mirror of
https://github.com/Unidata/netcdf-c.git
synced 2024-12-21 08:39:46 +08:00
fb40a72b45
re: Issue https://github.com/Unidata/netcdf-c/issues/2685 re: PR https://github.com/Unidata/netcdf-c/pull/2179 As noted in PR https://github.com/Unidata/netcdf-c/pull/2179, the old code did not allow for reclaiming instances of types, nor for properly copying them. That PR provided new functions capable of reclaiming/copying instances of arbitrary types. However, as noted by Issue https://github.com/Unidata/netcdf-c/issues/2685, using these most general functions resulted in a significant performance degradation, even for common cases. This PR attempts to mitigate the cost of using the general reclaim/copy functions in two ways. First, the previous functions operating at the top level by using ncid and typeid arguments. These functions were augmented with equivalent versions that used the netcdf-c library internal data structures to allow direct access to needed information. These new functions are used internally to the library. The second mitigation involves optimizing the internal functions by providing early tests for common cases. This avoids unnecessary recursive function calls. The overall result is a significant improvement in speed by a factor of roughly twenty -- your mileage may vary. These optimized functions are still not as fast as the original (more limited) functions, but they are getting close. Additional optimizations are possible. But the cost is a significant "uglification" of the code that I deemed a step too far, at least for now. ## Misc. Changes 1. Added a test case to check the proper reclamation/copy of complex types. 2. Found and fixed some places where nc_reclaim/copy should have been used. 3. Replaced, in the netcdf-c library, (almost all) occurrences of nc_reclaim_copy with calls to NC_reclaim/copy. This plus the optimizations is the primary speed-up mechanism. 4. In DAP4, the metadata is held in a substrate in-memory file; this required some changes so that the reclaim/copy code accessed that substrate dispatcher rather than the DAP4 dispatcher. 5. Re-factored and isolated the code that computes if a type is (transitively) variable-sized or not. 6. Clean up the reclamation code in ncgen; adding the use of nc_reclaim exposed some memory problems.
496 lines
16 KiB
C
496 lines
16 KiB
C
/*********************************************************************
|
|
* Copyright 2018, UCAR/Unidata
|
|
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
|
|
*********************************************************************/
|
|
|
|
#include "includes.h"
|
|
#include "ncoffsets.h"
|
|
#include "netcdf_aux.h"
|
|
|
|
/**************************************************/
|
|
/* Code for generating data lists*/
|
|
/**************************************************/
|
|
/* For datalist constant rules: see the rules on the man page */
|
|
|
|
/* Forward*/
|
|
static void generate_array(Symbol*,Bytebuffer*,Datalist*,Generator*,Writer);
|
|
static void generate_primdata(Symbol*, NCConstant*, Bytebuffer*, Datalist* fillsrc, Generator*);
|
|
static void generate_fieldarray(Symbol*, NCConstant*, Dimset*, Bytebuffer*, Datalist* fillsrc, Generator*);
|
|
|
|
/* Mnemonics */
|
|
#define VLENLIST1
|
|
#define FIELDARRAY 1
|
|
|
|
#define ITER_BUFSIZE_DEFAULT (2<<20)
|
|
|
|
void
|
|
pvec(int rank, size_t* vector)
|
|
{
|
|
int i;
|
|
fprintf(stderr,"(");
|
|
for(i=0;i<rank;i++)
|
|
fprintf(stderr," %lu",(long)vector[i]);
|
|
fprintf(stderr,")");
|
