hantro: postproc related fixes
Subsystems (2)
cover hantro: postproc related fixes
Doing some more extended testing, we found a few issues when the post-processor is enabled (currently for color conversion).
The two patches below should fix these issues.
Ezequiel Garcia (2):
hantro: h264: Get the correct fallback reference buffer
hantro: postproc: Fix motion vector space allocation
drivers/staging/media/hantro/hantro_h264.c | 2 +- drivers/staging/media/hantro/hantro_hw.h | 4 ++-- drivers/staging/media/hantro/hantro_postproc.c | 4 +++- 3 files changed, 6 insertions(+), 4 deletions(-)
Signature · 1 line
2.27.0
PATCH 1/2 hantro: h264: Get the correct fallback reference buffer
If the bitstream and the application are incorrectly configuring the reference pictures, the hardware will need to fallback to using some other reference picture.
When the post-processor is enabled, the fallback buffer should be a shadow buffer (postproc.dec_q), and not a CAPTURE queue buffer, since the latter is post-processed and not really the output of the decoder core.
--- drivers/staging/media/hantro/hantro_h264.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/media/hantro/hantro_h264.c +1 -1
@@ -325,7 +325,7 @@ dma_addr_t hantro_h264_get_ref_buf(struct hantro_ctx *ctx,*/ dst_buf =hantro_get_dst_buf (ctx); buf = &dst_buf->vb2_buf; - dma_addr =vb2_dma_contig_plane_dma_addr (buf,0 ); + dma_addr =hantro_get_dec_buf_addr (ctx, buf); }return dma_addr;
Signature · 1 line
2.27.0
PATCH 2/2 hantro: postproc: Fix motion vector space allocation
In reply to Ezequiel GarciaWhen the post-processor is enabled, the driver allocates "shadow buffers" which are used for the decoder core, and exposes the post-processed buffers to userspace.
For this reason, extra motion vector space has to be allocated on the shadow buffers, which the driver wasn't doing. Fix this.
This fix removes artifacts on high profile bitstreams, found on certain platforms.
While here, fix the MV layout comment, since the multicore (MC) word is found before the MV buffer.
--- drivers/staging/media/hantro/hantro_hw.h | 4 ++-- drivers/staging/media/hantro/hantro_postproc.c | 4 +++- 2 files changed, 5 insertions(+), 3 deletions(-)
drivers/staging/media/hantro/hantro_hw.h +2 -2
@@ -200,10 +200,10 @@ hantro_h264_mv_size(unsigned int width, unsigned int height)* +---------------------------+ * | UV-plane 128 bytes x MBs | * +---------------------------+ -* | MV buffer 64 bytes x MBs | -* +---------------------------+ * | MC sync 32 bytes | * +---------------------------+ +* | MV buffer 64 bytes x MBs | +* +---------------------------+ */ return 64 *MB_WIDTH (width) *MB_WIDTH (height) +32 ; }
drivers/staging/media/hantro/hantro_postproc.c +3 -1
@@ -118,7 +118,9 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)unsigned int num_buffers = cap_queue->num_buffers;unsigned int i, buf_size; - buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage; + buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage + +hantro_h264_mv_size (ctx->dst_fmt.width, + ctx->dst_fmt.height);for (i =0 ; i < num_buffers; ++i) {struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
Signature · 1 line
2.27.0
Re: [PATCH 2/2] hantro: postproc: Fix motion vector space allocation
Quoted 8 lines · Ezequiel Garcia
When the post-processor is enabled, the driver allocates "shadow buffers" which are used for the decoder core, and exposes the post-processed buffers to userspace.
For this reason, extra motion vector space has to be allocated on the shadow buffers, which the driver wasn't doing. Fix this.
Quoted 5 lines
This fix removes artifacts on high profile bitstreams, found on certain platforms.
While here, fix the MV layout comment, since the multicore (MC) word is found before the MV buffer.
If this is the case then the mv offset currently used is probably wrong for multicore hantro blocks.
The imx-vpu-hantro reference code mentions the following for H.264,
allocate 32 bytes for multicore status fields
locate it after picture and direct MV
and for VP8 it locate it directly after picture.
