[Mesa-dev] [Bug 102204] GL_ARB_buffer_storage crippled extension on r600, radeonsi and amdgpu Mesa drivers

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=102204

Matias N. Goldberg <dark_sylinc at yahoo.com.ar> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dark_sylinc at yahoo.com.ar

--- Comment #10 from Matias N. Goldberg <dark_sylinc at yahoo.com.ar> ---
OK this left me wondering.

I looked further into what happens when you call glBufferStorage, and it seems
that buffer_usage in st_cb_bufferobjects.c
https://github.com/anholt/mesa/blob/master/src/mesa/state_tracker/st_cb_bufferobjects.c#L226
stores all buffers in VRAM unless GL_CLIENT_STORAGE_BIT is used because it
always returns PIPE_USAGE_DEFAULT.

The chosen pipe_resource_usage will end up in si_init_resource_fields
(https://github.com/anholt/mesa/blob/master/src/gallium/drivers/radeonsi/si_buffer.c#L103)
which ends up putting the buffers all buffers in VRAM with Write Combining
(unless GL_CLIENT_STORAGE_BIT is set)

That is... an odd choice. This is specially bad for buffers requested with
GL_MAP_READ_BIT flags, which clearly should not be stored in VRAM with WC bits
set.

This indeed looks like a bug to me. Unfortunately, these hint flags don't map
really well to HW.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20181103/ebbb0c31/attachment-0001.html>