aoqi@0: /*
aoqi@0:  * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
aoqi@0:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
aoqi@0:  *
aoqi@0:  * This code is free software; you can redistribute it and/or modify it
aoqi@0:  * under the terms of the GNU General Public License version 2 only, as
aoqi@0:  * published by the Free Software Foundation.
aoqi@0:  *
aoqi@0:  * This code is distributed in the hope that it will be useful, but WITHOUT
aoqi@0:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
aoqi@0:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
aoqi@0:  * version 2 for more details (a copy is included in the LICENSE file that
aoqi@0:  * accompanied this code).
aoqi@0:  *
aoqi@0:  * You should have received a copy of the GNU General Public License version
aoqi@0:  * 2 along with this work; if not, write to the Free Software Foundation,
aoqi@0:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
aoqi@0:  *
aoqi@0:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
aoqi@0:  * or visit www.oracle.com if you need additional information or have any
aoqi@0:  * questions.
aoqi@0:  *
aoqi@0:  */
aoqi@0: 
aoqi@1: /*
aoqi@1:  * This file has been modified by Loongson Technology in 2015. These
aoqi@1:  * modifications are Copyright (c) 2015 Loongson Technology, and are made
aoqi@1:  * available on the same license terms set forth above.
aoqi@1:  */
aoqi@1: 
aoqi@0: #ifndef SHARE_VM_OPTO_OUTPUT_HPP
aoqi@0: #define SHARE_VM_OPTO_OUTPUT_HPP
aoqi@0: 
aoqi@0: #include "opto/block.hpp"
aoqi@0: #include "opto/node.hpp"
aoqi@0: #ifdef TARGET_ARCH_MODEL_x86_32
aoqi@0: # include "adfiles/ad_x86_32.hpp"
aoqi@0: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_x86_64
aoqi@0: # include "adfiles/ad_x86_64.hpp"
aoqi@0: #endif
aoqi@1: #ifdef TARGET_ARCH_MODEL_mips_64
aoqi@1: # include "adfiles/ad_mips_64.hpp"
aoqi@1: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_sparc
aoqi@0: # include "adfiles/ad_sparc.hpp"
aoqi@0: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_zero
aoqi@0: # include "adfiles/ad_zero.hpp"
aoqi@0: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_arm
aoqi@0: # include "adfiles/ad_arm.hpp"
aoqi@0: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_ppc_32
aoqi@0: # include "adfiles/ad_ppc_32.hpp"
aoqi@0: #endif
aoqi@0: #ifdef TARGET_ARCH_MODEL_ppc_64
aoqi@0: # include "adfiles/ad_ppc_64.hpp"
aoqi@0: #endif
aoqi@0: 
aoqi@0: class Arena;
aoqi@0: class Bundle;
aoqi@0: class Block;
aoqi@0: class Block_Array;
aoqi@0: class Node;
aoqi@0: class Node_Array;
aoqi@0: class Node_List;
aoqi@0: class PhaseCFG;
aoqi@0: class PhaseChaitin;
aoqi@0: class Pipeline_Use_Element;
aoqi@0: class Pipeline_Use;
aoqi@0: 
aoqi@0: #ifndef PRODUCT
aoqi@0: #define DEBUG_ARG(x) , x
aoqi@0: #else
aoqi@0: #define DEBUG_ARG(x)
aoqi@0: #endif
aoqi@0: 
aoqi@0: // Define the initial sizes for allocation of the resizable code buffer
aoqi@0: enum {
aoqi@0:   initial_code_capacity  =  16 * 1024,
aoqi@0:   initial_stub_capacity  =   4 * 1024,
aoqi@0:   initial_const_capacity =   4 * 1024,
aoqi@0:   initial_locs_capacity  =   3 * 1024
aoqi@0: };
aoqi@0: 
aoqi@0: //------------------------------Scheduling----------------------------------
aoqi@0: // This class contains all the information necessary to implement instruction
aoqi@0: // scheduling and bundling.
