Set allocation bit on allocation fast path (#103)

Author: YiluoWei

.github/scripts/ci-test-global-alloc-bit.sh

@@ -0,0 +1,56 @@
+set -xe
+
+unset JAVA_TOOL_OPTIONS
+
+# To OpenJDK folder
+root_dir=$(dirname "$0")/../../
+cd $root_dir/repos/openjdk
+
+# Choose build: use slowdebug for shorter build time (32m user time for release vs. 20m user time for slowdebug)
+export DEBUG_LEVEL=fastdebug
+export MMTK_EXTREME_ASSERTIONS=0
+export GLOBAL_ALLOC_BIT=1
+echo $RUSTUP_TOOLCHAIN
+
+# Build
+sh configure --disable-warnings-as-errors --with-debug-level=$DEBUG_LEVEL
+make CONF=linux-x86_64-normal-server-$DEBUG_LEVEL THIRD_PARTY_HEAP=$PWD/../../openjdk
+
+# --- SemiSpace ---
+export MMTK_PLAN=SemiSpace
+
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar antlr
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar fop
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar luindex
+
+
+# --- Immix ---
+export MMTK_PLAN=Immix
+
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar antlr
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar fop
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar luindex
+
+# --- GenCopy ---
+export MMTK_PLAN=GenCopy
+
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar antlr
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar fop
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar luindex
+
+# --- NoGC ---
+
+export MMTK_PLAN=NoGC
+
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms1G -Xmx1G -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar antlr
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms1G -Xmx1G -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar fop
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms1G -Xmx1G -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar luindex
+
+
+# --- MarkSweep ---
+export MMTK_PLAN=MarkSweep
+
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar antlr
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar fop
+build/linux-x86_64-normal-server-$DEBUG_LEVEL/jdk/bin/java -XX:+UseThirdPartyHeap -server -XX:MetaspaceSize=100M -Xms500M -Xmx500M -XX:-TieredCompilation -Xcomp -jar benchmarks/dacapo-2006-10-MR2.jar luindex
+

.github/scripts/ci-test.sh

@@ -5,3 +5,5 @@ cd $cur
 ./ci-test-normal.sh
 cd $cur
 ./ci-test-assertions.sh
+cd $cur
+./ci-test-global-alloc-bit.sh

.github/workflows/ci.yml

@@ -23,6 +23,8 @@ jobs:
       # Run the tests
       - name: Dacapo Tests
         run: ./.github/scripts/ci-test-normal.sh
+      - name: DaCapo Tests for Global Allocation Bit
+        run: ./.github/scripts/ci-test-global-alloc-bit.sh
 
       # Style check
       - name: Style checks

mmtk/Cargo.toml

@@ -20,7 +20,7 @@ lazy_static = "1.1"
 # - change branch
 # - change repo name
 # But other changes including adding/removing whitespaces in commented lines may break the CI.
-mmtk = { git = "https://github.com/mmtk/mmtk-core.git", rev = "53a46c559c6c48080730625b4749f43cf52841ad" }
+mmtk = { git = "https://github.com/mmtk/mmtk-core.git", rev = "e581539f25d050bd34a541e783351ff5f44095aa" }
 # Uncomment the following to build locally
 # mmtk = { path = "../repos/mmtk-core" }
 
@@ -29,6 +29,7 @@ default = []
 mmtk_extreme_assertions = ["mmtk/extreme_assertions"]
 nogc_lock_free = ["mmtk/nogc_lock_free"]
 nogc_no_zeroing = ["mmtk/nogc_no_zeroing"]
+global_alloc_bit = ["mmtk/global_alloc_bit"]
 
 # This compile time constant places the mark bit in the header of the object instead of on the side.
 mark_bit_in_header = []

mmtk/src/lib.rs

@@ -85,10 +85,18 @@ pub struct OpenJDK_Upcalls {
 
 pub static mut UPCALLS: *const OpenJDK_Upcalls = null_mut();
 
+#[no_mangle]
+pub static GLOBAL_SIDE_METADATA_BASE_ADDRESS: uintptr_t =
+    crate::mmtk::util::metadata::side_metadata::GLOBAL_SIDE_METADATA_BASE_ADDRESS.as_usize();
+
 #[no_mangle]
 pub static GLOBAL_SIDE_METADATA_VM_BASE_ADDRESS: uintptr_t =
     crate::mmtk::util::metadata::side_metadata::GLOBAL_SIDE_METADATA_VM_BASE_ADDRESS.as_usize();
 
