static int
OPTallConstant(Client cntxt, MalBlkPtr mb, InstrPtr p)
{
int i;
(void)cntxt;
if ( !(p->token == ASSIGNsymbol ||
getModuleId(p) == calcRef ||
getModuleId(p) == strRef ||
getModuleId(p) == mtimeRef ||
getModuleId(p) == mmathRef))
return FALSE;
if (getModuleId(p) == mmathRef && strcmp(getFunctionId(p), "rand") == 0)
return FALSE;
for (i = p->retc; i < p->argc; i++)
if (isVarConstant(mb, getArg(p, i)) == FALSE)
return FALSE;
for (i = 0; i < p->retc; i++) {
if (isaBatType(getArgType(mb, p, i)))
return FALSE;
if ( mb->unsafeProp )
return FALSE;
}
return TRUE;
}
/*
* Removal of elements from the symbol table should be
* done with care. For, it should be assured that
* there are no references to the definition at the
* moment of removal. This situation can not easily
* checked at runtime, without tremendous overhead.
*/
void deleteSymbol(Module scope, Symbol prg){
InstrPtr sig;
int t;
sig = getSignature(prg);
#ifdef _DEBUG_MODULE_
fprintf(stderr,"#delete symbol %s.%s from %s\n", getModuleId(sig), getFunctionId(sig), prg->name);
#endif
if (getModuleId(sig) && getModuleId(sig)!= scope->name ){
/* move the definition to the proper place */
/* default scope is the last resort */
Module c= findModule(scope, getModuleId(sig));
if(c )
scope = c;
}
t = getSymbolIndex(getFunctionId(sig));
if (scope->space[t] == prg) {
scope->space[t] = scope->space[t]->peer;
freeSymbol(prg);
} else {
Symbol nxt = scope->space[t];
while (nxt->peer != NULL) {
if (nxt->peer == prg) {
nxt->peer = prg->peer;
nxt->skip = prg->peer;
freeSymbol(prg);
return;
}
nxt = nxt->peer;
}
}
}
/*
* For clarity we show the last optimizer applied
* also as the last of the list, although it is linked with mb.
*/
void showMalBlkHistory(stream *out, MalBlkPtr mb)
{
MalBlkPtr m=mb;
InstrPtr p,sig;
int j=0;
str msg;
sig = getInstrPtr(mb,0);
m= m->history;
while(m){
p= getInstrPtr(m,m->stop-1);
if( p->token == REMsymbol){
msg= instruction2str(m, 0, p, FALSE);
if (msg ) {
mnstr_printf(out,"%s.%s[%2d] %s\n",
getModuleId(sig), getFunctionId(sig),j++,msg+3);
GDKfree(msg);
}
}
m= m->history;
}
p=getInstrPtr(mb,mb->stop-1);
if( p->token == REMsymbol){
msg= instruction2str(mb, 0, p, FALSE);
if (msg) {
mnstr_printf(out,"%s.%s[%2d] %s\n",
getModuleId(sig), getFunctionId(sig),j++,msg+3);
GDKfree(msg);
}
}
}
static int
malMatch(InstrPtr p1, InstrPtr p2)
{
int i, j;
if (getFunctionId(p1) == 0 && getFunctionId(p2) != 0)
return 0;
if (getModuleId(p1) == 0 && getModuleId(p2) != 0)
return 0;
if (getModuleId(p1) != getModuleId(p2))
return 0;
if (getFunctionId(p2) == 0)
return 0;
if (getFunctionId(p1) != getFunctionId(p2))
return 0;
if (p1->retc != p2->retc)
return 0;
if (p1->argc != p2->argc)
return 0;
if (p1->barrier != p2->barrier)
return 0;
for (i = 0; i < p1->argc; i++)
for (j = i + 1; j < p1->argc; j++)
if ((getArg(p1, i) == getArg(p1, j)) != (getArg(p2, i) == getArg(p2, j)))
return 0;
return 1;
}
int
OPTgroupsImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr p)
{
int i, actions=0;
InstrPtr q;
InstrPtr *old, *ref;
int limit,slimit;
(void) cntxt;
(void) stk;
if (varGetProp(mb, getArg(mb->stmt[0], 0), inlineProp) != NULL) {
return 0;
}
/* beware, new variables and instructions are introduced */
ref= (InstrPtr*) GDKzalloc(sizeof(InstrPtr) * mb->vtop); /* to find last assignment */
if ( ref == NULL) {
return 0;
}
old= mb->stmt;
limit= mb->stop;
slimit= mb->ssize;
if ( newMalBlkStmt(mb,mb->ssize) <0) {
GDKfree(ref);
return 0;
}
for (i = 0; i<limit; i++){
p= old[i];
if (getModuleId(p) == groupRef && p->argc == 4 && getFunctionId(p) == subgroupRef ){
setFunctionId(p, multicolumnsRef);
ref[getArg(p,0)] = p;
actions++;
OPTDEBUGgroups {
mnstr_printf(cntxt->fdout,"#new groups instruction\n");
printInstruction(cntxt->fdout,mb, 0, p, LIST_MAL_ALL);
}
}
if (getModuleId(p) == groupRef && p->argc == 5 && getFunctionId(p) == subgroupdoneRef && ref[getArg(p,4)] != NULL){
/*
* Try to expand its argument list with what we have found so far.
* This creates a series of derive paths, many of which will be removed during deadcode elimination.
*/
q= copyInstruction(ref[getArg(p,4)]);
q= pushArgument(mb, q, getArg(p,3));
getArg(q,0) = getArg(p,0);
getArg(q,1) = getArg(p,1);
getArg(q,2) = getArg(p,2);
ref[getArg(q,0)] = q;
freeInstruction(p);
p= q;
OPTDEBUGgroups{
mnstr_printf(cntxt->fdout,"#new groups instruction extension\n");
printInstruction(cntxt->fdout,mb, 0, p, LIST_MAL_ALL);
}
}
int
OPTjsonImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i, j, limit, slimit;
int bu = 0, br = 0, bj = 0;
str nme;
InstrPtr p,q;
int actions = 0;
InstrPtr *old;
(void) pci;
(void) cntxt;
(void) stk; /* to fool compilers */
old= mb->stmt;
limit= mb->stop;
slimit = mb->ssize;
if ( newMalBlkStmt(mb,mb->stop) < 0)
return 0;
for (i = 0; i < limit; i++) {
p = old[i];
if( getModuleId(p) == sqlRef && getFunctionId(p) == affectedRowsRef) {
q = newStmt(mb, jsonRef, resultSetRef);
q = pushArgument(mb, q, bu);
q = pushArgument(mb, q, br);
q = pushArgument(mb, q, bj);
j = getArg(q,0);
p= getInstrPtr(mb,0);
setVarType(mb,getArg(p,0),TYPE_str);
q = newReturnStmt(mb);
getArg(q,0)= getArg(p,0);
pushArgument(mb,q,j);
continue;
}
if( getModuleId(p) == sqlRef && getFunctionId(p) == rsColumnRef) {
nme = getVarConstant(mb,getArg(p,4)).val.sval;
if (strcmp(nme,"uuid")==0)
bu = getArg(p,7);
if (strcmp(nme,"lng")==0)
br = getArg(p,7);
if (strcmp(nme,"json")==0)
bj = getArg(p,7);
freeInstruction(p);
continue;
}
pushInstruction(mb,p);
}
for(; i<slimit; i++)
if (old[i])
freeInstruction(old[i]);
GDKfree(old);
return actions;
}
int
OPTinlineImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr p)
{
int i;
InstrPtr q,sig;
int actions = 0;
(void) p;
(void)stk;
for (i = 1; i < mb->stop; i++) {
q = getInstrPtr(mb, i);
if( q->blk ) {
sig = getInstrPtr(q->blk,0);
/*
* Time for inlining functions that are used in multiplex operations.
* They are produced by SQL compiler.
*/
if( getFunctionId(q)== multiplexRef &&
getModuleId(q) == malRef &&
OPTinlineMultiplex(cntxt,mb,q)) {
OPTDEBUGinline {
mnstr_printf(cntxt->fdout,"#multiplex inline function\n");
printInstruction(cntxt->fdout,mb,0,q,LIST_MAL_ALL);
}
varSetProp(mb, getArg(q,0), inlineProp, op_eq, NULL);
} else
/*
* Check if the function definition is tagged as being inlined.
*/
if (sig->token == FUNCTIONsymbol &&
/*
* After filling in a structure it is added to the multi-level symbol
* table. We keep a skip list of similarly named function symbols.
* This speeds up searching provided the modules adhere to the
* structure and group the functions as well.
