|
/*
* Copyright (c) 1998, 2022, 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.
*
*/
#include "precompiled.hpp"
#include "asm/assembler.inline.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "code/compiledIC.hpp"
#include "code/debugInfo.hpp"
#include "code/debugInfoRec.hpp"
#include "compiler/compileBroker.hpp"
#include "compiler/compilerDirectives.hpp"
#include "compiler/disassembler.hpp"
#include "compiler/oopMap.hpp"
#include "gc/shared/barrierSet.hpp"
#include "gc/shared/c2/barrierSetC2.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/allocation.hpp"
#include "opto/ad.hpp"
#include "opto/block.hpp"
#include "opto/c2compiler.hpp"
#include "opto/c2_MacroAssembler.hpp"
#include "opto/callnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/locknode.hpp"
#include "opto/machnode.hpp"
#include "opto/node.hpp"
#include "opto/optoreg.hpp"
#include "opto/output.hpp"
#include "opto/regalloc.hpp"
#include "opto/runtime.hpp"
#include "opto/subnode.hpp"
#include "opto/type.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/sharedRuntime.hpp"
#include "utilities/macros.hpp"
#include "utilities/powerOfTwo.hpp"
#include "utilities/xmlstream.hpp"
#ifndef PRODUCT
#define DEBUG_ARG(x) , x
#else
#define DEBUG_ARG(x)
#endif
//------------------------------Scheduling----------------------------------
// This class contains all the information necessary to implement instruction
// scheduling and bundling.
class Scheduling {
private:
// Arena to use
Arena *_arena;
// Control-Flow Graph info
PhaseCFG *_cfg;
// Register Allocation info
PhaseRegAlloc *_regalloc;
// Number of nodes in the method
uint _node_bundling_limit;
// List of scheduled nodes. Generated in reverse order
Node_List _scheduled;
// List of nodes currently available for choosing for scheduling
Node_List _available;
// For each instruction beginning a bundle, the number of following
// nodes to be bundled with it.
Bundle *_node_bundling_base;
// Mapping from register to Node
Node_List _reg_node;
// Free list for pinch nodes.
Node_List _pinch_free_list;
// Number of uses of this node within the containing basic block.
short *_uses;
// Schedulable portion of current block. Skips Region/Phi/CreateEx up
// front, branch+proj at end. Also skips Catch/CProj (same as
// branch-at-end), plus just-prior exception-throwing call.
uint _bb_start, _bb_end;
// Latency from the end of the basic block as scheduled
unsigned short *_current_latency;
// Remember the next node
Node *_next_node;
// Use this for an unconditional branch delay slot
Node *_unconditional_delay_slot;
// Pointer to a Nop
MachNopNode *_nop;
// Length of the current bundle, in instructions
uint _bundle_instr_count;
// Current Cycle number, for computing latencies and bundling
uint _bundle_cycle_number;
// Bundle information
Pipeline_Use_Element _bundle_use_elements[resource_count];
Pipeline_Use _bundle_use;
// Dump the available list
void dump_available() const;
public:
Scheduling(Arena *arena, Compile &compile);
// Destructor
NOT_PRODUCT( ~Scheduling(); )
// Step ahead "i" cycles
void step(uint i);
// Step ahead 1 cycle, and clear the bundle state (for example,
// at a branch target)
void step_and_clear();
Bundle* node_bundling(const Node *n) {
assert(valid_bundle_info(n), "oob");
return (&_node_bundling_base[n->_idx]);
}
bool valid_bundle_info(const Node *n) const {
return (_node_bundling_limit > n->_idx);
}
bool starts_bundle(const Node *n) const {
return (_node_bundling_limit > n->_idx && _node_bundling_base[n->_idx].starts_bundle());
}
// Do the scheduling
void DoScheduling();
// Compute the register antidependencies within a basic block
void ComputeRegisterAntidependencies(Block *bb);
void verify_do_def( Node *n, OptoReg::Name def, const char *msg );
void verify_good_schedule( Block *b, const char *msg );
void anti_do_def( Block *b, Node *def, OptoReg::Name def_reg, int is_def );
void anti_do_use( Block *b, Node *use, OptoReg::Name use_reg );
// Add a node to the current bundle
void AddNodeToBundle(Node *n, const Block *bb);
// Add a node to the list of available nodes
void AddNodeToAvailableList(Node *n);
// Compute the local use count for the nodes in a block, and compute
// the list of instructions with no uses in the block as available
void ComputeUseCount(const Block *bb);
// Choose an instruction from the available list to add to the bundle
Node * ChooseNodeToBundle();
// See if this Node fits into the currently accumulating bundle
bool NodeFitsInBundle(Node *n);
// Decrement the use count for a node
void DecrementUseCounts(Node *n, const Block *bb);
// Garbage collect pinch nodes for reuse by other blocks.
void garbage_collect_pinch_nodes();
// Clean up a pinch node for reuse (helper for above).
void cleanup_pinch( Node *pinch );
// Information for statistics gathering
#ifndef PRODUCT
private:
// Gather information on size of nops relative to total
uint _branches, _unconditional_delays;
static uint _total_nop_size, _total_method_size;
static uint _total_branches, _total_unconditional_delays;
static uint _total_instructions_per_bundle[Pipeline::_max_instrs_per_cycle+1];
public:
static void print_statistics();
static void increment_instructions_per_bundle(uint i) {
_total_instructions_per_bundle[i]++;
}
static void increment_nop_size(uint s) {
_total_nop_size += s;
}
static void increment_method_size(uint s) {
_total_method_size += s;
}
#endif
};
volatile int C2SafepointPollStubTable::_stub_size = 0;
Label& C2SafepointPollStubTable::add_safepoint(uintptr_t safepoint_offset) {
C2SafepointPollStub* entry = new (Compile::current()->comp_arena()) C2SafepointPollStub(safepoint_offset);
_safepoints.append(entry);
return entry->_stub_label;
}
void C2SafepointPollStubTable::emit(CodeBuffer& cb) {
MacroAssembler masm(&cb);
for (int i = _safepoints.length() - 1; i >= 0; i--) {
// Make sure there is enough space in the code buffer
if (cb.insts()->maybe_expand_to_ensure_remaining(PhaseOutput::MAX_inst_size) && cb.blob() == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
C2SafepointPollStub* entry = _safepoints.at(i);
emit_stub(masm, entry);
}
}
int C2SafepointPollStubTable::stub_size_lazy() const {
int size = Atomic::load(&_stub_size);
if (size != 0) {
return size;
}
Compile* const C = Compile::current();
BufferBlob* const blob = C->output()->scratch_buffer_blob();
CodeBuffer cb(blob->content_begin(), C->output()->scratch_buffer_code_size());
MacroAssembler masm(&cb);
C2SafepointPollStub* entry = _safepoints.at(0);
emit_stub(masm, entry);
size += cb.insts_size();
Atomic::store(&_stub_size, size);
return size;
}
int C2SafepointPollStubTable::estimate_stub_size() const {
if (_safepoints.length() == 0) {
return 0;
}
int result = stub_size_lazy() * _safepoints.length();
#ifdef ASSERT
Compile* const C = Compile::current();
BufferBlob* const blob = C->output()->scratch_buffer_blob();
int size = 0;
for (int i = _safepoints.length() - 1; i >= 0; i--) {
CodeBuffer cb(blob->content_begin(), C->output()->scratch_buffer_code_size());
MacroAssembler masm(&cb);
C2SafepointPollStub* entry = _safepoints.at(i);
emit_stub(masm, entry);
size += cb.insts_size();
}
assert(size == result, "stubs should not have variable size");
#endif
return result;
}
// Nmethod entry barrier stubs
C2EntryBarrierStub* C2EntryBarrierStubTable::add_entry_barrier() {
assert(_stub == NULL, "There can only be one entry barrier stub");
_stub = new (Compile::current()->comp_arena()) C2EntryBarrierStub();
return _stub;
}
void C2EntryBarrierStubTable::emit(CodeBuffer& cb) {
if (_stub == NULL) {
// No stub - nothing to do
return;
}
C2_MacroAssembler masm(&cb);
// Make sure there is enough space in the code buffer
if (cb.insts()->maybe_expand_to_ensure_remaining(PhaseOutput::MAX_inst_size) && cb.blob() == NULL) {
ciEnv::current()->record_failure("CodeCache is full");
return;
}
intptr_t before = masm.offset();
masm.emit_entry_barrier_stub(_stub);
intptr_t after = masm.offset();
int actual_size = (int)(after - before);
int expected_size = masm.entry_barrier_stub_size();
assert(actual_size == expected_size, "Estimated size is wrong, expected %d, was %d", expected_size, actual_size);
}
int C2EntryBarrierStubTable::estimate_stub_size() const {
if (BarrierSet::barrier_set()->barrier_set_nmethod() == NULL) {
// No nmethod entry barrier?
