+
#define POS(c_idx, x, y)
\
&s->frame->data[c_idx][((y) >> s->ps.sps->vshift[c_idx]) *
s->frame->linesize[c_idx] + \
(((x) >> s->ps.sps->hshift[c_idx]) <<
s->ps.sps->pixel_shift)]
@@ -1719,7 +1724,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
luma_mc(s, tmp, tmpstride, ref0->frame,
-¤t_mv.mv[0], x0, y0, nPbW, nPbH);
+¤t_mv.mv[0], x0, y0, nPbW, nPbH, pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
@@ -1755,7 +1760,7 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
luma_mc(s, tmp, tmpstride, ref1->frame,
-¤t_mv.mv[1], x0, y0, nPbW, nPbH);
+¤t_mv.mv[1], x0, y0, nPbW, nPbH, pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
@@ -1792,9 +1797,9 @@ static void hls_prediction_unit(HEVCContext *s, int x0,
int y0,
DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
luma_mc(s, tmp, tmpstride, ref0->frame,
-¤t_mv.mv[0], x0, y0, nPbW, nPbH);
+¤t_mv.mv[0], x0, y0, nPbW, nPbH, pred_idx);
luma_mc(s, tmp2, tmpstride, ref1->frame,
-¤t_mv.mv[1], x0, y0, nPbW, nPbH);
+¤t_mv.mv[1], x0, y0, nPbW, nPbH, pred_idx);
if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
(s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
diff --git a/libavcodec/hevcdsp.c b/libavcodec/hevcdsp.c
index 216101a..4e311a6 100644
--- a/libavcodec/hevcdsp.c
+++ b/libavcodec/hevcdsp.c
@@ -116,6 +116,12 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int
bit_depth)
#undef FUNC
#define FUNC(a, depth) a ## _ ## depth
+#define QPEL_FUNC(i, width, depth)\
+hevcdsp->put_hevc_qpel[0][0][i] = FUNC(put_hevc_qpel_pixels_ ## width,
depth); \
+hevcdsp->put_hevc_qpel[0][1][i] = FUNC(put_hevc_qpel_h_ ## width,
depth); \
+hevcdsp->put_hevc_qpel[1][0][i] = FUNC(put_hevc_qpel_v_ ## width,
depth); \
+hevcdsp->put_hevc_qpel[1][1][i] = FUNC(put_hevc_qpel_hv_ ## width,
depth); \
+
#define HEVC_DSP(depth) \
hevcdsp->put_pcm= FUNC(put_pcm, depth); \
hevcdsp->transquant_bypass[0] = FUNC(transquant_bypass4x4, depth);\
@@ -139,22 +145,14 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int
bit_depth)
hevcdsp->sao_edge_filter[2] = FUNC(sao_edge_filter_2, depth); \
hevcdsp->sao_edge_filter[3] = FUNC(sao_edge_filter_3, depth); \
Re: [libav-devel] [PATCH 1/2] imgutils: Use av_pix_fmt_get_chroma_sub_sample in avcodec_get_chroma_sub_sample
two newbie questions:
1) Is this function still needed? Could we just call
ac_pix_fmt_get_chroma_sub_sample directly?
2) What about the return code? Isn't it lost? What happens in case of error?
Federico
On Tue, Aug 18, 2015 at 4:22 PM, Luca Barbato wrote:
> Avoid a NULL dereference.
>
> CC: libav-sta...@libav.org
> ---
> libavcodec/imgconvert.c | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/libavcodec/imgconvert.c b/libavcodec/imgconvert.c
> index 2d32602..1f6d587 100644
> --- a/libavcodec/imgconvert.c
> +++ b/libavcodec/imgconvert.c
> @@ -41,9 +41,7 @@
>
> void avcodec_get_chroma_sub_sample(enum AVPixelFormat pix_fmt, int *h_shift,
> int *v_shift)
> {
> -const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
> -*h_shift = desc->log2_chroma_w;
> -*v_shift = desc->log2_chroma_h;
> +av_pix_fmt_get_chroma_sub_sample(pix_fmt, h_shift, v_shift);
> }
>
> static int is_gray(const AVPixFmtDescriptor *desc)
> --
> 2.5.0
>
> ___
> libav-devel mailing list
> libav-devel@libav.org
> https://lists.libav.org/mailman/listinfo/libav-devel
--
Website at http://tomassetti.me
GitHub https://github.com/ftomassetti
___
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
Re: [libav-devel] [PATCH 2/2] imgutils: Fix a typo in avcodec_get_pix_fmt_loss
It seems pretty straighforward to me, Federico On Tue, Aug 18, 2015 at 4:22 PM, Luca Barbato wrote: > If the candidate does not have alpha and the source does have alpha > report the loss of alpha. > > CC: libav-sta...@libav.org > --- > libavcodec/imgconvert.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libavcodec/imgconvert.c b/libavcodec/imgconvert.c > index 1f6d587..6b389ac 100644 > --- a/libavcodec/imgconvert.c > +++ b/libavcodec/imgconvert.c > @@ -76,7 +76,7 @@ int avcodec_get_pix_fmt_loss(enum AVPixelFormat dst_pix_fmt, > loss |= FF_LOSS_COLORSPACE; > > if (has_alpha && !(dst_desc->flags & AV_PIX_FMT_FLAG_ALPHA) && > - (dst_desc->flags & AV_PIX_FMT_FLAG_ALPHA)) > + (src_desc->flags & AV_PIX_FMT_FLAG_ALPHA)) > loss |= FF_LOSS_ALPHA; > > if (dst_pix_fmt == AV_PIX_FMT_PAL8 && !is_gray(src_desc)) > -- > 2.5.0 > > ___ > libav-devel mailing list > libav-devel@libav.org > https://lists.libav.org/mailman/listinfo/libav-devel -- Website at http://tomassetti.me GitHub https://github.com/ftomassetti ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
