[PATCH] D134415: [InstCombine] Improve fold icmp of truncated left shift

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:1561
+    if (C1->isPowerOf2()) {
+      // (trunc (2**C1 << Y) to iN) == 0 -> Y u>  N - C1 + 1
+      // (trunc (2**C1 << Y) to iN) != 0 -> Y u<= N - C1 + 1
----------------
That looks correct, but I'd prefer that this comment matches what the code and the alive proof actually do, so I'd change it to something like this:

```
// (trunc (2**C1 << Y) to iN) == 0 -> Y u> ctlz(trunc(C1))
```


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:1569-1570
+      }
+      // (trunc (2**C1 << Y) to iN) == 2**C --> Y == C - C1
+      // (trunc (2**C1 << Y) to iN) != 2**C --> Y != C - C1
+      if (C.isPowerOf2() && C.uge(*C1))
----------------
Add an Alive2 proof for this pattern to the patch description?


================
Comment at: llvm/test/Transforms/InstCombine/icmp-trunc.ll:409
 
 ; TODO: A > 4
 
----------------
Remove TODO.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134415/new/

https://reviews.llvm.org/D134415