Thomas Harte
274867579b
Deploys constexpr as a stricter const.
2019-12-22 00:22:17 -05:00
Thomas Harte
e9318efeb6
Switches to std::make_shared/make_unique in a bunch of applicable places.
...
No doubt many more similar improvements are available, these are just the ones that were easy to find.
2019-12-21 23:34:25 -05:00
Thomas Harte
cc7ff1ec9e
Corrects typo.
2019-11-09 14:59:35 -05:00
Thomas Harte
9d12ca691f
Makes meaning of 2048 here explicit.
2019-11-09 14:59:16 -05:00
Thomas Harte
68f9084d9e
Fixes timing of wrap.
2019-11-07 23:24:51 -05:00
Thomas Harte
f9a35c6636
Corrects potential wrap error for non-single-byte targets.
2019-11-07 22:23:26 -05:00
Thomas Harte
6c75c60149
Merge branch 'master' into AtariST
2019-11-05 22:05:51 -05:00
Thomas Harte
35f1a7ab10
Resolves a potential out-of-bounds array access.
...
Risk was: allocation exactly joins end of buffer. In which case the next get_data won't wrap the texture y coordinate since it won't spot an x overage.
2019-11-05 21:47:40 -05:00
Thomas Harte
1c154131f9
Expands size of storage in Cycles/HalfCycles; adjusts widely to compensate.
2019-10-29 22:36:29 -04:00
Thomas Harte
92847037b3
Merge branch 'master' into AtariST
2019-10-27 21:40:51 -04:00
Thomas Harte
4266264449
Switches to the more idiomatic .empty().
2019-10-27 21:31:31 -04:00
Thomas Harte
3794d94b68
Adds a sanity check on alignment.
2019-10-27 14:09:02 -04:00
Thomas Harte
d1dd6876b5
Adds the option to affix a standard prefix to log messages.
2019-10-20 20:38:55 -04:00
Thomas Harte
684644420a
Increases scan buffer availability.
2019-10-20 17:22:41 -04:00
Thomas Harte
9c7aa5f3fc
Attempts also to spot data writes without allocations.
2019-10-19 18:26:56 -04:00
Thomas Harte
418cd07e17
Adds a check against overrunning data.
2019-10-19 18:17:44 -04:00
Thomas Harte
19198ea665
Improves const usage.
2019-07-16 22:13:47 -04:00
Thomas Harte
60cedca97b
Adds cmath in support of ceilf.
2019-07-08 18:14:03 -04:00
Thomas Harte
f6f9024631
Corrects Macintosh aspect ratio (and framing).
2019-06-13 18:41:38 -04:00
Thomas Harte
66540ff86f
Permits end_data only after a begin_data.
2019-06-03 21:56:53 -04:00
Thomas Harte
b860ba2ee3
Disallows smaller buffer use for 'sharp' displays and tightens sampling window.
2019-06-03 15:58:14 -04:00
Thomas Harte
d25ab35d58
Finally gets setw usage correct.
2019-04-15 12:41:56 -04:00
Thomas Harte
6f7dd10d95
Reduces draw/update contention.
...
This won't yet have any effect on either port owing to the way they handle contexts, but here it is.
2019-03-07 19:28:32 -05:00
Thomas Harte
24fb95291a
Reverts to support a full RGBA colour buffer.
2019-03-07 19:22:40 -05:00
Thomas Harte
84422676cb
Switches to more coherent logic about buffer sizing.
2019-03-06 19:19:30 -05:00
Thomas Harte
7441e3f4c5
Corrects aspect ratio when changing accumulation texture size.
2019-03-05 22:10:32 -05:00
Thomas Harte
f18132d674
Makes effort to round out draft 1 of Outputs::Display::Metrics.
2019-03-05 22:01:58 -05:00
Thomas Harte
5660007221
Experimentally introduces adaptive quality intermediate buffers.
2019-03-05 21:41:20 -05:00
Thomas Harte
86a6b04d4a
Begins attempts to keep track of display metrics.
...
i.e. a system that can both make smart decisions about when to use a lower resolution, and hopefully allow some sort of flywheel-esque horizontal retrace synchronisation. And possibly some raster beam chasing?