*/
void insertSymbol(Module scope, Symbol prg){
InstrPtr sig;
int t;
Module c;
assert(scope);
sig = getSignature(prg);
#ifdef _DEBUG_MODULE_
fprintf(stderr,"#insertSymbol: %s.%s in %s ", getModuleId(sig), getFunctionId(sig), scope->name);
#endif
if(getModuleId(sig) && getModuleId(sig)!= scope->name){
/* move the definition to the proper place */
/* default scope is the last resort */
c= findModule(scope,getModuleId(sig));
if ( c )
scope = c;
#ifdef _DEBUG_MODULE_
fprintf(stderr," found alternative module %s ", scope->name);
#endif
}
t = getSymbolIndex(getFunctionId(sig));
if( scope->space == NULL) {
scope->space = (Symbol *) GDKzalloc(MAXSCOPE * sizeof(Symbol));
if (scope->space == NULL)
return;
}
assert(scope->space);
if (scope->space[t] == prg){
/* already known, last inserted */
#ifdef _DEBUG_MODULE_
fprintf(stderr," unexpected double insert ");
#endif
} else {
prg->peer= scope->space[t];
scope->space[t] = prg;
if( prg->peer &&
idcmp(prg->name,prg->peer->name) == 0)
prg->skip = prg->peer->skip;
else
prg->skip = prg->peer;
}
assert(prg != prg->peer);
#ifdef _DEBUG_MODULE_
fprintf(stderr,"\n");
#endif
}
static int
OPTallConstant(Client cntxt, MalBlkPtr mb, InstrPtr p)
{
int i;
(void)cntxt;
if ( !( p->token == ASSIGNsymbol ||
getModuleId(p) == calcRef ||
getModuleId(p) == strRef ||
getModuleId(p) == mmathRef ))
return FALSE;
for (i = p->retc; i < p->argc; i++)
if (isVarConstant(mb, getArg(p, i)) == FALSE)
return FALSE;
for (i = 0; i < p->retc; i++)
if (isaBatType(getArgType(mb, p, i)))
return FALSE;
return p->argc != p->retc;
}
int
OPTmultiplexImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
InstrPtr *old, p;
int i, limit, slimit, actions= 0;
str msg= MAL_SUCCEED;
(void) stk;
(void) pci;
old = mb->stmt;
limit = mb->stop;
slimit = mb->ssize;
if ( newMalBlkStmt(mb, mb->ssize) < 0 )
return 0;
for (i = 0; i < limit; i++) {
p = old[i];
if (msg == MAL_SUCCEED && getModuleId(p) == malRef &&
getFunctionId(p) == multiplexRef) {
if ( MANIFOLDtypecheck(cntxt,mb,p) != NULL){
setFunctionId(p, manifoldRef);
pushInstruction(mb, p);
actions++;
continue;
}
msg = OPTexpandMultiplex(cntxt, mb, stk, p);
if( msg== MAL_SUCCEED){
freeInstruction(p);
old[i]=0;
actions++;
continue;
}
pushInstruction(mb, p);
actions++;
} else if( old[i])
pushInstruction(mb, p);
}
for(;i<slimit; i++)
if( old[i])
freeInstruction(old[i]);
GDKfree(old);
if (mb->errors){
/* rollback */
}
GDKfree(msg);
return mb->errors? 0: actions;
}
static lng
SQLgetSpace(mvc *m, MalBlkPtr mb)
{
sql_trans *tr = m->session->tr;
lng space = 0, i;
for (i = 0; i < mb->stop; i++) {
InstrPtr p = mb->stmt[i];
char *f = getFunctionId(p);
if (getModuleId(p) == sqlRef && (f == bindRef || f == bindidxRef)) {
int upd = (p->argc == 7 || p->argc == 9), mode = 0;
char *sname = getVarConstant(mb, getArg(p, 2 + upd)).val.sval;
char *tname = getVarConstant(mb, getArg(p, 3 + upd)).val.sval;
char *cname = NULL;
sql_schema *s = mvc_bind_schema(m, sname);
if (!s || strcmp(s->base.name, dt_schema) == 0)
continue;
cname = getVarConstant(mb, getArg(p, 4 + upd)).val.sval;
mode = getVarConstant(mb, getArg(p, 5 + upd)).val.ival;
if (mode != 0 || !cname || !s)
continue;
if (f == bindidxRef) {
sql_idx *i = mvc_bind_idx(m, s, cname);
if (i && (!isRemote(i->t) && !isMergeTable(i->t))) {
BAT *b = store_funcs.bind_idx(tr, i, RDONLY);
if (b) {
space += getBatSpace(b);
BBPunfix(b->batCacheid);
}
}
} else if (f == bindRef) {
sql_table *t = mvc_bind_table(m, s, tname);
sql_column *c = mvc_bind_column(m, t, cname);
if (c && (!isRemote(c->t) && !isMergeTable(c->t))) {
BAT *b = store_funcs.bind_col(tr, c, RDONLY);
if (b) {
space += getBatSpace(b);
BBPunfix(b->batCacheid);
}
}
}
}
}
return space;
}
ATerm get_moduleid(int cid, ATerm sid) {
EM_Session session;
EM_ModuleId moduleId;
session = getSession(sid);
if (session == NULL) {
return sndValue(ATmake("no-such-session"));
}
moduleId = getModuleId(sid);
if (moduleId == NULL) {
return sndValue(ATmake("session-not-bound"));
}
else {
return sndValue(ATmake("moduleid(<term>)", moduleId));
}
}
int
OPTmatpackImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int v, i, j, limit, slimit;
InstrPtr p,q;
int actions = 0;
InstrPtr *old;
char *packIncrementRef = putName("packIncrement", 13);
(void) pci;
(void) cntxt;
(void) stk; /* to fool compilers */
old= mb->stmt;
limit= mb->stop;
slimit = mb->ssize;
if ( newMalBlkStmt(mb,mb->stop) < 0)
return 0;
for (i = 0; i < limit; i++) {
p = old[i];
if( getModuleId(p) == matRef && getFunctionId(p) == packRef && isaBatType(getArgType(mb,p,1))) {
q = newStmt(mb, matRef, packIncrementRef);
v = getArg(q,0);
setVarType(mb,v,getArgType(mb,p,1));
q = pushArgument(mb, q, getArg(p,1));
q = pushInt(mb,q, p->argc - p->retc);
for ( j = 2; j < p->argc; j++) {
q = newStmt(mb,matRef, packIncrementRef);
q = pushArgument(mb, q, v);
q = pushArgument(mb, q, getArg(p,j));
setVarType(mb,getArg(q,0),getVarType(mb,v));
v = getArg(q,0);
}
getArg(q,0) = getArg(p,0);
freeInstruction(p);
continue;
}
pushInstruction(mb,p);
}
for(; i<slimit; i++)
if (old[i])
freeInstruction(old[i]);
GDKfree(old);
return actions;
}
ATerm bind_session(int cid, ATerm sid, ATerm moduleId) {
assert(sid != NULL);
assert(moduleId != NULL);
if (getSession(sid) == NULL) {
return sndValue(ATmake("no-such-session"));
}
else {
EM_ModuleId id = getModuleId(sid);
EM_ModuleId module = EM_ModuleIdFromTerm(moduleId);
if (id != NULL && EM_isEqualModuleId(module, id) != 0) {
ATabort("editor-manager:bind_session: attempt to rebind %t (%t)\n",
sid, moduleId);
}
putModuleId(sid, module);
return sndValue(ATmake("session-bound"));
}
}
static void
propagateNonTarget(MalBlkPtr mb, int pc)
{
int i;
InstrPtr p;
str scheduler = putName("scheduler", 9);
for (; pc < mb->stop; pc++) {
p = getInstrPtr(mb, pc);
if (getModuleId(p) == scheduler)
continue;
for (i = 0; i < p->argc; i++)
if (isVarDisabled(mb, getArg(p, i)) && p->token >= 0)
p->token = -p->token; /* temporary NOOP */
if (p->token < 0)
for (i = 0; i < p->retc; i++)
setVarDisabled(mb, getArg(p, i));
}
}
static int
isNewSource(InstrPtr p) {
str mp= getModuleId(p);
if( mp == sqlRef && getFunctionId(p) == bindRef) return 1;
if( mp == calcRef) return 1;
if( mp == batcalcRef) return 1;
if( mp == strRef) return 1;
if( mp == batstrRef) return 1;
if( mp == putName("array",5)) return 1;
if( mp == putName("url",3)) return 1;
if( mp == putName("daytime",7)) return 1;
if( mp == putName("day",3)) return 1;
if( mp == putName("date",4)) return 1;
if( mp == putName("time",4)) return 1;
if( mp == putName("tzone",5)) return 1;
if( mp == putName("color",4)) return 1;
if( mp == putName("batcolor",8)) return 1;
if( mp == putName("blob",4)) return 1;
return 0;
}
/*
* The multiplexSimple is called by the MAL scenario. It bypasses
* the optimizer infrastructure, to avoid excessive space allocation
* and interpretation overhead.
*/
str
OPTmultiplexSimple(Client cntxt)
{
MalBlkPtr mb= cntxt->curprg->def;
int i, doit=0;
InstrPtr p;
if(mb)
for( i=0; i<mb->stop; i++){
p= getInstrPtr(mb,i);
if(getModuleId(p) == malRef && getFunctionId(p) == multiplexRef)
doit++;
}
if( doit) {
OPTmultiplexImplementation(cntxt, mb, 0, 0);
chkTypes(cntxt->fdout, cntxt->nspace, mb,TRUE);
if ( mb->errors == 0) {
chkFlow(cntxt->fdout, mb);
chkDeclarations(cntxt->fdout,mb);
}
}
return 0;
}
int
OPTpushselectImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i, j, limit, slimit, actions=0, *vars, push_down_delta = 0, nr_topn = 0, nr_likes = 0;
InstrPtr p, *old;
subselect_t subselects;
memset(&subselects, 0, sizeof(subselects));
if( mb->errors)
return 0;
OPTDEBUGpushselect
mnstr_printf(cntxt->fdout,"#Push select optimizer started\n");
(void) stk;
(void) pci;
vars= (int*) GDKzalloc(sizeof(int)* mb->vtop);
if( vars == NULL)
return 0;
limit = mb->stop;
slimit= mb->ssize;
old = mb->stmt;
/* check for bailout conditions */
for (i = 1; i < limit; i++) {
int lastbat;
p = old[i];
for (j = 0; j<p->retc; j++) {
int res = getArg(p, j);
vars[res] = i;
}
if (getModuleId(p) == algebraRef &&
(getFunctionId(p) == tintersectRef || getFunctionId(p) == tinterRef ||
getFunctionId(p) == tdifferenceRef || getFunctionId(p) == tdiffRef)) {
GDKfree(vars);
return 0;
}
if (getModuleId(p) == algebraRef && getFunctionId(p) == sliceRef)
nr_topn++;
if (isLikeOp(p))
nr_likes++;
if (getModuleId(p) == sqlRef && getFunctionId(p) == deltaRef)
push_down_delta++;
if (getModuleId(p) == sqlRef && getFunctionId(p) == tidRef) { /* rewrite equal table ids */
int sname = getArg(p, 2), tname = getArg(p, 3), s;
for (s = 0; s < subselects.nr; s++) {
InstrPtr q = old[vars[subselects.tid[s]]];
int Qsname = getArg(q, 2), Qtname = getArg(q, 3);
if (no_updates(old, vars, getArg(q,1), getArg(p,1)) &&
