diff options
Diffstat (limited to 'libavfilter')
| -rw-r--r-- | libavfilter/Makefile | 3 | ||||
| -rw-r--r-- | libavfilter/allfilters.c | 2 | ||||
| -rw-r--r-- | libavfilter/deshake.h | 24 | ||||
| -rw-r--r-- | libavfilter/deshake_opencl.c | 198 | ||||
| -rw-r--r-- | libavfilter/deshake_opencl.h | 45 | ||||
| -rw-r--r-- | libavfilter/deshake_opencl_kernel.h | 225 | ||||
| -rw-r--r-- | libavfilter/opencl_allkernels.c | 41 | ||||
| -rw-r--r-- | libavfilter/opencl_allkernels.h | 29 | ||||
| -rw-r--r-- | libavfilter/unsharp.h | 31 | ||||
| -rw-r--r-- | libavfilter/unsharp_opencl.c | 422 | ||||
| -rw-r--r-- | libavfilter/unsharp_opencl.h | 34 | ||||
| -rw-r--r-- | libavfilter/unsharp_opencl_kernel.h | 342 | ||||
| -rw-r--r-- | libavfilter/vf_deshake.c | 23 | ||||
| -rw-r--r-- | libavfilter/vf_unsharp.c | 26 |
14 files changed, 3 insertions, 1442 deletions
diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 30474ea77b..89737b5ad0 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -19,7 +19,6 @@ OBJS = allfilters.o \ framequeue.o \ graphdump.o \ graphparser.o \ - opencl_allkernels.o \ transform.o \ video.o \ @@ -250,7 +249,6 @@ OBJS-$(CONFIG_NOISE_FILTER) += vf_noise.o OBJS-$(CONFIG_NULL_FILTER) += vf_null.o OBJS-$(CONFIG_OCR_FILTER) += vf_ocr.o OBJS-$(CONFIG_OCV_FILTER) += vf_libopencv.o -OBJS-$(CONFIG_OPENCL) += deshake_opencl.o unsharp_opencl.o OBJS-$(CONFIG_OSCILLOSCOPE_FILTER) += vf_datascope.o OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o framesync.o OBJS-$(CONFIG_OVERLAY_OPENCL_FILTER) += vf_overlay_opencl.o opencl.o \ @@ -396,7 +394,6 @@ OBJS-$(CONFIG_MOVIE_FILTER) += src_movie.o SLIBOBJS-$(HAVE_GNU_WINDRES) += avfilterres.o SKIPHEADERS-$(CONFIG_LIBVIDSTAB) += vidstabutils.h -SKIPHEADERS-$(CONFIG_OPENCL) += opencl_internal.h deshake_opencl_kernel.h unsharp_opencl_kernel.h OBJS-$(CONFIG_SHARED) += log2_tab.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index beacfce612..50e9d9a0cb 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -22,7 +22,6 @@ #include "libavutil/thread.h" #include "avfilter.h" #include "config.h" -#include "opencl_allkernels.h" #define REGISTER_FILTER(X, x, y) \ @@ -407,7 +406,6 @@ static void register_all(void) REGISTER_FILTER_UNCONDITIONAL(vsink_buffer); REGISTER_FILTER_UNCONDITIONAL(af_afifo); REGISTER_FILTER_UNCONDITIONAL(vf_fifo); - ff_opencl_register_filter_kernel_code_all(); } void avfilter_register_all(void) diff --git a/libavfilter/deshake.h b/libavfilter/deshake.h index 5e25bb3a7a..406cbab2f6 100644 --- a/libavfilter/deshake.h +++ b/libavfilter/deshake.h @@ -26,9 +26,6 @@ #include "avfilter.h" #include "transform.h" #include "libavutil/pixelutils.h" -#if CONFIG_OPENCL -#include "libavutil/opencl.h" -#endif enum SearchMethod { @@ -53,24 +50,6 @@ typedef struct Transform { double zoom; ///< Zoom percentage } Transform; -#if CONFIG_OPENCL - -typedef struct DeshakeOpenclContext { - cl_command_queue command_queue; - cl_program program; - cl_kernel kernel_luma; - cl_kernel kernel_chroma; - int in_plane_size[8]; - int out_plane_size[8]; - int plane_num; - cl_mem cl_inbuf; - size_t cl_inbuf_size; - cl_mem cl_outbuf; - size_t cl_outbuf_size; -} DeshakeOpenclContext; - -#endif - #define MAX_R 64 typedef struct DeshakeContext { @@ -96,9 +75,6 @@ typedef struct DeshakeContext { int cy; char *filename; ///< Motion search detailed log filename int opencl; -#if CONFIG_OPENCL - DeshakeOpenclContext opencl_ctx; -#endif int (* transform)(AVFilterContext *ctx, int width, int height, int cw, int ch, const float *matrix_y, const float *matrix_uv, enum InterpolateMethod interpolate, enum FillMethod fill, AVFrame *in, AVFrame *out); diff --git a/libavfilter/deshake_opencl.c b/libavfilter/deshake_opencl.c deleted file mode 100644 index 877ec1dc03..0000000000 --- a/libavfilter/deshake_opencl.c +++ /dev/null @@ -1,198 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * Copyright (C) 2013 Lenny Wang - * - * This file is part of FFmpeg. - * - * 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 - * transform input video - */ - -#include "libavutil/common.h" -#include "libavutil/dict.h" -#include "libavutil/pixdesc.h" -#include "deshake_opencl.h" -#include "libavutil/opencl_internal.h" - -#define PLANE_NUM 3 -#define ROUND_TO_16(a) (((((a) - 1)/16)+1)*16) - -int ff_opencl_transform(AVFilterContext *ctx, - int width, int height, int cw, int ch, - const float *matrix_y, const float *matrix_uv, - enum InterpolateMethod interpolate, - enum FillMethod fill, AVFrame *in, AVFrame *out) -{ - int ret = 0; - cl_int status; - DeshakeContext *deshake = ctx->priv; - float4 packed_matrix_lu = {matrix_y[0], matrix_y[1], matrix_y[2], matrix_y[5]}; - float4 packed_matrix_ch = {matrix_uv[0], matrix_uv[1], matrix_uv[2], matrix_uv[5]}; - size_t global_worksize_lu[2] = {(size_t)ROUND_TO_16(width), (size_t)ROUND_TO_16(height)}; - size_t global_worksize_ch[2] = {(size_t)ROUND_TO_16(cw), (size_t)(2*ROUND_TO_16(ch))}; - size_t local_worksize[2] = {16, 16}; - FFOpenclParam param_lu = {0}; - FFOpenclParam param_ch = {0}; - param_lu.ctx = param_ch.ctx = ctx; - param_lu.kernel = deshake->opencl_ctx.kernel_luma; - param_ch.kernel = deshake->opencl_ctx.kernel_chroma; - - if ((unsigned int)interpolate > INTERPOLATE_BIQUADRATIC) { - av_log(ctx, AV_LOG_ERROR, "Selected interpolate method is invalid\n"); - return AVERROR(EINVAL); - } - ret = avpriv_opencl_set_parameter(¶m_lu, - FF_OPENCL_PARAM_INFO(deshake->opencl_ctx.cl_inbuf), - FF_OPENCL_PARAM_INFO(deshake->opencl_ctx.cl_outbuf), - FF_OPENCL_PARAM_INFO(packed_matrix_lu), - FF_OPENCL_PARAM_INFO(interpolate), - FF_OPENCL_PARAM_INFO(fill), - FF_OPENCL_PARAM_INFO(in->linesize[0]), - FF_OPENCL_PARAM_INFO(out->linesize[0]), - FF_OPENCL_PARAM_INFO(height), - FF_OPENCL_PARAM_INFO(width), - NULL); - if (ret < 0) - return ret; - ret = avpriv_opencl_set_parameter(¶m_ch, - FF_OPENCL_PARAM_INFO(deshake->opencl_ctx.cl_inbuf), - FF_OPENCL_PARAM_INFO(deshake->opencl_ctx.cl_outbuf), - FF_OPENCL_PARAM_INFO(packed_matrix_ch), - FF_OPENCL_PARAM_INFO(interpolate), - FF_OPENCL_PARAM_INFO(fill), - FF_OPENCL_PARAM_INFO(in->linesize[0]), - FF_OPENCL_PARAM_INFO(out->linesize[0]), - FF_OPENCL_PARAM_INFO(in->linesize[1]), - FF_OPENCL_PARAM_INFO(out->linesize[1]), - FF_OPENCL_PARAM_INFO(height), - FF_OPENCL_PARAM_INFO(width), - FF_OPENCL_PARAM_INFO(ch), - FF_OPENCL_PARAM_INFO(cw), - NULL); - if (ret < 0) - return ret; - status = clEnqueueNDRangeKernel(deshake->opencl_ctx.command_queue, - deshake->opencl_ctx.kernel_luma, 2, NULL, - global_worksize_lu, local_worksize, 0, NULL, NULL); - status |= clEnqueueNDRangeKernel(deshake->opencl_ctx.command_queue, - deshake->opencl_ctx.kernel_chroma, 2, NULL, - global_worksize_ch, local_worksize, 0, NULL, NULL); - if (status != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL run kernel error occurred: %s\n", av_opencl_errstr(status)); - return AVERROR_EXTERNAL; - } - ret = av_opencl_buffer_read_image(out->data, deshake->opencl_ctx.out_plane_size, - deshake->opencl_ctx.plane_num, deshake->opencl_ctx.cl_outbuf, - deshake->opencl_ctx.cl_outbuf_size); - if (ret < 0) - return ret; - return ret; -} - -int ff_opencl_deshake_init(AVFilterContext *ctx) -{ - int ret = 0; - DeshakeContext *deshake = ctx->priv; - ret = av_opencl_init(NULL); - if (ret < 0) - return ret; - deshake->opencl_ctx.plane_num = PLANE_NUM; - deshake->opencl_ctx.command_queue = av_opencl_get_command_queue(); - if (!deshake->opencl_ctx.command_queue) { - av_log(ctx, AV_LOG_ERROR, "Unable to get OpenCL command queue in filter 'deshake'\n"); - return AVERROR(EINVAL); - } - deshake->opencl_ctx.program = av_opencl_compile("avfilter_transform", NULL); - if (!deshake->opencl_ctx.program) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to compile program 'avfilter_transform'\n"); - return AVERROR(EINVAL); - } - if (!deshake->opencl_ctx.kernel_luma) { - deshake->opencl_ctx.kernel_luma = clCreateKernel(deshake->opencl_ctx.program, - "avfilter_transform_luma", &ret); - if (ret != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel 'avfilter_transform_luma'\n"); - return AVERROR(EINVAL); - } - } - if (!deshake->opencl_ctx.kernel_chroma) { - deshake->opencl_ctx.kernel_chroma = clCreateKernel(deshake->opencl_ctx.program, - "avfilter_transform_chroma", &ret); - if (ret != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel 'avfilter_transform_chroma'\n"); - return AVERROR(EINVAL); - } - } - return ret; -} - -void ff_opencl_deshake_uninit(AVFilterContext *ctx) -{ - DeshakeContext *deshake = ctx->priv; - av_opencl_buffer_release(&deshake->opencl_ctx.cl_inbuf); - av_opencl_buffer_release(&deshake->opencl_ctx.cl_outbuf); - clReleaseKernel(deshake->opencl_ctx.kernel_luma); - clReleaseKernel(deshake->opencl_ctx.kernel_chroma); - clReleaseProgram(deshake->opencl_ctx.program); - deshake->opencl_ctx.command_queue = NULL; - av_opencl_uninit(); -} - -int ff_opencl_deshake_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out) -{ - int ret = 0; - AVFilterLink *link = ctx->inputs[0]; - DeshakeContext *deshake = ctx->priv; - const int hshift = av_pix_fmt_desc_get(link->format)->log2_chroma_h; - int chroma_height = AV_CEIL_RSHIFT(link->h, hshift); - - if ((!deshake->opencl_ctx.cl_inbuf) || (!deshake->opencl_ctx.cl_outbuf)) { - deshake->opencl_ctx.in_plane_size[0] = (in->linesize[0] * in->height); - deshake->opencl_ctx.in_plane_size[1] = (in->linesize[1] * chroma_height); - deshake->opencl_ctx.in_plane_size[2] = (in->linesize[2] * chroma_height); - deshake->opencl_ctx.out_plane_size[0] = (out->linesize[0] * out->height); - deshake->opencl_ctx.out_plane_size[1] = (out->linesize[1] * chroma_height); - deshake->opencl_ctx.out_plane_size[2] = (out->linesize[2] * chroma_height); - deshake->opencl_ctx.cl_inbuf_size = deshake->opencl_ctx.in_plane_size[0] + - deshake->opencl_ctx.in_plane_size[1] + - deshake->opencl_ctx.in_plane_size[2]; - deshake->opencl_ctx.cl_outbuf_size = deshake->opencl_ctx.out_plane_size[0] + - deshake->opencl_ctx.out_plane_size[1] + - deshake->opencl_ctx.out_plane_size[2]; - if (!deshake->opencl_ctx.cl_inbuf) { - ret = av_opencl_buffer_create(&deshake->opencl_ctx.cl_inbuf, - deshake->opencl_ctx.cl_inbuf_size, - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - } - if (!deshake->opencl_ctx.cl_outbuf) { - ret = av_opencl_buffer_create(&deshake->opencl_ctx.cl_outbuf, - deshake->opencl_ctx.cl_outbuf_size, - CL_MEM_READ_WRITE, NULL); - if (ret < 0) - return ret; - } - } - ret = av_opencl_buffer_write_image(deshake->opencl_ctx.cl_inbuf, - deshake->opencl_ctx.cl_inbuf_size, - 0, in->data,deshake->opencl_ctx.in_plane_size, - deshake->opencl_ctx.plane_num); - return ret; -} diff --git a/libavfilter/deshake_opencl.h b/libavfilter/deshake_opencl.h deleted file mode 100644 index f3d96dc4e9..0000000000 --- a/libavfilter/deshake_opencl.h +++ /dev/null @@ -1,45 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * - * This file is part of FFmpeg. - * - * 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 - */ - -#ifndef AVFILTER_DESHAKE_OPENCL_H -#define AVFILTER_DESHAKE_OPENCL_H - -#include "deshake.h" - -typedef struct float4 { - float x; - float y; - float z; - float w; -} float4; - -int ff_opencl_deshake_init(AVFilterContext *ctx); - -void ff_opencl_deshake_uninit(AVFilterContext *ctx); - -int ff_opencl_deshake_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out); - -int ff_opencl_transform(AVFilterContext *ctx, - int width, int height, int cw, int ch, - const float *matrix_y, const float *matrix_uv, - enum InterpolateMethod interpolate, - enum FillMethod fill, AVFrame *in, AVFrame *out); - -#endif /* AVFILTER_DESHAKE_OPENCL_H */ diff --git a/libavfilter/deshake_opencl_kernel.h b/libavfilter/deshake_opencl_kernel.h deleted file mode 100644 index dd45d6f60b..0000000000 --- a/libavfilter/deshake_opencl_kernel.h +++ /dev/null @@ -1,225 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * Copyright (C) 2013 Lenny Wang - * - * - * This file is part of FFmpeg. - * - * 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 - */ - -#ifndef AVFILTER_DESHAKE_OPENCL_KERNEL_H -#define AVFILTER_DESHAKE_OPENCL_KERNEL_H - -#include "libavutil/opencl.h" - -const char *ff_kernel_deshake_opencl = AV_OPENCL_KERNEL( -inline unsigned char pixel(global const unsigned char *src, int x, int y, - int w, int h,int stride, unsigned char def) -{ - return (x < 0 || y < 0 || x >= w || y >= h) ? def : src[x + y * stride]; -} - -unsigned char interpolate_nearest(float x, float y, global const unsigned char *src, - int width, int height, int stride, unsigned char def) -{ - return pixel(src, (int)(x + 0.5f), (int)(y + 0.5f), width, height, stride, def); -} - -unsigned char interpolate_bilinear(float x, float