Mistake in generation of Egg expressions about diospyros HOT 6 CLOSED

Interesting… maybe I'm just being slow, but I guess I expected that these would be legible as S-expressions instead of this flat symbol list. (Put differently, I'm not exactly sure how to read this RecExpr dump as an expression.)

Is there a chance that this is the output of just formatting a RecExpr directly (i.e., .to_string()), but calling the pretty() method on RecExpr would yield a more legible S-expression?

from diospyros.

I should have started off by printing the S-expressions, and I realized that the mistake arises in how I convert the LLVM IR to Egg expressions, so I have changed the title of the issue to be less misleading.

The new S-expression input is:
(VecMAC (Vec no-temp-name4 (Get no-array-name4 1,) 0 0) (Vec 1 no-temp-name5 (Get no-array-name5 1,) 1) (Vec no-temp-name4 (Get no-array-name3 1,) (Get no-array-name6 3,) 0))
and the new S-expression output is
(VecMAC (Vec (Get no-array-name7 1,) (Get no-array-name9 2,) 0 0) (Vec 1 (Get no-array-name8 2,) (Get no-array-name10 3,) 1) (Vec no-array-name7 no-temp-name6 (Get no-array-name11 2,) 0))

The corrected reason for the problem is that no-array-7 appears by itself as a node in the input S-expression.

from diospyros.

Huh, interesting… this might be dumb, but should I be surprised that the symbol no-array-name7 only appears in the output? It seems weird that egg optimization would "invent" a new array instead of just rearranging accesses to existing arrays…

from diospyros.

This is completely my mistake; I made a copy and paste error, because the correct initial input should be:
(Vec (+ (Get no-array-name7 1,) no-array-name7) (+ (Get no-array-name9 2,) (* (Get no-array-name8 2,) no-temp-name6)) (* (Get no-array-name10 3,) (Get no-array-name11 2,)) 0)

from diospyros.

Got it. I'm still not 100% clear on why the conclusion is that the LLVM->egg translation is wrong... (or did I misunderstand)? Can you point to the subexpression that was generated incorrectly?

from diospyros.

Now resolved.

from diospyros.