((sname == Qsname && tname == Qtname) ||
(0 && strcmp(getVarConstant(mb, sname).val.sval, getVarConstant(mb, Qsname).val.sval) == 0 &&
strcmp(getVarConstant(mb, tname).val.sval, getVarConstant(mb, Qtname).val.sval) == 0))) {
clrFunction(p);
p->retc = 1;
p->argc = 2;
getArg(p, 1) = getArg(q, 0);
break;
}
}
}
lastbat = lastbat_arg(mb, p);
if (isSubSelect(p) && p->retc == 1 &&
/* no cand list */ getArgType(mb, p, lastbat) != newBatType(TYPE_oid, TYPE_oid)) {
int i1 = getArg(p, 1), tid = 0;
InstrPtr q = old[vars[i1]];
/* find the table ids */
while(!tid) {
if (getModuleId(q) == algebraRef && getFunctionId(q) == leftfetchjoinRef) {
int i1 = getArg(q, 1);
InstrPtr s = old[vars[i1]];
if (getModuleId(s) == sqlRef && getFunctionId(s) == tidRef)
tid = getArg(q, 1);
if (s->argc == 2 && s->retc == 1) {
int i1 = getArg(s, 1);
InstrPtr s = old[vars[i1]];
if (getModuleId(s) == sqlRef && getFunctionId(s) == tidRef)
tid = getArg(q, 1);
}
break;
} else if (isMapOp(q) && q->argc >= 2 && isaBatType(getArgType(mb, q, 1))) {
int i1 = getArg(q, 1);
q = old[vars[i1]];
} else if (isMapOp(q) && q->argc >= 3 && isaBatType(getArgType(mb, q, 2))) {
int i2 = getArg(q, 2);
q = old[vars[i2]];
} else {
break;
}
}
if (tid && subselect_add(&subselects, tid, getArg(p, 0)) < 0) {
GDKfree(vars);
return 0;
}
}
/* left hand side */
if ( (GDKdebug & (1<<15)) &&
isMatJoinOp(p) && p->retc == 2) {
int i1 = getArg(p, 2), tid = 0;
InstrPtr q = old[vars[i1]];
/* find the table ids */
while(!tid) {
if (getModuleId(q) == algebraRef && getFunctionId(q) == leftfetchjoinRef) {
int i1 = getArg(q, 1);
InstrPtr s = old[vars[i1]];
if (getModuleId(s) == sqlRef && getFunctionId(s) == tidRef)
tid = getArg(q, 1);
break;
} else if (isMapOp(q) && q->argc >= 2 && isaBatType(getArgType(mb, q, 1))) {
int i1 = getArg(q, 1);
q = old[vars[i1]];
} else if (isMapOp(q) && q->argc >= 3 && isaBatType(getArgType(mb, q, 2))) {
int i2 = getArg(q, 2);
q = old[vars[i2]];
} else {
break;
}
}
if (tid && subselect_add(&subselects, tid, getArg(p, 0)) < 0) {
GDKfree(vars);
return 0;
}
}
/* right hand side */
if ( (GDKdebug & (1<<15)) &&
isMatJoinOp(p) && p->retc == 2) {
int i1 = getArg(p, 3), tid = 0;
InstrPtr q = old[vars[i1]];
/* find the table ids */
while(!tid) {
if (getModuleId(q) == algebraRef && getFunctionId(q) == leftfetchjoinRef) {
int i1 = getArg(q, 1);
InstrPtr s = old[vars[i1]];
if (getModuleId(s) == sqlRef && getFunctionId(s) == tidRef)
tid = getArg(q, 1);
break;
} else if (isMapOp(q) && q->argc >= 2 && isaBatType(getArgType(mb, q, 1))) {
int i1 = getArg(q, 1);
q = old[vars[i1]];
} else if (isMapOp(q) && q->argc >= 3 && isaBatType(getArgType(mb, q, 2))) {
int i2 = getArg(q, 2);
q = old[vars[i2]];
} else {
break;
}
}
if (tid && subselect_add(&subselects, tid, getArg(p, 1)) < 0) {
GDKfree(vars);
return 0;
}
}
}
if ((!subselects.nr && !nr_topn && !nr_likes) || newMalBlkStmt(mb, mb->ssize) <0 ) {
GDKfree(vars);
return 0;
}
pushInstruction(mb,old[0]);
for (i = 1; i < limit; i++) {
p = old[i];
/* rewrite batalgebra.like + subselect -> likesubselect */
if (getModuleId(p) == algebraRef && p->retc == 1 && getFunctionId(p) == subselectRef) {
int var = getArg(p, 1);
InstrPtr q = mb->stmt[vars[var]]; /* BEWARE: the optimizer may not add or remove statements ! */
if (isLikeOp(q)) { /* TODO check if getArg(p, 3) value == TRUE */
InstrPtr r = newInstruction(mb, ASSIGNsymbol);
int has_cand = (getArgType(mb, p, 2) == newBatType(TYPE_oid, TYPE_oid));
int a, anti = (getFunctionId(q)[0] == 'n'), ignore_case = (getFunctionId(q)[anti?4:0] == 'i');
setModuleId(r, algebraRef);
setFunctionId(r, likesubselectRef);
getArg(r,0) = getArg(p,0);
r = pushArgument(mb, r, getArg(q, 1));
if (has_cand)
r = pushArgument(mb, r, getArg(p, 2));
for(a = 2; a<q->argc; a++)
r = pushArgument(mb, r, getArg(q, a));
if (r->argc < (4+has_cand))
r = pushStr(mb, r, ""); /* default esc */
if (r->argc < (5+has_cand))
r = pushBit(mb, r, ignore_case);
if (r->argc < (6+has_cand))
r = pushBit(mb, r, anti);
freeInstruction(p);
p = r;
actions++;
}
}
/* inject table ids into subselect
* s = subselect(c, C1..) => subselect(c, t, C1..)
*/
if (isSubSelect(p) && p->retc == 1) {
int tid = 0;
if ((tid = subselect_find_tids(&subselects, getArg(p, 0))) >= 0) {
int lastbat = lastbat_arg(mb, p);
if (getArgType(mb, p, lastbat) == TYPE_bat) /* empty candidate list bat_nil */
getArg(p, lastbat) = tid;
else
p = PushArgument(mb, p, tid, lastbat+1);
/* make sure to resolve again */
p->token = ASSIGNsymbol;
p->typechk = TYPE_UNKNOWN;
p->fcn = NULL;
p->blk = NULL;
actions++;
}
}
else if ( (GDKdebug & (1<<15)) &&
isMatJoinOp(p) && p->retc == 2
&& !(getFunctionId(p) == joinRef && p->argc > 4)
) {
int ltid = 0, rtid = 0, done = 0;
int range = 0;
if(getFunctionId(p) == joinRef)
range = (p->argc >= 4);
if ((ltid = subselect_find_tids(&subselects, getArg(p, 0))) >= 0 &&
(rtid = subselect_find_tids(&subselects, getArg(p, 1))) >= 0) {
p = PushArgument(mb, p, ltid, 4+range);
p = PushArgument(mb, p, rtid, 5+range);
done = 1;
} else if ((ltid = subselect_find_tids(&subselects, getArg(p, 0))) >= 0) {
p = PushArgument(mb, p, ltid, 4+range);
p = PushNil(mb, p, 5+range, TYPE_bat);
done = 1;
} else if ((rtid = subselect_find_tids(&subselects, getArg(p, 1))) >= 0) {
p = PushNil(mb, p, 4+range, TYPE_bat);
p = PushArgument(mb, p, rtid, 5+range);
done = 1;
}
if (done) {
if(getFunctionId(p) == antijoinRef)
p = pushInt(mb, p, JOIN_NE);
p = pushBit(mb, p, FALSE); /* do not match nils */
p = pushNil(mb, p, TYPE_lng); /* no estimate */
/* TODO join* -> subjoin* */
if(getFunctionId(p) == joinRef)
getFunctionId(p) = subjoinRef;
else if(getFunctionId(p) == antijoinRef)
getFunctionId(p) = subthetajoinRef;
else if(getFunctionId(p) == thetajoinRef)
getFunctionId(p) = subthetajoinRef;
else if(getFunctionId(p) == bandjoinRef)
getFunctionId(p) = subbandjoinRef;
/* make sure to resolve again */
p->token = ASSIGNsymbol;
p->typechk = TYPE_UNKNOWN;
p->fcn = NULL;
p->blk = NULL;
actions++;
}
}
/* Leftfetchjoins involving rewriten table ids need to be flattend
* l = leftfetchjoin(t, c); => l = c;
* and
* l = leftfetchjoin(s, ntids); => l = s;
*/
else if (getModuleId(p) == algebraRef && getFunctionId(p) == leftfetchjoinRef) {
int var = getArg(p, 1);
if (subselect_find_subselect(&subselects, var) > 0) {
InstrPtr q = newAssignment(mb);
getArg(q, 0) = getArg(p, 0);
(void) pushArgument(mb, q, getArg(p, 2));
actions++;
freeInstruction(p);
continue;
} else { /* deletes/updates use table ids */
int var = getArg(p, 2);
InstrPtr q = mb->stmt[vars[var]]; /* BEWARE: the optimizer may not add or remove statements ! */
if (q->token == ASSIGNsymbol) {
var = getArg(q, 1);
q = mb->stmt[vars[var]];
}
if (subselect_find_subselect(&subselects, var) > 0) {
InstrPtr qq = newAssignment(mb);
/* TODO: check result */
getArg(qq, 0) = getArg(p, 0);
(void) pushArgument(mb, qq, getArg(p, 1));
actions++;
freeInstruction(p);
continue;
}
/* c = sql.delta(b,uid,uval,ins);
* l = leftfetchjoin(x, c);
* into
* l = sql.projectdelta(x,b,uid,uval,ins);
*/
else if (getModuleId(q) == sqlRef && getFunctionId(q) == deltaRef && q->argc == 5) {
q = copyInstruction(q);
setFunctionId(q, projectdeltaRef);
getArg(q, 0) = getArg(p, 0);
q = PushArgument(mb, q, getArg(p, 1), 1);
freeInstruction(p);
p = q;
actions++;
}
}
}
pushInstruction(mb,p);
}
for (; i<limit; i++)
if (old[i])
pushInstruction(mb,old[i]);
for (; i<slimit; i++)
if (old[i])
freeInstruction(old[i]);
GDKfree(old);
if (!push_down_delta) {
GDKfree(vars);
return actions;
}
/* now push selects through delta's */
limit = mb->stop;
slimit= mb->ssize;
old = mb->stmt;
if (newMalBlkStmt(mb, mb->stop+(5*push_down_delta)) <0 ) {
mb->stmt = old;
GDKfree(vars);
return actions;
}
pushInstruction(mb,old[0]);
for (i = 1; i < limit; i++) {
int lastbat;
p = old[i];
for (j = 0; j<p->retc; j++) {
int res = getArg(p, j);
vars[res] = i;
}
/* push subslice under projectdelta */
if (isSlice(p) && p->retc == 1) {
int var = getArg(p, 1);
InstrPtr q = old[vars[var]];
if (getModuleId(q) == sqlRef && getFunctionId(q) == projectdeltaRef) {
InstrPtr r = copyInstruction(p);
InstrPtr s = copyInstruction(q);
ValRecord cst;
/* slice the candidates */
setFunctionId(r, sliceRef);
getArg(r, 0) = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
getArg(r, 1) = getArg(s, 1);
cst.vtype = getArgType(mb, r, 2);
cst.val.wval = 0;
getArg(r, 2) = defConstant(mb, cst.vtype, &cst); /* start from zero */
pushInstruction(mb,r);
/* dummy result for the old q, will be removed by deadcode optimizer */
getArg(q, 0) = newTmpVariable(mb, getArgType(mb, q, 0));
getArg(s, 1) = getArg(r, 0); /* use result of subslice */
pushInstruction(mb, s);
}
}
/* c = delta(b, uid, uvl, ins)
* s = subselect(c, C1..)
