Workers now directly operate on their state buffer.

To quote the documentation:
First, within a single shader invocation, if you perform an incoherent memory write, the value written will always be visible for reading. But only through that particular variable and only within the shader invocation that issued
the write. You need not do anything special to make this happen.

This means there is nothing stopping us from just reading and writing in the buffer as if the values were local variables, as they are specific to a given invocation. Also, buffer access happens infrequently, and furthermore
there's nothing stopping the GPU from caching the values, as memory access is incoherent anyhow.

Long story short: Now the code uses only 28 VGPRs, meaning that the AMD GCN hardware ideal of 32 is again achieved.

0 files changed, 0 insertions, 0 deletions