int main(int argc, char **argv)
{
FILE *out; /* Output data file */
char s[255],s2[255],delim[255],*pstr; /* Generic strings */
int *memcache; /* used to flush cache */
int len_buf_align, /* meaningful when args.cache is 0. buflen */
/* rounded up to be divisible by 8 */
num_buf_align; /* meaningful when args.cache is 0. number */
/* of aligned buffers in memtmp */
int c, /* option index */
i, j, n, nq, /* Loop indices */
asyncReceive=0, /* Pre-post a receive buffer? */
bufalign=16*1024,/* Boundary to align buffer to */
errFlag, /* Error occurred in inner testing loop */
nrepeat, /* Number of time to do the transmission */
nrepeat_const=0,/* Set if we are using a constant nrepeat */
len, /* Number of bytes to be transmitted */
inc=0, /* Increment value */
perturbation=DEFPERT, /* Perturbation value */
pert,
start= 1, /* Starting value for signature curve */
end=MAXINT, /* Ending value for signature curve */
streamopt=0, /* Streaming mode flag */
reset_connection;/* Reset the connection between trials */
ArgStruct args; /* Arguments for all the calls */
double t, t0, t1, t2, /* Time variables */
tlast, /* Time for the last transmission */
latency; /* Network message latency */
Data bwdata[NSAMP]; /* Bandwidth curve data */
int integCheck=0; /* Integrity check */
/* Initialize vars that may change from default due to arguments */
strcpy(s, "np.out"); /* Default output file */
/* Let modules initialize related vars, and possibly call a library init
function that requires argc and argv */
Init(&args, &argc, &argv); /* This will set args.tr and args.rcv */
args.preburst = 0; /* Default to not bursting preposted receives */
args.bidir = 0; /* Turn bi-directional mode off initially */
args.cache = 1; /* Default to use cache */
args.upper = end;
args.host = NULL;
args.soffset=0; /* default to no offsets */
args.roffset=0;
args.syncflag=0; /* use normal mpi_send */
/* TCGMSG launches NPtcgmsg with a -master master_hostname
* argument, so ignore all arguments and set them manually
* in netpipe.c instead.
*/
#if ! defined(TCGMSG)
/* Parse the arguments. See Usage for description */
while ((c = getopt(argc, argv, "SO:rIiPszgfaB2h:p:o:l:u:b:m:n:t:c:d:D:")) != -1)
{
switch(c)
{
case 'O':
strcpy(s2,optarg);
strcpy(delim,",");
if((pstr=strtok(s2,delim))!=NULL) {
args.soffset=atoi(pstr);
if((pstr=strtok((char *)NULL,delim))!=NULL)
args.roffset=atoi(pstr);
else /* only got one token */
args.roffset=args.soffset;
} else {
args.soffset=0; args.roffset=0;
}
printf("Transmit buffer offset: %d\nReceive buffer offset: %d\n",args.soffset,args.roffset);
break;
case 'p': perturbation = atoi(optarg);
if( perturbation > 0 ) {
printf("Using a perturbation value of %d\n\n", perturbation);
} else {
perturbation = 0;
printf("Using no perturbations\n\n");
}
break;
case 'B': if(integCheck == 1) {
fprintf(stderr, "Integrity check not supported with prepost burst\n");
exit(-1);
}
args.preburst = 1;
asyncReceive = 1;
printf("Preposting all receives before a timed run.\n");
printf("Some would consider this cheating,\n");
printf("but it is needed to match some vendor tests.\n"); fflush(stdout);
break;
case 'I': args.cache = 0;
printf("Performance measured without cache effects\n\n"); fflush(stdout);
break;
case 'o': memset(s,0,sizeof(s));strncpy(s,optarg,sizeof(s)-1);
printf("Sending output to %s\n", s); fflush(stdout);
break;
case 's': streamopt = 1;
printf("Streaming in one direction only.\n\n");
#if defined(TCP) && ! defined(INFINIBAND)
printf("Sockets are reset between trials to avoid\n");
printf("degradation from a collapsing window size.\n\n");
#endif
args.reset_conn = 1;
printf("Streaming does not provide an accurate\n");
printf("measurement of the latency since small\n");
printf("messages may get bundled together.\n\n");
if( args.bidir == 1 ) {
printf("You can't use -s and -2 together\n");