|
}
|
|
|
|
/**************************************************/
|
|
|
|
|
|
int
|
|
generator_getstate(Generator* generator ,void** statep)
|
|
{
|
|
if(statep) *statep = (void*)generator->globalstate;
|
|
return 1;
|
|
}
|
|
|
|
int generator_reset(Generator* generator, void* state)
|
|
{
|
|
generator->globalstate = state;
|
|
return 1;
|
|
}
|
|
/**************************************************/
|
|
|
|
void
|
|
generate_attrdata(Symbol* asym, Generator* generator, Writer writer, Bytebuffer* codebuf)
|
|
{
|
|
Symbol* basetype = asym->typ.basetype;
|
|
nc_type typecode = basetype->typ.typecode;
|
|
|
|
if(typecode == NC_CHAR) {
|
|
gen_charattr(asym->data,codebuf);
|
|
} else {
|
|
int uid;
|
|
size_t count;
|
|
generator->listbegin(generator,asym,NULL,LISTATTR,asym->data->length,codebuf,&uid);
|
|
for(count=0;count<asym->data->length;count++) {
|
|
NCConstant* con = datalistith(asym->data,count);
|
|
generator->list(generator,asym,NULL,LISTATTR,uid,count,codebuf);
|
|
generate_basetype(asym->typ.basetype,con,codebuf,NULL,generator);
|
|
}
|
|
generator->listend(generator,asym,NULL,LISTATTR,uid,count,codebuf);
|
|
}
|
|
writer(generator,asym,codebuf,0,NULL,NULL);
|
|
}
|
|
|
|
void
|
|
generate_vardata(Symbol* vsym, Generator* generator, Writer writer, Bytebuffer* code)
|
|
{
|
|
Dimset* dimset = &vsym->typ.dimset;
|
|
int rank = dimset->ndims;
|
|
Symbol* basetype = vsym->typ.basetype;
|
|
Datalist* filler = getfiller(vsym);
|
|
|
|
if(vsym->data == NULL) return;
|
|
|
|
if(rank == 0) {/*scalar case*/
|
|
NCConstant* c0 = datalistith(vsym->data,0);
|
|
generate_basetype(basetype,c0,code,filler,generator);
|
|
writer(generator,vsym,code,0,NULL,NULL);
|
|
} else {/*rank > 0*/
|
|
generate_array(vsym,code,filler,generator,writer);
|
|
}
|
|
}
|
|
|
|
/* Generate an instance of the basetype using the value of con*/
|
|
void
|
|
generate_basetype(Symbol* tsym, NCConstant* con, Bytebuffer* codebuf, Datalist* filler, Generator* generator)
|
|
{
|
|
Datalist* data;
|
|
int offsetbase = 0;
|
|
|
|
switch (tsym->subclass) {
|
|
|
|
case NC_ENUM:
|
|
case NC_OPAQUE:
|
|
case NC_PRIM:
|
|
if(con == NULL || isfillconst(con)) {
|
|
Datalist* fill = (filler==NULL?getfiller(tsym):filler);
|
|
ASSERT(fill->length == 1);
|
|
con = datalistith(fill,0);
|
|
}
|
|
if(islistconst(con)) {
|
|
semerror(constline(con),"Expected primitive found {..}");
|
|
}
|
|
generate_primdata(tsym,con,codebuf,filler,generator);
|
|
break;
|
|
|
|
case NC_COMPOUND: {
|
|
int i,uid, nfields, dllen;
|
|
if(con == NULL || isfillconst(con)) {
|
|
Datalist* fill = (filler==NULL?getfiller(tsym):filler);
|
|
ASSERT(fill->length == 1);
|
|
con = fill->data[0];
|
|
if(!islistconst(con)) {
|
|
if(con)
|
|
semerror(con->lineno,"Compound data fill value is not enclosed in {..}");
|
|
else
|
|
semerror(0,"Compound data fill value not enclosed in {..}, con is NULL.");
|
|
}
|
|
}
|
|
|
|
if(!con) { /* fail on null compound. */
|
|
semerror(constline(con),"NULL compound data.");
|
|
break;
|
|
}
|
|
|
|
if(!islistconst(con)) {/* fail on no compound*/
|
|
semerror(constline(con),"Compound data must be enclosed in {..}");
|
|
}
|
|
|
|
data = con->value.compoundv;
|
|
nfields = listlength(tsym->subnodes);
|
|
dllen = datalistlen(data);
|
|
if(dllen > nfields) {
|
|
semerror(con->lineno,"Datalist longer than the number of compound fields");
|
|
break;
|
|
}
|
|
generator->listbegin(generator,tsym,&offsetbase,LISTCOMPOUND,listlength(tsym->subnodes),codebuf,&uid);
|
|
for(i=0;i<nfields;i++) {
|
|
