From 8b34700cbd43b18950dc81cceb958d0b39126806 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Mon, 3 Dec 2018 23:48:04 +0100 Subject: [PATCH 01/18] avutil: add av_buffer_pool_flush() Used by V4L2 request API hwaccel Signed-off-by: Jonas Karlman --- libavutil/buffer.c | 13 +++++++++++++ libavutil/buffer.h | 5 +++++ 2 files changed, 18 insertions(+) diff --git a/libavutil/buffer.c b/libavutil/buffer.c index e4562a7..09da632 100644 --- a/libavutil/buffer.c +++ b/libavutil/buffer.c @@ -319,6 +319,19 @@ static void buffer_pool_free(AVBufferPool *pool) av_freep(&pool); } +void av_buffer_pool_flush(AVBufferPool *pool) +{ + ff_mutex_lock(&pool->mutex); + while (pool->pool) { + BufferPoolEntry *buf = pool->pool; + pool->pool = buf->next; + + buf->free(buf->opaque, buf->data); + av_freep(&buf); + } + ff_mutex_unlock(&pool->mutex); +} + void av_buffer_pool_uninit(AVBufferPool **ppool) { AVBufferPool *pool; diff --git a/libavutil/buffer.h b/libavutil/buffer.h index e1ef5b7..fde9bae 100644 --- a/libavutil/buffer.h +++ b/libavutil/buffer.h @@ -284,6 +284,11 @@ AVBufferPool *av_buffer_pool_init2(size_t size, void *opaque, AVBufferRef* (*alloc)(void *opaque, size_t size), void (*pool_free)(void *opaque)); +/** + * Free all available buffers in a buffer pool. + */ + void av_buffer_pool_flush(AVBufferPool *pool); + /** * Mark the pool as being available for freeing. It will actually be freed only * once all the allocated buffers associated with the pool are released. Thus it -- 2.44.0 From da33bb8f94c073c8fb6dc7a3f3efe72d78b64a14 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 15 Dec 2018 22:32:16 +0100 Subject: [PATCH 02/18] Add common V4L2 request API code Signed-off-by: Jonas Karlman Signed-off-by: Alex Bee --- configure | 12 + libavcodec/Makefile | 1 + libavcodec/hwconfig.h | 2 + libavcodec/v4l2_request.c | 1048 +++++++++++++++++++++++++++++++++++++ libavcodec/v4l2_request.h | 78 +++ 5 files changed, 1141 insertions(+) create mode 100644 libavcodec/v4l2_request.c create mode 100644 libavcodec/v4l2_request.h diff --git a/configure b/configure index a89cfa6..503eb06 100755 --- a/configure +++ b/configure @@ -279,6 +279,7 @@ External library support: if openssl, gnutls or mbedtls is not used [no] --enable-libtwolame enable MP2 encoding via libtwolame [no] --enable-libuavs3d enable AVS3 decoding via libuavs3d [no] + --enable-libudev enable libudev [no] --enable-libv4l2 enable libv4l2/v4l-utils [no] --enable-libvidstab enable video stabilization using vid.stab [no] --enable-libvmaf enable vmaf filter via libvmaf [no] @@ -347,6 +348,7 @@ External library support: --enable-omx-rpi enable OpenMAX IL code for Raspberry Pi [no] --enable-rkmpp enable Rockchip Media Process Platform code [no] --disable-v4l2-m2m disable V4L2 mem2mem code [autodetect] + --enable-v4l2-request enable V4L2 request API code [no] --disable-vaapi disable Video Acceleration API (mainly Unix/Intel) code [autodetect] --disable-vdpau disable Nvidia Video Decode and Presentation API for Unix code [autodetect] --disable-videotoolbox disable VideoToolbox code [autodetect] @@ -1896,6 +1898,7 @@ EXTERNAL_LIBRARY_LIST=" libtheora libtwolame libuavs3d + libudev libv4l2 libvmaf libvorbis @@ -1952,6 +1955,7 @@ HWACCEL_LIBRARY_LIST=" mmal omx opencl + v4l2_request " DOCUMENT_LIST=" @@ -3045,6 +3049,7 @@ dxva2_deps="dxva2api_h DXVA2_ConfigPictureDecode ole32 user32" ffnvcodec_deps_any="libdl LoadLibrary" mediacodec_deps="android" nvdec_deps="ffnvcodec" +v4l2_request_deps="linux_videodev2_h linux_media_h v4l2_timeval_to_ns libdrm libudev" vaapi_x11_deps="xlib_x11" videotoolbox_hwaccel_deps="videotoolbox pthreads" videotoolbox_hwaccel_extralibs="-framework QuartzCore" @@ -6807,6 +6812,7 @@ enabled libtwolame && require libtwolame twolame.h twolame_init -ltwolame { check_lib libtwolame twolame.h twolame_encode_buffer_float32_interleaved -ltwolame || die "ERROR: libtwolame must be installed and version must be >= 0.3.10"; } enabled libuavs3d && require_pkg_config libuavs3d "uavs3d >= 1.1.41" uavs3d.h uavs3d_decode +enabled libudev && require_pkg_config libudev libudev libudev.h udev_new enabled libv4l2 && require_pkg_config libv4l2 libv4l2 libv4l2.h v4l2_ioctl enabled libvidstab && require_pkg_config libvidstab "vidstab >= 0.98" vid.stab/libvidstab.h vsMotionDetectInit enabled libvmaf && require_pkg_config libvmaf "libvmaf >= 2.0.0" libvmaf.h vmaf_init @@ -6910,6 +6916,10 @@ enabled rkmpp && { require_pkg_config rkmpp rockchip_mpp rockchip/r { enabled libdrm || die "ERROR: rkmpp requires --enable-libdrm"; } } +enabled v4l2_request && { enabled libdrm || + die "ERROR: v4l2-request requires --enable-libdrm"; } && + { enabled libudev || + die "ERROR: v4l2-request requires --enable-libudev"; } enabled vapoursynth && require_pkg_config vapoursynth "vapoursynth-script >= 42" VSScript.h vsscript_init @@ -6992,6 +7002,8 @@ if enabled v4l2_m2m; then check_cc vp9_v4l2_m2m linux/videodev2.h "int i = V4L2_PIX_FMT_VP9;" fi +check_func_headers "linux/media.h linux/videodev2.h" v4l2_timeval_to_ns + check_headers sys/videoio.h test_code cc sys/videoio.h "struct v4l2_frmsizeenum vfse; vfse.discrete.width = 0;" && enable_sanitized struct_v4l2_frmivalenum_discrete diff --git a/libavcodec/Makefile b/libavcodec/Makefile index ec57e53..5e626c9 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -169,6 +169,7 @@ OBJS-$(CONFIG_VP3DSP) += vp3dsp.o OBJS-$(CONFIG_VP56DSP) += vp56dsp.o OBJS-$(CONFIG_VP8DSP) += vp8dsp.o OBJS-$(CONFIG_V4L2_M2M) += v4l2_m2m.o v4l2_context.o v4l2_buffers.o v4l2_fmt.o +OBJS-$(CONFIG_V4L2_REQUEST) += v4l2_request.o OBJS-$(CONFIG_WMA_FREQS) += wma_freqs.o OBJS-$(CONFIG_WMV2DSP) += wmv2dsp.o diff --git a/libavcodec/hwconfig.h b/libavcodec/hwconfig.h index e164722..ebafb6f 100644 --- a/libavcodec/hwconfig.h +++ b/libavcodec/hwconfig.h @@ -77,6 +77,8 @@ void ff_hwaccel_uninit(AVCodecContext *avctx); HW_CONFIG_HWACCEL(1, 1, 1, VULKAN, VULKAN, ff_ ## codec ## _vulkan_hwaccel) #define HWACCEL_D3D11VA(codec) \ HW_CONFIG_HWACCEL(0, 0, 1, D3D11VA_VLD, NONE, ff_ ## codec ## _d3d11va_hwaccel) +#define HWACCEL_V4L2REQUEST(codec) \ + HW_CONFIG_HWACCEL(1, 0, 0, DRM_PRIME, DRM, ff_ ## codec ## _v4l2request_hwaccel) #define HW_CONFIG_ENCODER(device, frames, ad_hoc, format, device_type_) \ &(const AVCodecHWConfigInternal) { \ diff --git a/libavcodec/v4l2_request.c b/libavcodec/v4l2_request.c new file mode 100644 index 0000000..595a53e --- /dev/null +++ b/libavcodec/v4l2_request.c @@ -0,0 +1,1048 @@ +/* + * 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 +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "decode.h" +#include "internal.h" +#include "v4l2_request.h" + +static const AVClass v4l2_request_context_class = { + .class_name = "V4L2RequestContext", + .item_name = av_default_item_name, + .version = LIBAVUTIL_VERSION_INT, +}; + +uint64_t ff_v4l2_request_get_capture_timestamp(AVFrame *frame) +{ + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)frame->data[0]; + return req ? v4l2_timeval_to_ns(&req->capture.buffer.timestamp) : 0; +} + +int ff_v4l2_request_reset_frame(AVCodecContext *avctx, AVFrame *frame) +{ + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)frame->data[0]; + memset(&req->drm, 0, sizeof(AVDRMFrameDescriptor)); + req->output.used = 0; + return 0; +} + +int ff_v4l2_request_append_output_buffer(AVCodecContext *avctx, AVFrame *frame, const uint8_t *data, uint32_t size) +{ + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)frame->data[0]; + if (req->output.used + size + (AV_INPUT_BUFFER_PADDING_SIZE * 4) <= req->output.size) { + memcpy(req->output.addr + req->output.used, data, size); + req->output.used += size; + } else { + av_log(avctx, AV_LOG_ERROR, "%s: output.used=%u output.size=%u size=%u\n", __func__, req->output.used, req->output.size, size); + } + return 0; +} + +static int v4l2_request_controls(V4L2RequestContext *ctx, int request_fd, unsigned long type, struct v4l2_ext_control *control, int count) +{ + struct v4l2_ext_controls controls = { + .controls = control, + .count = count, + .request_fd = request_fd, + .which = (request_fd >= 0) ? V4L2_CTRL_WHICH_REQUEST_VAL : 0, + }; + + if (!control || !count) + return 0; + + return ioctl(ctx->video_fd, type, &controls); +} + +static int v4l2_request_set_controls(V4L2RequestContext *ctx, int request_fd, struct v4l2_ext_control *control, int count) +{ + return v4l2_request_controls(ctx, request_fd, VIDIOC_S_EXT_CTRLS, control, count); +} + +int ff_v4l2_request_set_controls(AVCodecContext *avctx, struct v4l2_ext_control *control, int count) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret; + + ret = v4l2_request_controls(ctx, -1, VIDIOC_S_EXT_CTRLS, control, count); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: set controls failed, %s (%d)\n", __func__, strerror(errno), errno); + return AVERROR(EINVAL); + } + + return ret; +} + +int ff_v4l2_request_get_controls(AVCodecContext *avctx, struct v4l2_ext_control *control, int count) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret; + + ret = v4l2_request_controls(ctx, -1, VIDIOC_G_EXT_CTRLS, control, count); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get controls failed, %s (%d)\n", __func__, strerror(errno), errno); + return AVERROR(EINVAL); + } + + return ret; +} + +int ff_v4l2_request_query_control(AVCodecContext *avctx, struct v4l2_query_ext_ctrl *control) +{ + int ret; + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + + ret = ioctl(ctx->video_fd, VIDIOC_QUERY_EXT_CTRL, control); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: query control failed, %s (%d)\n", __func__, strerror(errno), errno); + return AVERROR(EINVAL); + } + + return 0; +} + +int ff_v4l2_request_query_control_default_value(AVCodecContext *avctx, uint32_t id) +{ + int ret; + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + struct v4l2_queryctrl control = { + .id = id, + }; + + ret = ioctl(ctx->video_fd, VIDIOC_QUERYCTRL, &control); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: query control failed, %s (%d)\n", __func__, strerror(errno), errno); + return AVERROR(EINVAL); + } + + return control.default_value; +} + +static int v4l2_request_queue_buffer(V4L2RequestContext *ctx, int request_fd, V4L2RequestBuffer *buf, uint32_t flags) +{ + struct v4l2_plane planes[1] = {}; + struct v4l2_buffer buffer = { + .type = buf->buffer.type, + .memory = buf->buffer.memory, + .index = buf->index, + .timestamp.tv_usec = buf->index + 1, + .bytesused = buf->used, + .request_fd = request_fd, + .flags = ((request_fd >= 0) ? V4L2_BUF_FLAG_REQUEST_FD : 0) | flags, + }; + + if (V4L2_TYPE_IS_MULTIPLANAR(buf->buffer.type)) { + planes[0].bytesused = buf->used; + buffer.bytesused = 0; + buffer.length = 1; + buffer.m.planes = planes; + } + + return ioctl(ctx->video_fd, VIDIOC_QBUF, &buffer); +} + +static int v4l2_request_dequeue_buffer(V4L2RequestContext *ctx, V4L2RequestBuffer *buf) +{ + int ret; + struct v4l2_plane planes[1] = {}; + struct v4l2_buffer buffer = { + .type = buf->buffer.type, + .memory = buf->buffer.memory, + .index = buf->index, + }; + + if (V4L2_TYPE_IS_MULTIPLANAR(buf->buffer.type)) { + buffer.length = 1; + buffer.m.planes = planes; + } + + ret = ioctl(ctx->video_fd, VIDIOC_DQBUF, &buffer); + if (ret < 0) + return ret; + + buf->buffer.timestamp = buffer.timestamp; + return 0; +} + +static const struct { + uint32_t pixelformat; + enum AVPixelFormat sw_format; + uint32_t drm_format; + uint64_t format_modifier; +} v4l2_request_capture_pixelformats[] = { + { V4L2_PIX_FMT_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_LINEAR }, + { V4L2_PIX_FMT_SUNXI_TILED_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_ALLWINNER_TILED }, +}; + +static int v4l2_request_set_drm_descriptor(V4L2RequestDescriptor *req, struct v4l2_format *format) +{ + AVDRMFrameDescriptor *desc = &req->drm; + AVDRMLayerDescriptor *layer = &desc->layers[0]; + uint32_t pixelformat = V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.pixelformat : format->fmt.pix.pixelformat; + + layer->format = 0; + for (int i = 0; i < FF_ARRAY_ELEMS(v4l2_request_capture_pixelformats); i++) { + if (pixelformat == v4l2_request_capture_pixelformats[i].pixelformat) { + layer->format = v4l2_request_capture_pixelformats[i].drm_format; + desc->objects[0].format_modifier = v4l2_request_capture_pixelformats[i].format_modifier; + break; + } + } + + if (!layer->format) + return -1; + + desc->nb_objects = 1; + desc->objects[0].fd = req->capture.fd; + desc->objects[0].size = req->capture.size; + + desc->nb_layers = 1; + layer->nb_planes = 2; + + layer->planes[0].object_index = 0; + layer->planes[0].offset = 0; + layer->planes[0].pitch = V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.plane_fmt[0].bytesperline : format->fmt.pix.bytesperline; + + layer->planes[1].object_index = 0; + layer->planes[1].offset = layer->planes[0].pitch * (V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.height : format->fmt.pix.height); + layer->planes[1].pitch = layer->planes[0].pitch; + + return 0; +} + +static int v4l2_request_queue_decode(AVCodecContext *avctx, AVFrame *frame, struct v4l2_ext_control *control, int count, int first_slice, int last_slice) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)frame->data[0]; + struct timeval tv = { 2, 0 }; + fd_set except_fds; + int ret; + + av_log(avctx, AV_LOG_DEBUG, "%s: ctx=%p used=%u controls=%d index=%d fd=%d request_fd=%d first_slice=%d last_slice=%d\n", __func__, + ctx, req->output.used, count, req->capture.index, req->capture.fd, req->request_fd, first_slice, last_slice); + + ret = v4l2_request_set_controls(ctx, req->request_fd, control, count); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: set controls failed for request %d, %s (%d)\n", __func__, req->request_fd, strerror(errno), errno); + return -1; + } + + memset(req->output.addr + req->output.used, 0, AV_INPUT_BUFFER_PADDING_SIZE * 4); + + ret = v4l2_request_queue_buffer(ctx, req->request_fd, &req->output, last_slice ? 0 : V4L2_BUF_FLAG_M2M_HOLD_CAPTURE_BUF); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: queue output buffer %d failed for request %d, %s (%d)\n", __func__, req->output.index, req->request_fd, strerror(errno), errno); + return -1; + } + + if (first_slice) { + ret = v4l2_request_queue_buffer(ctx, -1, &req->capture, 0); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: queue capture buffer %d failed for request %d, %s (%d)\n", __func__, req->capture.index, req->request_fd, strerror(errno), errno); + return -1; + } + } + + // NOTE: do we need to dequeue when request fails/timeout? + + // 4. queue request and wait + ret = ioctl(req->request_fd, MEDIA_REQUEST_IOC_QUEUE, NULL); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: queue request %d failed, %s (%d)\n", __func__, req->request_fd, strerror(errno), errno); + goto fail; + } + + FD_ZERO(&except_fds); + FD_SET(req->request_fd, &except_fds); + + ret = select(req->request_fd + 1, NULL, NULL, &except_fds, &tv); + if (ret == 0) { + av_log(avctx, AV_LOG_ERROR, "%s: request %d timeout\n", __func__, req->request_fd); + goto fail; + } else if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: select request %d failed, %s (%d)\n", __func__, req->request_fd, strerror(errno), errno); + goto fail; + } + + ret = v4l2_request_dequeue_buffer(ctx, &req->output); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: dequeue output buffer %d failed for request %d, %s (%d)\n", __func__, req->output.index, req->request_fd, strerror(errno), errno); + return -1; + } + + if (last_slice) { + ret = v4l2_request_dequeue_buffer(ctx, &req->capture); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: dequeue capture buffer %d failed for request %d, %s (%d)\n", __func__, req->capture.index, req->request_fd, strerror(errno), errno); + return -1; + } + } + + // TODO: check errors + // buffer.flags & V4L2_BUF_FLAG_ERROR + + ret = ioctl(req->request_fd, MEDIA_REQUEST_IOC_REINIT, NULL); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: reinit request %d failed, %s (%d)\n", __func__, req->request_fd, strerror(errno), errno); + return -1; + } + + if (last_slice) + return v4l2_request_set_drm_descriptor(req, &ctx->format); + + return 0; + +fail: + ret = v4l2_request_dequeue_buffer(ctx, &req->output); + if (ret < 0) + av_log(avctx, AV_LOG_ERROR, "%s: dequeue output buffer %d failed for request %d, %s (%d)\n", __func__, req->output.index, req->request_fd, strerror(errno), errno); + + ret = v4l2_request_dequeue_buffer(ctx, &req->capture); + if (ret < 0) + av_log(avctx, AV_LOG_ERROR, "%s: dequeue capture buffer %d failed for request %d, %s (%d)\n", __func__, req->capture.index, req->request_fd, strerror(errno), errno); + + ret = ioctl(req->request_fd, MEDIA_REQUEST_IOC_REINIT, NULL); + if (ret < 0) + av_log(avctx, AV_LOG_ERROR, "%s: reinit request %d failed, %s (%d)\n", __func__, req->request_fd, strerror(errno), errno); + + return -1; +} + +int ff_v4l2_request_decode_slice(AVCodecContext *avctx, AVFrame *frame, struct v4l2_ext_control *control, int count, int first_slice, int last_slice) +{ + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)frame->data[0]; + + // fall back to queue each slice as a full frame + if ((req->output.capabilities & V4L2_BUF_CAP_SUPPORTS_M2M_HOLD_CAPTURE_BUF) != V4L2_BUF_CAP_SUPPORTS_M2M_HOLD_CAPTURE_BUF) + return v4l2_request_queue_decode(avctx, frame, control, count, 1, 1); + + return v4l2_request_queue_decode(avctx, frame, control, count, first_slice, last_slice); +} + +int ff_v4l2_request_decode_frame(AVCodecContext *avctx, AVFrame *frame, struct v4l2_ext_control *control, int count) +{ + return v4l2_request_queue_decode(avctx, frame, control, count, 1, 1); +} + +static int v4l2_request_try_framesize(AVCodecContext *avctx, uint32_t pixelformat) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + struct v4l2_frmsizeenum frmsize = { + .index = 0, + .pixel_format = pixelformat, + }; + + if (ioctl(ctx->video_fd, VIDIOC_ENUM_FRAMESIZES, &frmsize) < 0) + return 0; + + /* + * We only validate min/max framesize for V4L2_FRMSIZE_TYPE_STEPWISE here, since the alignment + * which is eventually needed will be done driver-side later in VIDIOC_S_FMT and there is no need + * validate step_width/step_height here + */ + do { + if (frmsize.type == V4L2_FRMSIZE_TYPE_DISCRETE && frmsize.discrete.width == avctx->coded_width && + frmsize.discrete.height == avctx->coded_height) + return 0; + else if ((frmsize.type == V4L2_FRMSIZE_TYPE_STEPWISE || frmsize.type == V4L2_FRMSIZE_TYPE_CONTINUOUS) && + avctx->coded_width >= frmsize.stepwise.min_width && avctx->coded_height >= frmsize.stepwise.min_height && + avctx->coded_width <= frmsize.stepwise.max_width && avctx->coded_height <= frmsize.stepwise.max_height) + return 0; + + frmsize.index++; + } while (ioctl(ctx->video_fd, VIDIOC_ENUM_FRAMESIZES, &frmsize) >= 0); + + av_log(avctx, AV_LOG_INFO, "%s: pixelformat %s not supported for width %u height %u\n", __func__, av_fourcc2str(pixelformat), avctx->coded_width, avctx->coded_height); + return -1; +} + +static int v4l2_request_try_format(AVCodecContext *avctx, enum v4l2_buf_type type, uint32_t pixelformat) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + struct v4l2_fmtdesc fmtdesc = { + .index = 0, + .type = type, + }; + + if (V4L2_TYPE_IS_OUTPUT(type)) { + struct v4l2_create_buffers buffers = { + .count = 0, + .memory = V4L2_MEMORY_MMAP, + .format.type = type, + }; + + if (ioctl(ctx->video_fd, VIDIOC_CREATE_BUFS, &buffers) < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: create buffers failed for type %u, %s (%d)\n", __func__, type, strerror(errno), errno); + return -1; + } + + if ((buffers.capabilities & V4L2_BUF_CAP_SUPPORTS_REQUESTS) != V4L2_BUF_CAP_SUPPORTS_REQUESTS) { + av_log(avctx, AV_LOG_INFO, "%s: output buffer type do not support requests, capabilities %u\n", __func__, buffers.capabilities); + return -1; + } + } + + while (ioctl(ctx->video_fd, VIDIOC_ENUM_FMT, &fmtdesc) >= 0) { + if (fmtdesc.pixelformat == pixelformat) + return 0; + + fmtdesc.index++; + } + + av_log(avctx, AV_LOG_DEBUG, "%s: pixelformat %s not supported for type %u\n", __func__, av_fourcc2str(pixelformat), type); + return -1; +} + +static int v4l2_request_set_format(AVCodecContext *avctx, enum v4l2_buf_type type, uint32_t pixelformat, uint32_t buffersize) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + struct v4l2_format format = { + .type = type, + }; + + if (V4L2_TYPE_IS_MULTIPLANAR(type)) { + format.fmt.pix_mp.width = avctx->coded_width; + format.fmt.pix_mp.height = avctx->coded_height; + format.fmt.pix_mp.pixelformat = pixelformat; + format.fmt.pix_mp.plane_fmt[0].sizeimage = buffersize; + format.fmt.pix_mp.num_planes = 1; + } else { + format.fmt.pix.width = avctx->coded_width; + format.fmt.pix.height = avctx->coded_height; + format.fmt.pix.pixelformat = pixelformat; + format.fmt.pix.sizeimage = buffersize; + } + + return ioctl(ctx->video_fd, VIDIOC_S_FMT, &format); +} + +static int v4l2_request_select_capture_format(AVCodecContext *avctx) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + enum v4l2_buf_type type = ctx->format.type; + +#if 1 + struct v4l2_format format = { + .type = type, + }; + struct v4l2_fmtdesc fmtdesc = { + .index = 0, + .type = type, + }; + uint32_t pixelformat; + + if (ioctl(ctx->video_fd, VIDIOC_G_FMT, &format) < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get capture format failed, %s (%d)\n", __func__, strerror(errno), errno); + return -1; + } + + pixelformat = V4L2_TYPE_IS_MULTIPLANAR(type) ? format.fmt.pix_mp.pixelformat : format.fmt.pix.pixelformat; + + for (int i = 0; i < FF_ARRAY_ELEMS(v4l2_request_capture_pixelformats); i++) { + if (pixelformat == v4l2_request_capture_pixelformats[i].pixelformat) + return v4l2_request_set_format(avctx, type, pixelformat, 0); + } + + while (ioctl(ctx->video_fd, VIDIOC_ENUM_FMT, &fmtdesc) >= 0) { + for (int i = 0; i < FF_ARRAY_ELEMS(v4l2_request_capture_pixelformats); i++) { + if (fmtdesc.pixelformat == v4l2_request_capture_pixelformats[i].pixelformat) + return v4l2_request_set_format(avctx, type, fmtdesc.pixelformat, 0); + } + + fmtdesc.index++; + } +#else + for (int i = 0; i < FF_ARRAY_ELEMS(v4l2_request_capture_pixelformats); i++) { + uint32_t pixelformat = v4l2_request_capture_pixelformats[i].pixelformat; + if (!v4l2_request_try_format(avctx, type, pixelformat)) + return v4l2_request_set_format(avctx, type, pixelformat, 0); + } +#endif + + return -1; +} + +static int v4l2_request_probe_video_device(struct udev_device *device, AVCodecContext *avctx, uint32_t pixelformat, uint32_t buffersize, struct v4l2_ext_control *control, int count) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret = AVERROR(EINVAL); + struct v4l2_capability capability = {0}; + unsigned int capabilities = 0; + + const char *path = udev_device_get_devnode(device); + if (!path) { + av_log(avctx, AV_LOG_ERROR, "%s: get video device devnode failed\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + ctx->video_fd = open(path, O_RDWR | O_NONBLOCK, 0); + if (ctx->video_fd < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: opening %s failed, %s (%d)\n", __func__, path, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = ioctl(ctx->video_fd, VIDIOC_QUERYCAP, &capability); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get video capability failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + if (capability.capabilities & V4L2_CAP_DEVICE_CAPS) + capabilities = capability.device_caps; + else + capabilities = capability.capabilities; + + av_log(avctx, AV_LOG_DEBUG, "%s: ctx=%p path=%s capabilities=%u\n", __func__, ctx, path, capabilities); + + if ((capabilities & V4L2_CAP_STREAMING) != V4L2_CAP_STREAMING) { + av_log(avctx, AV_LOG_ERROR, "%s: missing required streaming capability\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + if ((capabilities & V4L2_CAP_VIDEO_M2M_MPLANE) == V4L2_CAP_VIDEO_M2M_MPLANE) { + ctx->output_type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE; + ctx->format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE; + } else if ((capabilities & V4L2_CAP_VIDEO_M2M) == V4L2_CAP_VIDEO_M2M) { + ctx->output_type = V4L2_BUF_TYPE_VIDEO_OUTPUT; + ctx->format.type = V4L2_BUF_TYPE_VIDEO_CAPTURE; + } else { + av_log(avctx, AV_LOG_ERROR, "%s: missing required mem2mem capability\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = v4l2_request_try_format(avctx, ctx->output_type, pixelformat); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "%s: try output format failed\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = v4l2_request_try_framesize(avctx, pixelformat); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "%s: try framesize failed\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = v4l2_request_set_format(avctx, ctx->output_type, pixelformat, buffersize); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: set output format failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = v4l2_request_set_controls(ctx, -1, control, count); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "%s: set controls failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = v4l2_request_select_capture_format(avctx); + if (ret < 0) { + av_log(avctx, AV_LOG_DEBUG, "%s: select capture format failed\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + return 0; + +fail: + if (ctx->video_fd >= 0) { + close(ctx->video_fd); + ctx->video_fd = -1; + } + return ret; +} + +static int v4l2_request_init_context(AVCodecContext *avctx) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret; + + ret = ioctl(ctx->video_fd, VIDIOC_G_FMT, &ctx->format); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get capture format failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + if (V4L2_TYPE_IS_MULTIPLANAR(ctx->format.type)) { + struct v4l2_pix_format_mplane *fmt = &ctx->format.fmt.pix_mp; + av_log(ctx, AV_LOG_DEBUG, "%s: pixelformat=%s width=%u height=%u bytesperline=%u sizeimage=%u num_planes=%u\n", __func__, + av_fourcc2str(fmt->pixelformat), fmt->width, fmt->height, fmt->plane_fmt[0].bytesperline, fmt->plane_fmt[0].sizeimage, fmt->num_planes); + } else { + struct v4l2_pix_format *fmt = &ctx->format.fmt.pix; + av_log(ctx, AV_LOG_DEBUG, "%s: pixelformat=%s width=%u height=%u bytesperline=%u sizeimage=%u\n", __func__, + av_fourcc2str(fmt->pixelformat), fmt->width, fmt->height, fmt->bytesperline, fmt->sizeimage); + } + + ret = ff_decode_get_hw_frames_ctx(avctx, AV_HWDEVICE_TYPE_DRM); + if (ret < 0) + goto fail; + + ret = ioctl(ctx->video_fd, VIDIOC_STREAMON, &ctx->output_type); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: output stream on failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = ioctl(ctx->video_fd, VIDIOC_STREAMON, &ctx->format.type); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: capture stream on failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + return 0; + +fail: + ff_v4l2_request_uninit(avctx); + return ret; +} + +static int v4l2_request_probe_media_device(struct udev_device *device, AVCodecContext *avctx, uint32_t pixelformat, uint32_t buffersize, struct v4l2_ext_control *control, int count) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret; + struct media_device_info device_info = {0}; + struct media_v2_topology topology = {0}; + struct media_v2_interface *interfaces = NULL; + struct udev *udev = udev_device_get_udev(device); + struct udev_device *video_device; + dev_t devnum; + + const char *path = udev_device_get_devnode(device); + if (!path) { + av_log(avctx, AV_LOG_ERROR, "%s: get media device devnode failed\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + ctx->media_fd = open(path, O_RDWR, 0); + if (ctx->media_fd < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: opening %s failed, %s (%d)\n", __func__, path, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = ioctl(ctx->media_fd, MEDIA_IOC_DEVICE_INFO, &device_info); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get media device info failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + av_log(avctx, AV_LOG_DEBUG, "%s: ctx=%p path=%s driver=%s\n", __func__, ctx, path, device_info.driver); + + ret = ioctl(ctx->media_fd, MEDIA_IOC_G_TOPOLOGY, &topology); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get media topology failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + if (topology.num_interfaces <= 0) { + av_log(avctx, AV_LOG_ERROR, "%s: media device has no interfaces\n", __func__); + ret = AVERROR(EINVAL); + goto fail; + } + + interfaces = av_mallocz(topology.num_interfaces * sizeof(struct media_v2_interface)); + if (!interfaces) { + av_log(avctx, AV_LOG_ERROR, "%s: allocating media interface struct failed\n", __func__); + ret = AVERROR(ENOMEM); + goto fail; + } + + topology.ptr_interfaces = (__u64)(uintptr_t)interfaces; + ret = ioctl(ctx->media_fd, MEDIA_IOC_G_TOPOLOGY, &topology); + if (ret < 0) { + av_log(avctx, AV_LOG_ERROR, "%s: get media topology failed, %s (%d)\n", __func__, strerror(errno), errno); + ret = AVERROR(EINVAL); + goto fail; + } + + ret = AVERROR(EINVAL); + for (int i = 0; i < topology.num_interfaces; i++) { + if (interfaces[i].intf_type != MEDIA_INTF_T_V4L_VIDEO) + continue; + + devnum = makedev(interfaces[i].devnode.major, interfaces[i].devnode.minor); + video_device = udev_device_new_from_devnum(udev, 'c', devnum); + if (!video_device) { + av_log(avctx, AV_LOG_ERROR, "%s: video_device=%p\n", __func__, video_device); + continue; + } + + ret = v4l2_request_probe_video_device(video_device, avctx, pixelformat, buffersize, control, count); + udev_device_unref(video_device); + + if (!ret) { + av_freep(&interfaces); + av_log(avctx, AV_LOG_INFO, "Using V4L2 media device %s (%s) for %s\n", path, device_info.driver, av_fourcc2str(pixelformat)); + return 0; + } + } + +fail: + av_freep(&interfaces); + if (ctx->media_fd >= 0) { + close(ctx->media_fd); + ctx->media_fd = -1; + } + return ret; +} + +int ff_v4l2_request_init(AVCodecContext *avctx, uint32_t pixelformat, uint32_t buffersize, struct v4l2_ext_control *control, int count) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret = AVERROR(EINVAL); + struct udev *udev; + struct udev_enumerate *enumerate; + struct udev_list_entry *devices; + struct udev_list_entry *entry; + struct udev_device *device; + + av_log(avctx, AV_LOG_DEBUG, "%s: ctx=%p hw_device_ctx=%p hw_frames_ctx=%p\n", __func__, ctx, avctx->hw_device_ctx, avctx->hw_frames_ctx); + + ctx->av_class = &v4l2_request_context_class; + ctx->media_fd = -1; + ctx->video_fd = -1; + + udev = udev_new(); + if (!udev) { + av_log(avctx, AV_LOG_ERROR, "%s: allocating udev context failed\n", __func__); + ret = AVERROR(ENOMEM); + goto fail; + } + + enumerate = udev_enumerate_new(udev); + if (!enumerate) { + av_log(avctx, AV_LOG_ERROR, "%s: allocating udev enumerator failed\n", __func__); + ret = AVERROR(ENOMEM); + goto fail; + } + + udev_enumerate_add_match_subsystem(enumerate, "media"); + udev_enumerate_scan_devices(enumerate); + + devices = udev_enumerate_get_list_entry(enumerate); + udev_list_entry_foreach(entry, devices) { + const char *path = udev_list_entry_get_name(entry); + if (!path) + continue; + + device = udev_device_new_from_syspath(udev, path); + if (!device) + continue; + + ret = v4l2_request_probe_media_device(device, avctx, pixelformat, buffersize, control, count); + udev_device_unref(device); + + if (!ret) + break; + } + + udev_enumerate_unref(enumerate); + + if (!ret) + ret = v4l2_request_init_context(avctx); + else + av_log(avctx, AV_LOG_INFO, "No V4L2 media device found for %s\n", av_fourcc2str(pixelformat)); + +fail: + udev_unref(udev); + return ret; +} + +int ff_v4l2_request_uninit(AVCodecContext *avctx) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + int ret; + + av_log(ctx, AV_LOG_DEBUG, "%s: avctx=%p\n", __func__, avctx); + + if (ctx->video_fd >= 0) { + ret = ioctl(ctx->video_fd, VIDIOC_STREAMOFF, &ctx->output_type); + if (ret < 0) + av_log(avctx, AV_LOG_ERROR, "%s: output stream off failed, %s (%d)\n", __func__, strerror(errno), errno); + + ret = ioctl(ctx->video_fd, VIDIOC_STREAMOFF, &ctx->format.type); + if (ret < 0) + av_log(avctx, AV_LOG_ERROR, "%s: capture stream off failed, %s (%d)\n", __func__, strerror(errno), errno); + } + + if (avctx->hw_frames_ctx) { + AVHWFramesContext *hwfc = (AVHWFramesContext*)avctx->hw_frames_ctx->data; + av_buffer_pool_flush(hwfc->pool); + } + + if (ctx->video_fd >= 0) + close(ctx->video_fd); + + if (ctx->media_fd >= 0) + close(ctx->media_fd); + + return 0; +} + +static int v4l2_request_buffer_alloc(V4L2RequestContext *ctx, V4L2RequestBuffer *buf, enum v4l2_buf_type type) +{ + int ret; + struct v4l2_plane planes[1] = {}; + struct v4l2_create_buffers buffers = { + .count = 1, + .memory = V4L2_MEMORY_MMAP, + .format.type = type, + }; + + av_log(ctx, AV_LOG_DEBUG, "%s: buf=%p type=%u\n", __func__, buf, type); + + ret = ioctl(ctx->video_fd, VIDIOC_G_FMT, &buffers.format); + if (ret < 0) { + av_log(ctx, AV_LOG_ERROR, "%s: get format failed for type %u, %s (%d)\n", __func__, type, strerror(errno), errno); + return ret; + } + + if (V4L2_TYPE_IS_MULTIPLANAR(buffers.format.type)) { + struct v4l2_pix_format_mplane *fmt = &buffers.format.fmt.pix_mp; + av_log(ctx, AV_LOG_DEBUG, "%s: pixelformat=%s width=%u height=%u bytesperline=%u sizeimage=%u num_planes=%u\n", __func__, + av_fourcc2str(fmt->pixelformat), fmt->width, fmt->height, fmt->plane_fmt[0].bytesperline, fmt->plane_fmt[0].sizeimage, fmt->num_planes); + } else { + struct v4l2_pix_format *fmt = &buffers.format.fmt.pix; + av_log(ctx, AV_LOG_DEBUG, "%s: pixelformat=%s width=%u height=%u bytesperline=%u sizeimage=%u\n", __func__, + av_fourcc2str(fmt->pixelformat), fmt->width, fmt->height, fmt->bytesperline, fmt->sizeimage); + } + + ret = ioctl(ctx->video_fd, VIDIOC_CREATE_BUFS, &buffers); + if (ret < 0) { + av_log(ctx, AV_LOG_ERROR, "%s: create buffers failed for type %u, %s (%d)\n", __func__, type, strerror(errno), errno); + return ret; + } + + if (V4L2_TYPE_IS_MULTIPLANAR(type)) { + buf->width = buffers.format.fmt.pix_mp.width; + buf->height = buffers.format.fmt.pix_mp.height; + buf->size = buffers.format.fmt.pix_mp.plane_fmt[0].sizeimage; + buf->buffer.length = 1; + buf->buffer.m.planes = planes; + } else { + buf->width = buffers.format.fmt.pix.width; + buf->height = buffers.format.fmt.pix.height; + buf->size = buffers.format.fmt.pix.sizeimage; + } + + buf->index = buffers.index; + buf->capabilities = buffers.capabilities; + buf->used = 0; + + buf->buffer.type = type; + buf->buffer.memory = V4L2_MEMORY_MMAP; + buf->buffer.index = buf->index; + + ret = ioctl(ctx->video_fd, VIDIOC_QUERYBUF, &buf->buffer); + if (ret < 0) { + av_log(ctx, AV_LOG_ERROR, "%s: query buffer %d failed, %s (%d)\n", __func__, buf->index, strerror(errno), errno); + return ret; + } + + buf->buffer.timestamp.tv_usec = buf->index + 1; + + if (V4L2_TYPE_IS_OUTPUT(type)) { + uint32_t offset = V4L2_TYPE_IS_MULTIPLANAR(type) ? buf->buffer.m.planes[0].m.mem_offset : buf->buffer.m.offset; + void *addr = mmap(NULL, buf->size, PROT_READ | PROT_WRITE, MAP_SHARED, ctx->video_fd, offset); + if (addr == MAP_FAILED) { + av_log(ctx, AV_LOG_ERROR, "%s: mmap failed, %s (%d)\n", __func__, strerror(errno), errno); + return -1; + } + + buf->addr = (uint8_t*)addr; + } else { + struct v4l2_exportbuffer exportbuffer = { + .type = type, + .index = buf->index, + .flags = O_RDONLY, + }; + + ret = ioctl(ctx->video_fd, VIDIOC_EXPBUF, &exportbuffer); + if (ret < 0) { + av_log(ctx, AV_LOG_ERROR, "%s: export buffer %d failed, %s (%d)\n", __func__, buf->index, strerror(errno), errno); + return ret; + } + + buf->fd = exportbuffer.fd; + } + + av_log(ctx, AV_LOG_DEBUG, "%s: buf=%p index=%d fd=%d addr=%p width=%u height=%u size=%u\n", __func__, + buf, buf->index, buf->fd, buf->addr, buf->width, buf->height, buf->size); + return 0; +} + +static void v4l2_request_buffer_free(V4L2RequestBuffer *buf) +{ + av_log(NULL, AV_LOG_DEBUG, "%s: buf=%p index=%d fd=%d addr=%p width=%u height=%u size=%u\n", __func__, + buf, buf->index, buf->fd, buf->addr, buf->width, buf->height, buf->size); + + if (buf->addr) + munmap(buf->addr, buf->size); + + if (buf->fd >= 0) + close(buf->fd); +} + +static void v4l2_request_frame_free(void *opaque, uint8_t *data) +{ + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)data; + + av_log(NULL, AV_LOG_DEBUG, "%s: opaque=%p data=%p request_fd=%d\n", __func__, opaque, data, req->request_fd); + + if (req->request_fd >= 0) + close(req->request_fd); + + v4l2_request_buffer_free(&req->capture); + v4l2_request_buffer_free(&req->output); + + av_free(data); +} + +static AVBufferRef *v4l2_request_frame_alloc(void *opaque, size_t size) +{ + V4L2RequestContext *ctx = opaque; + V4L2RequestDescriptor *req; + AVBufferRef *ref; + uint8_t *data; + int ret; + + data = av_mallocz(size); + if (!data) + return NULL; + + av_log(ctx, AV_LOG_DEBUG, "%s: size=%zu data=%p\n", __func__, size, data); + + ref = av_buffer_create(data, size, v4l2_request_frame_free, ctx, 0); + if (!ref) { + av_freep(&data); + return NULL; + } + + req = (V4L2RequestDescriptor*)data; + req->request_fd = -1; + req->output.fd = -1; + req->capture.fd = -1; + + ret = v4l2_request_buffer_alloc(ctx, &req->output, ctx->output_type); + if (ret < 0) { + av_buffer_unref(&ref); + return NULL; + } + + ret = v4l2_request_buffer_alloc(ctx, &req->capture, ctx->format.type); + if (ret < 0) { + av_buffer_unref(&ref); + return NULL; + } + + ret = ioctl(ctx->media_fd, MEDIA_IOC_REQUEST_ALLOC, &req->request_fd); + if (ret < 0) { + av_log(ctx, AV_LOG_ERROR, "%s: request alloc failed, %s (%d)\n", __func__, strerror(errno), errno); + av_buffer_unref(&ref); + return NULL; + } + + av_log(ctx, AV_LOG_DEBUG, "%s: size=%zu data=%p request_fd=%d\n", __func__, size, data, req->request_fd); + return ref; +} + +static void v4l2_request_pool_free(void *opaque) +{ + av_log(NULL, AV_LOG_DEBUG, "%s: opaque=%p\n", __func__, opaque); +} + +static void v4l2_request_hwframe_ctx_free(AVHWFramesContext *hwfc) +{ + av_log(NULL, AV_LOG_DEBUG, "%s: hwfc=%p pool=%p\n", __func__, hwfc, hwfc->pool); + + av_buffer_pool_flush(hwfc->pool); + av_buffer_pool_uninit(&hwfc->pool); +} + +int ff_v4l2_request_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx) +{ + V4L2RequestContext *ctx = avctx->internal->hwaccel_priv_data; + AVHWFramesContext *hwfc = (AVHWFramesContext*)hw_frames_ctx->data; + uint32_t pixelformat; + + hwfc->format = AV_PIX_FMT_DRM_PRIME; + hwfc->sw_format = AV_PIX_FMT_NONE; + + if (V4L2_TYPE_IS_MULTIPLANAR(ctx->format.type)) { + hwfc->width = ctx->format.fmt.pix_mp.width; + hwfc->height = ctx->format.fmt.pix_mp.height; + pixelformat = ctx->format.fmt.pix_mp.pixelformat; + } else { + hwfc->width = ctx->format.fmt.pix.width; + hwfc->height = ctx->format.fmt.pix.height; + pixelformat = ctx->format.fmt.pix.pixelformat; + } + + for (int i = 0; i < FF_ARRAY_ELEMS(v4l2_request_capture_pixelformats); i++) { + if (pixelformat == v4l2_request_capture_pixelformats[i].pixelformat) { + hwfc->sw_format = v4l2_request_capture_pixelformats[i].sw_format; + break; + } + } + + hwfc->pool = av_buffer_pool_init2(sizeof(V4L2RequestDescriptor), ctx, v4l2_request_frame_alloc, v4l2_request_pool_free); + if (!hwfc->pool) + return AVERROR(ENOMEM); + + hwfc->free = v4l2_request_hwframe_ctx_free; + + hwfc->initial_pool_size = 1; + + switch (avctx->codec_id) { + case AV_CODEC_ID_VP9: + hwfc->initial_pool_size += 8; + break; + case AV_CODEC_ID_VP8: + hwfc->initial_pool_size += 3; + break; + default: + hwfc->initial_pool_size += 2; + } + + av_log(ctx, AV_LOG_DEBUG, "%s: avctx=%p hw_frames_ctx=%p hwfc=%p pool=%p width=%d height=%d initial_pool_size=%d\n", __func__, + avctx, hw_frames_ctx, hwfc, hwfc->pool, hwfc->width, hwfc->height, hwfc->initial_pool_size); + return 0; +} diff --git a/libavcodec/v4l2_request.h b/libavcodec/v4l2_request.h new file mode 100644 index 0000000..5f105d8 --- /dev/null +++ b/libavcodec/v4l2_request.h @@ -0,0 +1,78 @@ +/* + * 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 AVCODEC_V4L2_REQUEST_H +#define AVCODEC_V4L2_REQUEST_H + +#include + +#include "libavutil/hwcontext_drm.h" + +typedef struct V4L2RequestContext { + const AVClass *av_class; + int video_fd; + int media_fd; + enum v4l2_buf_type output_type; + struct v4l2_format format; + int timestamp; +} V4L2RequestContext; + +typedef struct V4L2RequestBuffer { + int index; + int fd; + uint8_t *addr; + uint32_t width; + uint32_t height; + uint32_t size; + uint32_t used; + uint32_t capabilities; + struct v4l2_buffer buffer; +} V4L2RequestBuffer; + +typedef struct V4L2RequestDescriptor { + AVDRMFrameDescriptor drm; + int request_fd; + V4L2RequestBuffer output; + V4L2RequestBuffer capture; +} V4L2RequestDescriptor; + +uint64_t ff_v4l2_request_get_capture_timestamp(AVFrame *frame); + +int ff_v4l2_request_reset_frame(AVCodecContext *avctx, AVFrame *frame); + +int ff_v4l2_request_append_output_buffer(AVCodecContext *avctx, AVFrame *frame, const uint8_t *data, uint32_t size); + +int ff_v4l2_request_set_controls(AVCodecContext *avctx, struct v4l2_ext_control *control, int count); + +int ff_v4l2_request_get_controls(AVCodecContext *avctx, struct v4l2_ext_control *control, int count); + +int ff_v4l2_request_query_control(AVCodecContext *avctx, struct v4l2_query_ext_ctrl *control); + +int ff_v4l2_request_query_control_default_value(AVCodecContext *avctx, uint32_t id); + +int ff_v4l2_request_decode_slice(AVCodecContext *avctx, AVFrame *frame, struct v4l2_ext_control *control, int count, int first_slice, int last_slice); + +int ff_v4l2_request_decode_frame(AVCodecContext *avctx, AVFrame *frame, struct v4l2_ext_control *control, int count); + +int ff_v4l2_request_init(AVCodecContext *avctx, uint32_t pixelformat, uint32_t buffersize, struct v4l2_ext_control *control, int count); + +int ff_v4l2_request_uninit(AVCodecContext *avctx); + +int ff_v4l2_request_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx); + +#endif /* AVCODEC_V4L2_REQUEST_H */ -- 2.44.0 From 9c440906da4c4f0f63faa71a0b988d2447ca57bf Mon Sep 17 00:00:00 2001 From: Boris Brezillon Date: Wed, 22 May 2019 14:44:22 +0200 Subject: [PATCH 03/18] h264dec: add ref_pic_marking and pic_order_cnt bit_size to slice context Used by V4L2 request API h264 hwaccel Signed-off-by: Boris Brezillon Signed-off-by: Jonas Karlman --- libavcodec/h264_slice.c | 6 +++++- libavcodec/h264dec.h | 2 ++ 2 files changed, 7 insertions(+), 1 deletion(-) diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index 5c17edf..ea7e882 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -1685,7 +1685,7 @@ static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl, unsigned int slice_type, tmp, i; int field_pic_flag, bottom_field_flag; int first_slice = sl == h->slice_ctx && !h->current_slice; - int picture_structure; + int picture_structure, pos; if (first_slice) av_assert0(!h->setup_finished); @@ -1776,6 +1776,7 @@ static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl, sl->poc_lsb = 0; sl->delta_poc_bottom = 0; + pos = sl->gb.index; if (sps->poc_type == 0) { sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb); @@ -1790,6 +1791,7 @@ static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl, if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME) sl->delta_poc[1] = get_se_golomb(&sl->gb); } + sl->pic_order_cnt_bit_size = sl->gb.index - pos; sl->redundant_pic_count = 0; if (pps->redundant_pic_cnt_present) @@ -1829,9 +1831,11 @@ static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl, sl->explicit_ref_marking = 0; if (nal->ref_idc) { + pos = sl->gb.index; ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx); if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE)) return AVERROR_INVALIDDATA; + sl->ref_pic_marking_bit_size = sl->gb.index - pos; } if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) { diff --git a/libavcodec/h264dec.h b/libavcodec/h264dec.h index 675be9b..a9d515d 100644 --- a/libavcodec/h264dec.h +++ b/libavcodec/h264dec.h @@ -315,6 +315,7 @@ typedef struct H264SliceContext { MMCO mmco[H264_MAX_MMCO_COUNT]; int nb_mmco; int explicit_ref_marking; + int ref_pic_marking_bit_size; int frame_num; int idr_pic_id; @@ -323,6 +324,7 @@ typedef struct H264SliceContext { int delta_poc[2]; int curr_pic_num; int max_pic_num; + int pic_order_cnt_bit_size; } H264SliceContext; /** -- 2.44.0 From d507e050b4dbab847abcc3d97c902f8e01d39945 Mon Sep 17 00:00:00 2001 From: Jernej Skrabec Date: Sat, 15 Dec 2018 22:32:16 +0100 Subject: [PATCH 04/18] Add V4L2 request API h264 hwaccel Signed-off-by: Jernej Skrabec Signed-off-by: Jonas Karlman --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/h264_slice.c | 4 + libavcodec/h264dec.c | 3 + libavcodec/hwaccels.h | 1 + libavcodec/v4l2_request_h264.c | 462 +++++++++++++++++++++++++++++++++ 6 files changed, 474 insertions(+) create mode 100644 libavcodec/v4l2_request_h264.c diff --git a/configure b/configure index 503eb06..252d188 100755 --- a/configure +++ b/configure @@ -3081,6 +3081,8 @@ h264_dxva2_hwaccel_deps="dxva2" h264_dxva2_hwaccel_select="h264_decoder" h264_nvdec_hwaccel_deps="nvdec" h264_nvdec_hwaccel_select="h264_decoder" +h264_v4l2request_hwaccel_deps="v4l2_request h264_v4l2_request" +h264_v4l2request_hwaccel_select="h264_decoder" h264_vaapi_hwaccel_deps="vaapi" h264_vaapi_hwaccel_select="h264_decoder" h264_vdpau_hwaccel_deps="vdpau" @@ -7003,6 +7005,7 @@ if enabled v4l2_m2m; then fi check_func_headers "linux/media.h linux/videodev2.h" v4l2_timeval_to_ns +check_cc h264_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_H264_SLICE;" check_headers sys/videoio.h test_code cc sys/videoio.h "struct v4l2_frmsizeenum vfse; vfse.discrete.width = 0;" && enable_sanitized struct_v4l2_frmivalenum_discrete diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 5e626c9..4bfd477 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1001,6 +1001,7 @@ OBJS-$(CONFIG_H264_D3D11VA_HWACCEL) += dxva2_h264.o OBJS-$(CONFIG_H264_DXVA2_HWACCEL) += dxva2_h264.o OBJS-$(CONFIG_H264_NVDEC_HWACCEL) += nvdec_h264.o OBJS-$(CONFIG_H264_QSV_HWACCEL) += qsvdec.o +OBJS-$(CONFIG_H264_V4L2REQUEST_HWACCEL) += v4l2_request_h264.o OBJS-$(CONFIG_H264_VAAPI_HWACCEL) += vaapi_h264.o OBJS-$(CONFIG_H264_VDPAU_HWACCEL) += vdpau_h264.o OBJS-$(CONFIG_H264_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index ea7e882..236bdc6 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -786,6 +786,7 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \ (CONFIG_H264_D3D11VA_HWACCEL * 2) + \ CONFIG_H264_NVDEC_HWACCEL + \ + CONFIG_H264_V4L2REQUEST_HWACCEL + \ CONFIG_H264_VAAPI_HWACCEL + \ CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \ CONFIG_H264_VDPAU_HWACCEL + \ @@ -890,6 +891,9 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) #endif #if CONFIG_H264_VAAPI_HWACCEL *fmt++ = AV_PIX_FMT_VAAPI; +#endif +#if CONFIG_H264_V4L2REQUEST_HWACCEL + *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif if (h->avctx->color_range == AVCOL_RANGE_JPEG) *fmt++ = AV_PIX_FMT_YUVJ420P; diff --git a/libavcodec/h264dec.c b/libavcodec/h264dec.c index f346c65..8d6334d 100644 --- a/libavcodec/h264dec.c +++ b/libavcodec/h264dec.c @@ -1135,6 +1135,9 @@ const FFCodec ff_h264_decoder = { #endif #if CONFIG_H264_VULKAN_HWACCEL HWACCEL_VULKAN(h264), +#endif +#if CONFIG_H264_V4L2REQUEST_HWACCEL + HWACCEL_V4L2REQUEST(h264), #endif NULL }, diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index c463071..5211553 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -32,6 +32,7 @@ extern const struct FFHWAccel ff_h264_d3d11va_hwaccel; extern const struct FFHWAccel ff_h264_d3d11va2_hwaccel; extern const struct FFHWAccel ff_h264_dxva2_hwaccel; extern const struct FFHWAccel ff_h264_nvdec_hwaccel; +extern const struct FFHWAccel ff_h264_v4l2request_hwaccel; extern const struct FFHWAccel ff_h264_vaapi_hwaccel; extern const struct FFHWAccel ff_h264_vdpau_hwaccel; extern const struct FFHWAccel ff_h264_videotoolbox_hwaccel; diff --git a/libavcodec/v4l2_request_h264.c b/libavcodec/v4l2_request_h264.c new file mode 100644 index 0000000..980cd93 --- /dev/null +++ b/libavcodec/v4l2_request_h264.c @@ -0,0 +1,462 @@ +/* + * 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 "h264dec.h" +#include "hwaccel_internal.h" +#include "hwconfig.h" +#include "internal.h" +#include "v4l2_request.h" + +typedef struct V4L2RequestControlsH264 { + struct v4l2_ctrl_h264_sps sps; + struct v4l2_ctrl_h264_pps pps; + struct v4l2_ctrl_h264_scaling_matrix scaling_matrix; + struct v4l2_ctrl_h264_decode_params decode_params; + struct v4l2_ctrl_h264_slice_params slice_params; + struct v4l2_ctrl_h264_pred_weights pred_weights; + int pred_weights_required; + int first_slice; + int num_slices; +} V4L2RequestControlsH264; + +typedef struct V4L2RequestContextH264 { + V4L2RequestContext base; + int decode_mode; + int start_code; +} V4L2RequestContextH264; + +static uint8_t nalu_slice_start_code[] = { 0x00, 0x00, 0x01 }; + +static void fill_weight_factors(struct v4l2_h264_weight_factors *factors, int list, const H264SliceContext *sl) +{ + for (int i = 0; i < sl->ref_count[list]; i++) { + if (sl->pwt.luma_weight_flag[list]) { + factors->luma_weight[i] = sl->pwt.luma_weight[i][list][0]; + factors->luma_offset[i] = sl->pwt.luma_weight[i][list][1]; + } else { + factors->luma_weight[i] = 1 << sl->pwt.luma_log2_weight_denom; + factors->luma_offset[i] = 0; + } + for (int j = 0; j < 2; j++) { + if (sl->pwt.chroma_weight_flag[list]) { + factors->chroma_weight[i][j] = sl->pwt.chroma_weight[i][list][j][0]; + factors->chroma_offset[i][j] = sl->pwt.chroma_weight[i][list][j][1]; + } else { + factors->chroma_weight[i][j] = 1 << sl->pwt.chroma_log2_weight_denom; + factors->chroma_offset[i][j] = 0; + } + } + } +} + +static void fill_dpb_entry(struct v4l2_h264_dpb_entry *entry, const H264Picture *pic, int long_idx) +{ + entry->reference_ts = ff_v4l2_request_get_capture_timestamp(pic->f); + entry->pic_num = pic->pic_id; + entry->frame_num = pic->long_ref ? long_idx : pic->frame_num; + entry->fields = pic->reference & V4L2_H264_FRAME_REF; + entry->flags = V4L2_H264_DPB_ENTRY_FLAG_VALID; + if (entry->fields) + entry->flags |= V4L2_H264_DPB_ENTRY_FLAG_ACTIVE; + if (pic->long_ref) + entry->flags |= V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM; + if (pic->field_picture) + entry->flags |= V4L2_H264_DPB_ENTRY_FLAG_FIELD; + if (pic->field_poc[0] != INT_MAX) + entry->top_field_order_cnt = pic->field_poc[0]; + if (pic->field_poc[1] != INT_MAX) + entry->bottom_field_order_cnt = pic->field_poc[1]; +} + +static void fill_dpb(struct v4l2_ctrl_h264_decode_params *decode, const H264Context *h) +{ + int entries = 0; + + for (int i = 0; i < h->short_ref_count; i++) { + const H264Picture *pic = h->short_ref[i]; + if (pic && (pic->field_poc[0] != INT_MAX || pic->field_poc[1] != INT_MAX)) + fill_dpb_entry(&decode->dpb[entries++], pic, pic->pic_id); + } + + if (!h->long_ref_count) + return; + + for (int i = 0; i < FF_ARRAY_ELEMS(h->long_ref); i++) { + const H264Picture *pic = h->long_ref[i]; + if (pic && (pic->field_poc[0] != INT_MAX || pic->field_poc[1] != INT_MAX)) + fill_dpb_entry(&decode->dpb[entries++], pic, i); + } +} + +static void fill_ref_list(struct v4l2_h264_reference *reference, struct v4l2_ctrl_h264_decode_params *decode, const H264Ref *ref) +{ + uint64_t timestamp; + + if (!ref->parent) + return; + + timestamp = ff_v4l2_request_get_capture_timestamp(ref->parent->f); + + for (uint8_t i = 0; i < FF_ARRAY_ELEMS(decode->dpb); i++) { + struct v4l2_h264_dpb_entry *entry = &decode->dpb[i]; + if ((entry->flags & V4L2_H264_DPB_ENTRY_FLAG_VALID) && + entry->reference_ts == timestamp) { + reference->fields = ref->reference & V4L2_H264_FRAME_REF; + reference->index = i; + return; + } + } +} + +static void fill_sps(struct v4l2_ctrl_h264_sps *ctrl, const H264Context *h) +{ + const SPS *sps = h->ps.sps; + + *ctrl = (struct v4l2_ctrl_h264_sps) { + .profile_idc = sps->profile_idc, + .constraint_set_flags = sps->constraint_set_flags, + .level_idc = sps->level_idc, + .seq_parameter_set_id = sps->sps_id, + .chroma_format_idc = sps->chroma_format_idc, + .bit_depth_luma_minus8 = sps->bit_depth_luma - 8, + .bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8, + .log2_max_frame_num_minus4 = sps->log2_max_frame_num - 4, + .pic_order_cnt_type = sps->poc_type, + .log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4, + .max_num_ref_frames = sps->ref_frame_count, + .num_ref_frames_in_pic_order_cnt_cycle = sps->poc_cycle_length, + .offset_for_non_ref_pic = sps->offset_for_non_ref_pic, + .offset_for_top_to_bottom_field = sps->offset_for_top_to_bottom_field, + .pic_width_in_mbs_minus1 = h->mb_width - 1, + .pic_height_in_map_units_minus1 = sps->frame_mbs_only_flag ? h->mb_height - 1 : h->mb_height / 2 - 1, + }; + + if (sps->poc_cycle_length > 0 && sps->poc_cycle_length <= 255) + memcpy(ctrl->offset_for_ref_frame, sps->offset_for_ref_frame, sps->poc_cycle_length * sizeof(ctrl->offset_for_ref_frame[0])); + + if (sps->residual_color_transform_flag) + ctrl->flags |= V4L2_H264_SPS_FLAG_SEPARATE_COLOUR_PLANE; + if (sps->transform_bypass) + ctrl->flags |= V4L2_H264_SPS_FLAG_QPPRIME_Y_ZERO_TRANSFORM_BYPASS; + if (sps->delta_pic_order_always_zero_flag) + ctrl->flags |= V4L2_H264_SPS_FLAG_DELTA_PIC_ORDER_ALWAYS_ZERO; + if (sps->gaps_in_frame_num_allowed_flag) + ctrl->flags |= V4L2_H264_SPS_FLAG_GAPS_IN_FRAME_NUM_VALUE_ALLOWED; + if (sps->frame_mbs_only_flag) + ctrl->flags |= V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY; + if (sps->mb_aff) + ctrl->flags |= V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD; + if (sps->direct_8x8_inference_flag) + ctrl->flags |= V4L2_H264_SPS_FLAG_DIRECT_8X8_INFERENCE; +} + +static void fill_pps(struct v4l2_ctrl_h264_pps *ctrl, const H264Context *h) +{ + const SPS *sps = h->ps.sps; + const PPS *pps = h->ps.pps; + const H264SliceContext *sl = &h->slice_ctx[0]; + int qp_bd_offset = 6 * (sps->bit_depth_luma - 8); + + *ctrl = (struct v4l2_ctrl_h264_pps) { + .pic_parameter_set_id = sl->pps_id, + .seq_parameter_set_id = pps->sps_id, + .num_slice_groups_minus1 = pps->slice_group_count - 1, + .num_ref_idx_l0_default_active_minus1 = pps->ref_count[0] - 1, + .num_ref_idx_l1_default_active_minus1 = pps->ref_count[1] - 1, + .weighted_bipred_idc = pps->weighted_bipred_idc, + .pic_init_qp_minus26 = pps->init_qp - 26 - qp_bd_offset, + .pic_init_qs_minus26 = pps->init_qs - 26 - qp_bd_offset, + .chroma_qp_index_offset = pps->chroma_qp_index_offset[0], + .second_chroma_qp_index_offset = pps->chroma_qp_index_offset[1], + }; + + if (pps->cabac) + ctrl->flags |= V4L2_H264_PPS_FLAG_ENTROPY_CODING_MODE; + if (pps->pic_order_present) + ctrl->flags |= V4L2_H264_PPS_FLAG_BOTTOM_FIELD_PIC_ORDER_IN_FRAME_PRESENT; + if (pps->weighted_pred) + ctrl->flags |= V4L2_H264_PPS_FLAG_WEIGHTED_PRED; + if (pps->deblocking_filter_parameters_present) + ctrl->flags |= V4L2_H264_PPS_FLAG_DEBLOCKING_FILTER_CONTROL_PRESENT; + if (pps->constrained_intra_pred) + ctrl->flags |= V4L2_H264_PPS_FLAG_CONSTRAINED_INTRA_PRED; + if (pps->redundant_pic_cnt_present) + ctrl->flags |= V4L2_H264_PPS_FLAG_REDUNDANT_PIC_CNT_PRESENT; + if (pps->transform_8x8_mode) + ctrl->flags |= V4L2_H264_PPS_FLAG_TRANSFORM_8X8_MODE; + + /* FFmpeg always provide a scaling matrix */ + ctrl->flags |= V4L2_H264_PPS_FLAG_SCALING_MATRIX_PRESENT; +} + +static int v4l2_request_h264_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const H264Context *h = avctx->priv_data; + const PPS *pps = h->ps.pps; + const SPS *sps = h->ps.sps; + const H264SliceContext *sl = &h->slice_ctx[0]; + V4L2RequestControlsH264 *controls = h->cur_pic_ptr->hwaccel_picture_private; + + fill_sps(&controls->sps, h); + fill_pps(&controls->pps, h); + + memcpy(controls->scaling_matrix.scaling_list_4x4, pps->scaling_matrix4, sizeof(controls->scaling_matrix.scaling_list_4x4)); + memcpy(controls->scaling_matrix.scaling_list_8x8[0], pps->scaling_matrix8[0], sizeof(controls->scaling_matrix.scaling_list_8x8[0])); + memcpy(controls->scaling_matrix.scaling_list_8x8[1], pps->scaling_matrix8[3], sizeof(controls->scaling_matrix.scaling_list_8x8[1])); + + if (sps->chroma_format_idc == 3) { + memcpy(controls->scaling_matrix.scaling_list_8x8[2], pps->scaling_matrix8[1], sizeof(controls->scaling_matrix.scaling_list_8x8[2])); + memcpy(controls->scaling_matrix.scaling_list_8x8[3], pps->scaling_matrix8[4], sizeof(controls->scaling_matrix.scaling_list_8x8[3])); + memcpy(controls->scaling_matrix.scaling_list_8x8[4], pps->scaling_matrix8[2], sizeof(controls->scaling_matrix.scaling_list_8x8[4])); + memcpy(controls->scaling_matrix.scaling_list_8x8[5], pps->scaling_matrix8[5], sizeof(controls->scaling_matrix.scaling_list_8x8[5])); + } + + controls->decode_params = (struct v4l2_ctrl_h264_decode_params) { + .nal_ref_idc = h->nal_ref_idc, + .frame_num = h->poc.frame_num, + .top_field_order_cnt = h->cur_pic_ptr->field_poc[0] != INT_MAX ? h->cur_pic_ptr->field_poc[0] : 0, + .bottom_field_order_cnt = h->cur_pic_ptr->field_poc[1] != INT_MAX ? h->cur_pic_ptr->field_poc[1] : 0, + .idr_pic_id = sl->idr_pic_id, + .pic_order_cnt_lsb = sl->poc_lsb, + .delta_pic_order_cnt_bottom = sl->delta_poc_bottom, + .delta_pic_order_cnt0 = sl->delta_poc[0], + .delta_pic_order_cnt1 = sl->delta_poc[1], + /* Size in bits of dec_ref_pic_marking() syntax element. */ + .dec_ref_pic_marking_bit_size = sl->ref_pic_marking_bit_size, + /* Size in bits of pic order count syntax. */ + .pic_order_cnt_bit_size = sl->pic_order_cnt_bit_size, + .slice_group_change_cycle = 0, /* slice group not supported by FFmpeg */ + }; + + if (h->picture_idr) + controls->decode_params.flags |= V4L2_H264_DECODE_PARAM_FLAG_IDR_PIC; + if (FIELD_PICTURE(h)) + controls->decode_params.flags |= V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC; + if (h->picture_structure == PICT_BOTTOM_FIELD) + controls->decode_params.flags |= V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD; + if (sl->slice_type_nos == AV_PICTURE_TYPE_P) + controls->decode_params.flags |= V4L2_H264_DECODE_PARAM_FLAG_PFRAME; + if (sl->slice_type_nos == AV_PICTURE_TYPE_B) + controls->decode_params.flags |= V4L2_H264_DECODE_PARAM_FLAG_BFRAME; + + fill_dpb(&controls->decode_params, h); + + controls->first_slice = 1; + controls->num_slices = 0; + + return ff_v4l2_request_reset_frame(avctx, h->cur_pic_ptr->f); +} + +static int v4l2_request_h264_queue_decode(AVCodecContext *avctx, int last_slice) +{ + const H264Context *h = avctx->priv_data; + V4L2RequestControlsH264 *controls = h->cur_pic_ptr->hwaccel_picture_private; + V4L2RequestContextH264 *ctx = avctx->internal->hwaccel_priv_data; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_H264_SPS, + .ptr = &controls->sps, + .size = sizeof(controls->sps), + }, + { + .id = V4L2_CID_STATELESS_H264_PPS, + .ptr = &controls->pps, + .size = sizeof(controls->pps), + }, + { + .id = V4L2_CID_STATELESS_H264_SCALING_MATRIX, + .ptr = &controls->scaling_matrix, + .size = sizeof(controls->scaling_matrix), + }, + { + .id = V4L2_CID_STATELESS_H264_DECODE_PARAMS, + .ptr = &controls->decode_params, + .size = sizeof(controls->decode_params), + }, + { + .id = V4L2_CID_STATELESS_H264_SLICE_PARAMS, + .ptr = &controls->slice_params, + .size = sizeof(controls->slice_params), + }, + { + .id = V4L2_CID_STATELESS_H264_PRED_WEIGHTS, + .ptr = &controls->pred_weights, + .size = sizeof(controls->pred_weights), + }, + }; + + if (ctx->decode_mode == V4L2_STATELESS_H264_DECODE_MODE_SLICE_BASED) { + int count = FF_ARRAY_ELEMS(control) - (controls->pred_weights_required ? 