Patches για να δουλεύει το ffmpeg με xvba+libva1

mitsaras77 | Κυρ, 11/08/2009 - 15:49 | 20' | 1

[31/12/2009]Προσοχή: Τα patches που παραθέτω εδώ έχουν περάσει πια επίσημα στο svn repository του ffmpeg project αρκετά βελτιωμένα, αν και λείπουν κάτι μικροαλλαγές ακόμη που εμπεριέχονται στο patch του mplayer-vaapi. Δεν είναι δύσκολο να τις απομονόσετε και να τις περάσετε σαν patch στο ffmpeg που έχετε κατεβάσει στο δίσκο σας.

Πηγή των patches: Από το patch του πακέτου mplayer-vaapi. Για περισσότερα εδώ

Τα patches για Makefile, allcodecs έχουν τροποποιηθεί για την πιό πρόσφατη svn version του ffmpeg. Προσαρμόστε τα στη revision που έχετε.

1) Πρώτο patch

 

 

--- libavcodec/allcodecs.c  
+++ libavcodec/allcodecs.c
@@ -55,6 +55,7 @@
 
     /* hardware accelerators */
     REGISTER_HWACCEL (H263_VAAPI, h263_vaapi);
+    REGISTER_HWACCEL (H264_VAAPI, h264_vaapi);
     REGISTER_HWACCEL (MPEG2_VAAPI, mpeg2_vaapi);
     REGISTER_HWACCEL (MPEG4_VAAPI, mpeg4_vaapi);
     REGISTER_HWACCEL (VC1_VAAPI, vc1_vaapi);

 

2) Δευτερο patch

 

--- libavcodec/Makefile
+++ libavcodec/Makefile
@@ -137,6 +137,7 @@
                                           h264_parser.o cabac.o        \
                                           mpegvideo.o error_resilience.o
 OBJS-$(CONFIG_H264_ENCODER)            += h264enc.o h264dspenc.o
+OBJS-$(CONFIG_H264_VAAPI_HWACCEL)      += h264.o h264idct.o h264pred.o h264_parser.o cabac.o mpegvideo.o error_resilience.o vaapi.o vaapi_h264.o
 OBJS-$(CONFIG_HUFFYUV_DECODER)         += huffyuv.o
 OBJS-$(CONFIG_HUFFYUV_ENCODER)         += huffyuv.o
 OBJS-$(CONFIG_IDCIN_DECODER)           += idcinvideo.o

 

3) Τρίτο patch

 