aoqi@0: class Scheduling {
aoqi@0: 
aoqi@0: private:
aoqi@0:   // Arena to use
aoqi@0:   Arena *_arena;
aoqi@0: 
aoqi@0:   // Control-Flow Graph info
aoqi@0:   PhaseCFG *_cfg;
aoqi@0: 
aoqi@0:   // Register Allocation info
aoqi@0:   PhaseRegAlloc *_regalloc;
aoqi@0: 
aoqi@0:   // Number of nodes in the method
aoqi@0:   uint _node_bundling_limit;
aoqi@0: 
aoqi@0:   // List of scheduled nodes. Generated in reverse order
aoqi@0:   Node_List _scheduled;
aoqi@0: 
aoqi@0:   // List of nodes currently available for choosing for scheduling
aoqi@0:   Node_List _available;
aoqi@0: 
aoqi@0:   // For each instruction beginning a bundle, the number of following
aoqi@0:   // nodes to be bundled with it.
aoqi@0:   Bundle *_node_bundling_base;
aoqi@0: 
aoqi@0:   // Mapping from register to Node
aoqi@0:   Node_List _reg_node;
aoqi@0: 
aoqi@0:   // Free list for pinch nodes.
aoqi@0:   Node_List _pinch_free_list;
aoqi@0: 
aoqi@0:   // Latency from the beginning of the containing basic block (base 1)
aoqi@0:   // for each node.
aoqi@0:   unsigned short *_node_latency;
aoqi@0: 
aoqi@0:   // Number of uses of this node within the containing basic block.
aoqi@0:   short *_uses;
aoqi@0: 
aoqi@0:   // Schedulable portion of current block.  Skips Region/Phi/CreateEx up
aoqi@0:   // front, branch+proj at end.  Also skips Catch/CProj (same as
aoqi@0:   // branch-at-end), plus just-prior exception-throwing call.
aoqi@0:   uint _bb_start, _bb_end;
aoqi@0: 
aoqi@0:   // Latency from the end of the basic block as scheduled
aoqi@0:   unsigned short *_current_latency;
aoqi@0: 
aoqi@0:   // Remember the next node
aoqi@0:   Node *_next_node;
aoqi@0: 
aoqi@0:   // Use this for an unconditional branch delay slot
aoqi@0:   Node *_unconditional_delay_slot;
aoqi@0: 
aoqi@0:   // Pointer to a Nop
aoqi@0:   MachNopNode *_nop;
aoqi@0: 
aoqi@0:   // Length of the current bundle, in instructions
aoqi@0:   uint _bundle_instr_count;
aoqi@0: 
aoqi@0:   // Current Cycle number, for computing latencies and bundling
aoqi@0:   uint _bundle_cycle_number;
aoqi@0: 
aoqi@0:   // Bundle information
aoqi@0:   Pipeline_Use_Element _bundle_use_elements[resource_count];
aoqi@0:   Pipeline_Use         _bundle_use;
aoqi@0: 
aoqi@0:   // Dump the available list
aoqi@0:   void dump_available() const;
aoqi@0: 
aoqi@0: public:
aoqi@0:   Scheduling(Arena *arena, Compile &compile);
aoqi@0: 
aoqi@0:   // Destructor
aoqi@0:   NOT_PRODUCT( ~Scheduling(); )
aoqi@0: 
aoqi@0:   // Step ahead "i" cycles
aoqi@0:   void step(uint i);
aoqi@0: 
aoqi@0:   // Step ahead 1 cycle, and clear the bundle state (for example,
aoqi@0:   // at a branch target)
aoqi@0:   void step_and_clear();
aoqi@0: 
aoqi@0:   Bundle* node_bundling(const Node *n) {
aoqi@0:     assert(valid_bundle_info(n), "oob");
aoqi@0:     return (&_node_bundling_base[n->_idx]);
aoqi@0:   }
aoqi@0: 
aoqi@0:   bool valid_bundle_info(const Node *n) const {