0 : 1); + return ff_v4l2_request_decode_slice(avctx, h->cur_pic_ptr->f, control, count, controls->first_slice, last_slice); + } + + return ff_v4l2_request_decode_frame(avctx, h->cur_pic_ptr->f, control, FF_ARRAY_ELEMS(control) - 2); +} + +static int v4l2_request_h264_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const H264Context *h = avctx->priv_data; + const PPS *pps = h->ps.pps; + const H264SliceContext *sl = &h->slice_ctx[0]; + V4L2RequestControlsH264 *controls = h->cur_pic_ptr->hwaccel_picture_private; + V4L2RequestContextH264 *ctx = avctx->internal->hwaccel_priv_data; + int i, ret, count; + + if (ctx->decode_mode == V4L2_STATELESS_H264_DECODE_MODE_SLICE_BASED && controls->num_slices) { + ret = v4l2_request_h264_queue_decode(avctx, 0); + if (ret) + return ret; + + ff_v4l2_request_reset_frame(avctx, h->cur_pic_ptr->f); + controls->first_slice = 0; + } + + if (ctx->start_code == V4L2_STATELESS_H264_START_CODE_ANNEX_B) { + ret = ff_v4l2_request_append_output_buffer(avctx, h->cur_pic_ptr->f, nalu_slice_start_code, 3); + if (ret) + return ret; + } + + ret = ff_v4l2_request_append_output_buffer(avctx, h->cur_pic_ptr->f, buffer, size); + if (ret) + return ret; + + if (ctx->decode_mode != V4L2_STATELESS_H264_DECODE_MODE_SLICE_BASED) + return 0; + + controls->slice_params = (struct v4l2_ctrl_h264_slice_params) { + /* Offset in bits to slice_data() from the beginning of this slice. */ + .header_bit_size = get_bits_count(&sl->gb), + + .first_mb_in_slice = sl->first_mb_addr, + + .slice_type = ff_h264_get_slice_type(sl), + .colour_plane_id = 0, /* separate colour plane not supported by FFmpeg */ + .redundant_pic_cnt = sl->redundant_pic_count, + .cabac_init_idc = sl->cabac_init_idc, + .slice_qp_delta = sl->qscale - pps->init_qp, + .slice_qs_delta = 0, /* not implemented by FFmpeg */ + .disable_deblocking_filter_idc = sl->deblocking_filter < 2 ? !sl->deblocking_filter : sl->deblocking_filter, + .slice_alpha_c0_offset_div2 = sl->slice_alpha_c0_offset / 2, + .slice_beta_offset_div2 = sl->slice_beta_offset / 2, + .num_ref_idx_l0_active_minus1 = sl->list_count > 0 ? sl->ref_count[0] - 1 : 0, + .num_ref_idx_l1_active_minus1 = sl->list_count > 1 ? sl->ref_count[1] - 1 : 0, + }; + + if (sl->slice_type == AV_PICTURE_TYPE_B && sl->direct_spatial_mv_pred) + controls->slice_params.flags |= V4L2_H264_SLICE_FLAG_DIRECT_SPATIAL_MV_PRED; + + /* V4L2_H264_SLICE_FLAG_SP_FOR_SWITCH: not implemented by FFmpeg */ + + controls->pred_weights_required = V4L2_H264_CTRL_PRED_WEIGHTS_REQUIRED(&controls->pps, &controls->slice_params); + if (controls->pred_weights_required) { + controls->pred_weights.chroma_log2_weight_denom = sl->pwt.chroma_log2_weight_denom; + controls->pred_weights.luma_log2_weight_denom = sl->pwt.luma_log2_weight_denom; + } + + count = sl->list_count > 0 ? sl->ref_count[0] : 0; + for (i = 0; i < count; i++) + fill_ref_list(&controls->slice_params.ref_pic_list0[i], &controls->decode_params, &sl->ref_list[0][i]); + if (count && controls->pred_weights_required) + fill_weight_factors(&controls->pred_weights.weight_factors[0], 0, sl); + + count = sl->list_count > 1 ? sl->ref_count[1] : 0; + for (i = 0; i < count; i++) + fill_ref_list(&controls->slice_params.ref_pic_list1[i], &controls->decode_params, &sl->ref_list[1][i]); + if (count && controls->pred_weights_required) + fill_weight_factors(&controls->pred_weights.weight_factors[1], 1, sl); + + controls->num_slices++; + return 0; +} + +static int v4l2_request_h264_end_frame(AVCodecContext *avctx) +{ + return v4l2_request_h264_queue_decode(avctx, 1); +} + +static int v4l2_request_h264_set_controls(AVCodecContext *avctx) +{ + V4L2RequestContextH264 *ctx = avctx->internal->hwaccel_priv_data; + + struct v4l2_ext_control control[] = { + { .id = V4L2_CID_STATELESS_H264_DECODE_MODE, }, + { .id = V4L2_CID_STATELESS_H264_START_CODE, }, + }; + + ctx->decode_mode = ff_v4l2_request_query_control_default_value(avctx, V4L2_CID_STATELESS_H264_DECODE_MODE); + if (ctx->decode_mode != V4L2_STATELESS_H264_DECODE_MODE_SLICE_BASED && + ctx->decode_mode != V4L2_STATELESS_H264_DECODE_MODE_FRAME_BASED) { + av_log(avctx, AV_LOG_ERROR, "%s: unsupported decode mode, %d\n", __func__, ctx->decode_mode); + return AVERROR(EINVAL); + } + + ctx->start_code = ff_v4l2_request_query_control_default_value(avctx, V4L2_CID_STATELESS_H264_START_CODE); + if (ctx->start_code != V4L2_STATELESS_H264_START_CODE_NONE && + ctx->start_code != V4L2_STATELESS_H264_START_CODE_ANNEX_B) { + av_log(avctx, AV_LOG_ERROR, "%s: unsupported start code, %d\n", __func__, ctx->start_code); + return AVERROR(EINVAL); + } + + control[0].value = ctx->decode_mode; + control[1].value = ctx->start_code; + + return ff_v4l2_request_set_controls(avctx, control, FF_ARRAY_ELEMS(control)); +} + +static int v4l2_request_h264_init(AVCodecContext *avctx) +{ + const H264Context *h = avctx->priv_data; + struct v4l2_ctrl_h264_sps sps; + int ret; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_H264_SPS, + .ptr = &sps, + .size = sizeof(sps), + }, + }; + + fill_sps(&sps, h); + + ret = ff_v4l2_request_init(avctx, V4L2_PIX_FMT_H264_SLICE, 4 * 1024 * 1024, control, FF_ARRAY_ELEMS(control)); + if (ret) + return ret; + + return v4l2_request_h264_set_controls(avctx); +} + +const FFHWAccel ff_h264_v4l2request_hwaccel = { + .p.name = "h264_v4l2request", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_H264, + .p.pix_fmt = AV_PIX_FMT_DRM_PRIME, + .start_frame = v4l2_request_h264_start_frame, + .decode_slice = v4l2_request_h264_decode_slice, + .end_frame = v4l2_request_h264_end_frame, + .frame_priv_data_size = sizeof(V4L2RequestControlsH264), + .init = v4l2_request_h264_init, + .uninit = ff_v4l2_request_uninit, + .priv_data_size = sizeof(V4L2RequestContextH264), + .frame_params = ff_v4l2_request_frame_params, + .caps_internal = HWACCEL_CAP_ASYNC_SAFE, +}; -- 2.44.0 From f125ce861fcf1f4dfba26304f72a46dc15553d5a Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 15 Dec 2018 22:32:16 +0100 Subject: [PATCH 05/18] Add V4L2 request API mpeg2 hwaccel Signed-off-by: Jonas Karlman --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/hwaccels.h | 1 + libavcodec/mpeg12dec.c | 6 ++ libavcodec/v4l2_request_mpeg2.c | 160 ++++++++++++++++++++++++++++++++ 5 files changed, 171 insertions(+) create mode 100644 libavcodec/v4l2_request_mpeg2.c diff --git a/configure b/configure index 252d188..3127877 100755 --- a/configure +++ b/configure @@ -3125,6 +3125,8 @@ mpeg2_dxva2_hwaccel_deps="dxva2" mpeg2_dxva2_hwaccel_select="mpeg2video_decoder" mpeg2_nvdec_hwaccel_deps="nvdec" mpeg2_nvdec_hwaccel_select="mpeg2video_decoder" +mpeg2_v4l2request_hwaccel_deps="v4l2_request mpeg2_v4l2_request" +mpeg2_v4l2request_hwaccel_select="mpeg2video_decoder" mpeg2_vaapi_hwaccel_deps="vaapi" mpeg2_vaapi_hwaccel_select="mpeg2video_decoder" mpeg2_vdpau_hwaccel_deps="vdpau" @@ -7006,6 +7008,7 @@ fi check_func_headers "linux/media.h linux/videodev2.h" v4l2_timeval_to_ns check_cc h264_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_H264_SLICE;" +check_cc mpeg2_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_MPEG2_SLICE;" check_headers sys/videoio.h test_code cc sys/videoio.h "struct v4l2_frmsizeenum vfse; vfse.discrete.width = 0;" && enable_sanitized struct_v4l2_frmivalenum_discrete diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 4bfd477..668f06a 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1022,6 +1022,7 @@ OBJS-$(CONFIG_MPEG2_D3D11VA_HWACCEL) += dxva2_mpeg2.o OBJS-$(CONFIG_MPEG2_DXVA2_HWACCEL) += dxva2_mpeg2.o OBJS-$(CONFIG_MPEG2_NVDEC_HWACCEL) += nvdec_mpeg12.o OBJS-$(CONFIG_MPEG2_QSV_HWACCEL) += qsvdec.o +OBJS-$(CONFIG_MPEG2_V4L2REQUEST_HWACCEL) += v4l2_request_mpeg2.o OBJS-$(CONFIG_MPEG2_VAAPI_HWACCEL) += vaapi_mpeg2.o OBJS-$(CONFIG_MPEG2_VDPAU_HWACCEL) += vdpau_mpeg12.o OBJS-$(CONFIG_MPEG2_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 5211553..f5baf8a 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -54,6 +54,7 @@ extern const struct FFHWAccel ff_mpeg2_d3d11va_hwaccel; extern const struct FFHWAccel ff_mpeg2_d3d11va2_hwaccel; extern const struct FFHWAccel ff_mpeg2_dxva2_hwaccel; extern const struct FFHWAccel ff_mpeg2_nvdec_hwaccel; +extern const struct FFHWAccel ff_mpeg2_v4l2request_hwaccel; extern const struct FFHWAccel ff_mpeg2_vaapi_hwaccel; extern const struct FFHWAccel ff_mpeg2_vdpau_hwaccel; extern const struct FFHWAccel ff_mpeg2_videotoolbox_hwaccel; diff --git a/libavcodec/mpeg12dec.c b/libavcodec/mpeg12dec.c index 5dac83e..4aa341c 100644 --- a/libavcodec/mpeg12dec.c +++ b/libavcodec/mpeg12dec.c @@ -1120,6 +1120,9 @@ static const enum AVPixelFormat mpeg2_hwaccel_pixfmt_list_420[] = { #endif #if CONFIG_MPEG2_VIDEOTOOLBOX_HWACCEL AV_PIX_FMT_VIDEOTOOLBOX, +#endif +#if CONFIG_MPEG2_V4L2REQUEST_HWACCEL + AV_PIX_FMT_DRM_PRIME, #endif AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE @@ -2896,6 +2899,9 @@ const FFCodec ff_mpeg2video_decoder = { #endif #if CONFIG_MPEG2_VIDEOTOOLBOX_HWACCEL HWACCEL_VIDEOTOOLBOX(mpeg2), +#endif +#if CONFIG_MPEG2_V4L2REQUEST_HWACCEL + HWACCEL_V4L2REQUEST(mpeg2), #endif NULL }, diff --git a/libavcodec/v4l2_request_mpeg2.c b/libavcodec/v4l2_request_mpeg2.c new file mode 100644 index 0000000..ae3f043 --- /dev/null +++ b/libavcodec/v4l2_request_mpeg2.c @@ -0,0 +1,160 @@ +/* + * 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 "hwaccel_internal.h" +#include "hwconfig.h" +#include "mpegvideo.h" +#include "v4l2_request.h" + +typedef struct V4L2RequestControlsMPEG2 { + struct v4l2_ctrl_mpeg2_sequence sequence; + struct v4l2_ctrl_mpeg2_picture picture; + struct v4l2_ctrl_mpeg2_quantisation quantisation; +} V4L2RequestControlsMPEG2; + +static int v4l2_request_mpeg2_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const MpegEncContext *s = avctx->priv_data; + V4L2RequestControlsMPEG2 *controls = s->current_picture_ptr->hwaccel_picture_private; + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)s->current_picture_ptr->f->data[0]; + + controls->sequence = (struct v4l2_ctrl_mpeg2_sequence) { + /* ISO/IEC 13818-2, ITU-T Rec. H.262: Sequence header */ + .horizontal_size = s->width, + .vertical_size = s->height, + .vbv_buffer_size = req->output.size, + + /* ISO/IEC 13818-2, ITU-T Rec. H.262: Sequence extension */ + .profile_and_level_indication = 0, + .chroma_format = s->chroma_format, + }; + + if (s->progressive_sequence) + controls->sequence.flags |= V4L2_MPEG2_SEQ_FLAG_PROGRESSIVE; + + controls->picture = (struct v4l2_ctrl_mpeg2_picture) { + /* ISO/IEC 13818-2, ITU-T Rec. H.262: Picture header */ + .picture_coding_type = s->pict_type, + + /* ISO/IEC 13818-2, ITU-T Rec. H.262: Picture coding extension */ + .f_code[0][0] = s->mpeg_f_code[0][0], + .f_code[0][1] = s->mpeg_f_code[0][1], + .f_code[1][0] = s->mpeg_f_code[1][0], + .f_code[1][1] = s->mpeg_f_code[1][1], + .picture_structure = s->picture_structure, + .intra_dc_precision = s->intra_dc_precision, + }; + + if (s->top_field_first) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_TOP_FIELD_FIRST; + + if (s->frame_pred_frame_dct) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_FRAME_PRED_DCT; + + if (s->concealment_motion_vectors) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_CONCEALMENT_MV; + + if (s->intra_vlc_format) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_INTRA_VLC; + + if (s->q_scale_type) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_Q_SCALE_TYPE; + + if (s->alternate_scan) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_ALT_SCAN; + + if (s->repeat_first_field) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_REPEAT_FIRST; + + if (s->progressive_frame) + controls->picture.flags |= V4L2_MPEG2_PIC_FLAG_PROGRESSIVE; + + switch (s->pict_type) { + case AV_PICTURE_TYPE_B: + controls->picture.backward_ref_ts = ff_v4l2_request_get_capture_timestamp(s->next_picture.f); + // fall-through + case AV_PICTURE_TYPE_P: + controls->picture.forward_ref_ts = ff_v4l2_request_get_capture_timestamp(s->last_picture.f); + } + + for (int i = 0; i < 64; i++) { + int n = s->idsp.idct_permutation[ff_zigzag_direct[i]]; + controls->quantisation.intra_quantiser_matrix[i] = s->intra_matrix[n]; + controls->quantisation.non_intra_quantiser_matrix[i] = s->inter_matrix[n]; + controls->quantisation.chroma_intra_quantiser_matrix[i] = s->chroma_intra_matrix[n]; + controls->quantisation.chroma_non_intra_quantiser_matrix[i] = s->chroma_inter_matrix[n]; + } + + return ff_v4l2_request_reset_frame(avctx, s->current_picture_ptr->f); +} + +static int v4l2_request_mpeg2_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const MpegEncContext *s = avctx->priv_data; + + return ff_v4l2_request_append_output_buffer(avctx, s->current_picture_ptr->f, buffer, size); +} + +static int v4l2_request_mpeg2_end_frame(AVCodecContext *avctx) +{ + const MpegEncContext *s = avctx->priv_data; + V4L2RequestControlsMPEG2 *controls = s->current_picture_ptr->hwaccel_picture_private; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_MPEG2_SEQUENCE, + .ptr = &controls->sequence, + .size = sizeof(controls->sequence), + }, + { + .id = V4L2_CID_STATELESS_MPEG2_PICTURE, + .ptr = &controls->picture, + .size = sizeof(controls->picture), + }, + { + .id = V4L2_CID_STATELESS_MPEG2_QUANTISATION, + .ptr = &controls->quantisation, + .size = sizeof(controls->quantisation), + }, + }; + + return ff_v4l2_request_decode_frame(avctx, s->current_picture_ptr->f, control, FF_ARRAY_ELEMS(control)); +} + +static int v4l2_request_mpeg2_init(AVCodecContext *avctx) +{ + return ff_v4l2_request_init(avctx, V4L2_PIX_FMT_MPEG2_SLICE, 1024 * 1024, NULL, 0); +} + +const FFHWAccel ff_mpeg2_v4l2request_hwaccel = { + .p.name = "mpeg2_v4l2request", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_MPEG2VIDEO, + .p.pix_fmt = AV_PIX_FMT_DRM_PRIME, + .start_frame = v4l2_request_mpeg2_start_frame, + .decode_slice = v4l2_request_mpeg2_decode_slice, + .end_frame = v4l2_request_mpeg2_end_frame, + .frame_priv_data_size = sizeof(V4L2RequestControlsMPEG2), + .init = v4l2_request_mpeg2_init, + .uninit = ff_v4l2_request_uninit, + .priv_data_size = sizeof(V4L2RequestContext), + .frame_params = ff_v4l2_request_frame_params, + .caps_internal = HWACCEL_CAP_ASYNC_SAFE, +}; -- 2.44.0 From 54cf5e04d320bac0839879728d912c30787b983b Mon Sep 17 00:00:00 2001 From: Boris Brezillon Date: Wed, 22 May 2019 14:46:58 +0200 Subject: [PATCH 06/18] Add V4L2 request API vp8 hwaccel Signed-off-by: Boris Brezillon Signed-off-by: Ezequiel Garcia Signed-off-by: Jonas Karlman --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/hwaccels.h | 1 + libavcodec/v4l2_request_vp8.c | 181 ++++++++++++++++++++++++++++++++++ libavcodec/vp8.c | 6 ++ 5 files changed, 192 insertions(+) create mode 100644 libavcodec/v4l2_request_vp8.c diff --git a/configure b/configure index 3127877..3ea4a3f 100755 --- a/configure +++ b/configure @@ -3157,6 +3157,8 @@ vc1_vdpau_hwaccel_deps="vdpau" vc1_vdpau_hwaccel_select="vc1_decoder" vp8_nvdec_hwaccel_deps="nvdec" vp8_nvdec_hwaccel_select="vp8_decoder" +vp8_v4l2request_hwaccel_deps="v4l2_request vp8_v4l2_request" +vp8_v4l2request_hwaccel_select="vp8_decoder" vp8_vaapi_hwaccel_deps="vaapi" vp8_vaapi_hwaccel_select="vp8_decoder" vp9_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_VP9" @@ -7009,6 +7011,7 @@ fi check_func_headers "linux/media.h linux/videodev2.h" v4l2_timeval_to_ns check_cc h264_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_H264_SLICE;" check_cc mpeg2_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_MPEG2_SLICE;" +check_cc vp8_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_VP8_FRAME;" check_headers sys/videoio.h test_code cc sys/videoio.h "struct v4l2_frmsizeenum vfse; vfse.discrete.width = 0;" && enable_sanitized struct_v4l2_frmivalenum_discrete diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 668f06a..3de7182 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1037,6 +1037,7 @@ OBJS-$(CONFIG_VC1_QSV_HWACCEL) += qsvdec.o OBJS-$(CONFIG_VC1_VAAPI_HWACCEL) += vaapi_vc1.o OBJS-$(CONFIG_VC1_VDPAU_HWACCEL) += vdpau_vc1.o OBJS-$(CONFIG_VP8_NVDEC_HWACCEL) += nvdec_vp8.o +OBJS-$(CONFIG_VP8_V4L2REQUEST_HWACCEL) += v4l2_request_vp8.o OBJS-$(CONFIG_VP8_VAAPI_HWACCEL) += vaapi_vp8.o OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL) += dxva2_vp9.o OBJS-$(CONFIG_VP9_DXVA2_HWACCEL) += dxva2_vp9.o diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index f5baf8a..a531226 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -70,6 +70,7 @@ extern const struct FFHWAccel ff_vc1_nvdec_hwaccel; extern const struct FFHWAccel ff_vc1_vaapi_hwaccel; extern const struct FFHWAccel ff_vc1_vdpau_hwaccel; extern const struct FFHWAccel ff_vp8_nvdec_hwaccel; +extern const struct FFHWAccel ff_vp8_v4l2request_hwaccel; extern const struct FFHWAccel ff_vp8_vaapi_hwaccel; extern const struct FFHWAccel ff_vp9_d3d11va_hwaccel; extern const struct FFHWAccel ff_vp9_d3d11va2_hwaccel; diff --git a/libavcodec/v4l2_request_vp8.c b/libavcodec/v4l2_request_vp8.c new file mode 100644 index 0000000..415a4c4 --- /dev/null +++ b/libavcodec/v4l2_request_vp8.c @@ -0,0 +1,181 @@ +/* + * 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 "hwaccel_internal.h" +#include "hwconfig.h" +#include "v4l2_request.h" +#include "vp8.h" + +typedef struct V4L2RequestControlsVP8 { + struct v4l2_ctrl_vp8_frame ctrl; +} V4L2RequestControlsVP8; + +static int v4l2_request_vp8_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const VP8Context *s = avctx->priv_data; + V4L2RequestControlsVP8 *controls = s->framep[VP8_FRAME_CURRENT]->hwaccel_picture_private; + + memset(&controls->ctrl, 0, sizeof(controls->ctrl)); + return ff_v4l2_request_reset_frame(avctx, s->framep[VP8_FRAME_CURRENT]->tf.f); +} + +static int v4l2_request_vp8_end_frame(AVCodecContext *avctx) +{ + const VP8Context *s = avctx->priv_data; + V4L2RequestControlsVP8 *controls = s->framep[VP8_FRAME_CURRENT]->hwaccel_picture_private; + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_VP8_FRAME, + .ptr = &controls->ctrl, + .size = sizeof(controls->ctrl), + }, + }; + + return ff_v4l2_request_decode_frame(avctx, s->framep[VP8_FRAME_CURRENT]->tf.f, + control, FF_ARRAY_ELEMS(control)); +} + +static int v4l2_request_vp8_decode_slice(AVCodecContext *avctx, + const uint8_t *buffer, + uint32_t size) +{ + const VP8Context *s = avctx->priv_data; + V4L2RequestControlsVP8 *controls = s->framep[VP8_FRAME_CURRENT]->hwaccel_picture_private; + struct v4l2_ctrl_vp8_frame *frame = &controls->ctrl; + const uint8_t *data = buffer + 3 + 7 * s->keyframe; + unsigned int i, j, k; + + frame->version = s->profile & 0x3; + frame->width = avctx->width; + frame->height = avctx->height; + /* FIXME: set ->xx_scale */ + frame->prob_skip_false = s->prob->mbskip; + frame->prob_intra = s->prob->intra; + frame->prob_gf = s->prob->golden; + frame->prob_last = s->prob->last; + frame->first_part_size = s->header_partition_size; + frame->first_part_header_bits = (8 * (s->coder_state_at_header_end.input - data) - + s->coder_state_at_header_end.bit_count - 8); + frame->num_dct_parts = s->num_coeff_partitions; + for (i = 0; i < 8; i++) + frame->dct_part_sizes[i] = s->coeff_partition_size[i]; + + frame->coder_state.range = s->coder_state_at_header_end.range; + frame->coder_state.value = s->coder_state_at_header_end.value; + frame->coder_state.bit_count = s->coder_state_at_header_end.bit_count; + if (s->framep[VP8_FRAME_PREVIOUS]) + frame->last_frame_ts = ff_v4l2_request_get_capture_timestamp(s->framep[VP8_FRAME_PREVIOUS]->tf.f); + if (s->framep[VP8_FRAME_GOLDEN]) + frame->golden_frame_ts = ff_v4l2_request_get_capture_timestamp(s->framep[VP8_FRAME_GOLDEN]->tf.f); + if (s->framep[VP8_FRAME_ALTREF]) + frame->alt_frame_ts = ff_v4l2_request_get_capture_timestamp(s->framep[VP8_FRAME_ALTREF]->tf.f); + frame->flags |= s->invisible ? 