6953144: Tiered compilation
Summary: Infrastructure for tiered compilation support (interpreter + c1 + c2) for 32 and 64 bit. Simple tiered policy implementation.
Reviewed-by: kvn, never, phh, twisti
/*
* Copyright (c) 1997, 2007, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
// A frame represents a physical stack frame (an activation). Frames can be
// C or Java frames, and the Java frames can be interpreted or compiled.
// In contrast, vframes represent source-level activations, so that one physical frame
// can correspond to multiple source level frames because of inlining.
// A frame is comprised of {pc, fp, sp}
// ------------------------------ Asm interpreter ----------------------------------------
// Layout of asm interpreter frame:
// [expression stack ] * <- sp
// [monitors ] \
// ... | monitor block size
// [monitors ] /
// [monitor block size ]
// [byte code index/pointr] = bcx() bcx_offset
// [pointer to locals ] = locals() locals_offset
// [constant pool cache ] = cache() cache_offset
// [methodData ] = mdp() mdx_offset
// [methodOop ] = method() method_offset
// [last sp ] = last_sp() last_sp_offset
// [old stack pointer ] (sender_sp) sender_sp_offset
// [old frame pointer ] <- fp = link()
// [return pc ]
// [oop temp ] (only for native calls)
// [locals and parameters ]
// <- sender sp
// ------------------------------ Asm interpreter ----------------------------------------
// ------------------------------ C++ interpreter ----------------------------------------
//
// Layout of C++ interpreter frame: (While executing in BytecodeInterpreter::run)
//
// <- SP (current esp/rsp)
// [local variables ] BytecodeInterpreter::run local variables
// ... BytecodeInterpreter::run local variables
// [local variables ] BytecodeInterpreter::run local variables
// [old frame pointer ] fp [ BytecodeInterpreter::run's ebp/rbp ]
// [return pc ] (return to frame manager)
// [interpreter_state* ] (arg to BytecodeInterpreter::run) --------------
// [expression stack ] <- last_Java_sp |
// [... ] * <- interpreter_state.stack |
// [expression stack ] * <- interpreter_state.stack_base |
// [monitors ] \ |
// ... | monitor block size |
// [monitors ] / <- interpreter_state.monitor_base |
// [struct interpretState ] <-----------------------------------------|
// [return pc ] (return to callee of frame manager [1]
// [locals and parameters ]
// <- sender sp
// [1] When the c++ interpreter calls a new method it returns to the frame
// manager which allocates a new frame on the stack. In that case there
// is no real callee of this newly allocated frame. The frame manager is
// aware of the additional frame(s) and will pop them as nested calls
// complete. Howevers tTo make it look good in the debugger the frame
// manager actually installs a dummy pc pointing to RecursiveInterpreterActivation
// with a fake interpreter_state* parameter to make it easy to debug
// nested calls.
// Note that contrary to the layout for the assembly interpreter the
// expression stack allocated for the C++ interpreter is full sized.
// However this is not as bad as it seems as the interpreter frame_manager
// will truncate the unused space on succesive method calls.
//
// ------------------------------ C++ interpreter ----------------------------------------
public:
enum {
pc_return_offset = 0,
// All frames
link_offset = 0,
return_addr_offset = 1,
// non-interpreter frames
sender_sp_offset = 2,
#ifndef CC_INTERP
// Interpreter frames
interpreter_frame_result_handler_offset = 3, // for native calls only
interpreter_frame_oop_temp_offset = 2, // for native calls only
interpreter_frame_sender_sp_offset = -1,
// outgoing sp before a call to an invoked method
interpreter_frame_last_sp_offset = interpreter_frame_sender_sp_offset - 1,
interpreter_frame_method_offset = interpreter_frame_last_sp_offset - 1,
interpreter_frame_mdx_offset = interpreter_frame_method_offset - 1,
interpreter_frame_cache_offset = interpreter_frame_mdx_offset - 1,
interpreter_frame_locals_offset = interpreter_frame_cache_offset - 1,
interpreter_frame_bcx_offset = interpreter_frame_locals_offset - 1,
interpreter_frame_initial_sp_offset = interpreter_frame_bcx_offset - 1,
interpreter_frame_monitor_block_top_offset = interpreter_frame_initial_sp_offset,
interpreter_frame_monitor_block_bottom_offset = interpreter_frame_initial_sp_offset,
#endif // CC_INTERP
// Entry frames
#ifdef AMD64
#ifdef _WIN64
entry_frame_after_call_words = 8,
entry_frame_call_wrapper_offset = 2,
arg_reg_save_area_bytes = 32, // Register argument save area
#else
entry_frame_after_call_words = 13,
entry_frame_call_wrapper_offset = -6,
arg_reg_save_area_bytes = 0,
#endif // _WIN64
#else
entry_frame_call_wrapper_offset = 2,
#endif // AMD64
// Native frames
native_frame_initial_param_offset = 2
};
intptr_t ptr_at(int offset) const {
return *ptr_at_addr(offset);
}
void ptr_at_put(int offset, intptr_t value) {
*ptr_at_addr(offset) = value;
}
private:
// an additional field beyond _sp and _pc:
intptr_t* _fp; // frame pointer
// The interpreter and adapters will extend the frame of the caller.
// Since oopMaps are based on the sp of the caller before extension
// we need to know that value. However in order to compute the address
// of the return address we need the real "raw" sp. Since sparc already
// uses sp() to mean "raw" sp and unextended_sp() to mean the caller's
// original sp we use that convention.
intptr_t* _unextended_sp;
intptr_t* ptr_at_addr(int offset) const {
return (intptr_t*) addr_at(offset);
}
#if ASSERT
// Used in frame::sender_for_{interpreter,compiled}_frame
static void verify_deopt_original_pc( nmethod* nm, intptr_t* unextended_sp, bool is_method_handle_return = false);
static void verify_deopt_mh_original_pc(nmethod* nm, intptr_t* unextended_sp) {
verify_deopt_original_pc(nm, unextended_sp, true);
}
#endif
public:
// Constructors
frame(intptr_t* sp, intptr_t* fp, address pc);
frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address pc);
frame(intptr_t* sp, intptr_t* fp);
// accessors for the instance variables
intptr_t* fp() const { return _fp; }
inline address* sender_pc_addr() const;
// return address of param, zero origin index.
inline address* native_param_addr(int idx) const;
// expression stack tos if we are nested in a java call
intptr_t* interpreter_frame_last_sp() const;
#ifndef CC_INTERP
// deoptimization support
void interpreter_frame_set_last_sp(intptr_t* sp);
#endif // CC_INTERP
#ifdef CC_INTERP
inline interpreterState get_interpreterState() const;
#endif // CC_INTERP