bool_t diff_reg(mem_t oldr, mem_t newr, FILE *outfile)
{
word_t pos;
int len = oldr->len;
bool_t diff = FALSE;
if (newr->len < len)
len = newr->len;
//write back
if(oldr->ca) oldr->ca->clear(TRUE);
if(newr->ca) newr->ca->clear(TRUE);
for (pos = 0; (!diff || outfile) && pos < len; pos += 4) {
word_t ov = 0;
word_t nv = 0;
oldr->getWord(pos, &ov);
newr->getWord(pos, &nv);
if (nv != ov) {
diff = TRUE;
if (outfile)
fprintf(outfile, "%s:\t0x%.8x\t0x%.8x\n",
reg_table[pos/4].name, ov, nv);
}
}
return diff;
}
bool_t diff_mem(mem_t oldm, mem_t newm, FILE *outfile)
{
word_t pos;
int len = oldm->len;
bool_t diff = FALSE;
if (newm->len < len)
len = newm->len;
// write back
if(oldm->ca) oldm->ca->clear(TRUE);
if(newm->ca) newm->ca->clear(TRUE);
//remove bus
/*
if(oldm->bus) oldm->bus->remove(oldm);
if(newm->bus) newm->bus->remove(newm);
*/
for (pos = 0; (!diff || outfile) && pos+4 <= len; pos += 4) {
word_t ov = 0; word_t nv = 0;
oldm->getWord(pos, &ov);
newm->getWord(pos, &nv);
if (nv != ov) {
diff = TRUE;
if (outfile)
fprintf(outfile, "0x%.4x:\t0x%.8x\t0x%.8x\n", pos, ov, nv);
}
}
//return *oldm->m != *newm->m;
return diff;
}
void pinnedFree(T *ptr)
{
mem_iter iter = pinned_maps.find((void *)ptr);
if (iter != pinned_maps.end()) {
iter->second.is_free = true;
pinned_used_bytes -= iter->second.bytes;
} else {
pinnedFreeWrapper(ptr); // Free it because we are not sure what the size is
}
}
virtual void solve(int nt) final
{
// multiple calls to sovlve() are meant to advance the solution by nt
nt += timestep;
// being generous about out-of-loop barriers
if (timestep == 0)
{
mem->barrier();
#if !defined(NDEBUG)
hook_ante_loop_called = false;
#endif
hook_ante_loop(nt);
mem->barrier();
}
// moved here so that if an exception is thrown from hook_ante_loop these do not cause complaints
#if !defined(NDEBUG)
hook_ante_step_called = false;
hook_post_step_called = false;
#endif
while (timestep < nt)
{
// progress-bar info through thread name (check top -H)
monitor(float(timestep) / nt); // TODO: does this value make sanse with repeated advence() calls?
// might be used to implement multi-threaded signal handling
mem->barrier();
if (mem->panic) break;
// proper solver stuff
hook_ante_step();
for (int e = 0; e < n_eqns; ++e) scale(e, ct_params_t::hint_scale(e));
for (int e = 0; e < n_eqns; ++e) xchng(e);
for (int e = 0; e < n_eqns; ++e)
{
advop(e);
if (e != n_eqns - 1) mem->barrier();
}
for (int e = 0; e < n_eqns; ++e) cycle(e); // note: cycle assumes ascending loop index
for (int e = 0; e < n_eqns; ++e) scale(e, -ct_params_t::hint_scale(e));
timestep++;
hook_post_step();
}
mem->barrier();
// note: hook_post_loop was removed as conficling with multiple-advance()-call logic
}
void pinnedGarbageCollect()
{
for(mem_iter iter = pinned_maps.begin(); iter != pinned_maps.end(); ++iter) {
if ((iter->second).is_free) {
pinnedFreeWrapper(iter->first);
}
}
mem_iter memory_curr = pinned_maps.begin();
mem_iter memory_end = pinned_maps.end();
while(memory_curr != memory_end) {
if (memory_curr->second.is_free) {
pinned_maps.erase(memory_curr++);
} else {
++memory_curr;
}
}
}
T* pinnedAlloc(const size_t &elements)
{
managerInit();
T* ptr = NULL;
// Allocate the higher megabyte. Overhead of creating pinned memory is
// more so we want more resuable memory.