y, global const unsigned char *src, - int width, int height, int stride, unsigned char def) -{ - int x_c, x_f, y_c, y_f; - int v1, v2, v3, v4; - x_f = (int)x; - y_f = (int)y; - x_c = x_f + 1; - y_c = y_f + 1; - - if (x_f < -1 || x_f > width || y_f < -1 || y_f > height) { - return def; - } else { - v4 = pixel(src, x_f, y_f, width, height, stride, def); - v2 = pixel(src, x_c, y_f, width, height, stride, def); - v3 = pixel(src, x_f, y_c, width, height, stride, def); - v1 = pixel(src, x_c, y_c, width, height, stride, def); - return (v1*(x - x_f)*(y - y_f) + v2*((x - x_f)*(y_c - y)) + - v3*(x_c - x)*(y - y_f) + v4*((x_c - x)*(y_c - y))); - } -} - -unsigned char interpolate_biquadratic(float x, float y, global const unsigned char *src, - int width, int height, int stride, unsigned char def) -{ - int x_c, x_f, y_c, y_f; - unsigned char v1, v2, v3, v4; - float f1, f2, f3, f4; - x_f = (int)x; - y_f = (int)y; - x_c = x_f + 1; - y_c = y_f + 1; - - if (x_f < - 1 || x_f > width || y_f < -1 || y_f > height) - return def; - else { - v4 = pixel(src, x_f, y_f, width, height, stride, def); - v2 = pixel(src, x_c, y_f, width, height, stride, def); - v3 = pixel(src, x_f, y_c, width, height, stride, def); - v1 = pixel(src, x_c, y_c, width, height, stride, def); - - f1 = 1 - sqrt((x_c - x) * (y_c - y)); - f2 = 1 - sqrt((x_c - x) * (y - y_f)); - f3 = 1 - sqrt((x - x_f) * (y_c - y)); - f4 = 1 - sqrt((x - x_f) * (y - y_f)); - return (v1 * f1 + v2 * f2 + v3 * f3 + v4 * f4) / (f1 + f2 + f3 + f4); - } -} - -inline const float clipf(float a, float amin, float amax) -{ - if (a < amin) return amin; - else if (a > amax) return amax; - else return a; -} - -inline int mirror(int v, int m) -{ - while ((unsigned)v > (unsigned)m) { - v = -v; - if (v < 0) - v += 2 * m; - } - return v; -} - -kernel void avfilter_transform_luma(global unsigned char *src, - global unsigned char *dst, - float4 matrix, - int interpolate, - int fill, - int src_stride_lu, - int dst_stride_lu, - int height, - int width) -{ - int x = get_global_id(0); - int y = get_global_id(1); - int idx_dst = y * dst_stride_lu + x; - unsigned char def = 0; - float x_s = x * matrix.x + y * matrix.y + matrix.z; - float y_s = x * (-matrix.y) + y * matrix.x + matrix.w; - - if (x < width && y < height) { - switch (fill) { - case 0: //FILL_BLANK - def = 0; - break; - case 1: //FILL_ORIGINAL - def = src[y*src_stride_lu + x]; - break; - case 2: //FILL_CLAMP - y_s = clipf(y_s, 0, height - 1); - x_s = clipf(x_s, 0, width - 1); - def = src[(int)y_s * src_stride_lu + (int)x_s]; - break; - case 3: //FILL_MIRROR - y_s = mirror(y_s, height - 1); - x_s = mirror(x_s, width - 1); - def = src[(int)y_s * src_stride_lu + (int)x_s]; - break; - } - switch (interpolate) { - case 0: //INTERPOLATE_NEAREST - dst[idx_dst] = interpolate_nearest(x_s, y_s, src, width, height, src_stride_lu, def); - break; - case 1: //INTERPOLATE_BILINEAR - dst[idx_dst] = interpolate_bilinear(x_s, y_s, src, width, height, src_stride_lu, def); - break; - case 2: //INTERPOLATE_BIQUADRATIC - dst[idx_dst] = interpolate_biquadratic(x_s, y_s, src, width, height, src_stride_lu, def); - break; - default: - return; - } - } -} - -kernel void avfilter_transform_chroma(global unsigned char *src, - global unsigned char *dst, - float4 matrix, - int interpolate, - int fill, - int src_stride_lu, - int dst_stride_lu, - int src_stride_ch, - int dst_stride_ch, - int height, - int width, - int ch, - int cw) -{ - - int x = get_global_id(0); - int y = get_global_id(1); - int pad_ch = get_global_size(1)>>1; - global unsigned char *dst_u = dst + height * dst_stride_lu; - global unsigned char *src_u = src + height * src_stride_lu; - global unsigned char *dst_v = dst_u + ch * dst_stride_ch; - global unsigned char *src_v = src_u + ch * src_stride_ch; - src = y < pad_ch ? src_u : src_v; - dst = y < pad_ch ? dst_u : dst_v; - y = select(y - pad_ch, y, y < pad_ch); - float x_s = x * matrix.x + y * matrix.y + matrix.z; - float y_s = x * (-matrix.y) + y * matrix.x + matrix.w; - int idx_dst = y * dst_stride_ch + x; - unsigned char def; - - if (x < cw && y < ch) { - switch (fill) { - case 0: //FILL_BLANK - def = 0; - break; - case 1: //FILL_ORIGINAL - def = src[y*src_stride_ch + x]; - break; - case 2: //FILL_CLAMP - y_s = clipf(y_s, 0, ch - 1); - x_s = clipf(x_s, 0, cw - 1); - def = src[(int)y_s * src_stride_ch + (int)x_s]; - break; - case 3: //FILL_MIRROR - y_s = mirror(y_s, ch - 1); - x_s = mirror(x_s, cw - 1); - def = src[(int)y_s * src_stride_ch + (int)x_s]; - break; - } - switch (interpolate) { - case 0: //INTERPOLATE_NEAREST - dst[idx_dst] = interpolate_nearest(x_s, y_s, src, cw, ch, src_stride_ch, def); - break; - case 1: //INTERPOLATE_BILINEAR - dst[idx_dst] = interpolate_bilinear(x_s, y_s, src, cw, ch, src_stride_ch, def); - break; - case 2: //INTERPOLATE_BIQUADRATIC - dst[idx_dst] = interpolate_biquadratic(x_s, y_s, src, cw, ch, src_stride_ch, def); - break; - default: - return; - } - } -} -); - -#endif /* AVFILTER_DESHAKE_OPENCL_KERNEL_H */ diff --git a/libavfilter/opencl_allkernels.c b/libavfilter/opencl_allkernels.c deleted file mode 100644 index 6d80fa8598..0000000000 --- a/libavfilter/opencl_allkernels.c +++ /dev/null @@ -1,41 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * - * This file is part of FFmpeg. - * - * 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 "opencl_allkernels.h" -#if CONFIG_OPENCL -#include "libavutil/opencl.h" -#include "deshake_opencl_kernel.h" -#include "unsharp_opencl_kernel.h" -#endif - -#define OPENCL_REGISTER_KERNEL_CODE(X, x) \ - { \ - if (CONFIG_##X##_FILTER) { \ - av_opencl_register_kernel_code(ff_kernel_##x##_opencl); \ - } \ - } - -void ff_opencl_register_filter_kernel_code_all(void) -{ - #if CONFIG_OPENCL - OPENCL_REGISTER_KERNEL_CODE(DESHAKE, deshake); - OPENCL_REGISTER_KERNEL_CODE(UNSHARP, unsharp); - #endif -} diff --git a/libavfilter/opencl_allkernels.h b/libavfilter/opencl_allkernels.h deleted file mode 100644 index 57b650d27b..0000000000 --- a/libavfilter/opencl_allkernels.h +++ /dev/null @@ -1,29 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * - * This file is part of FFmpeg. - * - * 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 - */ - -#ifndef AVFILTER_OPENCL_ALLKERNELS_H -#define AVFILTER_OPENCL_ALLKERNELS_H - -#include "avfilter.h" -#include "config.h" - -void