- The new behavior for the Multisample Antialiasing checkbox: Checked - GFX3D_Renderer_MultisampleSize = 4, Unchecked - GFX3D_Renderer_MultisampleSize = 0. (If someone else wants to make some UI so that GFX3D_Renderer_MultisampleSize can be set to other sizes, then have at it.)
- Add a unique sequence number to fetched frames to ensure that older frames are not drawn after newer frames.
- After much research, finally settle on a method for fetching the NDS framebuffers -- using a MTLBlitCommandEncoder to blit a MTLBuffer to a MTLTexture. It is faster than uploading a texture using [id<MTLTexture> replaceRegion:mipmapLevel:withBytes:bytesPerRow:], and also faster than using a pinned-memory backed linear texture. This method will be the way going forward for fetching framebuffers in Metal.
- All frontends will need to be updated to use the new GFX3D_Renderer_MultisampleSize setting.
- This change obsoletes GFX3D_Renderer_Multisample, which currently does nothing at the moment. It will be removed after all frontends are updated.
- Received packets are now queued properly and should no longer be overwritten or lost.
- Received packets under Ad-hoc mode now use the same transfer delay as Infrastructure mode. (Read one halfword every 8 microseconds.)
- Received packet transfer delay only works when the emulation level is set to WifiEmulationLevel_Compatibility. Transfer delay can be disabled by setting the emulation level to WifiEmulationLevel_Normal, which will cause the entire received packet to be transferred immediately.
- If the WiFi emulation level is Off, then always set POWER_US.Disable = 1.
- If POWER_US.Disable == 1, then do not trigger any further WiFi actions.
- The name contained within DeSmuME's frame header has been changed from "NDSWIFI\0" to "DESMUME\0".
- DeSmuME's frame header size has been increased from 12 bytes to 16 bytes.
- Baseband data is now set to a default dataset on reset.
- Baseband data reads/writes now respect the actual R/W behavior of each data byte.
If the position test register is negative, conversion to unsigned
integer is undefined. This breaks games on arm64 where the behaviour is
defined as 'truncate to zero'. Converting to a signed integer first
guarantees the intended behaviour.
- GPUEngineBase::_LineCopy() optimizations only apply to 2x, 3x, and 4x scaling.
- Add SSE2 version of 3x CopyLineExpand() when using ELEMENTSIZE==1.
- Add SSE2 versions of CopyLineReduce() and add specific 2x/3x/4x versions of CopyLineReduce_*() algorithms.
- CopyLineExpand() now supports vertical scaling in addition to horizontal scaling.
- GPU buffers that were previously only cache-aligned are now page-aligned if appropriate.
- Also fix a depth bug for scrolling clear images on SSE2 systems by disabling the SSE2-specific code. This issue will need to be researched at a later date.
rogerman