#111 initial implementation of gc

CMakeLists.txt

@@ -17,12 +17,13 @@ if (CHAI_BENCH)
     add_subdirectory(bench)
 endif ()
 
+find_package(Python COMPONENTS Interpreter REQUIRED)
 # @todo #15:90min Generate autogen/operations.hpp into build directory.
 set(OPERATIONS_DIR ${CMAKE_SOURCE_DIR}/include/ChaiVM/interpreter/autogen/operations.hpp)
 add_custom_target(gen_operations_header
         COMMENT "Generating ${OPERATIONS_DIR}"
         OUTPUT ${OPERATIONS_DIR}
-        COMMAND python3 ${CMAKE_SOURCE_DIR}/tools/opcode2operation-generator.py ${OPERATIONS_DIR} ${CMAKE_SOURCE_DIR}/tools/resources/instructions.yml
+        COMMAND ${Python_EXECUTABLE} ${CMAKE_SOURCE_DIR}/tools/opcode2operation-generator.py ${OPERATIONS_DIR} ${CMAKE_SOURCE_DIR}/tools/resources/instructions.yml
 )
 
 add_library(chai_include INTERFACE)

ISA.md

@@ -1,71 +1,71 @@
-There is ChaiVM's accumulator(acc) based ISA.
+There is ChaiVM's accumulator(getAcc) based ISA.
 
 Notation: `imm` is raw value, `[imm]` is value from constant pool by number imm.
 
