[PATCH v5 00/20] H.264 Field Decoding Support for Frame-based Decoders

Nicolas Dufresne posted 20 patches 1 week, 3 days ago
.../vcodec/vdec/vdec_h264_req_common.c        |  21 +-
.../vcodec/vdec/vdec_h264_req_common.h        |  11 +-
.../mediatek/vcodec/vdec/vdec_h264_req_if.c   |  15 +-
.../vcodec/vdec/vdec_h264_req_multi_if.c      |  27 +-
.../media/platform/nvidia/tegra-vde/h264.c    |  19 +-
drivers/media/v4l2-core/v4l2-h264.c           | 271 +++++++++++++++---
.../staging/media/hantro/hantro_g1_h264_dec.c |  38 +--
drivers/staging/media/hantro/hantro_h264.c    | 134 +++++++--
drivers/staging/media/hantro/hantro_hw.h      |   8 +-
drivers/staging/media/hantro/hantro_v4l2.c    |  25 ++
.../media/hantro/rockchip_vpu2_hw_h264_dec.c  |  98 +++----
drivers/staging/media/rkvdec/rkvdec-h264.c    | 157 +++++++---
drivers/staging/media/rkvdec/rkvdec.c         |  35 +--
drivers/staging/media/rkvdec/rkvdec.h         |   2 +
include/media/v4l2-h264.h                     |  31 +-
15 files changed, 646 insertions(+), 246 deletions(-)
[PATCH v5 00/20] H.264 Field Decoding Support for Frame-based Decoders
Posted by Nicolas Dufresne 1 week, 3 days ago
NOTE: This is a rebased series to solve merge conflicts, some patches have been
      skipped as they are already in media_stage tree

Until now, only Cedrus (a slice base decoder) supported interlaced
decoding.  In order to support field decoding in our frame-based decoder,
the v4l2-h264 library needed adaptation to produce the appropriate
reference lists.

This patch extends the v4l2-h264 library to produce the larger references
list needed to represent fields separately. Hantro, MTK-VCODEC and RKVDEC
drivers have been adapted to accommodate the larger lists. Though, only
Hantro and RKVDEC actually have HW support for field decoding. So only
these two have been updated to make use of the larger lists. All this work
has been done using the H.264 specification, LibreELEC downstream kernel
patches, Rockchip MPP reference software and Hantro reference software.

For reviewers, the following is the map of all commit. Patches that could
be merge independently of this serie are marked as independent. Note that
the test results do depend on the generic fixes.

01-07 : v4l2-h264 field decoding support
08-14 : rkvdec h.264 generic fixes (independent)
15-16 : rkvdec h.264 field decoding support
17-20 : hantro h.264 field decoding support

All this work have been tested using GStreamer mainline implementation
but also with FFMPEG LibreELEC fork using the testing tool fluster
running through the ITU-T H.264 (2016-02) AVCv2 set of bitstream. Before
this patch, the scores were:

Hantro:
  FFMPEG:   88/135
  GSteamer: 90/135
RKVDEC:
  FFMPEG:   73/135
  GSteamer: 77/135

And after these changes:

Hantro:
  FFMPEG:   118/135
  GSteamer: 129/135
RKVDEC:
  FFMPEG:   118/135
  GSteamer: 129/135

Note that a bug in FFMPEG / LibreELEC fork was noticed and fixed with the
following change:

diff --git a/libavcodec/v4l2_request_h264.c b/libavcodec/v4l2_request_h264.c
index 88da8f0a2d..394bae0550 100644
--- a/libavcodec/v4l2_request_h264.c
+++ b/libavcodec/v4l2_request_h264.c
@@ -66,7 +66,7 @@ static void fill_dpb_entry(struct v4l2_h264_dpb_entry *entry, const H264Picture
 {
     entry->reference_ts = ff_v4l2_request_get_capture_timestamp(pic->f);
     entry->pic_num = pic->pic_id;
-    entry->frame_num = pic->frame_num;
+    entry->frame_num = pic->long_ref ? pic->pic_id : pic->frame_num;
     entry->fields = pic->reference & V4L2_H264_FRAME_REF;
     entry->flags = V4L2_H264_DPB_ENTRY_FLAG_VALID;
     if (entry->fields)

Some useful links:

Detailed Hantro Results:     https://gitlab.freedesktop.org/-/snippets/5189
Detailed RKVDEC Results:     https://gitlab.freedesktop.org/-/snippets/5253
ITU-T H.264 (2016-02) AVCv2: https://www.itu.int/net/itu-t/sigdb/spevideo/VideoForm-s.aspx?val=102002641
Fluster:                     https://github.com/fluendo/fluster
GStreamer:                   https://gitlab.freedesktop.org/gstreamer/gstreamer/
FFMPEG Fork:                 https://github.com/jernejsk/FFmpeg/tree/v4l2-request-hwaccel-4.4
Rockchip MPP:                https://github.com/rockchip-linux/mpp

Changes in v5:
- Rebased on media_stage tree
- Fixed merge conflict due to some Meditatek vcodec driver refactoring
- Fixes made by Hans in his pull request v2 (build fix)
- Fixed xmas coding style in Tegra driver (from a late review comment)

Changes in v4:
- Fixed mtk-vcodec port
- Ported tegra-vde driver (compiled tested only)
- Applied Hans' commit message, comment and documentation suggestions

Changes in v3:
- Improved debug message on timestamp miss-match
- Moved H264 SPS validation into rkvdec-h264
- Added more comments around H264 SPS validation
- Also validate at streamon (rkvdec start())
- Applied more Review-by and Fixes tag
- Fixed Signed-off-by chain in Jonas patch

Changes in v2:
- Applied most of Sebastian's suggestion in comments and commit messages.
- Use a bool for dpb_valid and dpb_bottom in rkvdec
- Dropped one wrong typo fix (media: v4l2-mem2mem: Fix typo in trace message)
- Dropped Alex fix (media: rkvdec-h264: Don't hardcode SPS/PPS parameters
  + I will carry this one later, it seems cosmetic

Jonas Karlman (5):
  media: rkvdec: h264: Fix bit depth wrap in pps packet
  media: rkvdec: h264: Validate and use pic width and height in mbs
  media: rkvdec: h264: Fix reference frame_num wrap for second field
  media: rkvdec: Ensure decoded resolution fit coded resolution
  media: hantro: h264: Make dpb entry management more robust

Nicolas Dufresne (15):
  media: h264: Use v4l2_h264_reference for reflist
  media: h264: Increase reference lists size to 32
  media: h264: Store current picture fields
  media: h264: Store all fields into the unordered list
  media: v4l2: Trace calculated p/b0/b1 initial reflist
  media: h264: Sort p/b reflist using frame_num
  media: v4l2: Reorder field reflist
  media: rkvdec: Stop overclocking the decoder
  media: rkvdec: h264: Fix dpb_valid implementation
  media: rkvdec: Move H264 SPS validation in rkvdec-h264
  media: rkvdec-h264: Add field decoding support
  media: rkvdec: Enable capture buffer holding for H264
  media: hantro: Stop using H.264 parameter pic_num
  media: hantro: Add H.264 field decoding support
  media: hantro: Enable HOLD_CAPTURE_BUF for H.264

 .../vcodec/vdec/vdec_h264_req_common.c        |  21 +-
 .../vcodec/vdec/vdec_h264_req_common.h        |  11 +-
 .../mediatek/vcodec/vdec/vdec_h264_req_if.c   |  15 +-
 .../vcodec/vdec/vdec_h264_req_multi_if.c      |  27 +-
 .../media/platform/nvidia/tegra-vde/h264.c    |  19 +-
 drivers/media/v4l2-core/v4l2-h264.c           | 271 +++++++++++++++---
 .../staging/media/hantro/hantro_g1_h264_dec.c |  38 +--
 drivers/staging/media/hantro/hantro_h264.c    | 134 +++++++--
 drivers/staging/media/hantro/hantro_hw.h      |   8 +-
 drivers/staging/media/hantro/hantro_v4l2.c    |  25 ++
 .../media/hantro/rockchip_vpu2_hw_h264_dec.c  |  98 +++----
 drivers/staging/media/rkvdec/rkvdec-h264.c    | 157 +++++++---
 drivers/staging/media/rkvdec/rkvdec.c         |  35 +--
 drivers/staging/media/rkvdec/rkvdec.h         |   2 +
 include/media/v4l2-h264.h                     |  31 +-
 15 files changed, 646 insertions(+), 246 deletions(-)

-- 
2.34.3
[PATCH v5 01/20] media: h264: Use v4l2_h264_reference for reflist
Posted by Nicolas Dufresne 1 week, 3 days ago
In preparation for adding field decoding support, convert the byte arrays
for reflist into array of struct v4l2_h264_reference. That struct will
allow us to mark which field of the reference picture is being referenced.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Tested-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
[hverkuil: top_field_order_cnt -> pic_order_count]
---
 .../vcodec/vdec/vdec_h264_req_common.c        | 21 +++-
 .../vcodec/vdec/vdec_h264_req_common.h        | 11 ++-
 .../mediatek/vcodec/vdec/vdec_h264_req_if.c   | 15 ++-
 .../vcodec/vdec/vdec_h264_req_multi_if.c      | 27 +++--
 .../media/platform/nvidia/tegra-vde/h264.c    | 19 ++--
 drivers/media/v4l2-core/v4l2-h264.c           | 33 ++++---
 .../staging/media/hantro/hantro_g1_h264_dec.c | 38 +++----
 drivers/staging/media/hantro/hantro_hw.h      |  6 +-
 .../media/hantro/rockchip_vpu2_hw_h264_dec.c  | 98 +++++++++----------
 drivers/staging/media/rkvdec/rkvdec-h264.c    | 12 +--
 include/media/v4l2-h264.h                     | 19 ++--
 11 files changed, 167 insertions(+), 132 deletions(-)

diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.c
index 3c75a7b4e845..b6931729e4ea 100644
--- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.c
+++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.c
@@ -12,11 +12,24 @@
 #define GET_MTK_VDEC_PARAM(param) \
 	{ dst_param->param = src_param->param; }
 
-/*
- * The firmware expects unused reflist entries to have the value 0x20.
- */
-void mtk_vdec_h264_fixup_ref_list(u8 *ref_list, size_t num_valid)
+void mtk_vdec_h264_get_ref_list(u8 *ref_list,
+				const struct v4l2_h264_reference * v4l2_ref_list,
+				int num_valid)
 {
+	u32 i;
+
+	/*
+	 * TODO The firmware does not support field decoding. Future
+	 * implementation must use v4l2_ref_list[i].fields to obtain
+	 * the reference field parity.
+	 */
+
+	for (i = 0; i < num_valid; i++)
+		ref_list[i] = v4l2_ref_list[i].index;
+
+	/*
+	 * The firmware expects unused reflist entries to have the value 0x20.
+	 */
 	memset(&ref_list[num_valid], 0x20, 32 - num_valid);
 }
 
diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.h b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.h
index 0113f380b491..416277c70af3 100644
--- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.h
+++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_common.h
@@ -164,12 +164,15 @@ struct h264_fb {
 };
 
 /**
- * mtk_vdec_h264_fixup_ref_list - fixup unused reference to 0x20.
+ * mtk_vdec_h264_get_ref_list - translate V4L2 reference list
  *
- * @ref_list:	reference picture list
- * @num_valid:	used reference number
+ * @ref_list:		Mediatek reference picture list
+ * @v4l2_ref_list:	V4L2 reference picture list
+ * @num_valid:		used reference number
  */
-void mtk_vdec_h264_fixup_ref_list(u8 *ref_list, size_t num_valid);
+void mtk_vdec_h264_get_ref_list(u8 *ref_list,
+				const struct v4l2_h264_reference * v4l2_ref_list,
+				int num_valid);
 
 /**
  * mtk_vdec_h264_get_ctrl_ptr - get each CID contrl address.
diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_if.c
index b055ceea481d..4bc05ab5afea 100644
--- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_if.c
+++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_if.c
@@ -102,6 +102,9 @@ static int get_vdec_decode_parameters(struct vdec_h264_slice_inst *inst)
 	const struct v4l2_ctrl_h264_scaling_matrix *scaling_matrix;
 	struct mtk_h264_dec_slice_param *slice_param = &inst->h264_slice_param;
 	struct v4l2_h264_reflist_builder reflist_builder;
+	struct v4l2_h264_reference v4l2_p0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b1_reflist[V4L2_H264_REF_LIST_LEN];
 	u8 *p0_reflist = slice_param->decode_params.ref_pic_list_p0;
 	u8 *b0_reflist = slice_param->decode_params.ref_pic_list_b0;
 	u8 *b1_reflist = slice_param->decode_params.ref_pic_list_b1;
@@ -137,12 +140,14 @@ static int get_vdec_decode_parameters(struct vdec_h264_slice_inst *inst)
 	/* Build the reference lists */
 	v4l2_h264_init_reflist_builder(&reflist_builder, dec_params, sps,
 				       inst->dpb);
-	v4l2_h264_build_p_ref_list(&reflist_builder, p0_reflist);
-	v4l2_h264_build_b_ref_lists(&reflist_builder, b0_reflist, b1_reflist);
+	v4l2_h264_build_p_ref_list(&reflist_builder, v4l2_p0_reflist);
+	v4l2_h264_build_b_ref_lists(&reflist_builder, v4l2_b0_reflist,
+				    v4l2_b1_reflist);
+
 	/* Adapt the built lists to the firmware's expectations */
-	mtk_vdec_h264_fixup_ref_list(p0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b1_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(p0_reflist, v4l2_p0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b0_reflist, v4l2_b0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b1_reflist, v4l2_b1_reflist, reflist_builder.num_valid);
 
 	memcpy(&inst->vsi_ctx.h264_slice_params, slice_param,
 	       sizeof(inst->vsi_ctx.h264_slice_params));
diff --git a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
index 1d9e753cf894..784d01f8bd50 100644
--- a/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
+++ b/drivers/media/platform/mediatek/vcodec/vdec/vdec_h264_req_multi_if.c
@@ -222,6 +222,9 @@ static int get_vdec_sig_decode_parameters(struct vdec_h264_slice_inst *inst)
 	const struct v4l2_ctrl_h264_scaling_matrix *scaling_matrix;
 	struct vdec_h264_slice_lat_dec_param *slice_param = &inst->h264_slice_param;
 	struct v4l2_h264_reflist_builder reflist_builder;
+	struct v4l2_h264_reference v4l2_p0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b1_reflist[V4L2_H264_REF_LIST_LEN];
 	u8 *p0_reflist = slice_param->decode_params.ref_pic_list_p0;
 	u8 *b0_reflist = slice_param->decode_params.ref_pic_list_b0;
 	u8 *b1_reflist = slice_param->decode_params.ref_pic_list_b1;
@@ -256,13 +259,14 @@ static int get_vdec_sig_decode_parameters(struct vdec_h264_slice_inst *inst)
 
