From 551c6775abb5e0ad34c26d7e23bc6fbbe8ccc9d4 Mon Sep 17 00:00:00 2001 From: Mark Thompson Date: Mon, 15 Feb 2016 19:32:45 +0000 Subject: lavu: VAAPI hwcontext implementation Signed-off-by: Anton Khirnov --- libavutil/Makefile | 3 + libavutil/hwcontext.c | 3 + libavutil/hwcontext.h | 1 + libavutil/hwcontext_internal.h | 1 + libavutil/hwcontext_vaapi.c | 850 +++++++++++++++++++++++++++++++++++++++++ libavutil/hwcontext_vaapi.h | 82 ++++ libavutil/version.h | 2 +- 7 files changed, 941 insertions(+), 1 deletion(-) create mode 100644 libavutil/hwcontext_vaapi.c create mode 100644 libavutil/hwcontext_vaapi.h (limited to 'libavutil') diff --git a/libavutil/Makefile b/libavutil/Makefile index a095f0dd6e..0956703ce8 100644 --- a/libavutil/Makefile +++ b/libavutil/Makefile @@ -25,6 +25,7 @@ HEADERS = adler32.h \ hmac.h \ hwcontext.h \ hwcontext_cuda.h \ + hwcontext_vaapi.h \ hwcontext_vdpau.h \ imgutils.h \ intfloat.h \ @@ -108,11 +109,13 @@ OBJS = adler32.o \ OBJS-$(CONFIG_LZO) += lzo.o OBJS-$(CONFIG_CUDA) += hwcontext_cuda.o +OBJS-$(CONFIG_VAAPI) += hwcontext_vaapi.o OBJS-$(CONFIG_VDPAU) += hwcontext_vdpau.o OBJS += $(COMPAT_OBJS:%=../compat/%) SKIPHEADERS-$(CONFIG_CUDA) += hwcontext_cuda.h +SKIPHEADERS-$(CONFIG_VAAPI) += hwcontext_vaapi.h SKIPHEADERS-$(CONFIG_VDPAU) += hwcontext_vdpau.h SKIPHEADERS-$(HAVE_ATOMICS_GCC) += atomic_gcc.h SKIPHEADERS-$(HAVE_ATOMICS_SUNCC) += atomic_suncc.h diff --git a/libavutil/hwcontext.c b/libavutil/hwcontext.c index 53e11b8eae..ac1e2c9d67 100644 --- a/libavutil/hwcontext.c +++ b/libavutil/hwcontext.c @@ -32,6 +32,9 @@ static const HWContextType *hw_table[] = { #if CONFIG_CUDA &ff_hwcontext_type_cuda, #endif +#if CONFIG_VAAPI + &ff_hwcontext_type_vaapi, +#endif #if CONFIG_VDPAU &ff_hwcontext_type_vdpau, #endif diff --git a/libavutil/hwcontext.h b/libavutil/hwcontext.h index 681b55566c..8502342409 100644 --- a/libavutil/hwcontext.h +++ b/libavutil/hwcontext.h @@ -27,6 +27,7 @@ enum AVHWDeviceType { AV_HWDEVICE_TYPE_VDPAU, AV_HWDEVICE_TYPE_CUDA, + AV_HWDEVICE_TYPE_VAAPI, }; typedef struct AVHWDeviceInternal AVHWDeviceInternal; diff --git a/libavutil/hwcontext_internal.h b/libavutil/hwcontext_internal.h index 27de1f9e62..b7828c0350 100644 --- a/libavutil/hwcontext_internal.h +++ b/libavutil/hwcontext_internal.h @@ -97,6 +97,7 @@ struct AVHWFramesInternal { }; extern const HWContextType ff_hwcontext_type_cuda; +extern const HWContextType ff_hwcontext_type_vaapi; extern const HWContextType ff_hwcontext_type_vdpau; #endif /* AVUTIL_HWCONTEXT_INTERNAL_H */ diff --git a/libavutil/hwcontext_vaapi.c b/libavutil/hwcontext_vaapi.c new file mode 100644 index 0000000000..1a385ba514 --- /dev/null +++ b/libavutil/hwcontext_vaapi.c @@ -0,0 +1,850 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "avassert.h" +#include "buffer.h" +#include "common.h" +#include "hwcontext.h" +#include "hwcontext_internal.h" +#include "hwcontext_vaapi.h" +#include "mem.h" +#include "pixdesc.h" +#include "pixfmt.h" + +typedef struct VAAPISurfaceFormat { + enum AVPixelFormat pix_fmt; + VAImageFormat image_format; +} VAAPISurfaceFormat; + +typedef struct VAAPIDeviceContext { + // Surface formats which can be used with this device. + VAAPISurfaceFormat *formats; + int nb_formats; +} VAAPIDeviceContext; + +typedef struct VAAPIFramesContext { + // Surface attributes set at create time. + VASurfaceAttrib *attributes; + int nb_attributes; + // RT format of the underlying surface (Intel driver ignores this anyway). + unsigned int rt_format; + // Whether vaDeriveImage works. + int derive_works; +} VAAPIFramesContext; + +enum { + VAAPI_MAP_READ = 0x01, + VAAPI_MAP_WRITE = 0x02, + VAAPI_MAP_DIRECT = 0x04, +}; + +typedef struct VAAPISurfaceMap { + // The source hardware frame of this mapping (with hw_frames_ctx set). + const AVFrame *source; + // VAAPI_MAP_* flags which apply to this mapping. + int flags; + // Handle to the derived or copied image which is mapped. + VAImage image; +} VAAPISurfaceMap; + +#define MAP(va, rt, av) { \ + VA_FOURCC_ ## va, \ + VA_RT_FORMAT_ ## rt, \ + AV_PIX_FMT_ ## av \ + } +// The map fourcc <-> pix_fmt isn't bijective because of the annoying U/V +// plane swap cases. The frame handling below tries to hide these. +static struct { + unsigned int fourcc; + unsigned int rt_format; + enum AVPixelFormat pix_fmt; +} vaapi_format_map[] = { + MAP(NV12, YUV420, NV12), + MAP(YV12, YUV420, YUV420P), // With U/V planes swapped. + MAP(IYUV, YUV420, YUV420P), + //MAP(I420, YUV420, YUV420P), // Not in libva but used by Intel driver. +#ifdef VA_FOURCC_YV16 + MAP(YV16, YUV422, YUV422P), // With U/V planes swapped. +#endif + MAP(422H, YUV422, YUV422P), + MAP(UYVY, YUV422, UYVY422), + MAP(YUY2, YUV422, YUYV422), + MAP(Y800, YUV400, GRAY8), +#ifdef VA_FOURCC_P010 + //MAP(P010, YUV420_10BPP, P010), +#endif + MAP(BGRA, RGB32, BGRA), + //MAP(BGRX, RGB32, BGR0), + MAP(RGBA, RGB32, RGBA), + //MAP(RGBX, RGB32, RGB0), + MAP(ABGR, RGB32, ABGR), + //MAP(XBGR, RGB32, 0BGR), + MAP(ARGB, RGB32, ARGB), + //MAP(XRGB, RGB32, 0RGB), +}; +#undef MAP + +static enum AVPixelFormat vaapi_pix_fmt_from_fourcc(unsigned int fourcc) +{ + int i; + for (i = 0; i < FF_ARRAY_ELEMS(vaapi_format_map); i++) + if (vaapi_format_map[i].fourcc == fourcc) + return vaapi_format_map[i].pix_fmt; + return AV_PIX_FMT_NONE; +} + +static int vaapi_get_image_format(AVHWDeviceContext *hwdev, + enum AVPixelFormat pix_fmt, + VAImageFormat **image_format) +{ + VAAPIDeviceContext *ctx = hwdev->internal->priv; + int i; + + for (i = 0; i < ctx->nb_formats; i++) { + if (ctx->formats[i].pix_fmt == pix_fmt) { + *image_format = &ctx->formats[i].image_format; + return 0; + } + } + return AVERROR(EINVAL); +} + +static int vaapi_frames_get_constraints(AVHWDeviceContext *hwdev, + const void *hwconfig, + AVHWFramesConstraints *constraints) +{ + AVVAAPIDeviceContext *hwctx = hwdev->hwctx; + const AVVAAPIHWConfig *config = hwconfig; + AVVAAPIHWConfig *tmp_config; + VASurfaceAttrib *attr_list = NULL; + VAStatus vas; + enum AVPixelFormat pix_fmt; + unsigned int fourcc; + int err, i, j, attr_count, pix_fmt_count; + + if (!hwconfig) { + // No configuration was provided, so we create a temporary pipeline + // configuration in order to query all supported image formats. + + tmp_config = av_mallocz(sizeof(*config)); + if (!tmp_config) + return AVERROR(ENOMEM); + + vas = vaCreateConfig(hwctx->display, + VAProfileNone, VAEntrypointVideoProc, + NULL, 0, &tmp_config->config_id); + if (vas != VA_STATUS_SUCCESS) { + // No vpp. We might still be able to do something useful if + // codecs are supported, so try to make the most-commonly + // supported decoder configuration we can to query instead. + vas = vaCreateConfig(hwctx->display, + VAProfileH264ConstrainedBaseline, + VAEntrypointVLD, NULL, 0, + &tmp_config->config_id); + if (vas != VA_STATUS_SUCCESS) { + av_freep(&tmp_config); + return AVERROR(ENOSYS); + } + } + + config = tmp_config; + } + + attr_count = 0; + vas = vaQuerySurfaceAttributes(hwctx->display, config->config_id, + 0, &attr_count); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwdev, AV_LOG_ERROR, "Failed to query surface attributes: " + "%d (%s).