ARM: Optimize Div/Rem by 2^n for non-negative dividends

When it can be proved that dividends are non-negative or the min integer
if their type is integral, there is no need to generate instructions
correcting the result.

The CL implements this optimization for ARM32/ARM64.

Test: 411-checker-hdiv-hrem-pow2
Test: test.py --host --optimizing --jit --gtest --interpreter
Test: test.py -target --optimizing --jit --interpreter
Test: run-gtests.sh

Change-Id: I11211a42918b5801fce8e78f305e69549739c23c

1 files changed, 11 insertions, 0 deletions

diff --git a/compiler/optimizing/code_generator_utils.cc b/compiler/optimizing/code_generator_utils.cc
index dd47a1fc6c..9da520161f 100644
--- a/compiler/optimizing/code_generator_utils.cc
+++ b/compiler/optimizing/code_generator_utils.cc
@@ -100,4 +100,15 @@ bool IsBooleanValueOrMaterializedCondition(HInstruction* cond_input) {
   return !cond_input->IsCondition() || !cond_input->IsEmittedAtUseSite();
 }
 
+bool HasNonNegativeResultOrMinInt(HInstruction* instruction) {
+  // 1. The instruction itself has always a non-negative result or the min value of
+  //    the integral type if the instruction has the integral type.
+  // 2. TODO: The instruction can be an expression which uses an induction variable.
+  //    Induction variable often start from 0 and are only increased. Such an
+  //    expression might be always non-negative.
+  return instruction->IsAbs() ||
+         IsInt64Value(instruction, DataType::MinValueOfIntegralType(instruction->GetType())) ||
+         IsGEZero(instruction);
+}
+
 }  // namespace art