size_t alloc_bytes = divup(sizeof(T) * elements, 1048576) * 1048576;
if (elements > 0) {
// FIXME: Add better checks for garbage collection
// Perhaps look at total memory available as a metric
if (pinned_maps.size() >= MAX_BUFFERS || pinned_used_bytes >= MAX_BYTES) {
pinnedGarbageCollect();
}
for(mem_iter iter = pinned_maps.begin();
iter != pinned_maps.end(); ++iter) {
mem_info info = iter->second;
if (info.is_free && info.bytes == alloc_bytes) {
iter->second.is_free = false;
pinned_used_bytes += alloc_bytes;
return (T *)iter->first;
}
}
// Perform garbage collection if memory can not be allocated
if (cudaMallocHost((void **)&ptr, alloc_bytes) != cudaSuccess) {
pinnedGarbageCollect();
CUDA_CHECK(cudaMallocHost((void **)(&ptr), alloc_bytes));
}
mem_info info = {false, false, alloc_bytes};
pinned_maps[ptr] = info;
pinned_used_bytes += alloc_bytes;
}
return (T*)ptr;
}
/* CACHE */
CacheRec::CacheRec(mem_t target, int b, int s, int E, int t):
cb(b),cB(1<<b),cs(s),cS(1<<s),cE(E),ct(t),nlines(E<<s){
len = target->getLen();
tm = target;
contents = new byte_t[nlines<<b];
valid = new bool_t[nlines];
dirty = new bool_t[nlines];
tags = new byte_t[nlines];
clear();
#ifdef CHECK_CACHE
checker = copy_mem(target);
#else
checker = NULL;
#endif
}
virtual void cycle(int e) final
{
n[e] = (n[e] + 1) % n_tlev - n_tlev; // -n_tlev so that n+1 does not give out of bounds
if (e == n_eqns - 1) mem->cycle(rank);
}
bool exec( int instruction )
{
// printf( "EXEC: %d\n", instruction );
switch( instruction )
{
// moo
case 0:
{
if( prog_pos == program.begin() )
quit( true );
prog_pos--; // skip previous command.
int level = 1;
while( level > 0 )
{
if( prog_pos == program.begin() )
break;
prog_pos--;
if( (*prog_pos) == 0 )
level++;
else
if( (*prog_pos) == 7 ) // look for MOO
level--;
}
if( level != 0 )
quit(true);
return exec( *prog_pos );
}
// mOo
case 1:
if( mem_poses[stomach] == memory[stomach].begin() )
quit( true );
else
mem_poses[stomach]--;
break;
// moO
case 2:
mem_poses[stomach]++;
if( mem_poses[stomach] == memory[stomach].end() )
{
memory[stomach].push_back(0);
mem_poses[stomach] = memory[stomach].end();
mem_poses[stomach]--;
}
break;
// mOO
case 3:
if( (*mem_poses[stomach]) == 3 )
quit( false );
return exec(*mem_poses[stomach]);
// Moo
case 4:
if( (*mem_poses[stomach]) != 0 )
printf( "%c", *mem_poses[stomach] );
else
{
(*mem_poses[stomach]) = getchar();
while( getchar() != '\n' );
}
break;
// MOo
case 5:
(*mem_poses[stomach])--;
break;
// MoO
case 6:
(*mem_poses[stomach])++;
break;
// MOO
case 7:
if( (*mem_poses[stomach]) == 0 )
{
int level = 1;
int prev = 0;
prog_pos++; // have to skip past next command when looking for next moo.
if( prog_pos == program.end() )
break;
while( level > 0 )
{
prev = *prog_pos;
prog_pos++;
if( prog_pos == program.end() )
break;
if( (*prog_pos) == 7 )
level++;
else
if( (*prog_pos) == 0 ) // look for moo command.
