Unified · Split
| @@ -6197,6 +6197,11 @@ Negative values will blur the input video, while positive values will · 6196 unmodified lines | |||
| 6197 | sharpen it, a value of zero will disable the effect. | 6197 | sharpen it, a value of zero will disable the effect. |
| 6198 | 6198 | ||
| 6199 | Default value is 0.0. | 6199 | Default value is 0.0. |
| 6200 | |||
| 6201 | @item opencl | ||
| 6202 | If set to 1, specify using OpenCL capabilities, only available if | ||
| 6203 | FFmpeg was configured with @code{--enable-opencl}. Default value is 0. | ||
| 6204 | |||
| 6200 | @end table | 6205 | @end table |
| 6201 | 6206 | ||
| 6202 | All parameters are optional and default to the equivalent of the | 6207 | All parameters are optional and default to the equivalent of the |
| @@ -149,7 +149,7 @@ OBJS-$(CONFIG_NOFORMAT_FILTER) += vf_format.o · 148 unmodified lines | |||
| 149 | OBJS-$(CONFIG_NOISE_FILTER) += vf_noise.o | 149 | OBJS-$(CONFIG_NOISE_FILTER) += vf_noise.o |
| 150 | OBJS-$(CONFIG_NULL_FILTER) += vf_null.o | 150 | OBJS-$(CONFIG_NULL_FILTER) += vf_null.o |
| 151 | OBJS-$(CONFIG_OCV_FILTER) += vf_libopencv.o | 151 | OBJS-$(CONFIG_OCV_FILTER) += vf_libopencv.o |
| 152 | OBJS-$(CONFIG_OPENCL) += deshake_opencl.o | 152 | OBJS-$(CONFIG_OPENCL) += deshake_opencl.o unsharp_opencl.o |
| 153 | OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o | 153 | OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o |
| 154 | OBJS-$(CONFIG_PAD_FILTER) += vf_pad.o | 154 | OBJS-$(CONFIG_PAD_FILTER) += vf_pad.o |
| 155 | OBJS-$(CONFIG_PERMS_FILTER) += f_perms.o | 155 | OBJS-$(CONFIG_PERMS_FILTER) += f_perms.o |
| @@ -22,6 +22,7 @@ · 21 unmodified lines | |||
| 22 | #if CONFIG_OPENCL | 22 | #if CONFIG_OPENCL |
| 23 | #include "libavutil/opencl.h" | 23 | #include "libavutil/opencl.h" |
| 24 | #include "deshake_kernel.h" | 24 | #include "deshake_kernel.h" |
| 25 | #include "unsharp_kernel.h" | ||
| 25 | #endif | 26 | #endif |
| 26 | 27 | ||
| 27 | #define OPENCL_REGISTER_KERNEL_CODE(X, x) \ | 28 | #define OPENCL_REGISTER_KERNEL_CODE(X, x) \ |
| @@ -35,5 +36,6 @@ void ff_opencl_register_filter_kernel_code_all(void) · 7 unmodified lines | |||
| 35 | { | 36 | { |
| 36 | #if CONFIG_OPENCL | 37 | #if CONFIG_OPENCL |
| 37 | OPENCL_REGISTER_KERNEL_CODE(DESHAKE, deshake); | 38 | OPENCL_REGISTER_KERNEL_CODE(DESHAKE, deshake); |
| 39 | OPENCL_REGISTER_KERNEL_CODE(UNSHARP, unsharp); | ||
| 38 | #endif | 40 | #endif |
| 39 | } | 41 | } |
| @@ -0,0 +1,78 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVFILTER_UNSHARP_H | ||
| 22 | #define AVFILTER_UNSHARP_H | ||
| 23 | |||
| 24 | #include "config.h" | ||
| 25 | #include "avfilter.h" | ||
| 26 | #if CONFIG_OPENCL | ||
| 27 | #include "libavutil/opencl.h" | ||
| 28 | #endif | ||
| 29 | |||
| 30 | #define MIN_MATRIX_SIZE 3 | ||
| 31 | #define MAX_MATRIX_SIZE 63 | ||
| 32 | |||
| 33 | /* right-shift and round-up */ | ||
| 34 | #define SHIFTUP(x,shift) (-((-(x))>>(shift))) | ||
| 35 | |||
| 36 | #if CONFIG_OPENCL | ||
| 37 | |||
| 38 | typedef struct { | ||
| 39 | cl_mem cl_luma_mask; | ||
| 40 | cl_mem cl_chroma_mask; | ||
| 41 | int in_plane_size[8]; | ||
| 42 | int out_plane_size[8]; | ||
| 43 | int plane_num; | ||
| 44 | cl_mem cl_inbuf; | ||
| 45 | size_t cl_inbuf_size; | ||
| 46 | cl_mem cl_outbuf; | ||
| 47 | size_t cl_outbuf_size; | ||
| 48 | AVOpenCLKernelEnv kernel_env; | ||
| 49 | } UnsharpOpenclContext; | ||
| 50 | |||
| 51 | #endif | ||
| 52 | |||
| 53 | typedef struct UnsharpFilterParam { | ||
| 54 | int msize_x; ///< matrix width | ||
| 55 | int msize_y; ///< matrix height | ||
| 56 | int amount; ///< effect amount | ||
| 57 | int steps_x; ///< horizontal step count | ||
| 58 | int steps_y; ///< vertical step count | ||
| 59 | int scalebits; ///< bits to shift pixel | ||
| 60 | int32_t halfscale; ///< amount to add to pixel | ||
| 61 | uint32_t *sc[MAX_MATRIX_SIZE - 1]; ///< finite state machine storage | ||