return 0;
}
return C2_MacroAssembler::entry_barrier_stub_size();
}
PhaseOutput::PhaseOutput()
: Phase(Phase::Output),
_code_buffer("Compile::Fill_buffer"),
_first_block_size(0),
_handler_table(),
_inc_table(),
_safepoint_poll_table(),
_entry_barrier_table(),
_oop_map_set(NULL),
_scratch_buffer_blob(NULL),
_scratch_locs_memory(NULL),
_scratch_const_size(-1),
_in_scratch_emit_size(false),
_frame_slots(0),
_code_offsets(),
_node_bundling_limit(0),
_node_bundling_base(NULL),
_orig_pc_slot(0),
_orig_pc_slot_offset_in_bytes(0),
_buf_sizes(),
_block(NULL),
_index(0) {
C->set_output(this);
if (C->stub_name() == NULL) {
_orig_pc_slot = C->fixed_slots() - (sizeof(address) / VMRegImpl::stack_slot_size);
}
}
PhaseOutput::~PhaseOutput() {
C->set_output(NULL);
if (_scratch_buffer_blob != NULL) {
BufferBlob::free(_scratch_buffer_blob);
}
}
void PhaseOutput::perform_mach_node_analysis() {
// Late barrier analysis must be done after schedule and bundle
// Otherwise liveness based spilling will fail
BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
bs->late_barrier_analysis();
pd_perform_mach_node_analysis();
C->print_method(CompilerPhaseType::PHASE_MACH_ANALYSIS, 4);
}
// Convert Nodes to instruction bits and pass off to the VM
void PhaseOutput::Output() {
// RootNode goes
assert( C->cfg()->get_root_block()->number_of_nodes() == 0, "" );
// The number of new nodes (mostly MachNop) is proportional to
// the number of java calls and inner loops which are aligned.
if ( C->check_node_count((NodeLimitFudgeFactor + C->java_calls()*3 +
C->inner_loops()*(OptoLoopAlignment-1)),
"out of nodes before code generation" ) ) {
return;
}
// Make sure I can find the Start Node
Block *entry = C->cfg()->get_block(1);
Block *broot = C->cfg()->get_root_block();
const StartNode *start = entry->head()->as_Start();
// Replace StartNode with prolog
MachPrologNode *prolog = new MachPrologNode();
entry->map_node(prolog, 0);
C->cfg()->map_node_to_block(prolog, entry);
C->cfg()->unmap_node_from_block(start); // start is no longer in any block
// Virtual methods need an unverified entry point
if( C->is_osr_compilation() ) {
if( PoisonOSREntry ) {
// TODO: Should use a ShouldNotReachHereNode...
C->cfg()->insert( broot, 0, new MachBreakpointNode() );
}
} else {
if( C->method() && !C->method()->flags().is_static() ) {
// Insert unvalidated entry point
C->cfg()->insert( broot, 0, new MachUEPNode() );
}
}
// Break before main entry point
if ((C->method() && C->directive()->BreakAtExecuteOption) ||
(OptoBreakpoint && C->is_method_compilation()) ||
(OptoBreakpointOSR && C->is_osr_compilation()) ||
(OptoBreakpointC2R && !C->method()) ) {
// checking for C->method() means that OptoBreakpoint does not apply to
// runtime stubs or frame converters
C->cfg()->insert( entry, 1, new MachBreakpointNode() );
}
// Insert epilogs before every return
for (uint i = 0; i < C->cfg()->number_of_blocks(); i++) {
Block* block = C->cfg()->get_block(i);
if (!block->is_connector() && block->non_connector_successor(0) == C->cfg()->get_root_block()) { // Found a program exit point?
Node* m = block->end();
if (m->is_Mach() && m->as_Mach()->ideal_Opcode() != Op_Halt) {
MachEpilogNode* epilog = new MachEpilogNode(m->as_Mach()->ideal_Opcode() == Op_Return);
block->add_inst(epilog);
C->cfg()->map_node_to_block(epilog, block);
}
}
}
// Keeper of sizing aspects
_buf_sizes = BufferSizingData();
// Initialize code buffer
estimate_buffer_size(_buf_sizes._const);
if (C->failing()) return;
// Pre-compute the length of blocks and replace
// long branches with short if machine supports it.
// Must be done before ScheduleAndBundle due to SPARC delay slots
uint* blk_starts = NEW_RESOURCE_ARRAY(uint, C->cfg()->number_of_blocks() + 1);
blk_starts[0] = 0;
shorten_branches(blk_starts);
ScheduleAndBundle();
if (C->failing()) {
return;
}
perform_mach_node_analysis();
// Complete sizing of codebuffer
CodeBuffer* cb = init_buffer();
if (cb == NULL || C->failing()) {
return;
}
BuildOopMaps();
if (C->failing()) {
return;
}
fill_buffer(cb, blk_starts);
}
bool PhaseOutput::need_stack_bang(int frame_size_in_bytes) const {
// Determine if we need to generate a stack overflow check.
// Do it if the method is not a stub function and
// has java calls or has frame size > vm_page_size/8.
// The debug VM checks that deoptimization doesn't trigger an
// unexpected stack overflow (compiled method stack banging should
// guarantee it doesn't happen) so we always need the stack bang in
// a debug VM.
return (C->stub_function() == NULL &&
(C->has_java_calls() || frame_size_in_bytes > os::vm_page_size()>>3
DEBUG_ONLY(|| true)));
}
bool PhaseOutput::need_register_stack_bang() const {
// Determine if we need to generate a register stack overflow check.
// This is only used on architectures which have split register
// and memory stacks (ie. IA64).
// Bang if the method is not a stub function and has java calls
return (C->stub_function() == NULL && C->has_java_calls());
}
// Compute the size of first NumberOfLoopInstrToAlign instructions at the top
// of a loop. When aligning a loop we need to provide enough instructions
// in cpu's fetch buffer to feed decoders. The loop alignment could be
// avoided if we have enough instructions in fetch buffer at the head of a loop.
// By default, the size is set to 999999 by Block's constructor so that
// a loop will be aligned if the size is not reset here.
//
// Note: Mach instructions could contain several HW instructions
// so the size is estimated only.
//
void PhaseOutput::compute_loop_first_inst_sizes() {
// The next condition is used to gate the loop alignment optimization.
// Don't aligned a loop if there are enough instructions at the head of a loop
// or alignment padding is larger then MaxLoopPad. By default, MaxLoopPad
// is equal to OptoLoopAlignment-1 except on new Intel cpus, where it is
// equal to 11 bytes which is the largest address NOP instruction.
if (MaxLoopPad < OptoLoopAlignment - 1) {
uint last_block = C->cfg()->number_of_blocks() - 1;
for (uint i = 1; i <= last_block; i++) {
Block* block = C->cfg()->get_block(i);
// Check the first loop's block which requires an alignment.
if (block->loop_alignment() > (uint)relocInfo::addr_unit()) {
uint sum_size = 0;
uint inst_cnt = NumberOfLoopInstrToAlign;
inst_cnt = block->compute_first_inst_size(sum_size, inst_cnt, C->regalloc());
// Check subsequent fallthrough blocks if the loop's first
// block(s) does not have enough instructions.
