Add per array size allocation entrypoints.

- Update architectures that have fast paths for
  array allocation to use it.
- Will add more fast paths in follow-up CLs.

Test: test-art-target test-art-host.
Change-Id: I138cccd16464a85de22a8ed31c915f876e78fb04

15 files changed, 281 insertions, 87 deletions

diff --git a/compiler/oat_test.cc b/compiler/oat_test.cc
index 34b33a13a3..d5842a8c9d 100644
--- a/compiler/oat_test.cc
+++ b/compiler/oat_test.cc
@@ -487,7 +487,7 @@ TEST_F(OatTest, OatHeaderSizeCheck) {
   EXPECT_EQ(72U, sizeof(OatHeader));
   EXPECT_EQ(4U, sizeof(OatMethodOffsets));
   EXPECT_EQ(20U, sizeof(OatQuickMethodHeader));
-  EXPECT_EQ(157 * static_cast<size_t>(GetInstructionSetPointerSize(kRuntimeISA)),
+  EXPECT_EQ(161 * static_cast<size_t>(GetInstructionSetPointerSize(kRuntimeISA)),
             sizeof(QuickEntryPoints));
 }
 
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 99427f05da..d68aa51b1b 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -1417,4 +1417,22 @@ void CodeGenerator::EmitJitRoots(uint8_t* code,
   EmitJitRootPatches(code, roots_data);
 }
 
+QuickEntrypointEnum CodeGenerator::GetArrayAllocationEntrypoint(Handle<mirror::Class> array_klass) {
+  ScopedObjectAccess soa(Thread::Current());
+  if (array_klass.Get() == nullptr) {
+    // This can only happen for non-primitive arrays, as primitive arrays can always
+    // be resolved.
+    return kQuickAllocArrayResolved32;
+  }
+
+  switch (array_klass->GetComponentSize()) {
+    case 1: return kQuickAllocArrayResolved8;
+    case 2: return kQuickAllocArrayResolved16;
+    case 4: return kQuickAllocArrayResolved32;
+    case 8: return kQuickAllocArrayResolved64;
+  }
+  LOG(FATAL) << "Unreachable";
+  return kQuickAllocArrayResolved;
+}
+
 }  // namespace art
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 2d129aff22..b912672792 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -573,6 +573,8 @@ class CodeGenerator : public DeletableArenaObject<kArenaAllocCodeGenerator> {
   uint32_t GetReferenceSlowFlagOffset() const;
   uint32_t GetReferenceDisableFlagOffset() const;
 
+  static QuickEntrypointEnum GetArrayAllocationEntrypoint(Handle<mirror::Class> array_klass);
+
  protected:
   // Patch info used for recording locations of required linker patches and their targets,
   // i.e. target method, string, type or code identified by their dex file and index.
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 9762ee81b1..1e89ba590c 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -4762,7 +4762,9 @@ void LocationsBuilderARM64::VisitNewArray(HNewArray* instruction) {
 void InstructionCodeGeneratorARM64::VisitNewArray(HNewArray* instruction) {
   // Note: if heap poisoning is enabled, the entry point takes cares
   // of poisoning the reference.
-  codegen_->InvokeRuntime(kQuickAllocArrayResolved, instruction, instruction->GetDexPc());
+  QuickEntrypointEnum entrypoint =
+      CodeGenerator::GetArrayAllocationEntrypoint(instruction->GetLoadClass()->GetClass());
+  codegen_->InvokeRuntime(entrypoint, instruction, instruction->GetDexPc());
   CheckEntrypointTypes<kQuickAllocArrayResolved, void*, mirror::Class*, int32_t>();
 }
 