 | Operation | Format | Description |
 |:---------:|:------:|:------------|
 | Nop | N | Does nothing |
-| Ret | N | Returns a value via acc |
+| Ret | N | Returns a value via getAcc |
 | Mov | RR | i64, moves value from ri to rj |
-| Ldia | I | i64, loads constant [i] to acc |
-| Ldra | R | i64, loads the R1 to acc |
-| Star | R | i64, copies val of acc into register ri |
-| Add | R | i64, Adds R1 to acc |
-| Addi | I | i64, Adds [imm] to acc |
-| Sub | R | i64, sub R1 from acc |
-| Subi | I | i64, sub [imm] from acc |
-| Mul | R | i64, mul acc by R1 |
-| Muli | I | i64, mul acc by [imm] |
-| Div | R | i64, divides acc by R1 |
-| Divi | I | i64, divides acc by [imm] |
-| Modi | I | i64, acc = acc % [imm] |
-| Ldiaf | I | f64, loads constant [i] to acc |
-| Addf | R | f64, Adds R1 to acc |
-| Addif | I | f64, Adds [imm] to acc |
-| Subf | R | f64, sub R1 from acc |
-| Subif | I | i64, sub [imm] from acc |
-| Mulf | R | f64, mul acc by R1 |
-| Mulif | I | f64, mul acc by [imm] |
-| Divf | R | f64, divides acc by R1 |
-| Divif | I | f64, divides acc by [imm] |
-| IcPrint | N | Intrinsic. Prints value of acc |
-| IcScani | N | Intrinsic. Scans i64 from stdin into acc |
-| IcScanf | N | Intrinsic. Scans f64 from stdin into acc |
-| IcSqrt | N | f64, Intrinsic. Calculates sqrt of acc. |
-| IcSin | N | f64, Intrinsic. Calculates sin of acc. |
-| IcCos | N | f64, Intrinsic. Calculates cos of acc. |
-| If_icmpeq | RI | i64, if acc == r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
-| If_icmpne | RI | i64, if acc != r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
-| If_icmpgt | RI | i64, if acc > r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
-| If_icmpge | RI | i64, if acc >= r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
-| If_icmplt | RI | i64, if acc < r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
-| If_icmple | RI | i64, if acc <= r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| Ldia | I | i64, loads constant [i] to getAcc |
+| Ldra | R | i64, loads the R1 to getAcc |
+| Star | R | i64, copies val of getAcc into register ri |
+| Add | R | i64, Adds R1 to getAcc |
+| Addi | I | i64, Adds [imm] to getAcc |
+| Sub | R | i64, sub R1 from getAcc |
+| Subi | I | i64, sub [imm] from getAcc |
+| Mul | R | i64, mul getAcc by R1 |
+| Muli | I | i64, mul getAcc by [imm] |
+| Div | R | i64, divides getAcc by R1 |
+| Divi | I | i64, divides getAcc by [imm] |
+| Modi | I | i64, getAcc = getAcc % [imm] |
+| Ldiaf | I | f64, loads constant [i] to getAcc |
+| Addf | R | f64, Adds R1 to getAcc |
+| Addif | I | f64, Adds [imm] to getAcc |
+| Subf | R | f64, sub R1 from getAcc |
+| Subif | I | i64, sub [imm] from getAcc |
+| Mulf | R | f64, mul getAcc by R1 |
+| Mulif | I | f64, mul getAcc by [imm] |
+| Divf | R | f64, divides getAcc by R1 |
+| Divif | I | f64, divides getAcc by [imm] |
+| IcPrint | N | Intrinsic. Prints value of getAcc |
+| IcScani | N | Intrinsic. Scans i64 from stdin into getAcc |
+| IcScanf | N | Intrinsic. Scans f64 from stdin into getAcc |
+| IcSqrt | N | f64, Intrinsic. Calculates sqrt of getAcc. |
+| IcSin | N | f64, Intrinsic. Calculates sin of getAcc. |
+| IcCos | N | f64, Intrinsic. Calculates cos of getAcc. |
+| If_icmpeq | RI | i64, if getAcc == r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| If_icmpne | RI | i64, if getAcc != r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| If_icmpgt | RI | i64, if getAcc > r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| If_icmpge | RI | i64, if getAcc >= r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| If_icmplt | RI | i64, if getAcc < r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
+| If_icmple | RI | i64, if getAcc <= r1 then branch to instruction at offset(instructions) imm otherwise just next instr |
 | If_acmpeq | RI | ref, if references are equal, branch to instruction at offset(instructions) imm |
 | If_acmpne | RI | ref, if references are not equal, branch to instruction at offset(instructions) imm |
-| If_null | I | ref, if reference in acc is null, branch to instruction at offset(instructions) imm |
-| Сmpgf | R | f64 -> i64, compare acc with r1. Acc became 1 i64 if greater than r1, 0 if equal, otherwise -1 |
