Accelerated blits of multisampled textures on gpus

Abstract

Systems, computer readable media, and methods for hardware accelerated blits of multisampled textures on graphics processing units (GPUs) are disclosed. For multisampled surfaces, texture-to-buffer blits cannot be trivially implemented because most GPUs do not support writing multisampled surfaces with a linear memory layout. Moreover, GPUs often have a maximum limit for row stride (i.e., the number of bytes from one row of pixels in memory to the next) and/or texture size. When the destination buffer for the blit of a multisampled texture is too large to be aliased by an equivalent non-multisampled texture view, the stride of the view has no spatial relationship with the destination buffer. Thus, to access the source texture correctly, a 'remapping' may be performed to determine the linear sample index of a fragment within the view, and the destination buffer stride may be used to compute the texture coordinates used to sample the source texture.