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index c4caf4bf9d..abd8246325 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -4096,7 +4096,9 @@ void LocationsBuilderX86_64::VisitNewArray(HNewArray* instruction) {
 void InstructionCodeGeneratorX86_64::VisitNewArray(HNewArray* instruction) {
   // Note: if heap poisoning is enabled, the entry point takes cares
   // of poisoning the reference.
-  codegen_->InvokeRuntime(kQuickAllocArrayResolved, instruction, instruction->GetDexPc());
+  QuickEntrypointEnum entrypoint =
+      CodeGenerator::GetArrayAllocationEntrypoint(instruction->GetLoadClass()->GetClass());
+  codegen_->InvokeRuntime(entrypoint, instruction, instruction->GetDexPc());
   CheckEntrypointTypes<kQuickAllocArrayResolved, void*, mirror::Class*, int32_t>();
   DCHECK(!codegen_->IsLeafMethod());
 }
diff --git a/compiler/utils/assembler_thumb_test_expected.cc.inc b/compiler/utils/assembler_thumb_test_expected.cc.inc
index f132e2737d..071cd575e1 100644
--- a/compiler/utils/assembler_thumb_test_expected.cc.inc
+++ b/compiler/utils/assembler_thumb_test_expected.cc.inc
@@ -5610,7 +5610,7 @@ const char* const VixlJniHelpersResults[] = {
   " 214:	ecbd 8a10 	vpop	{s16-s31}\n",
   " 218:	e8bd 8de0 	ldmia.w	sp!, {r5, r6, r7, r8, sl, fp, pc}\n",
   " 21c:	4660      	mov	r0, ip\n",
-  " 21e:	f8d9 c2a4 	ldr.w	ip, [r9, #676]	; 0x2a4\n",
+  " 21e:	f8d9 c2b4 	ldr.w	ip, [r9, #692]	; 0x2b4\n",
   " 222:	47e0      	blx	ip\n",
   nullptr
 };
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index 00518e16be..3bd6fe275b 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -1646,6 +1646,10 @@ GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_REGION_TLAB_ALLOCATORS
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB)
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
@@ -1731,29 +1735,7 @@ ENTRY art_quick_alloc_object_resolved_rosalloc
     RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 END art_quick_alloc_object_resolved_rosalloc
 
-
-// The common fast path code for art_quick_alloc_array_region_tlab.
-.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
-    // Array classes are never finalizable or uninitialized, no need to check.
-    ldr    \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type
-    UNPOISON_HEAP_REF \wTemp0
-    ldr    \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET]
-    lsr    \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16
-                                                              // bits.
-                                                              // xCount is holding a 32 bit value,
-                                                              // it can not overflow.
-    lsl    \xTemp1, \xCount, \xTemp0                          // Calculate data size
-    // Add array data offset and alignment.
-    add    \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
-#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4
-#error Long array data offset must be 4 greater than int array data offset.
-#endif
-
-    add    \xTemp0, \xTemp0, #1                               // Add 4 to the length only if the
-                                                              // component size shift is 3
-                                                              // (for 64 bit alignment).
-    and    \xTemp0, \xTemp0, #4
-    add    \xTemp1, \xTemp1, \xTemp0
+.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
     and    \xTemp1, \xTemp1, #OBJECT_ALIGNMENT_MASK_TOGGLED64 // Apply alignemnt mask
                                                               // (addr + 7) & ~7. The mask must
                                                               // be 64 bits to keep high bits in
@@ -1868,8 +1850,7 @@ GENERATE_ALLOC_OBJECT_RESOLVED_REGION_TLAB art_quick_alloc_object_initialized_re
 
 // TODO: We could use this macro for the normal tlab allocator too.
 
-// The common code for art_quick_alloc_array_*region_tlab
-.macro GENERATE_ALLOC_ARRAY_REGION_TLAB name, entrypoint, fast_path
+.macro GENERATE_ALLOC_ARRAY_REGION_TLAB name, entrypoint, size_setup
 ENTRY \name
     // Fast path array allocation for region tlab allocation.
     // x0: mirror::Class* type
@@ -1880,7 +1861,8 @@ ENTRY \name
     ret                                                       // Return -1.
 #endif
     mov    x3, x0
-    \fast_path .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6
+    \size_setup x3, w3, x1, w1, x4, w4, x5, w5, x6, w6
+    ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6
 .Lslow_path\name:
     // x0: mirror::Class* klass
     // x1: int32_t component_count
@@ -1893,7 +1875,60 @@ ENTRY \name
 END \name
 .endm
 
-GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED
+.macro COMPUTE_ARRAY_SIZE_UNKNOWN xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
+    // Array classes are never finalizable or uninitialized, no need to check.
+    ldr    \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type
+    UNPOISON_HEAP_REF \wTemp0
+    ldr    \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET]
+    lsr    \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16
+                                                              // bits.
+                                                              // xCount is holding a 32 bit value,
+                                                              // it can not overflow.
+    lsl    \xTemp1, \xCount, \xTemp0                          // Calculate data size
+    // Add array data offset and alignment.
+    add    \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
+#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4
+#error Long array data offset must be 4 greater than int array data offset.
+#endif
+
+    add    \xTemp0, \xTemp0, #1                               // Add 4 to the length only if the
+                                                              // component size shift is 3
+                                                              // (for 64 bit alignment).
+    and    \xTemp0, \xTemp0, #4
+    add    \xTemp1, \xTemp1, \xTemp0
+.endm
+
+.macro COMPUTE_ARRAY_SIZE_8 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
+    // Add array data offset and alignment.
+    add    \xTemp1, \xCount, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
+.endm
+
+.macro COMPUTE_ARRAY_SIZE_16 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
+    lsl    \xTemp1, \xCount, #1
+    // Add array data offset and alignment.
+    add    \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
+.endm
+
+.macro COMPUTE_ARRAY_SIZE_32 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
+    lsl    \xTemp1, \xCount, #2
+    // Add array data offset and alignment.
+    add    \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
+.endm
+
+.macro COMPUTE_ARRAY_SIZE_64 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2
+    lsl    \xTemp1, \xCount, #3
+    // Add array data offset and alignment.
+    // Add 4 to the size for 64 bit alignment.
+    add    \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK + 4)
+.endm
+
+# TODO(ngeoffray): art_quick_alloc_array_resolved_region_tlab is not used for arm64, remove
+# the entrypoint once all backends have been updated to use the size variants.
+GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN
+GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved8_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_8
+GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved16_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_16
+GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved32_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_32
+GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved64_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_64
 