*
* nc = subselect(b, C1..)
* ni = subselect(ins, C1..)
* nu = subselect(uvl, C1..)
* s = subdelta(nc, uid, nu, ni);
*
* doesn't handle Xsubselect(x, .. z, C1.. cases) ie multicolumn selects
*/
lastbat = lastbat_arg(mb, p);
if (isSubSelect(p) && p->retc == 1 && lastbat == 2) {
int var = getArg(p, 1);
InstrPtr q = old[vars[var]];
if (q->token == ASSIGNsymbol) {
var = getArg(q, 1);
q = old[vars[var]];
}
if (getModuleId(q) == sqlRef && getFunctionId(q) == deltaRef) {
InstrPtr r = copyInstruction(p);
InstrPtr s = copyInstruction(p);
InstrPtr t = copyInstruction(p);
InstrPtr u = copyInstruction(q);
getArg(r, 0) = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
getArg(r, 1) = getArg(q, 1); /* column */
pushInstruction(mb,r);
getArg(s, 0) = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
getArg(s, 1) = getArg(q, 3); /* updates */
s = ReplaceWithNil(mb, s, 2, TYPE_bat); /* no candidate list */
setArgType(mb, s, 2, newBatType(TYPE_oid,TYPE_oid));
/* make sure to resolve again */
s->token = ASSIGNsymbol;
s->typechk = TYPE_UNKNOWN;
s->fcn = NULL;
s->blk = NULL;
pushInstruction(mb,s);
getArg(t, 0) = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
getArg(t, 1) = getArg(q, 4); /* inserts */
pushInstruction(mb,t);
setFunctionId(u, subdeltaRef);
getArg(u, 0) = getArg(p,0);
getArg(u, 1) = getArg(r,0);
getArg(u, 2) = getArg(p,2); /* pre-cands */
getArg(u, 3) = getArg(q,2); /* update ids */
getArg(u, 4) = getArg(s,0);
u = pushArgument(mb, u, getArg(t,0));
pushInstruction(mb,u);
freeInstruction(p);
continue;
}
}
pushInstruction(mb,p);
}
for (; i<limit; i++)
if (old[i])
pushInstruction(mb,old[i]);
GDKfree(vars);
GDKfree(old);
return actions;
}
static int
OPTsql_appendImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
InstrPtr *old = NULL;
int i, limit, slimit, actions = 0;
(void) pci; /* Tell compilers that we know that we do not */
(void) stk; /* use these function parameters, here. */
/* In general, a MAL optimizer transforms a given MAL program into a
* modified one by sequentially walking through the given program
* and concurrently creating a new one from scratch by
* (1) copying statements as is, modified, or in a different order,
* or (2) omitting statements or (3) introducing new statements.
*/
/* check for logical error: mb must never be NULL */
assert (mb != NULL);
/* save the old stage of the MAL block */
old = mb->stmt;
limit= mb->stop;
slimit = mb->ssize;
/* initialize the statement list. Notice, the symbol table remains intact */
if (newMalBlkStmt(mb, mb->ssize) < 0)
return 0;
/* the plan signature can be copied safely */
pushInstruction(mb, old[0]);
/* iterate over the instructions of the input MAL program */
for (i = 1; i < limit; i++) {
InstrPtr p = old[i];
/* check for
* v3 := sql.append( ..., ..., ..., ..., v0 );
*/
if (getModuleId(p) == sqlRef &&
getFunctionId(p) == appendRef &&
p->argc > 5 &&
p->retc == 1 &&
isaBatType(getArgType(mb, p, 5))) {
/* found
* v3 := sql.append( ..., ..., ..., ..., v0 );
*/
int j = 0, k = 0;
InstrPtr q1 = NULL, q2 = NULL;
bit found = FALSE;
/* check whether next is
* v4 := aggr.count(v0);
*/
if (i+1 < limit) {
InstrPtr q = old[i+1];
if (getModuleId(q) == aggrRef &&
getFunctionId(q) == countRef &&
q->argc == 2 &&
q->retc == 1 &&
getArg(q, 1) == getArg(p, 5)) {
/* found
* v3 := sql.append( ..., ..., ..., ..., v0 );
* v4 := aggr.count(v0);
*/
/* issue/execute
* v4 := aggr.count(v0);
* before
* v3 := sql.append( ..., ..., ..., ..., v0 );
*/
pushInstruction(mb, q);
q1 = q;
i++;
actions++; /* to keep track if anything has been done */
}
}
/* look for
* v5 := ... v0 ...;
*/
/* an expensive loop, better would be to remember that v0
* has a different role. A typical method is to keep a
* map from variable -> instruction where it was
* detected. Then you can check each assignment for use of
* v0
*/
for (j = i+1; !found && j < limit; j++)
for (k = old[j]->retc; !found && k < old[j]->argc; k++)
found = (getArg(old[j], k) == getArg(p, 5));
if (found) {
/* replace
* v3 := sql.append( ..., ..., ..., ..., v0 );
* with
* v1 := aggr.count( v0 );
* v2 := algebra.slice( v0, 0, v1 );
* v3 := sql.append( ..., ..., ..., ..., v2 );
*/
/* push new v1 := aggr.count( v0 ); unless already available */
if (q1 == NULL) {
/* use mal_builder.h primitives
* q1 = newStmt(mb, aggrRef,countRef);
* setArgType(mb,q1,TYPE_wrd) */
/* it will be added to the block and even my
* re-use MAL instructions */
q1 = newInstruction(mb,ASSIGNsymbol);
getArg(q1,0) = newTmpVariable(mb, TYPE_wrd);
setModuleId(q1, aggrRef);
setFunctionId(q1, countRef);
q1 = pushArgument(mb, q1, getArg(p, 5));
pushInstruction(mb, q1);
}
/* push new v2 := algebra.slice( v0, 0, v1 ); */
/* use mal_builder.h primitives
* q1 = newStmt(mb, algebraRef,sliceRef); */
q2 = newInstruction(mb,ASSIGNsymbol);
getArg(q2,0) = newTmpVariable(mb, TYPE_any);
setModuleId(q2, algebraRef);
setFunctionId(q2, sliceRef);
q2 = pushArgument(mb, q2, getArg(p, 5));
q2 = pushWrd(mb, q2, 0);
q2 = pushArgument(mb, q2, getArg(q1, 0));
pushInstruction(mb, q2);
/* push modified v3 := sql.append( ..., ..., ..., ..., v2 ); */
getArg(p, 5) = getArg(q2, 0);
pushInstruction(mb, p);
actions++;
continue;
}
}
pushInstruction(mb, p);
if (p->token == ENDsymbol) break;
}
/* We would like to retain everything from the ENDsymbol
* up to the end of the plan, because after the ENDsymbol
* the remaining optimizer steps are stored.
*/
for(i++; i<limit; i++)
if (old[i])
pushInstruction(mb, old[i]);
/* any remaining MAL instruction records are removed */
for(; i<slimit; i++)
if (old[i])
freeInstruction(old[i]);
GDKfree(old);
/* for statistics we return if/how many patches have been made */
DEBUGoptimizers
mnstr_printf(cntxt->fdout,"#opt_sql_append: %d statements added\n",
actions);
return actions;
}
int malAtomProperty(MalBlkPtr mb, InstrPtr pci)
{
str name;
int tpe;
(void)mb; /* fool compilers */
assert(pci != 0);
name = getFunctionId(pci);
tpe = getTypeIndex(getModuleId(pci), (int)strlen(getModuleId(pci)), TYPE_any);
if (tpe < 0 || tpe >= GDKatomcnt || tpe >= MAXATOMS)
return 0;
assert(pci->fcn != NULL);
switch (name[0]) {
case 'd':
if (idcmp("del", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomDel = (void (*)(Heap *, var_t *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'c':
if (idcmp("cmp", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomCmp = (int (*)(const void *, const void *))pci->fcn;
BATatoms[tpe].linear = 1;
setAtomName(pci);
return 1;
}
break;
case 'f':
if (idcmp("fromstr", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomFromStr = (int (*)(const char *, int *, ptr *))pci->fcn;
setAtomName(pci);
return 1;
}
if (idcmp("fix", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomFix = (int (*)(const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'h':
if (idcmp("heap", name) == 0 && pci->argc == 1) {
/* heap function makes an atom varsized */
BATatoms[tpe].size = sizeof(var_t);
assert_shift_width(ATOMelmshift(ATOMsize(tpe)), ATOMsize(tpe));
BATatoms[tpe].align = sizeof(var_t);
BATatoms[tpe].atomHeap = (void (*)(Heap *, size_t))pci->fcn;
setAtomName(pci);
return 1;
}
if (idcmp("hash", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomHash = (BUN (*)(const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'l':
if (idcmp("length", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomLen = (int (*)(const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'n':
if (idcmp("null", name) == 0 && pci->argc == 1) {
ptr atmnull = ((ptr (*)(void))pci->fcn)();
BATatoms[tpe].atomNull = atmnull;
setAtomName(pci);
return 1;
}
if (idcmp("nequal", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomCmp = (int (*)(const void *, const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'p':
if (idcmp("put", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomPut = (var_t (*)(Heap *, var_t *, const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 's':
if (idcmp("storage", name) == 0 && pci->argc == 1) {
BATatoms[tpe].storage = (*(int (*)(void))pci->fcn)();
setAtomName(pci);
return 1;
}
break;
case 't':
if (idcmp("tostr", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomToStr = (int (*)(str *, int *, const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'u':
if (idcmp("unfix", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomUnfix = (int (*)(const void *))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'r':
if (idcmp("read", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomRead = (void *(*)(void *, stream *, size_t))pci->fcn;
setAtomName(pci);
return 1;
}
break;
case 'w':
if (idcmp("write", name) == 0 && pci->argc == 1) {
BATatoms[tpe].atomWrite = (gdk_return (*)(const void *, stream *, size_t))pci->fcn;
setAtomName(pci);
