mpp provides wrong hstride when decoding this hevc file

[hbiyik]

always reproducable. for the below 1080p hevc file with of NV12 frame mpp hevc decoder gives 2304 hstride which is completely wrong, surrely no idea where 2304 is coming from, it is supposed to be 1920?

This smells like a bug unless there is something wrong in the hevc file.

hevc file is at https://filetransfer.io/data-package/jD2ACyB0#link due to filesize limit

it is jusr decapsulated form of this well known test file: https://test-videos.co.uk/vids/bigbuckbunny/mp4/h265/1080/Big_Buck_Bunny_1080_10s_30MB.mp4

[HermanChen]

This is an optimization for hardware decoder. The stride for hevc decoder is aligned to 256 odd times value. 2304 is 256 * 9 which is 1920 aligned value. This behavior can be changed in the code.

[hbiyik]

ok, when i saw the huge hstride i thought something was wrong, thanks for the explanation

[hbiyik]

I am reopeening this, since this is really creating a big issue in my code. How can i disable this? I tried MPP_COMPAT_DEC_FBC_HDR_256_ODD=0 in mpp_compat.h but still decoder spits 256 oddaligned strides.

[hbiyik]

@HermanChen

[HermanChen]

The MPP_COMPAT_DEC_FBC_HDR_256_ODD is for fbc mode stride setup. Can you setup the correct hor_stirde in your code using decoder output? Or you code must set the hor_stride to be the same with width?

[hbiyik]

do you mean something like this? if so, should i create a buffer manually of will mpp take care of it?

mpp_frame_init(&mppframe);
mpp_frame_set_hor_stride(mppframe, EXPECTED_STRIDE_ALIGNED_TO16);
//should I set a buffer or will mpp handle this?
codec->mpi->decode_get_frame(codec->ctx, &mppframe);

[HermanChen]

7d39aba.diff.zip Try this patch and set MPP_COMPAT_DEC_NFBC_NO_256_ODD compat value to 1

[hbiyik]

@HermanChen

I just had the chance to test this and it works generally but there are 2 problems.

1st, the pointer checks should be fixed as below, otherwise it always disable 256 oddaling (due to address check not value check.)

diff --git a/mpp/hal/rkdec/h265d/hal_h265d_com.c b/mpp/hal/rkdec/h265d/hal_h265d_com.c
index d7e8db7..40639d4 100644
--- a/mpp/hal/rkdec/h265d/hal_h265d_com.c
+++ b/mpp/hal/rkdec/h265d/hal_h265d_com.c
@@ -88,7 +88,7 @@

 RK_U32 hevc_hor_align(RK_U32 val)
 {
-    if (compat_ext_non_fbc_no_256_odd)
+    if (*compat_ext_non_fbc_no_256_odd)
         return MPP_ALIGN(val, 8);

     return MPP_ALIGN(val, 256) | 256;
diff --git a/mpp/hal/rkdec/vp9d/hal_vp9d_com.c b/mpp/hal/rkdec/vp9d/hal_vp9d_com.c
index eacd11c..59d217b 100644
--- a/mpp/hal/rkdec/vp9d/hal_vp9d_com.c
+++ b/mpp/hal/rkdec/vp9d/hal_vp9d_com.c
@@ -997,7 +997,7 @@

 RK_U32 vp9_hor_align(RK_U32 val)
 {
-    if (compat_ext_non_fbc_no_256_odd)
+    if (*compat_ext_non_fbc_no_256_odd)
         return MPP_ALIGN(val, 64);

     return MPP_ALIGN(val, 256) | 256;

The VP9 frames are provided with 64 aligned hstrides but, the dispayed picture has another stride becase i can see the displayed picture is broken by a wrong stride.

When above fixes are applied, it works as expected for h264 & h265 frames.