Introduce more compact ReadBarrierMark slow-paths.

Replace entry point ReadBarrierMark with 32
ReadBarrierMarkRegX entry points, using register
number X as input and output (instead of the standard
runtime calling convention) to save two moves in Baker's
read barrier mark slow-path code.

Test: ART host and target (ARM, ARM64) tests.
Bug: 29506760
Bug: 12687968
Change-Id: I73cfb82831cf040b8b018e984163c865cc44ed87
diff --git a/compiler/oat_test.cc b/compiler/oat_test.cc
index 672018b..41b1960 100644
--- a/compiler/oat_test.cc
+++ b/compiler/oat_test.cc
@@ -444,7 +444,7 @@
   EXPECT_EQ(72U, sizeof(OatHeader));
   EXPECT_EQ(4U, sizeof(OatMethodOffsets));
   EXPECT_EQ(20U, sizeof(OatQuickMethodHeader));
-  EXPECT_EQ(133 * GetInstructionSetPointerSize(kRuntimeISA), sizeof(QuickEntryPoints));
+  EXPECT_EQ(164 * GetInstructionSetPointerSize(kRuntimeISA), sizeof(QuickEntryPoints));
 }
 
 TEST_F(OatTest, OatHeaderIsValid) {