[llvm] [KnownBits] Implement knownbits lshr/ashr with exact flag (PR #84254)

[Jay Foad via llvm-commits]

================
@@ -565,7 +565,9 @@ void GISelKnownBits::computeKnownBitsImpl(Register R, KnownBits &Known,
     KnownBits ExtKnown = KnownBits::makeConstant(APInt(BitWidth, BitWidth));
     KnownBits ShiftKnown = KnownBits::computeForAddSub(
         /*Add=*/false, /*NSW=*/false, /* NUW=*/false, ExtKnown, WidthKnown);
-    Known = KnownBits::ashr(KnownBits::shl(Known, ShiftKnown), ShiftKnown);
+    Known = KnownBits::ashr(KnownBits::shl(Known, ShiftKnown), ShiftKnown,
+                            /*ShAmtNonZero=*/false,
+                            /*Exact*/ true);
----------------
jayfoad wrote:

> Could you read the Exact flag from the llvm::MachineInstr?

No, this is not related to whether the G_SBFX itself is exact. This is only exact because we're operating on the result of extractBits, which will leave zero bits outside of the masked value.

https://github.com/llvm/llvm-project/pull/84254