[FFmpeg-devel] [PATCH v3] hwcontext_vaapi: Use PRIME_2 memory type for modifiers.
Xiang, Haihao
haihao.xiang at intel.com
Wed Jun 30 11:47:44 EEST 2021
On Sat, 2021-05-15 at 17:53 +0200, Bas Nieuwenhuizen wrote:
> This way we can pass explicit modifiers in. Sometimes the
> modifier matters for the number of memory planes that
> libva accepts, in particular when dealing with
> driver-compressed textures. Furthermore the driver might
> not actually be able to determine the implicit modifier
> if all the buffer-passing has used explicit modifier.
> All these issues should be resolved by passing in the
> modifier, and for that we switch to using the PRIME_2
> memory type.
>
> Tested with experimental radeonsi patches for modifiers
> and kmsgrab. Also tested with radeonsi without the
> patches to double-check it works without PRIME_2 support.
>
> v2:
> Cache PRIME_2 support to avoid doing two calls every time on
> libva drivers that do not support it.
>
> v3:
> Remove prime2_vas usage.
> ---
> libavutil/hwcontext_vaapi.c | 158 ++++++++++++++++++++++++++----------
> 1 file changed, 114 insertions(+), 44 deletions(-)
>
> diff --git a/libavutil/hwcontext_vaapi.c b/libavutil/hwcontext_vaapi.c
> index 83e542876d..75acc851d6 100644
> --- a/libavutil/hwcontext_vaapi.c
> +++ b/libavutil/hwcontext_vaapi.c
> @@ -79,6 +79,9 @@ typedef struct VAAPIFramesContext {
> unsigned int rt_format;
> // Whether vaDeriveImage works.
> int derive_works;
> + // Caches whether VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2 is unsupported
> for
> + // surface imports.
> + int prime_2_import_unsupported;
> } VAAPIFramesContext;
>
> typedef struct VAAPIMapping {
> @@ -1022,32 +1025,17 @@ static void vaapi_unmap_from_drm(AVHWFramesContext
> *dst_fc,
> static int vaapi_map_from_drm(AVHWFramesContext *src_fc, AVFrame *dst,
> const AVFrame *src, int flags)
> {
> + VAAPIFramesContext *src_vafc = src_fc->internal->priv;
The first parameter 'src_fc' is actually a AVHWFramesContext pointer for dst,
not for src.
vaapi_map_from_drm() is called in vaapi_map_to()
See
https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/hwcontext_vaapi.c#L1408-L1418
and vaapi_map_to() is the implementation of map_to callback. It passes the
AVHWFramesContext pointer for dst to map_to() when map_to() is called, see
https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/hwcontext.c#L833-L841
I ever provided http://ffmpeg.org/pipermail/ffmpeg-devel/2020-June/264765.htmlto fix it.
> AVHWFramesContext *dst_fc =
> (AVHWFramesContext*)dst->hw_frames_ctx->data;
> AVVAAPIDeviceContext *dst_dev = dst_fc->device_ctx->hwctx;
> const AVDRMFrameDescriptor *desc;
> const VAAPIFormatDescriptor *format_desc;
> VASurfaceID surface_id;
> - VAStatus vas;
> + VAStatus vas = VA_STATUS_SUCCESS;
> + int use_prime2;
> uint32_t va_fourcc;
> - int err, i, j, k;
> -
> - unsigned long buffer_handle;
> - VASurfaceAttribExternalBuffers buffer_desc;
> - VASurfaceAttrib attrs[2] = {
> - {
> - .type = VASurfaceAttribMemoryType,
> - .flags = VA_SURFACE_ATTRIB_SETTABLE,
> - .value.type = VAGenericValueTypeInteger,
> - .value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME,
> - },
> - {
> - .type = VASurfaceAttribExternalBufferDescriptor,
> - .flags = VA_SURFACE_ATTRIB_SETTABLE,
> - .value.type = VAGenericValueTypePointer,
> - .value.value.p = &buffer_desc,
> - }
> - };
> + int err, i, j;
>
> desc = (AVDRMFrameDescriptor*)src->data[0];
Can the check below removed for PRIME_2 ?