{
level--;
if( prev == 7 )
level--;
}
}
if( level != 0 )
quit( true );
}
break;
// OOO
case 8:
(*mem_poses[stomach]) = 0;
break;
// MMM
case 9:
if( has_register_val )
(*mem_poses[stomach]) = register_val;
else
register_val = (*mem_poses[stomach]);
has_register_val = !has_register_val;
break;
// OOM
case 10:
printf( "%d\n", *mem_poses[stomach] );
break;
// oom
case 11:
{
char buf[100];
unsigned int c = 0;
while( c < sizeof(buf)-1 )
{
buf[c] = getchar();
c++;
buf[c] = 0;
if( buf[c-1] == '\n' )
break;
}
// swallow, just in case.
if( c == sizeof(buf) )
while( getchar() != '\n' );
(*mem_poses[stomach]) = atoi( buf );
break;
}
// MMm
case 12:
{
stomach--; if(stomach < 0) stomach += num_stomachs;
break;
}
// MmM
case 13:
{
stomach++; if(stomach >= num_stomachs) stomach -= num_stomachs;
break;
}
// Oom
case 14:
{
for( int i=0; i<num_stomachs; ++i )
{
if( mem_poses[i] == memory[i].begin() )
quit( true );
else
mem_poses[i]--;
}
break;
}
// oOm
case 15:
{
for( int i=0; i<num_stomachs; ++i )
{
mem_poses[i]++;
if( mem_poses[i] == memory[i].end() )
{
memory[i].push_back(0);
mem_poses[i] = memory[i].end();
mem_poses[i]--;
}
}
break;
}
// OoM
case 16:
{
for( int i=0; i<num_stomachs; ++i )
{
(*mem_poses[i])--;
}
break;
}
// oOM
case 17:
{
for( int i=0; i<num_stomachs; ++i)
{
(*mem_poses[i])++;
}
break;
}
// ooo
case 18:
{
for( int i=0; i<num_stomachs; ++i)
{
(*mem_poses[i]) = 0;
}
break;
}
// mmm
case 19:
{
if( has_register_val )
{
int weight = 0;
for( int i=0; i<num_stomachs; ++i)
{
if( *mem_poses[i] > 0 )
weight += *mem_poses[i];
}
if( weight == 0 )
{
for( int i=0; i<num_stomachs; ++i )
{
*mem_poses[i] = register_val;
}
} else {
int div = register_val / weight;
int mod = register_val - div * weight;
for( int i=0; i<num_stomachs; ++i )
{
if( *mem_poses[i] > 0 )
*mem_poses[i] = *mem_poses[i] * div;
else if( *mem_poses[i] < 0 )
*mem_poses[i] = mod * -*mem_poses[i];
else
*mem_poses[i] = 0;
}
}
} else {
int sum = 0;
for( int i=0; i<num_stomachs; ++i)
{
sum += *mem_poses[i];
}
register_val = sum;
}
has_register_val = !has_register_val;
break;
}
// bad stuff
default:
quit( false );
};
prog_pos++;
return true;
}
int main( int argc, char** argv )
{
if( argc < 2 )
{
printf( "Usage: %s program.cow\n\n", argv[0] );
exit( 1 );
}
FILE* f = fopen( argv[1], "rb" );
if( f == NULL )
{
printf( "Cannot open source file [%s].\n", argv[1] );
exit( 1 );
}
char buf[3];
memset( buf, 0, 3 );
//int pos = 0; // tola: unused?
