ARM: Improved double to integer truncation....

src/arm/lithium-codegen-arm.h

Index: src/arm/lithium-codegen-arm.h
===================================================================
--- src/arm/lithium-codegen-arm.h	(revision 7092)
+++ src/arm/lithium-codegen-arm.h	(working copy)
@@ -215,6 +215,35 @@
   void DoMathCos(LUnaryMathOperation* instr);
   void DoMathSin(LUnaryMathOperation* instr);
 
+  // Helpers for math operations.
+
+  // Helper for EmitECMATruncate.
+  // Tries to truncate using a vcvt instruction.
+  enum TruncationType {
+    kUnsignedTruncation,
+    kSignedTruncation
+  };
+
+  void TryVcvtTruncation(Register result,
+                         DwVfpRegister double_input,
+                         Register saved_fpscr,
+                         Register current_fpscr,
+                         SwVfpRegister single_scratch,
+                         TruncationType type,
+                         Label* success);
+
+  // Performs a truncating conversion of a floating point number as used by
+  // the JS bitwise operations. See ECMA-262 9.5: ToInt32.
+  // Jumps to the 'done' label if the truncation is successful, and falls
+  // through if it was not.
+  void EmitECMATruncate(Register result,
+                        Register scratch1,
+                        Register scratch2,
+                        DwVfpRegister double_input,
+                        DwVfpRegister double_scratch1,
+                        DwVfpRegister double_scratch2,
+                        Label* done);

[Karl Klose, 2011/03/09 10:37:51]

These functions should perhaps be in the macro assembler, because they are not
specific for lithium code generation and the macro assembler already contains FP
conversion and truncation operations.

[Alexandre, 2011/03/15 08:45:39]

Done.

+
   // Support for recording safepoint and position information.
   void RecordSafepoint(LPointerMap* pointers,
                        Safepoint::Kind kind,