 	/* Build the reference lists */
 	v4l2_h264_init_reflist_builder(&reflist_builder, dec_params, sps, inst->dpb);
-	v4l2_h264_build_p_ref_list(&reflist_builder, p0_reflist);
+	v4l2_h264_build_p_ref_list(&reflist_builder, v4l2_p0_reflist);
+	v4l2_h264_build_b_ref_lists(&reflist_builder, v4l2_b0_reflist, v4l2_b1_reflist);
 
-	v4l2_h264_build_b_ref_lists(&reflist_builder, b0_reflist, b1_reflist);
 	/* Adapt the built lists to the firmware's expectations */
-	mtk_vdec_h264_fixup_ref_list(p0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b1_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(p0_reflist, v4l2_p0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b0_reflist, v4l2_b0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b1_reflist, v4l2_b1_reflist, reflist_builder.num_valid);
+
 	memcpy(&inst->vsi_ctx.h264_slice_params, slice_param,
 	       sizeof(inst->vsi_ctx.h264_slice_params));
 
@@ -276,6 +280,9 @@ static void vdec_h264_slice_fill_decode_reflist(struct vdec_h264_slice_inst *ins
 	struct v4l2_ctrl_h264_decode_params *dec_params = &share_info->dec_params;
 	struct v4l2_ctrl_h264_sps *sps = &share_info->sps;
 	struct v4l2_h264_reflist_builder reflist_builder;
+	struct v4l2_h264_reference v4l2_p0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b0_reflist[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference v4l2_b1_reflist[V4L2_H264_REF_LIST_LEN];
 	u8 *p0_reflist = slice_param->decode_params.ref_pic_list_p0;
 	u8 *b0_reflist = slice_param->decode_params.ref_pic_list_b0;
 	u8 *b1_reflist = slice_param->decode_params.ref_pic_list_b1;
@@ -291,13 +298,13 @@ static void vdec_h264_slice_fill_decode_reflist(struct vdec_h264_slice_inst *ins
 	/* Build the reference lists */
 	v4l2_h264_init_reflist_builder(&reflist_builder, dec_params, sps,
 				       inst->dpb);
-	v4l2_h264_build_p_ref_list(&reflist_builder, p0_reflist);
-	v4l2_h264_build_b_ref_lists(&reflist_builder, b0_reflist, b1_reflist);
+	v4l2_h264_build_p_ref_list(&reflist_builder, v4l2_p0_reflist);
+	v4l2_h264_build_b_ref_lists(&reflist_builder, v4l2_b0_reflist, v4l2_b1_reflist);
 
 	/* Adapt the built lists to the firmware's expectations */
-	mtk_vdec_h264_fixup_ref_list(p0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b0_reflist, reflist_builder.num_valid);
-	mtk_vdec_h264_fixup_ref_list(b1_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(p0_reflist, v4l2_p0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b0_reflist, v4l2_b0_reflist, reflist_builder.num_valid);
+	mtk_vdec_h264_get_ref_list(b1_reflist, v4l2_b1_reflist, reflist_builder.num_valid);
 }
 
 static int vdec_h264_slice_alloc_mv_buf(struct vdec_h264_slice_inst *inst,
diff --git a/drivers/media/platform/nvidia/tegra-vde/h264.c b/drivers/media/platform/nvidia/tegra-vde/h264.c
index d8e5534e80c8..4fb0aaad16d6 100644
--- a/drivers/media/platform/nvidia/tegra-vde/h264.c
+++ b/drivers/media/platform/nvidia/tegra-vde/h264.c
@@ -45,9 +45,9 @@ struct tegra_vde_h264_decoder_ctx {
 };
 
 struct h264_reflists {
-	u8 p[V4L2_H264_NUM_DPB_ENTRIES];
-	u8 b0[V4L2_H264_NUM_DPB_ENTRIES];
-	u8 b1[V4L2_H264_NUM_DPB_ENTRIES];
+	struct v4l2_h264_reference p[V4L2_H264_NUM_DPB_ENTRIES];
+	struct v4l2_h264_reference b0[V4L2_H264_NUM_DPB_ENTRIES];
+	struct v4l2_h264_reference b1[V4L2_H264_NUM_DPB_ENTRIES];
 };
 
 static int tegra_vde_wait_mbe(struct tegra_vde *vde)
@@ -765,10 +765,10 @@ static int tegra_vde_h264_setup_frames(struct tegra_ctx *ctx,
 	struct tegra_m2m_buffer *tb = vb_to_tegra_buf(&dst->vb2_buf);
 	struct tegra_ctx_h264 *h = &ctx->h264;
 	struct v4l2_h264_reflist_builder b;
+	struct v4l2_h264_reference *dpb_id;
 	struct h264_reflists reflists;
 	struct vb2_buffer *ref;
 	unsigned int i;
-	u8 *dpb_id;
 	int err;
 
 	/*
@@ -811,14 +811,16 @@ static int tegra_vde_h264_setup_frames(struct tegra_ctx *ctx,
 	}
 
 	for (i = 0; i < b.num_valid; i++) {
-		ref = get_ref_buf(ctx, dst, dpb_id[i]);
+		int dpb_idx = dpb_id[i].index;
 
-		err = tegra_vde_h264_setup_frame(ctx, h264, &b, ref, dpb_id[i],
+		ref = get_ref_buf(ctx, dst, dpb_idx);
+
+		err = tegra_vde_h264_setup_frame(ctx, h264, &b, ref, dpb_idx,
 						 h264->dpb_frames_nb++);
 		if (err)
 			return err;
 
-		if (b.refs[dpb_id[i]].pic_order_count < b.cur_pic_order_count)
+		if (b.refs[dpb_idx].pic_order_count < b.cur_pic_order_count)
 			h264->dpb_ref_frames_with_earlier_poc_nb++;
 	}
 
@@ -880,6 +882,9 @@ static int tegra_vde_h264_setup_context(struct tegra_ctx *ctx,
 	if (h->pps->flags & V4L2_H264_PPS_FLAG_ENTROPY_CODING_MODE)
 		return -EOPNOTSUPP;
 
+	if (h->decode_params->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC)
+		return -EOPNOTSUPP;
+
 	if (h->sps->profile_idc == 66)
 		h264->baseline_profile = 1;
 
diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index ac47519a9fbe..afbfcf78efe4 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -75,12 +75,12 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 			pic_order_count = dpb[i].top_field_order_cnt;
 
 		b->refs[i].pic_order_count = pic_order_count;
-		b->unordered_reflist[b->num_valid] = i;
+		b->unordered_reflist[b->num_valid].index = i;
 		b->num_valid++;
 	}
 
 	for (i = b->num_valid; i < ARRAY_SIZE(b->unordered_reflist); i++)
-		b->unordered_reflist[i] = i;
+		b->unordered_reflist[i].index = i;
 }
 EXPORT_SYMBOL_GPL(v4l2_h264_init_reflist_builder);
 
@@ -90,8 +90,8 @@ static int v4l2_h264_p_ref_list_cmp(const void *ptra, const void *ptrb,
 	const struct v4l2_h264_reflist_builder *builder = data;
 	u8 idxa, idxb;
 
-	idxa = *((u8 *)ptra);
-	idxb = *((u8 *)ptrb);
+	idxa = ((struct v4l2_h264_reference *)ptra)->index;
+	idxb = ((struct v4l2_h264_reference *)ptrb)->index;
 
 	if (WARN_ON(idxa >= V4L2_H264_NUM_DPB_ENTRIES ||
 		    idxb >= V4L2_H264_NUM_DPB_ENTRIES))
@@ -125,8 +125,8 @@ static int v4l2_h264_b0_ref_list_cmp(const void *ptra, const void *ptrb,
 	s32 poca, pocb;
 	u8 idxa, idxb;
 
-	idxa = *((u8 *)ptra);
-	idxb = *((u8 *)ptrb);
+	idxa = ((struct v4l2_h264_reference *)ptra)->index;
+	idxb = ((struct v4l2_h264_reference *)ptrb)->index;
 
 	if (WARN_ON(idxa >= V4L2_H264_NUM_DPB_ENTRIES ||
 		    idxb >= V4L2_H264_NUM_DPB_ENTRIES))
@@ -170,8 +170,8 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
 	s32 poca, pocb;
 	u8 idxa, idxb;
 
-	idxa = *((u8 *)ptra);
-	idxb = *((u8 *)ptrb);
+	idxa = ((struct v4l2_h264_reference *)ptra)->index;
+	idxb = ((struct v4l2_h264_reference *)ptrb)->index;
 
 	if (WARN_ON(idxa >= V4L2_H264_NUM_DPB_ENTRIES ||
 		    idxb >= V4L2_H264_NUM_DPB_ENTRIES))
@@ -212,8 +212,8 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
  * v4l2_h264_build_p_ref_list() - Build the P reference list
  *
  * @builder: reference list builder context
- * @reflist: 16-bytes array used to store the P reference list. Each entry
- *	     is an index in the DPB
+ * @reflist: 16 sized array used to store the P reference list. Each entry
+ *	     is a v4l2_h264_reference structure
  *
  * This functions builds the P reference lists. This procedure is describe in
  * section '8.2.4 Decoding process for reference picture lists construction'
@@ -222,7 +222,7 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
  */
 void
 v4l2_h264_build_p_ref_list(const struct v4l2_h264_reflist_builder *builder,
-			   u8 *reflist)
+			   struct v4l2_h264_reference *reflist)
 {
 	memcpy(reflist, builder->unordered_reflist,
 	       sizeof(builder->unordered_reflist[0]) * builder->num_valid);
@@ -235,10 +235,10 @@ EXPORT_SYMBOL_GPL(v4l2_h264_build_p_ref_list);
  * v4l2_h264_build_b_ref_lists() - Build the B0/B1 reference lists
  *
  * @builder: reference list builder context
- * @b0_reflist: 16-bytes array used to store the B0 reference list. Each entry
- *		is an index in the DPB
- * @b1_reflist: 16-bytes array used to store the B1 reference list. Each entry
- *		is an index in the DPB
+ * @b0_reflist: 16 sized array used to store the B0 reference list. Each entry
+ *		is a v4l2_h264_reference structure
+ * @b1_reflist: 16 sized array used to store the B1 reference list. Each entry
+ *		is a v4l2_h264_reference structure
  *
  * This functions builds the B0/B1 reference lists. This procedure is described
  * in section '8.2.4 Decoding process for reference picture lists construction'
@@ -247,7 +247,8 @@ EXPORT_SYMBOL_GPL(v4l2_h264_build_p_ref_list);
  */
 void
 v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
-			    u8 *b0_reflist, u8 *b1_reflist)
+			    struct v4l2_h264_reference *b0_reflist,
+			    struct v4l2_h264_reference *b1_reflist)
 {
 	memcpy(b0_reflist, builder->unordered_reflist,
 	       sizeof(builder->unordered_reflist[0]) * builder->num_valid);
diff --git a/drivers/staging/media/hantro/hantro_g1_h264_dec.c b/drivers/staging/media/hantro/hantro_g1_h264_dec.c
index f49dbfb8a843..9de7f05eff2a 100644
--- a/drivers/staging/media/hantro/hantro_g1_h264_dec.c
+++ b/drivers/staging/media/hantro/hantro_g1_h264_dec.c
@@ -126,7 +126,7 @@ static void set_params(struct hantro_ctx *ctx, struct vb2_v4l2_buffer *src_buf)
 
 static void set_ref(struct hantro_ctx *ctx)
 {
-	const u8 *b0_reflist, *b1_reflist, *p_reflist;
+	const struct v4l2_h264_reference *b0_reflist, *b1_reflist, *p_reflist;
 	struct hantro_dev *vpu = ctx->dev;
 	int reg_num;
 	u32 reg;
@@ -157,12 +157,12 @@ static void set_ref(struct hantro_ctx *ctx)
 	 */
 	reg_num = 0;
 	for (i = 0; i < 15; i += 3) {
-		reg = G1_REG_BD_REF_PIC_BINIT_RLIST_F0(b0_reflist[i]) |
-		      G1_REG_BD_REF_PIC_BINIT_RLIST_F1(b0_reflist[i + 1]) |
-		      G1_REG_BD_REF_PIC_BINIT_RLIST_F2(b0_reflist[i + 2]) |
-		      G1_REG_BD_REF_PIC_BINIT_RLIST_B0(b1_reflist[i]) |
-		      G1_REG_BD_REF_PIC_BINIT_RLIST_B1(b1_reflist[i + 1]) |
-		      G1_REG_BD_REF_PIC_BINIT_RLIST_B2(b1_reflist[i + 2]);
+		reg = G1_REG_BD_REF_PIC_BINIT_RLIST_F0(b0_reflist[i].index) |
+		      G1_REG_BD_REF_PIC_BINIT_RLIST_F1(b0_reflist[i + 1].index) |
+		      G1_REG_BD_REF_PIC_BINIT_RLIST_F2(b0_reflist[i + 2].index) |
+		      G1_REG_BD_REF_PIC_BINIT_RLIST_B0(b1_reflist[i].index) |
+		      G1_REG_BD_REF_PIC_BINIT_RLIST_B1(b1_reflist[i + 1].index) |
+		      G1_REG_BD_REF_PIC_BINIT_RLIST_B2(b1_reflist[i + 2].index);
 		vdpu_write_relaxed(vpu, reg, G1_REG_BD_REF_PIC(reg_num++));
 	}
 
@@ -171,12 +171,12 @@ static void set_ref(struct hantro_ctx *ctx)
 	 * of forward and backward reference picture lists and first 4 entries
 	 * of P forward picture list.
 	 */
-	reg = G1_REG_BD_P_REF_PIC_BINIT_RLIST_F15(b0_reflist[15]) |
-	      G1_REG_BD_P_REF_PIC_BINIT_RLIST_B15(b1_reflist[15]) |
-	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F0(p_reflist[0]) |
-	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F1(p_reflist[1]) |
-	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F2(p_reflist[2]) |
-	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F3(p_reflist[3]);
+	reg = G1_REG_BD_P_REF_PIC_BINIT_RLIST_F15(b0_reflist[15].index) |
+	      G1_REG_BD_P_REF_PIC_BINIT_RLIST_B15(b1_reflist[15].index) |
+	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F0(p_reflist[0].index) |
+	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F1(p_reflist[1].index) |
+	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F2(p_reflist[2].index) |
+	      G1_REG_BD_P_REF_PIC_PINIT_RLIST_F3(p_reflist[3].index);
 	vdpu_write_relaxed(vpu, reg, G1_REG_BD_P_REF_PIC);
 