Symbol* field = (Symbol*)listget(tsym->subnodes,i);
|
|
con = datalistith(data,i);
|
|
generator->list(generator,field,&offsetbase,LISTCOMPOUND,uid,i,codebuf);
|
|
generate_basetype(field,con,codebuf,NULL,generator);
|
|
}
|
|
generator->listend(generator,tsym,&offsetbase,LISTCOMPOUND,uid,i,codebuf);
|
|
} break;
|
|
|
|
case NC_VLEN: {
|
|
Bytebuffer* vlenbuf;
|
|
int uid;
|
|
size_t count;
|
|
|
|
if(con == NULL || isfillconst(con)) {
|
|
Datalist* fill = (filler==NULL?getfiller(tsym):filler);
|
|
ASSERT(fill->length == 1);
|
|
con = fill->data[0];
|
|
if(con->nctype != NC_COMPOUND) {
|
|
semerror(con->lineno,"Vlen data fill value is not enclosed in {..}");
|
|
}
|
|
}
|
|
|
|
if(!islistconst(con)) {
|
|
semerror(constline(con),"Vlen data must be enclosed in {..}");
|
|
}
|
|
data = con->value.compoundv;
|
|
/* generate the nc_vlen_t instance*/
|
|
vlenbuf = bbNew();
|
|
if(tsym->typ.basetype->typ.typecode == NC_CHAR) {
|
|
gen_charseq(data,vlenbuf);
|
|
generator->vlenstring(generator,tsym,vlenbuf,&uid,&count);
|
|
} else {
|
|
generator->listbegin(generator,tsym,NULL,LISTVLEN,data->length,codebuf,&uid);
|
|
for(count=0;count<data->length;count++) {
|
|
NCConstant* con;
|
|
generator->list(generator,tsym,NULL,LISTVLEN,uid,count,vlenbuf);
|
|
con = datalistith(data,count);
|
|
generate_basetype(tsym->typ.basetype,con,vlenbuf,NULL,generator);
|
|
}
|
|
generator->listend(generator,tsym,NULL,LISTVLEN,uid,count,codebuf,(void*)vlenbuf);
|
|
}
|
|
generator->vlendecl(generator,tsym,codebuf,uid,count,vlenbuf); /* Will extract contents of vlenbuf */
|
|
bbFree(vlenbuf);
|
|
} break;
|
|
|
|
case NC_FIELD:
|
|
if(tsym->typ.dimset.ndims > 0) {
|
|
/* Verify that we have a sublist (or fill situation) */
|
|
if(con != NULL && !isfillconst(con) && !islistconst(con))
|
|
semerror(constline(con),"Dimensioned fields must be enclose in {...}");
|
|
generate_fieldarray(tsym->typ.basetype,con,&tsym->typ.dimset,codebuf,filler,generator);
|
|
} else {
|
|
generate_basetype(tsym->typ.basetype,con,codebuf,NULL,generator);
|
|
}
|
|
break;
|
|
|
|
default: PANIC1("generate_basetype: unexpected subclass %d",tsym->subclass);
|
|
}
|
|
}
|
|
|
|
/* Used only for structure field arrays*/
|
|
static void
|
|
generate_fieldarray(Symbol* basetype, NCConstant* con, Dimset* dimset,
|
|
Bytebuffer* codebuf, Datalist* filler, Generator* generator)
|
|
{
|
|
int i;
|
|
int chartype = (basetype->typ.typecode == NC_CHAR);
|
|
Datalist* data;
|
|
int rank = rankfor(dimset);
|
|
|
|
ASSERT(dimset->ndims > 0);
|
|
|
|
if(con != NULL && !isfillconst(con))
|
|
data = con->value.compoundv;
|
|
else
|
|
data = NULL;
|
|
|
|
if(chartype) {
|
|
Bytebuffer* charbuf = bbNew();
|
|
gen_chararray(dimset,0,data,charbuf,filler);
|
|
generator->charconstant(generator,basetype,codebuf,charbuf);
|
|
bbFree(charbuf);
|
|
} else {
|
|
int uid;
|
|
size_t xproduct = crossproduct(dimset,0,rank); /* compute total number of elements */
|
|
generator->listbegin(generator,basetype,NULL,LISTFIELDARRAY,xproduct,codebuf,&uid);
|
|
for(i=0;i<xproduct;i++) {
|
|
con = (data == NULL ? NULL : datalistith(data,i));
|
|
generator->list(generator,basetype,NULL,LISTFIELDARRAY,uid,i,codebuf);
|
|
generate_basetype(basetype,con,codebuf,NULL,generator);
|
|
}
|
|
generator->listend(generator,basetype,NULL,LISTFIELDARRAY,uid,i,codebuf);
|
|
}
|
|
}
|
|
|
|
|
|
/* An opaque string value might not conform
|
|
to the size of the opaque to which it is being
|
|
assigned. Normalize it to match the required
|
|
opaque length (in bytes).