\n", vas, vaErrorStr(vas)); + err = AVERROR(ENOSYS); + goto fail; + } + + attr_list = av_malloc(attr_count * sizeof(*attr_list)); + if (!attr_list) { + err = AVERROR(ENOMEM); + goto fail; + } + + vas = vaQuerySurfaceAttributes(hwctx->display, config->config_id, + attr_list, &attr_count); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwdev, AV_LOG_ERROR, "Failed to query surface attributes: " + "%d (%s).\n", vas, vaErrorStr(vas)); + err = AVERROR(ENOSYS); + goto fail; + } + + pix_fmt_count = 0; + for (i = 0; i < attr_count; i++) { + switch (attr_list[i].type) { + case VASurfaceAttribPixelFormat: + fourcc = attr_list[i].value.value.i; + pix_fmt = vaapi_pix_fmt_from_fourcc(fourcc); + if (pix_fmt != AV_PIX_FMT_NONE) { + ++pix_fmt_count; + } else { + // Something unsupported - ignore. + } + break; + case VASurfaceAttribMinWidth: + constraints->min_width = attr_list[i].value.value.i; + break; + case VASurfaceAttribMinHeight: + constraints->min_height = attr_list[i].value.value.i; + break; + case VASurfaceAttribMaxWidth: + constraints->max_width = attr_list[i].value.value.i; + break; + case VASurfaceAttribMaxHeight: + constraints->max_height = attr_list[i].value.value.i; + break; + } + } + if (pix_fmt_count == 0) { + // Nothing usable found. Presumably there exists something which + // works, so leave the set null to indicate unknown. + constraints->valid_sw_formats = NULL; + } else { + constraints->valid_sw_formats = av_malloc_array(pix_fmt_count + 1, + sizeof(pix_fmt)); + if (!constraints->valid_sw_formats) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (i = j = 0; i < attr_count; i++) { + if (attr_list[i].type != VASurfaceAttribPixelFormat) + continue; + fourcc = attr_list[i].value.value.i; + pix_fmt = vaapi_pix_fmt_from_fourcc(fourcc); + if (pix_fmt != AV_PIX_FMT_NONE) + constraints->valid_sw_formats[j++] = pix_fmt; + } + av_assert0(j == pix_fmt_count); + constraints->valid_sw_formats[j] = AV_PIX_FMT_NONE; + } + + constraints->valid_hw_formats = av_malloc_array(2, sizeof(pix_fmt)); + if (!constraints->valid_hw_formats) { + err = AVERROR(ENOMEM); + goto fail; + } + constraints->valid_hw_formats[0] = AV_PIX_FMT_VAAPI; + constraints->valid_hw_formats[1] = AV_PIX_FMT_NONE; + + err = 0; +fail: + av_freep(&attr_list); + if (!hwconfig) { + vaDestroyConfig(hwctx->display, tmp_config->config_id); + av_freep(&tmp_config); + } + return err; +} + +static int vaapi_device_init(AVHWDeviceContext *hwdev) +{ + VAAPIDeviceContext *ctx = hwdev->internal->priv; + AVVAAPIDeviceContext *hwctx = hwdev->hwctx; + AVHWFramesConstraints *constraints = NULL; + VAImageFormat *image_list = NULL; + VAStatus vas; + int err, i, j, image_count; + enum AVPixelFormat pix_fmt; + unsigned int fourcc; + + constraints = av_mallocz(sizeof(*constraints)); + if (!constraints) + goto fail; + + err = vaapi_frames_get_constraints(hwdev, NULL, constraints); + if (err < 0) + goto fail; + + image_count = vaMaxNumImageFormats(hwctx->display); + if (image_count <= 0) { + err = AVERROR(EIO); + goto fail; + } + image_list = av_malloc(image_count * sizeof(*image_list)); + if (!image_list) { + err = AVERROR(ENOMEM); + goto fail; + } + vas = vaQueryImageFormats(hwctx->display, image_list, &image_count); + if (vas != VA_STATUS_SUCCESS) { + err = AVERROR(EIO); + goto fail; + } + + ctx->formats = av_malloc(image_count * sizeof(*ctx->formats)); + if (!ctx->formats) { + err = AVERROR(ENOMEM); + goto fail; + } + ctx->nb_formats = 0; + for (i = 0; i < image_count; i++) { + fourcc = image_list[i].fourcc; + pix_fmt = vaapi_pix_fmt_from_fourcc(fourcc); + for (j = 0; constraints->valid_sw_formats[j] != AV_PIX_FMT_NONE; j++) { + if (pix_fmt == constraints->valid_sw_formats[j]) + break; + } + if (constraints->valid_sw_formats[j] != AV_PIX_FMT_NONE) { + av_log(hwdev, AV_LOG_DEBUG, "Format %#x -> %s.