| 62 | } UnsharpFilterParam; | ||
| 63 | |||
| 64 | typedef struct { | ||
| 65 | const AVClass *class; | ||
| 66 | int lmsize_x, lmsize_y, cmsize_x, cmsize_y; | ||
| 67 | float lamount, camount; | ||
| 68 | UnsharpFilterParam luma; ///< luma parameters (width, height, amount) | ||
| 69 | UnsharpFilterParam chroma; ///< chroma parameters (width, height, amount) | ||
| 70 | int hsub, vsub; | ||
| 71 | int opencl; | ||
| 72 | #if CONFIG_OPENCL | ||
| 73 | UnsharpOpenclContext opencl_ctx; | ||
| 74 | #endif | ||
| 75 | int (* apply_unsharp)(AVFilterContext *ctx, AVFrame *in, AVFrame *out); | ||
| 76 | } UnsharpContext; | ||
| 77 | |||
| 78 | #endif /* AVFILTER_UNSHARP_H */ | ||
| @@ -0,0 +1,137 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVFILTER_UNSHARP_KERNEL_H | ||
| 22 | #define AVFILTER_UNSHARP_KERNEL_H | ||
| 23 | |||
| 24 | #include "libavutil/opencl.h" | ||
| 25 | |||
| 26 | const char *ff_kernel_unsharp_opencl = AV_OPENCL_KERNEL( | ||
| 27 | inline unsigned char clip_uint8(int a) | ||
| 28 | { | ||
| 29 | if (a & (~0xFF)) | ||
| 30 | return (-a)>>31; | ||
| 31 | else | ||
| 32 | return a; | ||
| 33 | } | ||
| 34 | |||
| 35 | kernel void unsharp(global unsigned char *src, | ||
| 36 | global unsigned char *dst, | ||
| 37 | const global unsigned int *mask_lu, | ||
| 38 | const global unsigned int *mask_ch, | ||
| 39 | int amount_lu, | ||
| 40 | int amount_ch, | ||
| 41 | int step_x_lu, | ||
| 42 | int step_y_lu, | ||
| 43 | int step_x_ch, | ||
| 44 | int step_y_ch, | ||
| 45 | int scalebits_lu, | ||
| 46 | int scalebits_ch, | ||
| 47 | int halfscale_lu, | ||
| 48 | int halfscale_ch, | ||
| 49 | int src_stride_lu, | ||
| 50 | int src_stride_ch, | ||
| 51 | int dst_stride_lu, | ||
| 52 | int dst_stride_ch, | ||
| 53 | int height, | ||
| 54 | int width, | ||
| 55 | int ch, | ||
| 56 | int cw) | ||
| 57 | { | ||
| 58 | global unsigned char *dst_y = dst; | ||
| 59 | global unsigned char *dst_u = dst_y + height * dst_stride_lu; | ||
| 60 | global unsigned char *dst_v = dst_u + ch * dst_stride_ch; | ||
| 61 | |||
| 62 | global unsigned char *src_y = src; | ||
| 63 | global unsigned char *src_u = src_y + height * src_stride_lu; | ||
| 64 | global unsigned char *src_v = src_u + ch * src_stride_ch; | ||
| 65 | |||
| 66 | global unsigned char *temp_dst; | ||
| 67 | global unsigned char *temp_src; | ||
| 68 | const global unsigned int *temp_mask; | ||
| 69 | int global_id = get_global_id(0); | ||
| 70 | int i, j, x, y, temp_src_stride, temp_dst_stride, temp_height, temp_width, temp_steps_x, temp_steps_y, | ||
| 71 | temp_amount, temp_scalebits, temp_halfscale, sum, idx_x, idx_y, temp, res; | ||
| 72 | if (global_id < width * height) { | ||
| 73 | y = global_id / width; | ||
| 74 | x = global_id % width; | ||
| 75 | temp_dst = dst_y; | ||
| 76 | temp_src = src_y; | ||
| 77 | temp_src_stride = src_stride_lu; | ||
| 78 | temp_dst_stride = dst_stride_lu; | ||
| 79 | temp_height = height; | ||
| 80 | temp_width = width; | ||
| 81 | temp_steps_x = step_x_lu; | ||
| 82 | temp_steps_y = step_y_lu; | ||
| 83 | temp_mask = mask_lu; | ||
| 84 | temp_amount = amount_lu; | ||
| 85 | temp_scalebits = scalebits_lu; | ||
| 86 | temp_halfscale = halfscale_lu; | ||
| 87 | } else if ((global_id >= width * height) && (global_id < width * height + ch * cw)) { | ||
| 88 | y = (global_id - width * height) / cw; | ||
| 89 | x = (global_id - width * height) % cw; | ||
| 90 | temp_dst = dst_u; | ||
| 91 | temp_src = src_u; | ||
| 92 | temp_src_stride = src_stride_ch; | ||
| 93 | temp_dst_stride = dst_stride_ch; | ||
| 94 | temp_height = ch; | ||
| 95 | temp_width = cw; | ||
| 96 | temp_steps_x = step_x_ch; | ||
| 97 | temp_steps_y = step_y_ch; | ||
| 98 | temp_mask = mask_ch; | ||
| 99 | temp_amount = amount_ch; | ||
| 100 | temp_scalebits = scalebits_ch; | ||
| 101 | temp_halfscale = halfscale_ch; | ||
| 102 | } else { | ||
| 103 | y = (global_id - width * height - ch * cw) / cw; | ||
| 104 | x = (global_id - width * height - ch * cw) % cw; | ||
| 105 | temp_dst = dst_v; | ||
| 106 | temp_src = src_v; | ||
| 107 | temp_src_stride = src_stride_ch; | ||
| 108 | temp_dst_stride = dst_stride_ch; | ||
| 109 | temp_height = ch; | ||
| 110 | temp_width = cw; | ||
| 111 | temp_steps_x = step_x_ch; | ||
| 112 | temp_steps_y = step_y_ch; | ||
| 113 | temp_mask = mask_ch; | ||
| 114 | temp_amount = amount_ch; | ||
| 115 | temp_scalebits = scalebits_ch; | ||
| 116 | temp_halfscale = halfscale_ch; | ||
| 117 | } | ||
| 118 | if (temp_amount) { | ||
| 119 | sum = 0; | ||
| 120 | for (j = 0; j <= 2 * temp_steps_y; j++) { | ||
| 121 | idx_y = (y - temp_steps_y + j) <= 0 ? 0 : (y - temp_steps_y + j) >= temp_height ? temp_height-1 : y - temp_steps_y + j; | ||
| 122 | for (i = 0; i <= 2 * temp_steps_x; i++) { | ||
| 123 | idx_x = (x - temp_steps_x + i) <= 0 ? 0 : (x - temp_steps_x + i) >= temp_width ? temp_width-1 : x - temp_steps_x + i; | ||
| 124 | sum += temp_mask[i + j * (2 * temp_steps_x + 1)] * temp_src[idx_x + idx_y * temp_src_stride]; | ||
| 125 | } | ||
| 126 | } | ||
| 127 | temp = (int)temp_src[x + y * temp_src_stride]; | ||
| 128 | res = temp + (((temp - (int)((sum + temp_halfscale) >> temp_scalebits)) * temp_amount) >> 16); | ||
| 129 | temp_dst[x + y * temp_dst_stride] = clip_uint8(res); | ||
| 130 | } else { | ||
| 131 | temp_dst[x + y * temp_dst_stride] = temp_src[x + y * temp_src_stride]; | ||
| 132 | } | ||
| 133 | } | ||
| 134 | |||
| 135 | ); | ||
| 136 | |||
| 137 | #endif /* AVFILTER_UNSHARP_KERNEL_H */ | ||
| @@ -0,0 +1,282 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | /** | ||
| 22 | * @file | ||
| 23 | * unsharp input video | ||
| 24 | */ | ||
| 25 | |||
| 26 | #include "unsharp_opencl.h" | ||
| 27 | #include "libavutil/opencl_internal.h" | ||
| 28 | |||
| 29 | #define PLANE_NUM 3 | ||
| 30 | |||
| 31 | static inline void add_mask_counter(uint32_t *dst, uint32_t *counter1, uint32_t *counter2, int len) | ||
| 32 | { | ||
| 33 | int i; | ||
| 34 | for (i = 0; i < len; i++) { | ||
| 35 | dst[i] = counter1[i] + counter2[i]; | ||
| 36 | } | ||
| 37 | } | ||
| 38 | |||
| 39 | static int compute_mask(int step, uint32_t *mask) | ||
| 40 | { | ||
| 41 | int i, z, ret = 0; | ||
| 42 | int counter_size = sizeof(uint32_t) * (2 * step + 1); | ||
| 43 | uint32_t *temp1_counter, *temp2_counter, **counter; | ||
| 44 | temp1_counter = av_mallocz(counter_size); | ||
| 45 | if (!temp1_counter) { | ||
| 46 | ret = AVERROR(ENOMEM); | ||
| 47 | goto end; | ||
| 48 | } | ||
| 49 | temp2_counter = av_mallocz(counter_size); | ||
| 50 | if (!temp2_counter) { | ||
| 51 | ret = AVERROR(ENOMEM); | ||
| 52 | goto end; | ||
| 53 | } | ||
| 54 | counter = av_mallocz(counter_size); | ||
| 55 | if (!counter) { | ||
| 56 | ret = AVERROR(ENOMEM); | ||
| 57 | goto end; | ||
| 58 | } | ||
| 59 | for (i = 0; i < 2 * step + 1; i++) { | ||
| 60 | counter[i] = av_mallocz(counter_size); | ||
| 61 | if (!counter[i]) { | ||
| 62 | ret = AVERROR(ENOMEM); | ||
| 63 | goto end; | ||
| 64 | } | ||
| 65 | } | ||
| 66 | for (i = 0; i < 2 * step + 1; i++) { | ||
| 67 | memset(temp1_counter, 0, counter_size); | ||
| 68 | temp1_counter[i] = 1; | ||
| 69 | for (z = 0; z < step * 2; z += 2) { | ||
| 70 | add_mask_counter(temp2_counter, counter[z], temp1_counter, step * 2); | ||
| 71 | memcpy(counter[z], temp1_counter, counter_size); | ||
| 72 | add_mask_counter(temp1_counter, counter[z + 1], temp2_counter, step * 2); | ||
| 73 | memcpy(counter[z + 1], temp2_counter, counter_size); | ||
| 74 | } | ||
| 75 | } | ||
| 76 | memcpy(mask, temp1_counter, counter_size); | ||
| 77 | end: | ||
| 78 | av_freep(&temp1_counter); | ||
| 79 | av_freep(&temp2_counter); | ||
| 80 | for (i = 0; i < 2 * step + 1; i++) { | ||
| 81 | av_freep(&counter[i]); | ||
