[llvm] a41226b - [ValueTracking] Fix KnownBits conflict for calls (range vs returned) (#84353)

via llvm-commits llvm-commits at lists.llvm.org
Thu Mar 7 12:32:53 PST 2024


Author: Björn Pettersson
Date: 2024-03-07T21:32:49+01:00
New Revision: a41226b05510a6f40d99fc622d78853460dc5599

URL: https://github.com/llvm/llvm-project/commit/a41226b05510a6f40d99fc622d78853460dc5599
DIFF: https://github.com/llvm/llvm-project/commit/a41226b05510a6f40d99fc622d78853460dc5599.diff

LOG: [ValueTracking] Fix KnownBits conflict for calls (range vs returned) (#84353)

If a function only exits for certain input values we can still derive
that an argument is "returned". We can also derive range metadata that
describe the possible value range returned by the function. However, it
turns out that those two analyses can result in conflicting information.

Example:
  declare i16 @foo(i16 returned)
  ...
  %A = call i16 @foo(i16 4095), !range !{i16 32, i16 33}

To avoid "Bits known to be one AND zero?" assertion failures we know
make sure to discard the known bits for this kind of scenario.

Added: 
    

Modified: 
    llvm/lib/Analysis/ValueTracking.cpp
    llvm/unittests/Analysis/ValueTrackingTest.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 52ae9f034e5d34..6d0e79e11eed43 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -1476,6 +1476,12 @@ static void computeKnownBitsFromOperator(const Operator *I,
       if (RV->getType() == I->getType()) {
         computeKnownBits(RV, Known2, Depth + 1, Q);
         Known = Known.unionWith(Known2);
+        // If the function doesn't return properly for all input values
+        // (e.g. unreachable exits) then there might be conflicts between the
+        // argument value and the range metadata. Simply discard the known bits
+        // in case of conflicts.
+        if (Known.hasConflict())
+          Known.resetAll();
       }
     }
     if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {

diff  --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index 9e0abe7a16df98..6c6897d83a256e 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -2359,6 +2359,20 @@ TEST_F(ComputeKnownBitsTest, ComputeKnownBitsFreeze) {
   EXPECT_EQ(Known.One.getZExtValue(), 0u);
 }
 
+TEST_F(ComputeKnownBitsTest, ComputeKnownBitsReturnedRangeConflict) {
+  parseAssembly(
+      "declare i16 @foo(i16 returned)\n"
+      "\n"
+      "define i16 @test() {\n"
+      "  %A = call i16 @foo(i16 4095), !range !{i16 32, i16 33}\n"
+      "  ret i16 %A\n"
+      "}\n");
+  // The call returns 32 according to range metadata, but 4095 according to the
+  // returned arg operand. Given the conflicting information we expect that the
+  // known bits information simply is cleared.
+  expectKnownBits(/*zero*/ 0u, /*one*/ 0u);
+}
+
 TEST_F(ComputeKnownBitsTest, ComputeKnownBitsAddWithRange) {
   parseAssembly("define void @test(ptr %p) {\n"
                 "  %A = load i64, ptr %p, !range !{i64 64, i64 65536}\n"


        


More information about the llvm-commits mailing list