-| Cmplf | R | f64 -> i64, compare acc with r1. Acc became 1 i64 if less than r1, 0 if equal, otherwise -1 |
+| If_null | I | ref, if reference in getAcc is null, branch to instruction at offset(instructions) imm |
+| Сmpgf | R | f64 -> i64, compare getAcc with r1. Acc became 1 i64 if greater than r1, 0 if equal, otherwise -1 |
+| Cmplf | R | f64 -> i64, compare getAcc with r1. Acc became 1 i64 if less than r1, 0 if equal, otherwise -1 |
 | Goto | I | Goes to another instruction at branchoffset(instructions) imm |
 | Call | I | Calls function [imm]. Imm is reference to function in constant pool named constant_func_name_and_type. |
-| NewI64Array | N | Allocates array of type i64 with number of elements from acc register. A reference to this new array is stored in acc. |
-| GetI64FromArr | R | Gets i64 from i64 array in acc and puts it to acc. |
-| SetI64InArr | RR | Sets i64 value to i64 array in acc and puts it to acc. |
-| NewF64Array | N | Allocates array of type f64 with number of elements from acc register. A reference to this new array is stored in acc. |
-| GetF64FromArr | R | Gets f64 from f64 array in acc and puts it to acc. |
-| SetF64InArr | RR | Sets f64 value to f64 array in acc and puts it to acc. |
-| NewRefArray | N | Allocates array of objects with number of elements from acc register. A reference to this new array is stored in acc. Value of every object is null. |
-| GetRefFromArr | R | Gets object by index in r1 from object array in acc and puts it to acc. |
-| SetRefInArr | RR | Sets ref value in r2 by index in r1 to object array in acc and puts it to acc. |
-| StringPrint | N | Prints String in acc |
-| StringConcat | R | Create new String, concatenating String in acc and String in r1. Result in acc. |
-| StringLen | N | Puts len of String to acc |
-| StringSlice | RR | Create new String slicing String placed in acc. r1 is start and r2 is finish of slicing |
-| AllocRef | I | Ref, Creates object of klass [imm], leaves ref in acc |
+| NewI64Array | N | Allocates array of type i64 with number of elements from getAcc register. A reference to this new array is stored in getAcc. |
+| GetI64FromArr | R | Gets i64 from i64 array in getAcc and puts it to acc. |
+| SetI64InArr | RR | Sets i64 value to i64 array in acc and puts it to getAcc. |
+| NewF64Array | N | Allocates array of type f64 with number of elements from getAcc register. A reference to this new array is stored in getAcc. |
+| GetF64FromArr | R | Gets f64 from f64 array in getAcc and puts it to acc. |
+| SetF64InArr | RR | Sets f64 value to f64 array in getAcc and puts it to acc. |
+| NewRefArray | N | Allocates array of objects with number of elements from getAcc register. A reference to this new array is stored in getAcc. Value of every object is null. |
+| GetRefFromArr | R | Gets object by index in r1 from object array in getAcc and puts it to acc. |
+| SetRefInArr | RR | Sets ref value in r2 by index in r1 to object array in getAcc and puts it to acc. |
+| StringPrint | N | Prints String in getAcc |
+| StringConcat | R | Create new String, concatenating String in getAcc and String in r1. Result in getAcc. |
+| StringLen | N | Puts len of String to getAcc |
+| StringSlice | RR | Create new String slicing String placed in getAcc. r1 is start and r2 is finish of slicing |
+| AllocRef | I | Ref, Creates object of klass [imm], leaves ref in getAcc |
 | MovRef | RR | Ref, copies ref from ri to rj |
-| LdraRef | R | Ref, copies ref of the R1 to acc |
-| StarRef | R | Ref, copies ref of acc into register ri |
-| GetField | I | Ref, load value of field to acc. imm is offset of field (starting from all fields) Value can be ref, array or String |
-| SetField | RI | Ref, set value of register to field imm of object in acc. Value should be consistent to field signature. There are no checks in runtime. |
+| LdraRef | R | Ref, copies ref of the R1 to getAcc |
+| StarRef | R | Ref, copies ref of getAcc into register ri |
+| GetField | I | Ref, load value of field to getAcc. imm is offset of field (starting from all fields) Value can be ref, array or String |
+| SetField | RI | Ref, set value of register to field imm of object in getAcc. Value should be consistent to field signature. There are no checks in runtime. |
 
