[PATCH] D29729: [InstCombine] don't lose nsw/nuw from add by converting to xor

[Sanjay Patel via Phabricator via llvm-commits]

spatel updated this revision to Diff 88097.
spatel added a comment.

Patch updated:
If llvm.assume can't be used to solve this, I don't know of an existing way to do it, so I'm making this an NFC documentation patch for the current behavior.


https://reviews.llvm.org/D29729

Files:
  lib/Transforms/InstCombine/InstCombineAddSub.cpp
  test/Transforms/InstCombine/add.ll
  test/Transforms/InstCombine/icmp-add.ll


Index: test/Transforms/InstCombine/icmp-add.ll
===================================================================
--- test/Transforms/InstCombine/icmp-add.ll
+++ test/Transforms/InstCombine/icmp-add.ll
@@ -229,7 +229,8 @@
   ret i1 %c
 }
 
-; FIXME: InstCombine should not lose wrapping information by changing the add to xor.
+; This is known 'true', but we canonicalize the add to xor and lose that information.
+; Metadata that preserves the nsw-ness of the add could be used to improve this. 
 
 define i1 @slt_zero_add_nsw_signbit(i8 %x) {
 ; CHECK-LABEL: @slt_zero_add_nsw_signbit(
@@ -241,7 +242,8 @@
   ret i1 %z
 }
 
-; FIXME: InstCombine should not lose wrapping information by changing the add to xor.
+; This is known 'true', but we canonicalize the add to xor and lose that information.
+; Metadata that preserves the nuw-ness of the add could be used to improve this. 
 
 define i1 @slt_zero_add_nuw_signbit(i8 %x) {
 ; CHECK-LABEL: @slt_zero_add_nuw_signbit(
Index: test/Transforms/InstCombine/add.ll
===================================================================
--- test/Transforms/InstCombine/add.ll
+++ test/Transforms/InstCombine/add.ll
@@ -265,7 +265,7 @@
   ret i32 %add
 }
 
-; Lose no-wrap info by converting to xor? %x is known non-negative
+; Lose no-wrap info by converting to xor. %x is known non-negative
 ; here, but not after converting to xor.
 
 define i8 @add_nsw_signbit(i8 %x) {
@@ -277,7 +277,7 @@
   ret i8 %y
 }
 
-; Lose no-wrap info by converting to xor? %x is known non-negative
+; Lose no-wrap info by converting to xor. %x is known non-negative
 ; (x < 128 unsigned), but not after converting to xor.
 
 define i8 @add_nuw_signbit(i8 %x) {
Index: lib/Transforms/InstCombine/InstCombineAddSub.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1044,7 +1044,9 @@
 
   const APInt *Val;
   if (match(RHS, m_APInt(Val))) {
-    // X + (signbit) --> X ^ signbit
+    // X + (signbit) --> X ^ signbit. We may drop nsw/nuw here and forgo
+    // optimizations based on that.
+    // TODO: We could add metadata to preserve the wrapping information? 
     if (Val->isSignBit())
       return BinaryOperator::CreateXor(LHS, RHS);
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D29729.88097.patch
Type: text/x-patch
Size: 2304 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170211/bc0494f5/attachment.bin>