return 1;
}
break;
}
return 0;
}
str
shortStmtRendering(MalBlkPtr mb, MalStkPtr stk, InstrPtr p)
{
int i;
str base, s, t, nme;
size_t len= (mb->stop < 1000? 1000: mb->stop) * 128 /* max realistic line length estimate */;
base = s = GDKmalloc(len);
if ( s == NULL)
return s;
*s =0;
t=s;
if (p->token == REMsymbol && !( getModuleId(p) && strcmp(getModuleId(p),"querylog") == 0 && getFunctionId(p) && strcmp(getFunctionId(p),"define") == 0))
return base;
if (p->barrier == LEAVEsymbol ||
p->barrier == REDOsymbol ||
p->barrier == RETURNsymbol ||
p->barrier == YIELDsymbol ||
p->barrier == EXITsymbol ||
p->barrier == RAISEsymbol) {
snprintf(t,(len-(t-base)), "%s ", operatorName(p->barrier));
advance(t,base,len);
}
if( p->token == FUNCTIONsymbol) {
snprintf(t,(len-(t-base)), "function %s.", getModuleId(p));
advance(t,base,len);
}
if (p->token == ENDsymbol ){
snprintf(t,(len-(t-base)), "end %s.%s", getModuleId(getInstrPtr(mb,0)), getFunctionId(getInstrPtr(mb,0)));
return base;
}
// handle the result variables
for (i = 0; i < p->retc; i++)
if ( !isTmpVar(mb,getArg(p,i)) || isVarUsed(mb, getArg(p, i)) || isVarUDFtype(mb,getArg(p,i)))
break;
if (i == p->retc) // no result arguments
goto short_end;
/* display optional multi-assignment list */
if( getArgType(mb,p,0) != TYPE_void){
if (p->retc > 1 && t < base + len-1){
*t++ = '(';
*t=0;
}
for (i = 0; i < p->retc; i++) {
nme = shortRenderingTerm(mb, stk, p,i);
snprintf(t,(len-(t-base)), "%s%s", (i?", ":""), nme);
GDKfree(nme);
advance(t,base,len);
}
if (p->retc > 1 && t< base+len)
*t++ = ')';
if( t < base +len) *t++ = ':';
if( t < base +len) *t++ = '=';
if( t < base +len) *t++ = ' ';
}
*t =0;
short_end:
advance(t,base,len);
// handle the instruction mapping
snprintf(t, (len-(t-base)),"%s", (getFunctionId(p)?getFunctionId(p):""));
advance(t,base,len);
// handle the arguments, constants should be shown including their non-default type
/* display optional multi-assignment list */
if( t< base + len) *t++ = '(';
for (i = p->retc; i < p->argc; i++) {
nme = shortRenderingTerm(mb, stk, p,i);
snprintf(t,(len-(t-base)), "%s%s", (i!= p->retc? ", ":" "), nme);
GDKfree(nme);
advance(t,base,len);
if (i < p->retc - 1 && t < base+len){
*t++ = ',';
*t++ = ' ';
}
}
if( t < base + len) *t++ = ' ';
if( t < base + len) *t++ = ')';
*t=0;
if (t >= s + len)
throw(MAL,"instruction2str:","instruction too long");
return base;
}
str
instruction2str(MalBlkPtr mb, MalStkPtr stk, InstrPtr p, int flg)
{
int i;
str base, t;
size_t len = 512 + (p->argc * 128); /* max realistic line length estimate */
str arg;
t = base = GDKmalloc(len);
if ( base == NULL)
return NULL;
if (!flg) {
*t++ = '#';
len--;
if (p->typechk == TYPE_UNKNOWN) {
*t++ = '!'; /* error */
len--;
}
}
*t = 0;
if (p->token == REMsymbol && !( getModuleId(p) && strcmp(getModuleId(p),"querylog") == 0 && getFunctionId(p) && strcmp(getFunctionId(p),"define") == 0)) {
/* do nothing */
} else if (p->barrier) {
if (p->barrier == LEAVEsymbol ||
p->barrier == REDOsymbol ||
p->barrier == RETURNsymbol ||
p->barrier == YIELDsymbol ||
p->barrier == RAISEsymbol) {
if (!copystring(&t, " ", &len))
return base;
}
arg = operatorName(p->barrier);
if (!copystring(&t, arg, &len) ||
!copystring(&t, " ", &len))
return base;
} else if( functionStart(p) && flg != LIST_MAL_CALL ){
return fcnDefinition(mb, p, t, flg, base, len + (t - base));
} else if (!functionExit(p) && flg!=LIST_MAL_CALL) {
// beautify with tabs
if (!copystring(&t, " ", &len))
return base;
}
switch (p->token<0?-p->token:p->token) {
case FCNcall:
case FACcall:
case PATcall:
case CMDcall:
case ASSIGNsymbol :
// is any variable explicit or used
for (i = 0; i < p->retc; i++)
if ( !isTmpVar(mb,getArg(p,i)) || isVarUsed(mb, getArg(p, i)) || isVarUDFtype(mb,getArg(p,i)))
break;
if (i == p->retc)
break;
/* display multi-assignment list */
if (p->retc > 1 && !copystring(&t, "(", &len))
return base;
for (i = 0; i < p->retc; i++) {
arg= renderTerm(mb, stk, p, i, flg);
if (arg) {
if (!copystring(&t, arg, &len)) {
GDKfree(arg);
return base;
}
GDKfree(arg);
}
if (i < p->retc - 1 && !copystring(&t, ", ", &len))
return base;
}
if (p->retc > 1 && !copystring(&t, ")", &len))
return base;
if (p->argc > p->retc || getFunctionId(p)) {
if (!copystring(&t, " := ", &len))
return base;
}
break;
case ENDsymbol:
if (!copystring(&t, "end ", &len) ||
!copystring(&t, getModuleId(getInstrPtr(mb,0)), &len) ||
!copystring(&t, ".", &len) ||
!copystring(&t, getFunctionId(getInstrPtr(mb, 0)), &len))
return base;
break;
case COMMANDsymbol:
case FUNCTIONsymbol:
case FACTORYsymbol:
case PATTERNsymbol:
if (flg & LIST_MAL_VALUE) {
if (!copystring(&t, operatorName(p->token), &len) ||
!copystring(&t, " ", &len))
return base;
}
return fcnDefinition(mb, p, t, flg, base, len + (t - base));
case REMsymbol:
case NOOPsymbol:
if (!copystring(&t, "#", &len))
return base;
if (getVar(mb, getArg(p, 0))->value.val.sval && getVar(mb, getArg(p, 0))->value.len > 0 &&
!copystring(&t, getVar(mb, getArg(p, 0))->value.val.sval, &len))
return base;
if (!copystring(&t, " ", &len))
return base;
break;
default:
i = snprintf(t, len, " unknown symbol ?%d? ", p->token);
if (i < 0 || (size_t) i >= len)
return base;
len -= (size_t) i;
t += i;
break;
}
if (getModuleId(p)) {
if (!copystring(&t, getModuleId(p), &len) ||
!copystring(&t, ".", &len))
return base;
}
if (getFunctionId(p)) {
if (!copystring(&t, getFunctionId(p), &len) ||
!copystring(&t, "(", &len))
return base;
} else if (p->argc > p->retc + 1) {
if (!copystring(&t, "(", &len))
return base;
}
for (i = p->retc; i < p->argc; i++) {
arg= renderTerm(mb, stk, p, i, flg);
if (arg) {
if (!copystring(&t, arg, &len)) {
GDKfree(arg);
return base;
}
GDKfree(arg);
}
if (i < p->argc -1 && !copystring(&t, ", ", &len))
return base;
}
if (getFunctionId(p) || p->argc > p->retc + 1) {
if (!copystring(&t, ")", &len))
return base;
}
if (p->token != REMsymbol){
if (!copystring(&t, ";", &len))
return base;
}
return base;
}
str
fcnDefinition(MalBlkPtr mb, InstrPtr p, str t, int flg, str base, size_t len)
{
int i;
str arg, tpe;
len -= t - base;
if (!flg && !copystring(&t, "#", &len))
return base;
if( mb->inlineProp && !copystring(&t, "inline ", &len))
return base;
if( mb->unsafeProp && !copystring(&t, "unsafe ", &len))
return base;
if( mb->sealedProp && !copystring(&t, "sealed ", &len))
return base;
if (!copystring(&t, operatorName(p->token), &len) ||
!copystring(&t, " ", &len) ||
!copystring(&t, getModuleId(p) ? getModuleId(p) : "user", &len) ||
!copystring(&t, ".", &len) ||
!copystring(&t, getFunctionId(p), &len) ||
!copystring(&t, "(", &len))
return base;
for (i = p->retc; i < p->argc; i++) {
arg = renderTerm(mb, 0, p, i, (LIST_MAL_NAME | LIST_MAL_TYPE | LIST_MAL_PROPS));
if (arg && !copystring(&t, arg, &len)) {
GDKfree(arg);
return base;
}
GDKfree(arg);
if( i<p->argc-1 && !copystring(&t, ", ", &len))
return base;
}
advance(t,base,len);
if (p->varargs & VARARGS && !copystring(&t, "...", &len))
return base;
if (p->retc == 1) {
if (!copystring(&t, "):", &len))
return base;
tpe = getTypeName(getVarType(mb, getArg(p,0)));
if (!copystring(&t, tpe, &len)) {
GDKfree(tpe);
return base;
}
GDKfree(tpe);
if (p->varargs & VARRETS && !copystring(&t, "...", &len))
return base;
} else {
if (!copystring(&t, ") (", &len))
return base;
for (i = 0; i < p->retc; i++) {
arg = renderTerm(mb, 0, p, i, (LIST_MAL_NAME | LIST_MAL_TYPE | LIST_MAL_PROPS));
if (arg && !copystring(&t, arg, &len)) {
GDKfree(arg);
return base;
}
GDKfree(arg);
if( i<p->retc-1 && !copystring(&t, ", ", &len))
return base;
}
if (p->varargs & VARRETS && !copystring(&t, "...", &len))
return base;
if (!copystring(&t, ")", &len))
return base;
}
if (mb->binding[0]) {
if (!copystring(&t, " address ", &len) ||
!copystring(&t, mb->binding, &len))
return base;
}
(void) copystring(&t, ";", &len);
return base;
}
str
OPTvolcanoImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i, limit;
int mvcvar = -1;
int count=0;
InstrPtr p,q, *old = mb->stmt;
char buf[256];
lng usec = GDKusec();
str msg = MAL_SUCCEED;
(void) pci;
(void) cntxt;
(void) stk; /* to fool compilers */
if ( mb->inlineProp )
return MAL_SUCCEED;
limit= mb->stop;
if ( newMalBlkStmt(mb, mb->ssize + 20) < 0)
throw(MAL,"optimizer.volcano", SQLSTATE(HY001) MAL_MALLOC_FAIL);
for (i = 0; i < limit; i++) {
p = old[i];
pushInstruction(mb,p);
if( getModuleId(p) == sqlRef && getFunctionId(p)== mvcRef ){
mvcvar = getArg(p,0);
continue;
}
if( count < MAXdelays && getModuleId(p) == algebraRef ){
if( getFunctionId(p) == selectRef ||
getFunctionId(p) == thetaselectRef ||
getFunctionId(p) == likeselectRef ||
getFunctionId(p) == joinRef
){
q= newInstruction(0,languageRef,blockRef);
setDestVar(q, newTmpVariable(mb,TYPE_any));