\n", + fourcc, av_get_pix_fmt_name(pix_fmt)); + ctx->formats[ctx->nb_formats].pix_fmt = pix_fmt; + ctx->formats[ctx->nb_formats].image_format = image_list[i]; + ++ctx->nb_formats; + } else { + av_log(hwdev, AV_LOG_DEBUG, "Format %#x -> unknown.\n", fourcc); + } + } + + av_free(image_list); + av_hwframe_constraints_free(&constraints); + return 0; +fail: + av_freep(&ctx->formats); + av_free(image_list); + av_hwframe_constraints_free(&constraints); + return err; +} + +static void vaapi_device_uninit(AVHWDeviceContext *hwdev) +{ + VAAPIDeviceContext *ctx = hwdev->internal->priv; + + av_freep(&ctx->formats); +} + +static void vaapi_buffer_free(void *opaque, uint8_t *data) +{ + AVHWFramesContext *hwfc = opaque; + AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx; + VASurfaceID surface_id; + VAStatus vas; + + surface_id = (VASurfaceID)(uintptr_t)data; + + vas = vaDestroySurfaces(hwctx->display, &surface_id, 1); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to destroy surface %#x: " + "%d (%s).\n", surface_id, vas, vaErrorStr(vas)); + } +} + +static AVBufferRef *vaapi_pool_alloc(void *opaque, int size) +{ + AVHWFramesContext *hwfc = opaque; + VAAPIFramesContext *ctx = hwfc->internal->priv; + AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx; + AVVAAPIFramesContext *avfc = hwfc->hwctx; + VASurfaceID surface_id; + VAStatus vas; + AVBufferRef *ref; + + vas = vaCreateSurfaces(hwctx->display, ctx->rt_format, + hwfc->width, hwfc->height, + &surface_id, 1, + ctx->attributes, ctx->nb_attributes); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to create surface: " + "%d (%s).\n", vas, vaErrorStr(vas)); + return NULL; + } + av_log(hwfc, AV_LOG_DEBUG, "Created surface %#x.\n", surface_id); + + ref = av_buffer_create((uint8_t*)(uintptr_t)surface_id, + sizeof(surface_id), &vaapi_buffer_free, + hwfc, AV_BUFFER_FLAG_READONLY); + if (!ref) { + vaDestroySurfaces(hwctx->display, &surface_id, 1); + return NULL; + } + + if (hwfc->initial_pool_size > 0) { + // This is a fixed-size pool, so we must still be in the initial + // allocation sequence. + av_assert0(avfc->nb_surfaces < hwfc->initial_pool_size); + avfc->surface_ids[avfc->nb_surfaces] = surface_id; + ++avfc->nb_surfaces; + } + + return ref; +} + +static int vaapi_frames_init(AVHWFramesContext *hwfc) +{ + AVVAAPIFramesContext *avfc = hwfc->hwctx; + VAAPIFramesContext *ctx = hwfc->internal->priv; + AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx; + VAImageFormat *expected_format; + AVBufferRef *test_surface = NULL; + VASurfaceID test_surface_id; + VAImage test_image; + VAStatus vas; + int err, i; + unsigned int fourcc, rt_format; + + for (i = 0; i < FF_ARRAY_ELEMS(vaapi_format_map); i++) { + if (vaapi_format_map[i].pix_fmt == hwfc->sw_format) { + fourcc = vaapi_format_map[i].fourcc; + rt_format = vaapi_format_map[i].rt_format; + break; + } + } + if (i >= FF_ARRAY_ELEMS(vaapi_format_map)) { + av_log(hwfc, AV_LOG_ERROR, "Unsupported format: %s.