exit(0);
}
fflush(stdout);
break;
case 'l': start = atoi(optarg);
if (start < 1)
{
fprintf(stderr,"Need a starting value >= 1\n");
exit(0);
}
break;
case 'u': end = atoi(optarg);
break;
#if defined(TCP) && ! defined(INFINIBAND)
case 'b': /* -b # resets the buffer size, -b 0 keeps system defs */
args.prot.sndbufsz = args.prot.rcvbufsz = atoi(optarg);
break;
#endif
case '2': args.bidir = 1; /* Both procs are transmitters */
/* end will be maxed at sndbufsz+rcvbufsz */
printf("Passing data in both directions simultaneously.\n");
printf("Output is for the combined bandwidth.\n");
#if defined(TCP) && ! defined(INFINIBAND)
printf("The socket buffer size limits the maximum test size.\n\n");
#endif
if( streamopt ) {
printf("You can't use -s and -2 together\n");
exit(0);
}
break;
case 'h': args.tr = 1; /* -h implies transmit node */
args.rcv = 0;
args.host = (char *)malloc(strlen(optarg)+1);
strcpy(args.host, optarg);
break;
#ifdef DISK
case 'd': args.tr = 1; /* -d to specify input/output file */
args.rcv = 0;
args.prot.read = 0;
args.prot.read_type = 'c';
args.prot.dfile_name = (char *)malloc(strlen(optarg)+1);
strcpy(args.prot.dfile_name, optarg);
break;
case 'D': if( optarg[0] == 'r' )
args.prot.read = 1;
else
args.prot.read = 0;
args.prot.read_type = optarg[1];
break;
#endif
case 'i': if(args.preburst == 1) {
fprintf(stderr, "Integrity check not supported with prepost burst\n");
exit(-1);
}
integCheck = 1;
perturbation = 0;
start = sizeof(int)+1; /* Start with integer size */
printf("Doing an integrity check instead of measuring performance\n"); fflush(stdout);
break;
#if defined(MPI)
case 'z': args.source_node = -1;
printf("Receive using the ANY_SOURCE flag\n"); fflush(stdout);
break;
case 'a': asyncReceive = 1;
printf("Preposting asynchronous receives\n"); fflush(stdout);
break;
case 'S': args.syncflag=1;
fprintf(stderr,"Using synchronous sends\n");
break;
#endif
#if defined(MPI2)
case 'g': if(args.prot.no_fence == 1) {
fprintf(stderr, "-f cannot be used with -g\n");
exit(-1);
}
args.prot.use_get = 1;
printf("Using MPI-2 Get instead of Put\n");
break;
case 'f': if(args.prot.use_get == 1) {
fprintf(stderr, "-f cannot be used with -g\n");
exit(-1);
}
args.prot.no_fence = 1;
bufalign = 0;
printf("Buffer alignment off (Required for no fence)\n");
break;
#endif /* MPI2 */
#if defined(INFINIBAND)
case 'm': switch(atoi(optarg)) {
case 256: args.prot.ib_mtu = MTU256;
break;
case 512: args.prot.ib_mtu = MTU512;
break;
case 1024: args.prot.ib_mtu = MTU1024;
break;
case 2048: args.prot.ib_mtu = MTU2048;
break;
case 4096: args.prot.ib_mtu = MTU4096;
break;
default:
fprintf(stderr, "Invalid MTU size, must be one of "
"256, 512, 1024, 2048, 4096\n");
exit(-1);
}
break;
case 't': if( !strcmp(optarg, "send_recv") ) {
printf("Using Send/Receive communications\n");
args.prot.commtype = NP_COMM_SENDRECV;
} else if( !strcmp(optarg, "send_recv_with_imm") ) {
printf("Using Send/Receive communications with immediate data\n");
args.prot.commtype = NP_COMM_SENDRECV_WITH_IMM;
} else if( !strcmp(optarg, "rdma_write") ) {
printf("Using RDMA Write communications\n");
args.prot.commtype = NP_COMM_RDMAWRITE;
} else if( !strcmp(optarg, "rdma_write_with_imm") ) {
printf("Using RDMA Write communications with immediate data\n");
args.prot.commtype = NP_COMM_RDMAWRITE_WITH_IMM;
} else {
fprintf(stderr, "Invalid transfer type "
"specified, please choose one of:\n\n"
"\tsend_recv\t\tUse Send/Receive communications\t(default)\n"
"\tsend_recv_with_imm\tSame as above with immediate data\n"
"\trdma_write\t\tUse RDMA Write communications\n"
"\trdma_write_with_imm\tSame as above with immediate data\n\n");
exit(-1);
}
break;
case 'c': if( !strcmp(optarg, "local_poll") ) {
printf("Using local polling completion\n");
args.prot.comptype = NP_COMP_LOCALPOLL;
} else if( !strcmp(optarg, "vapi_poll") ) {