q = pushArgument(mb,q,mvcvar);
q = pushArgument(mb,q,getArg(p,0));
mvcvar= getArg(q,0);
pushInstruction(mb,q);
count++;
}
continue;
}
if( count < MAXdelays && getModuleId(p) == groupRef ){
if( getFunctionId(p) == subgroupdoneRef || getFunctionId(p) == groupdoneRef ){
q= newInstruction(0,languageRef,blockRef);
setDestVar(q, newTmpVariable(mb,TYPE_any));
q = pushArgument(mb,q,mvcvar);
q = pushArgument(mb,q,getArg(p,0));
mvcvar= getArg(q,0);
pushInstruction(mb,q);
count++;
}
}
if( getModuleId(p) == sqlRef){
if ( getFunctionId(p) == bindRef ||
getFunctionId(p) == bindidxRef ||
getFunctionId(p)== tidRef ||
getFunctionId(p)== appendRef ||
getFunctionId(p)== updateRef ||
getFunctionId(p)== deleteRef
){
setArg(p,p->retc,mvcvar);
}
}
}
GDKfree(old);
/* Defense line against incorrect plans */
if( count){
chkTypes(cntxt->usermodule, mb, FALSE);
chkFlow(mb);
chkDeclarations(mb);
}
/* keep all actions taken as a post block comment */
usec = GDKusec()- usec;
snprintf(buf,256,"%-20s actions=%2d time=" LLFMT " usec","volcano",count,usec);
newComment(mb,buf);
if( count >= 0)
addtoMalBlkHistory(mb);
return msg;
}
int
OPTpushrangesImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i,j, limit,actions=0;
InstrPtr p, *old;
int x,y,z;
Range range;
if( mb->errors)
return 0;
range= (Range) GDKzalloc(mb->vtop * sizeof(RangeRec));
if (range == NULL)
return 0;
OPTDEBUGpushranges
mnstr_printf(cntxt->fdout,"#Range select optimizer started\n");
(void) stk;
(void) pci;
limit = mb->stop;
old = mb->stmt;
/*
* In phase I we collect information about constants
*/
for (i = 0; i < limit; i++) {
p = old[i];
if( p->barrier)
break; /* end of optimizer */
for(j=p->retc; j< p->argc; j++)
range[getArg(p,j)].used++;
for(j=0; j<p->retc; j++){
range[getArg(p,j)].lastupdate= i;
if( range[getArg(p,j)].lastrange == 0)
range[getArg(p,j)].lastrange= i;
}
if( getModuleId(p)== algebraRef &&
( getFunctionId(p)== selectRef || getFunctionId(p)== uselectRef) ){
/*
* The operation X:= algebra.select(Y,L,H,Li,Hi) is analysed.
* First, we attempt to propagate the range known for Y onto the
* requested range of X. This may lead to smaller range of
* even the conclusion that X is necessarily empty.
* Of course, only under the condition that Y has not been changed by a
* side-effect since it was bound to X.
*/
x= getArg(p,1);
y= getArg(p,2);
if( range[x].lcst && isVarConstant(mb,y) ){
/* merge lowerbound */
if( ATOMcmp( getVarGDKType(mb,y),
VALptr( &getVarConstant(mb,range[x].lcst)),
VALptr( &getVarConstant(mb,y)) ) > 0){
getArg(p,2)= range[x].lcst;
z= range[x].srcvar;
if( getArg(p,1) == x &&
range[z].lastupdate == range[z].lastrange){
getArg(p,1) = z;
actions++;
}
}
y= getArg(p,3);
/* merge higherbound */
if( ATOMcmp( getVarGDKType(mb,y),
VALptr( &getVarConstant(mb,range[x].hcst)),
VALptr( &getVarConstant(mb,y)) ) < 0 ||
ATOMcmp( getVarGDKType(mb,y),
VALptr( &getVarConstant(mb,y)),
ATOMnilptr(getVarType(mb,y)) ) == 0){
getArg(p,3)= range[x].hcst;
z= range[x].srcvar;
if( getArg(p,1) == x && range[z].lastupdate == range[z].lastrange){
getArg(p,1) = z;
actions++;
}
}
}
/*
* The second step is to assign the result of this exercise to the
* result variable.
*/
x= getArg(p,0);
if( isVarConstant(mb, getArg(p,2)) ){
range[x].lcst = getArg(p,2);
range[x].srcvar= getArg(p,1);
range[x].lastupdate= range[x].lastrange = i;
}
if( isVarConstant(mb, getArg(p,3)) ){
range[x].hcst = getArg(p,3);
range[x].srcvar= getArg(p,1);
range[x].lastupdate= range[x].lastrange = i;
}
/*
* If both range bounds are constant, we can also detect empty results.
* It is empty if L> H or when L=H and the bounds are !(true,true).
*/
x= getArg(p,2);
y= getArg(p,3);
if( isVarConstant(mb, x) &&
isVarConstant(mb, y) ){
z =ATOMcmp( getVarGDKType(mb,y),
VALptr( &getVarConstant(mb,x)),
VALptr( &getVarConstant(mb,y)));
x= p->argc > 4;
x= x && isVarConstant(mb,getArg(p,4));
x= x && isVarConstant(mb,getArg(p,5));
x= x && getVarConstant(mb,getArg(p,4)).val.btval;
x= x && getVarConstant(mb,getArg(p,5)).val.btval;
if( z > 0 || (z==0 && p->argc>4 && !x)) {
int var = getArg(p, 0);
wrd zero = 0;
ValRecord v, *vp;
vp = VALset(&v, TYPE_wrd, &zero);
varSetProp(mb, var, rowsProp, op_eq, vp);
/* create an empty replacement */
x = getArgType(mb, p, 1);
p->argc=1;
getModuleId(p)= batRef;
getFunctionId(p)= newRef;
p= pushArgument(mb,p, newTypeVariable(mb, getHeadType(x)));
(void) pushArgument(mb,p, newTypeVariable(mb, getTailType(x)));
actions++;
}
}
}
}
OPTDEBUGpushranges
for(j=0; j< mb->vtop; j++)
if( range[j].used )
printRange(cntxt, mb,range,j);
/*
* Phase II, if we succeeded in pushing constants around and
* changing instructions, we might as well try once more to perform
* aliasRemoval, constantExpression, and pushranges.
*/
GDKfree(range);
return actions;
}
static void
SQLgetStatistics(Client cntxt, mvc *m, MalBlkPtr mb)
{
InstrPtr *old = NULL;
int oldtop, i, actions = 0, size = 0;
lng clk = GDKusec();
sql_trans *tr = m->session->tr;
str msg;
old = mb->stmt;
oldtop = mb->stop;
size = (mb->stop * 1.2 < mb->ssize) ? mb->ssize : (int) (mb->stop * 1.2);
mb->stmt = (InstrPtr *) GDKzalloc(size * sizeof(InstrPtr));
mb->ssize = size;
mb->stop = 0;
for (i = 0; i < oldtop; i++) {
InstrPtr p = old[i];
char *f = getFunctionId(p);
ValRecord vr;
if (getModuleId(p) == sqlRef && f == tidRef) {
char *sname = getVarConstant(mb, getArg(p, 2)).val.sval;
char *tname = getVarConstant(mb, getArg(p, 3)).val.sval;
sql_schema *s = mvc_bind_schema(m, sname);
sql_table *t;
if (!s || strcmp(s->base.name, dt_schema) == 0) {
pushInstruction(mb, p);
continue;
}
t = mvc_bind_table(m, s, tname);
if (t && (!isRemote(t) && !isMergeTable(t)) && t->p) {
int k = getArg(p, 0), mt_member = t->p->base.id;
varSetProp(mb, k, mtProp, op_eq, VALset(&vr, TYPE_int, &mt_member));
}
}
if (getModuleId(p) == sqlRef && (f == bindRef || f == bindidxRef)) {
int upd = (p->argc == 7 || p->argc == 9);
char *sname = getVarConstant(mb, getArg(p, 2 + upd)).val.sval;
char *tname = getVarConstant(mb, getArg(p, 3 + upd)).val.sval;
char *cname = NULL;
int not_null = 0, mt_member = 0;
wrd rows = 1; /* default to cope with delta bats */
int mode = 0;
int k = getArg(p, 0);
sql_schema *s = mvc_bind_schema(m, sname);
BAT *b;
if (!s || strcmp(s->base.name, dt_schema) == 0) {
pushInstruction(mb, p);
continue;
}
cname = getVarConstant(mb, getArg(p, 4 + upd)).val.sval;
mode = getVarConstant(mb, getArg(p, 5 + upd)).val.ival;
if (s && f == bindidxRef && cname) {
size_t cnt;
sql_idx *i = mvc_bind_idx(m, s, cname);
if (i && (!isRemote(i->t) && !isMergeTable(i->t))) {
cnt = store_funcs.count_idx(tr, i, 1);
assert(cnt <= (size_t) GDK_oid_max);
b = store_funcs.bind_idx(m->session->tr, i, RDONLY);
if (b) {
str loc;
if (b->batPersistence == PERSISTENT && BATlocation(&loc, &b->batCacheid) && loc)
varSetProp(mb, k, fileProp, op_eq, VALset(&vr, TYPE_str, loc));
cnt = BATcount(b);
BBPunfix(b->batCacheid);
}
rows = (wrd) cnt;
if (i->t->p)
mt_member = i->t->p->base.id;
}
} else if (s && f == bindRef && cname) {
size_t cnt;
sql_table *t = mvc_bind_table(m, s, tname);
sql_column *c = mvc_bind_column(m, t, cname);
if (c && (!isRemote(c->t) && !isMergeTable(c->t))) {
not_null = !c->null;
cnt = store_funcs.count_col(tr, c, 1);
assert(cnt <= (size_t) GDK_oid_max);
b = store_funcs.bind_col(m->session->tr, c, RDONLY);
if (b) {
str loc;
if (b->batPersistence == PERSISTENT && BATlocation(&loc, &b->batCacheid) && loc)
varSetProp(mb, k, fileProp, op_eq, VALset(&vr, TYPE_str, loc));
cnt = BATcount(b);
BBPunfix(b->batCacheid);
}
rows = (wrd) cnt;
if (c->t->p)
mt_member = c->t->p->base.id;
}
}
if (rows > 1 && mode != RD_INS)
varSetProp(mb, k, rowsProp, op_eq, VALset(&vr, TYPE_wrd, &rows));
if (not_null)
varSetProp(mb, k, notnilProp, op_eq, NULL);
if (mt_member && mode != RD_INS)
varSetProp(mb, k, mtProp, op_eq, VALset(&vr, TYPE_int, &mt_member));
{
int lowprop = hlbProp, highprop = hubProp;
/* rows == cnt has been checked above to be <= GDK_oid_max */
oid low = 0, high = low + (oid) rows;
pushInstruction(mb, p);
if (mode == RD_INS) {
low = high;
high += 1024 * 1024;
}
varSetProp(mb, getArg(p, 0), lowprop, op_gte, VALset(&vr, TYPE_oid, &low));
varSetProp(mb, getArg(p, 0), highprop, op_lt, VALset(&vr, TYPE_oid, &high));
}
if (not_null)
actions++;
} else {
pushInstruction(mb, p);
}
}
GDKfree(old);
msg = optimizerCheck(cntxt, mb, "optimizer.SQLgetstatistics", actions, GDKusec() - clk);
if (msg) /* what to do with an error? */
GDKfree(msg);
}
int
OPTquerylogImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i, limit, slimit;
InstrPtr p = 0, *old= mb->stmt, q,r;
int argc, io, user,nice,sys,idle,iowait,load, arg, start,finish, name;
int xtime=0, rtime = 0, tuples=0;
InstrPtr defineQuery = NULL;
// query log needed?