     /*
      * Called by managed code when the thread has been asked to suspend.
diff --git a/runtime/arch/quick_alloc_entrypoints.S b/runtime/arch/quick_alloc_entrypoints.S
index e79dc6029b..9204d85279 100644
--- a/runtime/arch/quick_alloc_entrypoints.S
+++ b/runtime/arch/quick_alloc_entrypoints.S
@@ -30,6 +30,11 @@ FOUR_ARG_DOWNCALL art_quick_alloc_string_from_bytes\c_suffix, artAllocStringFrom
 THREE_ARG_DOWNCALL art_quick_alloc_string_from_chars\c_suffix, artAllocStringFromCharsFromCode\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 // Called by managed code to allocate a string from string
 ONE_ARG_DOWNCALL art_quick_alloc_string_from_string\c_suffix, artAllocStringFromStringFromCode\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+
+TWO_ARG_DOWNCALL art_quick_alloc_array_resolved8\c_suffix, artAllocArrayFromCodeResolved\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+TWO_ARG_DOWNCALL art_quick_alloc_array_resolved16\c_suffix, artAllocArrayFromCodeResolved\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+TWO_ARG_DOWNCALL art_quick_alloc_array_resolved32\c_suffix, artAllocArrayFromCodeResolved\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+TWO_ARG_DOWNCALL art_quick_alloc_array_resolved64\c_suffix, artAllocArrayFromCodeResolved\cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 .endm
 
 .macro GENERATE_ALL_ALLOC_ENTRYPOINTS
@@ -56,14 +61,22 @@ GENERATE_ALLOC_ENTRYPOINTS _region_tlab_instrumented, RegionTLABInstrumented
   ONE_ARG_DOWNCALL art_quick_alloc_object_initialized ## c_suffix, artAllocObjectFromCodeInitialized ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 #define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(c_suffix, cxx_suffix) \
   ONE_ARG_DOWNCALL art_quick_alloc_object_with_checks ## c_suffix, artAllocObjectFromCodeWithChecks ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
-#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(c_suffix, cxx_suffix) \
-  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 #define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(c_suffix, cxx_suffix) \
   FOUR_ARG_DOWNCALL art_quick_alloc_string_from_bytes ## c_suffix, artAllocStringFromBytesFromCode ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 #define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(c_suffix, cxx_suffix) \
   THREE_ARG_DOWNCALL art_quick_alloc_string_from_chars ## c_suffix, artAllocStringFromCharsFromCode ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 #define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(c_suffix, cxx_suffix) \
   ONE_ARG_DOWNCALL art_quick_alloc_string_from_string ## c_suffix, artAllocStringFromStringFromCode ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(c_suffix, cxx_suffix) \
+  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(c_suffix, cxx_suffix) \
+  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved8 ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(c_suffix, cxx_suffix) \
+  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved16 ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(c_suffix, cxx_suffix) \
+  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved32 ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
+#define GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(c_suffix, cxx_suffix) \
+  TWO_ARG_DOWNCALL art_quick_alloc_array_resolved64 ## c_suffix, artAllocArrayFromCodeResolved ## cxx_suffix, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
 
