nihstro
nihstro copied to clipboard
Check if input operands may simultaneously be used as output operands
It sounds likely that hardware does not actually support this. A hardware test should be written to verify that it doesn't, and a diagnostic should be added to make sure source code doesn't try to do this.
Example of what I'm talking about: add r4, i0, r4. Note that r4 is used both as an input and an output of the addition.
Caelina's shaders actually use the form add r4, r1, r4 quite a lot. Is there documentation suggesting that the hardware may not be supporting this?
Hmm interesting. There is no specific documentation which suggested this, it's just something which sounded reasonable.