0 : V4L2_VP8_FRAME_FLAG_SHOW_FRAME; + frame->flags |= s->mbskip_enabled ? V4L2_VP8_FRAME_FLAG_MB_NO_SKIP_COEFF : 0; + frame->flags |= (s->profile & 0x4) ? V4L2_VP8_FRAME_FLAG_EXPERIMENTAL : 0; + frame->flags |= s->keyframe ? V4L2_VP8_FRAME_FLAG_KEY_FRAME : 0; + frame->flags |= s->sign_bias[VP8_FRAME_GOLDEN] ? V4L2_VP8_FRAME_FLAG_SIGN_BIAS_GOLDEN : 0; + frame->flags |= s->sign_bias[VP8_FRAME_ALTREF] ? V4L2_VP8_FRAME_FLAG_SIGN_BIAS_ALT : 0; + frame->segment.flags |= s->segmentation.enabled ? V4L2_VP8_SEGMENT_FLAG_ENABLED : 0; + frame->segment.flags |= s->segmentation.update_map ? V4L2_VP8_SEGMENT_FLAG_UPDATE_MAP : 0; + frame->segment.flags |= s->segmentation.update_feature_data ? V4L2_VP8_SEGMENT_FLAG_UPDATE_FEATURE_DATA : 0; + frame->segment.flags |= s->segmentation.absolute_vals ? 0 : V4L2_VP8_SEGMENT_FLAG_DELTA_VALUE_MODE; + for (i = 0; i < 4; i++) { + frame->segment.quant_update[i] = s->segmentation.base_quant[i]; + frame->segment.lf_update[i] = s->segmentation.filter_level[i]; + } + + for (i = 0; i < 3; i++) + frame->segment.segment_probs[i] = s->prob->segmentid[i]; + + frame->lf.level = s->filter.level; + frame->lf.sharpness_level = s->filter.sharpness; + frame->lf.flags |= s->lf_delta.enabled ? V4L2_VP8_LF_ADJ_ENABLE : 0; + frame->lf.flags |= s->lf_delta.update ? V4L2_VP8_LF_DELTA_UPDATE : 0; + frame->lf.flags |= s->filter.simple ? V4L2_VP8_LF_FILTER_TYPE_SIMPLE : 0; + for (i = 0; i < 4; i++) { + frame->lf.ref_frm_delta[i] = s->lf_delta.ref[i]; + frame->lf.mb_mode_delta[i] = s->lf_delta.mode[i + MODE_I4x4]; + } + + // Probabilites + if (s->keyframe) { + static const uint8_t keyframe_y_mode_probs[4] = { + 145, 156, 163, 128 + }; + static const uint8_t keyframe_uv_mode_probs[3] = { + 142, 114, 183 + }; + + memcpy(frame->entropy.y_mode_probs, keyframe_y_mode_probs, 4); + memcpy(frame->entropy.uv_mode_probs, keyframe_uv_mode_probs, 3); + } else { + for (i = 0; i < 4; i++) + frame->entropy.y_mode_probs[i] = s->prob->pred16x16[i]; + for (i = 0; i < 3; i++) + frame->entropy.uv_mode_probs[i] = s->prob->pred8x8c[i]; + } + for (i = 0; i < 2; i++) + for (j = 0; j < 19; j++) + frame->entropy.mv_probs[i][j] = s->prob->mvc[i][j]; + + for (i = 0; i < 4; i++) { + for (j = 0; j < 8; j++) { + static const int coeff_bands_inverse[8] = { + 0, 1, 2, 3, 5, 6, 4, 15 + }; + int coeff_pos = coeff_bands_inverse[j]; + + for (k = 0; k < 3; k++) { + memcpy(frame->entropy.coeff_probs[i][j][k], + s->prob->token[i][coeff_pos][k], 11); + } + } + } + + frame->quant.y_ac_qi = s->quant.yac_qi; + frame->quant.y_dc_delta = s->quant.ydc_delta; + frame->quant.y2_dc_delta = s->quant.y2dc_delta; + frame->quant.y2_ac_delta = s->quant.y2ac_delta; + frame->quant.uv_dc_delta = s->quant.uvdc_delta; + frame->quant.uv_ac_delta = s->quant.uvac_delta; + + return ff_v4l2_request_append_output_buffer(avctx, s->framep[VP8_FRAME_CURRENT]->tf.f, buffer, size); +} + +static int v4l2_request_vp8_init(AVCodecContext *avctx) +{ + return ff_v4l2_request_init(avctx, V4L2_PIX_FMT_VP8_FRAME, 2 * 1024 * 1024, NULL, 0); +} + +const FFHWAccel ff_vp8_v4l2request_hwaccel = { + .p.name = "vp8_v4l2request", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_VP8, + .p.pix_fmt = AV_PIX_FMT_DRM_PRIME, + .start_frame = v4l2_request_vp8_start_frame, + .decode_slice = v4l2_request_vp8_decode_slice, + .end_frame = v4l2_request_vp8_end_frame, + .frame_priv_data_size = sizeof(V4L2RequestControlsVP8), + .init = v4l2_request_vp8_init, + .uninit = ff_v4l2_request_uninit, + .priv_data_size = sizeof(V4L2RequestContext), + .frame_params = ff_v4l2_request_frame_params, + .caps_internal = HWACCEL_CAP_ASYNC_SAFE, +}; diff --git a/libavcodec/vp8.c b/libavcodec/vp8.c index ffc430d..eb619c9 100644 --- a/libavcodec/vp8.c +++ b/libavcodec/vp8.c @@ -197,6 +197,9 @@ static enum AVPixelFormat get_pixel_format(VP8Context *s) #endif #if CONFIG_VP8_NVDEC_HWACCEL AV_PIX_FMT_CUDA, +#endif +#if CONFIG_VP8_V4L2REQUEST_HWACCEL + AV_PIX_FMT_DRM_PRIME, #endif AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE, @@ -3002,6 +3005,9 @@ const FFCodec ff_vp8_decoder = { #endif #if CONFIG_VP8_NVDEC_HWACCEL HWACCEL_NVDEC(vp8), +#endif +#if CONFIG_VP8_V4L2REQUEST_HWACCEL + HWACCEL_V4L2REQUEST(vp8), #endif NULL }, -- 2.44.0 From 8f67d3099fcb63251f3fe718bc464417a053eb8e Mon Sep 17 00:00:00 2001 From: Jernej Skrabec Date: Sat, 15 Dec 2018 22:32:16 +0100 Subject: [PATCH 07/18] Add V4L2 request API hevc hwaccel Signed-off-by: Jernej Skrabec Signed-off-by: Jonas Karlman Signed-off-by: Benjamin Gaignard Signed-off-by: Alex Bee --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/hevcdec.c | 10 + libavcodec/hwaccels.h | 1 + libavcodec/v4l2_request_hevc.c | 682 +++++++++++++++++++++++++++++++++ 5 files changed, 697 insertions(+) create mode 100644 libavcodec/v4l2_request_hevc.c diff --git a/configure b/configure index 3ea4a3f..ad9af0b 100755 --- a/configure +++ b/configure @@ -3099,6 +3099,8 @@ hevc_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_HEVC" hevc_dxva2_hwaccel_select="hevc_decoder" hevc_nvdec_hwaccel_deps="nvdec" hevc_nvdec_hwaccel_select="hevc_decoder" +hevc_v4l2request_hwaccel_deps="v4l2_request hevc_v4l2_request" +hevc_v4l2request_hwaccel_select="hevc_decoder" hevc_vaapi_hwaccel_deps="vaapi VAPictureParameterBufferHEVC" hevc_vaapi_hwaccel_select="hevc_decoder" hevc_vdpau_hwaccel_deps="vdpau VdpPictureInfoHEVC" @@ -7010,6 +7012,7 @@ fi check_func_headers "linux/media.h linux/videodev2.h" v4l2_timeval_to_ns check_cc h264_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_H264_SLICE;" +check_cc hevc_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_HEVC_SLICE;" check_cc mpeg2_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_MPEG2_SLICE;" check_cc vp8_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_VP8_FRAME;" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 3de7182..e32bd18 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1010,6 +1010,7 @@ OBJS-$(CONFIG_HEVC_D3D11VA_HWACCEL) += dxva2_hevc.o OBJS-$(CONFIG_HEVC_DXVA2_HWACCEL) += dxva2_hevc.o OBJS-$(CONFIG_HEVC_NVDEC_HWACCEL) += nvdec_hevc.o OBJS-$(CONFIG_HEVC_QSV_HWACCEL) += qsvdec.o +OBJS-$(CONFIG_HEVC_V4L2REQUEST_HWACCEL) += v4l2_request_hevc.o OBJS-$(CONFIG_HEVC_VAAPI_HWACCEL) += vaapi_hevc.o h265_profile_level.o OBJS-$(CONFIG_HEVC_VDPAU_HWACCEL) += vdpau_hevc.o h265_profile_level.o OBJS-$(CONFIG_HEVC_VULKAN_HWACCEL) += vulkan_decode.o vulkan_hevc.o diff --git a/libavcodec/hevcdec.c b/libavcodec/hevcdec.c index b267f6e..2086ef3 100644 --- a/libavcodec/hevcdec.c +++ b/libavcodec/hevcdec.c @@ -401,6 +401,7 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + \ CONFIG_HEVC_D3D11VA_HWACCEL * 2 + \ CONFIG_HEVC_NVDEC_HWACCEL + \ + CONFIG_HEVC_V4L2REQUEST_HWACCEL + \ CONFIG_HEVC_VAAPI_HWACCEL + \ CONFIG_HEVC_VIDEOTOOLBOX_HWACCEL + \ CONFIG_HEVC_VDPAU_HWACCEL + \ @@ -431,6 +432,9 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) #endif #if CONFIG_HEVC_VULKAN_HWACCEL *fmt++ = AV_PIX_FMT_VULKAN; +#endif +#if CONFIG_HEVC_V4L2REQUEST_HWACCEL + *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif break; case AV_PIX_FMT_YUV420P10: @@ -455,6 +459,9 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) #endif #if CONFIG_HEVC_NVDEC_HWACCEL *fmt++ = AV_PIX_FMT_CUDA; +#endif +#if CONFIG_HEVC_V4L2REQUEST_HWACCEL + *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif break; case AV_PIX_FMT_YUV444P: @@ -3728,6 +3735,9 @@ const FFCodec ff_hevc_decoder = { #endif #if CONFIG_HEVC_VULKAN_HWACCEL HWACCEL_VULKAN(hevc), +#endif +#if CONFIG_HEVC_V4L2REQUEST_HWACCEL + HWACCEL_V4L2REQUEST(hevc), #endif NULL }, diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index a531226..4a0de92 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -41,6 +41,7 @@ extern const struct FFHWAccel ff_hevc_d3d11va_hwaccel; extern const struct FFHWAccel ff_hevc_d3d11va2_hwaccel; extern const struct FFHWAccel ff_hevc_dxva2_hwaccel; extern const struct FFHWAccel ff_hevc_nvdec_hwaccel; +extern const struct FFHWAccel ff_hevc_v4l2request_hwaccel; extern const struct FFHWAccel ff_hevc_vaapi_hwaccel; extern const struct FFHWAccel ff_hevc_vdpau_hwaccel; extern const struct FFHWAccel ff_hevc_videotoolbox_hwaccel; diff --git a/libavcodec/v4l2_request_hevc.c b/libavcodec/v4l2_request_hevc.c new file mode 100644 index 0000000..2c3537d --- /dev/null +++ b/libavcodec/v4l2_request_hevc.c @@ -0,0 +1,682 @@ +/* + * 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 "hevcdec.h" +#include "hwaccel_internal.h" +#include "hwconfig.h" +#include "internal.h" +#include "v4l2_request.h" + +#define MAX_SLICES 600 // as per HEVC spec ? +#define V4L2_HEVC_CONTROLS_MAX 6 + +typedef struct V4L2RequestControlsHEVC { + struct v4l2_ctrl_hevc_sps sps; + struct v4l2_ctrl_hevc_pps pps; + struct v4l2_ctrl_hevc_decode_params dec_params; + struct v4l2_ctrl_hevc_scaling_matrix scaling_matrix; + struct v4l2_ctrl_hevc_slice_params slice_params[MAX_SLICES]; + __u32 *entry_point_offsets; + unsigned int num_entry_point_offsets; + int first_slice; + int num_slices; +} V4L2RequestControlsHEVC; + +typedef struct V4L2RequestContextHEVC { + V4L2RequestContext base; + unsigned int decode_mode; + unsigned int start_code; + __u32 max_slices; + unsigned int supports_entry_point_offsets; + unsigned int supports_slices; + unsigned int supports_scaling_matrix; +} V4L2RequestContextHEVC; + +static uint8_t nalu_slice_start_code[] = { 0x00, 0x00, 0x01 }; + +static void v4l2_request_hevc_fill_pred_table(const HEVCContext *h, struct v4l2_hevc_pred_weight_table *table) +{ + int32_t luma_weight_denom, chroma_weight_denom; + const SliceHeader *sh = &h->sh; + + if (sh->slice_type == HEVC_SLICE_I || + (sh->slice_type == HEVC_SLICE_P && !h->ps.pps->weighted_pred_flag) || + (sh->slice_type == HEVC_SLICE_B && !h->ps.pps->weighted_bipred_flag)) + return; + + table->luma_log2_weight_denom = sh->luma_log2_weight_denom; + + if (h->ps.sps->chroma_format_idc) + table->delta_chroma_log2_weight_denom = sh->chroma_log2_weight_denom - sh->luma_log2_weight_denom; + + luma_weight_denom = (1 << sh->luma_log2_weight_denom); + chroma_weight_denom = (1 << sh->chroma_log2_weight_denom); + + for (int i = 0; i < 15 && i < sh->nb_refs[L0]; i++) { + table->delta_luma_weight_l0[i] = sh->luma_weight_l0[i] - luma_weight_denom; + table->luma_offset_l0[i] = sh->luma_offset_l0[i]; + table->delta_chroma_weight_l0[i][0] = sh->chroma_weight_l0[i][0] - chroma_weight_denom; + table->delta_chroma_weight_l0[i][1] = sh->chroma_weight_l0[i][1] - chroma_weight_denom; + table->chroma_offset_l0[i][0] = sh->chroma_offset_l0[i][0]; + table->chroma_offset_l0[i][1] = sh->chroma_offset_l0[i][1]; + } + + if (sh->slice_type != HEVC_SLICE_B) + return; + + for (int i = 0; i < 15 && i < sh->nb_refs[L1]; i++) { + table->delta_luma_weight_l1[i] = sh->luma_weight_l1[i] - luma_weight_denom; + table->luma_offset_l1[i] = sh->luma_offset_l1[i]; + table->delta_chroma_weight_l1[i][0] = sh->chroma_weight_l1[i][0] - chroma_weight_denom; + table->delta_chroma_weight_l1[i][1] = sh->chroma_weight_l1[i][1] - chroma_weight_denom; + table->chroma_offset_l1[i][0] = sh->chroma_offset_l1[i][0]; + table->chroma_offset_l1[i][1] = sh->chroma_offset_l1[i][1]; + } +} + +static uint8_t get_ref_pic_index(const HEVCContext *h, const HEVCFrame *frame, + struct v4l2_ctrl_hevc_decode_params *dec_params) +{ + uint64_t timestamp; + + if (!frame) + return 0; + + timestamp = ff_v4l2_request_get_capture_timestamp(frame->frame); + + for (uint8_t i = 0; i < dec_params->num_active_dpb_entries; i++) { + struct v4l2_hevc_dpb_entry *entry = &dec_params->dpb[i]; + if (entry->timestamp == timestamp) + return i; + } + + return 0; +} + +static void fill_dec_params(struct v4l2_ctrl_hevc_decode_params *dec_params, const HEVCContext *h) +{ + const HEVCFrame *pic = h->ref; + const SliceHeader *sh = &h->sh; + int i, entries = 0; + + *dec_params = (struct v4l2_ctrl_hevc_decode_params) { + .pic_order_cnt_val = pic->poc, /* FIXME: is this same as slice_params->slice_pic_order_cnt ? */ + .short_term_ref_pic_set_size = sh->short_term_ref_pic_set_size, + .long_term_ref_pic_set_size = sh->long_term_ref_pic_set_size, + .num_poc_st_curr_before = h->rps[ST_CURR_BEF].nb_refs, + .num_poc_st_curr_after = h->rps[ST_CURR_AFT].nb_refs, + .num_poc_lt_curr = h->rps[LT_CURR].nb_refs, + }; + + for (i = 0; i < FF_ARRAY_ELEMS(h->DPB); i++) { + const HEVCFrame *frame = &h->DPB[i]; + if (frame != pic && (frame->flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF))) { + struct v4l2_hevc_dpb_entry *entry = &dec_params->dpb[entries++]; + + entry->timestamp = ff_v4l2_request_get_capture_timestamp(frame->frame); + entry->field_pic = !!(frame->frame->flags & AV_FRAME_FLAG_INTERLACED); + entry->flags = 0; + if (frame->flags & HEVC_FRAME_FLAG_LONG_REF) + entry->flags |= V4L2_HEVC_DPB_ENTRY_LONG_TERM_REFERENCE; + + entry->pic_order_cnt_val = frame->poc; + } + } + + dec_params->num_active_dpb_entries = entries; + + if (IS_IRAP(h)) + dec_params->flags |= V4L2_HEVC_DECODE_PARAM_FLAG_IRAP_PIC; + + if (IS_IDR(h)) + dec_params->flags |= V4L2_HEVC_DECODE_PARAM_FLAG_IDR_PIC; + + if (sh->no_output_of_prior_pics_flag) + dec_params->flags |= V4L2_HEVC_DECODE_PARAM_FLAG_NO_OUTPUT_OF_PRIOR; + + for (i = 0; i < V4L2_HEVC_DPB_ENTRIES_NUM_MAX; i++) { + dec_params->poc_st_curr_before[i] = get_ref_pic_index(h, h->rps[ST_CURR_BEF].ref[i], dec_params); + dec_params->poc_st_curr_after[i] = get_ref_pic_index(h, h->rps[ST_CURR_AFT].ref[i], dec_params); + dec_params->poc_lt_curr[i] = get_ref_pic_index(h, h->rps[LT_CURR].ref[i], dec_params); + } +} + +static int v4l2_request_hevc_fill_slice_params(const HEVCContext *h, + V4L2RequestControlsHEVC *controls, + int slice) +{ + struct v4l2_ctrl_hevc_slice_params *slice_params = &controls->slice_params[slice]; + struct v4l2_ctrl_hevc_decode_params *dec_params = &controls->dec_params; + const HEVCFrame *pic = h->ref; + const SliceHeader *sh = &h->sh; + RefPicList *rpl; + int i; + + *slice_params = (struct v4l2_ctrl_hevc_slice_params) { + .bit_size = 0, + .data_byte_offset = (get_bits_count(&h->HEVClc->gb) + 1 + 7) / 8, + .num_entry_point_offsets = sh->num_entry_point_offsets, + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: NAL unit header */ + .nal_unit_type = h->nal_unit_type, + .nuh_temporal_id_plus1 = h->temporal_id + 1, + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: General slice segment header */ + .slice_type = sh->slice_type, + .colour_plane_id = sh->colour_plane_id, + .slice_pic_order_cnt = pic->poc, + .num_ref_idx_l0_active_minus1 = sh->nb_refs[L0] ? sh->nb_refs[L0] - 1 : 0, + .num_ref_idx_l1_active_minus1 = sh->nb_refs[L1] ? sh->nb_refs[L1] - 1 : 0, + .collocated_ref_idx = sh->slice_temporal_mvp_enabled_flag ? sh->collocated_ref_idx : 0, + .five_minus_max_num_merge_cand = sh->slice_type == HEVC_SLICE_I ? 0 : 5 - sh->max_num_merge_cand, + .slice_qp_delta = sh->slice_qp_delta, + .slice_cb_qp_offset = sh->slice_cb_qp_offset, + .slice_cr_qp_offset = sh->slice_cr_qp_offset, + .slice_act_y_qp_offset = 0, + .slice_act_cb_qp_offset = 0, + .slice_act_cr_qp_offset = 0, + .slice_beta_offset_div2 = sh->beta_offset / 2, + .slice_tc_offset_div2 = sh->tc_offset / 2, + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: Picture timing SEI message */ + .pic_struct = h->sei.picture_timing.picture_struct, + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: General slice segment header */ + .slice_segment_addr = sh->slice_segment_addr, + .short_term_ref_pic_set_size = sh->short_term_ref_pic_set_size, + .long_term_ref_pic_set_size = sh->long_term_ref_pic_set_size, + }; + + if (sh->slice_sample_adaptive_offset_flag[0]) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_LUMA; + + if (sh->slice_sample_adaptive_offset_flag[1]) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_SAO_CHROMA; + + if (sh->slice_temporal_mvp_enabled_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_TEMPORAL_MVP_ENABLED; + + if (sh->mvd_l1_zero_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_MVD_L1_ZERO; + + if (sh->cabac_init_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_CABAC_INIT; + + if (sh->collocated_list == L0) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_COLLOCATED_FROM_L0; + + /* TODO: V4L2_HEVC_SLICE_PARAMS_FLAG_USE_INTEGER_MV */ + + if (sh->disable_deblocking_filter_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_DEBLOCKING_FILTER_DISABLED; + + if (sh->slice_loop_filter_across_slices_enabled_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_SLICE_LOOP_FILTER_ACROSS_SLICES_ENABLED; + + if (sh->dependent_slice_segment_flag) + slice_params->flags |= V4L2_HEVC_SLICE_PARAMS_FLAG_DEPENDENT_SLICE_SEGMENT; + + if (sh->slice_type != HEVC_SLICE_I) { + rpl = &h->ref->refPicList[0]; + for (i = 0; i < rpl->nb_refs; i++) + slice_params->ref_idx_l0[i] = get_ref_pic_index(h, rpl->ref[i], dec_params); + } + + if (sh->slice_type == HEVC_SLICE_B) { + rpl = &h->ref->refPicList[1]; + for (i = 0; i < rpl->nb_refs; i++) + slice_params->ref_idx_l1[i] = get_ref_pic_index(h, rpl->ref[i], dec_params); + } + + v4l2_request_hevc_fill_pred_table(h, &slice_params->pred_weight_table); + + if (controls->num_entry_point_offsets < sh->num_entry_point_offsets) { + av_freep(&controls->entry_point_offsets); + controls->entry_point_offsets = av_mallocz(sizeof(*controls->entry_point_offsets) * sh->num_entry_point_offsets); + if (!controls->entry_point_offsets) + return AVERROR(ENOMEM); + controls->num_entry_point_offsets = sh->num_entry_point_offsets; + } + + for (i = 0; i < sh->num_entry_point_offsets; i++) + controls->entry_point_offsets[i] = sh->entry_point_offset[i]; + + return 0; +} + +static void fill_sps(struct v4l2_ctrl_hevc_sps *ctrl, const HEVCContext *h) +{ + const HEVCSPS *sps = h->ps.sps; + const HEVCPPS *pps = h->ps.pps; + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: Sequence parameter set */ + *ctrl = (struct v4l2_ctrl_hevc_sps) { + .video_parameter_set_id = sps->vps_id, + .seq_parameter_set_id = pps->sps_id, + .pic_width_in_luma_samples = sps->width, + .pic_height_in_luma_samples = sps->height, + .bit_depth_luma_minus8 = sps->bit_depth - 8, + .bit_depth_chroma_minus8 = sps->bit_depth_chroma - 8, + .log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_poc_lsb - 4, + .sps_max_dec_pic_buffering_minus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1, + .sps_max_num_reorder_pics = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics, + .sps_max_latency_increase_plus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_latency_increase + 1, + .log2_min_luma_coding_block_size_minus3 = sps->log2_min_cb_size - 3, + .log2_diff_max_min_luma_coding_block_size = sps->log2_diff_max_min_coding_block_size, + .log2_min_luma_transform_block_size_minus2 = sps->log2_min_tb_size - 2, + .log2_diff_max_min_luma_transform_block_size = sps->log2_max_trafo_size - sps->log2_min_tb_size, + .max_transform_hierarchy_depth_inter = sps->max_transform_hierarchy_depth_inter, + .max_transform_hierarchy_depth_intra = sps->max_transform_hierarchy_depth_intra, + .pcm_sample_bit_depth_luma_minus1 = sps->pcm.bit_depth - 1, + .pcm_sample_bit_depth_chroma_minus1 = sps->pcm.bit_depth_chroma - 1, + .log2_min_pcm_luma_coding_block_size_minus3 = sps->pcm.log2_min_pcm_cb_size - 3, + .log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size, + .num_short_term_ref_pic_sets = sps->nb_st_rps, + .num_long_term_ref_pics_sps = sps->num_long_term_ref_pics_sps, + .chroma_format_idc = sps->chroma_format_idc, + .sps_max_sub_layers_minus1 = sps->max_sub_layers - 1, + }; + + if (sps->separate_colour_plane_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_SEPARATE_COLOUR_PLANE; + + if (sps->scaling_list_enable_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_SCALING_LIST_ENABLED; + + if (sps->amp_enabled_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_AMP_ENABLED; + + if (sps->sao_enabled) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_SAMPLE_ADAPTIVE_OFFSET; + + if (sps->pcm_enabled_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_PCM_ENABLED; + + if (sps->pcm.loop_filter_disable_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_PCM_LOOP_FILTER_DISABLED; + + if (sps->long_term_ref_pics_present_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_LONG_TERM_REF_PICS_PRESENT; + + if (sps->sps_temporal_mvp_enabled_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_SPS_TEMPORAL_MVP_ENABLED; + + if (sps->sps_strong_intra_smoothing_enable_flag) + ctrl->flags |= V4L2_HEVC_SPS_FLAG_STRONG_INTRA_SMOOTHING_ENABLED; +} + +static int v4l2_request_hevc_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const HEVCContext *h = avctx->priv_data; + const HEVCPPS *pps = h->ps.pps; + const HEVCSPS *sps = h->ps.sps; + const ScalingList *sl = pps->scaling_list_data_present_flag ? + &pps->scaling_list : + sps->scaling_list_enable_flag ? + &sps->scaling_list : NULL; + V4L2RequestControlsHEVC *controls = h->ref->hwaccel_picture_private; + const SliceHeader *sh = &h->sh; + + fill_sps(&controls->sps, h); + fill_dec_params(&controls->dec_params, h); + + if (sl) { + for (int i = 0; i < 6; i++) { + for (int j = 0; j < 16; j++) + controls->scaling_matrix.scaling_list_4x4[i][j] = sl->sl[0][i][j]; + for (int j = 0; j < 64; j++) { + controls->scaling_matrix.scaling_list_8x8[i][j] = sl->sl[1][i][j]; + controls->scaling_matrix.scaling_list_16x16[i][j] = sl->sl[2][i][j]; + if (i < 2) + controls->scaling_matrix.scaling_list_32x32[i][j] = sl->sl[3][i * 3][j]; + } + controls->scaling_matrix.scaling_list_dc_coef_16x16[i] = sl->sl_dc[0][i]; + if (i < 2) + controls->scaling_matrix.scaling_list_dc_coef_32x32[i] = sl->sl_dc[1][i * 3]; + } + } + + /* ISO/IEC 23008-2, ITU-T Rec. H.265: Picture parameter set */ + controls->pps = (struct v4l2_ctrl_hevc_pps) { + .pic_parameter_set_id = sh->pps_id, + .num_extra_slice_header_bits = pps->num_extra_slice_header_bits, + .num_ref_idx_l0_default_active_minus1 = pps->num_ref_idx_l0_default_active - 1, + .num_ref_idx_l1_default_active_minus1 = pps->num_ref_idx_l1_default_active - 1, + .init_qp_minus26 = pps->pic_init_qp_minus26, + .diff_cu_qp_delta_depth = pps->diff_cu_qp_delta_depth, + .pps_cb_qp_offset = pps->cb_qp_offset, + .pps_cr_qp_offset = pps->cr_qp_offset, + .pps_beta_offset_div2 = pps->beta_offset / 2, + .pps_tc_offset_div2 = pps->tc_offset / 2, + .log2_parallel_merge_level_minus2 = pps->log2_parallel_merge_level - 2, + }; + + if (pps->dependent_slice_segments_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_DEPENDENT_SLICE_SEGMENT_ENABLED; + + if (pps->output_flag_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_OUTPUT_FLAG_PRESENT; + + if (pps->sign_data_hiding_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_SIGN_DATA_HIDING_ENABLED; + + if (pps->cabac_init_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_CABAC_INIT_PRESENT; + + if (pps->constrained_intra_pred_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_CONSTRAINED_INTRA_PRED; + + if (pps->transform_skip_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_TRANSFORM_SKIP_ENABLED; + + if (pps->cu_qp_delta_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_CU_QP_DELTA_ENABLED; + + if (pps->pic_slice_level_chroma_qp_offsets_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_PPS_SLICE_CHROMA_QP_OFFSETS_PRESENT; + + if (pps->weighted_pred_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_WEIGHTED_PRED; + + if (pps->weighted_bipred_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_WEIGHTED_BIPRED; + + if (pps->transquant_bypass_enable_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_TRANSQUANT_BYPASS_ENABLED; + + if (pps->tiles_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_TILES_ENABLED; + + if (pps->entropy_coding_sync_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_ENTROPY_CODING_SYNC_ENABLED; + + if (pps->loop_filter_across_tiles_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_LOOP_FILTER_ACROSS_TILES_ENABLED; + + if (pps->seq_loop_filter_across_slices_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_PPS_LOOP_FILTER_ACROSS_SLICES_ENABLED; + + if (pps->deblocking_filter_override_enabled_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_DEBLOCKING_FILTER_OVERRIDE_ENABLED; + + if (pps->disable_dbf) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_PPS_DISABLE_DEBLOCKING_FILTER; + + if (pps->lists_modification_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_LISTS_MODIFICATION_PRESENT; + + if (pps->slice_header_extension_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_SLICE_SEGMENT_HEADER_EXTENSION_PRESENT; + + if (pps->deblocking_filter_control_present_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_DEBLOCKING_FILTER_CONTROL_PRESENT; + + if (pps->uniform_spacing_flag) + controls->pps.flags |= V4L2_HEVC_PPS_FLAG_UNIFORM_SPACING; + + if (pps->tiles_enabled_flag) { + controls->pps.num_tile_columns_minus1 = pps->num_tile_columns - 1; + controls->pps.num_tile_rows_minus1 = pps->num_tile_rows - 1; + + for (int i = 0; i < pps->num_tile_columns; i++) + controls->pps.column_width_minus1[i] = pps->column_width[i] - 1; + + for (int i = 0; i < pps->num_tile_rows; i++) + controls->pps.row_height_minus1[i] = pps->row_height[i] - 1; + } + + controls->first_slice = 1; + controls->num_slices = 0; + controls->num_entry_point_offsets = 0; + + return ff_v4l2_request_reset_frame(avctx, h->ref->frame); +} + +static int v4l2_request_hevc_queue_decode(AVCodecContext *avctx, int last_slice) +{ + const HEVCContext *h = avctx->priv_data; + V4L2RequestControlsHEVC *controls = h->ref->hwaccel_picture_private; + struct v4l2_ctrl_hevc_slice_params *first_slice_params = &controls->slice_params[0]; + V4L2RequestContextHEVC *ctx = avctx->internal->hwaccel_priv_data; + int num_controls = 0; + + struct v4l2_ext_control control[V4L2_HEVC_CONTROLS_MAX] = {}; + + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_SPS, + .ptr = &controls->sps, + .size = sizeof(controls->sps), + }; + + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_PPS, + .ptr = &controls->pps, + .size = sizeof(controls->pps), + }; + + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_DECODE_PARAMS, + .ptr = &controls->dec_params, + .size = sizeof(controls->dec_params), + }; + + if (ctx->supports_scaling_matrix) { + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_SCALING_MATRIX, + .ptr = &controls->scaling_matrix, + .size = sizeof(controls->scaling_matrix), + }; + } + + if (ctx->supports_slices) { + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_SLICE_PARAMS, + .ptr = &controls->slice_params, + .size = sizeof(*first_slice_params) * controls->num_slices, + }; + } + + //this assumes that decoders supporting entry_point_offsets submit a single slice per request + if (ctx->supports_entry_point_offsets && first_slice_params->num_entry_point_offsets > 0) { + control[num_controls++] = (struct v4l2_ext_control) { + .id = V4L2_CID_STATELESS_HEVC_ENTRY_POINT_OFFSETS, + .ptr = controls->entry_point_offsets, + .size = sizeof(*controls->entry_point_offsets) * first_slice_params->num_entry_point_offsets, + }; + } + + if (ctx->decode_mode == V4L2_STATELESS_HEVC_DECODE_MODE_SLICE_BASED) + return ff_v4l2_request_decode_slice(avctx, h->ref->frame, control, num_controls, controls->first_slice, last_slice); + + return ff_v4l2_request_decode_frame(avctx, h->ref->frame, control, num_controls); +} + +static int v4l2_request_hevc_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const HEVCContext *h = avctx->priv_data; + V4L2RequestControlsHEVC *controls = h->ref->hwaccel_picture_private; + V4L2RequestContextHEVC *ctx = avctx->internal->hwaccel_priv_data; + V4L2RequestDescriptor *req = (V4L2RequestDescriptor*)h->ref->frame->data[0]; + int ret, slice = FFMIN(controls->num_slices, MAX_SLICES - 1); + + if (ctx->decode_mode == V4L2_STATELESS_HEVC_DECODE_MODE_SLICE_BASED && slice) { + ret = v4l2_request_hevc_queue_decode(avctx, 0); + if (ret) + return ret; + + ff_v4l2_request_reset_frame(avctx, h->ref->frame); + slice = controls->num_slices = 0; + controls->first_slice = 0; + } + + ret = v4l2_request_hevc_fill_slice_params(h, controls, slice); + if (ret) + return ret; + + if (ctx->start_code == V4L2_STATELESS_HEVC_START_CODE_ANNEX_B) { + ret = ff_v4l2_request_append_output_buffer(avctx, h->ref->frame, nalu_slice_start_code, 3); + if (ret) + return ret; + } + + ret = ff_v4l2_request_append_output_buffer(avctx, h->ref->frame, buffer, size); + if (ret) + return ret; + + controls->slice_params[slice].bit_size = req->output.used * 8; //FIXME + controls->num_slices++; + return 0; +} + +static int v4l2_request_hevc_end_frame(AVCodecContext *avctx) +{ + const HEVCContext *h = avctx->priv_data; + V4L2RequestControlsHEVC *controls = h->ref->hwaccel_picture_private; + int ret; + + ret = v4l2_request_hevc_queue_decode(avctx, 1); + + av_freep(&controls->entry_point_offsets); + + return ret; +} + +static int v4l2_request_hevc_set_controls(AVCodecContext *avctx) +{ + V4L2RequestContextHEVC *ctx = avctx->internal->hwaccel_priv_data; + int ret; + + struct v4l2_ext_control control[] = { + { .id = V4L2_CID_STATELESS_HEVC_DECODE_MODE, }, + { .id = V4L2_CID_STATELESS_HEVC_START_CODE, }, + }; + struct v4l2_query_ext_ctrl entry_point_offsets = { + .id = V4L2_CID_STATELESS_HEVC_ENTRY_POINT_OFFSETS, + }; + struct v4l2_query_ext_ctrl slice_params = { + .id = V4L2_CID_STATELESS_HEVC_SLICE_PARAMS, + }; + struct v4l2_query_ext_ctrl scaling_matrix = { + .id = V4L2_CID_STATELESS_HEVC_SCALING_MATRIX, + }; + + ctx->decode_mode = ff_v4l2_request_query_control_default_value(avctx, V4L2_CID_STATELESS_HEVC_DECODE_MODE); + if (ctx->decode_mode != V4L2_STATELESS_HEVC_DECODE_MODE_SLICE_BASED && + ctx->decode_mode != V4L2_STATELESS_HEVC_DECODE_MODE_FRAME_BASED) { + av_log(avctx, AV_LOG_ERROR, "%s: unsupported decode mode, %d\n", __func__, ctx->decode_mode); + return AVERROR(EINVAL); + } + + ctx->start_code = ff_v4l2_request_query_control_default_value(avctx, V4L2_CID_STATELESS_HEVC_START_CODE); + if (ctx->start_code != V4L2_STATELESS_HEVC_START_CODE_NONE && + ctx->start_code != V4L2_STATELESS_HEVC_START_CODE_ANNEX_B) { + av_log(avctx, AV_LOG_ERROR, "%s: unsupported start code, %d\n", __func__, ctx->start_code); + return AVERROR(EINVAL); + } + + ret = ff_v4l2_request_query_control(avctx, &entry_point_offsets); + if (ret) + ctx->supports_entry_point_offsets = 0; + else + ctx->supports_entry_point_offsets = 1; + + ret = ff_v4l2_request_query_control(avctx, &slice_params); + if (ret) { + ctx->supports_slices = 0; + ctx->max_slices = 0; + if (ctx->decode_mode == V4L2_STATELESS_HEVC_DECODE_MODE_SLICE_BASED) { + av_log(avctx, AV_LOG_ERROR, "%s: decoder is slice-based, \ + but doesn't support V4L2_CID_STATELESS_HEVC_SLICE_PARAMS control \n", __func__); + return AVERROR(EINVAL); + } + + if (ctx->supports_entry_point_offsets) { + av_log(avctx, AV_LOG_ERROR, "%s: decoder supports entry_point_offsets, \ + but doesn't support V4L2_CID_STATELESS_HEVC_SLICE_PARAMS control \n", __func__); + return AVERROR(EINVAL); + } + } else { + ctx->supports_slices = 1; + ctx->max_slices = slice_params.dims[0]; + if (ctx->max_slices > MAX_SLICES) { + av_log(avctx, AV_LOG_ERROR, "%s: unsupported max slices, %u\n", __func__, ctx->max_slices); + return AVERROR(EINVAL); + } + } + + ret = ff_v4l2_request_query_control(avctx, &scaling_matrix); + if (ret) + ctx->supports_scaling_matrix = 0; + else + ctx->supports_scaling_matrix = 1; + + av_log(avctx, AV_LOG_DEBUG, "%s: decoder is %s and supports slices %d, supports entry_point_offsets: %d supports scaling_matrix: %d max slices: %u\n", + __func__, + ctx->decode_mode == V4L2_STATELESS_HEVC_DECODE_MODE_SLICE_BASED ? "slice based" : "frame based", + ctx->supports_slices, + ctx->supports_entry_point_offsets, + ctx->supports_scaling_matrix, + ctx->max_slices + ); + + control[0].value = ctx->decode_mode; + control[1].value = ctx->start_code; + + return ff_v4l2_request_set_controls(avctx, control, FF_ARRAY_ELEMS(control)); +} + +static int v4l2_request_hevc_init(AVCodecContext *avctx) +{ + const HEVCContext *h = avctx->priv_data; + struct v4l2_ctrl_hevc_sps sps; + int ret; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_HEVC_SPS, + .ptr = &sps, + .size = sizeof(sps), + }, + }; + + fill_sps(&sps, h); + + ret = ff_v4l2_request_init(avctx, V4L2_PIX_FMT_HEVC_SLICE, 4 * 1024 * 1024, control, FF_ARRAY_ELEMS(control)); + if (ret) + return ret; + + return v4l2_request_hevc_set_controls(avctx); +} + +const FFHWAccel ff_hevc_v4l2request_hwaccel = { + .p.name = "hevc_v4l2request", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_HEVC, + .p.pix_fmt = AV_PIX_FMT_DRM_PRIME, + .start_frame = v4l2_request_hevc_start_frame, + .decode_slice = v4l2_request_hevc_decode_slice, + .end_frame = v4l2_request_hevc_end_frame, + .frame_priv_data_size = sizeof(V4L2RequestControlsHEVC), + .init = v4l2_request_hevc_init, + .uninit = ff_v4l2_request_uninit, + .priv_data_size = sizeof(V4L2RequestContextHEVC), + .frame_params = ff_v4l2_request_frame_params, + .caps_internal = HWACCEL_CAP_ASYNC_SAFE, +}; -- 2.44.0 From 381025bd969276c176fd4fda0b6f2271cedfac8e Mon Sep 17 00:00:00 2001 From: Boris Brezillon Date: Thu, 12 Dec 2019 16:13:55 +0100 Subject: [PATCH 08/18] Add V4L2 request API vp9 hwaccel Signed-off-by: Boris Brezillon Signed-off-by: Jernej Skrabec --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/hwaccels.h | 1 + libavcodec/v4l2_request_vp9.c | 283 ++++++++++++++++++++++++++++++++++ libavcodec/vp9.c | 190 ++++++++++++++++------- libavcodec/vp9dec.h | 4 + libavcodec/vp9shared.h | 1 + 7 files changed, 429 insertions(+), 54 deletions(-) create mode 100644 libavcodec/v4l2_request_vp9.c diff --git a/configure b/configure index ad9af0b..810e57c 100755 --- a/configure +++ b/configure @@ -3171,6 +3171,8 @@ vp9_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_VP9" vp9_dxva2_hwaccel_select="vp9_decoder" vp9_nvdec_hwaccel_deps="nvdec" vp9_nvdec_hwaccel_select="vp9_decoder" +vp9_v4l2request_hwaccel_deps="v4l2_request" +vp9_v4l2request_hwaccel_select="vp9_decoder" vp9_vaapi_hwaccel_deps="vaapi VADecPictureParameterBufferVP9_bit_depth" vp9_vaapi_hwaccel_select="vp9_decoder" vp9_vdpau_hwaccel_deps="vdpau VdpPictureInfoVP9" @@ -7015,6 +7017,7 @@ check_cc h264_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_H264_SLICE;" check_cc hevc_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_HEVC_SLICE;" check_cc mpeg2_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_MPEG2_SLICE;" check_cc vp8_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_VP8_FRAME;" +check_cc vp9_v4l2_request linux/videodev2.h "int i = V4L2_PIX_FMT_VP9_FRAME;" check_headers sys/videoio.h test_code cc sys/videoio.h "struct v4l2_frmsizeenum vfse; vfse.discrete.width = 0;" && enable_sanitized struct_v4l2_frmivalenum_discrete diff --git a/libavcodec/Makefile b/libavcodec/Makefile index e32bd18..a2dffd5 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1043,6 +1043,7 @@ OBJS-$(CONFIG_VP8_VAAPI_HWACCEL) += vaapi_vp8.o OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL) += dxva2_vp9.o OBJS-$(CONFIG_VP9_DXVA2_HWACCEL) += dxva2_vp9.o OBJS-$(CONFIG_VP9_NVDEC_HWACCEL) += nvdec_vp9.o +OBJS-$(CONFIG_VP9_V4L2REQUEST_HWACCEL) += v4l2_request_vp9.o OBJS-$(CONFIG_VP9_VAAPI_HWACCEL) += vaapi_vp9.o OBJS-$(CONFIG_VP9_VDPAU_HWACCEL) += vdpau_vp9.o OBJS-$(CONFIG_VP9_VIDEOTOOLBOX_HWACCEL) += videotoolbox_vp9.o diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 4a0de92..8d0d348 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -77,6 +77,7 @@ extern const struct FFHWAccel ff_vp9_d3d11va_hwaccel; extern const struct FFHWAccel ff_vp9_d3d11va2_hwaccel; extern const struct FFHWAccel ff_vp9_dxva2_hwaccel; extern const struct FFHWAccel ff_vp9_nvdec_hwaccel; +extern const struct FFHWAccel ff_vp9_v4l2request_hwaccel; extern const struct FFHWAccel ff_vp9_vaapi_hwaccel; extern const struct FFHWAccel ff_vp9_vdpau_hwaccel; extern const struct FFHWAccel ff_vp9_videotoolbox_hwaccel; diff --git a/libavcodec/v4l2_request_vp9.c b/libavcodec/v4l2_request_vp9.c new file mode 100644 index 0000000..2c6a9e4 --- /dev/null +++ b/libavcodec/v4l2_request_vp9.c @@ -0,0 +1,283 @@ +/* + * 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 "hwaccel_internal.h" +#include "hwconfig.h" +#include "v4l2_request.h" +#include "vp9dec.h" + +typedef struct V4L2RequestControlsVP9 { + struct v4l2_ctrl_vp9_frame decode_params; + struct v4l2_ctrl_vp9_compressed_hdr chp; +} V4L2RequestControlsVP9; + +static void v4l2_request_vp9_set_frame_ctx(AVCodecContext *avctx) +{ + VP9Context *s = avctx->priv_data; + const VP9Frame *f = &s->s.frames[CUR_FRAME]; + V4L2RequestControlsVP9 *controls = f->hwaccel_picture_private; + struct v4l2_ctrl_vp9_compressed_hdr *chp = &controls->chp; + + memset(chp, 0, sizeof(&chp)); + + chp->tx_mode = s->s.h.txfmmode; + memcpy(chp->tx8, s->prob_raw.p.tx8p, sizeof(s->prob_raw.p.tx8p)); + memcpy(chp->tx16, s->prob_raw.p.tx16p, sizeof(s->prob_raw.p.tx16p)); + memcpy(chp->tx32, s->prob_raw.p.tx32p, sizeof(s->prob_raw.p.tx32p)); + for (unsigned i = 0; i < 4; i++) { + for (unsigned j = 0; j < 2; j++) { + for (unsigned k = 0; k < 2; k++) { + for (unsigned l = 0; l < 6; l++) { + for (unsigned m = 0; m < 6; m++) { + memcpy(chp->coef[i][j][k][l][m], s->prob_raw.coef[i][j][k][l][m], sizeof(chp->coef[0][0][0][0][0])); + } + } + } + } + } + memcpy(chp->skip, s->prob_raw.p.skip, sizeof(s->prob_raw.p.skip)); + memcpy(chp->inter_mode, s->prob_raw.p.mv_mode, sizeof(s->prob_raw.p.mv_mode)); + memcpy(chp->interp_filter, s->prob_raw.p.filter, sizeof(s->prob_raw.p.filter)); + memcpy(chp->is_inter, s->prob_raw.p.intra, sizeof(s->prob_raw.p.intra)); + memcpy(chp->comp_mode, s->prob_raw.p.comp, sizeof(s->prob_raw.p.comp)); + memcpy(chp->single_ref, s->prob_raw.p.single_ref, sizeof(s->prob_raw.p.single_ref)); + memcpy(chp->comp_ref, s->prob_raw.p.comp_ref, sizeof(s->prob_raw.p.comp_ref)); + memcpy(chp->y_mode, s->prob_raw.p.y_mode, sizeof(s->prob_raw.p.y_mode)); + for (unsigned i = 0; i < 10; i++) + memcpy(chp->uv_mode[i], s->prob.p.uv_mode[i], sizeof(s->prob.p.uv_mode[0])); + for (unsigned i = 0; i < 4; i++) + memcpy(chp->partition[i * 4], s->prob_raw.p.partition[i], sizeof(s->prob_raw.p.partition[0])); + memcpy(chp->mv.joint, s->prob_raw.p.mv_joint, sizeof(s->prob_raw.p.mv_joint)); + for (unsigned i = 0; i < 2; i++) { + chp->mv.sign[i] = s->prob_raw.p.mv_comp[i].sign; + memcpy(chp->mv.classes[i], s->prob_raw.p.mv_comp[i].classes, sizeof(s->prob_raw.p.mv_comp[0].classes)); + chp->mv.class0_bit[i] = s->prob_raw.p.mv_comp[i].class0; + memcpy(chp->mv.bits[i], s->prob_raw.p.mv_comp[i].bits, sizeof(s->prob_raw.p.mv_comp[0].bits)); + memcpy(chp->mv.class0_fr[i], s->prob_raw.p.mv_comp[i].class0_fp, sizeof(s->prob_raw.p.mv_comp[0].class0_fp)); + memcpy(chp->mv.fr[i], s->prob_raw.p.mv_comp[i].fp, sizeof(s->prob_raw.p.mv_comp[0].fp)); + chp->mv.class0_hp[i] = s->prob_raw.p.mv_comp[i].class0_hp; + chp->mv.hp[i] = s->prob_raw.p.mv_comp[i].hp; + } +} + +static void fill_frame(struct v4l2_ctrl_vp9_frame *dec_params, AVCodecContext *avctx) +{ + const VP9Context *s = avctx->priv_data; + const ThreadFrame *ref; + + memset(dec_params, 0, sizeof(*dec_params)); + + if (s->s.h.keyframe) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_KEY_FRAME; + if (!s->s.h.invisible) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_SHOW_FRAME; + if (s->s.h.errorres) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_ERROR_RESILIENT; + if (s->s.h.intraonly) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_INTRA_ONLY; + if (!s->s.h.keyframe && s->s.h.highprecisionmvs) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_ALLOW_HIGH_PREC_MV; + if (s->s.h.refreshctx) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_REFRESH_FRAME_CTX; + if (s->s.h.parallelmode) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_PARALLEL_DEC_MODE; + if (s->ss_h) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_X_SUBSAMPLING; + if (s->ss_v) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_Y_SUBSAMPLING; + if (avctx->color_range == AVCOL_RANGE_JPEG) + dec_params->flags |= V4L2_VP9_FRAME_FLAG_COLOR_RANGE_FULL_SWING; + + dec_params->compressed_header_size = s->s.h.compressed_header_size; + dec_params->uncompressed_header_size = s->s.h.uncompressed_header_size; + dec_params->profile = s->s.h.profile; + dec_params->reset_frame_context = s->s.h.resetctx > 0 ? s->s.h.resetctx - 1 : 0; + dec_params->frame_context_idx = s->s.h.framectxid; + dec_params->bit_depth = s->s.h.bpp; + + dec_params->interpolation_filter = s->s.h.filtermode ^ (s->s.h.filtermode <= 1); + dec_params->tile_cols_log2 = s->s.h.tiling.log2_tile_cols; + dec_params->tile_rows_log2 = s->s.h.tiling.log2_tile_rows; + dec_params->reference_mode = s->s.h.comppredmode; + dec_params->frame_width_minus_1 = s->w - 1; + dec_params->frame_height_minus_1 = s->h - 1; + //dec_params->render_width_minus_1 = avctx->width - 1; + //dec_params->render_height_minus_1 = avctx->height - 1; + + ref = &s->s.refs[s->s.h.refidx[0]]; + if (ref->f && ref->f->buf[0]) + dec_params->last_frame_ts = ff_v4l2_request_get_capture_timestamp(ref->f); + ref = &s->s.refs[s->s.h.refidx[1]]; + if (ref->f && ref->f->buf[0]) + dec_params->golden_frame_ts = ff_v4l2_request_get_capture_timestamp(ref->f); + ref = &s->s.refs[s->s.h.refidx[2]]; + if (ref->f && ref->f->buf[0]) + dec_params->alt_frame_ts = ff_v4l2_request_get_capture_timestamp(ref->f); + + if (s->s.h.signbias[0]) + dec_params->ref_frame_sign_bias |= V4L2_VP9_SIGN_BIAS_LAST; + if (s->s.h.signbias[1]) + dec_params->ref_frame_sign_bias |= V4L2_VP9_SIGN_BIAS_GOLDEN; + if (s->s.h.signbias[2]) + dec_params->ref_frame_sign_bias |= V4L2_VP9_SIGN_BIAS_ALT; + + if (s->s.h.lf_delta.enabled) + dec_params->lf.flags |= V4L2_VP9_LOOP_FILTER_FLAG_DELTA_ENABLED; + if (s->s.h.lf_delta.updated) + dec_params->lf.flags |= V4L2_VP9_LOOP_FILTER_FLAG_DELTA_UPDATE; + + dec_params->lf.level = s->s.h.filter.level; + dec_params->lf.sharpness = s->s.h.filter.sharpness; + for (unsigned i = 0; i < 4; i++) + dec_params->lf.ref_deltas[i] = s->s.h.lf_delta.ref[i]; + for (unsigned i = 0; i < 2; i++) + dec_params->lf.mode_deltas[i] = s->s.h.lf_delta.mode[i]; + + dec_params->quant.base_q_idx = s->s.h.yac_qi; + dec_params->quant.delta_q_y_dc = s->s.h.ydc_qdelta; + dec_params->quant.delta_q_uv_dc = s->s.h.uvdc_qdelta; + dec_params->quant.delta_q_uv_ac = s->s.h.uvac_qdelta; + + if (s->s.h.segmentation.enabled) + dec_params->seg.flags |= V4L2_VP9_SEGMENTATION_FLAG_ENABLED; + if (s->s.h.segmentation.update_map) + dec_params->seg.flags |= V4L2_VP9_SEGMENTATION_FLAG_UPDATE_MAP; + if (s->s.h.segmentation.temporal) + dec_params->seg.flags |= V4L2_VP9_SEGMENTATION_FLAG_TEMPORAL_UPDATE; + if (s->s.h.segmentation.update_data) + dec_params->seg.flags |= V4L2_VP9_SEGMENTATION_FLAG_UPDATE_DATA; + if (s->s.h.segmentation.absolute_vals) + dec_params->seg.flags |= V4L2_VP9_SEGMENTATION_FLAG_ABS_OR_DELTA_UPDATE; + + for (unsigned i = 0; i < 7; i++) + dec_params->seg.tree_probs[i] = s->s.h.segmentation.prob[i]; + + if (s->s.h.segmentation.temporal) { + for (unsigned i = 0; i < 3; i++) + dec_params->seg.pred_probs[i] = s->s.h.segmentation.pred_prob[i]; + } else { + memset(dec_params->seg.pred_probs, 255, sizeof(dec_params->seg.pred_probs)); + } + + for (unsigned i = 0; i < 8; i++) { + if (s->s.h.segmentation.feat[i].q_enabled) { + dec_params->seg.feature_enabled[i] |= 1 << V4L2_VP9_SEG_LVL_ALT_Q; + dec_params->seg.feature_data[i][V4L2_VP9_SEG_LVL_ALT_Q] = s->s.h.segmentation.feat[i].q_val; + } + + if (s->s.h.segmentation.feat[i].lf_enabled) { + dec_params->seg.feature_enabled[i] |= 1 << V4L2_VP9_SEG_LVL_ALT_L; + dec_params->seg.feature_data[i][V4L2_VP9_SEG_LVL_ALT_L] = s->s.h.segmentation.feat[i].lf_val; + } + + if (s->s.h.segmentation.feat[i].ref_enabled) { + dec_params->seg.feature_enabled[i] |= 1 << V4L2_VP9_SEG_LVL_REF_FRAME; + dec_params->seg.feature_data[i][V4L2_VP9_SEG_LVL_REF_FRAME] = s->s.h.segmentation.feat[i].ref_val; + } + + if (s->s.h.segmentation.feat[i].skip_enabled) + dec_params->seg.feature_enabled[i] |= 1 << V4L2_VP9_SEG_LVL_SKIP; + } +} + +static int v4l2_request_vp9_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const VP9Context *s = avctx->priv_data; + const VP9Frame *f = &s->s.frames[CUR_FRAME]; + V4L2RequestControlsVP9 *controls = f->hwaccel_picture_private; + + v4l2_request_vp9_set_frame_ctx(avctx); + + fill_frame(&controls->decode_params, avctx); + + return ff_v4l2_request_reset_frame(avctx, f->tf.f); +} + +static int v4l2_request_vp9_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const VP9Context *s = avctx->priv_data; + const VP9Frame *f = &s->s.frames[CUR_FRAME]; + + return ff_v4l2_request_append_output_buffer(avctx, f->tf.f, buffer, size); +} + +static int v4l2_request_vp9_end_frame(AVCodecContext *avctx) +{ + const VP9Context *s = avctx->priv_data; + const VP9Frame *f = &s->s.frames[CUR_FRAME]; + V4L2RequestControlsVP9 *controls = f->hwaccel_picture_private; + int ret; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_VP9_FRAME, + .ptr = &controls->decode_params, + .size = sizeof(controls->decode_params), + }, + { + .id = V4L2_CID_STATELESS_VP9_COMPRESSED_HDR, + .ptr = &controls->chp, + .size = sizeof(controls->chp), + }, + }; + + ret = ff_v4l2_request_decode_frame(avctx, f->tf.f, control, FF_ARRAY_ELEMS(control)); + if (ret) + return ret; + + if (!s->s.h.refreshctx) + return 0; + + return 0; +} + +static int v4l2_request_vp9_init(AVCodecContext *avctx) +{ + struct v4l2_ctrl_vp9_frame frame; + + struct v4l2_ext_control control[] = { + { + .id = V4L2_CID_STATELESS_VP9_FRAME, + .ptr = &frame, + .size = sizeof(frame), + }, + }; + + fill_frame(&frame, avctx); + + // TODO: check V4L2_CID_MPEG_VIDEO_VP9_PROFILE + return ff_v4l2_request_init(avctx, V4L2_PIX_FMT_VP9_FRAME, 3 * 1024 * 1024, control, FF_ARRAY_ELEMS(control)); +} + +const FFHWAccel ff_vp9_v4l2request_hwaccel = { + .p.name = "vp9_v4l2request", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_VP9, + .p.pix_fmt = AV_PIX_FMT_DRM_PRIME, + .start_frame = v4l2_request_vp9_start_frame, + .decode_slice = v4l2_request_vp9_decode_slice, + .end_frame = v4l2_request_vp9_end_frame, + .frame_priv_data_size = sizeof(V4L2RequestControlsVP9), + .init = v4l2_request_vp9_init, + .uninit = ff_v4l2_request_uninit, + .priv_data_size = sizeof(V4L2RequestContext), + .frame_params = ff_v4l2_request_frame_params, + .caps_internal = HWACCEL_CAP_ASYNC_SAFE, +}; diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 885500f..6ab88e5 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -177,6 +177,7 @@ static int update_size(AVCodecContext *avctx, int w, int h) #define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + \ CONFIG_VP9_D3D11VA_HWACCEL * 2 + \ CONFIG_VP9_NVDEC_HWACCEL + \ + CONFIG_VP9_V4L2REQUEST_HWACCEL + \ CONFIG_VP9_VAAPI_HWACCEL + \ CONFIG_VP9_VDPAU_HWACCEL + \ CONFIG_VP9_VIDEOTOOLBOX_HWACCEL) @@ -205,6 +206,9 @@ static int update_size(AVCodecContext *avctx, int w, int h) #if CONFIG_VP9_NVDEC_HWACCEL *fmtp++ = AV_PIX_FMT_CUDA; #endif +#if CONFIG_VP9_V4L2REQUEST_HWACCEL + *fmtp++ = AV_PIX_FMT_DRM_PRIME; +#endif #if CONFIG_VP9_VAAPI_HWACCEL *fmtp++ = AV_PIX_FMT_VAAPI; #endif @@ -219,6 +223,9 @@ static int update_size(AVCodecContext *avctx, int w, int h) #if CONFIG_VP9_NVDEC_HWACCEL *fmtp++ = AV_PIX_FMT_CUDA; #endif +#if CONFIG_VP9_V4L2REQUEST_HWACCEL + *fmtp++ = AV_PIX_FMT_DRM_PRIME; +#endif #if CONFIG_VP9_VAAPI_HWACCEL *fmtp++ = AV_PIX_FMT_VAAPI; #endif @@ -386,7 +393,7 @@ static av_always_inline int inv_recenter_nonneg(int v, int m) } // differential forward probability updates -static int update_prob(VPXRangeCoder *c, int p) +static int read_prob_delta(VPXRangeCoder *c) { static const uint8_t inv_map_table[255] = { 7, 20, 33, 46, 59, 72, 85, 98, 111, 124, 137, 150, 163, 176, @@ -440,8 +447,13 @@ static int update_prob(VPXRangeCoder *c, int p) av_assert2(d < FF_ARRAY_ELEMS(inv_map_table)); } - return p <= 128 ? 1 + inv_recenter_nonneg(inv_map_table[d], p - 1) : - 255 - inv_recenter_nonneg(inv_map_table[d], 255 - p); + return inv_map_table[d]; +} + +static int update_prob(int p, int d) +{ + return p <= 128 ? 1 + inv_recenter_nonneg(d, p - 1) : + 255 - inv_recenter_nonneg(d, 255 - p); } static int read_colorspace_details(AVCodecContext *avctx) @@ -707,7 +719,8 @@ static int decode_frame_header(AVCodecContext *avctx, get_bits(&s->gb, 8) : 255; } - if (get_bits1(&s->gb)) { + s->s.h.segmentation.update_data = get_bits1(&s->gb); + if (s->s.h.segmentation.update_data) { s->s.h.segmentation.absolute_vals = get_bits1(&s->gb); for (i = 0; i < 8; i++) { if ((s->s.h.segmentation.feat[i].q_enabled = get_bits1(&s->gb))) @@ -907,6 +920,8 @@ static int decode_frame_header(AVCodecContext *avctx, * as explicit copies if the fw update is missing (and skip the copy upon * fw update)? */ s->prob.p = s->prob_ctx[c].p; + memset(&s->prob_raw.p, 0, sizeof(s->prob_raw.p)); + memset(&s->prob_raw.coef, 0, sizeof(s->prob_raw.coef)); // txfm updates if (s->s.h.lossless) { @@ -918,18 +933,25 @@ static int decode_frame_header(AVCodecContext *avctx, if (s->s.h.txfmmode == TX_SWITCHABLE) { for (i = 0; i < 2; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.tx8p[i] = update_prob(&s->c, s->prob.p.tx8p[i]); + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.tx8p[i] = read_prob_delta(&s->c); + s->prob.p.tx8p[i] = update_prob(s->prob.p.tx8p[i], + s->prob_raw.p.tx8p[i]); + } for (i = 0; i < 2; i++) for (j = 0; j < 2; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.tx16p[i][j] = - update_prob(&s->c, s->prob.p.tx16p[i][j]); + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.tx16p[i][j] = read_prob_delta(&s->c); + s->prob.p.tx16p[i][j] = update_prob(s->prob.p.tx16p[i][j], + s->prob_raw.p.tx16p[i][j]); + } for (i = 0; i < 2; i++) for (j = 0; j < 3; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.tx32p[i][j] = - update_prob(&s->c, s->prob.p.tx32p[i][j]); + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.tx32p[i][j] = read_prob_delta(&s->c); + s->prob.p.tx32p[i][j] = update_prob(s->prob.p.tx32p[i][j], + s->prob_raw.p.tx32p[i][j]); + } } } @@ -941,15 +963,18 @@ static int decode_frame_header(AVCodecContext *avctx, for (k = 0; k < 2; k++) for (l = 0; l < 6; l++) for (m = 0; m < 6; m++) { + uint8_t *pd = s->prob_raw.coef[i][j][k][l][m]; uint8_t *p = s->prob.coef[i][j][k][l][m]; uint8_t *r = ref[j][k][l][m]; if (m >= 3 && l == 0) // dc only has 3 pt break; for (n = 0; n < 3; n++) { - if (vpx_rac_get_prob_branchy(&s->c, 252)) - p[n] = update_prob(&s->c, r[n]); - else + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + pd[n] = read_prob_delta(&s->c); + p[n] = update_prob(r[n], pd[n]); + } else { p[n] = r[n]; + } } memcpy(&p[3], ff_vp9_model_pareto8[p[2]], 8); } @@ -972,25 +997,37 @@ static int decode_frame_header(AVCodecContext *avctx, // mode updates for (i = 0; i < 3; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.skip[i] = update_prob(&s->c, s->prob.p.skip[i]); + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.skip[i] = read_prob_delta(&s->c); + s->prob.p.skip[i] = update_prob(s->prob.p.skip[i], + s->prob_raw.p.skip[i]); + } if (!s->s.h.keyframe && !s->s.h.intraonly) { for (i = 0; i < 7; i++) for (j = 0; j < 3; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_mode[i][j] = read_prob_delta(&s->c); s->prob.p.mv_mode[i][j] = - update_prob(&s->c, s->prob.p.mv_mode[i][j]); + update_prob(s->prob.p.mv_mode[i][j], + s->prob_raw.p.mv_mode[i][j]); + } if (s->s.h.filtermode == FILTER_SWITCHABLE) for (i = 0; i < 4; i++) for (j = 0; j < 2; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.filter[i][j] = read_prob_delta(&s->c); s->prob.p.filter[i][j] = - update_prob(&s->c, s->prob.p.filter[i][j]); + update_prob(s->prob.p.filter[i][j], + s->prob_raw.p.filter[i][j]); + } for (i = 0; i < 4; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.intra[i] = update_prob(&s->c, s->prob.p.intra[i]); + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.intra[i] = read_prob_delta(&s->c); + s->prob.p.intra[i] = update_prob(s->prob.p.intra[i], + s->prob_raw.p.intra[i]); + } if (s->s.h.allowcompinter) { s->s.h.comppredmode = vp89_rac_get(&s->c); @@ -998,92 +1035,134 @@ static int decode_frame_header(AVCodecContext *avctx, s->s.h.comppredmode += vp89_rac_get(&s->c); if (s->s.h.comppredmode == PRED_SWITCHABLE) for (i = 0; i < 5; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.comp[i] = read_prob_delta(&s->c); s->prob.p.comp[i] = - update_prob(&s->c, s->prob.p.comp[i]); + update_prob(s->prob.p.comp[i], s->prob_raw.p.comp[i]); + } } else { s->s.h.comppredmode = PRED_SINGLEREF; } if (s->s.h.comppredmode != PRED_COMPREF) { for (i = 0; i < 5; i++) { - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.single_ref[i][0] = read_prob_delta(&s->c); s->prob.p.single_ref[i][0] = - update_prob(&s->c, s->prob.p.single_ref[i][0]); - if (vpx_rac_get_prob_branchy(&s->c, 252)) + update_prob(s->prob.p.single_ref[i][0], + s->prob_raw.p.single_ref[i][0]); + } + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.single_ref[i][1] = read_prob_delta(&s->c); s->prob.p.single_ref[i][1] = - update_prob(&s->c, s->prob.p.single_ref[i][1]); + update_prob(s->prob.p.single_ref[i][1], + s->prob_raw.p.single_ref[i][1]); + } } } if (s->s.h.comppredmode != PRED_SINGLEREF) { for (i = 0; i < 5; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.comp_ref[i] = read_prob_delta(&s->c); s->prob.p.comp_ref[i] = - update_prob(&s->c, s->prob.p.comp_ref[i]); + update_prob(s->prob.p.comp_ref[i], + s->prob_raw.p.comp_ref[i]); + } } for (i = 0; i < 4; i++) for (j = 0; j < 9; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.y_mode[i][j] = read_prob_delta(&s->c); s->prob.p.y_mode[i][j] = - update_prob(&s->c, s->prob.p.y_mode[i][j]); + update_prob(s->prob.p.y_mode[i][j], + s->prob_raw.p.y_mode[i][j]); + } for (i = 0; i < 4; i++) for (j = 0; j < 4; j++) for (k = 0; k < 3; k++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.partition[i][j][k] = read_prob_delta(&s->c); s->prob.p.partition[3 - i][j][k] = - update_prob(&s->c, - s->prob.p.partition[3 - i][j][k]); + update_prob(s->prob.p.partition[3 - i][j][k], + s->prob_raw.p.partition[i][j][k]); + } // mv fields don't use the update_prob subexp model for some reason for (i = 0; i < 3; i++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_joint[i] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_joint[i] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_joint[i] = s->prob_raw.p.mv_joint[i]; + } for (i = 0; i < 2; i++) { - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].sign = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].sign = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].sign = + s->prob_raw.p.mv_comp[i].sign; + } for (j = 0; j < 10; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].classes[j] = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].classes[j] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].classes[j] = + s->prob_raw.p.mv_comp[i].classes[j]; + } - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].class0 = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].class0 = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].class0 = + s->prob_raw.p.mv_comp[i].class0; + } for (j = 0; j < 10; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].bits[j] = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].bits[j] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].bits[j] = + s->prob_raw.p.mv_comp[i].bits[j]; + } } for (i = 0; i < 2; i++) { for (j = 0; j < 2; j++) for (k = 0; k < 3; k++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].class0_fp[j][k] = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].class0_fp[j][k] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].class0_fp[j][k] = + s->prob_raw.p.mv_comp[i].class0_fp[j][k]; + } for (j = 0; j < 3; j++) - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].fp[j] = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].fp[j] = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].fp[j] = + s->prob_raw.p.mv_comp[i].fp[j]; + } } if (s->s.h.highprecisionmvs) { for (i = 0; i < 2; i++) { - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].class0_hp = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].class0_hp = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].class0_hp = + s->prob_raw.p.mv_comp[i].class0_hp; + } - if (vpx_rac_get_prob_branchy(&s->c, 252)) - s->prob.p.mv_comp[i].hp = + if (vpx_rac_get_prob_branchy(&s->c, 252)) { + s->prob_raw.p.mv_comp[i].hp = (vp89_rac_get_uint(&s->c, 7) << 1) | 1; + s->prob.p.mv_comp[i].hp = + s->prob_raw.p.mv_comp[i].hp; + } } } } @@ -1912,6 +1991,9 @@ const FFCodec ff_vp9_decoder = { #if CONFIG_VP9_VDPAU_HWACCEL HWACCEL_VDPAU(vp9), #endif +#if CONFIG_VP9_V4L2REQUEST_HWACCEL + HWACCEL_V4L2REQUEST(vp9), +#endif #if CONFIG_VP9_VIDEOTOOLBOX_HWACCEL HWACCEL_VIDEOTOOLBOX(vp9), #endif diff --git a/libavcodec/vp9dec.h b/libavcodec/vp9dec.h index de7aba0..5935ba6 100644 --- a/libavcodec/vp9dec.h +++ b/libavcodec/vp9dec.h @@ -135,6 +135,10 @@ typedef struct VP9Context { ProbContext p; uint8_t coef[4][2][2][6][6][11]; } prob; + struct { + ProbContext p; + uint8_t coef[4][2][2][6][6][11]; + } prob_raw; // contextual (above) cache uint8_t *above_partition_ctx; diff --git a/libavcodec/vp9shared.h b/libavcodec/vp9shared.h index e54f235..b8fef7d 100644 --- a/libavcodec/vp9shared.h +++ b/libavcodec/vp9shared.h @@ -136,6 +136,7 @@ typedef struct VP9BitstreamHeader { uint8_t temporal; uint8_t absolute_vals; uint8_t update_map; + uint8_t update_data; uint8_t prob[7]; uint8_t pred_prob[3]; struct { -- 2.44.0 From fdfd56866f020df1e333af7e8beb47b19b1c4ad0 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Mon, 29 Apr 2019 22:08:59 +0000 Subject: [PATCH 09/18] HACK: hwcontext_drm: do not require drm device Signed-off-by: Jonas Karlman --- libavutil/hwcontext_drm.