| 82 | } | ||
| 83 | av_freep(&counter); | ||
| 84 | return ret; | ||
| 85 | } | ||
| 86 | |||
| 87 | static int compute_mask_matrix(cl_mem cl_mask_matrix, int step_x, int step_y) | ||
| 88 | { | ||
| 89 | int i, j, ret = 0; | ||
| 90 | uint32_t *mask_matrix, *mask_x, *mask_y; | ||
| 91 | size_t size_matrix = sizeof(uint32_t) * (2 * step_x + 1) * (2 * step_y + 1); | ||
| 92 | mask_x = av_mallocz(sizeof(uint32_t) * (2 * step_x + 1)); | ||
| 93 | if (!mask_x) { | ||
| 94 | ret = AVERROR(ENOMEM); | ||
| 95 | goto end; | ||
| 96 | } | ||
| 97 | mask_y = av_mallocz(sizeof(uint32_t) * (2 * step_y + 1)); | ||
| 98 | if (!mask_y) { | ||
| 99 | ret = AVERROR(ENOMEM); | ||
| 100 | goto end; | ||
| 101 | } | ||
| 102 | mask_matrix = av_mallocz(size_matrix); | ||
| 103 | if (!mask_matrix) { | ||
| 104 | ret = AVERROR(ENOMEM); | ||
| 105 | goto end; | ||
| 106 | } | ||
| 107 | ret = compute_mask(step_x, mask_x); | ||
| 108 | if (ret < 0) | ||
| 109 | goto end; | ||
| 110 | ret = compute_mask(step_y, mask_y); | ||
| 111 | if (ret < 0) | ||
| 112 | goto end; | ||
| 113 | for (j = 0; j < 2 * step_y + 1; j++) { | ||
| 114 | for (i = 0; i < 2 * step_x + 1; i++) { | ||
| 115 | mask_matrix[i + j * (2 * step_x + 1)] = mask_y[j] * mask_x[i]; | ||
| 116 | } | ||
| 117 | } | ||
| 118 | ret = av_opencl_buffer_write(cl_mask_matrix, (uint8_t *)mask_matrix, size_matrix); | ||
| 119 | end: | ||
| 120 | av_freep(&mask_x); | ||
| 121 | av_freep(&mask_y); | ||
| 122 | av_freep(&mask_matrix); | ||
| 123 | return ret; | ||
| 124 | } | ||
| 125 | |||
| 126 | static int generate_mask(AVFilterContext *ctx) | ||
| 127 | { | ||
| 128 | UnsharpContext *unsharp = ctx->priv; | ||
| 129 | int i, ret = 0, step_x[2], step_y[2]; | ||
| 130 | cl_mem mask_matrix[2]; | ||
| 131 | mask_matrix[0] = unsharp->opencl_ctx.cl_luma_mask; | ||
| 132 | mask_matrix[1] = unsharp->opencl_ctx.cl_chroma_mask; | ||
| 133 | step_x[0] = unsharp->luma.steps_x; | ||
| 134 | step_x[1] = unsharp->chroma.steps_x; | ||
| 135 | step_y[0] = unsharp->luma.steps_y; | ||
| 136 | step_y[1] = unsharp->chroma.steps_y; | ||
| 137 | if (!mask_matrix[0] || !mask_matrix[1]) { | ||
| 138 | av_log(ctx, AV_LOG_ERROR, "Luma mask and chroma mask should not be NULL\n"); | ||
| 139 | return AVERROR(EINVAL); | ||
| 140 | } | ||
| 141 | for (i = 0; i < 2; i++) { | ||
| 142 | ret = compute_mask_matrix(mask_matrix[i], step_x[i], step_y[i]); | ||
| 143 | if (ret < 0) | ||
| 144 | return ret; | ||
| 145 | } | ||
| 146 | return ret; | ||
| 147 | } | ||
| 148 | |||
| 149 | int ff_opencl_apply_unsharp(AVFilterContext *ctx, AVFrame *in, AVFrame *out) | ||
| 150 | { | ||
| 151 | int ret; | ||
| 152 | AVFilterLink *link = ctx->inputs[0]; | ||
| 153 | UnsharpContext *unsharp = ctx->priv; | ||
| 154 | cl_int status; | ||
| 155 | int cw = SHIFTUP(link->w, unsharp->hsub); | ||
| 156 | int ch = SHIFTUP(link->h, unsharp->vsub); | ||
| 157 | const size_t global_work_size = link->w * link->h + 2 * ch * cw; | ||
| 158 | FFOpenclParam opencl_param = {0}; | ||
| 159 | |||
| 160 | opencl_param.ctx = ctx; | ||
| 161 | opencl_param.kernel = unsharp->opencl_ctx.kernel_env.kernel; | ||
| 162 | ret = ff_opencl_set_parameter(&opencl_param, | ||
| 163 | FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_inbuf), | ||
| 164 | FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_outbuf), | ||
| 165 | FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_luma_mask), | ||
| 166 | FF_OPENCL_PARAM_INFO(unsharp->opencl_ctx.cl_chroma_mask), | ||
| 167 | FF_OPENCL_PARAM_INFO(unsharp->luma.amount), | ||
| 168 | FF_OPENCL_PARAM_INFO(unsharp->chroma.amount), | ||
| 169 | FF_OPENCL_PARAM_INFO(unsharp->luma.steps_x), | ||
| 170 | FF_OPENCL_PARAM_INFO(unsharp->luma.steps_y), | ||
| 171 | FF_OPENCL_PARAM_INFO(unsharp->chroma.steps_x), | ||
| 172 | FF_OPENCL_PARAM_INFO(unsharp->chroma.steps_y), | ||
| 173 | FF_OPENCL_PARAM_INFO(unsharp->luma.scalebits), | ||