 To generate this file use the following python script:
 ```shell

Makefile

@@ -23,7 +23,7 @@ format:
 	./tools/clang-format.sh $(PWD)/$(BENCH_DIR)
 .PHONY: build
 build: init
-	cmake -S $(PWD) -B $(PWD)/$(BUILD_DIR) -DCHAIVM_ADD_MEM_SANITIZER=OFF -DCHAIVM_ADD_THREAD_SANITIZER=OFF
+	cmake -S $(PWD) -B $(PWD)/$(BUILD_DIR) -DCHAIVM_ADD_MEM_SANITIZER=OFF -DCHAIVM_ADD_THREAD_SANITIZER=OFF -DCMAKE_BUILD_TYPE=RELEASE
 	cmake --build $(PWD)/$(BUILD_DIR) --parallel $(JOBS)
 
 .PHONY: test

README.md

@@ -18,7 +18,7 @@ fn main 8 0 {
   Star r3
   Ldra r3
   Mul r2
-  Printc acc
+  Printc getAcc
   Mov 0 r0
   Ret
 }

include/ChaiVM/interpreter/code-manager/klass.hpp

@@ -25,6 +25,11 @@ struct Klass {
 
     chsize_t nFields() const;
     chsize_t instanceSize() const;
+
+    inline bool fieldIsObject(chsize_t offset) const {
+        assert(offset % sizeof(chsize_t) == 0);
+        return fields_[offset / sizeof(chsize_t)].isObject_;
+    }
 };
 
 } // namespace chai::interpreter

include/ChaiVM/interpreter/executor.hpp

@@ -6,6 +6,7 @@
 #include "ChaiVM/memory/traced-allocator.hpp"
 #include "decoder.hpp"
 #include "frame.hpp"
+#include "garbage-collector.hpp"
 #include "objects.hpp"
 
 namespace chai::interpreter {
@@ -26,9 +27,18 @@ class Executor {
     void run();
 
     chsize_t &acc();
-    chsize_t acc() const;
+    chsize_t getAcc() const;
+
+    /**
+     * Checks if reference is stored in accumulator.
+     * @return true if accumulator stores reference, false otherwise.
+     */
+    bool isAccRef() const;
 
     Frame const *getCurrentFrame() const;
+    const CodeManager* getCodeManager() const;
+    memory::TracedByteAllocator& getObjectAllocator();
+    const GarbageCollector& getGC() const;
 
 private:
     chsize_t &pc();
@@ -167,13 +177,17 @@ class Executor {
         &Executor::set_field,
     };
 
+    void triggerGC();
+
 private:
     chsize_t acc_;
+    bool isAccRef_ = false;
     CodeManager *codeManager_;
     memory::LinearBuffer &framesBuffer_;
     memory::LinearBuffer &primitivesBuffer_;
     memory::TracedByteAllocator &objectsAllocator_;
     Frame *currentFrame_ = nullptr;
+    GarbageCollector gc_{*this};
 };
 
 inline void Executor::advancePc() { pc() += sizeof(bytecode_t); }

include/ChaiVM/interpreter/frame.hpp

@@ -3,6 +3,7 @@
 #include "ChaiVM/interpreter/code-manager/func-struct.hpp"
 #include "ChaiVM/memory/linear-allocator.hpp"
 #include "ChaiVM/memory/linear-buffer.hpp"
+#include <bitset>
 
 namespace chai::interpreter {
 
@@ -15,14 +16,19 @@ class Frame {
 
     chai::chsize_t &operator[](size_t n) &;
     const chsize_t &operator[](size_t n) const &;
+    size_t size() const;
+
+    bool isRegisterReference(RegisterId reg_id) const;
+    void setRegisterIsRef(RegisterId reg_id, bool val);
 
     /**
      * Get state.
      * @return state.
      */
-    std::vector<chsize_t> copyState();
+    std::vector<chsize_t> copyState() const;
 
     Frame *back();
+    const Frame *back() const;
 
 public:
     Function const &func_;
@@ -32,6 +38,8 @@ class Frame {
     Frame *prev_;
     size_t regsize_;
     std::vector<chsize_t, memory::LinearAllocator<chsize_t>> registers_;
+    // @todo #111:60min use custom allocator here
+    std::bitset<256> isRegRef_;
 };
 
 } // namespace chai::interpreter

include/ChaiVM/interpreter/garbage-collector.hpp

@@ -0,0 +1,33 @@
+#pragma once
+
+#include "ChaiVM/interpreter/objects.hpp"
+#include <unordered_set>
+
+// forward declaration for executor due to cyclic dependency
+namespace chai::interpreter {
+class Executor;
+}
+
+namespace chai::interpreter {
+
+using chai::interpreter::Executor;
+
+class GarbageCollector {
+public:
+    GarbageCollector(Executor &exec) : exec_(exec) {}
+
+    void collect();
+    const std::vector<interpreter::Object> &getRoots() const;
+private:
+    void collectRoots();
+    void mark();
+    void markObjects(Object obj);
+    void markObjectArrays(Object obj);
+    void sweep();
+
+    std::vector<interpreter::Object> roots_{};
+    std::unordered_set<chsize_t> markedRefs_{};
+    Executor &exec_;
+};
+
+} // namespace chai::interpreter

include/ChaiVM/interpreter/objects.hpp

@@ -3,7 +3,6 @@
 #include "ChaiVM/interpreter/code-manager/klass.hpp"
 #include "ChaiVM/memory/linear-allocator.hpp"
 #include "ChaiVM/memory/linear-buffer.hpp"
-#include "executor.hpp"
 #include <cassert>
 
 namespace chai::interpreter {
@@ -25,6 +24,11 @@ struct ObjectHeader {
      * Number of klass in "klass pool".
      */
     Immidiate klassId_;
+
+    /**
+     * Flag that object still has references
+     */
+    bool isMarked_;
 };
 
 /**
@@ -66,6 +70,11 @@ class Object {
      */
     void setMember(Immidiate offset, chsize_t value) const;
 
+    bool isMarked() const;
+    void mark();
+    void unmark();
+    Immidiate klassId() const;
+
 protected:
     ObjectHeader *header_;
 
@@ -97,6 +106,13 @@ class ObjectArray final : private Object {
      */
     explicit ObjectArray(chsize_t ref);
 
+    /**
+     * Ctor.
+     * Create object array from object.
+     * @param obj The object
+     */
+    explicit ObjectArray(Object obj);
+
     chsize_t length() const;
 
     chai::chsize_t &operator[](int64_t i) &;

include/ChaiVM/memory/traced-allocator.hpp

@@ -1,12 +1,22 @@
 #pragma once
 
 #include <list>
+#include <exception>
 
 #include "ChaiVM/memory/free-list-trash/free-list-allocator.hpp"
 #include "ChaiVM/types.hpp"
 
 namespace chai::memory {
 
+class out_of_memory_exception final : public std::bad_alloc {
+public:
+    const char* what() const noexcept {
+        return msg;
+    }
+private:
+    static constexpr const char* msg = "VM is out of memory";
+};
+
 struct AllocationInfo final {
     void *ptr = nullptr;
     size_t size = 0;
@@ -16,7 +26,7 @@ struct AllocationInfo final {
 class TracedByteAllocator final {
 public:
     TracedByteAllocator(size_t size)
-        : allocator_(size, FreeListAllocator::PlacementPolicy::FIND_FIRST) {
+        : size_(size), allocator_(size, FreeListAllocator::PlacementPolicy::FIND_FIRST) {
         allocator_.Init();
     }
 
@@ -27,7 +37,7 @@ class TracedByteAllocator final {
             allocations_.push_back(AllocationInfo{out, bytes, false});
             allocated_ += bytes;
         } else {
-            throw std::bad_alloc();
+            throw out_of_memory_exception();
         }
 
         return out;
@@ -38,11 +48,13 @@ class TracedByteAllocator final {
         // allocations_ are changed by GC
     }
 
-    auto &allocated() noexcept { return allocated_; }
-    auto const &allocated() const noexcept { return allocated_; }
-    auto &allocations() noexcept { return allocations_; }
+    size_t &allocated() { return allocated_; }
+    size_t const &allocated() const noexcept { return allocated_; }
+    auto &allocations() { return allocations_; }
+    size_t size() { return size_; }
 
 private:
+    size_t size_ = 0;
     size_t allocated_ = 0;
     std::list<AllocationInfo> allocations_;
     FreeListAllocator allocator_;