Block *nb = block;
while(inst_cnt > 0 &&
i < last_block &&
!C->cfg()->get_block(i + 1)->has_loop_alignment() &&
!nb->has_successor(block)) {
i++;
nb = C->cfg()->get_block(i);
inst_cnt = nb->compute_first_inst_size(sum_size, inst_cnt, C->regalloc());
} // while( inst_cnt > 0 && i < last_block )
block->set_first_inst_size(sum_size);
} // f( b->head()->is_Loop() )
} // for( i <= last_block )
} // if( MaxLoopPad < OptoLoopAlignment-1 )
}
// The architecture description provides short branch variants for some long
// branch instructions. Replace eligible long branches with short branches.
void PhaseOutput::shorten_branches(uint* blk_starts) {
Compile::TracePhase tp("shorten branches", &timers[_t_shortenBranches]);
// Compute size of each block, method size, and relocation information size
uint nblocks = C->cfg()->number_of_blocks();
uint* jmp_offset = NEW_RESOURCE_ARRAY(uint,nblocks);
uint* jmp_size = NEW_RESOURCE_ARRAY(uint,nblocks);
int* jmp_nidx = NEW_RESOURCE_ARRAY(int ,nblocks);
// Collect worst case block paddings
int* block_worst_case_pad = NEW_RESOURCE_ARRAY(int, nblocks);
memset(block_worst_case_pad, 0, nblocks * sizeof(int));
DEBUG_ONLY( uint *jmp_target = NEW_RESOURCE_ARRAY(uint,nblocks); )
DEBUG_ONLY( uint *jmp_rule = NEW_RESOURCE_ARRAY(uint,nblocks); )
bool has_short_branch_candidate = false;
// Initialize the sizes to 0
int code_size = 0; // Size in bytes of generated code
int stub_size = 0; // Size in bytes of all stub entries
// Size in bytes of all relocation entries, including those in local stubs.
// Start with 2-bytes of reloc info for the unvalidated entry point
int reloc_size = 1; // Number of relocation entries
// Make three passes. The first computes pessimistic blk_starts,
// relative jmp_offset and reloc_size information. The second performs
// short branch substitution using the pessimistic sizing. The
// third inserts nops where needed.
// Step one, perform a pessimistic sizing pass.
uint last_call_adr = max_juint;
uint last_avoid_back_to_back_adr = max_juint;
uint nop_size = (new MachNopNode())->size(C->regalloc());
for (uint i = 0; i < nblocks; i++) { // For all blocks
Block* block = C->cfg()->get_block(i);
_block = block;
// During short branch replacement, we store the relative (to blk_starts)
// offset of jump in jmp_offset, rather than the absolute offset of jump.
// This is so that we do not need to recompute sizes of all nodes when
// we compute correct blk_starts in our next sizing pass.
jmp_offset[i] = 0;
jmp_size[i] = 0;
jmp_nidx[i] = -1;
DEBUG_ONLY( jmp_target[i] = 0; )
DEBUG_ONLY( jmp_rule[i] = 0; )
// Sum all instruction sizes to compute block size
uint last_inst = block->number_of_nodes();
uint blk_size = 0;
for (uint j = 0; j < last_inst; j++) {
_index = j;
Node* nj = block->get_node(_index);
// Handle machine instruction nodes
if (nj->is_Mach()) {
MachNode* mach = nj->as_Mach();
blk_size += (mach->alignment_required() - 1) * relocInfo::addr_unit(); // assume worst case padding
reloc_size += mach->reloc();
if (mach->is_MachCall()) {
// add size information for trampoline stub
// class CallStubImpl is platform-specific and defined in the *.ad files.
stub_size += CallStubImpl::size_call_trampoline();
reloc_size += CallStubImpl::reloc_call_trampoline();
MachCallNode *mcall = mach->as_MachCall();
// This destination address is NOT PC-relative
mcall->method_set((intptr_t)mcall->entry_point());
if (mcall->is_MachCallJava() && mcall->as_MachCallJava()->_method) {
stub_size += CompiledStaticCall::to_interp_stub_size();
reloc_size += CompiledStaticCall::reloc_to_interp_stub();
}
} else if (mach->is_MachSafePoint()) {
// If call/safepoint are adjacent, account for possible
// nop to disambiguate the two safepoints.
// ScheduleAndBundle() can rearrange nodes in a block,
// check for all offsets inside this block.
if (last_call_adr >= blk_starts[i]) {
blk_size += nop_size;
}
}
if (mach->avoid_back_to_back(MachNode::AVOID_BEFORE)) {
// Nop is inserted between "avoid back to back" instructions.
// ScheduleAndBundle() can rearrange nodes in a block,
// check for all offsets inside this block.
if (last_avoid_back_to_back_adr >= blk_starts[i]) {
blk_size += nop_size;
}
}
if (mach->may_be_short_branch()) {
if (!nj->is_MachBranch()) {
#ifndef PRODUCT
nj->dump(3);
#endif
Unimplemented();
}
assert(jmp_nidx[i] == -1, "block should have only one branch");
jmp_offset[i] = blk_size;
jmp_size[i] = nj->size(C->regalloc());
jmp_nidx[i] = j;
has_short_branch_candidate = true;
}
}
blk_size += nj->size(C->regalloc());
// Remember end of call offset
if (nj->is_MachCall() && !nj->is_MachCallLeaf()) {
last_call_adr = blk_starts[i]+blk_size;
}
// Remember end of avoid_back_to_back offset
if (nj->is_Mach() && nj->as_Mach()->avoid_back_to_back(MachNode::AVOID_AFTER)) {
last_avoid_back_to_back_adr = blk_starts[i]+blk_size;
}
}
// When the next block starts a loop, we may insert pad NOP
// instructions. Since we cannot know our future alignment,
// assume the worst.
if (i < nblocks - 1) {
Block* nb = C->cfg()->get_block(i + 1);
int max_loop_pad = nb->code_alignment()-relocInfo::addr_unit();
if (max_loop_pad > 0) {
assert(is_power_of_2(max_loop_pad+relocInfo::addr_unit()), "");
// Adjust last_call_adr and/or last_avoid_back_to_back_adr.
// If either is the last instruction in this block, bump by
// max_loop_pad in lock-step with blk_size, so sizing
// calculations in subsequent blocks still can conservatively
// detect that it may the last instruction in this block.
if (last_call_adr == blk_starts[i]+blk_size) {
last_call_adr += max_loop_pad;
}
if (last_avoid_back_to_back_adr == blk_starts[i]+blk_size) {
last_avoid_back_to_back_adr += max_loop_pad;
}
blk_size += max_loop_pad;
block_worst_case_pad[i + 1] = max_loop_pad;
}
}
// Save block size; update total method size
blk_starts[i+1] = blk_starts[i]+blk_size;
}
// Step two, replace eligible long jumps.
bool progress = true;
uint last_may_be_short_branch_adr = max_juint;
while (has_short_branch_candidate && progress) {
progress = false;
has_short_branch_candidate = false;
int adjust_block_start = 0;
for (uint i = 0; i < nblocks; i++) {
Block* block = C->cfg()->get_block(i);
int idx = jmp_nidx[i];
MachNode* mach = (idx == -1) ? NULL: block->get_node(idx)->as_Mach();
if (mach != NULL && mach->may_be_short_branch()) {
#ifdef ASSERT
assert(jmp_size[i] > 0 && mach->is_MachBranch(), "sanity");
int j;
// Find the branch; ignore trailing NOPs.
