New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 915051 link

Starred by 1 user
Status: Untriaged
Owner: ----
Cc:
hiroh@chromium.org
gurcheta...@chromium.org
█ marc...@chromium.org
tfiga@chromium.org
Components:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 3
Type: Bug



Sign in to add a comment

minigbm [NV12, x86]: UVPlane's size is the same as YPlane's size

Project Member Reported by hiroh@chromium.org, Dec 14 
In NV12, UV Plane's size is half of Y Plane's size.
However, on x86, minigbm allocates UV Plane whose size is the same as Y Plane's size and gbm_bo_get_plane_size(bo, 1) returns Y Plane's size.
For example, in the case of NV12 on x86,
gbm_bo_get_plane_size(bo, 0): 138240
gbm_bo_get_plane_size(bo, 1): 138240

This is not critical issue, because seems like minigbm actually allocates the size on the second plane. 

Memory usage will be saved if this is fixed.

Comment 1 by gurcheta...@chromium.org, Dec 14 

We have special alignment for i915:

https://chromium.googlesource.com/chromiumos/platform/minigbm/+/master/i915.c

What's the size dimensions of the buffer?  Also can you call gbm_bo_get_format_modifier so we can know which type of buffer it is?  That way we can tell if it's by design or a bug.

Comment 2 by hiroh@chromium.org, Dec 14 

I cannot recall dimension right now.
I remember modifier was 0.

Comment 3 by arthurhsu@google.com, Dec 14

Components: OS>Kernel>Graphics

Comment 4 by marcheu@google.com, Dec 14 

I thought we were doing this because of a limitation of intel hw...

Comment 5 by gurcheta...@chromium.org, Dec 14 

If the modifier was 0, then we have I915_TILING_NONE


https://chromium.googlesource.com/chromiumos/platform/minigbm/+/master/i915.c#219

For that, we have libva alignment requirements.  hiroh@, can you get the width/height whenever you get the chance?  That'll let us do a back of the envelope calculation about this.

Comment 6 by hiroh@chromium.org, Dec 15 

Sure. I observed it when I play video with SW decoder.
Let me check size on Monday.

Comment 7 by hiroh@chromium.org, Dec 17 

I checked those values again.
When the size is (640, 360), format is YUV_420_BIPLANAR (i.e. NV12) and modifier is zero (i.e. non-tiled), the size of planes are,
1st: 230400
2nd: 230400

Comment 8 by tfiga@chromium.org, Dec 17 

It's quite strange. I think we may need to do some printfs in minigbm, because my inspection of the code couldn't spot a way that would make the 2nd plane the same as 1st for this format and resolution.
►

Sign in to add a comment