\n", + av_get_pix_fmt_name(hwfc->sw_format)); + return AVERROR(EINVAL); + } + + if (!hwfc->pool) { + int need_memory_type = 1, need_pixel_format = 1; + for (i = 0; i < avfc->nb_attributes; i++) { + if (ctx->attributes[i].type == VASurfaceAttribMemoryType) + need_memory_type = 0; + if (ctx->attributes[i].type == VASurfaceAttribPixelFormat) + need_pixel_format = 0; + } + ctx->nb_attributes = + avfc->nb_attributes + need_memory_type + need_pixel_format; + + ctx->attributes = av_malloc(ctx->nb_attributes * + sizeof(*ctx->attributes)); + if (!ctx->attributes) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (i = 0; i < avfc->nb_attributes; i++) + ctx->attributes[i] = avfc->attributes[i]; + if (need_memory_type) { + ctx->attributes[i++] = (VASurfaceAttrib) { + .type = VASurfaceAttribMemoryType, + .flags = VA_SURFACE_ATTRIB_SETTABLE, + .value.type = VAGenericValueTypeInteger, + .value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_VA, + }; + } + if (need_pixel_format) { + ctx->attributes[i++] = (VASurfaceAttrib) { + .type = VASurfaceAttribPixelFormat, + .flags = VA_SURFACE_ATTRIB_SETTABLE, + .value.type = VAGenericValueTypeInteger, + .value.value.i = fourcc, + }; + } + av_assert0(i == ctx->nb_attributes); + + ctx->rt_format = rt_format; + + if (hwfc->initial_pool_size > 0) { + // This pool will be usable as a render target, so we need to store + // all of the surface IDs somewhere that vaCreateContext() calls + // will be able to access them. + avfc->nb_surfaces = 0; + avfc->surface_ids = av_malloc(hwfc->initial_pool_size * + sizeof(*avfc->surface_ids)); + if (!avfc->surface_ids) { + err = AVERROR(ENOMEM); + goto fail; + } + } else { + // This pool allows dynamic sizing, and will not be usable as a + // render target. + avfc->nb_surfaces = 0; + avfc->surface_ids = NULL; + } + + hwfc->internal->pool_internal = + av_buffer_pool_init2(sizeof(VASurfaceID), hwfc, + &vaapi_pool_alloc, NULL); + if (!hwfc->internal->pool_internal) { + av_log(hwfc, AV_LOG_ERROR, "Failed to create VAAPI surface pool.\n"); + err = AVERROR(ENOMEM); + goto fail; + } + } + + // Allocate a single surface to test whether vaDeriveImage() is going + // to work for the specific configuration. + if (hwfc->pool) { + test_surface = av_buffer_pool_get(hwfc->pool); + if (!test_surface) { + av_log(hwfc, AV_LOG_ERROR, "Unable to allocate a surface from " + "user-configured buffer pool.\n"); + err = AVERROR(ENOMEM); + goto fail; + } + } else { + test_surface = av_buffer_pool_get(hwfc->internal->pool_internal); + if (!test_surface) { + av_log(hwfc, AV_LOG_ERROR, "Unable to allocate a surface from " + "internal buffer pool.\n"); + err = AVERROR(ENOMEM); + goto fail; + } + } + test_surface_id = (VASurfaceID)(uintptr_t)test_surface->data; + + ctx->derive_works = 0; + + err = vaapi_get_image_format(hwfc->device_ctx, + hwfc->sw_format, &expected_format); + if (err == 0) { + vas = vaDeriveImage(hwctx->display, test_surface_id, &test_image); + if (vas == VA_STATUS_SUCCESS) { + if (expected_format->fourcc == test_image.format.fourcc) { + av_log(hwfc, AV_LOG_DEBUG, "Direct mapping possible.\n"); + ctx->derive_works = 1; + } else { + av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: " + "derived image format %08x does not match " + "expected format %08x.\n", + expected_format->fourcc, test_image.format.fourcc); + } + vaDestroyImage(hwctx->display, test_image.image_id); + } else { + av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: " + "deriving image does not work: " + "%d (%s).\n", vas, vaErrorStr(vas)); + } + } else { + av_log(hwfc, AV_LOG_DEBUG, "Direct mapping disabled: " + "image format is not supported.