for (j = block->number_of_nodes()-1; j>=0; j--) {
Node* n = block->get_node(j);
if (!n->is_Mach() || n->as_Mach()->ideal_Opcode() != Op_Con)
break;
}
assert(j >= 0 && j == idx && block->get_node(j) == (Node*)mach, "sanity");
#endif
int br_size = jmp_size[i];
int br_offs = blk_starts[i] + jmp_offset[i];
// This requires the TRUE branch target be in succs[0]
uint bnum = block->non_connector_successor(0)->_pre_order;
int offset = blk_starts[bnum] - br_offs;
if (bnum > i) { // adjust following block's offset
offset -= adjust_block_start;
}
// This block can be a loop header, account for the padding
// in the previous block.
int block_padding = block_worst_case_pad[i];
assert(i == 0 || block_padding == 0 || br_offs >= block_padding, "Should have at least a padding on top");
// In the following code a nop could be inserted before
// the branch which will increase the backward distance.
bool needs_padding = ((uint)(br_offs - block_padding) == last_may_be_short_branch_adr);
assert(!needs_padding || jmp_offset[i] == 0, "padding only branches at the beginning of block");
if (needs_padding && offset <= 0)
offset -= nop_size;
if (C->matcher()->is_short_branch_offset(mach->rule(), br_size, offset)) {
// We've got a winner. Replace this branch.
MachNode* replacement = mach->as_MachBranch()->short_branch_version();
// Update the jmp_size.
int new_size = replacement->size(C->regalloc());
int diff = br_size - new_size;
assert(diff >= (int)nop_size, "short_branch size should be smaller");
// Conservatively take into account padding between
// avoid_back_to_back branches. Previous branch could be
// converted into avoid_back_to_back branch during next
// rounds.
if (needs_padding && replacement->avoid_back_to_back(MachNode::AVOID_BEFORE)) {
jmp_offset[i] += nop_size;
diff -= nop_size;
}
adjust_block_start += diff;
block->map_node(replacement, idx);
mach->subsume_by(replacement, C);
mach = replacement;
progress = true;
jmp_size[i] = new_size;
DEBUG_ONLY( jmp_target[i] = bnum; );
DEBUG_ONLY( jmp_rule[i] = mach->rule(); );
} else {
// The jump distance is not short, try again during next iteration.
has_short_branch_candidate = true;
}
} // (mach->may_be_short_branch())
if (mach != NULL && (mach->may_be_short_branch() ||
mach->avoid_back_to_back(MachNode::AVOID_AFTER))) {
last_may_be_short_branch_adr = blk_starts[i] + jmp_offset[i] + jmp_size[i];
}
blk_starts[i+1] -= adjust_block_start;
}
}
#ifdef ASSERT
for (uint i = 0; i < nblocks; i++) { // For all blocks
if (jmp_target[i] != 0) {
int br_size = jmp_size[i];
int offset = blk_starts[jmp_target[i]]-(blk_starts[i] + jmp_offset[i]);
if (!C->matcher()->is_short_branch_offset(jmp_rule[i], br_size, offset)) {
tty->print_cr("target (%d) - jmp_offset(%d) = offset (%d), jump_size(%d), jmp_block B%d, target_block B%d", blk_starts[jmp_target[i]], blk_starts[i] + jmp_offset[i], offset, br_size, i, jmp_target[i]);
}
assert(C->matcher()->is_short_branch_offset(jmp_rule[i], br_size, offset), "Displacement too large for short jmp");
}
}
#endif
// Step 3, compute the offsets of all blocks, will be done in fill_buffer()
// after ScheduleAndBundle().
// ------------------
// Compute size for code buffer
code_size = blk_starts[nblocks];
// Relocation records
reloc_size += 1; // Relo entry for exception handler
// Adjust reloc_size to number of record of relocation info
// Min is 2 bytes, max is probably 6 or 8, with a tax up to 25% for
// a relocation index.
// The CodeBuffer will expand the locs array if this estimate is too low.
reloc_size *= 10 / sizeof(relocInfo);
_buf_sizes._reloc = reloc_size;
_buf_sizes._code = code_size;
_buf_sizes._stub = stub_size;
}
//------------------------------FillLocArray-----------------------------------
// Create a bit of debug info and append it to the array. The mapping is from
// Java local or expression stack to constant, register or stack-slot. For
// doubles, insert 2 mappings and return 1 (to tell the caller that the next
// entry has been taken care of and caller should skip it).
static LocationValue *new_loc_value( PhaseRegAlloc *ra, OptoReg::Name regnum, Location::Type l_type ) {
// This should never have accepted Bad before
assert(OptoReg::is_valid(regnum), "location must be valid");
return (OptoReg::is_reg(regnum))
? new LocationValue(Location::new_reg_loc(l_type, OptoReg::as_VMReg(regnum)) )
: new LocationValue(Location::new_stk_loc(l_type, ra->reg2offset(regnum)));
}
ObjectValue*
PhaseOutput::sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id) {
for (int i = 0; i < objs->length(); i++) {
assert(objs->at(i)->is_object(), "corrupt object cache");
ObjectValue* sv = (ObjectValue*) objs->at(i);
if (sv->id() == id) {
return sv;
}
}
// Otherwise..
return NULL;
}
void PhaseOutput::set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
ObjectValue* sv ) {
assert(sv_for_node_id(objs, sv->id()) == NULL, "Precondition");
objs->append(sv);
}
void PhaseOutput::FillLocArray( int idx, MachSafePointNode* sfpt, Node *local,
GrowableArray<ScopeValue*> *array,
GrowableArray<ScopeValue*> *objs ) {
assert( local, "use _top instead of null" );
if (array->length() != idx) {
assert(array->length() == idx + 1, "Unexpected array count");
// Old functionality:
// return
// New functionality:
// Assert if the local is not top. In product mode let the new node
// override the old entry.
assert(local == C->top(), "LocArray collision");
if (local == C->top()) {
return;
}
array->pop();
}
const Type *t = local->bottom_type();
// Is it a safepoint scalar object node?
if (local->is_SafePointScalarObject()) {
SafePointScalarObjectNode* spobj = local->as_SafePointScalarObject();
ObjectValue* sv = sv_for_node_id(objs, spobj->_idx);
if (sv == NULL) {
ciKlass* cik = t->is_oopptr()->exact_klass();
assert(cik->is_instance_klass() ||
cik->is_array_klass(), "Not supported allocation.");
sv = new ObjectValue(spobj->_idx,
new ConstantOopWriteValue(cik->java_mirror()->constant_encoding()));
set_sv_for_object_node(objs, sv);
uint first_ind = spobj->first_index(sfpt->jvms());
for (uint i = 0; i < spobj->n_fields(); i++) {
Node* fld_node = sfpt->in(first_ind+i);
(void)FillLocArray(sv->field_values()->length(), sfpt, fld_node, sv->field_values(), objs);
}
}
array->append(sv);
return;
}
// Grab the register number for the local
OptoReg::Name regnum = C->regalloc()->get_reg_first(local);
if( OptoReg::is_valid(regnum) ) {// Got a register/stack?
// Record the double as two float registers.
// The register mask for such a value always specifies two adjacent
// float registers, with the lower register number even.
// Normally, the allocation of high and low words to these registers
// is irrelevant, because nearly all operations on register pairs
// (e.g., StoreD) treat them as a single unit.
// Here, we assume in addition that the words in these two registers
// stored "naturally" (by operations like StoreD and double stores
// within the interpreter) such that the lower-numbered register
// is written to the lower memory address. This may seem like
// a machine dependency, but it is not--it is a requirement on
// the author of the <arch>.ad file to ensure that, for every
// even/odd double-register pair to which a double may be allocated,
// the word in the even single-register is stored to the first
// memory word. (Note that register numbers are completely
// arbitrary, and are not tied to any machine-level encodings.)