| 174 | FF_OPENCL_PARAM_INFO(unsharp->chroma.scalebits), | ||
| 175 | FF_OPENCL_PARAM_INFO(unsharp->luma.halfscale), | ||
| 176 | FF_OPENCL_PARAM_INFO(unsharp->chroma.halfscale), | ||
| 177 | FF_OPENCL_PARAM_INFO(in->linesize[0]), | ||
| 178 | FF_OPENCL_PARAM_INFO(in->linesize[1]), | ||
| 179 | FF_OPENCL_PARAM_INFO(out->linesize[0]), | ||
| 180 | FF_OPENCL_PARAM_INFO(out->linesize[1]), | ||
| 181 | FF_OPENCL_PARAM_INFO(link->h), | ||
| 182 | FF_OPENCL_PARAM_INFO(link->w), | ||
| 183 | FF_OPENCL_PARAM_INFO(ch), | ||
| 184 | FF_OPENCL_PARAM_INFO(cw), | ||
| 185 | NULL); | ||
| 186 | if (ret < 0) | ||
| 187 | return ret; | ||
| 188 | status = clEnqueueNDRangeKernel(unsharp->opencl_ctx.kernel_env.command_queue, | ||
| 189 | unsharp->opencl_ctx.kernel_env.kernel, 1, NULL, | ||
| 190 | &global_work_size, NULL, 0, NULL, NULL); | ||
| 191 | if (status != CL_SUCCESS) { | ||
| 192 | av_log(ctx, AV_LOG_ERROR, "OpenCL run kernel error occurred: %d\n", status); | ||
| 193 | return AVERROR_EXTERNAL; | ||
| 194 | } | ||
| 195 | clFinish(unsharp->opencl_ctx.kernel_env.command_queue); | ||
| 196 | return av_opencl_buffer_read_image(out->data, unsharp->opencl_ctx.out_plane_size, | ||
| 197 | unsharp->opencl_ctx.plane_num, unsharp->opencl_ctx.cl_outbuf, | ||
| 198 | unsharp->opencl_ctx.cl_outbuf_size); | ||
| 199 | } | ||
| 200 | |||
| 201 | int ff_opencl_unsharp_init(AVFilterContext *ctx) | ||
| 202 | { | ||
| 203 | int ret = 0; | ||
| 204 | UnsharpContext *unsharp = ctx->priv; | ||
| 205 | ret = av_opencl_init(NULL); | ||
| 206 | if (ret < 0) | ||
| 207 | return ret; | ||
| 208 | ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_luma_mask, | ||
| 209 | sizeof(uint32_t) * (2 * unsharp->luma.steps_x + 1) * (2 * unsharp->luma.steps_y + 1), | ||
| 210 | CL_MEM_READ_ONLY, NULL); | ||
| 211 | if (ret < 0) | ||
| 212 | return ret; | ||
| 213 | ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_chroma_mask, | ||
| 214 | sizeof(uint32_t) * (2 * unsharp->chroma.steps_x + 1) * (2 * unsharp->chroma.steps_y + 1), | ||
| 215 | CL_MEM_READ_ONLY, NULL); | ||
| 216 | if (ret < 0) | ||
| 217 | return ret; | ||
| 218 | ret = generate_mask(ctx); | ||
| 219 | if (ret < 0) | ||
| 220 | return ret; | ||
| 221 | unsharp->opencl_ctx.plane_num = PLANE_NUM; | ||
| 222 | if (!unsharp->opencl_ctx.kernel_env.kernel) { | ||
| 223 | ret = av_opencl_create_kernel(&unsharp->opencl_ctx.kernel_env, "unsharp"); | ||
| 224 | if (ret < 0) { | ||
| 225 | av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel with name 'unsharp'\n"); | ||
| 226 | return ret; | ||
| 227 | } | ||
| 228 | } | ||
| 229 | return ret; | ||
| 230 | } | ||
| 231 | |||
| 232 | void ff_opencl_unsharp_uninit(AVFilterContext *ctx) | ||
| 233 | { | ||
| 234 | UnsharpContext *unsharp = ctx->priv; | ||
| 235 | av_opencl_buffer_release(&unsharp->opencl_ctx.cl_inbuf); | ||
| 236 | av_opencl_buffer_release(&unsharp->opencl_ctx.cl_outbuf); | ||
| 237 | av_opencl_buffer_release(&unsharp->opencl_ctx.cl_luma_mask); | ||
| 238 | av_opencl_buffer_release(&unsharp->opencl_ctx.cl_chroma_mask); | ||
| 239 | av_opencl_release_kernel(&unsharp->opencl_ctx.kernel_env); | ||
| 240 | av_opencl_uninit(); | ||
| 241 | } | ||
| 242 | |||
| 243 | int ff_opencl_unsharp_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out) | ||
| 244 | { | ||
| 245 | int ret = 0; | ||
| 246 | AVFilterLink *link = ctx->inputs[0]; | ||
| 247 | UnsharpContext *unsharp = ctx->priv; | ||
| 248 | int ch = SHIFTUP(link->h, unsharp->vsub); | ||
| 249 | |||
| 250 | if ((!unsharp->opencl_ctx.cl_inbuf) || (!unsharp->opencl_ctx.cl_outbuf)) { | ||
| 251 | unsharp->opencl_ctx.in_plane_size[0] = (in->linesize[0] * in->height); | ||
| 252 | unsharp->opencl_ctx.in_plane_size[1] = (in->linesize[1] * ch); | ||
| 253 | unsharp->opencl_ctx.in_plane_size[2] = (in->linesize[2] * ch); | ||
| 254 | unsharp->opencl_ctx.out_plane_size[0] = (out->linesize[0] * out->height); | ||