https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/hwcontext_vaapi.c#L1054-L1058
if (desc->nb_objects != 1) {
av_log(dst_fc, AV_LOG_ERROR, "VAAPI can only map frames "
"made from a single DRM object.\n");
return AVERROR(EINVAL);
}
Thanks
Haihao
>
> @@ -1083,35 +1071,117 @@ static int vaapi_map_from_drm(AVHWFramesContext
> *src_fc, AVFrame *dst,
> format_desc = vaapi_format_from_fourcc(va_fourcc);
> av_assert0(format_desc);
>
> - buffer_handle = desc->objects[0].fd;
> - buffer_desc.pixel_format = va_fourcc;
> - buffer_desc.width = src_fc->width;
> - buffer_desc.height = src_fc->height;
> - buffer_desc.data_size = desc->objects[0].size;
> - buffer_desc.buffers = &buffer_handle;
> - buffer_desc.num_buffers = 1;
> - buffer_desc.flags = 0;
> -
> - k = 0;
> - for (i = 0; i < desc->nb_layers; i++) {
> - for (j = 0; j < desc->layers[i].nb_planes; j++) {
> - buffer_desc.pitches[k] = desc->layers[i].planes[j].pitch;
> - buffer_desc.offsets[k] = desc->layers[i].planes[j].offset;
> - ++k;
> + use_prime2 = !src_vafc->prime_2_import_unsupported &&
> + desc->objects[0].format_modifier != DRM_FORMAT_MOD_INVALID;
> + if (use_prime2) {
> + VADRMPRIMESurfaceDescriptor prime_desc;
> + VASurfaceAttrib prime_attrs[2] = {
> + {
> + .type = VASurfaceAttribMemoryType,
> + .flags = VA_SURFACE_ATTRIB_SETTABLE,
> + .value.type = VAGenericValueTypeInteger,
> + .value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2,
> + },
> + {
> + .type = VASurfaceAttribExternalBufferDescriptor,
> + .flags = VA_SURFACE_ATTRIB_SETTABLE,
> + .value.type = VAGenericValueTypePointer,
> + .value.value.p = &prime_desc,
> + }
> + };
> + prime_desc.fourcc = va_fourcc;
> + prime_desc.width = src_fc->width;
> + prime_desc.height = src_fc->height;
> + prime_desc.num_objects = desc->nb_objects;
> + for (i = 0; i < desc->nb_objects; ++i) {
> + prime_desc.objects[i].fd = desc->objects[i].fd;
> + prime_desc.objects[i].size = desc->objects[i].size;
> + prime_desc.objects[i].drm_format_modifier =
> + desc->objects[i].format_modifier;
> }
> - }
> - buffer_desc.num_planes = k;
>
> - if (format_desc->chroma_planes_swapped &&
> - buffer_desc.num_planes == 3) {
> - FFSWAP(uint32_t, buffer_desc.pitches[1], buffer_desc.pitches[2]);
> - FFSWAP(uint32_t, buffer_desc.offsets[1], buffer_desc.offsets[2]);
> + prime_desc.num_layers = desc->nb_layers;
> + for (i = 0; i < desc->nb_layers; ++i) {
> + prime_desc.layers[i].drm_format = desc->layers[i].format;
> + prime_desc.layers[i].num_planes = desc->layers[i].nb_planes;
> + for (j = 0; j < desc->layers[i].nb_planes; ++j) {
> + prime_desc.layers[i].object_index[j] =
> + desc->layers[i].planes[j].object_index;
> + prime_desc.layers[i].offset[j] = desc-
> >layers[i].planes[j].offset;
> + prime_desc.layers[i].pitch[j] = desc-
> >layers[i].planes[j].pitch;
> + }
> +
> + if (format_desc->chroma_planes_swapped &&
> + desc->layers[i].nb_planes == 3) {
> + FFSWAP(uint32_t, prime_desc.layers[i].pitch[1],
> + prime_desc.layers[i].pitch[2]);
> + FFSWAP(uint32_t, prime_desc.layers[i].offset[1],
> + prime_desc.layers[i].offset[2]);
> + }
> + }
> +
> + /*
> + * We can query for PRIME_2 support with vaQuerySurfaceAttributes,
> but that
> + * that needs the config_id which we don't have here . Both Intel and
> + * Gallium seem to do the correct error checks, so lets just try the
> + * PRIME_2 import first.
> + */
> + vas = vaCreateSurfaces(dst_dev->display, format_desc->rt_format,
> + src->width, src->height, &surface_id, 1,
> + prime_attrs, FF_ARRAY_ELEMS(prime_attrs));
> + if (vas != VA_STATUS_SUCCESS)
> + src_vafc->prime_2_import_unsupported = 1;
> }
>
> - vas = vaCreateSurfaces(dst_dev->display, format_desc->rt_format,
> - src->width, src->height,
> - &surface_id, 1,
> - attrs, FF_ARRAY_ELEMS(attrs));
> + if (!use_prime2 || vas != VA_STATUS_SUCCESS) {
> + int k;
> + unsigned long buffer_handle;
> + VASurfaceAttribExternalBuffers buffer_desc;
> + VASurfaceAttrib buffer_attrs[2] = {
> + {
> + .type = VASurfaceAttribMemoryType,
> + .flags = VA_SURFACE_ATTRIB_SETTABLE,
> + .value.type = VAGenericValueTypeInteger,
> + .value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME,
> + },
> + {
> + .type = VASurfaceAttribExternalBufferDescriptor,
> + .flags = VA_SURFACE_ATTRIB_SETTABLE,
> + .value.type = VAGenericValueTypePointer,
> + .value.value.p = &buffer_desc,
> + }
> + };
> +
> + buffer_handle = desc->objects[0].fd;
> + buffer_desc.pixel_format = va_fourcc;
> + buffer_desc.width = src_fc->width;
> + buffer_desc.height = src_fc->height;
> + buffer_desc.data_size = desc->objects[0].size;
> + buffer_desc.buffers = &buffer_handle;
> + buffer_desc.num_buffers = 1;
> + buffer_desc.flags = 0;
> +
> + k = 0;
> + for (i = 0; i < desc->nb_layers; i++) {
> + for (j = 0; j < desc->layers[i].nb_planes; j++) {
> + buffer_desc.pitches[k] = desc->layers[i].planes[j].pitch;
> + buffer_desc.offsets[k] = desc->layers[i].planes[j].offset;
> + ++k;
> + }
> + }
> + buffer_desc.num_planes = k;
> +
> + if (format_desc->chroma_planes_swapped &&
> + buffer_desc.num_planes == 3) {
> + FFSWAP(uint32_t, buffer_desc.pitches[1], buffer_desc.pitches[2]);
> + FFSWAP(uint32_t, buffer_desc.offsets[1], buffer_desc.offsets[2]);
> + }
> +
> + vas = vaCreateSurfaces(dst_dev->display, format_desc->rt_format,
> + src->width, src->height,
> + &surface_id, 1,
> + buffer_attrs, FF_ARRAY_ELEMS(buffer_attrs));
> + }
> if (vas != VA_STATUS_SUCCESS) {
> av_log(dst_fc, AV_LOG_ERROR, "Failed to create surface from DRM "
> "object: %d (%s).\n", vas, vaErrorStr(vas));
More information about the ffmpeg-devel
mailing list