On Thu, Oct 29, 2020 at 09:09:00PM +0800, Linjie Fu wrote: > On Thu, Oct 29, 2020 at 7:16 PM <lance.lmw...@gmail.com> wrote: > > > From: Limin Wang <lance.lmw...@gmail.com> > > > > Signed-off-by: Limin Wang <lance.lmw...@gmail.com> > > --- > > libavfilter/unsharp.h | 3 + > > libavfilter/vf_unsharp.c | 162 > > +++++++++++++++++++++++++---------------------- > > 2 files changed, 90 insertions(+), 75 deletions(-) > > > > diff --git a/libavfilter/unsharp.h b/libavfilter/unsharp.h > > index a60b30f..253e32d 100644 > > --- a/libavfilter/unsharp.h > > +++ b/libavfilter/unsharp.h > > @@ -48,9 +48,12 @@ typedef struct UnsharpContext { > > UnsharpFilterParam luma; ///< luma parameters (width, height, > > amount) > > UnsharpFilterParam chroma; ///< chroma parameters (width, height, > > amount) > > int hsub, vsub; > > + int bitdepth; > > + int bps; > > int nb_threads; > > int opencl; > > int (* apply_unsharp)(AVFilterContext *ctx, AVFrame *in, AVFrame > > *out); > > + int (* unsharp_slice)(AVFilterContext *ctx, void *arg, int jobnr, int > > nb_jobs); > > > > Just curious: > Any special reason for the function moving?
Sorry, I'm not clear about your question. > > > > } UnsharpContext; > > > > #endif /* AVFILTER_UNSHARP_H */ > > diff --git a/libavfilter/vf_unsharp.c b/libavfilter/vf_unsharp.c > > index 7b430b6..416bf1c 100644 > > --- a/libavfilter/vf_unsharp.c > > +++ b/libavfilter/vf_unsharp.c > > @@ -57,81 +57,90 @@ typedef struct TheadData { > > int height; > > } ThreadData; > > > > -static int unsharp_slice(AVFilterContext *ctx, void *arg, int jobnr, int > > nb_jobs) > > -{ > > - ThreadData *td = arg; > > - UnsharpFilterParam *fp = td->fp; > > - uint32_t **sc = fp->sc; > > - uint32_t *sr = fp->sr; > > - const uint8_t *src2 = NULL; //silence a warning > > - const int amount = fp->amount; > > - const int steps_x = fp->steps_x; > > - const int steps_y = fp->steps_y; > > - const int scalebits = fp->scalebits; > > - const int32_t halfscale = fp->halfscale; > > - > > - uint8_t *dst = td->dst; > > - const uint8_t *src = td->src; > > - const int dst_stride = td->dst_stride; > > - const int src_stride = td->src_stride; > > - const int width = td->width; > > - const int height = td->height; > > - const int sc_offset = jobnr * 2 * steps_y; > > - const int sr_offset = jobnr * (MAX_MATRIX_SIZE - 1); > > - const int slice_start = (height * jobnr) / nb_jobs; > > - const int slice_end = (height * (jobnr+1)) / nb_jobs; > > - > > - int32_t res; > > - int x, y, z; > > - uint32_t tmp1, tmp2; > > - > > - if (!amount) { > > - av_image_copy_plane(dst + slice_start * dst_stride, dst_stride, > > - src + slice_start * src_stride, src_stride, > > - width, slice_end - slice_start); > > - return 0; > > - } > > - > > - for (y = 0; y < 2 * steps_y; y++) > > - memset(sc[sc_offset + y], 0, sizeof(sc[y][0]) * (width + 2 * > > steps_x)); > > - > > - // if this is not the first tile, we start from (slice_start - > > steps_y), > > - // so we can get smooth result at slice boundary > > - if (slice_start > steps_y) { > > - src += (slice_start - steps_y) * src_stride; > > - dst += (slice_start - steps_y) * dst_stride; > > - } > > - > > - for (y = -steps_y + slice_start; y < steps_y + slice_end; y++) { > > - if (y < height) > > - src2 = src; > > - > > - memset(sr + sr_offset, 0, sizeof(sr[0]) * (2 * steps_x - 1)); > > - for (x = -steps_x; x < width + steps_x; x++) { > > - tmp1 = x <= 0 ? src2[0] : x >= width ? src2[width-1] : > > src2[x]; > > - for (z = 0; z < steps_x * 2; z += 2) { > > - tmp2 = sr[sr_offset + z + 0] + tmp1; sr[sr_offset + z + > > 0] = tmp1; > > - tmp1 = sr[sr_offset + z + 1] + tmp2; sr[sr_offset + z + > > 1] = tmp2; > > - } > > - for (z = 0; z < steps_y * 2; z += 2) { > > - tmp2 = sc[sc_offset + z + 0][x + steps_x] + tmp1; > > sc[sc_offset + z + 0][x + steps_x] = tmp1; > > - tmp1 = sc[sc_offset + z + 1][x + steps_x] + tmp2; > > sc[sc_offset + z + 1][x + steps_x] = tmp2; > > - } > > - if (x >= steps_x && y >= (steps_y + slice_start)) { > > - const uint8_t *srx = src - steps_y * src_stride + x - > > steps_x; > > - uint8_t *dsx = dst - steps_y * dst_stride + x - > > steps_x; > > - > > - res = (int32_t)*srx + ((((int32_t) * srx - > > (int32_t)((tmp1 + halfscale) >> scalebits)) * amount) >> 16); > > - *dsx = av_clip_uint8(res); > > - } > > - } > > - if (y >= 0) { > > - dst += dst_stride; > > - src += src_stride; > > - } > > - } > > - return 0; > > +#define DEF_UNSHARP_SLICE_FUNC(name, nbits) > > \ > > +static int name##_##nbits(AVFilterContext *ctx, void *arg, int jobnr, int > > nb_jobs) \ > > +{ > > \ > > + ThreadData *td = arg; > > \ > > + UnsharpFilterParam *fp = td->fp; > > \ > > + UnsharpContext *s = ctx->priv; > > \ > > + uint32_t **sc = fp->sc; > > \ > > + uint32_t *sr = fp->sr; > > \ > > + const uint##nbits##_t *src2 = NULL; > > \ > > + const int amount = fp->amount; > > \ > > + const int steps_x = fp->steps_x; > > \ > > + const int steps_y = fp->steps_y; > > \ > > + const int scalebits = fp->scalebits; > > \ > > + const int32_t halfscale = fp->halfscale; > > \ > > + > > \ > > + uint##nbits##_t *dst = (uint##nbits##_t*)td->dst; > > \ > > + const uint##nbits##_t *src = (const uint##nbits##_t *)td->src; > > \ > > + int dst_stride = td->dst_stride; > > \ > > + int src_stride = td->src_stride; > > \ > > + const int width = td->width; > > \ > > + const int height = td->height; > > \ > > + const int sc_offset = jobnr * 2 * steps_y; > > \ > > + const int sr_offset = jobnr * (MAX_MATRIX_SIZE - 1); > > \ > > + const int slice_start = (height * jobnr) / nb_jobs; > > \ > > + const int slice_end = (height * (jobnr+1)) / nb_jobs; > > \ > > + > > \ > > + int32_t res; > > \ > > + int x, y, z; > > \ > > + uint32_t tmp1, tmp2; > > \ > > + > > \ > > + if (!amount) { > > \ > > + av_image_copy_plane(td->dst + slice_start * dst_stride, > > dst_stride, \ > > + td->src + slice_start * src_stride, > > src_stride, \ > > + width * s->bps, slice_end - slice_start); > > \ > > + return 0; > > \ > > + } > > \ > > + > > \ > > + for (y = 0; y < 2 * steps_y; y++) > > \ > > + memset(sc[sc_offset + y], 0, sizeof(sc[y][0]) * (width + 2 * > > steps_x)); \ > > + > > \ > > + dst_stride = dst_stride / s->bps; > > \ > > + src_stride = src_stride / s->bps; > > \ > > + /* if this is not the first tile, we start from (slice_start - > > steps_y) */ \ > > + /* so we can get smooth result at slice boundary */ > > \ > > + if (slice_start > steps_y) { > > \ > > + src += (slice_start - steps_y) * src_stride; > > \ > > + dst += (slice_start - steps_y) * dst_stride; > > \ > > + } > > \ > > + > > \ > > + for (y = -steps_y + slice_start; y < steps_y + slice_end; y++) { > > \ > > + if (y < height) > > \ > > + src2 = src; > > \ > > + > > \ > > + memset(sr + sr_offset, 0, sizeof(sr[0]) * (2 * steps_x - 1)); > > \ > > + for (x = -steps_x; x < width + steps_x; x++) { > > \ > > + tmp1 = x <= 0 ? src2[0] : x >= width ? src2[width-1] : > > src2[x]; \ > > + for (z = 0; z < steps_x * 2; z += 2) { > > \ > > + tmp2 = sr[sr_offset + z + 0] + tmp1; sr[sr_offset + z + > > 0] = tmp1; \ > > + tmp1 = sr[sr_offset + z + 1] + tmp2; sr[sr_offset + z + > > 1] = tmp2; \ > > + } > > \ > > + for (z = 0; z < steps_y * 2; z += 2) { > > \ > > + tmp2 = sc[sc_offset + z + 0][x + steps_x] + tmp1; > > \ > > + sc[sc_offset + z + 0][x + steps_x] = tmp1; > > \ > > + tmp1 = sc[sc_offset + z + 1][x + steps_x] + tmp2; > > \ > > + sc[sc_offset + z + 1][x + steps_x] = tmp2; > > \ > > + } > > \ > > + if (x >= steps_x && y >= (steps_y + slice_start)) { > > \ > > + const uint##nbits##_t *srx = src - steps_y * src_stride + > > x - steps_x; \ > > + uint##nbits##_t *dsx = dst - steps_y * dst_stride + > > x - steps_x; \ > > + > > \ > > + res = (int32_t)*srx + ((((int32_t) * srx - > > \ > > + (int32_t)((tmp1 + halfscale) >> scalebits)) * > > amount) >> (8+nbits)); \ > > + *dsx = av_clip_uint##nbits(res); > > \ > > + } > > \ > > + } > > \ > > + if (y >= 0) { > > \ > > + dst += dst_stride; > > \ > > + src += src_stride; > > \ > > + } > > \ > > + } > > \ > > + return 0; > > \ > > } > > +DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 16); > > +DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 8); > > > > Since the only difference is the bit depth of src/dst/srx/dsx, would it be > better to use one function only, > pass the bitdepth into the function and make the decision(uint8_t or > uint16_t) internally?(than #define the whole funcion) This is one general way to support more bit depth. Use macro based funtion(vf_lut3d.c..) or template file(see vf_phase.c). So if you have better way for suggestion, please give one example in the codebase. I'll study it. > > - Linjie -- Thanks, Limin Wang _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".