| 255 | unsharp->opencl_ctx.out_plane_size[1] = (out->linesize[1] * ch); | ||
| 256 | unsharp->opencl_ctx.out_plane_size[2] = (out->linesize[2] * ch); | ||
| 257 | unsharp->opencl_ctx.cl_inbuf_size = unsharp->opencl_ctx.in_plane_size[0] + | ||
| 258 | unsharp->opencl_ctx.in_plane_size[1] + | ||
| 259 | unsharp->opencl_ctx.in_plane_size[2]; | ||
| 260 | unsharp->opencl_ctx.cl_outbuf_size = unsharp->opencl_ctx.out_plane_size[0] + | ||
| 261 | unsharp->opencl_ctx.out_plane_size[1] + | ||
| 262 | unsharp->opencl_ctx.out_plane_size[2]; | ||
| 263 | if (!unsharp->opencl_ctx.cl_inbuf) { | ||
| 264 | ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_inbuf, | ||
| 265 | unsharp->opencl_ctx.cl_inbuf_size, | ||
| 266 | CL_MEM_READ_ONLY, NULL); | ||
| 267 | if (ret < 0) | ||
| 268 | return ret; | ||
| 269 | } | ||
| 270 | if (!unsharp->opencl_ctx.cl_outbuf) { | ||
| 271 | ret = av_opencl_buffer_create(&unsharp->opencl_ctx.cl_outbuf, | ||
| 272 | unsharp->opencl_ctx.cl_outbuf_size, | ||
| 273 | CL_MEM_READ_WRITE, NULL); | ||
| 274 | if (ret < 0) | ||
| 275 | return ret; | ||
| 276 | } | ||
| 277 | } | ||
| 278 | return av_opencl_buffer_write_image(unsharp->opencl_ctx.cl_inbuf, | ||
| 279 | unsharp->opencl_ctx.cl_inbuf_size, | ||
| 280 | 0, in->data, unsharp->opencl_ctx.in_plane_size, | ||
| 281 | unsharp->opencl_ctx.plane_num); | ||
| 282 | } | ||
| @@ -0,0 +1,34 @@ | |||
| 1 | /* | ||
| 2 | * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVFILTER_UNSHARP_OPENCL_H | ||
| 22 | #define AVFILTER_UNSHARP_OPENCL_H | ||
| 23 | |||
| 24 | #include "unsharp.h" | ||
| 25 | |||
| 26 | int ff_opencl_unsharp_init(AVFilterContext *ctx); | ||
| 27 | |||
| 28 | void ff_opencl_unsharp_uninit(AVFilterContext *ctx); | ||
| 29 | |||
| 30 | int ff_opencl_unsharp_process_inout_buf(AVFilterContext *ctx, AVFrame *in, AVFrame *out); | ||
| 31 | |||
| 32 | int ff_opencl_apply_unsharp(AVFilterContext *ctx, AVFrame *in, AVFrame *out); | ||
| 33 | |||
| 34 | #endif /* AVFILTER_UNSHARP_OPENCL_H */ | ||
| @@ -30,7 +30,7 @@ · 29 unmodified lines | |||
| 30 | 30 | ||
| 31 | #define LIBAVFILTER_VERSION_MAJOR 3 | 31 | #define LIBAVFILTER_VERSION_MAJOR 3 |
| 32 | #define LIBAVFILTER_VERSION_MINOR 60 | 32 | #define LIBAVFILTER_VERSION_MINOR 60 |
| 33 | #define LIBAVFILTER_VERSION_MICRO 101 | 33 | #define LIBAVFILTER_VERSION_MICRO 102 |
| 34 | 34 | ||
| 35 | #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \ | 35 | #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \ |
| 36 | LIBAVFILTER_VERSION_MINOR, \ | 36 | LIBAVFILTER_VERSION_MINOR, \ |
| @@ -46,36 +46,12 @@ · 45 unmodified lines | |||
| 46 | #include "libavutil/mem.h" | 46 | #include "libavutil/mem.h" |
| 47 | #include "libavutil/opt.h" | 47 | #include "libavutil/opt.h" |
| 48 | #include "libavutil/pixdesc.h" | 48 | #include "libavutil/pixdesc.h" |
| 49 | 49 | #include "unsharp.h" | |
| 50 | #define MIN_MATRIX_SIZE 3 | 50 | #include "unsharp_opencl.h" |
| 51 | #define MAX_MATRIX_SIZE 63 | ||
| 52 | |||
| 53 | /* right-shift and round-up */ | ||
| 54 | #define SHIFTUP(x,shift) (-((-(x))>>(shift))) | ||
| 55 | |||
| 56 | typedef struct FilterParam { | ||
| 57 | int msize_x; ///< matrix width | ||
| 58 | int msize_y; ///< matrix height | ||
| 59 | int amount; ///< effect amount | ||
| 60 | int steps_x; ///< horizontal step count | ||
| 61 | int steps_y; ///< vertical step count | ||
| 62 | int scalebits; ///< bits to shift pixel | ||
| 63 | int32_t halfscale; ///< amount to add to pixel | ||
| 64 | uint32_t *sc[MAX_MATRIX_SIZE - 1]; ///< finite state machine storage | ||
| 65 | } FilterParam; | ||
| 66 | |||
| 67 | typedef struct { | ||
| 68 | const AVClass *class; | ||
| 69 | int lmsize_x, lmsize_y, cmsize_x, cmsize_y; | ||
| 70 | float lamount, camount; | ||
| 71 | FilterParam luma; ///< luma parameters (width, height, amount) | ||
