[InstCombine] Remove redundant shift folds (NFCI)

[nikic]

These are already handled by canEvaluateShifted/getShiftedValue (one-use only), and also in reassociateShiftAmtsOfTwoSameDirectionShifts (also multi-use), so let's at least get rid of the third implementation...

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Nikita Popov (nikic)

Changes

These are already handled by canEvaluateShifted/getShiftedValue (one-use only), and also in reassociateShiftAmtsOfTwoSameDirectionShifts (also multi-use), so let's at least get rid of the third implementation...

---

Full diff: https://github.com/llvm/llvm-project/pull/90016.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp (-16)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp
index 244f03a1bc2b4c..1cb21a1d81af4b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp
@@ -1120,14 +1120,6 @@ Instruction *InstCombinerImpl::visitShl(BinaryOperator &I) {
       return BinaryOperator::CreateAnd(Trunc, ConstantInt::get(Ty, Mask));
     }
 
-    if (match(Op0, m_Shl(m_Value(X), m_APInt(C1))) && C1->ult(BitWidth)) {
-      unsigned AmtSum = ShAmtC + C1->getZExtValue();
-      // Oversized shifts are simplified to zero in InstSimplify.
-      if (AmtSum < BitWidth)
-        // (X << C1) << C2 --> X << (C1 + C2)
-        return BinaryOperator::CreateShl(X, ConstantInt::get(Ty, AmtSum));
-    }
-
     // If we have an opposite shift by the same amount, we may be able to
     // reorder binops and shifts to eliminate math/logic.
     auto isSuitableBinOpcode = [](Instruction::BinaryOps BinOpcode) {
@@ -1394,14 +1386,6 @@ Instruction *InstCombinerImpl::visitLShr(BinaryOperator &I) {
       }
     }
 
-    // (X >>u C1) >>u C --> X >>u (C1 + C)
-    if (match(Op0, m_LShr(m_Value(X), m_APInt(C1)))) {
-      // Oversized shifts are simplified to zero in InstSimplify.
-      unsigned AmtSum = ShAmtC + C1->getZExtValue();
-      if (AmtSum < BitWidth)
-        return BinaryOperator::CreateLShr(X, ConstantInt::get(Ty, AmtSum));
-    }
-
     Instruction *TruncSrc;
     if (match(Op0, m_OneUse(m_Trunc(m_Instruction(TruncSrc)))) &&
         match(TruncSrc, m_LShr(m_Value(X), m_APInt(C1)))) {

[dtcxzyw]

LGTM. They are unreachable in practice:

https://dtcxzyw.github.io/llvm-opt-benchmark/coverage/home/dtcxzyw/llvm-project/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp.html#L1123

https://dtcxzyw.github.io/llvm-opt-benchmark/coverage/home/dtcxzyw/llvm-project/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp.html#L1398