This is a fork from the very nice work here https://github.com/Overv/vramfs. As I needed a fast enough userland ramdisk, none of the existing fuse examples work reliably.

A few changes from the vramfs:

• no more cl dependency and of course no more ability to allocate on gpu.
• convert to pure ramfs
• memory grow dynmically
• some bugs in the original where file permissions were not copied properly.

• dir size not getting reported properly. vram_readdir() buggy?
• performance is slower than /dev/shm. check if overhead of fuse can account for this disparity.