2019-03-04 21:54:50 -05:00
Thomas Harte
641e349f33
Switches to an 8bpp stencil, for Nvidia compatibility.
2019-03-03 20:38:24 -05:00
Thomas Harte
ccdeb3fbc8
Ensures draw is a no-op prior to pipeline setup.
2019-03-03 15:04:14 -05:00
Thomas Harte
bee0d09877
Splits display update and draw functions.
...
On the Mac, draw is now called without an update for resizing events, and
anything else requested by AppKit. In all other cases — including from
the SDL version — both are called as if they were still a single function.
2019-03-02 19:33:28 -05:00
Thomas Harte
d97348dd38
Eliminates dangling uses of printf.
2019-03-02 18:07:05 -05:00
Thomas Harte
d122535a65
Unifies the OpenGL screenshot code and corrects it for arbitrary alignment.
2019-02-27 21:05:02 -05:00
Thomas Harte
db078c7363
Minor tweak: don't start counting phase from zero.
...
This should ensure no first pixel issues resulting from clamping.
2019-02-27 18:51:51 -05:00
Thomas Harte
6b4f6971de
Disables upper limit on frame buffer size.
...
Filtering is sufficiently imperfect as to make this justifiable only when performance requires it. So I need a test for that. Marked as TODO.
2019-02-26 22:39:07 -05:00
Thomas Harte
79707a3c66
Improves filtering slightly, and ensures coefficients are always set when needed.
2019-02-26 22:35:55 -05:00
Thomas Harte
694783efe9
Brings S-Video inside the group that filters luminance.
...
Thereby revealing some sort of error in offset selection.
2019-02-26 22:27:40 -05:00
Thomas Harte
68c5474e36
Reintroduces basic filtering for RGB mode (and introduces it for monochrome composite).
2019-02-26 22:21:49 -05:00
Thomas Harte
cd055a0298
Introduces an upper bound on output resolution, and resolves full-screen clear colour.
2019-02-25 22:07:48 -05:00
Thomas Harte
8f2abab0d9
Ensures texture targets are initially clear.
2019-02-25 21:55:14 -05:00
Thomas Harte
4c5dee866b
Ensures a proper black fill for Luminance8Phase8 input data.
2019-02-25 21:32:15 -05:00
Thomas Harte
7bb90c78d9
Resolves out-of-bounds initial condition whenever this loop began with start_line = 2047.
...
This, I believe, was the remaining cause of screen flashes.
2019-02-24 19:50:19 -05:00
Thomas Harte
a6e61ef83b
Reverts the clear colour to black.
...
The change was related to debugging; it was not intentionally committed.
2019-02-24 14:36:08 -05:00
Thomas Harte
c775a6c0f8
Introduces but disables a couple of bits of logging that might be helpful again in the future.
2019-02-24 14:30:39 -05:00
Thomas Harte
2f9e825728
Forces the outward-communicated composite angles to have the same precision as the cycle counts.
...
While also making a minor improvement to output range. Which is neither here nor there.
2019-02-24 14:29:43 -05:00
Thomas Harte
2f491b5be1
Reintroduces fragment snapping for composite colour sampling.
...
Thereby uncovers some sort of slightly-off recording of scan lines. On the Apple II, individual scans reach the ScanTarget at a density of exactly 0.25 colour cycles per pixel. So that timing information propagates exactly. But the whole lines that are composed via ::announce end up trying to fit 0.250154 colour cycles per pixel. Which creates a phase error as the display progresses from left to right.
This will need to be resolved in order to be able to fix the Apple II's intended colour phase. But, also, it's probably what was wrong with the Oric. And, quite possibly, why the single-step shader didn't work.
2019-02-24 13:39:14 -05:00
Thomas Harte
de7ebead23
Ensures the line_allocation_has_failed_ condition can be exited.
2019-02-21 22:30:41 -05:00
Thomas Harte
c0c4704419
Ensures that failure to allocate a line blocks all other allocations.
2019-02-21 21:38:48 -05:00
Thomas Harte
080f949f89
Ensures OpenGL version is logged prior to any other GL calls.
2019-02-20 20:21:17 -05:00