 	/*
@@ -185,12 +185,12 @@ static void set_ref(struct hantro_ctx *ctx)
 	 */
 	reg_num = 0;
 	for (i = 4; i < HANTRO_H264_DPB_SIZE; i += 6) {
-		reg = G1_REG_FWD_PIC_PINIT_RLIST_F0(p_reflist[i]) |
-		      G1_REG_FWD_PIC_PINIT_RLIST_F1(p_reflist[i + 1]) |
-		      G1_REG_FWD_PIC_PINIT_RLIST_F2(p_reflist[i + 2]) |
-		      G1_REG_FWD_PIC_PINIT_RLIST_F3(p_reflist[i + 3]) |
-		      G1_REG_FWD_PIC_PINIT_RLIST_F4(p_reflist[i + 4]) |
-		      G1_REG_FWD_PIC_PINIT_RLIST_F5(p_reflist[i + 5]);
+		reg = G1_REG_FWD_PIC_PINIT_RLIST_F0(p_reflist[i].index) |
+		      G1_REG_FWD_PIC_PINIT_RLIST_F1(p_reflist[i + 1].index) |
+		      G1_REG_FWD_PIC_PINIT_RLIST_F2(p_reflist[i + 2].index) |
+		      G1_REG_FWD_PIC_PINIT_RLIST_F3(p_reflist[i + 3].index) |
+		      G1_REG_FWD_PIC_PINIT_RLIST_F4(p_reflist[i + 4].index) |
+		      G1_REG_FWD_PIC_PINIT_RLIST_F5(p_reflist[i + 5].index);
 		vdpu_write_relaxed(vpu, reg, G1_REG_FWD_PIC(reg_num++));
 	}
 
diff --git a/drivers/staging/media/hantro/hantro_hw.h b/drivers/staging/media/hantro/hantro_hw.h
index 04844bbcbd36..3608e463290e 100644
--- a/drivers/staging/media/hantro/hantro_hw.h
+++ b/drivers/staging/media/hantro/hantro_hw.h
@@ -69,9 +69,9 @@ struct hantro_h264_dec_ctrls {
  * @b1:		B1 reflist
  */
 struct hantro_h264_dec_reflists {
-	u8 p[HANTRO_H264_DPB_SIZE];
-	u8 b0[HANTRO_H264_DPB_SIZE];
-	u8 b1[HANTRO_H264_DPB_SIZE];
+	struct v4l2_h264_reference p[HANTRO_H264_DPB_SIZE];
+	struct v4l2_h264_reference b0[HANTRO_H264_DPB_SIZE];
+	struct v4l2_h264_reference b1[HANTRO_H264_DPB_SIZE];
 };
 
 /**
diff --git a/drivers/staging/media/hantro/rockchip_vpu2_hw_h264_dec.c b/drivers/staging/media/hantro/rockchip_vpu2_hw_h264_dec.c
index 64a6330475eb..46c1a83bcc4e 100644
--- a/drivers/staging/media/hantro/rockchip_vpu2_hw_h264_dec.c
+++ b/drivers/staging/media/hantro/rockchip_vpu2_hw_h264_dec.c
@@ -298,7 +298,7 @@ static void set_params(struct hantro_ctx *ctx, struct vb2_v4l2_buffer *src_buf)
 
 static void set_ref(struct hantro_ctx *ctx)
 {
-	const u8 *b0_reflist, *b1_reflist, *p_reflist;
+	const struct v4l2_h264_reference *b0_reflist, *b1_reflist, *p_reflist;
 	struct hantro_dev *vpu = ctx->dev;
 	u32 reg;
 	int i;
@@ -307,20 +307,20 @@ static void set_ref(struct hantro_ctx *ctx)
 	b1_reflist = ctx->h264_dec.reflists.b1;
 	p_reflist = ctx->h264_dec.reflists.p;
 
-	reg = VDPU_REG_PINIT_RLIST_F9(p_reflist[9]) |
-	      VDPU_REG_PINIT_RLIST_F8(p_reflist[8]) |
-	      VDPU_REG_PINIT_RLIST_F7(p_reflist[7]) |
-	      VDPU_REG_PINIT_RLIST_F6(p_reflist[6]) |
-	      VDPU_REG_PINIT_RLIST_F5(p_reflist[5]) |
-	      VDPU_REG_PINIT_RLIST_F4(p_reflist[4]);
+	reg = VDPU_REG_PINIT_RLIST_F9(p_reflist[9].index) |
+	      VDPU_REG_PINIT_RLIST_F8(p_reflist[8].index) |
+	      VDPU_REG_PINIT_RLIST_F7(p_reflist[7].index) |
+	      VDPU_REG_PINIT_RLIST_F6(p_reflist[6].index) |
+	      VDPU_REG_PINIT_RLIST_F5(p_reflist[5].index) |
+	      VDPU_REG_PINIT_RLIST_F4(p_reflist[4].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(74));
 
-	reg = VDPU_REG_PINIT_RLIST_F15(p_reflist[15]) |
-	      VDPU_REG_PINIT_RLIST_F14(p_reflist[14]) |
-	      VDPU_REG_PINIT_RLIST_F13(p_reflist[13]) |
-	      VDPU_REG_PINIT_RLIST_F12(p_reflist[12]) |
-	      VDPU_REG_PINIT_RLIST_F11(p_reflist[11]) |
-	      VDPU_REG_PINIT_RLIST_F10(p_reflist[10]);
+	reg = VDPU_REG_PINIT_RLIST_F15(p_reflist[15].index) |
+	      VDPU_REG_PINIT_RLIST_F14(p_reflist[14].index) |
+	      VDPU_REG_PINIT_RLIST_F13(p_reflist[13].index) |
+	      VDPU_REG_PINIT_RLIST_F12(p_reflist[12].index) |
+	      VDPU_REG_PINIT_RLIST_F11(p_reflist[11].index) |
+	      VDPU_REG_PINIT_RLIST_F10(p_reflist[10].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(75));
 
 	reg = VDPU_REG_REFER1_NBR(hantro_h264_get_ref_nbr(ctx, 1)) |
@@ -355,54 +355,54 @@ static void set_ref(struct hantro_ctx *ctx)
 	      VDPU_REG_REFER14_NBR(hantro_h264_get_ref_nbr(ctx, 14));
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(83));
 
-	reg = VDPU_REG_BINIT_RLIST_F5(b0_reflist[5]) |
-	      VDPU_REG_BINIT_RLIST_F4(b0_reflist[4]) |
-	      VDPU_REG_BINIT_RLIST_F3(b0_reflist[3]) |
-	      VDPU_REG_BINIT_RLIST_F2(b0_reflist[2]) |
-	      VDPU_REG_BINIT_RLIST_F1(b0_reflist[1]) |
-	      VDPU_REG_BINIT_RLIST_F0(b0_reflist[0]);
+	reg = VDPU_REG_BINIT_RLIST_F5(b0_reflist[5].index) |
+	      VDPU_REG_BINIT_RLIST_F4(b0_reflist[4].index) |
+	      VDPU_REG_BINIT_RLIST_F3(b0_reflist[3].index) |
+	      VDPU_REG_BINIT_RLIST_F2(b0_reflist[2].index) |
+	      VDPU_REG_BINIT_RLIST_F1(b0_reflist[1].index) |
+	      VDPU_REG_BINIT_RLIST_F0(b0_reflist[0].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(100));
 
-	reg = VDPU_REG_BINIT_RLIST_F11(b0_reflist[11]) |
-	      VDPU_REG_BINIT_RLIST_F10(b0_reflist[10]) |
-	      VDPU_REG_BINIT_RLIST_F9(b0_reflist[9]) |
-	      VDPU_REG_BINIT_RLIST_F8(b0_reflist[8]) |
-	      VDPU_REG_BINIT_RLIST_F7(b0_reflist[7]) |
-	      VDPU_REG_BINIT_RLIST_F6(b0_reflist[6]);
+	reg = VDPU_REG_BINIT_RLIST_F11(b0_reflist[11].index) |
+	      VDPU_REG_BINIT_RLIST_F10(b0_reflist[10].index) |
+	      VDPU_REG_BINIT_RLIST_F9(b0_reflist[9].index) |
+	      VDPU_REG_BINIT_RLIST_F8(b0_reflist[8].index) |
+	      VDPU_REG_BINIT_RLIST_F7(b0_reflist[7].index) |
+	      VDPU_REG_BINIT_RLIST_F6(b0_reflist[6].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(101));
 
-	reg = VDPU_REG_BINIT_RLIST_F15(b0_reflist[15]) |
-	      VDPU_REG_BINIT_RLIST_F14(b0_reflist[14]) |
-	      VDPU_REG_BINIT_RLIST_F13(b0_reflist[13]) |
-	      VDPU_REG_BINIT_RLIST_F12(b0_reflist[12]);
+	reg = VDPU_REG_BINIT_RLIST_F15(b0_reflist[15].index) |
+	      VDPU_REG_BINIT_RLIST_F14(b0_reflist[14].index) |
+	      VDPU_REG_BINIT_RLIST_F13(b0_reflist[13].index) |
+	      VDPU_REG_BINIT_RLIST_F12(b0_reflist[12].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(102));
 
-	reg = VDPU_REG_BINIT_RLIST_B5(b1_reflist[5]) |
-	      VDPU_REG_BINIT_RLIST_B4(b1_reflist[4]) |
-	      VDPU_REG_BINIT_RLIST_B3(b1_reflist[3]) |
-	      VDPU_REG_BINIT_RLIST_B2(b1_reflist[2]) |
-	      VDPU_REG_BINIT_RLIST_B1(b1_reflist[1]) |
-	      VDPU_REG_BINIT_RLIST_B0(b1_reflist[0]);
+	reg = VDPU_REG_BINIT_RLIST_B5(b1_reflist[5].index) |
+	      VDPU_REG_BINIT_RLIST_B4(b1_reflist[4].index) |
+	      VDPU_REG_BINIT_RLIST_B3(b1_reflist[3].index) |
+	      VDPU_REG_BINIT_RLIST_B2(b1_reflist[2].index) |
+	      VDPU_REG_BINIT_RLIST_B1(b1_reflist[1].index) |
+	      VDPU_REG_BINIT_RLIST_B0(b1_reflist[0].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(103));
 
-	reg = VDPU_REG_BINIT_RLIST_B11(b1_reflist[11]) |
-	      VDPU_REG_BINIT_RLIST_B10(b1_reflist[10]) |
-	      VDPU_REG_BINIT_RLIST_B9(b1_reflist[9]) |
-	      VDPU_REG_BINIT_RLIST_B8(b1_reflist[8]) |
-	      VDPU_REG_BINIT_RLIST_B7(b1_reflist[7]) |
-	      VDPU_REG_BINIT_RLIST_B6(b1_reflist[6]);
+	reg = VDPU_REG_BINIT_RLIST_B11(b1_reflist[11].index) |
+	      VDPU_REG_BINIT_RLIST_B10(b1_reflist[10].index) |
+	      VDPU_REG_BINIT_RLIST_B9(b1_reflist[9].index) |
+	      VDPU_REG_BINIT_RLIST_B8(b1_reflist[8].index) |
+	      VDPU_REG_BINIT_RLIST_B7(b1_reflist[7].index) |
+	      VDPU_REG_BINIT_RLIST_B6(b1_reflist[6].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(104));
 
-	reg = VDPU_REG_BINIT_RLIST_B15(b1_reflist[15]) |
-	      VDPU_REG_BINIT_RLIST_B14(b1_reflist[14]) |
-	      VDPU_REG_BINIT_RLIST_B13(b1_reflist[13]) |
-	      VDPU_REG_BINIT_RLIST_B12(b1_reflist[12]);
+	reg = VDPU_REG_BINIT_RLIST_B15(b1_reflist[15].index) |
+	      VDPU_REG_BINIT_RLIST_B14(b1_reflist[14].index) |
+	      VDPU_REG_BINIT_RLIST_B13(b1_reflist[13].index) |
+	      VDPU_REG_BINIT_RLIST_B12(b1_reflist[12].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(105));
 
-	reg = VDPU_REG_PINIT_RLIST_F3(p_reflist[3]) |
-	      VDPU_REG_PINIT_RLIST_F2(p_reflist[2]) |
-	      VDPU_REG_PINIT_RLIST_F1(p_reflist[1]) |
-	      VDPU_REG_PINIT_RLIST_F0(p_reflist[0]);
+	reg = VDPU_REG_PINIT_RLIST_F3(p_reflist[3].index) |
+	      VDPU_REG_PINIT_RLIST_F2(p_reflist[2].index) |
+	      VDPU_REG_PINIT_RLIST_F1(p_reflist[1].index) |
+	      VDPU_REG_PINIT_RLIST_F0(p_reflist[0].index);
 	vdpu_write_relaxed(vpu, reg, VDPU_SWREG(106));
 
 	reg = VDPU_REG_REFER_LTERM_E(ctx->h264_dec.dpb_longterm);
diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index 951e19231da2..3c7f3d87fab4 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -100,9 +100,9 @@ struct rkvdec_h264_priv_tbl {
 #define RKVDEC_H264_DPB_SIZE 16
 
 struct rkvdec_h264_reflists {
-	u8 p[RKVDEC_H264_DPB_SIZE];
-	u8 b0[RKVDEC_H264_DPB_SIZE];
-	u8 b1[RKVDEC_H264_DPB_SIZE];
+	struct v4l2_h264_reference p[RKVDEC_H264_DPB_SIZE];
+	struct v4l2_h264_reference b0[RKVDEC_H264_DPB_SIZE];
+	struct v4l2_h264_reference b1[RKVDEC_H264_DPB_SIZE];
 	u8 num_valid;
 };
 
@@ -767,13 +767,13 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 
 			switch (j) {
 			case 0:
-				idx = h264_ctx->reflists.p[i];
+				idx = h264_ctx->reflists.p[i].index;
 				break;
 			case 1:
-				idx = h264_ctx->reflists.b0[i];
+				idx = h264_ctx->reflists.b0[i].index;
 				break;
 			case 2:
-				idx = h264_ctx->reflists.b1[i];
+				idx = h264_ctx->reflists.b1[i].index;
 				break;
 			}
 
diff --git a/include/media/v4l2-h264.h b/include/media/v4l2-h264.h
index 4b1c71c935e0..ef9a894e3c32 100644
--- a/include/media/v4l2-h264.h
+++ b/include/media/v4l2-h264.h
@@ -37,7 +37,7 @@ struct v4l2_h264_reflist_builder {
 		u16 longterm : 1;
 	} refs[V4L2_H264_NUM_DPB_ENTRIES];
 	s32 cur_pic_order_count;
-	u8 unordered_reflist[V4L2_H264_NUM_DPB_ENTRIES];
+	struct v4l2_h264_reference unordered_reflist[V4L2_H264_NUM_DPB_ENTRIES];
 	u8 num_valid;
 };
 