\n"); + } + + av_buffer_unref(&test_surface); + return 0; + +fail: + av_buffer_unref(&test_surface); + av_freep(&avfc->surface_ids); + av_freep(&ctx->attributes); + return err; +} + +static void vaapi_frames_uninit(AVHWFramesContext *hwfc) +{ + AVVAAPIFramesContext *avfc = hwfc->hwctx; + VAAPIFramesContext *ctx = hwfc->internal->priv; + + av_freep(&avfc->surface_ids); + av_freep(&ctx->attributes); +} + +static int vaapi_get_buffer(AVHWFramesContext *hwfc, AVFrame *frame) +{ + frame->buf[0] = av_buffer_pool_get(hwfc->pool); + if (!frame->buf[0]) + return AVERROR(ENOMEM); + + frame->data[3] = frame->buf[0]->data; + frame->format = AV_PIX_FMT_VAAPI; + frame->width = hwfc->width; + frame->height = hwfc->height; + + return 0; +} + +static int vaapi_transfer_get_formats(AVHWFramesContext *hwfc, + enum AVHWFrameTransferDirection dir, + enum AVPixelFormat **formats) +{ + VAAPIDeviceContext *ctx = hwfc->device_ctx->internal->priv; + enum AVPixelFormat *pix_fmts, preferred_format; + int i, k; + + preferred_format = hwfc->sw_format; + + pix_fmts = av_malloc((ctx->nb_formats + 1) * sizeof(*pix_fmts)); + if (!pix_fmts) + return AVERROR(ENOMEM); + + pix_fmts[0] = preferred_format; + k = 1; + for (i = 0; i < ctx->nb_formats; i++) { + if (ctx->formats[i].pix_fmt == preferred_format) + continue; + av_assert0(k < ctx->nb_formats); + pix_fmts[k++] = ctx->formats[i].pix_fmt; + } + av_assert0(k == ctx->nb_formats); + pix_fmts[k] = AV_PIX_FMT_NONE; + + *formats = pix_fmts; + return 0; +} + +static void vaapi_unmap_frame(void *opaque, uint8_t *data) +{ + AVHWFramesContext *hwfc = opaque; + AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx; + VAAPISurfaceMap *map = (VAAPISurfaceMap*)data; + const AVFrame *src; + VASurfaceID surface_id; + VAStatus vas; + + src = map->source; + surface_id = (VASurfaceID)(uintptr_t)src->data[3]; + av_log(hwfc, AV_LOG_DEBUG, "Unmap surface %#x.\n", surface_id); + + vas = vaUnmapBuffer(hwctx->display, map->image.buf); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to unmap image from surface " + "%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas)); + } + + if ((map->flags & VAAPI_MAP_WRITE) && + !(map->flags & VAAPI_MAP_DIRECT)) { + vas = vaPutImage(hwctx->display, surface_id, map->image.image_id, + 0, 0, hwfc->width, hwfc->height, + 0, 0, hwfc->width, hwfc->height); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to write image to surface " + "%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas)); + } + } + + vas = vaDestroyImage(hwctx->display, map->image.image_id); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to destroy image from surface " + "%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas)); + } + + av_free(map); +} + +static int vaapi_map_frame(AVHWFramesContext *hwfc, + AVFrame *dst, const AVFrame *src, int flags) +{ + AVVAAPIDeviceContext *hwctx = hwfc->device_ctx->hwctx; + VAAPIFramesContext *ctx = hwfc->internal->priv; + VASurfaceID surface_id; + VAImageFormat *image_format; + VAAPISurfaceMap *map; + VAStatus vas; + void *address = NULL; + int err, i; + + surface_id = (VASurfaceID)(uintptr_t)src->data[3]; + av_log(hwfc, AV_LOG_DEBUG, "Map surface %#x.\n", surface_id); + + if (!ctx->derive_works && (flags & VAAPI_MAP_DIRECT)) { + // Requested direct mapping but it is not possible. + return AVERROR(EINVAL); + } + if (dst->format == AV_PIX_FMT_NONE) + dst->format = hwfc->sw_format; + if (dst->format != hwfc->sw_format && (flags & VAAPI_MAP_DIRECT)) { + // Requested direct mapping but the formats do not match. + return AVERROR(EINVAL); + } + + err = vaapi_get_image_format(hwfc->device_ctx, dst->format, &image_format); + if (err < 0) { + // Requested format is not a valid output format. + return AVERROR(EINVAL); + } + + map = av_malloc(sizeof(VAAPISurfaceMap)); + if (!map) + return AVERROR(ENOMEM); + + map->source = src; + map->flags = flags; + map->image.image_id = VA_INVALID_ID; + + vas = vaSyncSurface(hwctx->display, surface_id); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to sync surface " + "%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas)); + err = AVERROR(EIO); + goto fail; + } + + // The memory which we map using derive need not be connected to the CPU + // in a way conducive to fast access. On Gen7-Gen9 Intel graphics, the + // memory is mappable but not cached, so normal memcpy()-like access is + // very slow to read it (but writing is ok). It is possible to read much + // faster with a copy routine which is aware of the limitation, but we + // assume for now that the user is not aware of that and would therefore + // prefer not to be given direct-mapped memory if they request read access. + if (ctx->derive_works && + ((flags & VAAPI_MAP_DIRECT) || !(flags & VAAPI_MAP_READ))) { + vas = vaDeriveImage(hwctx->display, surface_id, &map->image); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to derive image from " + "surface %#x: %d (%s).\n", + surface_id, vas, vaErrorStr(vas)); + err = AVERROR(EIO); + goto fail; + } + if (map->image.format.fourcc != image_format->fourcc) { + av_log(hwfc, AV_LOG_ERROR, "Derive image of surface %#x " + "is in wrong format: expected %#08x, got %#08x.\n", + surface_id, image_format->fourcc, map->image.format.fourcc); + err = AVERROR(EIO); + goto fail; + } + map->flags |= VAAPI_MAP_DIRECT; + } else { + vas = vaCreateImage(hwctx->display, image_format, + hwfc->width, hwfc->height, &map->image); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to create image for " + "surface %#x: %d (%s).\n", + surface_id, vas, vaErrorStr(vas)); + err = AVERROR(EIO); + goto fail; + } + if (flags & VAAPI_MAP_READ) { + vas = vaGetImage(hwctx->display, surface_id, 0, 0, + hwfc->width, hwfc->height, map->image.image_id); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to read image from " + "surface %#x: %d (%s).\n", + surface_id, vas, vaErrorStr(vas)); + err = AVERROR(EIO); + goto fail; + } + } + } + + vas = vaMapBuffer(hwctx->display, map->image.buf, &address); + if (vas != VA_STATUS_SUCCESS) { + av_log(hwfc, AV_LOG_ERROR, "Failed to map image from surface " + "%#x: %d (%s).\n", surface_id, vas, vaErrorStr(vas)); + err = AVERROR(EIO); + goto fail; + } + + dst->width = src->width; + dst->height = src->height; + + for (i = 0; i < map->image.num_planes; i++) { + dst->data[i] = (uint8_t*)address + map->image.offsets[i]; + dst->linesize[i] = map->image.pitches[i]; + } + if ( +#ifdef VA_FOURCC_YV16 + map->image.format.fourcc == VA_FOURCC_YV16 || +#endif + map->image.format.fourcc == VA_FOURCC_YV12) { + // Chroma planes are YVU rather than YUV, so swap them. + FFSWAP(uint8_t*, dst->data[1], dst->data[2]); + } + + dst->buf[0] = av_buffer_create((uint8_t*)map, sizeof(*map), + &vaapi_unmap_frame, hwfc, 0); + if (!dst->buf[0]) { + err = AVERROR(ENOMEM); + goto fail; + } + + return 0; + +fail: + if (map) { + if (address) + vaUnmapBuffer(hwctx->display, map->image.buf); + if (map->image.image_id != VA_INVALID_ID) + vaDestroyImage(hwctx->display, map->image.image_id); + av_free(map); + } + return err; +} + +static int vaapi_transfer_data_from(AVHWFramesContext *hwfc, + AVFrame *dst, const AVFrame *src) +{ + AVFrame *map; + int err; + + map = av_frame_alloc(); + if (!map) + return AVERROR(ENOMEM); + map->format = dst->format; + + err = vaapi_map_frame(hwfc, map, src, VAAPI_MAP_READ); + if (err) + goto fail; + + err = av_frame_copy(dst, map); + if (err) + goto fail; + + err = 0; +fail: + av_frame_free(&map); + return err; +} + +static