if ( !QLOGisset() )
return 0;
(void) pci;
(void) stk; /* to fool compilers */
(void) cntxt;
/* gather information */
for (i = 1; i < mb->stop; i++) {
p = getInstrPtr(mb,i);
if ( getModuleId(p) && idcmp(getModuleId(p), "querylog") == 0 && idcmp(getFunctionId(p),"define")==0){
defineQuery= p;
getVarConstant(mb,getArg(p,3)).val.lval = GDKusec()-getVarConstant(mb,getArg(p,3)).val.lval ;
}
}
if ( defineQuery == NULL)
/* nothing to do */
return 0;
limit= mb->stop;
slimit= mb->ssize;
if ( newMalBlkStmt(mb, mb->ssize) < 0)
return 0;
pushInstruction(mb, old[0]);
/* run the querylog.define operation */
defineQuery = copyInstruction(defineQuery);
setFunctionId(defineQuery, insertRef);
getArg(defineQuery,0) = newTmpVariable(mb,TYPE_any);
defineQuery->token = ASSIGNsymbol;
setModuleId(defineQuery,querylogRef);
/* collect the initial statistics */
q = newStmt(mb, "clients", "getUsername");
name= getArg(q,0)= newVariable(mb,GDKstrdup("name"),TYPE_str);
defineQuery = pushArgument(mb,defineQuery,name);
q = newStmt(mb, "mtime", "current_timestamp");
start= getArg(q,0)= newVariable(mb,GDKstrdup("start"),TYPE_timestamp);
defineQuery = pushArgument(mb,defineQuery,start);
pushInstruction(mb, defineQuery);
q = newStmt1(mb, sqlRef, "argRecord");
for ( argc=1; argc < old[0]->argc; argc++)
q = pushArgument(mb, q, getArg(old[0],argc));
arg= getArg(q,0)= newVariable(mb,GDKstrdup("args"),TYPE_str);
q = newStmt(mb, "alarm", "usec");
xtime = getArg(q,0)= newVariable(mb,GDKstrdup("xtime"),TYPE_lng);
user = newVariable(mb,GDKstrdup("user"),TYPE_lng);
nice = newVariable(mb,GDKstrdup("nice"),TYPE_lng);
sys = newVariable(mb,GDKstrdup("sys"),TYPE_lng);
idle = newVariable(mb,GDKstrdup("idle"),TYPE_lng);
iowait = newVariable(mb,GDKstrdup("iowait"),TYPE_lng);
q = newStmt(mb, "profiler", "cpustats");
q->retc= q->argc =0;
q = pushReturn(mb,q,user);
q = pushReturn(mb,q,nice);
q = pushReturn(mb,q,sys);
q = pushReturn(mb,q,idle);
q = pushReturn(mb,q,iowait);
q = newAssignment(mb);
tuples= getArg(q,0) = newVariable(mb,GDKstrdup("tuples"),TYPE_wrd);
(void) pushWrd(mb,q,1);
for (i = 1; i < limit; i++) {
p = old[i];
if (getModuleId(p)==sqlRef &&
(idcmp(getFunctionId(p),"exportValue")==0 ||
idcmp(getFunctionId(p),"exportResult")==0 ) ) {
q = newStmt(mb, "alarm", "usec");
r = newStmt1(mb, calcRef, "-");
r = pushArgument(mb, r, getArg(q,0));
r = pushArgument(mb, r, xtime);
getArg(r,0)=xtime;
q = newStmt(mb, "alarm", "usec");
rtime= getArg(q,0)= newVariable(mb,GDKstrdup("rtime"),TYPE_lng);
pushInstruction(mb,p);
continue;
}
if ( getModuleId(p) == sqlRef && idcmp(getFunctionId(p),"resultSet")==0 && isaBatType(getVarType(mb,getArg(p,3)))){
q = newStmt(mb, "aggr", "count");
getArg(q,0) = tuples;
(void) pushArgument(mb,q, getArg(p,3));
pushInstruction(mb,p);
continue;
}
if ( p->token== ENDsymbol || p->barrier == RETURNsymbol || p->barrier == YIELDsymbol){
if ( rtime == 0){
q = newStmt(mb, "alarm", "usec");
r = newStmt1(mb, calcRef, "-");
r = pushArgument(mb, r, getArg(q,0));
r = pushArgument(mb, r, xtime);
getArg(r,0)=xtime;
q = newStmt(mb, "alarm", "usec");
rtime= getArg(q,0)= newVariable(mb,GDKstrdup("rtime"),TYPE_lng);
}
q = newStmt(mb, "alarm", "usec");
r = newStmt1(mb, calcRef, "-");
r = pushArgument(mb, r, getArg(q,0));
r = pushArgument(mb, r, rtime);
getArg(r,0)=rtime;
/*
* Post execution statistics gathering
*/
q = newStmt(mb, "mtime", "current_timestamp");
finish= getArg(q,0)= newVariable(mb,GDKstrdup("finish"),TYPE_any);
q = newStmt(mb, "profiler", "cpuload");
load = newVariable(mb,GDKstrdup("load"),TYPE_int);
getArg(q,0)= load;
io = newVariable(mb,GDKstrdup("io"),TYPE_int);
q= pushReturn(mb,q,io);
q = pushArgument(mb,q,user);
q = pushArgument(mb,q,nice);
q = pushArgument(mb,q,sys);
q = pushArgument(mb,q,idle);
q = pushArgument(mb,q,iowait);
q = newStmt(mb, querylogRef, "call");
q = pushArgument(mb, q, start);
q = pushArgument(mb, q, finish);
q = pushArgument(mb, q, arg);
q = pushArgument(mb, q, tuples);
q = pushArgument(mb, q, xtime);
q = pushArgument(mb, q, rtime);
q = pushArgument(mb, q, load);
q = pushArgument(mb, q, io);
pushInstruction(mb,p);
continue;
}
pushInstruction(mb,p);
if (p->barrier == YIELDsymbol){
/* the factory yield may return */
q = newStmt(mb, "mtime", "current_timestamp");
start= getArg(q,0)= newVariable(mb,GDKstrdup("start"),TYPE_any);
q = newStmt1(mb, sqlRef, "argRecord");
for ( argc=1; argc < old[0]->argc; argc++)
q = pushArgument(mb, q, getArg(old[0],argc));
arg= getArg(q,0)= newVariable(mb,GDKstrdup("args"),TYPE_str);
q = newAssignment(mb);
q = pushLng(mb,q,0);
q = newAssignment(mb);
q = pushWrd(mb,q,0);
tuples= getArg(q,0)= newVariable(mb,GDKstrdup("tuples"),TYPE_wrd);
newFcnCall(mb,"profiler","setMemoryFlag");
q->argc--;
pushWrd(mb,q,1);
q = newStmt(mb, "alarm", "usec");
xtime = getArg(q,0)= newVariable(mb,GDKstrdup("xtime"),TYPE_lng);
}
}
for( ; i<slimit; i++)
if(old[i])
freeInstruction(old[i]);
GDKfree(old);
return 1;
}
int
OPTstrengthReductionImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
int i, j = 0, k, se= FALSE;
InstrPtr p;
int bk, ik, blk, blkbegin, blkexit, actions = 0;
InstrPtr *before, *within, *old = mb->stmt;
Lifespan span;
(void) cntxt;
(void) pci;
(void) stk; /* to fool compilers */
before = (InstrPtr *) GDKmalloc((mb->ssize + 1) * sizeof(InstrPtr));
within = (InstrPtr *) GDKmalloc((mb->ssize + 1) * sizeof(InstrPtr));
if (before== NULL || within == NULL){
if(before) GDKfree(before);
if(within) GDKfree(within);
return 0;
}
bk = 0;
ik = 0;
blk = 0;
blkexit= blkbegin = 0;
for (i = 0; i < mb->stop; i++)
before[i] = within[i] = 0;
before[bk++] = getInstrPtr(mb, 0);
span =setLifespan(mb);
if( span == NULL)
return 0;
for (i = 1; i < mb->stop - 1; i++) {
p = getInstrPtr(mb, i);
if (blockStart(p)) {
if (blkbegin == 0){
if( isLoopBarrier(mb,i) ){
blkbegin = i;
blkexit = getBlockExit(mb,i);
}
OPTDEBUGstrengthReduction
mnstr_printf(cntxt->fdout, "#check block %d-%d\n", blkbegin, blkexit);
}
within[ik++] = p;
blk++;
continue;
}
if (blockExit(p)) {
blk--;
if (blk == 0)
blkexit= blkbegin = 0;
/* move the saved instruction into place */
OPTDEBUGstrengthReduction
mnstr_printf(cntxt->fdout, "#combine both %d %d\n", bk, ik);
for (k = 0; k < ik; k++)
before[bk++] = within[k];
ik = 0;
before[bk++] = p;
continue;
}
/*
* @-
* Strength reduction is only relevant inside a block;
*/
if( blkexit == 0) {
within[ik++] = p;
continue;
}
/*
* @-
* Flow control statements may not be moved around
*/
if ( p->barrier != 0){
within[ik++] = p;
continue;
}
/*
* @-
* Limit strength reduction to the type modules and the batcalc, batstr, batcolor
* and sql.bind.
*/
if(getModuleId(p) && !isNewSource(p) ) {
within[ik++] = p;
continue;
}
/*
* @-
* Search the prospective new block and make sure that
* none of the arguments is assigned a value.
*/
for (j = ik-1; j > 0; j--) {
InstrPtr q = within[j];
for (k = 0; k < q->retc; k++)
if (SRoverwritten(p, getArg(q, k))) {
se = TRUE;
OPTDEBUGstrengthReduction
mnstr_printf(cntxt->fdout, "variable is set in loop %d\n", getArg(p, k));
goto noreduction;
}
}
/*
* @-
* Make sure the variables are not declared before the loop and used
* after the loop, because then you may not simple move an expression.
*/
for (k = 0; k < p->retc; k++)
if ( getBeginLifespan(span, getArg(p, k))<= blkbegin ||
getEndLifespan(span, getArg(p, k))> blkexit) {
se = TRUE;
OPTDEBUGstrengthReduction
mnstr_printf(cntxt->fdout, "variable %d may not be moved %d-%d\n",
getArg(p, k),getBeginLifespan(span, getArg(p, k)), getEndLifespan(span, getArg(p, k)));
goto noreduction;
}
noreduction:
OPTDEBUGstrengthReduction{
mnstr_printf(cntxt->fdout,"move %d to stack %s\n", i, (se ?"within":"before"));
printInstruction(cntxt->fdout, mb, 0, p, LIST_MAL_ALL);
}
if (blkexit && se == FALSE && !hasSideEffects(p, TRUE) && !isUpdateInstruction(p) )
before[bk++] = p;
else
within[ik++] = p;
}
actions += ik;
for (k = 0; k < ik; k++)
before[bk++] = within[k];
before[bk++] = getInstrPtr(mb, i);
GDKfree(mb->stmt);
mb->stmt = (InstrPtr *) GDKzalloc((mb->ssize) * sizeof(InstrPtr));
if ( mb->stmt == NULL){
GDKfree(span);
GDKfree(before);
GDKfree(within);
mb->stmt = old;
return 0;
}
mb->stop = 0;
OPTDEBUGstrengthReduction
mnstr_printf(cntxt->fdout,"stop= %d bk=%d\n",mb->stop,bk);
for (i = 0; i < bk; i++)
if( before[i])
pushInstruction(mb, before[i]);
GDKfree(span);
GDKfree(before);
GDKfree(within);
return actions;
}
static lng
SQLgetSpace(mvc *m, MalBlkPtr mb, int prepare)
{
sql_trans *tr = m->session->tr;
lng size,space = 0, i;
str lasttable = 0;
for (i = 0; i < mb->stop; i++) {
InstrPtr p = mb->stmt[i];
/* now deal with the update binds, it is only necessary to identify that there are updats
* The actual size is not that important */
if (getModuleId(p) == sqlRef && getFunctionId(p) == bindRef && p->retc <= 2){
char *sname = getVarConstant(mb, getArg(p, 1 + p->retc)).val.sval;
char *tname = getVarConstant(mb, getArg(p, 2 + p->retc)).val.sval;
char *cname = getVarConstant(mb, getArg(p, 3 + p->retc)).val.sval;
int access = getVarConstant(mb, getArg(p, 4 + p->retc)).val.ival;
sql_schema *s = mvc_bind_schema(m, sname);
sql_table *t = 0;
sql_column *c = 0;
if (!s || strcmp(s->base.name, dt_schema) == 0)
continue;
t = mvc_bind_table(m, s, tname);
if (!t)
continue;
c = mvc_bind_column(m, t, cname);
if (!s)
continue;
/* we have to sum the cost of all three components of a BAT */
if (c && (!isRemote(c->t) && !isMergeTable(c->t)) && (lasttable == 0 || strcmp(lasttable,tname)==0)) {
size = SQLgetColumnSize(tr, c, access);
space += size; // accumulate once per table
//lasttable = tname; invalidate this attempt
if( !prepare && size == 0 && ! t->system){
//mnstr_printf(GDKout,"found empty column %s.%s.%s prepare %d size "LLFMT"\n",sname,tname,cname,prepare,size);
setFunctionId(p, emptybindRef);
}
}
}
if (getModuleId(p) == sqlRef && (getFunctionId(p) == bindidxRef)) {
char *sname = getVarConstant(mb, getArg(p, 1 + p->retc)).val.sval;
//char *tname = getVarConstant(mb, getArg(p, 2 + p->retc)).val.sval;
char *idxname = getVarConstant(mb, getArg(p, 3 + p->retc)).val.sval;
int access = getVarConstant(mb, getArg(p, 4 + p->retc)).val.ival;
sql_schema *s = mvc_bind_schema(m, sname);
BAT *b;
if (getFunctionId(p) == bindidxRef) {
sql_idx *i = mvc_bind_idx(m, s, idxname);
if (i && (!isRemote(i->t) && !isMergeTable(i->t))) {
b = store_funcs.bind_idx(tr, i, RDONLY);
if (b) {
space += (size =getBatSpace(b));
if (!size) {
sql_column *c = i->t->columns.set->h->data;
size = SQLgetColumnSize(tr, c, access);
}
if( !prepare && size == 0 && ! i->t->system){
setFunctionId(p, emptybindidxRef);
//mnstr_printf(GDKout,"found empty column %s.%s.%s prepare %d size "LLFMT"\n",sname,tname,idxname,prepare,size);
}
BBPunfix(b->batCacheid);
}
}
}
}
}
return space;
}
int
inlineMALblock(MalBlkPtr mb, int pc, MalBlkPtr mc)
{
int i, k, l, n;
InstrPtr *ns, p,q;
int *nv, *np = NULL;
p = getInstrPtr(mb, pc);
q = getInstrPtr(mc, 0);
ns = GDKzalloc((l = (mb->ssize + mc->ssize + p->retc - 3)) * sizeof(InstrPtr));
if (ns == NULL)
return -1;
if ( mc->ptop > 0){
np = (int*) GDKmalloc(mc->ptop * sizeof(int));
if (np == 0){
GDKfree(ns);
return -1;
}
}
nv = (int*) GDKmalloc(mc->vtop * sizeof(int));
if (nv == 0){
GDKfree(ns);
if( np)
GDKfree(np);
return -1;
}
/* add all properties of the new block to the target environment */
for (n = 0; n < mc->ptop; n++) {
int propid = newProperty(mb);
if (propid < 0) {
assert(0);
return -1;
}
np[n] = propid;
mb->prps[propid].idx = mc->prps[n].idx;
mb->prps[propid].op = mc->prps[n].op;
mb->prps[propid].var = mc->prps[n].var; /* fixed later */
}
/* add all variables of the new block to the target environment */
for (n = 0; n < mc->vtop; n++) {
VarPtr ov, v;
if (isExceptionVariable(mc->var[n]->name)) {
nv[n] = newVariable(mb,GDKstrdup(mc->var[n]->name),TYPE_str);
if (isVarUDFtype(mc,n))
setVarUDFtype(mb,nv[n]);
if (isVarUsed(mc,n))
setVarUsed(mb,nv[n]);
} else if (isVarTypedef(mc,n)) {
nv[n] = newTypeVariable(mb,getVarType(mc,n));
} else if (isVarConstant(mc,n)) {
nv[n] = cpyConstant(mb,getVar(mc,n));
} else {
nv[n] = newTmpVariable(mb, getVarType(mc, n));
if (isVarUDFtype(mc,n))
setVarUDFtype(mb,nv[n]);
if (isVarUsed(mc,n))
setVarUsed(mb,nv[n]);
}
/* remap the properties */
ov = getVar(mc, n);
v = getVar(mb, nv[n]);
if (ov->propc > v->maxprop) {
int size = sizeof(VarRecord);
VarPtr vnew = (VarPtr) GDKzalloc(size + ov->propc * sizeof(int));
memcpy((char*) vnew, (char*) v, size);
vnew->maxprop = ov->propc;
mb->var[nv[n]] = vnew;
GDKfree(v);
v = getVar(mb, nv[n]);
}
for (i = 0; i < ov->propc; i++)
v->prps[i] = np[ov->prps[i]];
v->propc = ov->propc;
}
/* change the property variables to the new context */
for (n = 0; n < mc->ptop; n++) {
if (mc->prps[n].var)
mb->prps[np[n]].var = nv[mc->prps[n].var];
assert( mb->prps[np[n]].var >= 0);
}
/* use an alias mapping to keep track of the actual arguments */
for (n = p->retc; n < p->argc; n++)
nv[getArg(q,n)] = getArg(p, n);
k = 0;
/* find the return statement of the inline function */
for (i = 1; i < mc->stop - 1; i++) {
q = mc->stmt[i];
if( q->barrier== RETURNsymbol || q->barrier== YIELDsymbol){
/* add the mapping of the return variables */
for(n=0; n<p->retc; n++)
nv[getArg(q,n)] = getArg(p,n);
}
}
/* copy the stable part */
for (i = 0; i < pc; i++)
ns[k++] = mb->stmt[i];
for (i = 1; i < mc->stop - 1; i++) {
q = mc->stmt[i];
if( q->token == ENDsymbol)
break;
/* copy the instruction and fix variable references */
ns[k] = copyInstruction(q);
for (n = 0; n < q->argc; n++)
getArg(ns[k], n) = nv[getArg(q, n)];
if (q->barrier == RETURNsymbol || q->barrier == YIELDsymbol) {
for(n=0; n<q->retc; n++)
clrVarFixed(mb,getArg(ns[k],n)); /* for typing */
setModuleId(ns[k],getModuleId(q));
setFunctionId(ns[k],getFunctionId(q));
ns[k]->barrier = 0;
ns[k]->token = ASSIGNsymbol;
}
k++;
}
/* copy the remainder of the stable part */
freeInstruction(p);
for (i = pc + 1; i < mb->stop; i++){
ns[k++] = mb->stmt[i];
}
/* remove any free instruction */
for(; i<mb->ssize; i++)
if( mb->stmt[i]){
freeInstruction(mb->stmt[i]);
mb->stmt[i]= 0;
}
GDKfree(mb->stmt);
mb->stmt = ns;
mb->ssize = l;
mb->stop = k;
GDKfree(np);
GDKfree(nv);
return pc;
}
