[InstCombine] Teach canEvaluateTruncated to handle arithmetic shift (including those with vector splat shift amount)

Differential Revision: https://reviews.llvm.org/D36784

llvm-svn: 311050

2 files changed, 43 insertions, 0 deletions

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
index 298120e1f69..05728c25db7 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -373,6 +373,23 @@ static bool canEvaluateTruncated(Value *V, Type *Ty, InstCombiner &IC,
     }
     break;
   }
+  case Instruction::AShr: {
+    // If this is a truncate of an arithmetic shr, we can truncate it to a
+    // smaller ashr iff we know that all the bits from the sign bit of the
+    // original type and the sign bit of the truncate type are similar.
+    // TODO: It is enough to check that the bits we would be shifting in are
+    //       similar to sign bit of the truncate type.
+    const APInt *Amt;
+    if (match(I->getOperand(1), m_APInt(Amt))) {
+      uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits();
+      uint32_t BitWidth = Ty->getScalarSizeInBits();
+      if (Amt->getLimitedValue(BitWidth) < BitWidth &&
+          OrigBitWidth - BitWidth <
+              IC.ComputeNumSignBits(I->getOperand(0), 0, CxtI))
+        return canEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI);
+    }
+    break;
+  }
   case Instruction::Trunc:
     // trunc(trunc(x)) -> trunc(x)
     return true;
diff --git a/llvm/test/Transforms/InstCombine/trunc.ll b/llvm/test/Transforms/InstCombine/trunc.ll
index 6d0e0ac5cca..40956d3ff8c 100644
--- a/llvm/test/Transforms/InstCombine/trunc.ll
+++ b/llvm/test/Transforms/InstCombine/trunc.ll
@@ -89,6 +89,32 @@ define i32 @test6(i64 %A) {
   ret i32 %D
 }
 
+define i32 @trunc_ashr(i32 %X) {
+; CHECK-LABEL: @trunc_ashr(
+; CHECK-NEXT:    [[B:%.*]] = or i32 [[X:%.*]], -2147483648
+; CHECK-NEXT:    [[C:%.*]] = ashr i32 [[B]], 8
+; CHECK-NEXT:    ret i32 [[C]]
+;
+  %A = zext i32 %X to i36
+  %B = or i36 %A, -2147483648 ; 0xF80000000
+  %C = ashr i36 %B, 8
+  %T = trunc i36 %C to i32
+  ret i32  %T
+}
+
+define <2 x i32> @trunc_ashr_vec(<2 x i32> %X) {
+; CHECK-LABEL: @trunc_ashr_vec(
+; CHECK-NEXT:    [[B:%.*]] = or <2 x i32> [[X:%.*]], <i32 -2147483648, i32 -2147483648>
+; CHECK-NEXT:    [[C:%.*]] = ashr <2 x i32> [[B]], <i32 8, i32 8>
+; CHECK-NEXT:    ret <2 x i32> [[C]]
+;
+  %A = zext <2 x i32> %X to <2 x i36>
+  %B = or <2 x i36> %A, <i36 -2147483648, i36 -2147483648> ; 0xF80000000
+  %C = ashr <2 x i36> %B, <i36 8, i36 8>
+  %T = trunc <2 x i36> %C to <2 x i32>
+  ret <2 x i32>  %T
+}
+
 define i92 @test7(i64 %A) {
 ; CHECK-LABEL: @test7(
 ; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 %A, 32