[AutoDiff] Do not propagate same adjoint buffer multiple times

[asl]

Adjoint buffers of projections (e.g. obtained via begin_access) are same as adjoint buffer of underlying struct value. As a result, when propagating adjoint values to pullback successor blocks we tend to produce lots of identical copies (essentially for every struct access and in every basic block) of adjoint buffers.

These copy_addrs instructions are then lowered down to plain loads and stores and while the redundant copies are usually optimized away by subsequent optimization passes, presence of such copies leads to elevated memory consumption and compilation time as one needs to track liveness of these values being copied.

Track the values being propagated and simply do not generate extra copies if the same value was already propagated.

One step towards #61773

[asl]

Tagging @BradLarson

[asl]

@swift-ci please test

[dan-zheng]

Nice fix!

(Duplicate copy_addr for adjoint buffers has been a longstanding control flow autodiff issue, huh?)

[asl]

@dan-zheng This is only tip of the iceberg as we need to collect the cases that impact performance. Here essentially every struct access would generate new adjoint buffer copy for every basic block, so number of copies might scale quadratically as # struct accesses x # of bb's Later these copy_addrs would be exploded to multiple instructions leading to even more "fun" SIL / LLVM IR to optimize. Poor GVN :)

[asl]

@swift-ci please test

[asl]

@swift-ci please test macos