lavfi/unsharp: add opencl unsharp filter

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

9 files changed, 592 insertions, 38 deletions

diff --git a/doc/filters.texi b/doc/filters.texi
index 9456aa4cb7..412981d8b8 100644
--- a/doc/filters.texi
+++ b/doc/filters.texi
@@ -6197,6 +6197,11 @@ Negative values will blur the input video, while positive values will
 sharpen it, a value of zero will disable the effect.
 
 Default value is 0.0.
+
+@item opencl
+If set to 1, specify using OpenCL capabilities, only available if
+FFmpeg was configured with @code{--enable-opencl}. Default value is 0.
+
 @end table
 
 All parameters are optional and default to the equivalent of the
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 084bd302ee..963a71bb11 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -149,7 +149,7 @@ OBJS-$(CONFIG_NOFORMAT_FILTER)               += vf_format.o
 OBJS-$(CONFIG_NOISE_FILTER)                  += vf_noise.o
 OBJS-$(CONFIG_NULL_FILTER)                   += vf_null.o
 OBJS-$(CONFIG_OCV_FILTER)                    += vf_libopencv.o
-OBJS-$(CONFIG_OPENCL)                        += deshake_opencl.o
+OBJS-$(CONFIG_OPENCL)                        += deshake_opencl.o unsharp_opencl.o
 OBJS-$(CONFIG_OVERLAY_FILTER)                += vf_overlay.o
 OBJS-$(CONFIG_PAD_FILTER)                    += vf_pad.o
 OBJS-$(CONFIG_PERMS_FILTER)                  += f_perms.o
diff --git a/libavfilter/opencl_allkernels.c b/libavfilter/opencl_allkernels.c
index 021eec2626..b4530c0b2e 100644
--- a/libavfilter/opencl_allkernels.c
+++ b/libavfilter/opencl_allkernels.c
@@ -22,6 +22,7 @@
 #if CONFIG_OPENCL
 #include "libavutil/opencl.h"
 #include "deshake_kernel.h"
+#include "unsharp_kernel.h"
 #endif
 