ff_opencl_register_filter_kernel_code_all(void); - -#endif /* AVFILTER_OPENCL_ALLKERNELS_H */ diff --git a/libavfilter/unsharp.h b/libavfilter/unsharp.h index 340a6a0caf..caff986fc1 100644 --- a/libavfilter/unsharp.h +++ b/libavfilter/unsharp.h @@ -24,38 +24,10 @@ #include "config.h" #include "avfilter.h" -#if CONFIG_OPENCL -#include "libavutil/opencl.h" -#endif #define MIN_MATRIX_SIZE 3 #define MAX_MATRIX_SIZE 63 -#if CONFIG_OPENCL - -typedef struct UnsharpOpenclContext { - cl_command_queue command_queue; - cl_program program; - cl_kernel kernel_default; - cl_kernel kernel_luma; - cl_kernel kernel_chroma; - cl_mem cl_luma_mask; - cl_mem cl_chroma_mask; - cl_mem cl_luma_mask_x; - cl_mem cl_chroma_mask_x; - cl_mem cl_luma_mask_y; - cl_mem cl_chroma_mask_y; - int in_plane_size[8]; - int out_plane_size[8]; - int plane_num; - cl_mem cl_inbuf; - size_t cl_inbuf_size; - cl_mem cl_outbuf; - size_t cl_outbuf_size; - int use_fast_kernels; -} UnsharpOpenclContext; - -#endif typedef struct UnsharpFilterParam { int msize_x; ///< matrix width @@ -76,9 +48,6 @@ typedef struct UnsharpContext { UnsharpFilterParam chroma; ///< chroma parameters (width, height, amount) int hsub, vsub; int opencl; -#if CONFIG_OPENCL - UnsharpOpenclContext opencl_ctx; -#endif int (* apply_unsharp)(AVFilterContext *ctx, AVFrame *in, AVFrame *out); } UnsharpContext; diff --git a/libavfilter/unsharp_opencl.c b/libavfilter/unsharp_opencl.c deleted file mode 100644 index 1545455846..0000000000 --- a/libavfilter/unsharp_opencl.c +++ /dev/null @@ -1,422 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * Copyright (C) 2013 Lenny Wang - * - * This file is part of FFmpeg. - * - * 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 - * unsharp input video - */ - -#include "unsharp_opencl.h" -#include "libavutil/common.h" -#include "libavutil/opencl_internal.h" - -#define PLANE_NUM 3 -#define ROUND_TO_16(a) (((((a) - 1)/16)+1)*16) - -static inline void add_mask_counter(uint32_t *dst, uint32_t *counter1, uint32_t *counter2, int len) -{ - int i; - for (i = 0; i < len; i++) { - dst[i] = counter1[i] + counter2[i]; - } -} - -static int compute_mask(int step, uint32_t *mask) -{ - int i, z, ret = 0; - int counter_size = sizeof(uint32_t) * (2 * step + 1); - uint32_t *temp1_counter, *temp2_counter, **counter = NULL; - temp1_counter = av_mallocz(counter_size); - if (!temp1_counter) { - ret = AVERROR(ENOMEM); - goto end; - } - temp2_counter = av_mallocz(counter_size); - if (!temp2_counter) { - ret = AVERROR(ENOMEM); - goto end; - } - counter = av_mallocz_array(2 * step + 1, sizeof(uint32_t *)); - if (!counter) { - ret = AVERROR(ENOMEM); - goto end; - } - for (i = 0; i < 2 * step + 1; i++) { - counter[i] = av_mallocz(counter_size); - if (!counter[i]) { - ret = AVERROR(ENOMEM); - goto end; - } - } - for (i = 0; i < 2 * step + 1; i++) { - memset(temp1_counter, 0, counter_size); - temp1_counter[i] = 1; - for (z = 0; z < step * 2; z += 2) { - add_mask_counter(temp2_counter, counter[z], temp1_counter, step * 2); - memcpy(counter[z], temp1_counter, counter_size); - add_mask_counter(temp1_counter, counter[z + 1], temp2_counter, step * 2); - memcpy(counter[z + 1], temp2_counter, counter_size); - } - } - memcpy(mask, temp1_counter, counter_size); -end: - av_freep(&temp1_counter); - av_freep(&temp2_counter); - for (i = 0; counter && i < 2 * step + 1; i++) { - av_freep(&counter[i]); - } - av_freep(&counter); - return ret; -} - -static int copy_separable_masks(cl_mem cl_mask_x, cl_mem cl_mask_y, int step_x, int step_y) -{ - int ret = 0; - uint32_t *mask_x, *mask_y; - size_t size_mask_x = sizeof(uint32_t) * (2 * step_x + 1); - size_t size_mask_y = sizeof(uint32_t) * (2 * step_y + 1); - mask_x = av_mallocz_array(2 * step_x + 1, sizeof(uint32_t)); - if (!mask_x) { - ret = AVERROR(ENOMEM); - goto end; - } - mask_y = av_mallocz_array(2 * step_y + 1, sizeof(uint32_t)); - if (!mask_y) { - ret = AVERROR(ENOMEM); - goto end; - } - - ret = compute_mask(step_x, mask_x); - if (ret < 0) - goto end; - ret = compute_mask(step_y, mask_y); - if (ret < 0) - goto end; - - ret = av_opencl_buffer_write(cl_mask_x, (uint8_t *)mask_x, size_mask_x); - ret = av_opencl_buffer_write(cl_mask_y, (uint8_t *)mask_y, size_mask_y); -end: - av_freep(&mask_x); - av_freep(&mask_y); - - return ret; -} - -static int generate_mask(AVFilterContext *ctx) -{ - cl_mem masks[4]; - cl_mem mask_matrix[2]; - int i, ret = 0, step_x[2], step_y[2]; - - UnsharpContext *unsharp = ctx->priv; - mask_matrix[0] = unsharp->opencl_ctx.cl_luma_mask; - mask_matrix[1] = unsharp->opencl_ctx.cl_chroma_mask; - masks[0] = unsharp->opencl_ctx.cl_luma_mask_x; - masks[1] = unsharp->opencl_ctx.cl_luma_mask_y; - masks[2] = unsharp->opencl_ctx.cl_chroma_mask_x; - masks[3] = unsharp->opencl_ctx.cl_chroma_mask_y; - step_x[0] = unsharp->luma.steps_x; - step_x[1] = unsharp->chroma.steps_x; - step_y[0] = unsharp->luma.steps_y; - step_y[1] = unsharp->chroma.steps_y; - - /* use default kernel if any matrix dim larger than 8 due to limited local mem size */ - if (step_x[0]>8 || step_x[1]>8 || step_y[0]>8 || step_y[1]>8) - unsharp->opencl_ctx.use_fast_kernels = 0; - else - unsharp->opencl_ctx.use_fast_kernels = 1; - - if (!masks[0] || !masks[1] || !masks[2] || !masks[3]) { - av_log(ctx, AV_LOG_ERROR, "Luma mask and chroma mask should not be NULL\n"); - return AVERROR(EINVAL); - } - if (!mask_matrix[0] || !mask_matrix[1]) { - av_log(ctx, AV_LOG_ERROR, "Luma mask and chroma mask should not be NULL\n"); - return AVERROR(EINVAL); - } - for (i = 0; i < 2; i++) { - ret = copy_separable_masks(masks[2*i], masks[2*i+1], step_x[i], step_y[i]); - if (ret < 0) - return ret; - } - return ret; -} - -int ff_opencl_apply_unsharp(AVFilterContext *ctx, AVFrame *in, AVFrame *out) -{ - int ret; - AVFilterLink *link = ctx->inputs[0]; - UnsharpContext *unsharp = ctx->priv; - cl_int status; - FFOpenclParam kernel1 = {0}; - FFOpenclParam kernel2 = {0}; - int width = link->w; - int height = link->h; - int cw = AV_CEIL_RSHIFT(link->w, unsharp->hsub); - int ch = AV_CEIL_RSHIFT(link->h, unsharp->vsub); - size_t globalWorkSize1d = width * height + 2 * ch * cw; - size_t globalWorkSize2dLuma[2]; - size_t globalWorkSize2dChroma[2]; - size_t localWorkSize2d[2] = {16, 16}; - - if (unsharp->opencl_ctx.use_fast_kernels) { - globalWorkSize2dLuma[0] = (size_t)ROUND_TO_16(width); - globalWorkSize2dLuma[1] = (size_t)ROUND_TO_16(height); - globalWorkSize2dChroma[0] = (size_t)ROUND_TO_16(cw); - globalWorkSize2dChroma[1] = (size_t)(2*ROUND_TO_16(ch)); - - kernel1.ctx = ctx; - kernel1.kernel = unsharp->opencl_ctx.kernel_luma; - ret = avpriv_opencl_set_parameter(&kernel1, - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_inbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_outbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_luma_mask_x), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_luma_mask_y), - FF_OPENCL_PARAM_INFO(unsharp->luma.amount), - FF_OPENCL_PARAM_INFO(unsharp->luma.scalebits), - FF_OPENCL_PARAM_INFO(unsharp->luma.halfscale), - FF_OPENCL_PARAM_INFO(in->linesize[0]), - FF_OPENCL_PARAM_INFO(out->linesize[0]), - FF_OPENCL_PARAM_INFO(width), - FF_OPENCL_PARAM_INFO(height), - NULL); - if (ret < 0) - return ret; - - kernel2.ctx = ctx; - kernel2.kernel = unsharp->opencl_ctx.kernel_chroma; - ret = avpriv_opencl_set_parameter(&kernel2, - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_inbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_outbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_chroma_mask_x), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_chroma_mask_y), - FF_OPENCL_PARAM_INFO(unsharp->chroma.amount), - FF_OPENCL_PARAM_INFO(unsharp->chroma.scalebits), - FF_OPENCL_PARAM_INFO(unsharp->chroma.halfscale), - FF_OPENCL_PARAM_INFO(in->linesize[0]), - FF_OPENCL_PARAM_INFO(in->linesize[1]), - FF_OPENCL_PARAM_INFO(out->linesize[0]), - FF_OPENCL_PARAM_INFO(out->linesize[1]), - FF_OPENCL_PARAM_INFO(link->w), - FF_OPENCL_PARAM_INFO(link->h), - FF_OPENCL_PARAM_INFO(cw), - FF_OPENCL_PARAM_INFO(ch), - NULL); - if (ret < 0) - return ret; - status = clEnqueueNDRangeKernel(unsharp->opencl_ctx.command_queue, - unsharp->opencl_ctx.kernel_luma, 2, NULL, - globalWorkSize2dLuma, localWorkSize2d, 0, NULL, NULL); - status |=clEnqueueNDRangeKernel(unsharp->opencl_ctx.command_queue, - unsharp->opencl_ctx.kernel_chroma, 2, NULL, - globalWorkSize2dChroma, localWorkSize2d, 0, NULL, NULL); - if (status != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL run kernel error occurred: %s\n", av_opencl_errstr(status)); - return AVERROR_EXTERNAL; - } - } else { /* use default kernel */ - kernel1.ctx = ctx; - kernel1.kernel = unsharp->opencl_ctx.kernel_default; - - ret = avpriv_opencl_set_parameter(&kernel1, - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_inbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_outbuf), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_luma_mask), - FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_chroma_mask), - FF_OPENCL_PARAM_INFO(unsharp->luma.amount), - FF_OPENCL_PARAM_INFO(unsharp->chroma.amount), - FF_OPENCL_PARAM_INFO(unsharp->luma.steps_x), - FF_OPENCL_PARAM_INFO(unsharp->luma.steps_y), - FF_OPENCL_PARAM_INFO(unsharp->chroma.steps_x), - FF_OPENCL_PARAM_INFO(unsharp->chroma.steps_y), - FF_OPENCL_PARAM_INFO(unsharp->luma.scalebits), - FF_OPENCL_PARAM_INFO(unsharp->chroma.scalebits), - FF_OPENCL_PARAM_INFO(unsharp->luma.halfscale), - FF_OPENCL_PARAM_INFO(unsharp->chroma.halfscale), - FF_OPENCL_PARAM_INFO(in->linesize[0]), - FF_OPENCL_PARAM_INFO(in->linesize[1]), - FF_OPENCL_PARAM_INFO(out->linesize[0]), - FF_OPENCL_PARAM_INFO(out->linesize[1]), - FF_OPENCL_PARAM_INFO(link->h), - FF_OPENCL_PARAM_INFO(link->w), - FF_OPENCL_PARAM_INFO(ch), - FF_OPENCL_PARAM_INFO(cw), - NULL); - if (ret < 0) - return ret; - status = clEnqueueNDRangeKernel(unsharp->opencl_ctx.command_queue, - unsharp->opencl_ctx.kernel_default, 1, NULL, - &globalWorkSize1d, NULL, 0, NULL, NULL); - if (status != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL run kernel error occurred: %s\n", av_opencl_errstr(status)); - return AVERROR_EXTERNAL; - } - } - //blocking map is suffficient, no need for clFinish - //clFinish(unsharp->opencl_ctx.command_queue); - - return av_opencl_buffer_read_image(out->data, unsharp->opencl_ctx.out_plane_size, - unsharp->opencl_ctx.plane_num, unsharp->opencl_ctx.cl_outbuf, - unsharp->opencl_ctx.cl_outbuf_size); -} - -int ff_opencl_unsharp_init(AVFilterContext *ctx) -{ - int ret = 0; - char build_opts[96]; - UnsharpContext *unsharp = ctx->priv; - ret = av_opencl_init(NULL); - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_luma_mask, - sizeof(uint32_t) * (2 * unsharp->luma.steps_x + 1) * (2 * unsharp->luma.steps_y + 1), - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_chroma_mask, - sizeof(uint32_t) * (2 * unsharp->chroma.steps_x + 1) * (2 * unsharp->chroma.steps_y + 1), - CL_MEM_READ_ONLY, NULL); - // separable filters - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_luma_mask_x, - sizeof(uint32_t) * (2 * unsharp->luma.steps_x + 1), - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_luma_mask_y, - sizeof(uint32_t) * (2 * unsharp->luma.steps_y + 1), - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_chroma_mask_x, - sizeof(uint32_t) * (2 * unsharp->chroma.steps_x + 1), - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_chroma_mask_y, - sizeof(uint32_t) * (2 * unsharp->chroma.steps_y + 1), - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - ret = generate_mask(ctx); - if (ret < 0) - return ret; - unsharp->opencl_ctx.plane_num = PLANE_NUM; - unsharp->opencl_ctx.command_queue = av_opencl_get_command_queue(); - if (!unsharp->opencl_ctx.command_queue) { - av_log(ctx, AV_LOG_ERROR, "Unable to get OpenCL command queue in filter 'unsharp'\n"); - return AVERROR(EINVAL); - } - snprintf(build_opts, 96, "-D LU_RADIUS_X=%d -D LU_RADIUS_Y=%d -D CH_RADIUS_X=%d -D CH_RADIUS_Y=%d", - 2*unsharp->luma.steps_x+1, 2*unsharp->luma.steps_y+1, 2*unsharp->chroma.steps_x+1, 2*unsharp->chroma.steps_y+1); - unsharp->opencl_ctx.program = av_opencl_compile("unsharp", build_opts); - if (!unsharp->opencl_ctx.program) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to compile program 'unsharp'\n"); - return AVERROR(EINVAL); - } - if (unsharp->opencl_ctx.use_fast_kernels) { - if (!unsharp->opencl_ctx.kernel_luma) { - unsharp->opencl_ctx.kernel_luma = clCreateKernel(unsharp->opencl_ctx.program, "unsharp_luma", &ret); - if (ret != CL_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel 'unsharp_luma'\n"); - return ret; - } - } - if (!unsharp->opencl_ctx.kernel_chroma) { - unsharp->opencl_ctx.kernel_chroma = clCreateKernel(unsharp->opencl_ctx.program, "unsharp_chroma", &ret); - if (ret < 0) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel 'unsharp_chroma'\n"); - return ret; - } - } - } - else { - if (!unsharp->opencl_ctx.kernel_default) { - unsharp->opencl_ctx.kernel_default = clCreateKernel(unsharp->opencl_ctx.program, "unsharp_default", &ret); - if (ret < 0) { - av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel 'unsharp_default'\n"); - return ret; - } - } - } - return ret; -} - -void ff_opencl_unsharp_uninit(AVFilterContext *ctx) -{ - UnsharpContext *unsharp = ctx->priv; - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_inbuf); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_outbuf); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_luma_mask); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_chroma_mask); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_luma_mask_x); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_chroma_mask_x); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_luma_mask_y); - av_opencl_buffer_release(&unsharp->opencl_ctx.cl_chroma_mask_y); - clReleaseKernel(unsharp->opencl_ctx.kernel_default); - clReleaseKernel(unsharp->opencl_ctx.kernel_luma); - clReleaseKernel(unsharp->opencl_ctx.kernel_chroma); - clReleaseProgram(unsharp->opencl_ctx.program); - unsharp->opencl_ctx.command_queue = NULL; - av_opencl_uninit(); -} - -int ff_opencl_unsharp_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out) -{ - int ret = 0; - AVFilterLink *link = ctx->inputs[0]; - UnsharpContext *unsharp = ctx->priv; - int ch = AV_CEIL_RSHIFT(link->h, unsharp->vsub); - - if ((!unsharp->opencl_ctx.cl_inbuf) || (!unsharp->opencl_ctx.cl_outbuf)) { - unsharp->opencl_ctx.in_plane_size[0] = (in->linesize[0] * in->height); - unsharp->opencl_ctx.in_plane_size[1] = (in->linesize[1] * ch); - unsharp->opencl_ctx.in_plane_size[2] = (in->linesize[2] * ch); - unsharp->opencl_ctx.out_plane_size[0] = (out->linesize[0] * out->height); - unsharp->opencl_ctx.out_plane_size[1] = (out->linesize[1] * ch); - unsharp->opencl_ctx.out_plane_size[2] = (out->linesize[2] * ch); - unsharp->opencl_ctx.cl_inbuf_size = unsharp->opencl_ctx.in_plane_size[0] + - unsharp->opencl_ctx.in_plane_size[1] + - unsharp->opencl_ctx.in_plane_size[2]; - unsharp->opencl_ctx.cl_outbuf_size = unsharp->opencl_ctx.out_plane_size[0] + - unsharp->opencl_ctx.out_plane_size[1] + - unsharp->opencl_ctx.out_plane_size[2]; - if (!unsharp->opencl_ctx.cl_inbuf) { - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_inbuf, - unsharp->opencl_ctx.cl_inbuf_size, - CL_MEM_READ_ONLY, NULL); - if (ret < 0) - return ret; - } - if (!unsharp->opencl_ctx.cl_outbuf) { - ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_outbuf, - unsharp->opencl_ctx.cl_outbuf_size, - CL_MEM_READ_WRITE, NULL); - if (ret < 0) - return ret; - } - } - return av_opencl_buffer_write_image(unsharp->opencl_ctx.cl_inbuf, - unsharp->opencl_ctx.cl_inbuf_size, - 0, in->data, unsharp->opencl_ctx.in_plane_size, - unsharp->opencl_ctx.plane_num); -} diff --git a/libavfilter/unsharp_opencl.h b/libavfilter/unsharp_opencl.h deleted file mode 100644 index 3aefab62e0..0000000000 --- a/libavfilter/unsharp_opencl.h +++ /dev/null @@ -1,34 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * - * This file is part of FFmpeg. - * - * 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 - */ - -#ifndef AVFILTER_UNSHARP_OPENCL_H -#define AVFILTER_UNSHARP_OPENCL_H - -#include "unsharp.h" - -int ff_opencl_unsharp_init(AVFilterContext *ctx); - -void ff_opencl_unsharp_uninit(AVFilterContext *ctx); - -int ff_opencl_unsharp_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out); - -int ff_opencl_apply_unsharp(AVFilterContext *ctx, AVFrame *in, AVFrame *out); - -#endif /* AVFILTER_UNSHARP_OPENCL_H */ diff --git a/libavfilter/unsharp_opencl_kernel.h b/libavfilter/unsharp_opencl_kernel.h deleted file mode 100644 index 307d0f1814..0000000000 --- a/libavfilter/unsharp_opencl_kernel.h +++ /dev/null @@ -1,342 +0,0 @@ -/* - * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> - * Copyright (C) 2013 Lenny Wang - * - * This file is part of FFmpeg. - * - * 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 - */ - -#ifndef AVFILTER_UNSHARP_OPENCL_KERNEL_H -#define AVFILTER_UNSHARP_OPENCL_KERNEL_H - -#include "libavutil/opencl.h" - -const char *ff_kernel_unsharp_opencl = AV_OPENCL_KERNEL( -inline unsigned char clip_uint8(int a) -{ - if (a & (~0xFF)) - return (-a)>>31; - else - return a; -} - -kernel void unsharp_luma( - global unsigned char *src, - global unsigned char *dst, - global int *mask_x, - global int *mask_y, - int amount, - int scalebits, - int halfscale, - int src_stride, - int dst_stride, - int width, - int height) -{ - int2 threadIdx, blockIdx, globalIdx; - threadIdx.x = get_local_id(0); - threadIdx.y = get_local_id(1); - blockIdx.x = get_group_id(0); - blockIdx.y = get_group_id(1); - globalIdx.x = get_global_id(0); - globalIdx.y = get_global_id(1); - - if (!amount) { - if (globalIdx.x < width && globalIdx.y < height) - dst[globalIdx.x + globalIdx.y*dst_stride] = src[globalIdx.x + globalIdx.y*src_stride]; - return; - } - - local unsigned int l[32][32]; - local unsigned int lcx[LU_RADIUS_X]; - local unsigned int lcy[LU_RADIUS_Y]; - int indexIx, indexIy, i, j; - - //load up tile: actual workspace + halo of 8 points in x and y \n - for(i = 0; i <= 1; i++) { - indexIy = -8 + (blockIdx.y + i) * 16 + threadIdx.y; - indexIy = indexIy < 0 ? 0 : indexIy; - indexIy = indexIy >= height ? height - 1: indexIy; - for(j = 0; j <= 1; j++) { - indexIx = -8 + (blockIdx.x + j) * 16 + threadIdx.x; - indexIx = indexIx < 0 ? 