Thomas Harte
246b474a25
Removes ONE_BIG_LOCK, having effectively neutered it anyway.
...
Starts work on more explicit API usage validation. Maybe the issue isn't a race condition?
2020-08-16 22:09:25 -04:00
Thomas Harte
745797b596
Introduces a stencil buffer plus the inter-frame clearing it allows.
2020-08-16 16:42:32 -04:00
Thomas Harte
940e9e037e
Adds first_scan to LineMetadata.
...
Also reorders `Line` fields to match `Scan` fields, just for visual consistency.
2020-08-16 08:59:37 -04:00
Thomas Harte
230b9fc9e6
Permits multiple simultaneous scan reading ranges.
...
Also updates the OpenGL scan target as per the latest movements of things.
2020-08-12 22:08:41 -04:00
Thomas Harte
6384ff3ee7
Add fix for data_type_size_ for owners that don't change texture pointer upon new modals.
2020-08-09 21:17:51 -04:00
Thomas Harte
ced4a75a1a
Adds note on the buffering scan target's minor adaptation of data_offset.
2020-08-09 20:44:46 -04:00
Thomas Harte
bdcf266e45
Having learnt a bit more: eliminates Metal attribute tags, switches to more natural expression of structs.
...
Also thereby eliminates the need for a forced alignas(4) on various structs.
2020-08-08 17:27:32 -04:00
Thomas Harte
583f6b1ba2
Modifies BufferingScanTarget to allow has-a relationship.
...
I might switch fully to has-a. Further consideration required.
2020-08-07 22:03:27 -04:00
Thomas Harte
4c9418f59a
Guarantees alignof(4) on all GPU-bound structures.
...
Taken as given: Metal's requirement here is reasonable enough that it'll either be the same as other frameworks, or at least possibly help them down a fast path.
2020-08-07 21:18:08 -04:00
Thomas Harte
6bc66d8b96
Tidies, ensures ::will_change_owner acquires the producer mutex.
2020-07-29 23:18:03 -04:00
Thomas Harte
23f381f381
Fixes frame_is_complete_, gets rid of active_line_, explains ONE_BIG_LOCK in set_write_area.
2020-07-29 23:03:38 -04:00
Thomas Harte
51ad423eca
Resolves off-by-one error in line writing, adds diagnostic one-big-lock option.
2020-07-29 22:45:13 -04:00
Thomas Harte
72a8fef989
Switches to much more straightforward Line/LineMetadata storage.
...
Spoiler: covering this whole segment behind producer_mutex_ seems to resolve all output issues, so clearly the existing logic isn't functioning correctly. Making it simpler seems like a pretty obvious way to get to the bottom of that.
2020-07-29 21:49:17 -04:00
Thomas Harte
02f41ee513
This has become the general producer mutex, might as well name it as such.
2020-07-29 21:34:07 -04:00
Thomas Harte
b7760bb052
Reorders code, gets explicit about memory ordering.
2020-07-28 22:02:22 -04:00
Thomas Harte
2470055d90
Hides the modals.
2020-07-27 23:33:39 -04:00
Thomas Harte
77c5b86acc
Moves ownership of the scan and line buffers out of the BufferingScanTarget.
2020-07-26 22:46:03 -04:00
Thomas Harte
bc6426313e
Localises three of the four macros.
2020-07-26 17:54:33 -04:00
Thomas Harte
8bef7ff4c5
Makes all three PointerSets and is_updating_ private.
2020-07-26 17:27:19 -04:00
Thomas Harte
89acb70091
Slightly reorganise.
2020-07-24 16:20:20 -04:00
Thomas Harte
66165a6dea
Add missing include files.
2020-07-23 23:24:24 -04:00
Thomas Harte
ee1d7eb61f
Makes more buffer-specific stuff private.
2020-07-23 23:06:14 -04:00
Thomas Harte
e260f92988
Privatises write_pointers_mutex_ and write_pointers_.
...
Also gives subclasses control over write-area texture space allocation.
2020-07-23 22:54:40 -04:00
Thomas Harte
74788ccf8e
Pulls the BufferingScanTarget into a separate file.
2020-07-22 22:16:47 -04:00