 .macro GENERATE_ALLOC_ENTRYPOINTS_FOR_EACH_ALLOCATOR
 GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_REGION_TLAB_ALLOCATORS
@@ -76,6 +89,10 @@ GENERATE_ALLOC_ENTRYPOINTS_FOR_REGION_TLAB_ALLOCATOR
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
@@ -87,6 +104,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab, TLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab, TLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab, TLAB)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab, TLAB)
@@ -102,6 +123,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_dlmalloc, DlMalloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_dlmalloc, DlMalloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_dlmalloc, DlMalloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_dlmalloc, DlMalloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_dlmalloc, DlMalloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_dlmalloc, DlMalloc)
@@ -110,6 +135,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_dlmalloc_instrumented, DlMallo
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_dlmalloc_instrumented, DlMallocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_dlmalloc_instrumented, DlMallocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_dlmalloc_instrumented, DlMallocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_dlmalloc_instrumented, DlMallocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_dlmalloc_instrumented, DlMallocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_dlmalloc_instrumented, DlMallocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_dlmalloc_instrumented, DlMallocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_dlmalloc_instrumented, DlMallocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_dlmalloc_instrumented, DlMallocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_dlmalloc_instrumented, DlMallocInstrumented)
@@ -119,6 +148,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_dlmalloc_instrumented, DlMa
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_rosalloc, RosAlloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_rosalloc, RosAlloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_rosalloc, RosAlloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_rosalloc, RosAlloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_rosalloc, RosAlloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_rosalloc, RosAlloc)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_rosalloc, RosAlloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_rosalloc, RosAlloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_rosalloc, RosAlloc)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_rosalloc, RosAlloc)
@@ -127,6 +160,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_rosalloc_instrumented, RosAllo
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_rosalloc_instrumented, RosAllocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_rosalloc_instrumented, RosAllocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_rosalloc_instrumented, RosAllocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_rosalloc_instrumented, RosAllocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_rosalloc_instrumented, RosAllocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_rosalloc_instrumented, RosAllocInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_rosalloc_instrumented, RosAllocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_rosalloc_instrumented, RosAllocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_rosalloc_instrumented, RosAllocInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_rosalloc_instrumented, RosAllocInstrumented)
@@ -135,6 +172,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_bump_pointer, BumpPointer)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_bump_pointer, BumpPointer)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_bump_pointer, BumpPointer)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_bump_pointer, BumpPointer)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_bump_pointer, BumpPointer)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_bump_pointer, BumpPointer)
@@ -143,6 +184,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_bump_pointer_instrumented, Bum
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_bump_pointer_instrumented, BumpPointerInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_bump_pointer_instrumented, BumpPointerInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_bump_pointer_instrumented, BumpPointerInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_bump_pointer_instrumented, BumpPointerInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_bump_pointer_instrumented, BumpPointerInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_bump_pointer_instrumented, BumpPointerInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_bump_pointer_instrumented, BumpPointerInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_bump_pointer_instrumented, BumpPointerInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_bump_pointer_instrumented, BumpPointerInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_bump_pointer_instrumented, BumpPointerInstrumented)
@@ -151,6 +196,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_tlab_instrumented, TLABInstrum
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab_instrumented, TLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab_instrumented, TLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab_instrumented, TLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab_instrumented, TLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab_instrumented, TLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab_instrumented, TLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab_instrumented, TLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab_instrumented, TLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab_instrumented, TLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab_instrumented, TLABInstrumented)
@@ -159,6 +208,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region, Region)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region, Region)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region, Region)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region, Region)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region, Region)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region, Region)
@@ -167,6 +220,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_instrumented, RegionIns
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_instrumented, RegionInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_instrumented, RegionInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_instrumented, RegionInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_instrumented, RegionInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_instrumented, RegionInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_instrumented, RegionInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_instrumented, RegionInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_instrumented, RegionInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_instrumented, RegionInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_instrumented, RegionInstrumented)
@@ -175,6 +232,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab_instrumented, Regi
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab_instrumented, RegionTLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab_instrumented, RegionTLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab_instrumented, RegionTLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab_instrumented, RegionTLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab_instrumented, RegionTLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab_instrumented, RegionTLABInstrumented)
+GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab_instrumented, RegionTLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab_instrumented, RegionTLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab_instrumented, RegionTLABInstrumented)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab_instrumented, RegionTLABInstrumented)
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index 46bee399f4..f3d4256ea3 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -987,6 +987,10 @@ GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_TLAB_ALLOCATORS
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB)
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
@@ -995,6 +999,10 @@ GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab, TLAB)
 // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab, TLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab, TLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab, TLAB)
+// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab, TLAB)
 GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab, TLAB)
@@ -1112,26 +1120,11 @@ MACRO1(ALLOC_OBJECT_INITIALIZED_TLAB_FAST_PATH, slowPathLabel)
 END_MACRO
 
