08-24-2017, 03:28 PM
Oh, DX9 is old school 
For OpenGL, all data is copied into a circular pixel buffer object. And I guess a dma read command is sent to the GPU. So before the draw, the GPU will get the data, and then will do the draw. So it should be lock-less.
For the code, basically we need to split GSDevice::FetchSurface/GSDevice::CreateSurface into 2 code paths. One will manage the standard 2d texture, the others one will manage the texture. So when you need to create a palette, we first check that if any palette with same data already exists. Or maybe it can be done in GSTextureCache (lookup stuff) without any modification of GSDevice function. It won't be easy to not impact the performance of the others games.
For OpenGL, all data is copied into a circular pixel buffer object. And I guess a dma read command is sent to the GPU. So before the draw, the GPU will get the data, and then will do the draw. So it should be lock-less.For the code, basically we need to split GSDevice::FetchSurface/GSDevice::CreateSurface into 2 code paths. One will manage the standard 2d texture, the others one will manage the texture. So when you need to create a palette, we first check that if any palette with same data already exists. Or maybe it can be done in GSTextureCache (lookup stuff) without any modification of GSDevice function. It won't be easy to not impact the performance of the others games.