The Rockchip hantro devices I have is not multicore (to my knowledge), is the iMX8M hantro a multicore block?
Best regards,
Jonas
Quoted 41 lines
Fixes 8c2d66b036c77 ("media: hantro: Support color conversion via post-processing")Signed-off-by Ezequiel Garcia <[email protected]>--- drivers/staging/media/hantro/hantro_hw.h | 4 ++-- drivers/staging/media/hantro/hantro_postproc.c | 4 +++- 2 files changed, 5 insertions(+), 3 deletions(-)drivers/staging/media/hantro/hantro_hw.h +2
-2@@ -200,10 +200,10 @@ hantro_h264_mv_size(unsigned int width, unsigned int height)* +---------------------------+ * | UV-plane 128 bytes x MBs | * +---------------------------+ -* | MV buffer 64 bytes x MBs | -* +---------------------------+ * | MC sync 32 bytes | * +---------------------------+ +* | MV buffer 64 bytes x MBs | +* +---------------------------+ */ return 64 *MB_WIDTH (width) *MB_WIDTH (height) +32 ; }drivers/staging/media/hantro/hantro_postproc.c +3
-1@@ -118,7 +118,9 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)unsigned int num_buffers = cap_queue->num_buffers;unsigned int i, buf_size; - buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage; + buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage + +hantro_h264_mv_size (ctx->dst_fmt.width, + ctx->dst_fmt.height);for (i =0 ; i < num_buffers; ++i) {struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];
Re: [PATCH 2/2] hantro: postproc: Fix motion vector space allocation
Quoted 10 lines · Jonas Karlman
Quoted 8 lines · Ezequiel Garcia
When the post-processor is enabled, the driver allocates "shadow buffers" which are used for the decoder core, and exposes the post-processed buffers to userspace.
For this reason, extra motion vector space has to be allocated on the shadow buffers, which the driver wasn't doing. Fix this.
Quoted 9 lines
Quoted 5 lines
This fix removes artifacts on high profile bitstreams, found on certain platforms.
While here, fix the MV layout comment, since the multicore (MC) word is found before the MV buffer.
If this is the case then the mv offset currently used is probably wrong for multicore hantro blocks.
Oh, yes, my bad.
Quoted 10 lines
The imx-vpu-hantro reference code mentions the following for H.264,
allocate 32 bytes for multicore status fields
locate it after picture and direct MVand for VP8 it locate it directly after picture.
The Rockchip hantro devices I have is not multicore (to my knowledge), is the iMX8M hantro a multicore block?
Newer cores (newer than G1 and G2) specify the MC status before the MV buffer, and I was too quick to assume it was the case also for G1.
I'll get a v2.
Thanks for the catch,
Ezequiel
Quoted 44 lines
Best regards,
JonasQuoted 41 lines
Fixes 8c2d66b036c77 ("media: hantro: Support color conversion via post-processing")Signed-off-by Ezequiel Garcia <[email protected]>--- drivers/staging/media/hantro/hantro_hw.h | 4 ++-- drivers/staging/media/hantro/hantro_postproc.c | 4 +++- 2 files changed, 5 insertions(+), 3 deletions(-)drivers/staging/media/hantro/hantro_hw.h +2
-2@@ -200,10 +200,10 @@ hantro_h264_mv_size(unsigned int width, unsigned int height)* +---------------------------+ * | UV-plane 128 bytes x MBs | * +---------------------------+ -* | MV buffer 64 bytes x MBs | -* +---------------------------+ * | MC sync 32 bytes | * +---------------------------+ +* | MV buffer 64 bytes x MBs | +* +---------------------------+ */ return 64 *MB_WIDTH (width) *MB_WIDTH (height) +32 ; }drivers/staging/media/hantro/hantro_postproc.c +3
-1@@ -118,7 +118,9 @@ int hantro_postproc_alloc(struct hantro_ctx *ctx)unsigned int num_buffers = cap_queue->num_buffers;unsigned int i, buf_size; - buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage; + buf_size = ctx->dst_fmt.plane_fmt[0 ].sizeimage + +hantro_h264_mv_size (ctx->dst_fmt.width, + ctx->dst_fmt.height);for (i =0 ; i < num_buffers; ++i) {struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i];