--- libavcodec/vaapi_h264.c	(révision 0)
+++ libavcodec/vaapi_h264.c	(révision 0)
@@ -0,0 +1,360 @@
+/*
+ * H.264 HW decode acceleration through VAAPI
+ *
+ * Copyright (C) 2008-2009 Splitted-Desktop Systems
+ *
+ * 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 "vaapi_internal.h"
+#include "h264.h"
+
+/** @file 
+ *  This file implements the glue code between FFmpeg's and VAAPI's structures.
+ */
+
+/** Reconstruct bitstream slice_type.
+ */
+static int get_slice_type(H264Context *h)
+{
+    switch (h->slice_type) {
+    case FF_P_TYPE:  return 0;
+    case FF_B_TYPE:  return 1;
+    case FF_I_TYPE:  return 2;
+    case FF_SP_TYPE: return 3;
+    case FF_SI_TYPE: return 4;
+    default:         return -1;
+    }
+}
+
+/** Initialize an empty VAAPI picture.
+ */
+static void init_vaapi_pic(VAPictureH264 *va_pic)
+{
+    va_pic->picture_id          = 0xffffffff;
+    va_pic->flags               = VA_PICTURE_H264_INVALID;
+    va_pic->TopFieldOrderCnt    = 0;
+    va_pic->BottomFieldOrderCnt = 0;
+}
+
+/** Translate an FFmpeg Picture into its VAAPI form.
+ *  @param[out] va_pic          A pointer to VAAPI's own picture struct
+ *  @param[in]  pic             A pointer to the FFmpeg picture struct to convert
+ *  @param[in]  pic_structure   The picture field type (as defined in mpegvideo.h),
+ *                              supersede pic's field type if nonzero.
+ */
+static void fill_vaapi_pic(VAPictureH264  *va_pic,
+                           Picture        *pic,
+                           int             pic_structure)
+{
+    if (pic_structure == 0)
+        pic_structure = pic->reference;
+
+    va_pic->picture_id = ff_vaapi_get_surface(pic);
+    va_pic->frame_idx  = pic->long_ref ? pic->pic_id : pic->frame_num;
+
+    va_pic->flags = 0;
+    if (pic_structure != PICT_FRAME)
+        va_pic->flags |= (pic_structure & PICT_TOP_FIELD) ? VA_PICTURE_H264_TOP_FIELD : VA_PICTURE_H264_BOTTOM_FIELD;
+    if (pic->reference)
+        va_pic->flags |= pic->long_ref ? VA_PICTURE_H264_LONG_TERM_REFERENCE : VA_PICTURE_H264_SHORT_TERM_REFERENCE;
+
+    va_pic->TopFieldOrderCnt = 0;
+    if ((pic_structure & PICT_TOP_FIELD) && pic->field_poc[0] != INT_MAX)
+        va_pic->TopFieldOrderCnt = pic->field_poc[0];
+
+    va_pic->BottomFieldOrderCnt = 0;
+    if ((pic_structure & PICT_BOTTOM_FIELD) && pic->field_poc[1] != INT_MAX)
+        va_pic->BottomFieldOrderCnt = pic->field_poc[1];
+}
+
+/** Decoded picture buffer.
+ */
+typedef struct DPB {
+    unsigned int   size;
+    unsigned int   max_size;
+    VAPictureH264 *pics;
+} DPB;
+
+/** Append picture to the decoded picture buffer, in a VAAPI form that
+ *  merges the second field picture attributes with the first, if available.
+ *  The decoded picture buffer's size must be large enough
+ *  to receive the new VAAPI picture object.
+ *
+ *  VAAPI decoded pictures merge both field when present.
+ */
+static int append_to_vaapi_decoded_pic_buf(DPB *dpb, Picture *pic)
+{
+    unsigned int i;
+
+    if (dpb->size >= dpb->max_size)
+        return -1;
+
+    for (i = 0; i < dpb->size; i++) {
+        VAPictureH264 * const merged_va_pic = &dpb->pics[i];
+
+        /* Check if the reference picture was already refered in a previous pass */
+        if (merged_va_pic->picture_id == ff_vaapi_get_surface(pic)) {
+            /* Don't overwrite our reference, use a temporary */
+            VAPictureH264 tmp_va_pic;
+            fill_vaapi_pic(&tmp_va_pic, pic, 0);
+
+            if ((tmp_va_pic.flags ^ merged_va_pic->flags) & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD)) {
+                /* Merge second field. */
+                merged_va_pic->flags |= tmp_va_pic.flags & (VA_PICTURE_H264_TOP_FIELD | VA_PICTURE_H264_BOTTOM_FIELD);
+
+                if (tmp_va_pic.flags & VA_PICTURE_H264_TOP_FIELD) {
+                    merged_va_pic->TopFieldOrderCnt    = tmp_va_pic.TopFieldOrderCnt;
+                } else {
+                    merged_va_pic->BottomFieldOrderCnt = tmp_va_pic.BottomFieldOrderCnt;
+                }
+            }
+            return 0;
+        }
+    }
+
+    fill_vaapi_pic(&dpb->pics[dpb->size++], pic, 0);
+    return 0;
+}
+
+/** Prepare VAAPI's reference frames array.