while( !feof(f) )
{
int found = 0;
buf[2] = fgetc( f );
if(( found = !strncmp( "moo", buf, 3 ) ))
program.push_back( 0 );
else if(( found = !strncmp( "mOo", buf, 3 ) ))
program.push_back( 1 );
else if(( found = !strncmp( "moO", buf, 3 ) ))
program.push_back( 2 );
else if(( found = !strncmp( "mOO", buf, 3 ) ))
program.push_back( 3 );
else if(( found = !strncmp( "Moo", buf, 3 ) ))
program.push_back( 4 );
else if(( found = !strncmp( "MOo", buf, 3 ) ))
program.push_back( 5 );
else if(( found = !strncmp( "MoO", buf, 3 ) ))
program.push_back( 6 );
else if(( found = !strncmp( "MOO", buf, 3 ) ))
program.push_back( 7 );
else if(( found = !strncmp( "OOO", buf, 3 ) ))
program.push_back( 8 );
else if(( found = !strncmp( "MMM", buf, 3 ) ))
program.push_back( 9 );
else if(( found = !strncmp( "OOM", buf, 3 ) ))
program.push_back( 10 );
else if(( found = !strncmp( "oom", buf, 3 ) ))
program.push_back( 11 );
// from here down, distributed digestion specific calls
else if(( found = !strncmp( "MMm", buf, 3 ) ))
program.push_back( 12 );
else if(( found = !strncmp( "MmM", buf, 3 ) ))
program.push_back( 13 );
else if(( found = !strncmp( "Oom", buf, 3 ) ))
program.push_back( 14 );
else if(( found = !strncmp( "oOm", buf, 3 ) ))
program.push_back( 15 );
else if(( found = !strncmp( "OoM", buf, 3 ) ))
program.push_back( 16 );
else if(( found = !strncmp( "oOM", buf, 3 ) ))
program.push_back( 17 );
else if(( found = !strncmp( "ooo", buf, 3 ) ))
program.push_back( 18 );
else if(( found = !strncmp( "mmm", buf, 3 ) ))
program.push_back( 19 );
if( found )
{
memset( buf, 0, 3 );
}
else
{
buf[0] = buf[1];
buf[1] = buf[2];
buf[2] = 0;
}
}
fclose( f );
printf( "Welcome to COW!\n\nExecuting [%s]...\n\n", argv[1] );
// init main memory.
for(int i=0; i<num_stomachs; ++i)
{
memory[i].push_back(0);
mem_poses[i] = memory[i].begin();
}
stomach = 0;
prog_pos = program.begin();
while( prog_pos != program.end() )
if( !exec( *prog_pos ) )
break;
quit( false );
return 0;
}
bool compile( int instruction, bool advance )
{
switch( instruction )
{
// moo
case 0:
{
int level = 1;
int num = MOOcount + 1;
mem_t::iterator t = prog_pos;
t--; // skip past previous command when searching for MOO.
if( t != program.begin() )
{
if((*t) == 7 )
num--;
while( level > 0 )
{
if( t == program.begin() )
break;
t--;
if( (*t) == 0 )
level++;
else
if( (*t) == 7 ) // look for MOO
{
level--;
num--;
}
}
}
if( level != 0 && advance )
quit();
else if( level != 0 )
{
fprintf( output, "rterr();" );
break;
}
moocount++;
fprintf( output, "goto M%d;", num );
fprintf( output, "m%d:", moocount );
PRETTY( "moo" );
}
break;
// mOo
case 1:
fprintf( output, "if(p==m.begin()){rterr();}else{p--;}" );
PRETTY( "mOo" );
break;
// moO
case 2:
fprintf( output, "p++; if(p==m.end()){m.push_back(0);p=m.end();p--;}" );
PRETTY( "moO" );
break;
// mOO
case 3:
// I think it should be possible to build a switch statement here and then
// use the compile function itself to fill in the possibilities.
// printf( "NOT IMPLEMENTED: mOO\n\n" );
// quit();
fprintf( output, "switch(*p){" );
fprintf( output, "case 0:{" ); compile( 0, false ); fprintf( output, "}break;" );
fprintf( output, "case 1:{" ); compile( 1, false ); fprintf( output, "}break;" );
fprintf( output, "case 2:{" ); compile( 2, false ); fprintf( output, "}break;" );
fprintf( output, "case 4:{" ); compile( 4, false ); fprintf( output, "}break;" );
fprintf( output, "case 5:{" ); compile( 5, false ); fprintf( output, "}break;" );
fprintf( output, "case 6:{" ); compile( 6, false ); fprintf( output, "}break;" );
fprintf( output, "case 7:{" ); compile( 7, false ); fprintf( output, "}break;" );
fprintf( output, "case 8:{" ); compile( 8, false ); fprintf( output, "}break;" );
fprintf( output, "case 9:{" ); compile( 9, false ); fprintf( output, "}break;" );
fprintf( output, "case 10:{" ); compile( 10, false ); fprintf( output, "}break;" );
fprintf( output, "case 11:{" ); compile( 11, false ); fprintf( output, "}break;" );
fprintf( output, "default:{goto x;}};" );
PRETTY( "mOO" );
break;
// Moo
case 4:
fprintf( output, "if((*p)!=0){putchar(*p);}else{(*p)=getchar();while(getchar()!='\\n');}" );
PRETTY( "Moo" );
break;
// MOo
case 5:
fprintf( output, "(*p)--;" );
PRETTY( "MOo" );
break;
// MoO
case 6:
fprintf( output, "(*p)++;" );
PRETTY( "MoO" );
break;
// MOO
case 7:
{
int level = 1;
int num = moocount;
int prev = 0;
mem_t::iterator t = prog_pos;
t++; // have to skip past next command when looking for next moo.
