Tracking issue for rendering bugs on certain SDL_Renderer backends

[nmlgc]

Trying to run every SDL_Renderer backend on every GPU ever released will invariably uncover a potentially overwhelming number of bugs. Since these aren't really my fault, the policy is as follows: If the game works without issues on at least one of the backends (including the software one), the bug is in SDL_Renderer or (more likely) the GPU or its driver, and it's not eligible for a free fix according to the ReC98 bugfix policy. If these backend-specific bugs really bother you, file an issue with SDL and investigate it there. If SDL fixes or works around such a bug, I will release a fixed build of the latest version of this fork, with a recompiled SDL DLL, for free.

OpenGL ES 2 × Intel iGPUs × Geometry scaling: Last render call on each frame is omitted

Exit is not rendered.

OpenGL.ES.2.x.Intel.iGPUs.3.webm

Note how the game flips between omitting the number of credits and the Y of DEMO PLAY, depending on what is rendered last on each frame.

Introduced in SDL commit f9b918ff403782986f2a6712e6e2a462767a0457 with the move to client-side arrays. Can be reproduced with the testsprite2 example program:

testsprite2 --renderer opengles2 2 will only show a single bouncing smiley face on this iGPU.

Confirmed on two separate Intel iGPUs (HD Graphics 630 and UHD Graphics 630); switching to a discrete Nvidia GPU made the error go away, as does using framebuffer scaling.

Direct3D 12: Slowdowns with lots of sprites on screen (particularly with large bullet cancels after the Extra Stage midboss)

Tracked in #35.

Direct3D 11: Not working on Wine

SDL 2's Direct3D 11 backend uses the IDXGISwapChain1::SetRotation() method, which Wine (as of version 9.18) doesn't implement: