gecko/js/src/nanojit/Fragmento.cpp

606 lines
17 KiB
C++

/* -*- Mode: C++; c-basic-offset: 4; indent-tabs-mode: t; tab-width: 4 -*- */
/* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla Public License Version
* 1.1 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
* http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis,
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
* for the specific language governing rights and limitations under the
* License.
*
* The Original Code is [Open Source Virtual Machine].
*
* The Initial Developer of the Original Code is
* Adobe System Incorporated.
* Portions created by the Initial Developer are Copyright (C) 2004-2007
* the Initial Developer. All Rights Reserved.
*
* Contributor(s):
* Adobe AS3 Team
* Mozilla TraceMonkey Team
* Asko Tontti <atontti@cc.hut.fi>
*
* Alternatively, the contents of this file may be used under the terms of
* either the GNU General Public License Version 2 or later (the "GPL"), or
* the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
* in which case the provisions of the GPL or the LGPL are applicable instead
* of those above. If you wish to allow use of your version of this file only
* under the terms of either the GPL or the LGPL, and not to allow others to
* use your version of this file under the terms of the MPL, indicate your
* decision by deleting the provisions above and replace them with the notice
* and other provisions required by the GPL or the LGPL. If you do not delete
* the provisions above, a recipient may use your version of this file under
* the terms of any one of the MPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
#include "nanojit.h"
#undef MEMORY_INFO
namespace nanojit
{
#ifdef FEATURE_NANOJIT
using namespace avmplus;
static uint32_t calcSaneCacheSize(uint32_t in)
{
if (in < uint32_t(NJ_LOG2_PAGE_SIZE)) return NJ_LOG2_PAGE_SIZE; // at least 1 page
if (in > 30) return 30; // 1GB should be enough for anyone
return in;
}
/**
* This is the main control center for creating and managing fragments.
*/
Fragmento::Fragmento(AvmCore* core, uint32_t cacheSizeLog2)
: _frags(core->GetGC()),
_freePages(core->GetGC(), 1024),
_allocList(core->GetGC()),
_max_pages(1 << (calcSaneCacheSize(cacheSizeLog2) - NJ_LOG2_PAGE_SIZE)),
_pagesGrowth(1)
{
#ifdef _DEBUG
{
// XXX These belong somewhere else, but I can't find the
// right location right now.
NanoStaticAssert((LIR_lt ^ 3) == LIR_ge);
NanoStaticAssert((LIR_le ^ 3) == LIR_gt);
NanoStaticAssert((LIR_ult ^ 3) == LIR_uge);
NanoStaticAssert((LIR_ule ^ 3) == LIR_ugt);
NanoStaticAssert((LIR_flt ^ 3) == LIR_fge);
NanoStaticAssert((LIR_fle ^ 3) == LIR_fgt);
/* Opcodes must be strictly increasing without holes. */
uint32_t count = 0;
#define OPDEF(op, number, operands) \
NanoAssertMsg(LIR_##op == count++, "misnumbered opcode");
#define OPDEF64(op, number, operands) OPDEF(op, number, operands)
#include "LIRopcode.tbl"
#undef OPDEF
#undef OPDEF64
}
#endif
#ifdef MEMORY_INFO
_allocList.set_meminfo_name("Fragmento._allocList");
#endif
NanoAssert(_max_pages > _pagesGrowth); // shrink growth if needed
_core = core;
GC *gc = core->GetGC();
_assm = NJ_NEW(gc, nanojit::Assembler)(this);
verbose_only( enterCounts = NJ_NEW(gc, BlockHist)(gc); )
verbose_only( mergeCounts = NJ_NEW(gc, BlockHist)(gc); )
}
Fragmento::~Fragmento()
{
AllocEntry *entry;
clearFrags();
_frags.clear();
_freePages.clear();
while( _allocList.size() > 0 )
{
//fprintf(stderr,"dealloc %x\n", (intptr_t)_allocList.get(_allocList.size()-1));
#ifdef MEMORY_INFO
ChangeSizeExplicit("NanoJitMem", -1, _gcHeap->Size(_allocList.last()));
#endif
entry = _allocList.removeLast();
_gcHeap->Free( entry->page, entry->allocSize );
NJ_DELETE(entry);
}
NJ_DELETE(_assm);
#if defined(NJ_VERBOSE)
NJ_DELETE(enterCounts);
NJ_DELETE(mergeCounts);
#endif
}
void Fragmento::trackPages()
{
const uint32_t pageUse = _stats.pages - _freePages.size();
if (_stats.maxPageUse < pageUse)
_stats.maxPageUse = pageUse;
}
Page* Fragmento::pageAlloc()
{
NanoAssert(sizeof(Page) == NJ_PAGE_SIZE);
if (!_freePages.size())
pagesGrow(_pagesGrowth); // try to get more mem
if ((_pagesGrowth << 1) < _max_pages)
_pagesGrowth <<= 1;
trackPages();
Page* page = 0;
if (_freePages.size())
page = _freePages.removeLast();
return page;
}
void Fragmento::pagesRelease(PageList& l)
{
_freePages.add(l);
l.clear();
NanoAssert(_freePages.size() <= _stats.pages);
}
void Fragmento::pageFree(Page* page)
{
_freePages.add(page);
NanoAssert(_freePages.size() <= _stats.pages);
}
void Fragmento::pagesGrow(int32_t count)
{
NanoAssert(!_freePages.size());
MMGC_MEM_TYPE("NanojitFragmentoMem");
Page* memory = 0;
GC *gc = _core->GetGC();
if (_stats.pages < _max_pages)
{
AllocEntry *entry;
// make sure we don't grow beyond _max_pages
if (_stats.pages + count > _max_pages)
count = _max_pages - _stats.pages;
if (count < 0)
count = 0;
// @todo nastiness that needs a fix'n
_gcHeap = gc->GetGCHeap();
NanoAssert(int32_t(NJ_PAGE_SIZE)<=_gcHeap->kNativePageSize);
// convert _max_pages to gc page count
int32_t gcpages = (count*NJ_PAGE_SIZE) / _gcHeap->kNativePageSize;
MMGC_MEM_TYPE("NanoJitMem");
memory = (Page*)_gcHeap->Alloc(gcpages);
#ifdef MEMORY_INFO
ChangeSizeExplicit("NanoJitMem", 1, _gcHeap->Size(memory));
#endif
NanoAssert((int*)memory == pageTop(memory));
//fprintf(stderr,"head alloc of %d at %x of %d pages using nj page size of %d\n", gcpages, (intptr_t)memory, (intptr_t)_gcHeap->kNativePageSize, NJ_PAGE_SIZE);
entry = NJ_NEW(gc, AllocEntry);
entry->page = memory;
entry->allocSize = gcpages;
_allocList.add(entry);
_stats.pages += count;
Page* page = memory;
while(--count >= 0)
{
//fprintf(stderr,"Fragmento::pageGrow adding page %x ; %d\n", (unsigned)page, _freePages.size()+1);
_freePages.add(page++);
}
trackPages();
}
}
// Clear the fragment. This *does not* remove the fragment from the
// map--the caller must take care of this.
void Fragmento::clearFragment(Fragment* f)
{
Fragment *peer = f->peer;
while (peer) {
Fragment *next = peer->peer;
peer->releaseTreeMem(this);
NJ_DELETE(peer);
peer = next;
}
f->releaseTreeMem(this);
NJ_DELETE(f);
}
void Fragmento::clearFrag(const void* ip)
{
if (_frags.containsKey(ip)) {
clearFragment(_frags.remove(ip));
}
}
void Fragmento::clearFrags()
{
// reclaim any dangling native pages
_assm->pageReset();
while (!_frags.isEmpty()) {
clearFragment(_frags.removeLast());
}
verbose_only( enterCounts->clear();)
verbose_only( mergeCounts->clear();)
verbose_only( _stats.flushes++ );
verbose_only( _stats.compiles = 0 );
//fprintf(stderr, "Fragmento.clearFrags %d free pages of %d\n", _stats.freePages, _stats.pages);
}
Assembler* Fragmento::assm()
{
return _assm;
}
AvmCore* Fragmento::core()
{
return _core;
}
Fragment* Fragmento::getAnchor(const void* ip)
{
Fragment *f = newFrag(ip);
Fragment *p = _frags.get(ip);
if (p) {
f->first = p;
/* append at the end of the peer list */
Fragment* next;
while ((next = p->peer) != NULL)
p = next;
p->peer = f;
} else {
f->first = f;
_frags.put(ip, f); /* this is the first fragment */
}
f->anchor = f;
f->root = f;
f->kind = LoopTrace;
verbose_only( addLabel(f, "T", _frags.size()); )
return f;
}
Fragment* Fragmento::getLoop(const void* ip)
{
return _frags.get(ip);
}
#ifdef NJ_VERBOSE
void Fragmento::addLabel(Fragment *f, const char *prefix, int id)
{
char fragname[20];
sprintf(fragname,"%s%d", prefix, id);
labels->add(f, sizeof(Fragment), 0, fragname);
}
#endif
Fragment *Fragmento::getMerge(GuardRecord *lr, const void* ip)
{
Fragment *anchor = lr->exit->from->anchor;
for (Fragment *f = anchor->branches; f != 0; f = f->nextbranch) {
if (f->kind == MergeTrace && f->ip == ip /*&& f->calldepth == lr->calldepth*/) {
// found existing shared branch on anchor
return f;
}
}
Fragment *f = newBranch(anchor, ip);
f->root = f;
f->kind = MergeTrace;
verbose_only(
int mergeid = 1;
for (Fragment *g = anchor->branches; g != 0; g = g->nextbranch)
if (g->kind == MergeTrace)
mergeid++;
addLabel(f, "M", mergeid);
)
return f;
}
Fragment *Fragmento::createBranch(SideExit* exit, const void* ip)
{
Fragment *f = newBranch(exit->from, ip);
f->kind = BranchTrace;
f->treeBranches = f->root->treeBranches;
f->root->treeBranches = f;
return f;
}
#ifdef NJ_VERBOSE
struct fragstats {
int size;
uint64_t traceDur;
uint64_t interpDur;
int lir, lirbytes;
};
void Fragmento::dumpFragStats(Fragment *f, int level, fragstats &stat)
{
char buf[50];
sprintf(buf, "%*c%s", 1+level, ' ', labels->format(f));
int called = f->hits();
if (called >= 0)
called += f->_called;
else
called = -(1<<f->blacklistLevel) - called - 1;
uint32_t main = f->_native - f->_exitNative;
char cause[200];
if (f->_token && strcmp(f->_token,"loop")==0)
sprintf(cause,"%s %d", f->_token, f->xjumpCount);
else if (f->_token) {
if (f->eot_target) {
sprintf(cause,"%s %s", f->_token, labels->format(f->eot_target));
} else {
strcpy(cause, f->_token);
}
}
else
cause[0] = 0;
_assm->outputf("%-10s %7d %6d %6d %6d %4d %9llu %9llu %-12s %s", buf,
called, f->guardCount, main, f->_native, f->compileNbr, f->traceTicks/1000, f->interpTicks/1000,
cause, labels->format(f->ip));
stat.size += main;
stat.traceDur += f->traceTicks;
stat.interpDur += f->interpTicks;
stat.lir += f->_lir;
stat.lirbytes += f->_lirbytes;
for (Fragment *x = f->branches; x != 0; x = x->nextbranch)
if (x->kind != MergeTrace)
dumpFragStats(x,level+1,stat);
for (Fragment *x = f->branches; x != 0; x = x->nextbranch)
if (x->kind == MergeTrace)
dumpFragStats(x,level+1,stat);
if (f->isAnchor() && f->branches != 0) {
_assm->output("");
}
}
class DurData { public:
DurData(): frag(0), traceDur(0), interpDur(0), size(0) {}
DurData(int): frag(0), traceDur(0), interpDur(0), size(0) {}
DurData(Fragment* f, uint64_t td, uint64_t id, int32_t s)
: frag(f), traceDur(td), interpDur(id), size(s) {}
Fragment* frag;
uint64_t traceDur;
uint64_t interpDur;
int32_t size;
};
void Fragmento::dumpRatio(const char *label, BlockHist *hist)
{
int total=0, unique=0;
for (int i = 0, n=hist->size(); i < n; i++) {
const void * id = hist->keyAt(i);
int c = hist->get(id);
if (c > 1) {
//_assm->outputf("%d %X", c, id);
unique += 1;
}
else if (c == 1) {
unique += 1;
}
total += c;
}
_assm->outputf("%s total %d unique %d ratio %.1f%", label, total, unique, double(total)/unique);
}
void Fragmento::dumpStats()
{
bool vsave = _assm->_verbose;
_assm->_verbose = true;
_assm->output("");
dumpRatio("inline", enterCounts);
dumpRatio("merges", mergeCounts);
_assm->outputf("abc %d il %d (%.1fx) abc+il %d (%.1fx)",
_stats.abcsize, _stats.ilsize, (double)_stats.ilsize/_stats.abcsize,
_stats.abcsize + _stats.ilsize,
double(_stats.abcsize+_stats.ilsize)/_stats.abcsize);
int32_t count = _frags.size();
int32_t pages = _stats.pages;
int32_t maxPageUse = _stats.maxPageUse;
int32_t free = _freePages.size();
int32_t flushes = _stats.flushes;
if (!count)
{
_assm->outputf("No fragments in cache, %d flushes", flushes);
_assm->_verbose = vsave;
return;
}
_assm->outputf("\nFragment statistics");
_assm->outputf(" loop trees: %d", count);
_assm->outputf(" flushes: %d", flushes);
_assm->outputf(" compiles: %d / %d", _stats.compiles, _stats.totalCompiles);
_assm->outputf(" used: %dk / %dk", (pages-free)<<(NJ_LOG2_PAGE_SIZE-10), pages<<(NJ_LOG2_PAGE_SIZE-10));
_assm->outputf(" maxPageUse: %dk", (maxPageUse)<<(NJ_LOG2_PAGE_SIZE-10));
_assm->output("\ntrace calls guards main native gen T-trace T-interp");
avmplus::SortedMap<uint64_t, DurData, avmplus::LIST_NonGCObjects> durs(_core->gc);
uint64_t totaldur=0;
fragstats totalstat = { 0,0,0,0,0 };
for (int32_t i=0; i<count; i++)
{
Fragment *f = _frags.at(i);
while (true) {
fragstats stat = { 0,0,0,0,0 };
dumpFragStats(f, 0, stat);
if (stat.lir) {
totalstat.lir += stat.lir;
totalstat.lirbytes += stat.lirbytes;
}
uint64_t bothDur = stat.traceDur + stat.interpDur;
if (bothDur) {
totalstat.interpDur += stat.interpDur;
totalstat.traceDur += stat.traceDur;
totalstat.size += stat.size;
totaldur += bothDur;
while (durs.containsKey(bothDur)) bothDur++;
DurData d(f, stat.traceDur, stat.interpDur, stat.size);
durs.put(bothDur, d);
}
if (!f->peer)
break;
f = f->peer;
}
}
uint64_t totaltrace = totalstat.traceDur;
int totalsize = totalstat.size;
_assm->outputf("");
_assm->outputf("lirbytes %d / lir %d = %.1f bytes/lir", totalstat.lirbytes,
totalstat.lir, double(totalstat.lirbytes)/totalstat.lir);
_assm->outputf(" trace interp");
_assm->outputf("%9lld (%2d%%) %9lld (%2d%%)",
totaltrace/1000, int(100.0*totaltrace/totaldur),
(totaldur-totaltrace)/1000, int(100.0*(totaldur-totaltrace)/totaldur));
_assm->outputf("");
_assm->outputf("trace ticks trace interp size");
for (int32_t i=durs.size()-1; i >= 0; i--) {
uint64_t bothDur = durs.keyAt(i);
DurData d = durs.get(bothDur);
int size = d.size;
_assm->outputf("%-4s %9lld (%2d%%) %9lld (%2d%%) %9lld (%2d%%) %6d (%2d%%) %s",
labels->format(d.frag),
bothDur/1000, int(100.0*bothDur/totaldur),
d.traceDur/1000, int(100.0*d.traceDur/totaldur),
d.interpDur/1000, int(100.0*d.interpDur/totaldur),
size, int(100.0*size/totalsize),
labels->format(d.frag->ip));
}
_assm->_verbose = vsave;
}
void Fragmento::countBlock(BlockHist *hist, const void* ip)
{
int c = hist->count(ip);
if (_assm->_verbose)
_assm->outputf("++ %s %d", labels->format(ip), c);
}
void Fragmento::countIL(uint32_t il, uint32_t abc)
{
_stats.ilsize += il;
_stats.abcsize += abc;
}
#ifdef AVMPLUS_VERBOSE
void Fragmento::drawTrees(char *fileName) {
drawTraceTrees(this, this->_frags, this->_core, fileName);
}
#endif
#endif // NJ_VERBOSE
//
// Fragment
//
Fragment::Fragment(const void* _ip) : ip(_ip)
{
// Fragment is a gc object which is zero'd by the GC, no need to clear fields
}
Fragment::~Fragment()
{
onDestroy();
NanoAssert(_pages == 0);
}
void Fragment::resetHits()
{
blacklistLevel >>= 1;
_hits = 0;
}
void Fragment::blacklist()
{
blacklistLevel++;
_hits = -(1<<blacklistLevel);
}
Fragment *Fragmento::newFrag(const void* ip)
{
GC *gc = _core->gc;
Fragment *f = NJ_NEW(gc, Fragment)(ip);
f->blacklistLevel = 5;
f->recordAttempts = 0;
return f;
}
Fragment *Fragmento::newBranch(Fragment *from, const void* ip)
{
Fragment *f = newFrag(ip);
f->anchor = from->anchor;
f->root = from->root;
f->xjumpCount = from->xjumpCount;
/*// prepend
f->nextbranch = from->branches;
from->branches = f;*/
// append
if (!from->branches) {
from->branches = f;
} else {
Fragment *p = from->branches;
while (p->nextbranch != 0)
p = p->nextbranch;
p->nextbranch = f;
}
return f;
}
void Fragment::releaseLirBuffer()
{
lastIns = 0;
}
void Fragment::releaseCode(Fragmento* frago)
{
_code = 0;
while(_pages)
{
Page* next = _pages->next;
frago->pageFree(_pages);
_pages = next;
}
}
void Fragment::releaseTreeMem(Fragmento* frago)
{
releaseLirBuffer();
releaseCode(frago);
// now do it for all branches
Fragment* branch = branches;
while(branch)
{
Fragment* next = branch->nextbranch;
branch->releaseTreeMem(frago); // @todo safer here to recurse in case we support nested trees
NJ_DELETE(branch);
branch = next;
}
}
#endif /* FEATURE_NANOJIT */
}