[VAN-798] correct missing "run" functions and subcmp arguments #71
Conversation
| //CHECK-NEXT: %call.fr_sub = call i256 @fr_sub(i256 %4, i256 1) | ||
| //CHECK-NEXT: %5 = getelementptr i256, i256* %subc_[[X3]], i32 0 | ||
| //CHECK-NEXT: store i256 %call.fr_sub, i256* %5, align 4 | ||
| //CHECK-NEXT: call void @Sum_0_run([0 x i256]* %sub_[[X3]]) |
There was a problem hiding this comment.
Ah I see, interesting consequence of us immediately emitting the run call after assigning all the input signals is that we effectively have 4 copies of the same loop body when we could have 2. Avoiding this in all cases seems impossible, but at least in this case, we would avoid it if we generated the run call right before using the subcomponent's output rather than right after the last input signal is assigned (since the assignment is outside of the loop). My guess is that this is not a very important optimization, though (and it's certainly outside the scope of this PR---just some thought I had while trying to understand why there were more functions than I expected).
There was a problem hiding this comment.
We would have to be careful to only call the run function once even if the output is read more than once or there is more than one output signal.
There was a problem hiding this comment.
By construction we know that all input signals will be assigned before we can read any outputs so I like Ian's idea in that regards but there are components that sometimes you don't need to read the output (i.e Num2Bits because its constrained within) so we will be missing the run function of that one in some cases
also stabilize lit test output ordering for flattened function version
tim-hoffman
force-pushed
the
th/van-798
branch
from
abfb4c5 to
9a31d04
Compare
also stabilize lit test output ordering for flattened function version