if( t != program.end() )
{
if( (*t) == 0 )
num++;
while( level > 0 )
{
prev = *t;
t++;
if( t == program.end() )
break;
if( (*t) == 7 ) // look for MOO command.
level++;
else
if( (*t) == 0 ) // look for moo command.
{
if( prev == 7 )
level--;
level--;
num++;
}
if( level == 0 )
break;
}
}
if( advance && level != 0 )
quit();
else if( level != 0 )
{
fprintf( output, "rterr();" );
break;
}
MOOcount++;
fprintf( output, "M%d:", MOOcount );
fprintf( output, "if(!(*p))goto m%d;", num );
PRETTY( "MOO" );
}
break;
// OOO
case 8:
fprintf( output, "(*p)=0;" );
PRETTY( "OOO" );
break;
// MMM
case 9:
fprintf( output, "if(h){(*p)=r;}else{r=(*p);}h=!h;" );
PRETTY( "MMM" );
break;
// OOM
case 10:
fprintf( output, "printf(\"%%d\\n\",*p);" );
PRETTY( "OOM" );
break;
// oom
case 11:
fprintf( output, "char b[100];int c=0;" );
fprintf( output, "while(c<sizeof(b)-1){b[c]=getchar();c++;b[c]=0;if(b[c-1]=='\\n')break;}" );
fprintf( output, "if(c==sizeof(b))while(getchar()!='\\n');(*p)=atoi(b);" );
PRETTY( "oom" );
break;
// bad stuff
default:
return false;
};
if( advance )
prog_pos++;
return true;
}
int main( int argc, char** argv )
{
if( argc < 2 )
{
printf( "Usage: %s program.cow\n\n", argv[0] );
exit( 1 );
}
FILE* f = fopen( argv[1], "rb" );
if( f == NULL )
{
printf( "Cannot open source file [%s].\n", argv[1] );
exit( 1 );
}
char buf[3];
memset( buf, 0, 3 );
int pos = 0;
while( !feof(f) )
{
int found = 0;
buf[2] = fgetc( f );
if( found = !strncmp( "moo", buf, 3 ) )
program.push_back( 0 );
else if( found = !strncmp( "mOo", buf, 3 ) )
program.push_back( 1 );
else if( found = !strncmp( "moO", buf, 3 ) )
program.push_back( 2 );
else if( found = !strncmp( "mOO", buf, 3 ) )
program.push_back( 3 );
else if( found = !strncmp( "Moo", buf, 3 ) )
program.push_back( 4 );
else if( found = !strncmp( "MOo", buf, 3 ) )
program.push_back( 5 );
else if( found = !strncmp( "MoO", buf, 3 ) )
program.push_back( 6 );
else if( found = !strncmp( "MOO", buf, 3 ) )
program.push_back( 7 );
else if( found = !strncmp( "OOO", buf, 3 ) )
program.push_back( 8 );
else if( found = !strncmp( "MMM", buf, 3 ) )
program.push_back( 9 );
else if( found = !strncmp( "OOM", buf, 3 ) )
program.push_back( 10 );
else if( found = !strncmp( "oom", buf, 3 ) )
program.push_back( 11 );
if( found )
{
memset( buf, 0, 3 );
}
else
{
buf[0] = buf[1];
buf[1] = buf[2];
buf[2] = 0;
}
}
fclose( f );
printf( "Compiling [%s]...\n", argv[1] );
// init main memory.