+ *  Called when starting decoding a frame, this function fills the
+ *  reference frame array with per-frame values.
+ *  This is required because VAAPI need interlaced fields reference
+ *  to be merged in its reference array.
+ */
+static int prepare_vaapi_reference_frame_array(VAPictureParameterBufferH264 *pic_param, 
+                                               H264Context                  *h)
+{
+    DPB dpb;
+    unsigned int i, list;
+
+    dpb.size     = 0;
+    dpb.max_size = FF_ARRAY_ELEMS(pic_param->ReferenceFrames);
+    dpb.pics     = pic_param->ReferenceFrames;
+    for (i = 0; i < dpb.max_size; i++)
+        init_vaapi_pic(&dpb.pics[i]);
+
+    /* For each refered picture in the current decoding context, 
+       prepare a decoded picture buffer to hold it */
+    for (list = 0; list < h->list_count; list++)
+        for (i = 0; i < h->ref_count[list]; i++) {
+            Picture * const pic = &h->ref_list[list][i];
+            if (pic->reference && append_to_vaapi_decoded_pic_buf(&dpb, pic) < 0)
+                return -1;
+        }
+    return 0;
+}
+
+/** Fill VAAPI reference picture lists.
+ *  Called for each field in slice, it fills the reference picture
+ *  list for that field.
+ *  @param[out] RefPicList  VAAPI internal reference picture list
+ *  @param[in]  ref_list    A pointer to the FFmpeg reference list
+ *  @param[in]  ref_count   The number of reference pictures in ref_list
+ */
+static void fill_vaapi_reference_pic_list(VAPictureH264 RefPicList[32], 
+                                          Picture *     ref_list, 
+                                          unsigned int  ref_count)
+{
+    unsigned int i, n = 0;
+    for (i = 0; i < ref_count; i++)
+        if (ref_list[i].reference)
+            fill_vaapi_pic(&RefPicList[n++], &ref_list[i], 0);
+
+    for (; n < 32; n++)
+        init_vaapi_pic(&RefPicList[n]);
+}
+
+/** Initialize prediction weight table.
+ *  VAAPI requires a plain prediction weight table as it does not infer 
+ *  any value.
+ *  @param[in]  h                   A pointer to the current H.264 context
+ *  @param[in]  list                The reference frame list index to use
+ *  @param[out] luma_weight_flag    VAAPI plain luma weight flag 
+ *  @param[out] luma_weight         VAAPI plain luma weight table 
+ *  @param[out] luma_offset         VAAPI plain luma offset table  
+ *  @param[out] chroma_weight_flag  VAAPI plain chroma weight flag  
+ *  @param[out] chroma_weight       VAAPI plain chroma weight table
+ *  @param[out] chroma_offset       VAAPI plain chroma offset table
+ */
+static void fill_vaapi_plain_pred_weight_table(H264Context *h, int list,
+                                               unsigned char *luma_weight_flag,
+                                               short          luma_weight[32],
+                                               short          luma_offset[32],
+                                               unsigned char *chroma_weight_flag,
+                                               short          chroma_weight[32][2],
+                                               short          chroma_offset[32][2])
+{
+    unsigned int i, j;
+
+    *luma_weight_flag    = h->luma_weight_flag[list];
+    *chroma_weight_flag  = h->chroma_weight_flag[list];
+
+    for (i = 0; i < h->ref_count[list]; i++) {
+        /* VAAPI also wants the inferred (default) values, not
+           only what is available in the bitstream (7.4.3.2). */
+        if (h->luma_weight_flag[list]) {
+            luma_weight[i] = h->luma_weight[list][i];
+            luma_offset[i] = h->luma_offset[list][i];
+        } else {
+            luma_weight[i] = 1 << h->luma_log2_weight_denom;
+            luma_offset[i] = 0;
+        }
+        for (j = 0; j < 2; j++) {
+            if (h->chroma_weight_flag[list]) {
+                chroma_weight[i][j] = h->chroma_weight[list][i][j];
+                chroma_offset[i][j] = h->chroma_offset[list][i][j];
+            } else {
+                chroma_weight[i][j] = 1 << h->chroma_log2_weight_denom;
+                chroma_offset[i][j] = 0;
+            }
+        }
+    }
+}
+
+/** Initialize and start decoding a frame with VAAPI.
