https://bugs.winehq.org/show_bug.cgi?id=54223
--- Comment #17 from Zeb Figura z.figura12@gmail.com --- Okay, I am sure I tried to reproduce this before on an AMD machine, but I just tried now on a different machine and I can reproduce the performance difference.
The constant buffers used by the application are *all* small (mostly < 256 bytes, some 1536 and 3072 bytes), GPU-only, uploaded with UpdateSubResource().
I don't think the buffer chunk triggers any different logic in wined3d. In every case we're using glBufferSubData() and glBindBufferRange() with the size of the buffer.
Do drivers just not like these GPU-only buffers to be suballocated? Since they're not mappable we don't get anything out of it, so maybe that's the right call. But I'd kind of like to know why.