|
|
Note that the string is a sequence of nibbles (4 bits).
|
|
*/
|
|
static void
|
|
normalizeopaquelength(NCConstant* prim, unsigned long nbytes)
|
|
{
|
|
int nnibs = 2*nbytes;
|
|
ASSERT(prim->nctype==NC_OPAQUE);
|
|
if(prim->value.opaquev.len == nnibs) {
|
|
/* do nothing*/
|
|
} else if(prim->value.opaquev.len > nnibs) { /* truncate*/
|
|
prim->value.opaquev.stringv[nnibs] = '\0';
|
|
prim->value.opaquev.len = nnibs;
|
|
} else {/* prim->value.opaquev.len < nnibs => expand*/
|
|
char* s;
|
|
s = (char*)ecalloc(nnibs+1);
|
|
memset(s,'0',nnibs); /* Fill with '0' characters */
|
|
memcpy(s,prim->value.opaquev.stringv,prim->value.opaquev.len);
|
|
s[nnibs] = '\0';
|
|
efree(prim->value.opaquev.stringv);
|
|
prim->value.opaquev.stringv=s;
|
|
prim->value.opaquev.len = nnibs;
|
|
}
|
|
}
|
|
|
|
static void
|
|
generate_primdata(Symbol* basetype, NCConstant* prim, Bytebuffer* codebuf,
|
|
Datalist* filler, Generator* generator)
|
|
{
|
|
NCConstant* target;
|
|
int match;
|
|
|
|
if(prim == NULL || isfillconst(prim)) {
|
|
Datalist* fill = (filler==NULL?getfiller(basetype):filler);
|
|
ASSERT(fill->length == 1);
|
|
prim = datalistith(fill,0);
|
|
}
|
|
|
|
ASSERT((prim->nctype != NC_COMPOUND));
|
|
|
|
/* Verify that the constant is consistent with the type */
|
|
match = 1;
|
|
switch (prim->nctype) {
|
|
case NC_CHAR:
|
|
case NC_BYTE:
|
|
case NC_SHORT:
|
|
case NC_INT:
|
|
case NC_FLOAT:
|
|
case NC_DOUBLE:
|
|
case NC_UBYTE:
|
|
case NC_USHORT:
|
|
case NC_UINT:
|
|
case NC_INT64:
|
|
case NC_UINT64:
|
|
case NC_STRING:
|
|
match = (basetype->subclass == NC_PRIM ? 1 : 0);
|
|
break;
|
|
|
|
#ifdef USE_NETCDF4
|
|
case NC_NIL:
|
|
match = (basetype->subclass == NC_PRIM && basetype->typ.typecode == NC_STRING ? 1 : 0);
|
|
break;
|
|
|
|
case NC_OPAQUE:
|
|
/* OPAQUE is also consistent with numbers */
|
|
match = (basetype->subclass == NC_OPAQUE
|
|
|| basetype->subclass == NC_PRIM ? 1 : 0);
|
|
break;
|
|
case NC_ECONST:
|
|
match = (basetype->subclass == NC_ENUM ? 1 : 0);
|
|
if(match) {
|
|
/* Make sure this econst belongs to this enum */
|
|
Symbol* ec = prim->value.enumv;
|
|
Symbol* en = ec->container;
|
|
match = (en == basetype);
|
|
}
|
|
break;
|
|
#endif
|
|
default:
|
|
match = 0;
|
|
}
|
|
if(!match) {
|
|
semerror(constline(prim),"Data value is not consistent with the expected type: %s",
|
|
basetype->name);
|
|
}
|
|
|
|
target = nullconst();
|
|
target->nctype = basetype->typ.typecode;
|
|
|
|
if(target->nctype != NC_ECONST) {
|
|
convert1(prim,target);
|
|
}
|
|
|
|
switch (target->nctype) {
|
|
case NC_ECONST:
|
|
if(basetype->subclass != NC_ENUM) {