+ */
+static int start_frame(AVCodecContext *          avctx, 
+                       av_unused const uint8_t * buffer, 
+                       av_unused uint32_t        size)
+{
+    H264Context * const h = avctx->priv_data;
+    MpegEncContext * const s = &h->s;
+    struct vaapi_context * const vactx = avctx->hwaccel_context;
+    VAPictureParameterBufferH264 *pic_param;
+    VAIQMatrixBufferH264 *iq_matrix;
+
+    dprintf(avctx, "start_frame()n");
+
+    vactx->slice_param_size = sizeof(VASliceParameterBufferH264);
+
+    /* Fill in VAPictureParameterBufferH264. */
+    pic_param = ff_vaapi_alloc_picture(vactx, sizeof(VAPictureParameterBufferH264));
+    if (!pic_param)
+        return -1;
+    fill_vaapi_pic(&pic_param->CurrPic, s->current_picture_ptr, s->picture_structure);
+    if (prepare_vaapi_reference_frame_array(pic_param, h) < 0)
+        return -1;
+    pic_param->picture_width_in_mbs_minus1                      = s->mb_width - 1;
+    pic_param->picture_height_in_mbs_minus1                     = s->mb_height - 1;
+    pic_param->bit_depth_luma_minus8                            = h->sps.bit_depth_luma   - 8;
+    pic_param->bit_depth_chroma_minus8                          = h->sps.bit_depth_chroma - 8;
+    pic_param->num_ref_frames                                   = h->sps.ref_frame_count;
+    pic_param->seq_fields.value                                 = 0; /* reset all bits */
+    pic_param->seq_fields.bits.chroma_format_idc                = h->sps.chroma_format_idc;
+    pic_param->seq_fields.bits.residual_colour_transform_flag   = h->sps.residual_color_transform_flag; /* XXX: only for 4:4:4 high profile? */
+    pic_param->seq_fields.bits.gaps_in_frame_num_value_allowed_flag = h->sps.gaps_in_frame_num_allowed_flag;
+    pic_param->seq_fields.bits.frame_mbs_only_flag              = h->sps.frame_mbs_only_flag;
+    pic_param->seq_fields.bits.mb_adaptive_frame_field_flag     = h->sps.mb_aff;
+    pic_param->seq_fields.bits.direct_8x8_inference_flag        = h->sps.direct_8x8_inference_flag;
+    pic_param->seq_fields.bits.MinLumaBiPredSize8x8             = h->sps.level_idc >= 31; /* A.3.3.2 */
+    pic_param->seq_fields.bits.log2_max_frame_num_minus4        = h->sps.log2_max_frame_num - 4;
+    pic_param->seq_fields.bits.pic_order_cnt_type               = h->sps.poc_type;
+    pic_param->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = h->sps.log2_max_poc_lsb - 4;
+    pic_param->seq_fields.bits.delta_pic_order_always_zero_flag = h->sps.delta_pic_order_always_zero_flag;
+    pic_param->num_slice_groups_minus1                          = h->pps.slice_group_count - 1;
+    pic_param->slice_group_map_type                             = h->pps.mb_slice_group_map_type;
+    pic_param->slice_group_change_rate_minus1                   = 0; /* XXX: unimplemented in FFmpeg */
+    pic_param->pic_init_qp_minus26                              = h->pps.init_qp - 26;
+    pic_param->pic_init_qs_minus26                              = h->pps.init_qs - 26;
+    pic_param->chroma_qp_index_offset                           = h->pps.chroma_qp_index_offset[0];
+    pic_param->second_chroma_qp_index_offset                    = h->pps.chroma_qp_index_offset[1];
+    pic_param->pic_fields.value                                 = 0; /* reset all bits */
+    pic_param->pic_fields.bits.entropy_coding_mode_flag         = h->pps.cabac;
+    pic_param->pic_fields.bits.weighted_pred_flag               = h->pps.weighted_pred;
+    pic_param->pic_fields.bits.weighted_bipred_idc              = h->pps.weighted_bipred_idc;
+    pic_param->pic_fields.bits.transform_8x8_mode_flag          = h->pps.transform_8x8_mode;
+    pic_param->pic_fields.bits.field_pic_flag                   = h->s.picture_structure != PICT_FRAME;
+    pic_param->pic_fields.bits.constrained_intra_pred_flag      = h->pps.constrained_intra_pred;
+    pic_param->pic_fields.bits.pic_order_present_flag           = h->pps.pic_order_present;
+    pic_param->pic_fields.bits.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
+    pic_param->pic_fields.bits.redundant_pic_cnt_present_flag   = h->pps.redundant_pic_cnt_present;