/*
memory.push_back( 0 );
mem_pos = memory.begin();
*/
output = fopen( "cow.out.cpp", "wb" );
fprintf( output, "#include <stdio.h>\n" );
fprintf( output, "#include <vector>\n" );
fprintf( output, "typedef std::vector<int> t_;t_ m;t_::iterator p;\n" );
fprintf( output, "bool h;int r;\n" );
fprintf( output, "void rterr(){puts(\"Runtime error.\\n\");}\n" );
fprintf( output, "int main(int a,char** v){\n" );
fprintf( output, "m.push_back(0);p=m.begin();h=false;\n" );
prog_pos = program.begin();
while( prog_pos != program.end() )
if( !compile( *prog_pos, true ) )
{
printf( "ERROR!\n" );
break;
}
fprintf( output, "x:return(0);}\n" );
fclose( output );
printf( "C++ source code: cow.out.cpp\n" );
#ifdef COMPILER
std::string path( (const char*)COMPILER );
path.append( " " );
path.append( (const char*)NAME_FLAG );
path.append( " " );
path.append( (const char*)OUTPUT_EXEC );
path.append( " " );
path.append( (const char*)FLAGS );
path.append( " " );
path.append( (const char*)OUTPUT_CPP );
if( system( path.c_str() ) )
printf( "\n\nCould not compile. Possible causes: C++ compiler is not installed, not in path, or not named '%s' or there is a bug in this compiler.\n\n", COMPILER );
else
printf( "Executable created: cow.out\n" );
#endif
return 0;
}
bool op_ring::execute()
{
if( pos != 0 ) switch( pos )
{
// One
case 2:
value=1;
break;
// Zero
case 3:
value=0;
break;
// Load
case 4:
value=*mem_pos;
break;
// Store
case 5:
*mem_pos = value;
break;
// DAdd
case 7:
{
int t = mem_pos - memory.begin();
t += value;
if( t < 0 )
return false; // out of bounds!
if( t < memory.size() )
mem_pos = memory.begin() + t;
else
{
while( memory.size() <= t )
memory.push_back(0);
mem_pos = memory.begin() + t;
}
}
break;
// Logic
case 8:
if( *mem_pos != 0 )
value = value && 1;
else
value = 0;
break;
// If
case 9:
if( *mem_pos == 0 )
break;
// else fall through!
// PAdd
case 6:
{
int t = prog_pos - program.begin();
t += value;
if( t < 0 || t >= program.size() )
return false; // out of bounds!
prog_pos = program.begin() + t;
next_instruction = false;
}
break;
// IntIO
case 10:
if( value == 0 )
{
// read integer
char buf[100];
int c = 0;
while( c < sizeof(buf)-1 )
{
buf[c] = getchar();
c++;
buf[c] = 0;
if( buf[c-1] == '\n' )
break;
}
// swallow, just in case.
if( c == sizeof(buf) )
while( getchar() != '\n' );
(*mem_pos) = atoi( buf );
}
else
printf( "%d", *mem_pos );
break;
// AscIO
case 11:
if( value == 0 )
{
// read character
(*mem_pos) = getchar();
while( getchar() != '\n' );
}
else
printf( "%c", *mem_pos );
break;
// Exit
case 1:
default:
return false;
}
return true;
}
int main( int argc, char** argv )
{
if( argc < 2 )
{
printf( "Usage: %s program.wrl\n\n", argv[0] );
exit( 1 );
}
FILE* f = fopen( argv[1], "rb" );
if( f == NULL )
{
printf( "Cannot open source file [%s].\n", argv[1] );
exit( 1 );
}
while( !feof(f) )
{
char buf = fgetc( f );
if( buf == '1' )
program.push_back( true );
else if( buf == '0' )
program.push_back( false );
}
fclose( f );
printf( "Welcome to Whirl!\n\nExecuting [%s]...\n\n", argv[1] );
// init main memory.
memory.push_back( 0 );
mem_pos = memory.begin();
prog_pos = program.begin();
bool execute = false;
while( prog_pos != program.end() )
{
next_instruction = true;
if( *prog_pos )
{
cur->rotate();
execute = false;
}
else
{
cur->switch_dir();
if( execute )
{
if( !cur->execute() )
return 0;
(cur == &ops)? cur = &math: cur = &ops;
execute = false;
}
else
execute = true;
}
if( next_instruction )
prog_pos++;
}
printf( "\n" );
return 0;
}