 #define OPENCL_REGISTER_KERNEL_CODE(X, x)                                              \
@@ -35,5 +36,6 @@ 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/unsharp.h b/libavfilter/unsharp.h
new file mode 100644
index 0000000000..0add8aa7fa
--- /dev/null
+++ b/libavfilter/unsharp.h
@@ -0,0 +1,78 @@
+/*
+ * 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_H
+#define AVFILTER_UNSHARP_H
+
+#include "config.h"
+#include "avfilter.h"
+#if CONFIG_OPENCL
+#include "libavutil/opencl.h"
+#endif
+
+#define MIN_MATRIX_SIZE 3
+#define MAX_MATRIX_SIZE 63
+
+/* right-shift and round-up */
+#define SHIFTUP(x,shift) (-((-(x))>>(shift)))
+
+#if CONFIG_OPENCL
+
+typedef struct {
+    cl_mem cl_luma_mask;
+    cl_mem cl_chroma_mask;
+    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;
+    AVOpenCLKernelEnv kernel_env;
+} UnsharpOpenclContext;
+
+#endif
+
+typedef struct UnsharpFilterParam {
+    int msize_x;                             ///< matrix width
+    int msize_y;                             ///< matrix height
+    int amount;                              ///< effect amount
+    int steps_x;                             ///< horizontal step count
+    int steps_y;                             ///< vertical step count
+    int scalebits;                           ///< bits to shift pixel
+    int32_t halfscale;                       ///< amount to add to pixel
+    uint32_t *sc[MAX_MATRIX_SIZE - 1];       ///< finite state machine storage
+} UnsharpFilterParam;
+
+typedef struct {
+    const AVClass *class;
+    int lmsize_x, lmsize_y, cmsize_x, cmsize_y;
+    float lamount, camount;
+    UnsharpFilterParam luma;   ///< luma parameters (width, height, amount)
+    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;
+
+#endif /* AVFILTER_UNSHARP_H */
diff --git a/libavfilter/unsharp_kernel.h b/libavfilter/unsharp_kernel.h
new file mode 100644
index 0000000000..dd65bc0a9a
--- /dev/null
+++ b/libavfilter/unsharp_kernel.h
@@ -0,0 +1,137 @@
+/*
+ * 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_KERNEL_H
+#define AVFILTER_UNSHARP_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(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_KERNEL_H */
diff --git a/libavfilter/unsharp_opencl.c b/libavfilter/unsharp_opencl.c
new file mode 100644
index 0000000000..e9a4c930e9
--- /dev/null
+++ b/libavfilter/unsharp_opencl.c
@@ -0,0 +1,282 @@
+/*
+ * 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
+ */
+
+/**
+ * @file
+ * unsharp input video
+ */
+
+#include "unsharp_opencl.h"
+#include "libavutil/opencl_internal.h"
+
+#define PLANE_NUM 3
+
+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;
+    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(counter_size);
+    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; i < 2 * step + 1; i++) {
+        av_freep(&counter[i]);
+    }
+    av_freep(&counter);
+    return ret;
+}
+
+static int compute_mask_matrix(cl_mem cl_mask_matrix, int step_x, int step_y)
+{
+    int i, j, ret = 0;
+    uint32_t *mask_matrix, *mask_x, *mask_y;
+    size_t size_matrix = sizeof(uint32_t) * (2 * step_x + 1) * (2 * step_y + 1);
+    mask_x = av_mallocz(sizeof(uint32_t) * (2 * step_x + 1));
+    if (!mask_x) {
+        ret = AVERROR(ENOMEM);
+        goto end;
+    }
+    mask_y = av_mallocz(sizeof(uint32_t) * (2 * step_y + 1));
+    if (!mask_y) {
+        ret = AVERROR(ENOMEM);
+        goto end;
+    }
+    mask_matrix = av_mallocz(size_matrix);
+    if (!mask_matrix) {
+        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;
+    for (j = 0; j < 2 * step_y + 1; j++) {
+        for (i = 0; i < 2 * step_x + 1; i++) {
+            mask_matrix[i + j * (2 * step_x + 1)] = mask_y[j] * mask_x[i];
+        }
+    }
+    ret = av_opencl_buffer_write(cl_mask_matrix, (uint8_t *)mask_matrix, size_matrix);
+end:
+    av_freep(&mask_x);
+    av_freep(&mask_y);
+    av_freep(&mask_matrix);
+    return ret;
+}
+
+static int generate_mask(AVFilterContext *ctx)
+{
+    UnsharpContext *unsharp = ctx->priv;
+    int i, ret = 0, step_x[2], step_y[2];
+    cl_mem mask_matrix[2];
+    mask_matrix[0] = unsharp->opencl_ctx.cl_luma_mask;
+    mask_matrix[1] = unsharp->opencl_ctx.cl_chroma_mask;
+    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;
+    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 = compute_mask_matrix(mask_matrix[i], 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;
+    int cw = SHIFTUP(link->w, unsharp->hsub);
+    int ch = SHIFTUP(link->h, unsharp->vsub);
+    const size_t global_work_size = link->w * link->h + 2 * ch * cw;
+    FFOpenclParam opencl_param = {0};
+
+    opencl_param.ctx = ctx;
+    opencl_param.kernel = unsharp->opencl_ctx.kernel_env.kernel;
+    ret = ff_opencl_set_parameter(&opencl_param,
+                                  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.kernel_env.command_queue,
+                                    unsharp->opencl_ctx.kernel_env.kernel, 1, NULL,
+                                    &global_work_size, NULL, 0, NULL, NULL);
+    if (status != CL_SUCCESS) {
+        av_log(ctx, AV_LOG_ERROR, "OpenCL run kernel error occurred: %d\n", status);
+        return AVERROR_EXTERNAL;
+    }
+    clFinish(unsharp->opencl_ctx.kernel_env.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;
+    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);
+    if (ret < 0)
+        return ret;
+    ret = generate_mask(ctx);
+    if (ret < 0)
+        return ret;
+    unsharp->opencl_ctx.plane_num = PLANE_NUM;
+    if (!unsharp->opencl_ctx.kernel_env.kernel) {
+        ret = av_opencl_create_kernel(&unsharp->opencl_ctx.kernel_env, "unsharp");
+        if (ret < 0) {
+            av_log(ctx, AV_LOG_ERROR, "OpenCL failed to create kernel with name 'unsharp'\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_release_kernel(&unsharp->opencl_ctx.kernel_env);
+    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 = SHIFTUP(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
new file mode 100644
index 0000000000..3aefab62e0
--- /dev/null
+++ b/libavfilter/unsharp_opencl.h
@@ -0,0 +1,34 @@
+/*
+ * 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/version.h b/libavfilter/version.h
index 9e8e83a38d..7f5eefef52 100644
--- a/libavfilter/version.h
+++ b/libavfilter/version.h
@@ -30,7 +30,7 @@
 