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/libavutil/hwcontext_drm.c b/libavutil/hwcontext_drm.c index 7a9fdbd..6297d1f 100644 --- a/libavutil/hwcontext_drm.c +++ b/libavutil/hwcontext_drm.c @@ -53,6 +53,11 @@ static int drm_device_create(AVHWDeviceContext *hwdev, const char *device, AVDRMDeviceContext *hwctx = hwdev->hwctx; drmVersionPtr version; + if (device == NULL) { + hwctx->fd = -1; + return 0; + } + hwctx->fd = open(device, O_RDWR); if (hwctx->fd < 0) return AVERROR(errno); -- 2.44.0 From 38d09d83e6131bca6473674bf5979a13c979278b Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Fri, 15 May 2020 16:54:05 +0000 Subject: [PATCH 10/18] WIP: add NV15 and NV20 support Signed-off-by: Jonas Karlman --- libavcodec/h264_slice.c | 9 ++++++--- libavcodec/v4l2_request.c | 7 +++++++ 2 files changed, 13 insertions(+), 3 deletions(-) diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index 236bdc6..75f8235 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -812,6 +812,9 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) #endif #if CONFIG_H264_VULKAN_HWACCEL *fmt++ = AV_PIX_FMT_VULKAN; +#endif +#if CONFIG_H264_V4L2REQUEST_HWACCEL + *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif if (CHROMA444(h)) { if (h->avctx->colorspace == AVCOL_SPC_RGB) { @@ -868,6 +871,9 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL if (h->avctx->colorspace != AVCOL_SPC_RGB) *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX; +#endif +#if CONFIG_H264_V4L2REQUEST_HWACCEL + *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif if (CHROMA444(h)) { if (h->avctx->colorspace == AVCOL_SPC_RGB) @@ -891,9 +897,6 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) #endif #if CONFIG_H264_VAAPI_HWACCEL *fmt++ = AV_PIX_FMT_VAAPI; -#endif -#if CONFIG_H264_V4L2REQUEST_HWACCEL - *fmt++ = AV_PIX_FMT_DRM_PRIME; #endif if (h->avctx->color_range == AVCOL_RANGE_JPEG) *fmt++ = AV_PIX_FMT_YUVJ420P; diff --git a/libavcodec/v4l2_request.c b/libavcodec/v4l2_request.c index 595a53e..8d2a248 100644 --- a/libavcodec/v4l2_request.c +++ b/libavcodec/v4l2_request.c @@ -196,6 +196,13 @@ static const struct { } v4l2_request_capture_pixelformats[] = { { V4L2_PIX_FMT_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_LINEAR }, { V4L2_PIX_FMT_SUNXI_TILED_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_ALLWINNER_TILED }, +#if defined(V4L2_PIX_FMT_NV15) && defined(DRM_FORMAT_NV15) + { V4L2_PIX_FMT_NV15, AV_PIX_FMT_NONE, DRM_FORMAT_NV15, DRM_FORMAT_MOD_LINEAR }, +#endif + { V4L2_PIX_FMT_NV16, AV_PIX_FMT_NV16, DRM_FORMAT_NV16, DRM_FORMAT_MOD_LINEAR }, +#if defined(V4L2_PIX_FMT_NV20) && defined(DRM_FORMAT_NV20) + { V4L2_PIX_FMT_NV20, AV_PIX_FMT_NONE, DRM_FORMAT_NV20, DRM_FORMAT_MOD_LINEAR }, +#endif }; static int v4l2_request_set_drm_descriptor(V4L2RequestDescriptor *req, struct v4l2_format *format) -- 2.44.0 From 71a1f3f8cb5d81fa8b8239bc302dc27580558550 Mon Sep 17 00:00:00 2001 From: Jernej Skrabec Date: Mon, 9 Oct 2023 22:19:18 +0200 Subject: [PATCH 11/18] v4l2 request api: add new codecs --- libavcodec/v4l2_request.c | 32 ++++++++++++++++++++++++++++---- 1 file changed, 28 insertions(+), 4 deletions(-) diff --git a/libavcodec/v4l2_request.c b/libavcodec/v4l2_request.c index 8d2a248..813c1b3 100644 --- a/libavcodec/v4l2_request.c +++ b/libavcodec/v4l2_request.c @@ -202,6 +202,27 @@ static const struct { { V4L2_PIX_FMT_NV16, AV_PIX_FMT_NV16, DRM_FORMAT_NV16, DRM_FORMAT_MOD_LINEAR }, #if defined(V4L2_PIX_FMT_NV20) && defined(DRM_FORMAT_NV20) { V4L2_PIX_FMT_NV20, AV_PIX_FMT_NONE, DRM_FORMAT_NV20, DRM_FORMAT_MOD_LINEAR }, +#endif + { V4L2_PIX_FMT_P010, AV_PIX_FMT_P010, DRM_FORMAT_P010, DRM_FORMAT_MOD_LINEAR }, +#if defined(V4L2_PIX_FMT_YUV420_10_AFBC_16X16_SPLIT) + { + .pixelformat = V4L2_PIX_FMT_YUV420_10_AFBC_16X16_SPLIT, + .sw_format = AV_PIX_FMT_NONE, + .drm_format = DRM_FORMAT_YUV420_10BIT, + .format_modifier = DRM_FORMAT_MOD_ARM_AFBC(AFBC_FORMAT_MOD_BLOCK_SIZE_16x16 | + AFBC_FORMAT_MOD_SPARSE | + AFBC_FORMAT_MOD_SPLIT), + }, +#endif +#if defined(V4L2_PIX_FMT_YUV420_8_AFBC_16X16_SPLIT) + { + .pixelformat = V4L2_PIX_FMT_YUV420_8_AFBC_16X16_SPLIT, + .sw_format = AV_PIX_FMT_NONE, + .drm_format = DRM_FORMAT_YUV420_8BIT, + .format_modifier = DRM_FORMAT_MOD_ARM_AFBC(AFBC_FORMAT_MOD_BLOCK_SIZE_16x16 | + AFBC_FORMAT_MOD_SPARSE | + AFBC_FORMAT_MOD_SPLIT), + }, #endif }; @@ -228,15 +249,18 @@ static int v4l2_request_set_drm_descriptor(V4L2RequestDescriptor *req, struct v4 desc->objects[0].size = req->capture.size; desc->nb_layers = 1; - layer->nb_planes = 2; + layer->nb_planes = 1; layer->planes[0].object_index = 0; layer->planes[0].offset = 0; layer->planes[0].pitch = V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.plane_fmt[0].bytesperline : format->fmt.pix.bytesperline; - layer->planes[1].object_index = 0; - layer->planes[1].offset = layer->planes[0].pitch * (V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.height : format->fmt.pix.height); - layer->planes[1].pitch = layer->planes[0].pitch; + if (!fourcc_mod_is_vendor(desc->objects[0].format_modifier, ARM)) { + layer->nb_planes = 2; + layer->planes[1].object_index = 0; + layer->planes[1].offset = layer->planes[0].pitch * (V4L2_TYPE_IS_MULTIPLANAR(format->type) ? format->fmt.pix_mp.height : format->fmt.pix.height); + layer->planes[1].pitch = layer->planes[0].pitch; + } return 0; } -- 2.44.0 From b6ec3a75b3ebaae839257c5a6de8bbdd0fdc237c Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 4 Nov 2023 21:14:21 +0000 Subject: [PATCH 12/18] WIP: libavutil: add NV15 and NV20 pix fmt --- libavutil/pixdesc.c | 24 ++++++++++++++++++++++++ libavutil/pixfmt.h | 3 +++ 2 files changed, 27 insertions(+) diff --git a/libavutil/pixdesc.c b/libavutil/pixdesc.c index 6ded946..9747b5a 100644 --- a/libavutil/pixdesc.c +++ b/libavutil/pixdesc.c @@ -584,6 +584,18 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { }, .flags = AV_PIX_FMT_FLAG_PLANAR, }, + [AV_PIX_FMT_NV15] = { + .name = "nv15", + .nb_components = 3, + .log2_chroma_w = 1, + .log2_chroma_h = 1, + .comp = { + { 0, 1, 0, 0, 10 }, /* Y */ + { 1, 2, 0, 0, 10 }, /* U */ + { 1, 2, 1, 0, 10 }, /* V */ + }, + .flags = AV_PIX_FMT_FLAG_PLANAR, + }, [AV_PIX_FMT_ARGB] = { .name = "argb", .nb_components = 4, @@ -2085,6 +2097,18 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { }, .flags = AV_PIX_FMT_FLAG_PLANAR, }, + [AV_PIX_FMT_NV20NP] = { + .name = "nv20", + .nb_components = 3, + .log2_chroma_w = 1, + .log2_chroma_h = 0, + .comp = { + { 0, 1, 0, 0, 10 }, /* Y */ + { 1, 2, 0, 0, 10 }, /* U */ + { 1, 2, 1, 0, 10 }, /* V */ + }, + .flags = AV_PIX_FMT_FLAG_PLANAR, + }, [AV_PIX_FMT_NV20LE] = { .name = "nv20le", .nb_components = 3, diff --git a/libavutil/pixfmt.h b/libavutil/pixfmt.h index a26c72d..08e7aac 100644 --- a/libavutil/pixfmt.h +++ b/libavutil/pixfmt.h @@ -429,6 +429,9 @@ enum AVPixelFormat { AV_PIX_FMT_GBRAP14BE, ///< planar GBR 4:4:4:4 56bpp, big-endian AV_PIX_FMT_GBRAP14LE, ///< planar GBR 4:4:4:4 56bpp, little-endian + AV_PIX_FMT_NV15, ///< packed planar YUV 4:2:0, 15bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (first component U and the following component V), no padding between components + AV_PIX_FMT_NV20NP, ///< packed planar YUV 4:2:2, 20bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (first component U and the following component V), no padding between components + AV_PIX_FMT_NB ///< number of pixel formats, DO NOT USE THIS if you want to link with shared libav* because the number of formats might differ between versions }; -- 2.44.0 From 5f3f0bf02e8728a9d36050260055feaabdccf687 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 4 Nov 2023 22:35:38 +0000 Subject: [PATCH 13/18] WIP: libavutil: force NV15 and NV20 linesize --- libavutil/imgutils.c | 2 ++ libavutil/pixdesc.c | 4 ++-- libavutil/pixdesc.h | 5 +++++ 3 files changed, 9 insertions(+), 2 deletions(-) diff --git a/libavutil/imgutils.c b/libavutil/imgutils.c index da38126..0cd8d3d 100644 --- a/libavutil/imgutils.c +++ b/libavutil/imgutils.c @@ -70,6 +70,8 @@ int image_get_linesize(int width, int plane, if (desc->flags & AV_PIX_FMT_FLAG_BITSTREAM) linesize = (linesize + 7) >> 3; + if (desc->flags & AV_PIX_FMT_FLAG_10BIT_NO_PADDING) + linesize = linesize * 5 / 4; return linesize; } diff --git a/libavutil/pixdesc.c b/libavutil/pixdesc.c index 9747b5a..42c4acd 100644 --- a/libavutil/pixdesc.c +++ b/libavutil/pixdesc.c @@ -594,7 +594,7 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { { 1, 2, 0, 0, 10 }, /* U */ { 1, 2, 1, 0, 10 }, /* V */ }, - .flags = AV_PIX_FMT_FLAG_PLANAR, + .flags = AV_PIX_FMT_FLAG_PLANAR | AV_PIX_FMT_FLAG_10BIT_NO_PADDING, }, [AV_PIX_FMT_ARGB] = { .name = "argb", @@ -2107,7 +2107,7 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { { 1, 2, 0, 0, 10 }, /* U */ { 1, 2, 1, 0, 10 }, /* V */ }, - .flags = AV_PIX_FMT_FLAG_PLANAR, + .flags = AV_PIX_FMT_FLAG_PLANAR | AV_PIX_FMT_FLAG_10BIT_NO_PADDING, }, [AV_PIX_FMT_NV20LE] = { .name = "nv20le", diff --git a/libavutil/pixdesc.h b/libavutil/pixdesc.h index 0df73e6..73714c2 100644 --- a/libavutil/pixdesc.h +++ b/libavutil/pixdesc.h @@ -157,6 +157,11 @@ typedef struct AVPixFmtDescriptor { */ #define AV_PIX_FMT_FLAG_FLOAT (1 << 9) +/** + * HACK: force use of linesize = linesize * 5 / 4 + */ +#define AV_PIX_FMT_FLAG_10BIT_NO_PADDING (1 << 10) + /** * Return the number of bits per pixel used by the pixel format * described by pixdesc. Note that this is not the same as the number -- 2.44.0 From c941547fb6068b09140ad618f7cea7fcb2806189 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 4 Nov 2023 21:14:21 +0000 Subject: [PATCH 14/18] WIP: libswscale: add NV15 and NV20 unscaled conversion --- libswscale/swscale_unscaled.c | 55 +++++++++++++++++++++++++++++++++++ libswscale/utils.c | 2 ++ 2 files changed, 57 insertions(+) diff --git a/libswscale/swscale_unscaled.c b/libswscale/swscale_unscaled.c index a5c9917..5d054c0 100644 --- a/libswscale/swscale_unscaled.c +++ b/libswscale/swscale_unscaled.c @@ -221,6 +221,56 @@ static int nv24ToPlanarWrapper(SwsContext *c, const uint8_t *src[], return srcSliceH; } +static int nv15ToPlanarWrapper(SwsContext *c, const uint8_t *src[], + int srcStride[], int srcSliceY, + int srcSliceH, uint8_t *dstParam[], + int dstStride[]) +{ + const AVPixFmtDescriptor *dst_format = av_pix_fmt_desc_get(c->dstFormat); + int vsub = 1 << dst_format->log2_chroma_h; + uint16_t *dstY = (uint16_t*)(dstParam[0] + dstStride[0] * srcSliceY); + uint16_t *dstU = (uint16_t*)(dstParam[1] + dstStride[1] * srcSliceY / vsub); + uint16_t *dstV = (uint16_t*)(dstParam[2] + dstStride[2] * srcSliceY / vsub); + int x, y; + + const int shift[3] = { + dst_format->comp[0].shift, + dst_format->comp[1].shift, + dst_format->comp[2].shift, + }; + + for (y = srcSliceH; y > 0; y--) { + const uint8_t *tsrcY = src[0]; + uint16_t *tdstY = dstY; + for (x = c->srcW / 4; x > 0; x--) { + *tdstY++ = (((tsrcY[1] & 0x3) << 8) | (tsrcY[0] & 0xff)) << shift[0]; + *tdstY++ = (((tsrcY[2] & 0xf) << 6) | ((tsrcY[1] >> 2) & 0x3f)) << shift[0]; + *tdstY++ = (((tsrcY[3] & 0x3f) << 4) | ((tsrcY[2] >> 4) & 0xf)) << shift[0]; + *tdstY++ = (((tsrcY[4] & 0xff) << 2) | ((tsrcY[3] >> 6) & 0x3)) << shift[0]; + tsrcY += 5; + } + src[0] += srcStride[0]; + dstY += dstStride[0] / sizeof(uint16_t); + } + + for (y = srcSliceH / vsub; y > 0; y--) { + const uint8_t *tsrcUV = src[1]; + uint16_t *tdstU = dstU, *tdstV = dstV; + for (x = c->chrSrcW / 2; x > 0; x--) { + *tdstU++ = (((tsrcUV[1] & 0x3) << 8) | (tsrcUV[0] & 0xff)) << shift[1]; + *tdstV++ = (((tsrcUV[2] & 0xf) << 6) | ((tsrcUV[1] >> 2) & 0x3f)) << shift[2]; + *tdstU++ = (((tsrcUV[3] & 0x3f) << 4) | ((tsrcUV[2] >> 4) & 0xf)) << shift[1]; + *tdstV++ = (((tsrcUV[4] & 0xff) << 2) | ((tsrcUV[3] >> 6) & 0x3)) << shift[2]; + tsrcUV += 5; + } + src[1] += srcStride[1]; + dstU += dstStride[1] / sizeof(uint16_t); + dstV += dstStride[2] / sizeof(uint16_t); + } + + return srcSliceH; +} + static int planarToP01xWrapper(SwsContext *c, const uint8_t *src8[], int srcStride[], int srcSliceY, int srcSliceH, uint8_t *dstParam8[], @@ -2004,6 +2054,11 @@ void ff_get_unscaled_swscale(SwsContext *c) (srcFormat == AV_PIX_FMT_NV24 || srcFormat == AV_PIX_FMT_NV42)) { c->convert_unscaled = nv24ToPlanarWrapper; } + /* nv15_to_yuv420p10 */ + if ((dstFormat == AV_PIX_FMT_YUV420P10 && srcFormat == AV_PIX_FMT_NV15) || + (dstFormat == AV_PIX_FMT_YUV422P10 && srcFormat == AV_PIX_FMT_NV20NP)) { + c->convert_unscaled = nv15ToPlanarWrapper; + } /* yuv2bgr */ if ((srcFormat == AV_PIX_FMT_YUV420P || srcFormat == AV_PIX_FMT_YUV422P || srcFormat == AV_PIX_FMT_YUVA420P) && isAnyRGB(dstFormat) && diff --git a/libswscale/utils.c b/libswscale/utils.c index e1ad685..a878456 100644 --- a/libswscale/utils.c +++ b/libswscale/utils.c @@ -267,6 +267,8 @@ static const FormatEntry format_entries[] = { [AV_PIX_FMT_RGBAF16LE] = { 1, 0 }, [AV_PIX_FMT_XV30LE] = { 1, 1 }, [AV_PIX_FMT_XV36LE] = { 1, 1 }, + [AV_PIX_FMT_NV15] = { 1, 0 }, + [AV_PIX_FMT_NV20NP] = { 1, 0 }, }; int ff_shuffle_filter_coefficients(SwsContext *c, int *filterPos, -- 2.44.0 From 49ff7fcf6ae1de10007f529c979f7478f2d7a451 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 11 Nov 2023 21:27:42 +0000 Subject: [PATCH 15/18] WIP: v4l2-request: use NV15 and NV20 pix fmt --- libavcodec/v4l2_request.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/libavcodec/v4l2_request.c b/libavcodec/v4l2_request.c index 813c1b3..bf1e3c4 100644 --- a/libavcodec/v4l2_request.c +++ b/libavcodec/v4l2_request.c @@ -197,11 +197,11 @@ static const struct { { V4L2_PIX_FMT_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_LINEAR }, { V4L2_PIX_FMT_SUNXI_TILED_NV12, AV_PIX_FMT_NV12, DRM_FORMAT_NV12, DRM_FORMAT_MOD_ALLWINNER_TILED }, #if defined(V4L2_PIX_FMT_NV15) && defined(DRM_FORMAT_NV15) - { V4L2_PIX_FMT_NV15, AV_PIX_FMT_NONE, DRM_FORMAT_NV15, DRM_FORMAT_MOD_LINEAR }, + { V4L2_PIX_FMT_NV15, AV_PIX_FMT_NV15, DRM_FORMAT_NV15, DRM_FORMAT_MOD_LINEAR }, #endif { V4L2_PIX_FMT_NV16, AV_PIX_FMT_NV16, DRM_FORMAT_NV16, DRM_FORMAT_MOD_LINEAR }, #if defined(V4L2_PIX_FMT_NV20) && defined(DRM_FORMAT_NV20) - { V4L2_PIX_FMT_NV20, AV_PIX_FMT_NONE, DRM_FORMAT_NV20, DRM_FORMAT_MOD_LINEAR }, + { V4L2_PIX_FMT_NV20, AV_PIX_FMT_NV20NP, DRM_FORMAT_NV20, DRM_FORMAT_MOD_LINEAR }, #endif { V4L2_PIX_FMT_P010, AV_PIX_FMT_P010, DRM_FORMAT_P010, DRM_FORMAT_MOD_LINEAR }, #if defined(V4L2_PIX_FMT_YUV420_10_AFBC_16X16_SPLIT) -- 2.44.0 From 0db39ac91179f6893ecaede6936fb12f3e642306 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 4 Nov 2023 19:13:43 +0000 Subject: [PATCH 16/18] WIP: v4l2-request: change to autodetect fixes a review note from prior upstreaming attempt --- configure | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/configure b/configure index 810e57c..bde3d46 100755 --- a/configure +++ b/configure @@ -348,7 +348,7 @@ External library support: --enable-omx-rpi enable OpenMAX IL code for Raspberry Pi [no] --enable-rkmpp enable Rockchip Media Process Platform code [no] --disable-v4l2-m2m disable V4L2 mem2mem code [autodetect] - --enable-v4l2-request enable V4L2 request API code [no] + --disable-v4l2-request disable V4L2 request API code [autodetect] --disable-vaapi disable Video Acceleration API (mainly Unix/Intel) code [autodetect] --disable-vdpau disable Nvidia Video Decode and Presentation API for Unix code [autodetect] --disable-videotoolbox disable VideoToolbox code [autodetect] @@ -1934,6 +1934,7 @@ HWACCEL_AUTODETECT_LIBRARY_LIST=" videotoolbox vulkan v4l2_m2m + v4l2_request " # catchall list of things that require external libs to link @@ -1955,7 +1956,6 @@ HWACCEL_LIBRARY_LIST=" mmal omx opencl - v4l2_request " DOCUMENT_LIST=" -- 2.44.0 From 36a27318e0f280b448fb5ab6e59f4151f6001517 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sat, 4 Nov 2023 21:13:27 +0000 Subject: [PATCH 17/18] libswscale: fix NV16 unscaled conversion --- libswscale/swscale_unscaled.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/libswscale/swscale_unscaled.c b/libswscale/swscale_unscaled.c index 5d054c0..833c2c6 100644 --- a/libswscale/swscale_unscaled.c +++ b/libswscale/swscale_unscaled.c @@ -190,7 +190,7 @@ static int planarToNv24Wrapper(SwsContext *c, const uint8_t *src[], copyPlane(src[0], srcStride[0], srcSliceY, srcSliceH, c->srcW, dstParam[0], dstStride[0]); - if (c->dstFormat == AV_PIX_FMT_NV24) + if (c->dstFormat == AV_PIX_FMT_NV24 || c->dstFormat == AV_PIX_FMT_NV16) interleaveBytes(src[1], src[2], dst, c->chrSrcW, srcSliceH, srcStride[1], srcStride[2], dstStride[1]); else @@ -211,7 +211,7 @@ static int nv24ToPlanarWrapper(SwsContext *c, const uint8_t *src[], copyPlane(src[0], srcStride[0], srcSliceY, srcSliceH, c->srcW, dstParam[0], dstStride[0]); - if (c->srcFormat == AV_PIX_FMT_NV24) + if (c->srcFormat == AV_PIX_FMT_NV24 || c->srcFormat == AV_PIX_FMT_NV16) deinterleaveBytes(src[1], dst1, dst2, c->chrSrcW, srcSliceH, srcStride[1], dstStride[1], dstStride[2]); else @@ -2039,6 +2039,11 @@ void ff_get_unscaled_swscale(SwsContext *c) (dstFormat == AV_PIX_FMT_NV12 || dstFormat == AV_PIX_FMT_NV21)) { c->convert_unscaled = planarToNv12Wrapper; } + /* yv16_to_nv16 */ + if ((srcFormat == AV_PIX_FMT_YUV422P || srcFormat == AV_PIX_FMT_YUVA422P) && + dstFormat == AV_PIX_FMT_NV16) { + c->convert_unscaled = planarToNv24Wrapper; + } /* yv24_to_nv24 */ if ((srcFormat == AV_PIX_FMT_YUV444P || srcFormat == AV_PIX_FMT_YUVA444P) && (dstFormat == AV_PIX_FMT_NV24 || dstFormat == AV_PIX_FMT_NV42)) { @@ -2049,6 +2054,10 @@ void ff_get_unscaled_swscale(SwsContext *c) (srcFormat == AV_PIX_FMT_NV12 || srcFormat == AV_PIX_FMT_NV21)) { c->convert_unscaled = nv12ToPlanarWrapper; } + /* nv16_to_yv16 */ + if (dstFormat == AV_PIX_FMT_YUV422P && srcFormat == AV_PIX_FMT_NV16) { + c->convert_unscaled = nv24ToPlanarWrapper; + } /* nv24_to_yv24 */ if (dstFormat == AV_PIX_FMT_YUV444P && (srcFormat == AV_PIX_FMT_NV24 || srcFormat == AV_PIX_FMT_NV42)) { -- 2.44.0 From 502d19c9564969538952051c2d395c714e41e567 Mon Sep 17 00:00:00 2001 From: Jonas Karlman Date: Sun, 31 Dec 2023 08:56:14 +0000 Subject: [PATCH 18/18] WIP: limit v4l2 udev enum sysname match --- libavcodec/v4l2_request.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/libavcodec/v4l2_request.c b/libavcodec/v4l2_request.c index bf1e3c4..e249dea 100644 --- a/libavcodec/v4l2_request.c +++ b/libavcodec/v4l2_request.c @@ -764,6 +764,7 @@ int ff_v4l2_request_init(AVCodecContext *avctx, uint32_t pixelformat, uint32_t b struct udev_list_entry *devices; struct udev_list_entry *entry; struct udev_device *device; + const char *sysname; av_log(avctx, AV_LOG_DEBUG, "%s: ctx=%p hw_device_ctx=%p hw_frames_ctx=%p\n", __func__, ctx, avctx->hw_device_ctx, avctx->hw_frames_ctx); @@ -786,6 +787,9 @@ int ff_v4l2_request_init(AVCodecContext *avctx, uint32_t pixelformat, uint32_t b } udev_enumerate_add_match_subsystem(enumerate, "media"); + sysname = getenv("V4L2_SYSNAME"); + if (sysname) + udev_enumerate_add_match_sysname(enumerate, sysname); udev_enumerate_scan_devices(enumerate); devices = udev_enumerate_get_list_entry(enumerate); -- 2.44.0