+#[no_mangle]
+pub static GLOBAL_ALLOC_BIT_ADDRESS: uintptr_t =
+    crate::mmtk::util::metadata::side_metadata::ALLOC_SIDE_METADATA_ADDR.as_usize();
+
 #[derive(Default)]
 pub struct OpenJDK;
 

openjdk/CompileThirdPartyHeap.gmk

@@ -6,6 +6,14 @@ ifdef MMTK_PLAN
   GC_FEATURES=--features $(MMTK_PLAN)
 endif
 
+ifeq ($(GLOBAL_ALLOC_BIT), 1)
+  ifndef GC_FEATURES
+    GC_FEATURES=--features global_alloc_bit
+  else
+    GC_FEATURES:=$(strip $(GC_FEATURES))",global_alloc_bit"
+  endif
+endif
+
 ifeq ($(MARK_IN_HEADER), 1)
   ifndef GC_FEATURES
     GC_FEATURES=--features mark_bit_in_header
@@ -44,5 +52,8 @@ FORCE:
 TARGETS += $(LIB_MMTK)
 JVM_SRC_DIRS += $(MMTK_CPP_ROOT) $(MMTK_CPP_ROOT)/barriers
 JVM_CFLAGS += -std=c++11 -DINCLUDE_THIRD_PARTY_HEAP -DTHIRD_PARTY_HEAP_SRC=$(MMTK_CPP_ROOT)
+ifeq ($(GLOBAL_ALLOC_BIT), 1)
+  JVM_CFLAGS += -DMMTK_ENABLE_GLOBAL_ALLOC_BIT
+endif
 
 $(BUILD_LIBJVM): $(LIB_MMTK)

openjdk/mmtk.h

@@ -12,7 +12,9 @@ extern "C" {
 typedef void* MMTk_Mutator;
 typedef void* MMTk_TraceLocal;
 
+extern const uintptr_t GLOBAL_SIDE_METADATA_BASE_ADDRESS;
 extern const uintptr_t GLOBAL_SIDE_METADATA_VM_BASE_ADDRESS;
+extern const uintptr_t GLOBAL_ALLOC_BIT_ADDRESS;
 
 /**
  * Allocation

openjdk/mmtkBarrierSet.hpp

@@ -38,6 +38,8 @@
 
 #define MMTK_ENABLE_ALLOCATION_FASTPATH true
 
+const intptr_t ALLOC_BIT_BASE_ADDRESS = GLOBAL_ALLOC_BIT_ADDRESS;
+
 class MMTkBarrierSetRuntime: public CHeapObj<mtGC> {
 public:
   virtual void record_modified_node(oop object) {};

openjdk/mmtkBarrierSetAssembler_x86.cpp

@@ -110,6 +110,39 @@ void MMTkBarrierSetAssembler::eden_allocate(MacroAssembler* masm, Register threa
     __ jcc(Assembler::above, slow_case);
     // lab.cursor = end
     __ movptr(cursor, end);
+
+#ifdef MMTK_ENABLE_GLOBAL_ALLOC_BIT
+    Register tmp3 = rscratch2;
+    Register tmp2 = rscratch1;
+    Register tmp4 = t1;
+    assert_different_registers(obj, tmp2, tmp3, tmp4, rcx);
+
+    // tmp2 = load-byte (SIDE_METADATA_BASE_ADDRESS + (obj >> 6));
+    __ movptr(tmp3, obj);
+    __ shrptr(tmp3, 6);
+    __ movptr(tmp2, ALLOC_BIT_BASE_ADDRESS);
+    __ movb(tmp2, Address(tmp2, tmp3));
+    // tmp3 = 1 << ((obj >> 3) & 7)
+    //   1. tmp4 = (obj >> 3) & 7
+    __ movptr(tmp4, obj);
+    __ shrptr(tmp4, 3);
+    __ andptr(tmp4, 7);
+    //   2. tmp4 = 1 << tmp4
+    __ movptr(tmp3, rcx);
+    __ movl(rcx, tmp4);
+    __ movptr(tmp4, 1);
+    __ shlptr(tmp4);
+    __ movptr(rcx, tmp3);
+    // tmp2 = tmp2 | tmp4
+    __ orptr(tmp2, tmp4);
+
+    // store-byte tmp2 (SIDE_METADATA_BASE_ADDRESS + (obj >> 6))
+    __ movptr(tmp3, obj);
+    __ shrptr(tmp3, 6);
+    __ movptr(tmp4, ALLOC_BIT_BASE_ADDRESS);
+    __ movb(Address(tmp4, tmp3), tmp2);  
+#endif
+
     // BarrierSetAssembler::incr_allocated_bytes
     if (var_size_in_bytes->is_valid()) {
       __ addq(Address(r15_thread, in_bytes(JavaThread::allocated_bytes_offset())), var_size_in_bytes);