#ifdef _LP64
if( t->base() == Type::DoubleBot || t->base() == Type::DoubleCon ) {
array->append(new ConstantIntValue((jint)0));
array->append(new_loc_value( C->regalloc(), regnum, Location::dbl ));
} else if ( t->base() == Type::Long ) {
array->append(new ConstantIntValue((jint)0));
array->append(new_loc_value( C->regalloc(), regnum, Location::lng ));
} else if ( t->base() == Type::RawPtr ) {
// jsr/ret return address which must be restored into the full
// width 64-bit stack slot.
array->append(new_loc_value( C->regalloc(), regnum, Location::lng ));
}
#else //_LP64
if( t->base() == Type::DoubleBot || t->base() == Type::DoubleCon || t->base() == Type::Long ) {
// Repack the double/long as two jints.
// The convention the interpreter uses is that the second local
// holds the first raw word of the native double representation.
// This is actually reasonable, since locals and stack arrays
// grow downwards in all implementations.
// (If, on some machine, the interpreter's Java locals or stack
// were to grow upwards, the embedded doubles would be word-swapped.)
array->append(new_loc_value( C->regalloc(), OptoReg::add(regnum,1), Location::normal ));
array->append(new_loc_value( C->regalloc(), regnum , Location::normal ));
}
#endif //_LP64
else if( (t->base() == Type::FloatBot || t->base() == Type::FloatCon) &&
OptoReg::is_reg(regnum) ) {
array->append(new_loc_value( C->regalloc(), regnum, Matcher::float_in_double()
? Location::float_in_dbl : Location::normal ));
} else if( t->base() == Type::Int && OptoReg::is_reg(regnum) ) {
array->append(new_loc_value( C->regalloc(), regnum, Matcher::int_in_long
? Location::int_in_long : Location::normal ));
} else if( t->base() == Type::NarrowOop ) {
array->append(new_loc_value( C->regalloc(), regnum, Location::narrowoop ));
} else if (t->base() == Type::VectorA || t->base() == Type::VectorS ||
t->base() == Type::VectorD || t->base() == Type::VectorX ||
t->base() == Type::VectorY || t->base() == Type::VectorZ) {
array->append(new_loc_value( C->regalloc(), regnum, Location::vector ));
} else if (C->regalloc()->is_oop(local)) {
assert(t->base() == Type::OopPtr || t->base() == Type::InstPtr ||
t->base() == Type::AryPtr,
"Unexpected type: %s", t->msg());
array->append(new_loc_value( C->regalloc(), regnum, Location::oop ));
} else {
assert(t->base() == Type::Int || t->base() == Type::Half ||
t->base() == Type::FloatCon || t->base() == Type::FloatBot,
"Unexpected type: %s", t->msg());
array->append(new_loc_value( C->regalloc(), regnum, Location::normal ));
}
return;
}
// No register. It must be constant data.
switch (t->base()) {
case Type::Half: // Second half of a double
ShouldNotReachHere(); // Caller should skip 2nd halves
break;
case Type::AnyPtr:
array->append(new ConstantOopWriteValue(NULL));
break;
case Type::AryPtr:
case Type::InstPtr: // fall through
array->append(new ConstantOopWriteValue(t->isa_oopptr()->const_oop()->constant_encoding()));
break;
case Type::NarrowOop:
if (t == TypeNarrowOop::NULL_PTR) {
array->append(new ConstantOopWriteValue(NULL));
} else {
array->append(new ConstantOopWriteValue(t->make_ptr()->isa_oopptr()->const_oop()->constant_encoding()));
}
break;
case Type::Int:
array->append(new ConstantIntValue(t->is_int()->get_con()));
break;
case Type::RawPtr:
// A return address (T_ADDRESS).
assert((intptr_t)t->is_ptr()->get_con() < (intptr_t)0x10000, "must be a valid BCI");
#ifdef _LP64
// Must be restored to the full-width 64-bit stack slot.
array->append(new ConstantLongValue(t->is_ptr()->get_con()));
#else
array->append(new ConstantIntValue(t->is_ptr()->get_con()));
#endif
break;
case Type::FloatCon: {
float f = t->is_float_constant()->getf();
array->append(new ConstantIntValue(jint_cast(f)));
break;
}
case Type::DoubleCon: {
jdouble d = t->is_double_constant()->getd();
#ifdef _LP64
array->append(new ConstantIntValue((jint)0));
array->append(new ConstantDoubleValue(d));
#else
// Repack the double as two jints.
// The convention the interpreter uses is that the second local
// holds the first raw word of the native double representation.
// This is actually reasonable, since locals and stack arrays
// grow downwards in all implementations.
// (If, on some machine, the interpreter's Java locals or stack
// were to grow upwards, the embedded doubles would be word-swapped.)
jlong_accessor acc;
acc.long_value = jlong_cast(d);
array->append(new ConstantIntValue(acc.words[1]));
array->append(new ConstantIntValue(acc.words[0]));
#endif
break;
}
case Type::Long: {
jlong d = t->is_long()->get_con();
#ifdef _LP64
array->append(new ConstantIntValue((jint)0));
array->append(new ConstantLongValue(d));
#else
// Repack the long as two jints.
// The convention the interpreter uses is that the second local
// holds the first raw word of the native double representation.
// This is actually reasonable, since locals and stack arrays
// grow downwards in all implementations.
// (If, on some machine, the interpreter's Java locals or stack
// were to grow upwards, the embedded doubles would be word-swapped.)
jlong_accessor acc;
acc.long_value = d;
array->append(new ConstantIntValue(acc.words[1]));
array->append(new ConstantIntValue(acc.words[0]));
#endif
break;
}
case Type::Top: // Add an illegal value here
array->append(new LocationValue(Location()));
break;
default:
ShouldNotReachHere();
break;
}
}
// Determine if this node starts a bundle
bool PhaseOutput::starts_bundle(const Node *n) const {
return (_node_bundling_limit > n->_idx &&
_node_bundling_base[n->_idx].starts_bundle());
}
//--------------------------Process_OopMap_Node--------------------------------
void PhaseOutput::Process_OopMap_Node(MachNode *mach, int current_offset) {
// Handle special safepoint nodes for synchronization
MachSafePointNode *sfn = mach->as_MachSafePoint();
MachCallNode *mcall;
int safepoint_pc_offset = current_offset;
bool is_method_handle_invoke = false;
bool return_oop = false;
bool has_ea_local_in_scope = sfn->_has_ea_local_in_scope;
bool arg_escape = false;
// Add the safepoint in the DebugInfoRecorder
if( !mach->is_MachCall() ) {
mcall = NULL;
C->debug_info()->add_safepoint(safepoint_pc_offset, sfn->_oop_map);
} else {
mcall = mach->as_MachCall();
// Is the call a MethodHandle call?
if (mcall->is_MachCallJava()) {
if (mcall->as_MachCallJava()->_method_handle_invoke) {
assert(C->has_method_handle_invokes(), "must have been set during call generation");
is_method_handle_invoke = true;
}
arg_escape = mcall->as_MachCallJava()->_arg_escape;
}
// Check if a call returns an object.
if (mcall->returns_pointer()) {
return_oop = true;
}
safepoint_pc_offset += mcall->ret_addr_offset();
C->debug_info()->add_safepoint(safepoint_pc_offset, mcall->_oop_map);
}
// Loop over the JVMState list to add scope information
// Do not skip safepoints with a NULL method, they need monitor info
JVMState* youngest_jvms = sfn->jvms();
int max_depth = youngest_jvms->depth();
// Allocate the object pool for scalar-replaced objects -- the map from
// small-integer keys (which can be recorded in the local and ostack
// arrays) to descriptions of the object state.