|
|
semerror(constline(prim),"Conversion to enum not supported (yet)");
|
|
} break;
|
|
case NC_OPAQUE:
|
|
normalizeopaquelength(target,basetype->typ.size);
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
generator->constant(generator,basetype,target,codebuf);
|
|
reclaimconstant(target);
|
|
target = NULL;
|
|
return;
|
|
}
|
|
|
|
/* Avoid long argument lists */
|
|
struct Args {
|
|
Symbol* vsym;
|
|
Dimset* dimset;
|
|
int typecode;
|
|
int storage;
|
|
int rank;
|
|
Generator* generator;
|
|
Writer writer;
|
|
Bytebuffer* code;
|
|
Datalist* filler;
|
|
size_t dimsizes[NC_MAX_VAR_DIMS];
|
|
size_t chunksizes[NC_MAX_VAR_DIMS];
|
|
};
|
|
|
|
static void
|
|
generate_arrayR(struct Args* args, int dimindex, size_t* index, Datalist* data)
|
|
{
|
|
size_t counter,stop;
|
|
size_t count[NC_MAX_VAR_DIMS];
|
|
Datalist* actual;
|
|
Symbol* dim = args->dimset->dimsyms[dimindex];
|
|
|
|
stop = args->dimsizes[dimindex];
|
|
|
|
/* Four cases: (dimindex==rank-1|dimindex<rank-1) X (unlimited|!unlimited) */
|
|
if(dimindex == (args->rank - 1)) {/* base case */
|
|
int uid;
|
|
if(dimindex > 0 && dim->dim.isunlimited) {
|
|
/* Get the unlimited list */
|
|
NCConstant* con = datalistith(data,0);
|
|
actual = compoundfor(con);
|
|
} else
|
|
actual = data;
|
|
/* For last index, dump all of its elements */
|
|
args->generator->listbegin(args->generator,args->vsym,NULL,LISTDATA,datalistlen(actual),args->code,&uid);
|
|
for(counter=0;counter<stop;counter++) {
|
|
NCConstant* con = datalistith(actual,counter);
|
|
generate_basetype(args->vsym->typ.basetype,con,args->code,args->filler,args->generator);
|
|
args->generator->list(args->generator,args->vsym,NULL,LISTDATA,uid,counter,args->code);
|
|
}
|
|
args->generator->listend(args->generator,args->vsym,NULL,LISTDATA,uid,counter,args->code);
|
|
memcpy(count,onesvector,sizeof(size_t)*dimindex);
|
|
count[dimindex] = stop;
|
|
/* Write the data; also reclaims written data */
|
|
args->writer(args->generator,args->vsym,args->code,args->rank,index,count);
|
|
bbClear(args->code);
|
|
} else {
|
|
actual = data;
|
|
/* Iterate over this dimension */
|
|
for(counter = 0;counter < stop; counter++) {
|
|
Datalist* subdata = NULL;
|
|
NCConstant* con = datalistith(actual,counter);
|
|
if(con == NULL)
|
|
subdata = filldatalist;
|
|
else {
|
|
ASSERT(islistconst(con));
|
|
if(islistconst(con)) subdata = compoundfor(con);
|
|
}
|
|
index[dimindex] = counter;
|
|
generate_arrayR(args,dimindex+1,index,subdata); /* recurse */
|
|
}
|
|
}
|
|
}
|
|
|
|
static void
|
|
