- HLSL changes: [MooglyGuy]
* Upped vertex buffer size to 64k verts, fixes assert in
starwars and alphaone, please include the printed error
message in any subsequent encounterings of the assert.
* Improved vector rendering (beam width 1.5 suggested)
* Ducked raster bloom default to 0.225 to reduce washout
openly, for there was nothing left to see with" [MooglyGuy]
* Enabled vector bloom and associated .ini controls
* Added raster bloom and associated .ini controls, each bloom "level" is the
linear weight of successively half-sized render targets
* Removed D3D8 mode
* Mass renaming in D3D renderer to use namespaces, initial planning step to
HAL-based renderer implementation on Windows (i.e., GL on Windows)
* Converted d3d_info, d3d_poly_info, and d3d_texture_info into classes
* Added batching of vectors for possible speed increase
* Minor cleanup of shader state setting
- Fixed set_vector functionality and simplified shaders as a result
- Fixed HLSL presets, 0 to 3, in increasing level of terribleness
- Reduced options footprint from RGB triplets
Next plan: Separate INI writing.
- Reworked default shadow mask settings, eliminating rainbow banding and matching reference shots more closely
- Moved color power to occur after shadow mask, as it is intended to simulate nonlinear phosphor response
- Added a variable-width notch filter to the Y channel in NTSC post-processing, eliminating luma banding on e.g. CoCo 2 and Apple II
Attempting to fix the HLSL 'blurriness' reported by a few people. Now HLSL will auto-prescale to the nearest texture size that is greater than the target screen size on both axes and is also an even multiple of the raw bitmap's size.
- Added the ability to render screenshots at arbitrary resolutions.
- Added the ability to record AVI videos (albeit with no audio) at arbitrary resolutions.
- Added a 43-tap-wide FIR-based NTSC filter with tunable Y, I and Q frequency response.
- Updated scanlines to have a user-tunable pixel-height ratio in addition to the current screen-height ratio.
- Fixed a VRAM leak that was causing many dynamic-resolution drivers to run out of memory mid-run.
- Switched to point-sampling through most of the shader chain to avoid non-intentional pixel smearing
- Corrected alignment and scaling on the shadow mask and scanlines; dramatically reduces rainbow banding at non-oversampled resolutions.
- Re-worked render target handling to align pixels better, reducing unintentional blurring
- Made major fixes to CVBS simulation, significantly increasing color saturation
- Potential fix for some crashing reported by John IV
- Split color convolution and deconvergence into separate shaders for potential GPU savings down the line
- Added light and heavy variants of the color convolution shader, the former with YIQ colorspace removed
- Re-worked defocus to occur prior to shadow mask application, as it would be on a real monitor.
- Removed Edge Detection, as it was just for fun and can easily be added in by users if desired.
- Split "pincushion" into "Pincushion" and "Screen Curvature", the former affecting the only the displayed image and the latter only affecting the shadow mask.
Focus: 8-sample blur that averages 7 samples around a center sample.
Phosphor: Not currently used, treated as a pass-through by drawd3d.c, but could be used to implement additional convolutions in a second pass.
Pincushion: Used (when commented in in drawd3d.c) to pincushion an entire full-screen texture but not otherwise apply any convolutions.
Post: The meat and potatoes. It does scanlines, it does aperture masking, it does dot crawl, it does chroma subsampling, it does YIQ colorspace convolution, it does RGB colorspace convolution, it does pincushioning, it walks, it talks, it does the dishes, it'll screw your wife for you, and if you don't have a wife it will find one for you, get you married to her, and screw her for you, IT IS THAT GOOD, LADIES AND GENTLEMEN.
Primary: Simple passthrough for UI and artwork.