 // The fast path code for art_quick_alloc_array_region_tlab.
-// Inputs: RDI: the class, RSI: int32_t component_count
-// Free temps: RCX, RDX, R8, R9
+// Inputs: RDI: the class, RSI: int32_t component_count, R9: total_size
+// Free temps: RCX, RDX, R8
 // Output: RAX: return value.
-MACRO1(ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED, slowPathLabel)
-    movl MIRROR_CLASS_COMPONENT_TYPE_OFFSET(%rdi), %ecx        // Load component type.
-    UNPOISON_HEAP_REF ecx
-    movl MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET(%rcx), %ecx // Load primitive type.
-    shrq LITERAL(PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT), %rcx        // Get component size shift.
-    movq %rsi, %r9
-    salq %cl, %r9                                              // Calculate array count shifted.
-    // Add array header + alignment rounding.
-    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK), %r9
-    // Add 4 extra bytes if we are doing a long array.
-    addq LITERAL(1), %rcx
-    andq LITERAL(4), %rcx
-    addq %rcx, %r9
+MACRO1(ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE, slowPathLabel)
     movq %gs:THREAD_SELF_OFFSET, %rcx                          // rcx = thread
-#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4
-#error Long array data offset must be 4 greater than int array data offset.
-#endif
     // Mask out the unaligned part to make sure we are 8 byte aligned.
     andq LITERAL(OBJECT_ALIGNMENT_MASK_TOGGLED64), %r9
     movq THREAD_LOCAL_POS_OFFSET(%rcx), %rax
@@ -1149,7 +1142,6 @@ MACRO1(ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED, slowPathLabel)
     ret                                                        // Fast path succeeded.
 END_MACRO
 
-
 // The common slow path code for art_quick_alloc_object_{resolved, initialized}_tlab
 // and art_quick_alloc_object_{resolved, initialized}_region_tlab.
 MACRO1(ALLOC_OBJECT_TLAB_SLOW_PATH, cxx_name)
@@ -1161,16 +1153,6 @@ MACRO1(ALLOC_OBJECT_TLAB_SLOW_PATH, cxx_name)
     RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER                // return or deliver exception
 END_MACRO
 
-// The slow path code for art_quick_alloc_array_region_tlab.
-MACRO1(ALLOC_ARRAY_TLAB_SLOW_PATH, cxx_name)
-    SETUP_SAVE_REFS_ONLY_FRAME                                 // save ref containing registers for GC
-    // Outgoing argument set up
-    movq %gs:THREAD_SELF_OFFSET, %rdx                          // pass Thread::Current()
-    call CALLVAR(cxx_name)                                     // cxx_name(arg0, arg1, Thread*)
-    RESTORE_SAVE_REFS_ONLY_FRAME                               // restore frame up to return address
-    RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER                    // return or deliver exception
-END_MACRO
-
 // A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_tlab, TLAB). May be
 // called with CC if the GC is not active.
 DEFINE_FUNCTION art_quick_alloc_object_resolved_tlab
@@ -1191,25 +1173,92 @@ DEFINE_FUNCTION art_quick_alloc_object_initialized_tlab
     ALLOC_OBJECT_TLAB_SLOW_PATH artAllocObjectFromCodeInitializedTLAB
 END_FUNCTION art_quick_alloc_object_initialized_tlab
 
-// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB).
-DEFINE_FUNCTION art_quick_alloc_array_resolved_tlab
+MACRO0(COMPUTE_ARRAY_SIZE_UNKNOWN)
+    movl MIRROR_CLASS_COMPONENT_TYPE_OFFSET(%rdi), %ecx        // Load component type.
+    UNPOISON_HEAP_REF ecx
+    movl MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET(%rcx), %ecx // Load primitive type.
+    shrq LITERAL(PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT), %rcx        // Get component size shift.
+    movq %rsi, %r9
+    salq %cl, %r9                                              // Calculate array count shifted.
+    // Add array header + alignment rounding.
+    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK), %r9
+    // Add 4 extra bytes if we are doing a long array.
+    addq LITERAL(1), %rcx
+    andq LITERAL(4), %rcx
+#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4
+#error Long array data offset must be 4 greater than int array data offset.
+#endif
+    addq %rcx, %r9
+END_MACRO
+
+MACRO0(COMPUTE_ARRAY_SIZE_8)
     // RDI: mirror::Class* klass, RSI: int32_t component_count
     // RDX, RCX, R8, R9: free. RAX: return val.
-    ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED .Lart_quick_alloc_array_resolved_tlab_slow_path
-.Lart_quick_alloc_array_resolved_tlab_slow_path:
-    ALLOC_ARRAY_TLAB_SLOW_PATH artAllocArrayFromCodeResolvedTLAB
-END_FUNCTION art_quick_alloc_array_resolved_tlab
+    movq %rsi, %r9
+    // Add array header + alignment rounding.
+    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK), %r9
+END_MACRO
 
-// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB).
-DEFINE_FUNCTION art_quick_alloc_array_resolved_region_tlab
-    // Fast path region tlab allocation.
+MACRO0(COMPUTE_ARRAY_SIZE_16)
     // RDI: mirror::Class* klass, RSI: int32_t component_count
-    // RCX, RDX, R8, R9: free. RAX: return val.
-    ASSERT_USE_READ_BARRIER
-    ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED .Lart_quick_alloc_array_resolved_region_tlab_slow_path
-.Lart_quick_alloc_array_resolved_region_tlab_slow_path:
-    ALLOC_ARRAY_TLAB_SLOW_PATH artAllocArrayFromCodeResolvedRegionTLAB
-END_FUNCTION art_quick_alloc_array_resolved_region_tlab
+    // RDX, RCX, R8, R9: free. RAX: return val.
+    movq %rsi, %r9
+    salq LITERAL(1), %r9
+    // Add array header + alignment rounding.
+    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK), %r9
+END_MACRO
+
+MACRO0(COMPUTE_ARRAY_SIZE_32)
+    // RDI: mirror::Class* klass, RSI: int32_t component_count
+    // RDX, RCX, R8, R9: free. RAX: return val.
+    movq %rsi, %r9
+    salq LITERAL(2), %r9
+    // Add array header + alignment rounding.
+    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK), %r9
+END_MACRO
+
+MACRO0(COMPUTE_ARRAY_SIZE_64)
+    // RDI: mirror::Class* klass, RSI: int32_t component_count
+    // RDX, RCX, R8, R9: free. RAX: return val.
+    movq %rsi, %r9
+    salq LITERAL(3), %r9
+    // Add array header + alignment rounding.
+    // Add 4 extra bytes for array data alignment
+    addq LITERAL(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK + 4), %r9
+END_MACRO
+
+// The slow path code for art_quick_alloc_array_*tlab.
+MACRO1(ALLOC_ARRAY_TLAB_SLOW_PATH, cxx_name)
+END_MACRO
+
+MACRO3(GENERATE_ALLOC_ARRAY_TLAB, c_entrypoint, cxx_name, size_setup)
+    DEFINE_FUNCTION VAR(c_entrypoint)
+    // RDI: mirror::Class* klass, RSI: int32_t component_count
+    // RDX, RCX, R8, R9: free. RAX: return val.
+    CALL_MACRO(size_setup)
+    ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE .Lslow_path\c_entrypoint
+.Lslow_path\c_entrypoint:
+    SETUP_SAVE_REFS_ONLY_FRAME                                 // save ref containing registers for GC
+    // Outgoing argument set up
+    movq %gs:THREAD_SELF_OFFSET, %rdx                          // pass Thread::Current()
+    call CALLVAR(cxx_name)                                     // cxx_name(arg0, arg1, Thread*)
+    RESTORE_SAVE_REFS_ONLY_FRAME                               // restore frame up to return address
+    RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER                    // return or deliver exception
+    END_FUNCTION VAR(c_entrypoint)
+END_MACRO
+
+
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_8
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_16
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_32
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_64
+
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_8
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_16
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_32
+GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_64
 
 // A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB).
 DEFINE_FUNCTION art_quick_alloc_object_resolved_region_tlab
diff --git a/runtime/asm_support.h b/runtime/asm_support.h
index ed83f1c57f..46f2c08663 100644
--- a/runtime/asm_support.h
+++ b/runtime/asm_support.h
@@ -104,7 +104,7 @@ ADD_TEST_EQ(THREAD_LOCAL_OBJECTS_OFFSET,
 
 // Offset of field Thread::tlsPtr_.mterp_current_ibase.
 #define THREAD_CURRENT_IBASE_OFFSET \
-    (THREAD_LOCAL_OBJECTS_OFFSET + __SIZEOF_SIZE_T__ + (1 + 157) * __SIZEOF_POINTER__)
+    (THREAD_LOCAL_OBJECTS_OFFSET + __SIZEOF_SIZE_T__ + (1 + 161) * __SIZEOF_POINTER__)
 ADD_TEST_EQ(THREAD_CURRENT_IBASE_OFFSET,
             art::Thread::MterpCurrentIBaseOffset<POINTER_SIZE>().Int32Value())
 // Offset of field Thread::tlsPtr_.mterp_default_ibase.
diff --git a/runtime/entrypoints/quick/quick_alloc_entrypoints.cc b/runtime/entrypoints/quick/quick_alloc_entrypoints.cc
index e9f09b2bd9..582f0cff48 100644
--- a/runtime/entrypoints/quick/quick_alloc_entrypoints.cc
+++ b/runtime/entrypoints/quick/quick_alloc_entrypoints.cc
@@ -129,29 +129,34 @@ GENERATE_ENTRYPOINTS_FOR_ALLOCATOR(RegionTLAB, gc::kAllocatorTypeRegionTLAB)
 