@@ -51,10 +51,10 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
  * v4l2_h264_build_b_ref_lists() - Build the B0/B1 reference lists
  *
  * @builder: reference list builder context
- * @b0_reflist: 16-bytes array used to store the B0 reference list. Each entry
- *		is an index in the DPB
- * @b1_reflist: 16-bytes array used to store the B1 reference list. Each entry
- *		is an index in the DPB
+ * @b0_reflist: 16 sized array used to store the B0 reference list. Each entry
+ *		is a v4l2_h264_reference structure
+ * @b1_reflist: 16 sized array used to store the B1 reference list. Each entry
+ *		is a v4l2_h264_reference structure
  *
  * This functions builds the B0/B1 reference lists. This procedure is described
  * in section '8.2.4 Decoding process for reference picture lists construction'
@@ -63,14 +63,15 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
  */
 void
 v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
-			    u8 *b0_reflist, u8 *b1_reflist);
+			    struct v4l2_h264_reference *b0_reflist,
+			    struct v4l2_h264_reference *b1_reflist);
 
 /**
  * v4l2_h264_build_p_ref_list() - Build the P reference list
  *
  * @builder: reference list builder context
- * @reflist: 16-bytes array used to store the P reference list. Each entry
- *	     is an index in the DPB
+ * @reflist: 16 sized array used to store the P reference list. Each entry
+ *	     is a v4l2_h264_reference structure
  *
  * This functions builds the P reference lists. This procedure is describe in
  * section '8.2.4 Decoding process for reference picture lists construction'
@@ -79,6 +80,6 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
  */
 void
 v4l2_h264_build_p_ref_list(const struct v4l2_h264_reflist_builder *builder,
-			   u8 *reflist);
+			   struct v4l2_h264_reference *reflist);
 
 #endif /* _MEDIA_V4L2_H264_H */
-- 
2.34.3
[PATCH v5 02/20] media: h264: Increase reference lists size to 32
Posted by Nicolas Dufresne 1 week, 3 days ago
This is to accommodate support for field decoding, which splits the top
and the bottom references into the reference list.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/media/v4l2-core/v4l2-h264.c        | 6 +++---
 drivers/staging/media/hantro/hantro_hw.h   | 6 +++---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 6 +++---
 include/media/v4l2-h264.h                  | 8 ++++----
 4 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index afbfcf78efe4..4b46b36526c0 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -212,7 +212,7 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
  * v4l2_h264_build_p_ref_list() - Build the P reference list
  *
  * @builder: reference list builder context
- * @reflist: 16 sized array used to store the P reference list. Each entry
+ * @reflist: 32 sized array used to store the P reference list. Each entry
  *	     is a v4l2_h264_reference structure
  *
  * This functions builds the P reference lists. This procedure is describe in
@@ -235,9 +235,9 @@ EXPORT_SYMBOL_GPL(v4l2_h264_build_p_ref_list);
  * v4l2_h264_build_b_ref_lists() - Build the B0/B1 reference lists
  *
  * @builder: reference list builder context
- * @b0_reflist: 16 sized array used to store the B0 reference list. Each entry
+ * @b0_reflist: 32 sized array used to store the B0 reference list. Each entry
  *		is a v4l2_h264_reference structure
- * @b1_reflist: 16 sized array used to store the B1 reference list. Each entry
+ * @b1_reflist: 32 sized array used to store the B1 reference list. Each entry
  *		is a v4l2_h264_reference structure
  *
  * This functions builds the B0/B1 reference lists. This procedure is described
diff --git a/drivers/staging/media/hantro/hantro_hw.h b/drivers/staging/media/hantro/hantro_hw.h
index 3608e463290e..0b5b9da86c43 100644
--- a/drivers/staging/media/hantro/hantro_hw.h
+++ b/drivers/staging/media/hantro/hantro_hw.h
@@ -69,9 +69,9 @@ struct hantro_h264_dec_ctrls {
  * @b1:		B1 reflist
  */
 struct hantro_h264_dec_reflists {
-	struct v4l2_h264_reference p[HANTRO_H264_DPB_SIZE];
-	struct v4l2_h264_reference b0[HANTRO_H264_DPB_SIZE];
-	struct v4l2_h264_reference b1[HANTRO_H264_DPB_SIZE];
+	struct v4l2_h264_reference p[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference b0[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference b1[V4L2_H264_REF_LIST_LEN];
 };
 
 /**
diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index 3c7f3d87fab4..dff89732ddd0 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -100,9 +100,9 @@ struct rkvdec_h264_priv_tbl {
 #define RKVDEC_H264_DPB_SIZE 16
 
 struct rkvdec_h264_reflists {
-	struct v4l2_h264_reference p[RKVDEC_H264_DPB_SIZE];
-	struct v4l2_h264_reference b0[RKVDEC_H264_DPB_SIZE];
-	struct v4l2_h264_reference b1[RKVDEC_H264_DPB_SIZE];
+	struct v4l2_h264_reference p[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference b0[V4L2_H264_REF_LIST_LEN];
+	struct v4l2_h264_reference b1[V4L2_H264_REF_LIST_LEN];
 	u8 num_valid;
 };
 
diff --git a/include/media/v4l2-h264.h b/include/media/v4l2-h264.h
index ef9a894e3c32..e282fb16ac58 100644
--- a/include/media/v4l2-h264.h
+++ b/include/media/v4l2-h264.h
@@ -37,7 +37,7 @@ struct v4l2_h264_reflist_builder {
 		u16 longterm : 1;
 	} refs[V4L2_H264_NUM_DPB_ENTRIES];
 	s32 cur_pic_order_count;
-	struct v4l2_h264_reference unordered_reflist[V4L2_H264_NUM_DPB_ENTRIES];
+	struct v4l2_h264_reference unordered_reflist[V4L2_H264_REF_LIST_LEN];
 	u8 num_valid;
 };
 
@@ -51,9 +51,9 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
  * v4l2_h264_build_b_ref_lists() - Build the B0/B1 reference lists
  *
  * @builder: reference list builder context
- * @b0_reflist: 16 sized array used to store the B0 reference list. Each entry
+ * @b0_reflist: 32 sized array used to store the B0 reference list. Each entry
  *		is a v4l2_h264_reference structure
- * @b1_reflist: 16 sized array used to store the B1 reference list. Each entry
+ * @b1_reflist: 32 sized array used to store the B1 reference list. Each entry
  *		is a v4l2_h264_reference structure
  *
  * This functions builds the B0/B1 reference lists. This procedure is described
@@ -70,7 +70,7 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
  * v4l2_h264_build_p_ref_list() - Build the P reference list
  *
  * @builder: reference list builder context
- * @reflist: 16 sized array used to store the P reference list. Each entry
+ * @reflist: 32 sized array used to store the P reference list. Each entry
  *	     is a v4l2_h264_reference structure
  *
  * This functions builds the P reference lists. This procedure is describe in
-- 
2.34.3
[PATCH v5 03/20] media: h264: Store current picture fields
Posted by Nicolas Dufresne 1 week, 3 days ago
This information, also called picture structure, is required in field
decoding mode to construct reference lists.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/media/v4l2-core/v4l2-h264.c | 10 +++++++---
 include/media/v4l2-h264.h           |  4 ++++
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index 4b46b36526c0..58f18bb0afb6 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -34,13 +34,17 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 	cur_frame_num = dec_params->frame_num;
 
 	memset(b, 0, sizeof(*b));
-	if (!(dec_params->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC))
+	if (!(dec_params->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC)) {
 		b->cur_pic_order_count = min(dec_params->bottom_field_order_cnt,
 					     dec_params->top_field_order_cnt);
-	else if (dec_params->flags & V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD)
+		b->cur_pic_fields = V4L2_H264_FRAME_REF;
+	} else if (dec_params->flags & V4L2_H264_DECODE_PARAM_FLAG_BOTTOM_FIELD) {
 		b->cur_pic_order_count = dec_params->bottom_field_order_cnt;
-	else
+		b->cur_pic_fields = V4L2_H264_BOTTOM_FIELD_REF;
+	} else {
 		b->cur_pic_order_count = dec_params->top_field_order_cnt;
+		b->cur_pic_fields = V4L2_H264_TOP_FIELD_REF;
+	}
 
 	for (i = 0; i < V4L2_H264_NUM_DPB_ENTRIES; i++) {
 		u32 pic_order_count;
diff --git a/include/media/v4l2-h264.h b/include/media/v4l2-h264.h
index e282fb16ac58..e165a54c68fa 100644
--- a/include/media/v4l2-h264.h
+++ b/include/media/v4l2-h264.h
@@ -21,6 +21,7 @@
  * @refs.longterm: set to true for a long term reference
  * @refs: array of references
  * @cur_pic_order_count: picture order count of the frame being decoded
+ * @cur_pic_fields: fields present in the frame being decoded
  * @unordered_reflist: unordered list of references. Will be used to generate
  *		       ordered P/B0/B1 lists
  * @num_valid: number of valid references in the refs array
@@ -36,7 +37,10 @@ struct v4l2_h264_reflist_builder {
 		u32 pic_num;
 		u16 longterm : 1;
 	} refs[V4L2_H264_NUM_DPB_ENTRIES];
+
 	s32 cur_pic_order_count;
+	u8 cur_pic_fields;
+
 	struct v4l2_h264_reference unordered_reflist[V4L2_H264_REF_LIST_LEN];
 	u8 num_valid;
 };
-- 
2.34.3
[PATCH v5 04/20] media: h264: Store all fields into the unordered list
Posted by Nicolas Dufresne 1 week, 3 days ago
When the current picture is a field, store each field into the
unordered_list and preserve both top and bottom picture order
count.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 .../media/platform/nvidia/tegra-vde/h264.c    |  2 +-
 drivers/media/v4l2-core/v4l2-h264.c           | 65 ++++++++++++++-----
 include/media/v4l2-h264.h                     |  6 +-
 3 files changed, 52 insertions(+), 21 deletions(-)

diff --git a/drivers/media/platform/nvidia/tegra-vde/h264.c b/drivers/media/platform/nvidia/tegra-vde/h264.c
index 4fb0aaad16d6..88f81a134ba0 100644
--- a/drivers/media/platform/nvidia/tegra-vde/h264.c
+++ b/drivers/media/platform/nvidia/tegra-vde/h264.c
@@ -820,7 +820,7 @@ static int tegra_vde_h264_setup_frames(struct tegra_ctx *ctx,
 		if (err)
 			return err;
 
-		if (b.refs[dpb_idx].pic_order_count < b.cur_pic_order_count)
+		if (b.refs[dpb_idx].top_field_order_cnt < b.cur_pic_order_count)
 			h264->dpb_ref_frames_with_earlier_poc_nb++;
 	}
 
diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index 58f18bb0afb6..38d8dbda0045 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -47,8 +47,6 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 	}
 
 	for (i = 0; i < V4L2_H264_NUM_DPB_ENTRIES; i++) {
-		u32 pic_order_count;
-
 		if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE))
 			continue;
 
@@ -59,8 +57,6 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 		/*
 		 * Handle frame_num wraparound as described in section
 		 * '8.2.4.1 Decoding process for picture numbers' of the spec.
-		 * TODO: This logic will have to be adjusted when we start
-		 * supporting interlaced content.
 		 * For long term references, frame_num is set to
 		 * long_term_frame_idx which requires no wrapping.
 		 */
@@ -70,17 +66,33 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 		else
 			b->refs[i].frame_num = dpb[i].frame_num;
 
-		if (dpb[i].fields == V4L2_H264_FRAME_REF)
-			pic_order_count = min(dpb[i].top_field_order_cnt,
-					      dpb[i].bottom_field_order_cnt);
-		else if (dpb[i].fields & V4L2_H264_BOTTOM_FIELD_REF)
-			pic_order_count = dpb[i].bottom_field_order_cnt;
-		else
-			pic_order_count = dpb[i].top_field_order_cnt;
+		b->refs[i].top_field_order_cnt = dpb[i].top_field_order_cnt;
+		b->refs[i].bottom_field_order_cnt = dpb[i].bottom_field_order_cnt;
+
+		if (b->cur_pic_fields == V4L2_H264_FRAME_REF) {
+			u8 fields = V4L2_H264_FRAME_REF;
+
+			b->unordered_reflist[b->num_valid].index = i;
+			b->unordered_reflist[b->num_valid].fields = fields;
+			b->num_valid++;
+			continue;
+		}
+
+		if (dpb[i].fields & V4L2_H264_TOP_FIELD_REF) {
+			u8 fields = V4L2_H264_TOP_FIELD_REF;
+
+			b->unordered_reflist[b->num_valid].index = i;
+			b->unordered_reflist[b->num_valid].fields = fields;
+			b->num_valid++;
+		}
 
-		b->refs[i].pic_order_count = pic_order_count;
-		b->unordered_reflist[b->num_valid].index = i;
-		b->num_valid++;
+		if (dpb[i].fields & V4L2_H264_BOTTOM_FIELD_REF) {
+			u8 fields = V4L2_H264_BOTTOM_FIELD_REF;
+
+			b->unordered_reflist[b->num_valid].index = i;
+			b->unordered_reflist[b->num_valid].fields = fields;
+			b->num_valid++;
+		}
 	}
 
 	for (i = b->num_valid; i < ARRAY_SIZE(b->unordered_reflist); i++)
@@ -88,6 +100,23 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 }
 EXPORT_SYMBOL_GPL(v4l2_h264_init_reflist_builder);
 
+static s32 v4l2_h264_get_poc(const struct v4l2_h264_reflist_builder *b,
+			     const struct v4l2_h264_reference *ref)
+{
+	switch (ref->fields) {
+	case V4L2_H264_FRAME_REF:
+		return min(b->refs[ref->index].top_field_order_cnt,
+				b->refs[ref->index].bottom_field_order_cnt);
+	case V4L2_H264_TOP_FIELD_REF:
+		return b->refs[ref->index].top_field_order_cnt;
+	case V4L2_H264_BOTTOM_FIELD_REF:
+		return b->refs[ref->index].bottom_field_order_cnt;
+	}
+
+	/* not reached */
+	return 0;
+}
+
 static int v4l2_h264_p_ref_list_cmp(const void *ptra, const void *ptrb,
 				    const void *data)
 {
@@ -150,8 +179,8 @@ static int v4l2_h264_b0_ref_list_cmp(const void *ptra, const void *ptrb,
 		       builder->refs[idxb].pic_num ?
 		       -1 : 1;
 
-	poca = builder->refs[idxa].pic_order_count;
-	pocb = builder->refs[idxb].pic_order_count;
+	poca = v4l2_h264_get_poc(builder, ptra);
+	pocb = v4l2_h264_get_poc(builder, ptrb);
 
 	/*
 	 * Short term pics with POC < cur POC first in POC descending order
@@ -195,8 +224,8 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
 		       builder->refs[idxb].pic_num ?
 		       -1 : 1;
 
-	poca = builder->refs[idxa].pic_order_count;
-	pocb = builder->refs[idxb].pic_order_count;
+	poca = v4l2_h264_get_poc(builder, ptra);
+	pocb = v4l2_h264_get_poc(builder, ptrb);
 
 	/*
 	 * Short term pics with POC > cur POC first in POC ascending order
diff --git a/include/media/v4l2-h264.h b/include/media/v4l2-h264.h
index e165a54c68fa..4cef717b3f18 100644
--- a/include/media/v4l2-h264.h
+++ b/include/media/v4l2-h264.h
@@ -15,7 +15,8 @@
 /**
  * struct v4l2_h264_reflist_builder - Reference list builder object
  *
- * @refs.pic_order_count: reference picture order count
+ * @refs.top_field_order_cnt: top field order count
+ * @refs.bottom_field_order_cnt: bottom field order count
  * @refs.frame_num: reference frame number
  * @refs.pic_num: reference picture number
  * @refs.longterm: set to true for a long term reference
@@ -32,7 +33,8 @@
  */
 struct v4l2_h264_reflist_builder {
 	struct {
-		s32 pic_order_count;
+		s32 top_field_order_cnt;
+		s32 bottom_field_order_cnt;
 		int frame_num;
 		u32 pic_num;
 		u16 longterm : 1;
-- 
2.34.3
[PATCH v5 05/20] media: v4l2: Trace calculated p/b0/b1 initial reflist
Posted by Nicolas Dufresne 1 week, 3 days ago
Add debug print statements to print the content of P & B reference
lists, to verify that the ordering of the generated reference lists is
correct. This is especially important for the field decoding mode,
where sorting is more complex.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Tested-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/media/v4l2-core/v4l2-h264.c | 100 ++++++++++++++++++++++++++++
 1 file changed, 100 insertions(+)

diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index 38d8dbda0045..853f54e0fe67 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -12,6 +12,12 @@
 
 #include <media/v4l2-h264.h>
 
+/*
+ * Size of the tempory buffer allocated when printing reference lists. The
+ * output will be truncated if the size is too small.
+ */
+static const int tmp_str_size = 1024;
+
 /**
  * v4l2_h264_init_reflist_builder() - Initialize a P/B0/B1 reference list
  *				      builder
@@ -241,6 +247,95 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
 	return poca < pocb ? -1 : 1;
 }
 
+static char ref_type_to_char(u8 ref_type)
+{
+	switch (ref_type) {
+	case V4L2_H264_FRAME_REF:
+		return 'f';
+	case V4L2_H264_TOP_FIELD_REF:
+		return 't';
+	case V4L2_H264_BOTTOM_FIELD_REF:
+		return 'b';
+	}
+
+	return '?';
+}
+
+static const char *format_ref_list_p(const struct v4l2_h264_reflist_builder *builder,
+				     struct v4l2_h264_reference *reflist,
+				     char **out_str)
+{
+	int n = 0, i;
+
+	*out_str = kmalloc(tmp_str_size, GFP_KERNEL);
+
+	n += snprintf(*out_str + n, tmp_str_size - n, "|");
+
+	for (i = 0; i < builder->num_valid; i++) {
+		/* this is pic_num for frame and frame_num (wrapped) for field,
+		 * but for frame pic_num is equal to frame_num (wrapped).
+		 */
+		int frame_num = builder->refs[reflist[i].index].frame_num;
+		bool longterm = builder->refs[reflist[i].index].longterm;
+
+		n += scnprintf(*out_str + n, tmp_str_size - n, "%i%c%c|",
+			       frame_num, longterm ? 'l' : 's',
+			       ref_type_to_char(reflist[i].fields));
+	}
+
+	return *out_str;
+}
+
+static void print_ref_list_p(const struct v4l2_h264_reflist_builder *builder,
+			     struct v4l2_h264_reference *reflist)
+{
+	char *buf = NULL;
+
+	pr_debug("ref_pic_list_p (cur_poc %u%c) %s\n",
+		 builder->cur_pic_order_count,
+		 ref_type_to_char(builder->cur_pic_fields),
+		 format_ref_list_p(builder, reflist, &buf));
+
+	kfree(buf);
+}
+
+static const char *format_ref_list_b(const struct v4l2_h264_reflist_builder *builder,
+				     struct v4l2_h264_reference *reflist,
+				     char **out_str)
+{
+	int n = 0, i;
+
+	*out_str = kmalloc(tmp_str_size, GFP_KERNEL);
+
+	n += snprintf(*out_str + n, tmp_str_size - n, "|");
+
+	for (i = 0; i < builder->num_valid; i++) {
+		int frame_num = builder->refs[reflist[i].index].frame_num;
+		u32 poc = v4l2_h264_get_poc(builder, reflist + i);
+		bool longterm = builder->refs[reflist[i].index].longterm;
+
+		n += scnprintf(*out_str + n, tmp_str_size - n, "%i%c%c|",
+			       longterm ? frame_num : poc,
+			       longterm ? 'l' : 's',
+			       ref_type_to_char(reflist[i].fields));
+	}
+
+	return *out_str;
+}
+
+static void print_ref_list_b(const struct v4l2_h264_reflist_builder *builder,
+			     struct v4l2_h264_reference *reflist, u8 list_num)
+{
+	char *buf = NULL;
+
+	pr_debug("ref_pic_list_b%u (cur_poc %u%c) %s",
+		 list_num, builder->cur_pic_order_count,
+		 ref_type_to_char(builder->cur_pic_fields),
+		 format_ref_list_b(builder, reflist, &buf));
+
+	kfree(buf);
+}
+
 /**
  * v4l2_h264_build_p_ref_list() - Build the P reference list
  *
@@ -261,6 +356,8 @@ v4l2_h264_build_p_ref_list(const struct v4l2_h264_reflist_builder *builder,
 	       sizeof(builder->unordered_reflist[0]) * builder->num_valid);
 	sort_r(reflist, builder->num_valid, sizeof(*reflist),
 	       v4l2_h264_p_ref_list_cmp, NULL, builder);
+
+	print_ref_list_p(builder, reflist);
 }
 EXPORT_SYMBOL_GPL(v4l2_h264_build_p_ref_list);
 
@@ -296,6 +393,9 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
 	if (builder->num_valid > 1 &&
 	    !memcmp(b1_reflist, b0_reflist, builder->num_valid))
 		swap(b1_reflist[0], b1_reflist[1]);
+
+	print_ref_list_b(builder, b0_reflist, 0);
+	print_ref_list_b(builder, b1_reflist, 1);
 }
 EXPORT_SYMBOL_GPL(v4l2_h264_build_b_ref_lists);
 
-- 
2.34.3
[PATCH v5 06/20] media: h264: Sort p/b reflist using frame_num
Posted by Nicolas Dufresne 1 week, 3 days ago
In the reference list builder, frame_num refers to FrameNumWrap
in the spec, which is the same as the pic_num for frame decoding.
The same applies for long_term_pic_num and long_term_frame_idx.

Sort all type of references by frame_num so the sort can be reused
for fields reflist were the sorting is done using frame_num instead.
In short, pic_num is never actually used for building reference
lists.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/media/v4l2-core/v4l2-h264.c | 23 +++++++++++++----------
 include/media/v4l2-h264.h           |  2 --
 2 files changed, 13 insertions(+), 12 deletions(-)

diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index 853f54e0fe67..fe215035d9e8 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -56,7 +56,6 @@ v4l2_h264_init_reflist_builder(struct v4l2_h264_reflist_builder *b,
 		if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE))
 			continue;
 
-		b->refs[i].pic_num = dpb[i].pic_num;
 		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM)
 			b->refs[i].longterm = true;
 
@@ -145,15 +144,19 @@ static int v4l2_h264_p_ref_list_cmp(const void *ptra, const void *ptrb,
 	}
 
 	/*
-	 * Short term pics in descending pic num order, long term ones in
-	 * ascending order.
+	 * For frames, short term pics are in descending pic num order and long
+	 * term ones in ascending order. For fields, the same direction is used
+	 * but with frame_num (wrapped). For frames, the value of pic_num and
+	 * frame_num are the same (see formula (8-28) and (8-29)). For this
+	 * reason we can use frame_num only and share this function between
+	 * frames and fields reflist.
 	 */
 	if (!builder->refs[idxa].longterm)
 		return builder->refs[idxb].frame_num <
 		       builder->refs[idxa].frame_num ?
 		       -1 : 1;
 
-	return builder->refs[idxa].pic_num < builder->refs[idxb].pic_num ?
+	return builder->refs[idxa].frame_num < builder->refs[idxb].frame_num ?
 	       -1 : 1;
 }
 
@@ -179,10 +182,10 @@ static int v4l2_h264_b0_ref_list_cmp(const void *ptra, const void *ptrb,
 			return 1;
 	}
 
-	/* Long term pics in ascending pic num order. */
+	/* Long term pics in ascending frame num order. */
 	if (builder->refs[idxa].longterm)
-		return builder->refs[idxa].pic_num <
-		       builder->refs[idxb].pic_num ?
+		return builder->refs[idxa].frame_num <
+		       builder->refs[idxb].frame_num ?
 		       -1 : 1;
 
 	poca = v4l2_h264_get_poc(builder, ptra);
@@ -224,10 +227,10 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
 			return 1;
 	}
 
-	/* Long term pics in ascending pic num order. */
+	/* Long term pics in ascending frame num order. */
 	if (builder->refs[idxa].longterm)
-		return builder->refs[idxa].pic_num <
-		       builder->refs[idxb].pic_num ?
+		return builder->refs[idxa].frame_num <
+		       builder->refs[idxb].frame_num ?
 		       -1 : 1;
 
 	poca = v4l2_h264_get_poc(builder, ptra);
diff --git a/include/media/v4l2-h264.h b/include/media/v4l2-h264.h
index 4cef717b3f18..0d9eaa956123 100644
--- a/include/media/v4l2-h264.h
+++ b/include/media/v4l2-h264.h
@@ -18,7 +18,6 @@
  * @refs.top_field_order_cnt: top field order count
  * @refs.bottom_field_order_cnt: bottom field order count
  * @refs.frame_num: reference frame number
- * @refs.pic_num: reference picture number
  * @refs.longterm: set to true for a long term reference
  * @refs: array of references
  * @cur_pic_order_count: picture order count of the frame being decoded
@@ -36,7 +35,6 @@ struct v4l2_h264_reflist_builder {
 		s32 top_field_order_cnt;
 		s32 bottom_field_order_cnt;
 		int frame_num;
-		u32 pic_num;
 		u16 longterm : 1;
 	} refs[V4L2_H264_NUM_DPB_ENTRIES];
 
-- 
2.34.3
[PATCH v5 07/20] media: v4l2: Reorder field reflist
Posted by Nicolas Dufresne 1 week, 3 days ago
As per spec, the field reflist requires interleaving top and bottom
field in a specific way that does not fit inside the sort operation.
The process consist of alternating references parity, starting with
a reference of the same parity as the current picture. This
processs is done twice, once for short term references and a second
time for the long term references.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/media/v4l2-core/v4l2-h264.c | 42 +++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index fe215035d9e8..72bd64f65198 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -250,6 +250,40 @@ static int v4l2_h264_b1_ref_list_cmp(const void *ptra, const void *ptrb,
 	return poca < pocb ? -1 : 1;
 }
 
+/*
+ * The references need to be reordered so that references are alternating
+ * between top and bottom field references starting with the current picture
+ * parity. This has to be done for short term and long term references
+ * separately.
+ */
+static void reorder_field_reflist(const struct v4l2_h264_reflist_builder *b,
+				  struct v4l2_h264_reference *reflist)
+{
+	struct v4l2_h264_reference tmplist[V4L2_H264_REF_LIST_LEN];
+	u8 lt, i = 0, j = 0, k = 0;
+
+	memcpy(tmplist, reflist, sizeof(tmplist[0]) * b->num_valid);
+
+	for (lt = 0; lt <= 1; lt++) {
+		do {
+			for (; i < b->num_valid && b->refs[tmplist[i].index].longterm == lt; i++) {
+				if (tmplist[i].fields == b->cur_pic_fields) {
+					reflist[k++] = tmplist[i++];
+					break;
+				}
+			}
+
+			for (; j < b->num_valid && b->refs[tmplist[j].index].longterm == lt; j++) {
+				if (tmplist[j].fields != b->cur_pic_fields) {
+					reflist[k++] = tmplist[j++];
+					break;
+				}
+			}
+		} while ((i < b->num_valid && b->refs[tmplist[i].index].longterm == lt) ||
+			 (j < b->num_valid && b->refs[tmplist[j].index].longterm == lt));
+	}
+}
+
 static char ref_type_to_char(u8 ref_type)
 {
 	switch (ref_type) {
@@ -360,6 +394,9 @@ v4l2_h264_build_p_ref_list(const struct v4l2_h264_reflist_builder *builder,
 	sort_r(reflist, builder->num_valid, sizeof(*reflist),
 	       v4l2_h264_p_ref_list_cmp, NULL, builder);
 
+	if (builder->cur_pic_fields != V4L2_H264_FRAME_REF)
+		reorder_field_reflist(builder, reflist);
+
 	print_ref_list_p(builder, reflist);
 }
 EXPORT_SYMBOL_GPL(v4l2_h264_build_p_ref_list);
@@ -393,6 +430,11 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
 	sort_r(b1_reflist, builder->num_valid, sizeof(*b1_reflist),
 	       v4l2_h264_b1_ref_list_cmp, NULL, builder);
 
+	if (builder->cur_pic_fields != V4L2_H264_FRAME_REF) {
+		reorder_field_reflist(builder, b0_reflist);
+		reorder_field_reflist(builder, b1_reflist);
+	}
+
 	if (builder->num_valid > 1 &&
 	    !memcmp(b1_reflist, b0_reflist, builder->num_valid))
 		swap(b1_reflist[0], b1_reflist[1]);
-- 
2.34.3
[PATCH v5 08/20] media: rkvdec: Stop overclocking the decoder
Posted by Nicolas Dufresne 1 week, 3 days ago
While this overclock hack seems to work on some implementations
(some ChromeBooks, RockPi4) it also causes instability on other
implementations (notably LibreComputer Renegade, but there were more
reports in the LibreELEC project, where this has been removed). While
performance is indeed affected (tested with GStreamer), 4K playback
still works as long as you don't operate in lock step and keep at
least 1 frame ahead of time in the decode queue.

After discussion with ChromeOS members, it would seem that their
implementation indeed used to synchronously decode each frame, so
this hack was simply compensating for their code being less
efficient. In my opinion, this hack should not have been included
upstream.

