[PATCH] D65692: [ValueTracking] When calculating known bits for integer abs, make sure we're looking at a negate and not just any instruction with the nsw flag set.

Craig Topper via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Aug 2 23:36:34 PDT 2019


craig.topper created this revision.
craig.topper added reviewers: RKSimon, spatel.
Herald added a subscriber: hiraditya.
Herald added a project: LLVM.

The matchSelectPattern code can match patterns like (x >= 0) ? x : -x
for absolute value. But it can also match ((x-y) >= 0) ? (x-y) : (y-x).
If the latter form was matched we can only use the nsw flag if its
set on both subtracts.

This match makes sure we're looking at the former case only.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D65692

Files:
  llvm/lib/Analysis/ValueTracking.cpp
  llvm/test/Transforms/InstSimplify/icmp-abs-nabs.ll


Index: llvm/test/Transforms/InstSimplify/icmp-abs-nabs.ll
===================================================================
--- llvm/test/Transforms/InstSimplify/icmp-abs-nabs.ll
+++ llvm/test/Transforms/InstSimplify/icmp-abs-nabs.ll
@@ -404,7 +404,12 @@
 ; We can't fold this to false unless both subs have nsw.
 define i1 @abs_sub_sub_missing_nsw(i32 %x, i32 %y) {
 ; CHECK-LABEL: @abs_sub_sub_missing_nsw(
-; CHECK-NEXT:    ret i1 false
+; CHECK-NEXT:    [[A:%.*]] = sub i32 [[X:%.*]], [[Y:%.*]]
+; CHECK-NEXT:    [[B:%.*]] = sub nsw i32 [[Y]], [[X]]
+; CHECK-NEXT:    [[C:%.*]] = icmp sgt i32 [[A]], -1
+; CHECK-NEXT:    [[D:%.*]] = select i1 [[C]], i32 [[A]], i32 [[B]]
+; CHECK-NEXT:    [[E:%.*]] = icmp slt i32 [[D]], 0
+; CHECK-NEXT:    ret i1 [[E]]
 ;
   %a = sub i32 %x, %y
   %b = sub nsw i32 %y, %x
Index: llvm/lib/Analysis/ValueTracking.cpp
===================================================================
--- llvm/lib/Analysis/ValueTracking.cpp
+++ llvm/lib/Analysis/ValueTracking.cpp
@@ -1095,7 +1095,7 @@
       // RHS from matchSelectPattern returns the negation part of abs pattern.
       // If the negate has an NSW flag we can assume the sign bit of the result
       // will be 0 because that makes abs(INT_MIN) undefined.
-      if (Q.IIQ.hasNoSignedWrap(cast<Instruction>(RHS)))
+      if (match(RHS, m_NSWSub(m_ZeroInt(), m_Specific(LHS))))
         MaxHighZeros = 1;
     }
 
@@ -5624,7 +5624,7 @@
     // then the result of abs(X) is [0..SIGNED_MAX],
     // otherwise it is [0..SIGNED_MIN], as -SIGNED_MIN == SIGNED_MIN.
     Lower = APInt::getNullValue(BitWidth);
-    if (cast<Instruction>(RHS)->hasNoSignedWrap())
+    if (match(RHS, m_NSWSub(m_ZeroInt(), m_Specific(LHS))))
       Upper = APInt::getSignedMaxValue(BitWidth) + 1;
     else
       Upper = APInt::getSignedMinValue(BitWidth) + 1;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D65692.213181.patch
Type: text/x-patch
Size: 1831 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190803/6b6a3787/attachment.bin>


More information about the llvm-commits mailing list