 #define GENERATE_ENTRYPOINTS(suffix) \
 extern "C" void* art_quick_alloc_array_resolved##suffix(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved8##suffix(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved16##suffix(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved32##suffix(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved64##suffix(mirror::Class* klass, int32_t); \
 extern "C" void* art_quick_alloc_object_resolved##suffix(mirror::Class* klass); \
 extern "C" void* art_quick_alloc_object_initialized##suffix(mirror::Class* klass); \
 extern "C" void* art_quick_alloc_object_with_checks##suffix(mirror::Class* klass); \
-extern "C" void* art_quick_check_and_alloc_array##suffix(uint32_t, int32_t, ArtMethod* ref); \
-extern "C" void* art_quick_check_and_alloc_array_with_access_check##suffix(uint32_t, int32_t, ArtMethod* ref); \
 extern "C" void* art_quick_alloc_string_from_bytes##suffix(void*, int32_t, int32_t, int32_t); \
 extern "C" void* art_quick_alloc_string_from_chars##suffix(int32_t, int32_t, void*); \
 extern "C" void* art_quick_alloc_string_from_string##suffix(void*); \
-extern "C" void* art_quick_alloc_array##suffix##_instrumented(uint32_t, int32_t, ArtMethod* ref); \
 extern "C" void* art_quick_alloc_array_resolved##suffix##_instrumented(mirror::Class* klass, int32_t); \
-extern "C" void* art_quick_alloc_array_with_access_check##suffix##_instrumented(uint32_t, int32_t, ArtMethod* ref); \
-extern "C" void* art_quick_alloc_object##suffix##_instrumented(uint32_t type_idx, ArtMethod* ref); \
+extern "C" void* art_quick_alloc_array_resolved8##suffix##_instrumented(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved16##suffix##_instrumented(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved32##suffix##_instrumented(mirror::Class* klass, int32_t); \
+extern "C" void* art_quick_alloc_array_resolved64##suffix##_instrumented(mirror::Class* klass, int32_t); \
 extern "C" void* art_quick_alloc_object_resolved##suffix##_instrumented(mirror::Class* klass); \
 extern "C" void* art_quick_alloc_object_initialized##suffix##_instrumented(mirror::Class* klass); \
 extern "C" void* art_quick_alloc_object_with_checks##suffix##_instrumented(mirror::Class* klass); \
-extern "C" void* art_quick_check_and_alloc_array##suffix##_instrumented(uint32_t, int32_t, ArtMethod* ref); \
-extern "C" void* art_quick_check_and_alloc_array_with_access_check##suffix##_instrumented(uint32_t, int32_t, ArtMethod* ref); \
 extern "C" void* art_quick_alloc_string_from_bytes##suffix##_instrumented(void*, int32_t, int32_t, int32_t); \
 extern "C" void* art_quick_alloc_string_from_chars##suffix##_instrumented(int32_t, int32_t, void*); \
 extern "C" void* art_quick_alloc_string_from_string##suffix##_instrumented(void*); \
 void SetQuickAllocEntryPoints##suffix(QuickEntryPoints* qpoints, bool instrumented) { \
   if (instrumented) { \
     qpoints->pAllocArrayResolved = art_quick_alloc_array_resolved##suffix##_instrumented; \
+    qpoints->pAllocArrayResolved8 = art_quick_alloc_array_resolved8##suffix##_instrumented; \
+    qpoints->pAllocArrayResolved16 = art_quick_alloc_array_resolved16##suffix##_instrumented; \
+    qpoints->pAllocArrayResolved32 = art_quick_alloc_array_resolved32##suffix##_instrumented; \
+    qpoints->pAllocArrayResolved64 = art_quick_alloc_array_resolved64##suffix##_instrumented; \
     qpoints->pAllocObjectResolved = art_quick_alloc_object_resolved##suffix##_instrumented; \
     qpoints->pAllocObjectInitialized = art_quick_alloc_object_initialized##suffix##_instrumented; \
     qpoints->pAllocObjectWithChecks = art_quick_alloc_object_with_checks##suffix##_instrumented; \
@@ -160,6 +165,10 @@ void SetQuickAllocEntryPoints##suffix(QuickEntryPoints* qpoints, bool instrument
     qpoints->pAllocStringFromString = art_quick_alloc_string_from_string##suffix##_instrumented; \
   } else { \
     qpoints->pAllocArrayResolved = art_quick_alloc_array_resolved##suffix; \
+    qpoints->pAllocArrayResolved8 = art_quick_alloc_array_resolved8##suffix; \
+    qpoints->pAllocArrayResolved16 = art_quick_alloc_array_resolved16##suffix; \
+    qpoints->pAllocArrayResolved32 = art_quick_alloc_array_resolved32##suffix; \
+    qpoints->pAllocArrayResolved64 = art_quick_alloc_array_resolved64##suffix; \
     qpoints->pAllocObjectResolved = art_quick_alloc_object_resolved##suffix; \
     qpoints->pAllocObjectInitialized = art_quick_alloc_object_initialized##suffix; \
     qpoints->pAllocObjectWithChecks = art_quick_alloc_object_with_checks##suffix; \
diff --git a/runtime/entrypoints/quick/quick_entrypoints_list.h b/runtime/entrypoints/quick/quick_entrypoints_list.h
index 22b0f921c7..e0a2e3cf8f 100644
--- a/runtime/entrypoints/quick/quick_entrypoints_list.h
+++ b/runtime/entrypoints/quick/quick_entrypoints_list.h
@@ -21,6 +21,10 @@
 