Fixes: cd33c830448ba ("media: rkvdec: Add the rkvdec driver")
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec.c b/drivers/staging/media/rkvdec/rkvdec.c
index c0cf3488f970..2df8cf4883e2 100644
--- a/drivers/staging/media/rkvdec/rkvdec.c
+++ b/drivers/staging/media/rkvdec/rkvdec.c
@@ -1027,12 +1027,6 @@ static int rkvdec_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
-	/*
-	 * Bump ACLK to max. possible freq. (500 MHz) to improve performance
-	 * When 4k video playback.
-	 */
-	clk_set_rate(rkvdec->clocks[0].clk, 500 * 1000 * 1000);
-
 	rkvdec->regs = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(rkvdec->regs))
 		return PTR_ERR(rkvdec->regs);
-- 
2.34.3
[PATCH v5 09/20] media: rkvdec: h264: Fix dpb_valid implementation
Posted by Nicolas Dufresne 1 week, 3 days ago
The ref builder only provided references that are marked as valid in the
dpb. Thus the current implementation of dpb_valid would always set the
flag to 1. This is not representing missing frames (this is called
'non-existing' pictures in the spec). In some context, these non-existing
pictures still need to occupy a slot in the reference list according to
the spec.

Fixes: cd33c830448ba ("media: rkvdec: Add the rkvdec driver")
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 33 ++++++++++++++++------
 1 file changed, 24 insertions(+), 9 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index dff89732ddd0..bcde37d72244 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -112,6 +112,7 @@ struct rkvdec_h264_run {
 	const struct v4l2_ctrl_h264_sps *sps;
 	const struct v4l2_ctrl_h264_pps *pps;
 	const struct v4l2_ctrl_h264_scaling_matrix *scaling_matrix;
+	int ref_buf_idx[V4L2_H264_NUM_DPB_ENTRIES];
 };
 
 struct rkvdec_h264_ctx {
@@ -725,6 +726,26 @@ static void assemble_hw_pps(struct rkvdec_ctx *ctx,
 	}
 }
 
+static void lookup_ref_buf_idx(struct rkvdec_ctx *ctx,
+			       struct rkvdec_h264_run *run)
+{
+	const struct v4l2_ctrl_h264_decode_params *dec_params = run->decode_params;
+	u32 i;
+
+	for (i = 0; i < ARRAY_SIZE(dec_params->dpb); i++) {
+		struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
+		const struct v4l2_h264_dpb_entry *dpb = run->decode_params->dpb;
+		struct vb2_queue *cap_q = &m2m_ctx->cap_q_ctx.q;
+		int buf_idx = -1;
+
+		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
+			buf_idx = vb2_find_timestamp(cap_q,
+						     dpb[i].reference_ts, 0);
+
+		run->ref_buf_idx[i] = buf_idx;
+	}
+}
+
 static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 			    struct rkvdec_h264_run *run)
 {
@@ -762,7 +783,7 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 
 	for (j = 0; j < RKVDEC_NUM_REFLIST; j++) {
 		for (i = 0; i < h264_ctx->reflists.num_valid; i++) {
-			u8 dpb_valid = 0;
+			bool dpb_valid = run->ref_buf_idx[i] >= 0;
 			u8 idx = 0;
 
 			switch (j) {
@@ -779,8 +800,6 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 
 			if (idx >= ARRAY_SIZE(dec_params->dpb))
 				continue;
-			dpb_valid = !!(dpb[idx].flags &
-				       V4L2_H264_DPB_ENTRY_FLAG_ACTIVE);
 
 			set_ps_field(hw_rps, DPB_INFO(i, j),
 				     idx | dpb_valid << 4);
@@ -859,13 +878,8 @@ get_ref_buf(struct rkvdec_ctx *ctx, struct rkvdec_h264_run *run,
 	    unsigned int dpb_idx)
 {
 	struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx;
-	const struct v4l2_h264_dpb_entry *dpb = run->decode_params->dpb;
 	struct vb2_queue *cap_q = &m2m_ctx->cap_q_ctx.q;
-	int buf_idx = -1;
-
-	if (dpb[dpb_idx].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
-		buf_idx = vb2_find_timestamp(cap_q,
-					     dpb[dpb_idx].reference_ts, 0);
+	int buf_idx = run->ref_buf_idx[dpb_idx];
 
 	/*
 	 * If a DPB entry is unused or invalid, address of current destination
@@ -1102,6 +1116,7 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
 
 	assemble_hw_scaling_list(ctx, &run);
 	assemble_hw_pps(ctx, &run);
+	lookup_ref_buf_idx(ctx, &run);
 	assemble_hw_rps(ctx, &run);
 	config_registers(ctx, &run);
 
-- 
2.34.3
[PATCH v5 10/20] media: rkvdec: h264: Fix bit depth wrap in pps packet
Posted by Nicolas Dufresne 1 week, 3 days ago
From: Jonas Karlman <jonas@kwiboo.se>

The luma and chroma bit depth fields in the pps packet are 3 bits wide.
8 is wrongly added to the bit depth values written to these 3 bit fields.
Because only the 3 LSB are written, the hardware was configured
correctly.

Correct this by not adding 8 to the luma and chroma bit depth value.

Fixes: cd33c830448ba ("media: rkvdec: Add the rkvdec driver")
Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index bcde37d72244..8d44a884a52e 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -662,8 +662,8 @@ static void assemble_hw_pps(struct rkvdec_ctx *ctx,
 	WRITE_PPS(0xff, PROFILE_IDC);
 	WRITE_PPS(1, CONSTRAINT_SET3_FLAG);
 	WRITE_PPS(sps->chroma_format_idc, CHROMA_FORMAT_IDC);
-	WRITE_PPS(sps->bit_depth_luma_minus8 + 8, BIT_DEPTH_LUMA);
-	WRITE_PPS(sps->bit_depth_chroma_minus8 + 8, BIT_DEPTH_CHROMA);
+	WRITE_PPS(sps->bit_depth_luma_minus8, BIT_DEPTH_LUMA);
+	WRITE_PPS(sps->bit_depth_chroma_minus8, BIT_DEPTH_CHROMA);
 	WRITE_PPS(0, QPPRIME_Y_ZERO_TRANSFORM_BYPASS_FLAG);
 	WRITE_PPS(sps->log2_max_frame_num_minus4, LOG2_MAX_FRAME_NUM_MINUS4);
 	WRITE_PPS(sps->max_num_ref_frames, MAX_NUM_REF_FRAMES);
-- 
2.34.3
[PATCH v5 11/20] media: rkvdec: Move H264 SPS validation in rkvdec-h264
Posted by Nicolas Dufresne 1 week, 3 days ago
No functional change, this moves H264 specific validation into the H264
specific code. This is in preparation of improving this validation and
reusing it when VIDIOC_STREAMON is called.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 23 ++++++++++++++++++++++
 drivers/staging/media/rkvdec/rkvdec.c      | 23 ++++++----------------
 drivers/staging/media/rkvdec/rkvdec.h      |  1 +
 3 files changed, 30 insertions(+), 17 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index 8d44a884a52e..0dcbcb1bac80 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -1137,9 +1137,32 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
 	return 0;
 }
 
+static int rkvdec_h264_try_ctrl(struct rkvdec_ctx *ctx, struct v4l2_ctrl *ctrl)
+{
+	if (ctrl->id == V4L2_CID_STATELESS_H264_SPS) {
+		const struct v4l2_ctrl_h264_sps *sps = ctrl->p_new.p_h264_sps;
+		/*
+		 * TODO: The hardware supports 10-bit and 4:2:2 profiles,
+		 * but it's currently broken in the driver.
+		 * Reject them for now, until it's fixed.
+		 */
+		if (sps->chroma_format_idc > 1)
+			/* Only 4:0:0 and 4:2:0 are supported */
+			return -EINVAL;
+		if (sps->bit_depth_luma_minus8 != sps->bit_depth_chroma_minus8)
+			/* Luma and chroma bit depth mismatch */
+			return -EINVAL;
+		if (sps->bit_depth_luma_minus8 != 0)
+			/* Only 8-bit is supported */
+			return -EINVAL;
+	}
+	return 0;
+}
+
 const struct rkvdec_coded_fmt_ops rkvdec_h264_fmt_ops = {
 	.adjust_fmt = rkvdec_h264_adjust_fmt,
 	.start = rkvdec_h264_start,
 	.stop = rkvdec_h264_stop,
 	.run = rkvdec_h264_run,
+	.try_ctrl = rkvdec_h264_try_ctrl,
 };
diff --git a/drivers/staging/media/rkvdec/rkvdec.c b/drivers/staging/media/rkvdec/rkvdec.c
index 2df8cf4883e2..e3d44d5b35f3 100644
--- a/drivers/staging/media/rkvdec/rkvdec.c
+++ b/drivers/staging/media/rkvdec/rkvdec.c
@@ -29,23 +29,12 @@
 
 static int rkvdec_try_ctrl(struct v4l2_ctrl *ctrl)
 {
-	if (ctrl->id == V4L2_CID_STATELESS_H264_SPS) {
-		const struct v4l2_ctrl_h264_sps *sps = ctrl->p_new.p_h264_sps;
-		/*
-		 * TODO: The hardware supports 10-bit and 4:2:2 profiles,
-		 * but it's currently broken in the driver.
-		 * Reject them for now, until it's fixed.
-		 */
-		if (sps->chroma_format_idc > 1)
-			/* Only 4:0:0 and 4:2:0 are supported */
-			return -EINVAL;
-		if (sps->bit_depth_luma_minus8 != sps->bit_depth_chroma_minus8)
-			/* Luma and chroma bit depth mismatch */
-			return -EINVAL;
-		if (sps->bit_depth_luma_minus8 != 0)
-			/* Only 8-bit is supported */
-			return -EINVAL;
-	}
+	struct rkvdec_ctx *ctx = container_of(ctrl->handler, struct rkvdec_ctx, ctrl_hdl);
+	const struct rkvdec_coded_fmt_desc *desc = ctx->coded_fmt_desc;
+
+	if (desc->ops->try_ctrl)
+		return desc->ops->try_ctrl(ctx, ctrl);
+
 	return 0;
 }
 
diff --git a/drivers/staging/media/rkvdec/rkvdec.h b/drivers/staging/media/rkvdec/rkvdec.h
index 2f4ea1786b93..9df0fba799a4 100644
--- a/drivers/staging/media/rkvdec/rkvdec.h
+++ b/drivers/staging/media/rkvdec/rkvdec.h
@@ -72,6 +72,7 @@ struct rkvdec_coded_fmt_ops {
 	void (*done)(struct rkvdec_ctx *ctx, struct vb2_v4l2_buffer *src_buf,
 		     struct vb2_v4l2_buffer *dst_buf,
 		     enum vb2_buffer_state result);
+	int (*try_ctrl)(struct rkvdec_ctx *ctx, struct v4l2_ctrl *ctrl);
 };
 
 struct rkvdec_coded_fmt_desc {
-- 
2.34.3
[PATCH v5 12/20] media: rkvdec: h264: Validate and use pic width and height in mbs
Posted by Nicolas Dufresne 1 week, 3 days ago
From: Jonas Karlman <jonas@kwiboo.se>

The width and height in macroblocks is currently configured based on OUTPUT
buffer resolution, this works for frame pictures but can cause issues for
field pictures.

When frame_mbs_only_flag is 0 the height in mbs should be height of
the field instead of height of frame.

Validate pic_width_in_mbs_minus1 and pic_height_in_map_units_minus1
against OUTPUT buffer resolution and use these values to configure HW.
The validation is happening in both try_ctrt() and start() since it is
otherwise possible to trick the driver during initialization by changing
the OUTPUT format after having set a valid control.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
[hverkuil: when -> When (first word in a comment block)]
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 81 +++++++++++++++++-----
 1 file changed, 62 insertions(+), 19 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index 0dcbcb1bac80..fb41e2fd8359 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -672,8 +672,17 @@ static void assemble_hw_pps(struct rkvdec_ctx *ctx,
 		  LOG2_MAX_PIC_ORDER_CNT_LSB_MINUS4);
 	WRITE_PPS(!!(sps->flags & V4L2_H264_SPS_FLAG_DELTA_PIC_ORDER_ALWAYS_ZERO),
 		  DELTA_PIC_ORDER_ALWAYS_ZERO_FLAG);
-	WRITE_PPS(DIV_ROUND_UP(ctx->coded_fmt.fmt.pix_mp.width, 16), PIC_WIDTH_IN_MBS);
-	WRITE_PPS(DIV_ROUND_UP(ctx->coded_fmt.fmt.pix_mp.height, 16), PIC_HEIGHT_IN_MBS);
+
+	/*
+	 * Use the SPS values since they are already in macroblocks
+	 * dimensions, height can be field height (halved) if
+	 * V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY is not set and also it allows
+	 * decoding smaller images into larger allocation which can be used
+	 * to implementing SVC spatial layer support.
+	 */
+	WRITE_PPS(sps->pic_width_in_mbs_minus1 + 1, PIC_WIDTH_IN_MBS);
+	WRITE_PPS(sps->pic_height_in_map_units_minus1 + 1, PIC_HEIGHT_IN_MBS);
+
 	WRITE_PPS(!!(sps->flags & V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY),
 		  FRAME_MBS_ONLY_FLAG);
 	WRITE_PPS(!!(sps->flags & V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD),
@@ -1035,13 +1044,61 @@ static int rkvdec_h264_adjust_fmt(struct rkvdec_ctx *ctx,
 	return 0;
 }
 
+static int rkvdec_h264_validate_sps(struct rkvdec_ctx *ctx,
+				    const struct v4l2_ctrl_h264_sps *sps)
+{
+	unsigned int width, height;
+
+	/*
+	 * TODO: The hardware supports 10-bit and 4:2:2 profiles,
+	 * but it's currently broken in the driver.
+	 * Reject them for now, until it's fixed.
+	 */
+	if (sps->chroma_format_idc > 1)
+		/* Only 4:0:0 and 4:2:0 are supported */
+		return -EINVAL;
+	if (sps->bit_depth_luma_minus8 != sps->bit_depth_chroma_minus8)
+		/* Luma and chroma bit depth mismatch */
+		return -EINVAL;
+	if (sps->bit_depth_luma_minus8 != 0)
+		/* Only 8-bit is supported */
+		return -EINVAL;
+
+	width = (sps->pic_width_in_mbs_minus1 + 1) * 16;
+	height = (sps->pic_height_in_map_units_minus1 + 1) * 16;
+
+	/*
+	 * When frame_mbs_only_flag is not set, this is field height,
+	 * which is half the final height (see (7-18) in the
+	 * specification)
+	 */
+	if (!(sps->flags & V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY))
+		height *= 2;
+
+	if (width > ctx->coded_fmt.fmt.pix_mp.width ||
+	    height > ctx->coded_fmt.fmt.pix_mp.height)
+		return -EINVAL;
+
+	return 0;
+}
+
 static int rkvdec_h264_start(struct rkvdec_ctx *ctx)
 {
 	struct rkvdec_dev *rkvdec = ctx->dev;
 	struct rkvdec_h264_priv_tbl *priv_tbl;
 	struct rkvdec_h264_ctx *h264_ctx;
+	struct v4l2_ctrl *ctrl;
 	int ret;
 
+	ctrl = v4l2_ctrl_find(&ctx->ctrl_hdl,
+			      V4L2_CID_STATELESS_H264_SPS);
+	if (!ctrl)
+		return -EINVAL;
+
+	ret = rkvdec_h264_validate_sps(ctx, ctrl->p_new.p_h264_sps);
+	if (ret)
+		return ret;
+
 	h264_ctx = kzalloc(sizeof(*h264_ctx), GFP_KERNEL);
 	if (!h264_ctx)
 		return -ENOMEM;
@@ -1139,23 +1196,9 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
 
 static int rkvdec_h264_try_ctrl(struct rkvdec_ctx *ctx, struct v4l2_ctrl *ctrl)
 {
-	if (ctrl->id == V4L2_CID_STATELESS_H264_SPS) {
-		const struct v4l2_ctrl_h264_sps *sps = ctrl->p_new.p_h264_sps;
-		/*
-		 * TODO: The hardware supports 10-bit and 4:2:2 profiles,
-		 * but it's currently broken in the driver.
-		 * Reject them for now, until it's fixed.
-		 */
-		if (sps->chroma_format_idc > 1)
-			/* Only 4:0:0 and 4:2:0 are supported */
-			return -EINVAL;
-		if (sps->bit_depth_luma_minus8 != sps->bit_depth_chroma_minus8)
-			/* Luma and chroma bit depth mismatch */
-			return -EINVAL;
-		if (sps->bit_depth_luma_minus8 != 0)
-			/* Only 8-bit is supported */
-			return -EINVAL;
-	}
+	if (ctrl->id == V4L2_CID_STATELESS_H264_SPS)
+		return rkvdec_h264_validate_sps(ctx, ctrl->p_new.p_h264_sps);
+
 	return 0;
 }
 
-- 
2.34.3
[PATCH v5 13/20] media: rkvdec: h264: Fix reference frame_num wrap for second field
Posted by Nicolas Dufresne 1 week, 3 days ago
From: Jonas Karlman <jonas@kwiboo.se>

When decoding the second field in a complementary field pair the second
field is sharing the same frame_num with the first field.