0 : indexIx; - indexIx = indexIx >= width ? width - 1: indexIx; - l[i*16 + threadIdx.y][j*16 + threadIdx.x] = src[indexIy*src_stride + indexIx]; - } - } - - int indexL = threadIdx.y*16 + threadIdx.x; - if (indexL < LU_RADIUS_X) - lcx[indexL] = mask_x[indexL]; - if (indexL < LU_RADIUS_Y) - lcy[indexL] = mask_y[indexL]; - barrier(CLK_LOCAL_MEM_FENCE); - - //needed for unsharp mask application in the end \n - int orig_value = (int)l[threadIdx.y + 8][threadIdx.x + 8]; - - int idx, idy, maskIndex; - int temp[2] = {0}; - int steps_x = (LU_RADIUS_X-1)/2; - int steps_y = (LU_RADIUS_Y-1)/2; - - // compute the actual workspace + left&right halos \n - \n#pragma unroll\n - for (j = 0; j <=1; j++) { - //extra work to cover left and right halos \n - idx = 16*j + threadIdx.x; - \n#pragma unroll\n - for (i = -steps_y; i <= steps_y; i++) { - idy = 8 + i + threadIdx.y; - maskIndex = (i + steps_y); - temp[j] += (int)l[idy][idx] * lcy[maskIndex]; - } - } - barrier(CLK_LOCAL_MEM_FENCE); - //save results from the vertical filter in local memory \n - idy = 8 + threadIdx.y; - \n#pragma unroll\n - for (j = 0; j <=1; j++) { - idx = 16*j + threadIdx.x; - l[idy][idx] = temp[j]; - } - barrier(CLK_LOCAL_MEM_FENCE); - - //compute results with the horizontal filter \n - int sum = 0; - idy = 8 + threadIdx.y; - \n#pragma unroll\n - for (j = -steps_x; j <= steps_x; j++) { - idx = 8 + j + threadIdx.x; - maskIndex = j + steps_x; - sum += (int)l[idy][idx] * lcx[maskIndex]; - } - - int res = orig_value + (((orig_value - (int)((sum + halfscale) >> scalebits)) * amount) >> 16); - - if (globalIdx.x < width && globalIdx.y < height) - dst[globalIdx.x + globalIdx.y*dst_stride] = clip_uint8(res); -} - -kernel void unsharp_chroma( - global unsigned char *src_y, - global unsigned char *dst_y, - global int *mask_x, - global int *mask_y, - int amount, - int scalebits, - int halfscale, - int src_stride_lu, - int src_stride_ch, - int dst_stride_lu, - int dst_stride_ch, - int width, - int height, - int cw, - int ch) -{ - global unsigned char *dst_u = dst_y + height * dst_stride_lu; - global unsigned char *dst_v = dst_u + ch * dst_stride_ch; - global unsigned char *src_u = src_y + height * src_stride_lu; - global unsigned char *src_v = src_u + ch * src_stride_ch; - int2 threadIdx, blockIdx, globalIdx; - threadIdx.x = get_local_id(0); - threadIdx.y = get_local_id(1); - blockIdx.x = get_group_id(0); - blockIdx.y = get_group_id(1); - globalIdx.x = get_global_id(0); - globalIdx.y = get_global_id(1); - int padch = get_global_size(1)/2; - global unsigned char *src = globalIdx.y>=padch ? src_v : src_u; - global unsigned char *dst = globalIdx.y>=padch ? dst_v : dst_u; - - blockIdx.y = globalIdx.y>=padch ? blockIdx.y - get_num_groups(1)/2 : blockIdx.y; - globalIdx.y = globalIdx.y>=padch ? globalIdx.y - padch : globalIdx.y; - - if (!amount) { - if (globalIdx.x < cw && globalIdx.y < ch) - dst[globalIdx.x + globalIdx.y*dst_stride_ch] = src[globalIdx.x + globalIdx.y*src_stride_ch]; - return; - } - - local unsigned int l[32][32]; - local unsigned int lcx[CH_RADIUS_X]; - local unsigned int lcy[CH_RADIUS_Y]; - int indexIx, indexIy, i, j; - for(i = 0; i <= 1; i++) { - indexIy = -8 + (blockIdx.y + i) * 16 + threadIdx.y; - indexIy = indexIy < 0 ? 0 : indexIy; - indexIy = indexIy >= ch ? ch - 1: indexIy; - for(j = 0; j <= 1; j++) { - indexIx = -8 + (blockIdx.x + j) * 16 + threadIdx.x; - indexIx = indexIx < 0 ? 0 : indexIx; - indexIx = indexIx >= cw ? cw - 1: indexIx; - l[i*16 + threadIdx.y][j*16 + threadIdx.x] = src[indexIy * src_stride_ch + indexIx]; - } - } - - int indexL = threadIdx.y*16 + threadIdx.x; - if (indexL < CH_RADIUS_X) - lcx[indexL] = mask_x[indexL]; - if (indexL < CH_RADIUS_Y) - lcy[indexL] = mask_y[indexL]; - barrier(CLK_LOCAL_MEM_FENCE); - - int orig_value = (int)l[threadIdx.y + 8][threadIdx.x + 8]; - - int idx, idy, maskIndex; - int steps_x = CH_RADIUS_X/2; - int steps_y = CH_RADIUS_Y/2; - int temp[2] = {0,0}; - - \n#pragma unroll\n - for (j = 0; j <= 1; j++) { - idx = 16*j + threadIdx.x; - \n#pragma unroll\n - for (i = -steps_y; i <= steps_y; i++) { - idy = 8 + i + threadIdx.y; - maskIndex = i + steps_y; - temp[j] += (int)l[idy][idx] * lcy[maskIndex]; - } - } - - barrier(CLK_LOCAL_MEM_FENCE); - idy = 8 + threadIdx.y; - \n#pragma unroll\n - for (j = 0; j <= 1; j++) { - idx = 16*j + threadIdx.x; - l[idy][idx] = temp[j]; - } - barrier(CLK_LOCAL_MEM_FENCE); - - //compute results with the horizontal filter \n - int sum = 0; - idy = 8 + threadIdx.y; - \n#pragma unroll\n - for (j = -steps_x; j <= steps_x; j++) { - idx = 8 + j + threadIdx.x; - maskIndex = j + steps_x; - sum += (int)l[idy][idx] * lcx[maskIndex]; - } - - int res = orig_value + (((orig_value - (int)((sum + halfscale) >> scalebits)) * amount) >> 16); - - if (globalIdx.x < cw && globalIdx.y < ch) - dst[globalIdx.x + globalIdx.y*dst_stride_ch] = clip_uint8(res); -} - -kernel void unsharp_default(global unsigned char *src, - global unsigned char *dst, - const global unsigned int *mask_lu, - const global unsigned int *mask_ch, - int amount_lu, - int amount_ch, - int step_x_lu, - int step_y_lu, - int step_x_ch, - int step_y_ch, - int scalebits_lu, - int scalebits_ch, - int halfscale_lu, - int halfscale_ch, - int src_stride_lu, - int src_stride_ch, - int dst_stride_lu, - int dst_stride_ch, - int height, - int width, - int ch, - int cw) -{ - global unsigned char *dst_y = dst; - global unsigned char *dst_u = dst_y + height * dst_stride_lu; - global unsigned char *dst_v = dst_u + ch * dst_stride_ch; - - global unsigned char *src_y = src; - global unsigned char *src_u = src_y + height * src_stride_lu; - global unsigned char *src_v = src_u + ch * src_stride_ch; - - global unsigned char *temp_dst; - global unsigned char *temp_src; - const global unsigned int *temp_mask; - int global_id = get_global_id(0); - int i, j, x, y, temp_src_stride, temp_dst_stride, temp_height, temp_width, temp_steps_x, temp_steps_y, - temp_amount, temp_scalebits, temp_halfscale, sum, idx_x, idx_y, temp, res; - if (global_id < width * height) { - y = global_id / width; - x = global_id % width; - temp_dst = dst_y; - temp_src = src_y; - temp_src_stride = src_stride_lu; - temp_dst_stride = dst_stride_lu; - temp_height = height; - temp_width = width; - temp_steps_x = step_x_lu; - temp_steps_y = step_y_lu; - temp_mask = mask_lu; - temp_amount = amount_lu; - temp_scalebits = scalebits_lu; - temp_halfscale = halfscale_lu; - } else if ((global_id >= width * height) && (global_id < width * height + ch * cw)) { - y = (global_id - width * height) / cw; - x = (global_id - width * height) % cw; - temp_dst = dst_u; - temp_src = src_u; - temp_src_stride = src_stride_ch; - temp_dst_stride = dst_stride_ch; - temp_height = ch; - temp_width = cw; - temp_steps_x = step_x_ch; - temp_steps_y = step_y_ch; - temp_mask = mask_ch; - temp_amount = amount_ch; - temp_scalebits = scalebits_ch; - temp_halfscale = halfscale_ch; - } else { - y = (global_id - width * height - ch * cw) / cw; - x = (global_id - width * height - ch * cw) % cw; - temp_dst = dst_v; - temp_src = src_v; - temp_src_stride = src_stride_ch; - temp_dst_stride = dst_stride_ch; - temp_height = ch; - temp_width = cw; - temp_steps_x = step_x_ch; - temp_steps_y = step_y_ch; - temp_mask = mask_ch; - temp_amount = amount_ch; - temp_scalebits = scalebits_ch; - temp_halfscale = halfscale_ch; - } - if (temp_amount) { - sum = 0; - for (j = 0; j <= 2 * temp_steps_y; j++) { - idx_y = (y - temp_steps_y + j) <= 0 ? 