aoqi@0:     return (_node_bundling_limit > n->_idx);
aoqi@0:   }
aoqi@0: 
aoqi@0:   bool starts_bundle(const Node *n) const {
aoqi@0:     return (_node_bundling_limit > n->_idx && _node_bundling_base[n->_idx].starts_bundle());
aoqi@0:   }
aoqi@0: 
aoqi@0:   // Do the scheduling
aoqi@0:   void DoScheduling();
aoqi@0: 
aoqi@0:   // Compute the local latencies walking forward over the list of
aoqi@0:   // nodes for a basic block
aoqi@0:   void ComputeLocalLatenciesForward(const Block *bb);
aoqi@0: 
aoqi@0:   // Compute the register antidependencies within a basic block
aoqi@0:   void ComputeRegisterAntidependencies(Block *bb);
aoqi@0:   void verify_do_def( Node *n, OptoReg::Name def, const char *msg );
aoqi@0:   void verify_good_schedule( Block *b, const char *msg );
aoqi@0:   void anti_do_def( Block *b, Node *def, OptoReg::Name def_reg, int is_def );
aoqi@0:   void anti_do_use( Block *b, Node *use, OptoReg::Name use_reg );
aoqi@0: 
aoqi@0:   // Add a node to the current bundle
aoqi@0:   void AddNodeToBundle(Node *n, const Block *bb);
aoqi@0: 
aoqi@0:   // Add a node to the list of available nodes
aoqi@0:   void AddNodeToAvailableList(Node *n);
aoqi@0: 
aoqi@0:   // Compute the local use count for the nodes in a block, and compute
aoqi@0:   // the list of instructions with no uses in the block as available
aoqi@0:   void ComputeUseCount(const Block *bb);
aoqi@0: 
aoqi@0:   // Choose an instruction from the available list to add to the bundle
aoqi@0:   Node * ChooseNodeToBundle();
aoqi@0: 
aoqi@0:   // See if this Node fits into the currently accumulating bundle
aoqi@0:   bool NodeFitsInBundle(Node *n);
aoqi@0: 
aoqi@0:   // Decrement the use count for a node
aoqi@0:  void DecrementUseCounts(Node *n, const Block *bb);
aoqi@0: 
aoqi@0:   // Garbage collect pinch nodes for reuse by other blocks.
aoqi@0:   void garbage_collect_pinch_nodes();
aoqi@0:   // Clean up a pinch node for reuse (helper for above).
aoqi@0:   void cleanup_pinch( Node *pinch );
aoqi@0: 
aoqi@0:   // Information for statistics gathering
aoqi@0: #ifndef PRODUCT
aoqi@0: private:
aoqi@0:   // Gather information on size of nops relative to total
aoqi@0:   uint _branches, _unconditional_delays;
aoqi@0: 
aoqi@0:   static uint _total_nop_size, _total_method_size;
aoqi@0:   static uint _total_branches, _total_unconditional_delays;
aoqi@0:   static uint _total_instructions_per_bundle[Pipeline::_max_instrs_per_cycle+1];
aoqi@0: 
aoqi@0: public:
aoqi@0:   static void print_statistics();
aoqi@0: 
aoqi@0:   static void increment_instructions_per_bundle(uint i) {
aoqi@0:     _total_instructions_per_bundle[i]++;
aoqi@0:   }
aoqi@0: 
aoqi@0:   static void increment_nop_size(uint s) {
aoqi@0:     _total_nop_size += s;
aoqi@0:   }
aoqi@0: 
aoqi@0:   static void increment_method_size(uint s) {
aoqi@0:     _total_method_size += s;
aoqi@0:   }
aoqi@0: #endif
aoqi@0: 
aoqi@0: };
aoqi@0: 
aoqi@0: #endif // SHARE_VM_OPTO_OUTPUT_HPP