GrowableArray<ScopeValue*> *objs = new GrowableArray<ScopeValue*>();
// Visit scopes from oldest to youngest.
for (int depth = 1; depth <= max_depth; depth++) {
JVMState* jvms = youngest_jvms->of_depth(depth);
int idx;
ciMethod* method = jvms->has_method() ? jvms->method() : NULL;
// Safepoints that do not have method() set only provide oop-map and monitor info
// to support GC; these do not support deoptimization.
int num_locs = (method == NULL) ? 0 : jvms->loc_size();
int num_exps = (method == NULL) ? 0 : jvms->stk_size();
int num_mon = jvms->nof_monitors();
assert(method == NULL || jvms->bci() < 0 || num_locs == method->max_locals(),
"JVMS local count must match that of the method");
// Add Local and Expression Stack Information
// Insert locals into the locarray
GrowableArray<ScopeValue*> *locarray = new GrowableArray<ScopeValue*>(num_locs);
for( idx = 0; idx < num_locs; idx++ ) {
FillLocArray( idx, sfn, sfn->local(jvms, idx), locarray, objs );
}
// Insert expression stack entries into the exparray
GrowableArray<ScopeValue*> *exparray = new GrowableArray<ScopeValue*>(num_exps);
for( idx = 0; idx < num_exps; idx++ ) {
FillLocArray( idx, sfn, sfn->stack(jvms, idx), exparray, objs );
}
// Add in mappings of the monitors
assert( !method ||
!method->is_synchronized() ||
method->is_native() ||
num_mon > 0 ||
!GenerateSynchronizationCode,
"monitors must always exist for synchronized methods");
// Build the growable array of ScopeValues for exp stack
GrowableArray<MonitorValue*> *monarray = new GrowableArray<MonitorValue*>(num_mon);
// Loop over monitors and insert into array
for (idx = 0; idx < num_mon; idx++) {
// Grab the node that defines this monitor
Node* box_node = sfn->monitor_box(jvms, idx);
Node* obj_node = sfn->monitor_obj(jvms, idx);
// Create ScopeValue for object
ScopeValue *scval = NULL;
if (obj_node->is_SafePointScalarObject()) {
SafePointScalarObjectNode* spobj = obj_node->as_SafePointScalarObject();
scval = PhaseOutput::sv_for_node_id(objs, spobj->_idx);
if (scval == NULL) {
const Type *t = spobj->bottom_type();
ciKlass* cik = t->is_oopptr()->exact_klass();
assert(cik->is_instance_klass() ||
cik->is_array_klass(), "Not supported allocation.");
ObjectValue* sv = new ObjectValue(spobj->_idx,
new ConstantOopWriteValue(cik->java_mirror()->constant_encoding()));
PhaseOutput::set_sv_for_object_node(objs, sv);
uint first_ind = spobj->first_index(youngest_jvms);
for (uint i = 0; i < spobj->n_fields(); i++) {
Node* fld_node = sfn->in(first_ind+i);
(void)FillLocArray(sv->field_values()->length(), sfn, fld_node, sv->field_values(), objs);
}
scval = sv;
}
} else if (!obj_node->is_Con()) {
OptoReg::Name obj_reg = C->regalloc()->get_reg_first(obj_node);
if( obj_node->bottom_type()->base() == Type::NarrowOop ) {
scval = new_loc_value( C->regalloc(), obj_reg, Location::narrowoop );
} else {
scval = new_loc_value( C->regalloc(), obj_reg, Location::oop );
}
} else {
const TypePtr *tp = obj_node->get_ptr_type();
scval = new ConstantOopWriteValue(tp->is_oopptr()->const_oop()->constant_encoding());
}
OptoReg::Name box_reg = BoxLockNode::reg(box_node);
Location basic_lock = Location::new_stk_loc(Location::normal,C->regalloc()->reg2offset(box_reg));
bool eliminated = (box_node->is_BoxLock() && box_node->as_BoxLock()->is_eliminated());
monarray->append(new MonitorValue(scval, basic_lock, eliminated));
}
// We dump the object pool first, since deoptimization reads it in first.
C->debug_info()->dump_object_pool(objs);
// Build first class objects to pass to scope
DebugToken *locvals = C->debug_info()->create_scope_values(locarray);
DebugToken *expvals = C->debug_info()->create_scope_values(exparray);
DebugToken *monvals = C->debug_info()->create_monitor_values(monarray);
// Make method available for all Safepoints
ciMethod* scope_method = method ? method : C->method();
// Describe the scope here
assert(jvms->bci() >= InvocationEntryBci && jvms->bci() <= 0x10000, "must be a valid or entry BCI");
assert(!jvms->should_reexecute() || depth == max_depth, "reexecute allowed only for the youngest");
// Now we can describe the scope.
methodHandle null_mh;
bool rethrow_exception = false;
C->debug_info()->describe_scope(
safepoint_pc_offset,
null_mh,
scope_method,
jvms->bci(),
jvms->should_reexecute(),
rethrow_exception,
is_method_handle_invoke,
return_oop,
has_ea_local_in_scope,
arg_escape,
locvals,
expvals,
monvals
);
} // End jvms loop
// Mark the end of the scope set.
C->debug_info()->end_safepoint(safepoint_pc_offset);
}
// A simplified version of Process_OopMap_Node, to handle non-safepoints.
class NonSafepointEmitter {
Compile* C;
JVMState* _pending_jvms;
int _pending_offset;
void emit_non_safepoint();
public:
NonSafepointEmitter(Compile* compile) {
this->C = compile;
_pending_jvms = NULL;
_pending_offset = 0;
}
void observe_instruction(Node* n, int pc_offset) {
if (!C->debug_info()->recording_non_safepoints()) return;
Node_Notes* nn = C->node_notes_at(n->_idx);
if (nn == NULL || nn->jvms() == NULL) return;
if (_pending_jvms != NULL &&
_pending_jvms->same_calls_as(nn->jvms())) {
// Repeated JVMS? Stretch it up here.
_pending_offset = pc_offset;
} else {
if (_pending_jvms != NULL &&
_pending_offset < pc_offset) {
emit_non_safepoint();
}
_pending_jvms = NULL;
if (pc_offset > C->debug_info()->last_pc_offset()) {
// This is the only way _pending_jvms can become non-NULL:
_pending_jvms = nn->jvms();
_pending_offset = pc_offset;
}
}
}
// Stay out of the way of real safepoints:
void observe_safepoint(JVMState* jvms, int pc_offset) {
if (_pending_jvms != NULL &&
!_pending_jvms->same_calls_as(jvms) &&
_pending_offset < pc_offset) {
emit_non_safepoint();
}
_pending_jvms = NULL;
}
void flush_at_end() {
if (_pending_jvms != NULL) {
emit_non_safepoint();
}
_pending_jvms = NULL;
}
};
void NonSafepointEmitter::emit_non_safepoint() {
JVMState* youngest_jvms = _pending_jvms;
int pc_offset = _pending_offset;
// Clear it now:
_pending_jvms = NULL;
DebugInformationRecorder* debug_info = C->debug_info();
assert(debug_info->recording_non_safepoints(), "sanity");
debug_info->add_non_safepoint(pc_offset);
int max_depth = youngest_jvms->depth();
// Visit scopes from oldest to youngest.
for (int depth = 1; depth <= max_depth; depth++) {
JVMState* jvms = youngest_jvms->of_depth(depth);
ciMethod* method = jvms->has_method() ? jvms->method() : NULL;
assert(!jvms->should_reexecute() || depth==max_depth, "reexecute allowed only for the youngest");
methodHandle null_mh;
debug_info->describe_scope(pc_offset, null_mh, method, jvms->bci(), jvms->should_reexecute());
}
// Mark the end of the scope set.
debug_info->end_non_safepoint(pc_offset);
}
//------------------------------init_buffer------------------------------------
void PhaseOutput::estimate_buffer_size(int& const_req) {
// Set the initially allocated size
const_req = initial_const_capacity;
// The extra spacing after the code is necessary on some platforms.