0 : (y - temp_steps_y + j) >= temp_height ? temp_height-1 : y - temp_steps_y + j; - for (i = 0; i <= 2 * temp_steps_x; i++) { - idx_x = (x - temp_steps_x + i) <= 0 ? 0 : (x - temp_steps_x + i) >= temp_width ? temp_width-1 : x - temp_steps_x + i; - sum += temp_mask[i + j * (2 * temp_steps_x + 1)] * temp_src[idx_x + idx_y * temp_src_stride]; - } - } - temp = (int)temp_src[x + y * temp_src_stride]; - res = temp + (((temp - (int)((sum + temp_halfscale) >> temp_scalebits)) * temp_amount) >> 16); - temp_dst[x + y * temp_dst_stride] = clip_uint8(res); - } else { - temp_dst[x + y * temp_dst_stride] = temp_src[x + y * temp_src_stride]; - } -} -); - -#endif /* AVFILTER_UNSHARP_OPENCL_KERNEL_H */ diff --git a/libavfilter/vf_deshake.c b/libavfilter/vf_deshake.c index 64b48c6d02..fb4eb355b8 100644 --- a/libavfilter/vf_deshake.c +++ b/libavfilter/vf_deshake.c @@ -60,7 +60,6 @@ #include "libavutil/qsort.h" #include "deshake.h" -#include "deshake_opencl.h" #define OFFSET(x) offsetof(DeshakeContext, x) #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM @@ -83,7 +82,7 @@ static const AVOption deshake_options[] = { { "exhaustive", "exhaustive search", 0, AV_OPT_TYPE_CONST, {.i64=EXHAUSTIVE}, INT_MIN, INT_MAX, FLAGS, "smode" }, { "less", "less exhaustive search", 0, AV_OPT_TYPE_CONST, {.i64=SMART_EXHAUSTIVE}, INT_MIN, INT_MAX, FLAGS, "smode" }, { "filename", "set motion search detailed log file name", OFFSET(filename), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, - { "opencl", "use OpenCL filtering capabilities", OFFSET(opencl), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, .flags = FLAGS }, + { "opencl", "ignored", OFFSET(opencl), AV_OPT_TYPE_BOOL, {.i64=0}, 0, 1, .flags = FLAGS }, { NULL } }; @@ -341,7 +340,6 @@ static int deshake_transform_c(AVFilterContext *ctx, static av_cold int init(AVFilterContext *ctx) { - int ret; DeshakeContext *deshake = ctx->priv; deshake->sad = av_pixelutils_get_sad_fn(4, 4, 1, deshake); // 16x16, 2nd source unaligned @@ -369,17 +367,7 @@ static av_cold int init(AVFilterContext *ctx) deshake->cx &= ~15; } deshake->transform = deshake_transform_c; - if (!CONFIG_OPENCL && deshake->opencl) { - av_log(ctx, AV_LOG_ERROR, "OpenCL support was not enabled in this build, cannot be selected\n"); - return AVERROR(EINVAL); - } - if (CONFIG_OPENCL && deshake->opencl) { - deshake->transform = ff_opencl_transform; - ret = ff_opencl_deshake_init(ctx); - if (ret < 0) - return ret; - } av_log(ctx, AV_LOG_VERBOSE, "cx: %d, cy: %d, cw: %d, ch: %d, rx: %d, ry: %d, edge: %d blocksize: %d contrast: %d search: %d\n", deshake->cx, deshake->cy, deshake->cw, deshake->ch, deshake->rx, deshake->ry, deshake->edge, deshake->blocksize * 2, deshake->contrast, deshake->search); @@ -416,9 +404,6 @@ static int config_props(AVFilterLink *link) static av_cold void uninit(AVFilterContext *ctx) { DeshakeContext *deshake = ctx->priv; - if (CONFIG_OPENCL && deshake->opencl) { - ff_opencl_deshake_uninit(ctx); - } av_frame_free(&deshake->ref); av_freep(&deshake->angles); deshake->angles_size = 0; @@ -447,12 +432,6 @@ static int filter_frame(AVFilterLink *link, AVFrame *in) } av_frame_copy_props(out, in); - if (CONFIG_OPENCL && deshake->opencl) { - ret = ff_opencl_deshake_process_inout_buf(link->dst,in, out); - if (ret < 0) - goto fail; - } - if (deshake->cx < 0 || deshake->cy < 0 || deshake->cw < 0 || deshake->ch < 0) { // Find the most likely global motion for the current frame find_motion(deshake, (deshake->ref == NULL) ? in->data[0] : deshake->ref->data[0], in->data[0], link->w, link->h, in->linesize[0], &t); diff --git a/libavfilter/vf_unsharp.c b/libavfilter/vf_unsharp.c index 438ff6d8fb..41ccc56942 100644 --- a/libavfilter/vf_unsharp.c +++ b/libavfilter/vf_unsharp.c @@ -46,7 +46,6 @@ #include "libavutil/opt.h" #include "libavutil/pixdesc.h" #include "unsharp.h" -#include "unsharp_opencl.h" static void apply_unsharp( uint8_t *dst, int dst_stride, const uint8_t *src, int src_stride, @@ -134,10 +133,8 @@ static void set_filter_param(UnsharpFilterParam *fp, int msize_x, int msize_y, f static av_cold int init(AVFilterContext *ctx) { - int ret = 0; UnsharpContext *s = ctx->priv; - set_filter_param(&s->luma, s->lmsize_x, s->lmsize_y, s->lamount); set_filter_param(&s->chroma, s->cmsize_x, s->cmsize_y, s->camount); @@ -146,16 +143,6 @@ static av_cold int init(AVFilterContext *ctx) return AVERROR(EINVAL); } s->apply_unsharp = apply_unsharp_c; - if (!CONFIG_OPENCL && s->opencl) { - av_log(ctx, AV_LOG_ERROR, "OpenCL support was not enabled in this build, cannot be selected\n"); - return AVERROR(EINVAL); - } - if (CONFIG_OPENCL && s->opencl) { - s->apply_unsharp = ff_opencl_apply_unsharp; - ret = ff_opencl_unsharp_init(ctx); - if (ret < 0) - return ret; - } return 0; } @@ -227,10 +214,6 @@ static av_cold void uninit(AVFilterContext *ctx) { UnsharpContext *s = ctx->priv; - if (CONFIG_OPENCL && s->opencl) { - ff_opencl_unsharp_uninit(ctx); - } - free_filter_param(&s->luma); free_filter_param(&s->chroma); } @@ -248,14 +231,9 @@ static int filter_frame(AVFilterLink *link, AVFrame *in) return AVERROR(ENOMEM); } av_frame_copy_props(out, in); - if (CONFIG_OPENCL && s->opencl) { - ret = ff_opencl_unsharp_process_inout_buf(link->dst, in, out); - if (ret < 0) - goto end; - } ret = s->apply_unsharp(link->dst, in, out); -end: + av_frame_free(&in); if (ret < 0) { @@ -282,7 +260,7 @@ static const AVOption unsharp_options[] = { { "cy", "set chroma matrix vertical size", OFFSET(cmsize_y), AV_OPT_TYPE_INT, { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS }, { "chroma_amount", "set chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, { "ca", "set chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, - { "opencl", "use OpenCL filtering capabilities", OFFSET(opencl), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS }, + { "opencl", "ignored", OFFSET(opencl), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS }, { NULL } }; |