openjdk/mmtkBarrierSetC2.cpp

@@ -277,6 +277,65 @@ void MMTkBarrierSetC2::expand_allocate(PhaseMacroExpand* x,
     fast_oop_ctrl = needgc_false; // No contention, so this is the fast path
     fast_oop_rawmem = store_eden_top;
 
+#ifdef MMTK_ENABLE_GLOBAL_ALLOC_BIT
+    // set the alloc bit:          
+    // intptr_t addr = (intptr_t) (void*) fast_oop;
+    // uint8_t* meta_addr = (uint8_t*) (ALLOC_BIT_BASE_ADDRESS + (addr >> 6));
+    // intptr_t shift = (addr >> 3) & 0b111;
+    // uint8_t byte_val = *meta_addr;
+    // uint8_t new_byte_val = byte_val | (1 << shift);
+    // *meta_addr = new_byte_val;  
+    Node *obj_addr = new CastP2XNode(fast_oop_ctrl, fast_oop);
+    x->transform_later(obj_addr);
+
+    Node *addr_shift = ConINode::make(6);
+    x->transform_later(addr_shift);
+
+    Node *meta_offset = new URShiftLNode(obj_addr, addr_shift);
+    x->transform_later(meta_offset);
+
+    Node *meta_base = ConLNode::make(ALLOC_BIT_BASE_ADDRESS);
+    x->transform_later(meta_base);
+
+    Node *meta_addr = new AddLNode(meta_base, meta_offset);
+    x->transform_later(meta_addr);
+
+    Node *meta_addr_p = new CastX2PNode(meta_addr);
+    x->transform_later(meta_addr_p);
+
+    Node *meta_val = new LoadUBNode(fast_oop_ctrl, fast_oop_rawmem, meta_addr_p, TypePtr::BOTTOM, TypeInt::BYTE, MemNode::unordered);
+    x->transform_later(meta_val);
+
+    Node *meta_val_shift = ConINode::make(3);
+    x->transform_later(meta_val_shift);
+
+    Node *shifted_addr = new URShiftLNode(obj_addr, meta_val_shift);
+    x->transform_later(shifted_addr);
+
+    Node *meta_val_mask = ConLNode::make(0b111);
+    x->transform_later(meta_val_mask);
+
+    Node *shifted_masked_addr = new AndLNode(shifted_addr, meta_val_mask);
+    x->transform_later(shifted_masked_addr);
+
+    Node *const_one =  ConINode::make(1);
+    x->transform_later(const_one);
+
+    Node *shifted_masked_addr_i = new ConvL2INode(shifted_masked_addr);   
+    x->transform_later(shifted_masked_addr_i);
+
+    Node *set_bit = new LShiftINode(const_one, shifted_masked_addr_i);
+    x->transform_later(set_bit);
+
+    Node *new_meta_val = new OrINode(meta_val, set_bit);
+    x->transform_later(new_meta_val);
+
+    Node *set_alloc_bit = new StoreBNode(fast_oop_ctrl, fast_oop_rawmem, meta_addr_p, TypeRawPtr::BOTTOM, new_meta_val, MemNode::unordered);
+    x->transform_later(set_alloc_bit);
+
+    fast_oop_rawmem = set_alloc_bit;
+#endif
+
     InitializeNode* init = alloc->initialization();
     fast_oop_rawmem = x->initialize_object(alloc,
                                            fast_oop_ctrl, fast_oop_rawmem, fast_oop,