[PATCH] D63505: [InstCombine] Fold icmp eq/ne (and %x, ~C), 0 -> %x u</u>= 0 earlier, C+1 is power of 2.

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

Looks good for IR, but looks like this needs an undo fold for backend?
https://godbolt.org/z/giY9Cp
At least for aarch64 this seems to result in worse ASM?



================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:1643-1645
   // Don't perform the following transforms if the AND has multiple uses
   if (!And->hasOneUse())
     return nullptr;
----------------
Why are we restricting this fold to single-use `and`?


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:1655
+
+    // ((X & ~7) == 0) --> X u< 8
+    // If X is (BinOp Y, C3), allow other rules to fold C3 with C2.
----------------
`((%x & C) == 0) --> %x u< (-C)  iff (-C) is power of two.`


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:1656-1657
+    // ((X & ~7) == 0) --> X u< 8
+    // If X is (BinOp Y, C3), allow other rules to fold C3 with C2.
+    if (!match(X, m_c_BinOp(m_Value(), m_Constant())) &&
+        (~(*C2) + 1).isPowerOf2()) {
----------------
Why do we care about that? Seems rather arbitrary.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63505/new/

https://reviews.llvm.org/D63505