Currently the frame_num for the first field is wrapped when it matches the
field being decoded, this caused issues decoding the second field in a
complementary field pair.

Fix this by using inclusive comparison: 'less than or equal'.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Ezequiel Garcia <ezequiel@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index fb41e2fd8359..57821ee3b213 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -782,7 +782,7 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 			continue;
 
 		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM ||
-		    dpb[i].frame_num < dec_params->frame_num) {
+		    dpb[i].frame_num <= dec_params->frame_num) {
 			p[i] = dpb[i].frame_num;
 			continue;
 		}
-- 
2.34.3
[PATCH v5 14/20] media: rkvdec: Ensure decoded resolution fit coded resolution
Posted by Nicolas Dufresne 1 week, 3 days ago
From: Jonas Karlman <jonas@kwiboo.se>

Ensure decoded CAPTURE buffer resolution is larger or equal to the coded
OUTPUT buffer resolution.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/staging/media/rkvdec/rkvdec.c b/drivers/staging/media/rkvdec/rkvdec.c
index e3d44d5b35f3..2bc4b1a40989 100644
--- a/drivers/staging/media/rkvdec/rkvdec.c
+++ b/drivers/staging/media/rkvdec/rkvdec.c
@@ -257,6 +257,8 @@ static int rkvdec_try_capture_fmt(struct file *file, void *priv,
 		pix_mp->pixelformat = coded_desc->decoded_fmts[0];
 
 	/* Always apply the frmsize constraint of the coded end. */
+	pix_mp->width = max(pix_mp->width, ctx->coded_fmt.fmt.pix_mp.width);
+	pix_mp->height = max(pix_mp->height, ctx->coded_fmt.fmt.pix_mp.height);
 	v4l2_apply_frmsize_constraints(&pix_mp->width,
 				       &pix_mp->height,
 				       &coded_desc->frmsize);
-- 
2.34.3
[PATCH v5 15/20] media: rkvdec-h264: Add field decoding support
Posted by Nicolas Dufresne 1 week, 3 days ago
This makes use of the new feature in the reference builder to program
up to 32 references when doing field decoding. It also signals the
parity (top or bottom) of the field to the hardware.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec-h264.c | 48 ++++++++++------------
 1 file changed, 21 insertions(+), 27 deletions(-)

diff --git a/drivers/staging/media/rkvdec/rkvdec-h264.c b/drivers/staging/media/rkvdec/rkvdec-h264.c
index 57821ee3b213..2992fb87cf72 100644
--- a/drivers/staging/media/rkvdec/rkvdec-h264.c
+++ b/drivers/staging/media/rkvdec/rkvdec-h264.c
@@ -97,13 +97,10 @@ struct rkvdec_h264_priv_tbl {
 	u8 err_info[RKV_ERROR_INFO_SIZE];
 };
 
-#define RKVDEC_H264_DPB_SIZE 16
-
 struct rkvdec_h264_reflists {
 	struct v4l2_h264_reference p[V4L2_H264_REF_LIST_LEN];
 	struct v4l2_h264_reference b0[V4L2_H264_REF_LIST_LEN];
 	struct v4l2_h264_reference b1[V4L2_H264_REF_LIST_LEN];
-	u8 num_valid;
 };
 
 struct rkvdec_h264_run {
@@ -747,23 +744,26 @@ static void lookup_ref_buf_idx(struct rkvdec_ctx *ctx,
 		struct vb2_queue *cap_q = &m2m_ctx->cap_q_ctx.q;
 		int buf_idx = -1;
 
-		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
+		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE) {
 			buf_idx = vb2_find_timestamp(cap_q,
 						     dpb[i].reference_ts, 0);
+			if (buf_idx < 0)
+				pr_debug("No buffer for reference_ts %llu",
+					 dpb[i].reference_ts);
+		}
 
 		run->ref_buf_idx[i] = buf_idx;
 	}
 }
 
 static void assemble_hw_rps(struct rkvdec_ctx *ctx,
+			    struct v4l2_h264_reflist_builder *builder,
 			    struct rkvdec_h264_run *run)
 {
 	const struct v4l2_ctrl_h264_decode_params *dec_params = run->decode_params;
 	const struct v4l2_h264_dpb_entry *dpb = dec_params->dpb;
 	struct rkvdec_h264_ctx *h264_ctx = ctx->priv;
-	const struct v4l2_ctrl_h264_sps *sps = run->sps;
 	struct rkvdec_h264_priv_tbl *priv_tbl = h264_ctx->priv_tbl.cpu;
-	u32 max_frame_num = 1 << (sps->log2_max_frame_num_minus4 + 4);
 
 	u32 *hw_rps = priv_tbl->rps;
 	u32 i, j;
@@ -781,37 +781,36 @@ static void assemble_hw_rps(struct rkvdec_ctx *ctx,
 		if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE))
 			continue;
 
-		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM ||
-		    dpb[i].frame_num <= dec_params->frame_num) {
-			p[i] = dpb[i].frame_num;
-			continue;
-		}
-
-		p[i] = dpb[i].frame_num - max_frame_num;
+		p[i] = builder->refs[i].frame_num;
 	}
 
 	for (j = 0; j < RKVDEC_NUM_REFLIST; j++) {
-		for (i = 0; i < h264_ctx->reflists.num_valid; i++) {
-			bool dpb_valid = run->ref_buf_idx[i] >= 0;
-			u8 idx = 0;
+		for (i = 0; i < builder->num_valid; i++) {
+			struct v4l2_h264_reference *ref;
+			bool dpb_valid;
+			bool bottom;
 
 			switch (j) {
 			case 0:
-				idx = h264_ctx->reflists.p[i].index;
+				ref = &h264_ctx->reflists.p[i];
 				break;
 			case 1:
-				idx = h264_ctx->reflists.b0[i].index;
+				ref = &h264_ctx->reflists.b0[i];
 				break;
 			case 2:
-				idx = h264_ctx->reflists.b1[i].index;
+				ref = &h264_ctx->reflists.b1[i];
 				break;
 			}
 
-			if (idx >= ARRAY_SIZE(dec_params->dpb))
+			if (WARN_ON(ref->index >= ARRAY_SIZE(dec_params->dpb)))
 				continue;
 
+			dpb_valid = run->ref_buf_idx[ref->index] >= 0;
+			bottom = ref->fields == V4L2_H264_BOTTOM_FIELD_REF;
+
 			set_ps_field(hw_rps, DPB_INFO(i, j),
-				     idx | dpb_valid << 4);
+				     ref->index | dpb_valid << 4);
+			set_ps_field(hw_rps, BOTTOM_FLAG(i, j), bottom);
 		}
 	}
 }
@@ -999,10 +998,6 @@ static void config_registers(struct rkvdec_ctx *ctx,
 				       rkvdec->regs + RKVDEC_REG_H264_BASE_REFER15);
 	}
 
-	/*
-	 * Since support frame mode only
-	 * top_field_order_cnt is the same as bottom_field_order_cnt
-	 */
 	reg = RKVDEC_CUR_POC(dec_params->top_field_order_cnt);
 	writel_relaxed(reg, rkvdec->regs + RKVDEC_REG_CUR_POC0);
 
@@ -1166,7 +1161,6 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
 	/* Build the P/B{0,1} ref lists. */
 	v4l2_h264_init_reflist_builder(&reflist_builder, run.decode_params,
 				       run.sps, run.decode_params->dpb);
-	h264_ctx->reflists.num_valid = reflist_builder.num_valid;
 	v4l2_h264_build_p_ref_list(&reflist_builder, h264_ctx->reflists.p);
 	v4l2_h264_build_b_ref_lists(&reflist_builder, h264_ctx->reflists.b0,
 				    h264_ctx->reflists.b1);
@@ -1174,7 +1168,7 @@ static int rkvdec_h264_run(struct rkvdec_ctx *ctx)
 	assemble_hw_scaling_list(ctx, &run);
 	assemble_hw_pps(ctx, &run);
 	lookup_ref_buf_idx(ctx, &run);
-	assemble_hw_rps(ctx, &run);
+	assemble_hw_rps(ctx, &reflist_builder, &run);
 	config_registers(ctx, &run);
 
 	rkvdec_run_postamble(ctx, &run.base);
-- 
2.34.3
[PATCH v5 16/20] media: rkvdec: Enable capture buffer holding for H264
Posted by Nicolas Dufresne 1 week, 3 days ago
In order to support interlaced video decoding, the driver must
allow holding the capture buffer so that the second field can
be decoded into it.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/rkvdec/rkvdec.c | 4 ++++
 drivers/staging/media/rkvdec/rkvdec.h | 1 +
 2 files changed, 5 insertions(+)

diff --git a/drivers/staging/media/rkvdec/rkvdec.c b/drivers/staging/media/rkvdec/rkvdec.c
index 2bc4b1a40989..7bab7586918c 100644
--- a/drivers/staging/media/rkvdec/rkvdec.c
+++ b/drivers/staging/media/rkvdec/rkvdec.c
@@ -127,6 +127,7 @@ static const struct rkvdec_coded_fmt_desc rkvdec_coded_fmts[] = {
 		.ops = &rkvdec_h264_fmt_ops,
 		.num_decoded_fmts = ARRAY_SIZE(rkvdec_h264_vp9_decoded_fmts),
 		.decoded_fmts = rkvdec_h264_vp9_decoded_fmts,
+		.subsystem_flags = VB2_V4L2_FL_SUPPORTS_M2M_HOLD_CAPTURE_BUF,
 	},
 	{
 		.fourcc = V4L2_PIX_FMT_VP9_FRAME,
@@ -385,6 +386,9 @@ static int rkvdec_s_output_fmt(struct file *file, void *priv,
 	cap_fmt->fmt.pix_mp.ycbcr_enc = f->fmt.pix_mp.ycbcr_enc;
 	cap_fmt->fmt.pix_mp.quantization = f->fmt.pix_mp.quantization;
 
+	/* Enable format specific queue features */
+	vq->subsystem_flags |= desc->subsystem_flags;
+
 	return 0;
 }
 
diff --git a/drivers/staging/media/rkvdec/rkvdec.h b/drivers/staging/media/rkvdec/rkvdec.h
index 9df0fba799a4..633335ebb9c4 100644
--- a/drivers/staging/media/rkvdec/rkvdec.h
+++ b/drivers/staging/media/rkvdec/rkvdec.h
@@ -82,6 +82,7 @@ struct rkvdec_coded_fmt_desc {
 	const struct rkvdec_coded_fmt_ops *ops;
 	unsigned int num_decoded_fmts;
 	const u32 *decoded_fmts;
+	u32 subsystem_flags;
 };
 
 struct rkvdec_dev {
-- 
2.34.3
[PATCH v5 17/20] media: hantro: Stop using H.264 parameter pic_num
Posted by Nicolas Dufresne 1 week, 3 days ago
The hardware expects FrameNumWrap or long_term_frame_idx. Picture
numbers are per field, and are mostly used during the memory
management process, which is done in userland. This fixes two
ITU conformance tests:

  - MR6_BT_B
  - MR8_BT_B

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/hantro/hantro_h264.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/staging/media/hantro/hantro_h264.c b/drivers/staging/media/hantro/hantro_h264.c
index 0b4d2491be3b..228629fb3cdf 100644
--- a/drivers/staging/media/hantro/hantro_h264.c
+++ b/drivers/staging/media/hantro/hantro_h264.c
@@ -354,8 +354,6 @@ u16 hantro_h264_get_ref_nbr(struct hantro_ctx *ctx, unsigned int dpb_idx)
 
 	if (!(dpb->flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE))
 		return 0;
-	if (dpb->flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM)
-		return dpb->pic_num;
 	return dpb->frame_num;
 }
 
-- 
2.34.3
[PATCH v5 18/20] media: hantro: h264: Make dpb entry management more robust
Posted by Nicolas Dufresne 1 week, 3 days ago
From: Jonas Karlman <jonas@kwiboo.se>

The driver maintains stable slot locations for reference pictures. This
change makes the code more robust by using the reference_ts as key and
by marking all entries invalid right from the start.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/hantro/hantro_h264.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/staging/media/hantro/hantro_h264.c b/drivers/staging/media/hantro/hantro_h264.c
index 228629fb3cdf..7377fc26f780 100644
--- a/drivers/staging/media/hantro/hantro_h264.c
+++ b/drivers/staging/media/hantro/hantro_h264.c
@@ -258,8 +258,7 @@ static void prepare_table(struct hantro_ctx *ctx)
 static bool dpb_entry_match(const struct v4l2_h264_dpb_entry *a,
 			    const struct v4l2_h264_dpb_entry *b)
 {
-	return a->top_field_order_cnt == b->top_field_order_cnt &&
-	       a->bottom_field_order_cnt == b->bottom_field_order_cnt;
+	return a->reference_ts == b->reference_ts;
 }
 
 static void update_dpb(struct hantro_ctx *ctx)
@@ -273,13 +272,13 @@ static void update_dpb(struct hantro_ctx *ctx)
 
 	/* Disable all entries by default. */
 	for (i = 0; i < ARRAY_SIZE(ctx->h264_dec.dpb); i++)
-		ctx->h264_dec.dpb[i].flags &= ~V4L2_H264_DPB_ENTRY_FLAG_ACTIVE;
+		ctx->h264_dec.dpb[i].flags = 0;
 
 	/* Try to match new DPB entries with existing ones by their POCs. */
 	for (i = 0; i < ARRAY_SIZE(dec_param->dpb); i++) {
 		const struct v4l2_h264_dpb_entry *ndpb = &dec_param->dpb[i];
 
-		if (!(ndpb->flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE))
+		if (!(ndpb->flags & V4L2_H264_DPB_ENTRY_FLAG_VALID))
 			continue;
 
 		/*
@@ -290,8 +289,7 @@ static void update_dpb(struct hantro_ctx *ctx)
 			struct v4l2_h264_dpb_entry *cdpb;
 
 			cdpb = &ctx->h264_dec.dpb[j];
-			if (cdpb->flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE ||
-			    !dpb_entry_match(cdpb, ndpb))
+			if (!dpb_entry_match(cdpb, ndpb))
 				continue;
 
 			*cdpb = *ndpb;
-- 
2.34.3
[PATCH v5 19/20] media: hantro: Add H.264 field decoding support
Posted by Nicolas Dufresne 1 week, 3 days ago
This adds the required code to support field decoding. While most of
the code is derived from Rockchip and VSI reference code, the
reduction of the reference list to 16 entries was found by
trial and errors. The list consists of all the references with the
opposite field parity.