// Sometimes we need to patch in a jump after the last instruction,
// if the nmethod has been deoptimized. (See 4932387, 4894843.)
// Compute the byte offset where we can store the deopt pc.
if (C->fixed_slots() != 0) {
_orig_pc_slot_offset_in_bytes = C->regalloc()->reg2offset(OptoReg::stack2reg(_orig_pc_slot));
}
// Compute prolog code size
_method_size = 0;
_frame_slots = OptoReg::reg2stack(C->matcher()->_old_SP) + C->regalloc()->_framesize;
assert(_frame_slots >= 0 && _frame_slots < 1000000, "sanity check");
if (C->has_mach_constant_base_node()) {
uint add_size = 0;
// Fill the constant table.
// Note: This must happen before shorten_branches.
for (uint i = 0; i < C->cfg()->number_of_blocks(); i++) {
Block* b = C->cfg()->get_block(i);
for (uint j = 0; j < b->number_of_nodes(); j++) {
Node* n = b->get_node(j);
// If the node is a MachConstantNode evaluate the constant
// value section.
if (n->is_MachConstant()) {
MachConstantNode* machcon = n->as_MachConstant();
machcon->eval_constant(C);
} else if (n->is_Mach()) {
// On Power there are more nodes that issue constants.
add_size += (n->as_Mach()->ins_num_consts() * 8);
}
}
}
// Calculate the offsets of the constants and the size of the
// constant table (including the padding to the next section).
constant_table().calculate_offsets_and_size();
const_req = constant_table().size() + add_size;
}
// Initialize the space for the BufferBlob used to find and verify
// instruction size in MachNode::emit_size()
init_scratch_buffer_blob(const_req);
}
CodeBuffer* PhaseOutput::init_buffer() {
int stub_req = _buf_sizes._stub;
int code_req = _buf_sizes._code;
int const_req = _buf_sizes._const;
int pad_req = NativeCall::instruction_size;
BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
stub_req += bs->estimate_stub_size();
stub_req += safepoint_poll_table()->estimate_stub_size();
stub_req += entry_barrier_table()->estimate_stub_size();
// nmethod and CodeBuffer count stubs & constants as part of method's code.
// class HandlerImpl is platform-specific and defined in the *.ad files.
int exception_handler_req = HandlerImpl::size_exception_handler() + MAX_stubs_size; // add marginal slop for handler
int deopt_handler_req = HandlerImpl::size_deopt_handler() + MAX_stubs_size; // add marginal slop for handler
stub_req += MAX_stubs_size; // ensure per-stub margin
code_req += MAX_inst_size; // ensure per-instruction margin
if (StressCodeBuffers)
code_req = const_req = stub_req = exception_handler_req = deopt_handler_req = 0x10; // force expansion
int total_req =
const_req +
code_req +
pad_req +
stub_req +
exception_handler_req +
deopt_handler_req; // deopt handler
if (C->has_method_handle_invokes())
total_req += deopt_handler_req; // deopt MH handler
CodeBuffer* cb = code_buffer();
cb->initialize(total_req, _buf_sizes._reloc);
// Have we run out of code space?
if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) {
C->record_failure("CodeCache is full");
return NULL;
}
// Configure the code buffer.
cb->initialize_consts_size(const_req);
cb->initialize_stubs_size(stub_req);
cb->initialize_oop_recorder(C->env()->oop_recorder());
// fill in the nop array for bundling computations
MachNode *_nop_list[Bundle::_nop_count];
Bundle::initialize_nops(_nop_list);
return cb;
}
//------------------------------fill_buffer------------------------------------
void PhaseOutput::fill_buffer(CodeBuffer* cb, uint* blk_starts) {
// blk_starts[] contains offsets calculated during short branches processing,
// offsets should not be increased during following steps.
// Compute the size of first NumberOfLoopInstrToAlign instructions at head
// of a loop. It is used to determine the padding for loop alignment.
Compile::TracePhase tp("fill buffer", &timers[_t_fillBuffer]);
compute_loop_first_inst_sizes();
// Create oopmap set.
_oop_map_set = new OopMapSet();
// !!!!! This preserves old handling of oopmaps for now
C->debug_info()->set_oopmaps(_oop_map_set);
uint nblocks = C->cfg()->number_of_blocks();
// Count and start of implicit null check instructions
uint inct_cnt = 0;
uint* inct_starts = NEW_RESOURCE_ARRAY(uint, nblocks+1);
// Count and start of calls
uint* call_returns = NEW_RESOURCE_ARRAY(uint, nblocks+1);
uint return_offset = 0;
int nop_size = (new MachNopNode())->size(C->regalloc());
int previous_offset = 0;
int current_offset = 0;
int last_call_offset = -1;
int last_avoid_back_to_back_offset = -1;
#ifdef ASSERT
uint* jmp_target = NEW_RESOURCE_ARRAY(uint,nblocks);
uint* jmp_offset = NEW_RESOURCE_ARRAY(uint,nblocks);
uint* jmp_size = NEW_RESOURCE_ARRAY(uint,nblocks);
uint* jmp_rule = NEW_RESOURCE_ARRAY(uint,nblocks);
#endif
// Create an array of unused labels, one for each basic block, if printing is enabled
#if defined(SUPPORT_OPTO_ASSEMBLY)
int* node_offsets = NULL;
uint node_offset_limit = C->unique();
if (C->print_assembly()) {
node_offsets = NEW_RESOURCE_ARRAY(int, node_offset_limit);
}
if (node_offsets != NULL) {
// We need to initialize. Unused array elements may contain garbage and mess up PrintOptoAssembly.
memset(node_offsets, 0, node_offset_limit*sizeof(int));
}
#endif
NonSafepointEmitter non_safepoints(C); // emit non-safepoints lazily
// Emit the constant table.
if (C->has_mach_constant_base_node()) {
if (!constant_table().emit(*cb)) {
C->record_failure("consts section overflow");
return;
}
}
// Create an array of labels, one for each basic block
Label* blk_labels = NEW_RESOURCE_ARRAY(Label, nblocks+1);
for (uint i = 0; i <= nblocks; i++) {
blk_labels[i].init();
}
// Now fill in the code buffer
Node* delay_slot = NULL;
for (uint i = 0; i < nblocks; i++) {
Block* block = C->cfg()->get_block(i);
_block = block;
Node* head = block->head();
// If this block needs to start aligned (i.e, can be reached other
// than by falling-thru from the previous block), then force the
// start of a new bundle.
if (Pipeline::requires_bundling() && starts_bundle(head)) {
cb->flush_bundle(true);
}
#ifdef ASSERT
if (!block->is_connector()) {
stringStream st;
block->dump_head(C->cfg(), &st);
MacroAssembler(cb).block_comment(st.freeze());
}
jmp_target[i] = 0;
jmp_offset[i] = 0;
jmp_size[i] = 0;
jmp_rule[i] = 0;
#endif
int blk_offset = current_offset;
// Define the label at the beginning of the basic block
MacroAssembler(cb).bind(blk_labels[block->_pre_order]);
uint last_inst = block->number_of_nodes();
// Emit block normally, except for last instruction.
// Emit means "dump code bits into code buffer".
for (uint j = 0; j<last_inst; j++) {
_index = j;
// Get the node
Node* n = block->get_node(j);
// See if delay slots are supported
if (valid_bundle_info(n) && node_bundling(n)->used_in_unconditional_delay()) {
assert(delay_slot == NULL, "no use of delay slot node");
assert(n->size(C->regalloc()) == Pipeline::instr_unit_size(), "delay slot instruction wrong size");
delay_slot = n;
continue;
}
// If this starts a new instruction group, then flush the current one
// (but allow split bundles)
if (Pipeline::requires_bundling() && starts_bundle(n))
cb->flush_bundle(false);
// Special handling for SafePoint/Call Nodes
bool is_mcall = false;
if (n->is_Mach()) {
MachNode *mach = n->as_Mach();
is_mcall = n->is_MachCall();
bool is_sfn = n->is_MachSafePoint();
// If this requires all previous instructions be flushed, then do so
if (is_sfn || is_mcall || mach->alignment_required() != 1) {
cb->flush_bundle(true);
current_offset = cb->insts_size();
}
// A padding may be needed again since a previous instruction
// could be moved to delay slot.