printf("Using VAPI polling completion\n");
args.prot.comptype = NP_COMP_VAPIPOLL;
} else if( !strcmp(optarg, "event") ) {
printf("Using VAPI event completion\n");
args.prot.comptype = NP_COMP_EVENT;
} else {
fprintf(stderr, "Invalid completion type specified, "
"please choose one of:\n\n"
"\tlocal_poll\tWait for last byte of data\t(default)\n"
"\tvapi_poll\tUse VAPI polling function\n"
"\tevent\t\tUse VAPI event handling function\n\n");
exit(-1);
}
break;
#endif
case 'n': nrepeat_const = atoi(optarg);
break;
#if defined(TCP) && ! defined(INFINIBAND)
case 'r': args.reset_conn = 1;
printf("Resetting connection after every trial\n");
break;
#endif
default:
PrintUsage();
exit(-12);
}
}
#endif /* ! defined TCGMSG */
#if defined(INFINIBAND)
asyncReceive = 1;
fprintf(stderr, "Preposting asynchronous receives (required for Infiniband)\n");
if(args.bidir && (
(args.cache && args.prot.commtype == NP_COMM_RDMAWRITE) || /* rdma_write only works with no-cache mode */
(!args.preburst && args.prot.commtype != NP_COMM_RDMAWRITE) || /* anything besides rdma_write requires prepost burst */
(args.preburst && args.prot.comptype == NP_COMP_LOCALPOLL && args.cache) || /* preburst with local polling in cache mode doesn't work */
0)) {
fprintf(stderr,
"\n"
"Bi-directional mode currently only works with a subset of the\n"
"Infiniband options. Restrictions are:\n"
"\n"
" RDMA write (-t rdma_write) requires no-cache mode (-I).\n"
"\n"
" Local polling (-c local_poll, default if no -c given) requires\n"
" no-cache mode (-I), and if not using RDMA write communication,\n"
" burst mode (-B).\n"
"\n"
" Any other communication type and any other completion type\n"
" require burst mode (-B). No-cache mode (-I) may be used\n"
" optionally.\n"
"\n"
" All other option combinations will fail.\n"
"\n");
exit(-1);
}
#endif
if (start > end)
{
fprintf(stderr, "Start MUST be LESS than end\n");
exit(420132);
}
args.nbuff = TRIALS;
args.port = DEFPORT;
Setup(&args);
if( args.bidir && end > args.upper ) {
end = args.upper;
if( args.tr ) {
printf("The upper limit is being set to %d Bytes\n", end);
#if defined(TCP) && ! defined(INFINIBAND)
printf("due to socket buffer size limitations\n\n");
#endif
} }
#if defined(GM)
if(streamopt && (!nrepeat_const || nrepeat_const > args.prot.num_stokens)) {
printf("\nGM is currently limited by the driver software to %d\n",
args.prot.num_stokens);
printf("outstanding sends. The number of repeats will be set\n");
printf("to this limit for every trial in streaming mode. You\n");
printf("may use the -n switch to set a smaller number of repeats\n\n");
nrepeat_const = args.prot.num_stokens;
}
#endif
if( args.tr ) /* Primary transmitter */
{
if ((out = fopen(s, "w")) == NULL)
{
fprintf(stderr,"Can't open %s for output\n", s);
exit(1);
}
}
else out = stdout;
/* Set a starting value for the message size increment. */
inc = (start > 1) ? start / 2 : 1;
nq = (start > 1) ? 1 : 0;
/* Test the timing to set tlast for the first test */
args.bufflen = start;
MyMalloc(&args, args.bufflen, 0, 0);
InitBufferData(&args, args.bufflen, 0, 0);
if(args.cache) args.s_buff = args.r_buff;
args.r_ptr = args.r_buff_orig = args.r_buff;
args.s_ptr = args.s_buff_orig = args.s_buff;
AfterAlignmentInit(&args); /* MPI-2 needs this to create a window */
/* Infiniband requires use of asynchronous communications, so we need
* the PrepareToReceive calls below
*/
if( asyncReceive )
PrepareToReceive(&args);
Sync(&args); /* Sync to prevent race condition in armci module */
/* For simplicity's sake, even if the real test below will be done in
* bi-directional mode, we still do the ping-pong one-way-at-a-time test
* here to estimate the one-way latency. Unless it takes significantly
* longer to send data in both directions at once than it does to send data
* one way at a time, this shouldn't be too far off anyway.