int vaapi_transfer_data_to(AVHWFramesContext *hwfc, + AVFrame *dst, const AVFrame *src) +{ + AVFrame *map; + int err; + + map = av_frame_alloc(); + if (!map) + return AVERROR(ENOMEM); + map->format = src->format; + + err = vaapi_map_frame(hwfc, map, dst, VAAPI_MAP_WRITE); + if (err) + goto fail; + + err = av_frame_copy(map, src); + if (err) + goto fail; + + err = 0; +fail: + av_frame_free(&map); + return err; +} + +const HWContextType ff_hwcontext_type_vaapi = { + .type = AV_HWDEVICE_TYPE_VAAPI, + .name = "VAAPI", + + .device_hwctx_size = sizeof(AVVAAPIDeviceContext), + .device_priv_size = sizeof(VAAPIDeviceContext), + .device_hwconfig_size = sizeof(AVVAAPIHWConfig), + .frames_hwctx_size = sizeof(AVVAAPIFramesContext), + .frames_priv_size = sizeof(VAAPIFramesContext), + + .device_init = &vaapi_device_init, + .device_uninit = &vaapi_device_uninit, + .frames_get_constraints = &vaapi_frames_get_constraints, + .frames_init = &vaapi_frames_init, + .frames_uninit = &vaapi_frames_uninit, + .frames_get_buffer = &vaapi_get_buffer, + .transfer_get_formats = &vaapi_transfer_get_formats, + .transfer_data_to = &vaapi_transfer_data_to, + .transfer_data_from = &vaapi_transfer_data_from, + + .pix_fmts = (const enum AVPixelFormat[]) { + AV_PIX_FMT_VAAPI, + AV_PIX_FMT_NONE + }, +}; diff --git a/libavutil/hwcontext_vaapi.h b/libavutil/hwcontext_vaapi.h new file mode 100644 index 0000000000..1c87f5db5e --- /dev/null +++ b/libavutil/hwcontext_vaapi.h @@ -0,0 +1,82 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVUTIL_HWCONTEXT_VAAPI_H +#define AVUTIL_HWCONTEXT_VAAPI_H + +#include + +/** + * @file + * API-specific header for AV_HWDEVICE_TYPE_VAAPI. + * + * Dynamic frame pools are supported, but note that any pool used as a render + * target is required to be of fixed size in order to be be usable as an + * argument to vaCreateContext(). + * + * For user-allocated pools, AVHWFramesContext.pool must return AVBufferRefs + * with the data pointer set to a VASurfaceID. + */ + +/** + * VAAPI connection details. + * + * Allocated as AVHWDeviceContext.hwctx + */ +typedef struct AVVAAPIDeviceContext { + /** + * The VADisplay handle, to be filled by the user. + */ + VADisplay display; +} AVVAAPIDeviceContext; + +/** + * VAAPI-specific data associated with a frame pool. + * + * Allocated as AVHWFramesContext.hwctx. + */ +typedef struct AVVAAPIFramesContext { + /** + * Set by the user to apply surface attributes to all surfaces in + * the frame pool. If null, default settings are used. + */ + VASurfaceAttrib *attributes; + int nb_attributes; + /** + * The surfaces IDs of all surfaces in the pool after creation. + * Only valid if AVHWFramesContext.initial_pool_size was positive. + * These are intended to be used as the render_targets arguments to + * vaCreateContext(). + */ + VASurfaceID *surface_ids; + int nb_surfaces; +} AVVAAPIFramesContext; + +/** + * VAAPI hardware pipeline configuration details. + * + * Allocated with av_hwdevice_hwconfig_alloc(). + */ +typedef struct AVVAAPIHWConfig { + /** + * ID of a VAAPI pipeline configuration. + */ + VAConfigID config_id; +} AVVAAPIHWConfig; + +#endif /* AVUTIL_HWCONTEXT_VAAPI_H */ diff --git a/libavutil/version.h b/libavutil/version.h index 72eb5f4605..fdd27e3f0b 100644 --- a/libavutil/version.h +++ b/libavutil/version.h @@ -54,7 +54,7 @@ */ #define LIBAVUTIL_VERSION_MAJOR 55 -#define LIBAVUTIL_VERSION_MINOR 8 +#define LIBAVUTIL_VERSION_MINOR 9 #define LIBAVUTIL_VERSION_MICRO 0 #define LIBAVUTIL_VERSION_INT AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \ -- cgit v1.2.3