 #define LIBAVFILTER_VERSION_MAJOR  3
 #define LIBAVFILTER_VERSION_MINOR  60
-#define LIBAVFILTER_VERSION_MICRO 101
+#define LIBAVFILTER_VERSION_MICRO 102
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
                                                LIBAVFILTER_VERSION_MINOR, \
diff --git a/libavfilter/vf_unsharp.c b/libavfilter/vf_unsharp.c
index 2d7aab26aa..4ea4348801 100644
--- a/libavfilter/vf_unsharp.c
+++ b/libavfilter/vf_unsharp.c
@@ -46,36 +46,12 @@
 #include "libavutil/mem.h"
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
-
-#define MIN_MATRIX_SIZE 3
-#define MAX_MATRIX_SIZE 63
-
-/* right-shift and round-up */
-#define SHIFTUP(x,shift) (-((-(x))>>(shift)))
-
-typedef struct FilterParam {
-    int msize_x;                             ///< matrix width
-    int msize_y;                             ///< matrix height
-    int amount;                              ///< effect amount
-    int steps_x;                             ///< horizontal step count
-    int steps_y;                             ///< vertical step count
-    int scalebits;                           ///< bits to shift pixel
-    int32_t halfscale;                       ///< amount to add to pixel
-    uint32_t *sc[MAX_MATRIX_SIZE - 1];       ///< finite state machine storage
-} FilterParam;
-
-typedef struct {
-    const AVClass *class;
-    int lmsize_x, lmsize_y, cmsize_x, cmsize_y;
-    float lamount, camount;
-    FilterParam luma;   ///< luma parameters (width, height, amount)
-    FilterParam chroma; ///< chroma parameters (width, height, amount)
-    int hsub, vsub;
-} UnsharpContext;
+#include "unsharp.h"
+#include "unsharp_opencl.h"
 
 static void apply_unsharp(      uint8_t *dst, int dst_stride,
                           const uint8_t *src, int src_stride,
-                          int width, int height, FilterParam *fp)
+                          int width, int height, UnsharpFilterParam *fp)
 {
     uint32_t **sc = fp->sc;
     uint32_t sr[MAX_MATRIX_SIZE - 1], tmp1, tmp2;
@@ -131,7 +107,25 @@ static void apply_unsharp(      uint8_t *dst, int dst_stride,
     }
 }
 
-static void set_filter_param(FilterParam *fp, int msize_x, int msize_y, float amount)
+static int apply_unsharp_c(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
+{
+    AVFilterLink *inlink = ctx->inputs[0];
+    UnsharpContext *unsharp = ctx->priv;
+    int i, plane_w[3], plane_h[3];
+    UnsharpFilterParam *fp[3];
+    plane_w[0] = inlink->w;
+    plane_w[1] = plane_w[2] = SHIFTUP(inlink->w, unsharp->hsub);
+    plane_h[0] = inlink->h;
+    plane_h[1] = plane_h[2] = SHIFTUP(inlink->h, unsharp->vsub);
+    fp[0] = &unsharp->luma;
+    fp[1] = fp[2] = &unsharp->chroma;
+    for (i = 0; i < 3; i++) {
+        apply_unsharp(out->data[i], out->linesize[i], in->data[i], in->linesize[i], plane_w[i], plane_h[i], fp[i]);
+    }
+    return 0;
+}
+
+static void set_filter_param(UnsharpFilterParam *fp, int msize_x, int msize_y, float amount)
 {
     fp->msize_x = msize_x;
     fp->msize_y = msize_y;
@@ -145,12 +139,24 @@ static void set_filter_param(FilterParam *fp, int msize_x, int msize_y, float am
 
 static av_cold int init(AVFilterContext *ctx)
 {
+    int ret = 0;
     UnsharpContext *unsharp = ctx->priv;
 
 
     set_filter_param(&unsharp->luma,   unsharp->lmsize_x, unsharp->lmsize_y, unsharp->lamount);
     set_filter_param(&unsharp->chroma, unsharp->cmsize_x, unsharp->cmsize_y, unsharp->camount);
 
+    unsharp->apply_unsharp = apply_unsharp_c;
+    if (!CONFIG_OPENCL && unsharp->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 && unsharp->opencl) {
+        unsharp->apply_unsharp = ff_opencl_apply_unsharp;
+        ret = ff_opencl_unsharp_init(ctx);
+        if (ret < 0)
+            return ret;
+    }
     return 0;
 }
 
@@ -167,7 +173,7 @@ static int query_formats(AVFilterContext *ctx)
     return 0;
 }
 
-static int init_filter_param(AVFilterContext *ctx, FilterParam *fp, const char *effect_type, int width)
+static int init_filter_param(AVFilterContext *ctx, UnsharpFilterParam *fp, const char *effect_type, int width)
 {
     int z;
     const char *effect = fp->amount == 0 ? "none" : fp->amount < 0 ? "blur" : "sharpen";
@@ -208,7 +214,7 @@ static int config_props(AVFilterLink *link)
     return 0;
 }
 
-static void free_filter_param(FilterParam *fp)
+static void free_filter_param(UnsharpFilterParam *fp)
 {
     int z;
 
@@ -220,6 +226,10 @@ static av_cold void uninit(AVFilterContext *ctx)
 {
     UnsharpContext *unsharp = ctx->priv;
 
+    if (CONFIG_OPENCL && unsharp->opencl) {
+        ff_opencl_unsharp_uninit(ctx);
+    }
+
     free_filter_param(&unsharp->luma);
     free_filter_param(&unsharp->chroma);
 }
@@ -229,8 +239,7 @@ static int filter_frame(AVFilterLink *link, AVFrame *in)
     UnsharpContext *unsharp = link->dst->priv;
     AVFilterLink *outlink   = link->dst->outputs[0];
     AVFrame *out;
-    int cw = SHIFTUP(link->w, unsharp->hsub);
-    int ch = SHIFTUP(link->h, unsharp->vsub);
+    int ret = 0;
 
     out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
     if (!out) {
@@ -238,12 +247,18 @@ static int filter_frame(AVFilterLink *link, AVFrame *in)
         return AVERROR(ENOMEM);
     }
     av_frame_copy_props(out, in);
+    if (CONFIG_OPENCL && unsharp->opencl) {
+        ret = ff_opencl_unsharp_process_inout_buf(link->dst, in, out);
+        if (ret < 0)
+            goto end;
+    }
 
-    apply_unsharp(out->data[0], out->linesize[0], in->data[0], in->linesize[0], link->w, link->h, &unsharp->luma);
-    apply_unsharp(out->data[1], out->linesize[1], in->data[1], in->linesize[1], cw,      ch,      &unsharp->chroma);
-    apply_unsharp(out->data[2], out->linesize[2], in->data[2], in->linesize[2], cw,      ch,      &unsharp->chroma);
-
+    ret = unsharp->apply_unsharp(link->dst, in, out);
+end:
     av_frame_free(&in);
+
+    if (ret < 0)
+        return ret;
     return ff_filter_frame(outlink, out);
 }
 
@@ -264,6 +279,7 @@ static const AVOption unsharp_options[] = {
     { "cy",             "chroma matrix vertical size",   OFFSET(cmsize_y), AV_OPT_TYPE_INT,   { .i64 = 5 }, MIN_SIZE, MAX_SIZE, FLAGS },
     { "chroma_amount",  "chroma effect strength",        OFFSET(camount),  AV_OPT_TYPE_FLOAT, { .dbl = 0 },       -2,        5, FLAGS },
     { "ca",             "chroma effect strength",        OFFSET(camount),  AV_OPT_TYPE_FLOAT, { .dbl = 0 },       -2,        5, FLAGS },
+    { "opencl",         "use OpenCL filtering capabilities", OFFSET(opencl), AV_OPT_TYPE_INT, { .i64 = 0 },        0,        1, FLAGS },
     { NULL },
 };