Allow gpu ringbuffer to wrap around.

Allow gpu ringbuffer to wrap around.

The ringbuffer was unwilling to wrap around unless the last block filled it up completely - otherwise it would have to wait until it was completely empty. Fix this in the most straightforward way by filling the ringbuffer up with a padding block if necessary.

BUG=80946
TEST=

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=83924

[piman, 2011-05-03 16:30:20]

LGTM + nit.

http://codereview.chromium.org/6901138/diff/1/gpu/command_buffer/client/ring_...
File gpu/command_buffer/client/ring_buffer.cc (right):

http://codereview.chromium.org/6901138/diff/1/gpu/command_buffer/client/ring_...
gpu/command_buffer/client/ring_buffer.cc:63: //Add padding to fill space before
wrapping around
1 space after //

[greggman, 2011-05-03 17:22:31]

Is there a reason you need a state for the blocks? It's a ring buffer so the
status of individual blocks don't matter. All that matters is which tokens have
passed which tells you which blocks are now free and which are still in use.
They don't need their own status do they?

[jbauman, 2011-05-03 17:31:03]

On 2011/05/03 17:22:31, greggman wrote:
> Is there a reason you need a state for the blocks? It's a ring buffer so the
> status of individual blocks don't matter. All that matters is which tokens
have
> passed which tells you which blocks are now free and which are still in use.
> They don't need their own status do they?

Not really, but I wanted some way to distinguish blocks that legitimately didn't
have a token, as that lets us continue to check the requirement that we have
enough space even counting blocks which are in use and haven't yet been freed
pending a token. It also doesn't require that we throw in an unnecessary token,
as that could cause pointless stalls or be complex in other ways.

[greggman, 2011-05-03 17:59:34]

lgtm