+    pic_param->pic_fields.bits.reference_pic_flag               = h->nal_ref_idc != 0;
+    pic_param->frame_num                                        = h->frame_num;
+
+    /* Fill in VAIQMatrixBufferH264. */
+    iq_matrix = ff_vaapi_alloc_iq_matrix(vactx, sizeof(VAIQMatrixBufferH264));
+    if (!iq_matrix)
+        return -1;
+    memcpy(iq_matrix->ScalingList4x4, h->pps.scaling_matrix4, sizeof(iq_matrix->ScalingList4x4));
+    memcpy(iq_matrix->ScalingList8x8, h->pps.scaling_matrix8, sizeof(iq_matrix->ScalingList8x8));
+    return 0;
+}
+
+/** Finish hardware decoding of the frame.
+ */
+static int end_frame(AVCodecContext *avctx)
+{
+    H264Context * const h = avctx->priv_data;
+
+    dprintf(avctx, "end_frame()n");
+    return ff_vaapi_common_end_frame(&h->s);
+}
+
+/** Decode the given H.264 slice with VAAPI.
+ */
+static int decode_slice(AVCodecContext *avctx,
+                        const uint8_t  *buffer,
+                        uint32_t        size)
+{
+    H264Context * const h = avctx->priv_data;
+    MpegEncContext * const s = &h->s;
+    VASliceParameterBufferH264 *slice_param;
+
+    dprintf(avctx, "decode_slice(): buffer %p, size %dn", buffer, size);
+
+    /* Fill in VASliceParameterBufferH264. */
+    slice_param = (VASliceParameterBufferH264 *)ff_vaapi_alloc_slice(avctx->hwaccel_context, buffer, size);
+    if (!slice_param)
+        return -1;
+    slice_param->slice_data_bit_offset          = get_bits_count(&h->s.gb) + 8; /* bit buffer started beyond nal_unit_type */
+    slice_param->first_mb_in_slice              = (s->mb_y >> FIELD_OR_MBAFF_PICTURE) * s->mb_width + s->mb_x;
+    slice_param->slice_type                     = get_slice_type(h);
+    slice_param->direct_spatial_mv_pred_flag    = h->slice_type == FF_B_TYPE ? h->direct_spatial_mv_pred : 0;
+    slice_param->num_ref_idx_l0_active_minus1   = h->ref_count[0] - 1;
+    slice_param->num_ref_idx_l1_active_minus1   = h->ref_count[1] - 1;
+    slice_param->cabac_init_idc                 = h->cabac_init_idc;
+    slice_param->slice_qp_delta                 = s->qscale - h->pps.init_qp;
+    slice_param->disable_deblocking_filter_idc  = h->deblocking_filter < 2 ? !h->deblocking_filter : h->deblocking_filter;
+    slice_param->slice_alpha_c0_offset_div2     = h->slice_alpha_c0_offset / 2;
+    slice_param->slice_beta_offset_div2         = h->slice_beta_offset / 2;
+    fill_vaapi_reference_pic_list(slice_param->RefPicList0, h->ref_list[0], h->list_count > 0 ? h->ref_count[0] : 0);
+    fill_vaapi_reference_pic_list(slice_param->RefPicList1, h->ref_list[1], h->list_count > 1 ? h->ref_count[1] : 0);
+    slice_param->luma_log2_weight_denom         = h->luma_log2_weight_denom;
+    slice_param->chroma_log2_weight_denom       = h->chroma_log2_weight_denom;
+
+    fill_vaapi_plain_pred_weight_table(h, 0,
+                                       &slice_param->luma_weight_l0_flag,   slice_param->luma_weight_l0,   slice_param->luma_offset_l0,
+                                       &slice_param->chroma_weight_l0_flag, slice_param->chroma_weight_l0, slice_param->chroma_offset_l0);
+    fill_vaapi_plain_pred_weight_table(h, 1,
+                                       &slice_param->luma_weight_l1_flag,   slice_param->luma_weight_l1,   slice_param->luma_offset_l1,
+                                       &slice_param->chroma_weight_l1_flag, slice_param->chroma_weight_l1, slice_param->chroma_offset_l1);
+    return 0;
+}
+
+AVHWAccel h264_vaapi_hwaccel = {
+    .name           = "h264_vaapi",
+    .type           = CODEC_TYPE_VIDEO,
+    .id             = CODEC_ID_H264,
+    .pix_fmt        = PIX_FMT_VAAPI_VLD,
+    .capabilities   = 0,
+    .start_frame    = start_frame,
+    .end_frame      = end_frame,
+    .decode_slice   = decode_slice,
+    .priv_data_size = 0,
+}

  

 

Εφαρμόστε copy/paste σε κάθε ένα patch στον αγαπημένο σας text editor και δώστε τους ένα όνομα. Μετά από τερματικό εφαρμόστε αυτές τις εντολές

svn checkout svn://svn.mplayerhq.hu/ffmpeg/trunk ffmpeg

cd ffmpeg

patch -p0 < /full_path_of_saved_patches

./configure ...

κλπ

Οσοι έχουν debian, μπορούν να χρησιμοποιήσουν και το gquilt (γραφικό εργαλείο διαχήρεισης patches).

Καλό hacking!!!

Δώσε αστέρια!
Σχόλια