| 72 | FilterParam chroma; ///< chroma parameters (width, height, amount) | ||
| 73 | int hsub, vsub; | ||
| 74 | } UnsharpContext; | ||
| 75 | 51 | ||
| 76 | static void apply_unsharp( uint8_t *dst, int dst_stride, | 52 | static void apply_unsharp( uint8_t *dst, int dst_stride, |
| 77 | const uint8_t *src, int src_stride, | 53 | const uint8_t *src, int src_stride, |
| 78 | int width, int height, FilterParam *fp) | 54 | int width, int height, UnsharpFilterParam *fp) |
| 79 | { | 55 | { |
| 80 | uint32_t **sc = fp->sc; | 56 | uint32_t **sc = fp->sc; |
| 81 | uint32_t sr[MAX_MATRIX_SIZE - 1], tmp1, tmp2; | 57 | uint32_t sr[MAX_MATRIX_SIZE - 1], tmp1, tmp2; |
| @@ -131,7 +107,25 @@ static void apply_unsharp( uint8_t *dst, int dst_stride, · 49 unmodified lines | |||
| 131 | } | 107 | } |
| 132 | } | 108 | } |
| 133 | 109 | ||
| 134 | static void set_filter_param(FilterParam *fp, int msize_x, int msize_y, float amount) | 110 | static int apply_unsharp_c(AVFilterContext *ctx, AVFrame *in, AVFrame *out) |
| 111 | { | ||
| 112 | AVFilterLink *inlink = ctx->inputs[0]; | ||
| 113 | UnsharpContext *unsharp = ctx->priv; | ||
| 114 | int i, plane_w[3], plane_h[3]; | ||
| 115 | UnsharpFilterParam *fp[3]; | ||
| 116 | plane_w[0] = inlink->w; | ||
| 117 | plane_w[1] = plane_w[2] = SHIFTUP(inlink->w, unsharp->hsub); | ||
| 118 | plane_h[0] = inlink->h; | ||
| 119 | plane_h[1] = plane_h[2] = SHIFTUP(inlink->h, unsharp->vsub); | ||
| 120 | fp[0] = &unsharp->luma; | ||
| 121 | fp[1] = fp[2] = &unsharp->chroma; | ||
| 122 | for (i = 0; i < 3; i++) { | ||
| 123 | apply_unsharp(out->data[i], out->linesize[i], in->data[i], in->linesize[i], plane_w[i], plane_h[i], fp[i]); | ||
| 124 | } | ||
| 125 | return 0; | ||
| 126 | } | ||
| 127 | |||
| 128 | static void set_filter_param(UnsharpFilterParam *fp, int msize_x, int msize_y, float amount) | ||
| 135 | { | 129 | { |
| 136 | fp->msize_x = msize_x; | 130 | fp->msize_x = msize_x; |
| 137 | fp->msize_y = msize_y; | 131 | fp->msize_y = msize_y; |
| @@ -145,12 +139,24 @@ static void set_filter_param(FilterParam *fp, int msize_x, int msize_y, float am · 7 unmodified lines | |||
| 145 | 139 | ||
| 146 | static av_cold int init(AVFilterContext *ctx) | 140 | static av_cold int init(AVFilterContext *ctx) |
| 147 | { | 141 | { |
| 142 | int ret = 0; | ||
| 148 | UnsharpContext *unsharp = ctx->priv; | 143 | UnsharpContext *unsharp = ctx->priv; |
| 149 | 144 | ||
| 150 | 145 | ||
| 151 | set_filter_param(&unsharp->luma, unsharp->lmsize_x, unsharp->lmsize_y, unsharp->lamount); | 146 | set_filter_param(&unsharp->luma, unsharp->lmsize_x, unsharp->lmsize_y, unsharp->lamount); |
| 152 | set_filter_param(&unsharp->chroma, unsharp->cmsize_x, unsharp->cmsize_y, unsharp->camount); | 147 | set_filter_param(&unsharp->chroma, unsharp->cmsize_x, unsharp->cmsize_y, unsharp->camount); |
| 153 | 148 | ||
| 149 | unsharp->apply_unsharp = apply_unsharp_c; | ||
| 150 | if (!CONFIG_OPENCL && unsharp->opencl) { | ||
| 151 | av_log(ctx, AV_LOG_ERROR, "OpenCL support was not enabled in this build, cannot be selected\n"); | ||
| 152 | return AVERROR(EINVAL); | ||
| 153 | } | ||
| 154 | if (CONFIG_OPENCL && unsharp->opencl) { | ||
| 155 | unsharp->apply_unsharp = ff_opencl_apply_unsharp; | ||
| 156 | ret = ff_opencl_unsharp_init(ctx); | ||
| 157 | if (ret < 0) | ||
| 158 | return ret; | ||
| 159 | } | ||
| 154 | return 0; | 160 | return 0; |
| 155 | } | 161 | } |
| 156 | 162 | ||
| @@ -167,7 +173,7 @@ static int query_formats(AVFilterContext *ctx) · 10 unmodified lines | |||
| 167 | return 0; | 173 | return 0; |
| 168 | } | 174 | } |
| 169 | 175 | ||
| 170 | static int init_filter_param(AVFilterContext *ctx, FilterParam *fp, const char *effect_type, int width) | 176 | static int init_filter_param(AVFilterContext *ctx, UnsharpFilterParam *fp, const char *effect_type, int width) |
| 171 | { | 177 | { |
| 172 | int z; | 178 | int z; |
