[InstCombine] Fold "x ?% y ==/!= 0" to "x & (y-1) ==/!= 0" iff y is power-of-two
authorRoman Lebedev <lebedev.ri@gmail.com>
Tue, 30 Jul 2019 15:28:22 +0000 (15:28 +0000)
committerRoman Lebedev <lebedev.ri@gmail.com>
Tue, 30 Jul 2019 15:28:22 +0000 (15:28 +0000)
commitbe612ea471bf3cef443f04588434b1648c27d532
tree788589941730b5424b84d5401195206bb11d45dd
parentc960c0a4915b18592691288c14d2cebd806197a9
[InstCombine] Fold "x ?% y ==/!= 0" to  "x & (y-1) ==/!= 0" iff y is power-of-two

Summary:
I have stumbled into this by accident while preparing to extend backend `x s% C ==/!= 0` handling.

While we did happen to handle this fold in most of the cases,
the folding is indirect - we fold `x u% y` to `x & (y-1)` (iff `y` is power-of-two),
or first turn `x s% -y` to `x u% y`; that does handle most of the cases.
But we can't turn `x s% INT_MIN` to `x u% -INT_MIN`,
and thus we end up being stuck with `(x s% INT_MIN) == 0`.

There is no such restriction for the more general fold:
https://rise4fun.com/Alive/IIeS

To be noted, the fold does not enforce that `y` is a constant,
so it may indeed increase instruction count.
This is consistent with what `x u% y`->`x & (y-1)` already does.
I think it makes sense, it's at most one (simple) extra instruction,
while `rem`ainder is really much more un-simple (and likely **very** costly).

Reviewers: spatel, RKSimon, nikic, xbolva00, craig.topper

Reviewed By: RKSimon

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65046

llvm-svn: 367322
llvm/include/llvm/IR/PatternMatch.h
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
llvm/lib/Transforms/InstCombine/InstCombineInternal.h
llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
llvm/test/Transforms/InstCombine/rem.ll