// align the instruction if necessary
int padding = mach->compute_padding(current_offset);
// Make sure safepoint node for polling is distinct from a call's
// return by adding a nop if needed.
if (is_sfn && !is_mcall && padding == 0 && current_offset == last_call_offset) {
padding = nop_size;
}
if (padding == 0 && mach->avoid_back_to_back(MachNode::AVOID_BEFORE) &&
current_offset == last_avoid_back_to_back_offset) {
// Avoid back to back some instructions.
padding = nop_size;
}
if (padding > 0) {
assert((padding % nop_size) == 0, "padding is not a multiple of NOP size");
int nops_cnt = padding / nop_size;
MachNode *nop = new MachNopNode(nops_cnt);
block->insert_node(nop, j++);
last_inst++;
C->cfg()->map_node_to_block(nop, block);
// Ensure enough space.
cb->insts()->maybe_expand_to_ensure_remaining(MAX_inst_size);
if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) {
C->record_failure("CodeCache is full");
return;
}
nop->emit(*cb, C->regalloc());
cb->flush_bundle(true);
current_offset = cb->insts_size();
}
bool observe_safepoint = is_sfn;
// Remember the start of the last call in a basic block
if (is_mcall) {
MachCallNode *mcall = mach->as_MachCall();
// This destination address is NOT PC-relative
mcall->method_set((intptr_t)mcall->entry_point());
// Save the return address
call_returns[block->_pre_order] = current_offset + mcall->ret_addr_offset();
observe_safepoint = mcall->guaranteed_safepoint();
}
// sfn will be valid whenever mcall is valid now because of inheritance
if (observe_safepoint) {
// Handle special safepoint nodes for synchronization
if (!is_mcall) {
MachSafePointNode *sfn = mach->as_MachSafePoint();
// !!!!! Stubs only need an oopmap right now, so bail out
if (sfn->jvms()->method() == NULL) {
// Write the oopmap directly to the code blob??!!
continue;
}
} // End synchronization
non_safepoints.observe_safepoint(mach->as_MachSafePoint()->jvms(),
current_offset);
Process_OopMap_Node(mach, current_offset);
} // End if safepoint
// If this is a null check, then add the start of the previous instruction to the list
else if( mach->is_MachNullCheck() ) {
inct_starts[inct_cnt++] = previous_offset;
}
// If this is a branch, then fill in the label with the target BB's label
else if (mach->is_MachBranch()) {
// This requires the TRUE branch target be in succs[0]
uint block_num = block->non_connector_successor(0)->_pre_order;
// Try to replace long branch if delay slot is not used,
// it is mostly for back branches since forward branch's
// distance is not updated yet.
bool delay_slot_is_used = valid_bundle_info(n) &&
C->output()->node_bundling(n)->use_unconditional_delay();
if (!delay_slot_is_used && mach->may_be_short_branch()) {
assert(delay_slot == NULL, "not expecting delay slot node");
int br_size = n->size(C->regalloc());
int offset = blk_starts[block_num] - current_offset;
if (block_num >= i) {
// Current and following block's offset are not
// finalized yet, adjust distance by the difference
// between calculated and final offsets of current block.
offset -= (blk_starts[i] - blk_offset);
}
// In the following code a nop could be inserted before
// the branch which will increase the backward distance.
bool needs_padding = (current_offset == last_avoid_back_to_back_offset);
if (needs_padding && offset <= 0)
offset -= nop_size;
if (C->matcher()->is_short_branch_offset(mach->rule(), br_size, offset)) {
// We've got a winner. Replace this branch.
MachNode* replacement = mach->as_MachBranch()->short_branch_version();
// Update the jmp_size.
int new_size = replacement->size(C->regalloc());
assert((br_size - new_size) >= (int)nop_size, "short_branch size should be smaller");
// Insert padding between avoid_back_to_back branches.
if (needs_padding && replacement->avoid_back_to_back(MachNode::AVOID_BEFORE)) {
MachNode *nop = new MachNopNode();
block->insert_node(nop, j++);
C->cfg()->map_node_to_block(nop, block);
last_inst++;
nop->emit(*cb, C->regalloc());
cb->flush_bundle(true);
current_offset = cb->insts_size();
}
#ifdef ASSERT
jmp_target[i] = block_num;
jmp_offset[i] = current_offset - blk_offset;
jmp_size[i] = new_size;
jmp_rule[i] = mach->rule();
#endif
block->map_node(replacement, j);
mach->subsume_by(replacement, C);
n = replacement;
mach = replacement;
}
}
mach->as_MachBranch()->label_set( &blk_labels[block_num], block_num );
} else if (mach->ideal_Opcode() == Op_Jump) {
for (uint h = 0; h < block->_num_succs; h++) {
Block* succs_block = block->_succs[h];
for (uint j = 1; j < succs_block->num_preds(); j++) {
Node* jpn = succs_block->pred(j);
if (jpn->is_JumpProj() && jpn->in(0) == mach) {
uint block_num = succs_block->non_connector()->_pre_order;
Label *blkLabel = &blk_labels[block_num];
mach->add_case_label(jpn->as_JumpProj()->proj_no(), blkLabel);
}
}
}
}
#ifdef ASSERT
// Check that oop-store precedes the card-mark
else if (mach->ideal_Opcode() == Op_StoreCM) {
uint storeCM_idx = j;
int count = 0;
for (uint prec = mach->req(); prec < mach->len(); prec++) {
Node *oop_store = mach->in(prec); // Precedence edge
if (oop_store == NULL) continue;
count++;
uint i4;
for (i4 = 0; i4 < last_inst; ++i4) {
if (block->get_node(i4) == oop_store) {
break;
}
}
// Note: This test can provide a false failure if other precedence
// edges have been added to the storeCMNode.
assert(i4 == last_inst || i4 < storeCM_idx, "CM card-mark executes before oop-store");
}
assert(count > 0, "storeCM expects at least one precedence edge");
}
#endif
else if (!n->is_Proj()) {
// Remember the beginning of the previous instruction, in case
// it's followed by a flag-kill and a null-check. Happens on
// Intel all the time, with add-to-memory kind of opcodes.
previous_offset = current_offset;
}
// Not an else-if!
// If this is a trap based cmp then add its offset to the list.
if (mach->is_TrapBasedCheckNode()) {
inct_starts[inct_cnt++] = current_offset;
}
}
// Verify that there is sufficient space remaining
cb->insts()->maybe_expand_to_ensure_remaining(MAX_inst_size);
if ((cb->blob() == NULL) || (!CompileBroker::should_compile_new_jobs())) {
C->record_failure("CodeCache is full");
return;
}
// Save the offset for the listing
#if defined(SUPPORT_OPTO_ASSEMBLY)
if ((node_offsets != NULL) && (n->_idx < node_offset_limit)) {
node_offsets[n->_idx] = cb->insts_size();
}
#endif
assert(!C->failing(), "Should not reach here if failing.");
// "Normal" instruction case
DEBUG_ONLY(uint instr_offset = cb->insts_size());
n->emit(*cb, C->regalloc());
current_offset = cb->insts_size();
--> --------------------
--> maximum size reached
--> --------------------
¤ Dauer der Verarbeitung: 0.78 Sekunden
(vorverarbeitet)
¤
|
Haftungshinweis
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung ist noch experimentell.
|
