ART: Implement loop full unrolling.
Performs whole loop unrolling for small loops with small
trip count to eliminate the loop check overhead, to have
more opportunities for inter-iteration optimizations.
caffeinemark/FloatAtom: 1.2x performance on arm64 Cortex-A57.
Test: 530-checker-peel-unroll.
Test: test-art-host, test-art-target.
Change-Id: Idf3fe3cb611376935d176c60db8c49907222e28a
diff --git a/compiler/optimizing/loop_optimization.h b/compiler/optimizing/loop_optimization.h
index bc47924..644b740 100644
--- a/compiler/optimizing/loop_optimization.h
+++ b/compiler/optimizing/loop_optimization.h
@@ -155,6 +155,12 @@
bool TryPeelingForLoopInvariantExitsElimination(LoopAnalysisInfo* analysis_info,
bool generate_code = true);
+ // Tries to perform whole loop unrolling for a small loop with a small trip count to eliminate
+ // the loop check overhead and to have more opportunities for inter-iteration optimizations.
+ // Returns whether transformation happened. 'generate_code' determines whether the optimization
+ // should be actually applied.
+ bool TryFullUnrolling(LoopAnalysisInfo* analysis_info, bool generate_code = true);
+
// Tries to apply scalar loop peeling and unrolling.
bool TryPeelingAndUnrolling(LoopNode* node);