| 173 | const char *effect = fp->amount == 0 ? "none" : fp->amount < 0 ? "blur" : "sharpen"; | 179 | const char *effect = fp->amount == 0 ? "none" : fp->amount < 0 ? "blur" : "sharpen"; |
| @@ -208,7 +214,7 @@ static int config_props(AVFilterLink *link) · 34 unmodified lines | |||
| 208 | return 0; | 214 | return 0; |
| 209 | } | 215 | } |
| 210 | 216 | ||
| 211 | static void free_filter_param(FilterParam *fp) | 217 | static void free_filter_param(UnsharpFilterParam *fp) |
| 212 | { | 218 | { |
| 213 | int z; | 219 | int z; |
| 214 | 220 | ||
| @@ -220,6 +226,10 @@ static av_cold void uninit(AVFilterContext *ctx) · 5 unmodified lines | |||
| 220 | { | 226 | { |
| 221 | UnsharpContext *unsharp = ctx->priv; | 227 | UnsharpContext *unsharp = ctx->priv; |
| 222 | 228 | ||
| 229 | if (CONFIG_OPENCL && unsharp->opencl) { | ||
| 230 | ff_opencl_unsharp_uninit(ctx); | ||
| 231 | } | ||
| 232 | |||
| 223 | free_filter_param(&unsharp->luma); | 233 | free_filter_param(&unsharp->luma); |
| 224 | free_filter_param(&unsharp->chroma); | 234 | free_filter_param(&unsharp->chroma); |
| 225 | } | 235 | } |
| @@ -229,8 +239,7 @@ static int filter_frame(AVFilterLink *link, AVFrame *in) · 3 unmodified lines | |||
| 229 | UnsharpContext *unsharp = link->dst->priv; | 239 | UnsharpContext *unsharp = link->dst->priv; |
| 230 | AVFilterLink *outlink = link->dst->outputs[0]; | 240 | AVFilterLink *outlink = link->dst->outputs[0]; |
| 231 | AVFrame *out; | 241 | AVFrame *out; |
| 232 | int cw = SHIFTUP(link->w, unsharp->hsub); | 242 | int ret = 0; |
| 233 | int ch = SHIFTUP(link->h, unsharp->vsub); | ||
| 234 | 243 | ||
| 235 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); | 244 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
| 236 | if (!out) { | 245 | if (!out) { |
| @@ -238,12 +247,18 @@ static int filter_frame(AVFilterLink *link, AVFrame *in) · 1 unmodified line | |||
| 238 | return AVERROR(ENOMEM); | 247 | return AVERROR(ENOMEM); |
| 239 | } | 248 | } |
| 240 | av_frame_copy_props(out, in); | 249 | av_frame_copy_props(out, in); |
| 250 | if (CONFIG_OPENCL && unsharp->opencl) { | ||
| 251 | ret = ff_opencl_unsharp_process_inout_buf(link->dst, in, out); | ||
| 252 | if (ret < 0) | ||
| 253 | goto end; | ||
| 254 | } | ||
| 241 | 255 | ||
| 242 | apply_unsharp(out->data[0], out->linesize[0], in->data[0], in->linesize[0], link->w, link->h, &unsharp->luma); | 256 | ret = unsharp->apply_unsharp(link->dst, in, out); |
| 243 | apply_unsharp(out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw, ch, &unsharp->chroma); | 257 | end: |
| 244 | apply_unsharp(out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw, ch, &unsharp->chroma); | ||
| 245 | |||
| 246 | av_frame_free(&in); | 258 | av_frame_free(&in); |
| 259 | |||
| 260 | if (ret < 0) | ||
| 261 | return ret; | ||
| 247 | return ff_filter_frame(outlink, out); | 262 | return ff_filter_frame(outlink, out); |
| 248 | } | 263 | } |
| 249 | 264 | ||
| @@ -264,6 +279,7 @@ static const AVOption unsharp_options[] = { · 14 unmodified lines | |||
| 264 | { "cy", "chroma matrix vertical size", OFFSET(cmsize_y), AV_OPT_TYPE_INT, { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS }, | 279 | { "cy", "chroma matrix vertical size", OFFSET(cmsize_y), AV_OPT_TYPE_INT, { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS }, |
| 265 | { "chroma_amount", "chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, | 280 | { "chroma_amount", "chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, |
| 266 | { "ca", "chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, | 281 | { "ca", "chroma effect strength", OFFSET(camount), AV_OPT_TYPE_FLOAT, { .dbl = 0 }, -2, 5, FLAGS }, |
| 282 | { "opencl", "use OpenCL filtering capabilities", OFFSET(opencl), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS }, | ||
| 267 | { NULL }, | 283 | { NULL }, |
| 268 | }; | 284 | }; |
| 269 | 285 | ||