 #define QUICK_ENTRYPOINT_LIST(V) \
   V(AllocArrayResolved, void*, mirror::Class*, int32_t) \
+  V(AllocArrayResolved8, void*, mirror::Class*, int32_t) \
+  V(AllocArrayResolved16, void*, mirror::Class*, int32_t) \
+  V(AllocArrayResolved32, void*, mirror::Class*, int32_t) \
+  V(AllocArrayResolved64, void*, mirror::Class*, int32_t) \
   V(AllocObjectResolved, void*, mirror::Class*) \
   V(AllocObjectInitialized, void*, mirror::Class*) \
   V(AllocObjectWithChecks, void*, mirror::Class*) \
diff --git a/runtime/entrypoints_order_test.cc b/runtime/entrypoints_order_test.cc
index 8e84d76e83..d0687ce7b0 100644
--- a/runtime/entrypoints_order_test.cc
+++ b/runtime/entrypoints_order_test.cc
@@ -152,7 +152,15 @@ class EntrypointsOrderTest : public CommonRuntimeTest {
   void CheckQuickEntryPoints() {
     CHECKED(OFFSETOF_MEMBER(QuickEntryPoints, pAllocArrayResolved) == 0,
             QuickEntryPoints_start_with_allocarray_resoved);
-    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved, pAllocObjectResolved,
+    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved, pAllocArrayResolved8,
+                         sizeof(void*));
+    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved8, pAllocArrayResolved16,
+                         sizeof(void*));
+    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved16, pAllocArrayResolved32,
+                         sizeof(void*));
+    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved32, pAllocArrayResolved64,
+                         sizeof(void*));
+    EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocArrayResolved64, pAllocObjectResolved,
                          sizeof(void*));
     EXPECT_OFFSET_DIFFNP(QuickEntryPoints, pAllocObjectResolved, pAllocObjectInitialized,
                          sizeof(void*));
diff --git a/runtime/oat.h b/runtime/oat.h
index 29821a2eea..4a68036e00 100644
--- a/runtime/oat.h
+++ b/runtime/oat.h
@@ -32,7 +32,7 @@ class InstructionSetFeatures;
 class PACKED(4) OatHeader {
  public:
   static constexpr uint8_t kOatMagic[] = { 'o', 'a', 't', '\n' };
-  static constexpr uint8_t kOatVersion[] = { '1', '0', '3', '\0' };  // Native pc change
+  static constexpr uint8_t kOatVersion[] = { '1', '0', '4', '\0' };  // Array allocation entrypoints
 
   static constexpr const char* kImageLocationKey = "image-location";
   static constexpr const char* kDex2OatCmdLineKey = "dex2oat-cmdline";
diff --git a/runtime/thread.cc b/runtime/thread.cc
index d93eab10a0..66a03a6826 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -2663,6 +2663,10 @@ void Thread::DumpThreadOffset(std::ostream& os, uint32_t offset) {
       return; \
     }
   QUICK_ENTRY_POINT_INFO(pAllocArrayResolved)
+  QUICK_ENTRY_POINT_INFO(pAllocArrayResolved8)
+  QUICK_ENTRY_POINT_INFO(pAllocArrayResolved16)
+  QUICK_ENTRY_POINT_INFO(pAllocArrayResolved32)
+  QUICK_ENTRY_POINT_INFO(pAllocArrayResolved64)
   QUICK_ENTRY_POINT_INFO(pAllocObjectResolved)
   QUICK_ENTRY_POINT_INFO(pAllocObjectInitialized)
   QUICK_ENTRY_POINT_INFO(pAllocObjectWithChecks)