generate_array(Symbol* vsym, Bytebuffer* code, Datalist* filler, Generator* generator, Writer writer)
|
|
{
|
|
int i;
|
|
size_t index[NC_MAX_VAR_DIMS];
|
|
struct Args args;
|
|
size_t totalsize;
|
|
int nunlimited = 0;
|
|
|
|
assert(vsym->typ.dimset.ndims > 0);
|
|
|
|
args.vsym = vsym;
|
|
args.dimset = &vsym->typ.dimset;
|
|
args.generator = generator;
|
|
args.writer = writer;
|
|
args.filler = filler;
|
|
args.code = code;
|
|
args.rank = args.dimset->ndims;
|
|
args.storage = vsym->var.special._Storage;
|
|
args.typecode = vsym->typ.basetype->typ.typecode;
|
|
|
|
assert(args.rank > 0);
|
|
|
|
totalsize = 1; /* total # elements in the array */
|
|
for(i=0;i<args.rank;i++) {
|
|
args.dimsizes[i] = args.dimset->dimsyms[i]->dim.declsize;
|
|
totalsize *= args.dimsizes[i];
|
|
}
|
|
nunlimited = countunlimited(args.dimset);
|
|
|
|
if(vsym->var.special._Storage == NC_CHUNKED) {
|
|
if(vsym->var.special._ChunkSizes)
|
|
memcpy(args.chunksizes,vsym->var.special._ChunkSizes,sizeof(size_t)*args.rank);
|
|
}
|
|
|
|
memset(index,0,sizeof(index));
|
|
|
|
/* Special case for NC_CHAR */
|
|
if(args.typecode == NC_CHAR) {
|
|
size_t start[NC_MAX_VAR_DIMS];
|
|
size_t count[NC_MAX_VAR_DIMS];
|
|
Bytebuffer* charbuf = bbNew();
|
|
gen_chararray(args.dimset,0,args.vsym->data,charbuf,args.filler);
|
|
args.generator->charconstant(args.generator,args.vsym,args.code,charbuf);
|
|
memset(start,0,sizeof(size_t)*args.rank);
|
|
memcpy(count,args.dimsizes,sizeof(size_t)*args.rank);
|
|
args.writer(args.generator,args.vsym,args.code,args.rank,start,count);
|
|
bbFree(charbuf);
|
|
bbClear(args.code);
|
|
return;
|
|
}
|
|
|
|
/* If the total no. of elements is less than some max and no unlimited,
|
|
then generate a single vara that covers the whole array */
|
|
if(totalsize <= wholevarsize && nunlimited == 0) {
|
|
Symbol* basetype = args.vsym->typ.basetype;
|
|
size_t counter;
|
|
int uid;
|
|
Datalist* flat = flatten(vsym->data,args.rank);
|
|
args.generator->listbegin(args.generator,basetype,NULL,LISTDATA,totalsize,args.code,&uid);
|
|
for(counter=0;counter<totalsize;counter++) {
|
|
NCConstant* con = datalistith(flat,counter);
|
|
if(con == NULL)
|
|
con = &fillconstant;
|
|
generate_basetype(basetype,con,args.code,args.filler,args.generator);
|
|
args.generator->list(args.generator,args.vsym,NULL,LISTDATA,uid,counter,args.code);
|
|
}
|
|
args.generator->listend(args.generator,args.vsym,NULL,LISTDATA,uid,counter,args.code);
|
|
args.writer(args.generator,args.vsym,args.code,args.rank,zerosvector,args.dimsizes);
|
|
freedatalist(flat);
|
|
} else
|
|
generate_arrayR(&args, 0, index, vsym->data);
|
|
}
|