The strategy is to deduplicate the reference picture that points
to the same storage (same index). The choice of opposite parity has
been made to keep the other field of the current field pair in the
list. This method may not be robust if a field was lost.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
[hverkuil: fix typos in the comment before deduplicate_reflist()]
[hverkuil: document new cur_poc field]
---
 drivers/staging/media/hantro/hantro_h264.c | 122 ++++++++++++++++++---
 drivers/staging/media/hantro/hantro_hw.h   |   2 +
 2 files changed, 110 insertions(+), 14 deletions(-)

diff --git a/drivers/staging/media/hantro/hantro_h264.c b/drivers/staging/media/hantro/hantro_h264.c
index 7377fc26f780..4e9a0ecf5c13 100644
--- a/drivers/staging/media/hantro/hantro_h264.c
+++ b/drivers/staging/media/hantro/hantro_h264.c
@@ -22,6 +22,12 @@
 #define POC_BUFFER_SIZE			34
 #define SCALING_LIST_SIZE		(6 * 16 + 2 * 64)
 
+/*
+ * For valid and long term reference marking, index are reversed, so bit 31
+ * indicates the status of the picture 0.
+ */
+#define REF_BIT(i)			BIT(32 - 1 - (i))
+
 /* Data structure describing auxiliary buffer format. */
 struct hantro_h264_dec_priv_tbl {
 	u32 cabac_table[CABAC_INIT_BUFFER_SIZE];
@@ -227,6 +233,7 @@ static void prepare_table(struct hantro_ctx *ctx)
 {
 	const struct hantro_h264_dec_ctrls *ctrls = &ctx->h264_dec.ctrls;
 	const struct v4l2_ctrl_h264_decode_params *dec_param = ctrls->decode;
+	const struct v4l2_ctrl_h264_sps *sps = ctrls->sps;
 	struct hantro_h264_dec_priv_tbl *tbl = ctx->h264_dec.priv.cpu;
 	const struct v4l2_h264_dpb_entry *dpb = ctx->h264_dec.dpb;
 	u32 dpb_longterm = 0;
@@ -237,20 +244,45 @@ static void prepare_table(struct hantro_ctx *ctx)
 		tbl->poc[i * 2] = dpb[i].top_field_order_cnt;
 		tbl->poc[i * 2 + 1] = dpb[i].bottom_field_order_cnt;
 
+		if (!(dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_VALID))
+			continue;
+
 		/*
 		 * Set up bit maps of valid and long term DPBs.
-		 * NOTE: The bits are reversed, i.e. MSb is DPB 0.
+		 * NOTE: The bits are reversed, i.e. MSb is DPB 0. For frame
+		 * decoding, bit 31 to 15 are used, while for field decoding,
+		 * all bits are used, with bit 31 being a top field, 30 a bottom
+		 * field and so on.
 		 */
-		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
-			dpb_valid |= BIT(HANTRO_H264_DPB_SIZE - 1 - i);
-		if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM)
-			dpb_longterm |= BIT(HANTRO_H264_DPB_SIZE - 1 - i);
+		if (dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC) {
+			if (dpb[i].fields & V4L2_H264_TOP_FIELD_REF)
+				dpb_valid |= REF_BIT(i * 2);
+
+			if (dpb[i].fields & V4L2_H264_BOTTOM_FIELD_REF)
+				dpb_valid |= REF_BIT(i * 2 + 1);
+
+			if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM) {
+				dpb_longterm |= REF_BIT(i * 2);
+				dpb_longterm |= REF_BIT(i * 2 + 1);
+			}
+		} else {
+			dpb_valid |= REF_BIT(i);
+
+			if (dpb[i].flags & V4L2_H264_DPB_ENTRY_FLAG_LONG_TERM)
+				dpb_longterm |= REF_BIT(i);
+		}
+	}
+	ctx->h264_dec.dpb_valid = dpb_valid;
+	ctx->h264_dec.dpb_longterm = dpb_longterm;
+
+	if ((dec_param->flags & V4L2_H264_DECODE_PARAM_FLAG_FIELD_PIC) ||
+	    !(sps->flags & V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD)) {
+		tbl->poc[32] = ctx->h264_dec.cur_poc;
+		tbl->poc[33] = 0;
+	} else {
+		tbl->poc[32] = dec_param->top_field_order_cnt;
+		tbl->poc[33] = dec_param->bottom_field_order_cnt;
 	}
-	ctx->h264_dec.dpb_valid = dpb_valid << 16;
-	ctx->h264_dec.dpb_longterm = dpb_longterm << 16;
-
-	tbl->poc[32] = dec_param->top_field_order_cnt;
-	tbl->poc[33] = dec_param->bottom_field_order_cnt;
 
 	assemble_scaling_list(ctx);
 }
@@ -326,6 +358,8 @@ dma_addr_t hantro_h264_get_ref_buf(struct hantro_ctx *ctx,
 {
 	struct v4l2_h264_dpb_entry *dpb = ctx->h264_dec.dpb;
 	dma_addr_t dma_addr = 0;
+	s32 cur_poc = ctx->h264_dec.cur_poc;
+	u32 flags;
 
 	if (dpb[dpb_idx].flags & V4L2_H264_DPB_ENTRY_FLAG_ACTIVE)
 		dma_addr = hantro_get_ref(ctx, dpb[dpb_idx].reference_ts);
@@ -343,7 +377,12 @@ dma_addr_t hantro_h264_get_ref_buf(struct hantro_ctx *ctx,
 		dma_addr = hantro_get_dec_buf_addr(ctx, buf);
 	}
 
-	return dma_addr;
+	flags = dpb[dpb_idx].flags & V4L2_H264_DPB_ENTRY_FLAG_FIELD ? 0x2 : 0;
+	flags |= abs(dpb[dpb_idx].top_field_order_cnt - cur_poc) <
+		 abs(dpb[dpb_idx].bottom_field_order_cnt - cur_poc) ?
+		 0x1 : 0;
+
+	return dma_addr | flags;
 }
 
 u16 hantro_h264_get_ref_nbr(struct hantro_ctx *ctx, unsigned int dpb_idx)
@@ -355,6 +394,47 @@ u16 hantro_h264_get_ref_nbr(struct hantro_ctx *ctx, unsigned int dpb_idx)
 	return dpb->frame_num;
 }
 
+/*
+ * Removes all references with the same parity as the current picture from the
+ * reference list. The remaining list will have references with the opposite
+ * parity. This is effectively a deduplication of references since each buffer
+ * stores two fields. For this reason, each buffer is found twice in the
+ * reference list.
+ *
+ * This technique has been chosen through trial and error. This simple approach
+ * resulted in the highest conformance score. Note that this method may suffer
+ * worse quality in the case an opposite reference frame has been lost. If this
+ * becomes a problem in the future, it should be possible to add a preprocessing
+ * to identify un-paired fields and avoid removing them.
+ */
+static void deduplicate_reflist(struct v4l2_h264_reflist_builder *b,
+				struct v4l2_h264_reference *reflist)
+{
+	int write_idx = 0;
+	int i;
+
+	if (b->cur_pic_fields == V4L2_H264_FRAME_REF) {
+		write_idx = b->num_valid;
+		goto done;
+	}
+
+	for (i = 0; i < b->num_valid; i++) {
+		if (!(b->cur_pic_fields == reflist[i].fields)) {
+			reflist[write_idx++] = reflist[i];
+			continue;
+		}
+	}
+
+done:
+	/* Should not happen unless we have a bug in the reflist builder. */
+	if (WARN_ON(write_idx > 16))
+		write_idx = 16;
+
+	/* Clear the remaining, some streams fails otherwise */
+	for (; write_idx < 16; write_idx++)
+		reflist[write_idx].index = 15;
+}
+
 int hantro_h264_dec_prepare_run(struct hantro_ctx *ctx)
 {
 	struct hantro_h264_dec_hw_ctx *h264_ctx = &ctx->h264_dec;
@@ -386,15 +466,29 @@ int hantro_h264_dec_prepare_run(struct hantro_ctx *ctx)
 	/* Update the DPB with new refs. */
 	update_dpb(ctx);
 
-	/* Prepare data in memory. */
-	prepare_table(ctx);
-
 	/* Build the P/B{0,1} ref lists. */
 	v4l2_h264_init_reflist_builder(&reflist_builder, ctrls->decode,
 				       ctrls->sps, ctx->h264_dec.dpb);
+	h264_ctx->cur_poc = reflist_builder.cur_pic_order_count;
+
+	/* Prepare data in memory. */
+	prepare_table(ctx);
+
 	v4l2_h264_build_p_ref_list(&reflist_builder, h264_ctx->reflists.p);
 	v4l2_h264_build_b_ref_lists(&reflist_builder, h264_ctx->reflists.b0,
 				    h264_ctx->reflists.b1);
+
+	/*
+	 * Reduce ref lists to at most 16 entries, Hantro hardware will deduce
+	 * the actual picture lists in field through the dpb_valid,
+	 * dpb_longterm bitmap along with the current frame parity.
+	 */
+	if (reflist_builder.cur_pic_fields != V4L2_H264_FRAME_REF) {
+		deduplicate_reflist(&reflist_builder, h264_ctx->reflists.p);
+		deduplicate_reflist(&reflist_builder, h264_ctx->reflists.b0);
+		deduplicate_reflist(&reflist_builder, h264_ctx->reflists.b1);
+	}
+
 	return 0;
 }
 
diff --git a/drivers/staging/media/hantro/hantro_hw.h b/drivers/staging/media/hantro/hantro_hw.h
index 0b5b9da86c43..52a960f6fa4a 100644
--- a/drivers/staging/media/hantro/hantro_hw.h
+++ b/drivers/staging/media/hantro/hantro_hw.h
@@ -83,6 +83,7 @@ struct hantro_h264_dec_reflists {
  * @ctrls:	V4L2 controls attached to a run
  * @dpb_longterm: DPB long-term
  * @dpb_valid:	  DPB valid
+ * @cur_poc:	Current picture order count
  */
 struct hantro_h264_dec_hw_ctx {
 	struct hantro_aux_buf priv;
@@ -91,6 +92,7 @@ struct hantro_h264_dec_hw_ctx {
 	struct hantro_h264_dec_ctrls ctrls;
 	u32 dpb_longterm;
 	u32 dpb_valid;
+	s32 cur_poc;
 };
 
 /**
-- 
2.34.3
[PATCH v5 20/20] media: hantro: Enable HOLD_CAPTURE_BUF for H.264
Posted by Nicolas Dufresne 1 week, 3 days ago
This is needed to optimize field decoding. Each field will be
decoded into the same capture buffer. To be able to queue multiple
buffers, we need to be able to ask the driver to hold the capture
buffer.

Signed-off-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Reviewed-by: Sebastian Fricke <sebastian.fricke@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
---
 drivers/staging/media/hantro/hantro_v4l2.c | 25 ++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/drivers/staging/media/hantro/hantro_v4l2.c b/drivers/staging/media/hantro/hantro_v4l2.c
index ed458866257a..22ad182ee972 100644
--- a/drivers/staging/media/hantro/hantro_v4l2.c
+++ b/drivers/staging/media/hantro/hantro_v4l2.c
@@ -408,6 +408,30 @@ hantro_update_requires_request(struct hantro_ctx *ctx, u32 fourcc)
 	}
 }
 
+static void
+hantro_update_requires_hold_capture_buf(struct hantro_ctx *ctx, u32 fourcc)
+{
+	struct vb2_queue *vq;
+
+	vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx,
+			     V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE);
+
+	switch (fourcc) {
+	case V4L2_PIX_FMT_JPEG:
+	case V4L2_PIX_FMT_MPEG2_SLICE:
+	case V4L2_PIX_FMT_VP8_FRAME:
+	case V4L2_PIX_FMT_HEVC_SLICE:
+	case V4L2_PIX_FMT_VP9_FRAME:
+		vq->subsystem_flags &= ~(VB2_V4L2_FL_SUPPORTS_M2M_HOLD_CAPTURE_BUF);
+		break;
+	case V4L2_PIX_FMT_H264_SLICE:
+		vq->subsystem_flags |= VB2_V4L2_FL_SUPPORTS_M2M_HOLD_CAPTURE_BUF;
+		break;
+	default:
+		break;
+	}
+}
+
 static int hantro_set_fmt_out(struct hantro_ctx *ctx,
 			      struct v4l2_pix_format_mplane *pix_mp)
 {
@@ -471,6 +495,7 @@ static int hantro_set_fmt_out(struct hantro_ctx *ctx,
 	ctx->dst_fmt.quantization = pix_mp->quantization;
 
 	hantro_update_requires_request(ctx, pix_mp->pixelformat);
+	hantro_update_requires_hold_capture_buf(ctx, pix_mp->pixelformat);
 
 	vpu_debug(0, "OUTPUT codec mode: %d\n", ctx->vpu_src_fmt->codec_mode);
 	vpu_debug(0, "fmt - w: %d, h: %d\n",
-- 
2.34.3