*/
t0 = When();
for( n=0; n<100; n++) {
if( args.tr) {
SendData(&args);
RecvData(&args);
if( asyncReceive && n<99 )
PrepareToReceive(&args);
} else if( args.rcv) {
RecvData(&args);
if( asyncReceive && n<99 )
PrepareToReceive(&args);
SendData(&args);
}
}
tlast = (When() - t0)/200;
/* Sync up and Reset before freeing the buffers */
Sync(&args);
Reset(&args);
/* Free the buffers and any other module-specific resources. */
if(args.cache)
FreeBuff(args.r_buff_orig, NULL);
else
FreeBuff(args.r_buff_orig, args.s_buff_orig);
/* Do setup for no-cache mode, using two distinct buffers. */
if (!args.cache)
{
/* Allocate dummy pool of memory to flush cache with */
if ( (memcache = (int *)malloc(MEMSIZE)) == NULL)
{
perror("malloc");
exit(1);
}
mymemset(memcache, 0, MEMSIZE/sizeof(int));
/* Allocate large memory pools */
MyMalloc(&args, MEMSIZE+bufalign, args.soffset, args.roffset);
/* Save buffer addresses */
args.s_buff_orig = args.s_buff;
args.r_buff_orig = args.r_buff;
/* Align buffers */
args.s_buff = AlignBuffer(args.s_buff, bufalign);
args.r_buff = AlignBuffer(args.r_buff, bufalign);
/* Post alignment initialization */
AfterAlignmentInit(&args);
/* Initialize send buffer pointer */
/* both soffset and roffset should be zero if we don't have any offset stuff, so this should be fine */
args.s_ptr = args.s_buff+args.soffset;
args.r_ptr = args.r_buff+args.roffset;
}
/**************************
* Main loop of benchmark *
**************************/
if( args.tr ) fprintf(stderr,"Now starting the main loop\n");
for ( n = 0, len = start, errFlag = 0;
n < NSAMP - 3 && tlast < STOPTM && len <= end && !errFlag;
len = len + inc, nq++ )
{
/* Exponentially increase the block size. */
if (nq > 2) inc = ((nq % 2))? inc + inc: inc;
/* This is a perturbation loop to test nearby values */
for (pert = ((perturbation > 0) && (inc > perturbation+1)) ? -perturbation : 0;
pert <= perturbation;
n++, pert += ((perturbation > 0) && (inc > perturbation+1)) ? perturbation : perturbation+1)
{
Sync(&args); /* Sync to prevent race condition in armci module */
/* Calculate how many times to repeat the experiment. */
if( args.tr )
{
if (nrepeat_const) {
nrepeat = nrepeat_const;
/* } else if (len == start) {*/
/* nrepeat = MAX( RUNTM/( 0.000020 + start/(8*1000) ), TRIALS);*/
} else {
nrepeat = MAX((RUNTM / ((double)args.bufflen /
(args.bufflen - inc + 1.0) * tlast)),TRIALS);
}
SendRepeat(&args, nrepeat);
}
else if( args.rcv )
{
RecvRepeat(&args, &nrepeat);
}
args.bufflen = len + pert;
if( args.tr )
fprintf(stderr,"%3d: %7d bytes %6d times --> ",
n,args.bufflen,nrepeat);
if (args.cache) /* Allow cache effects. We use only one buffer */
{
/* Allocate the buffer with room for alignment*/
MyMalloc(&args, args.bufflen+bufalign, args.soffset, args.roffset);
/* Save buffer address */
args.r_buff_orig = args.r_buff;
args.s_buff_orig = args.r_buff;
/* Align buffer */
args.r_buff = AlignBuffer(args.r_buff, bufalign);
args.s_buff = args.r_buff;
/* Initialize buffer with data
*
* NOTE: The buffers should be initialized with some sort of
* valid data, whether it is actually used for anything else,
* to get accurate results. Performance increases noticeably
* if the buffers are left uninitialized, but this does not
* give very useful results as realworld apps tend to actually
* have data stored in memory. We are not sure what causes
* the difference in performance at this time.
*/
InitBufferData(&args, args.bufflen, args.soffset, args.roffset);
/* Post-alignment initialization */
AfterAlignmentInit(&args);
/* Initialize buffer pointers (We use r_ptr and s_ptr for
* compatibility with no-cache mode, as this makes the code
* simpler)
*/
/* offsets are zero by default so this saves an #ifdef */
args.r_ptr = args.r_buff+args.roffset;
args.s_ptr = args.r_buff+args.soffset;
}
else /* Eliminate cache effects. We use two distinct buffers */
{
/* this isn't truly set up for offsets yet */
/* Size of an aligned memory block including trailing padding */
len_buf_align = args.bufflen;
if(bufalign != 0)
len_buf_align += bufalign - args.bufflen % bufalign;
/* Initialize the buffers with data
*
* See NOTE above.
*/
InitBufferData(&args, MEMSIZE, args.soffset, args.roffset);
/* Reset buffer pointers to beginning of pools */
args.r_ptr = args.r_buff+args.roffset;
args.s_ptr = args.s_buff+args.soffset;
}
bwdata[n].t = LONGTIME;
/* t2 = t1 = 0;*/
/* Finally, we get to transmit or receive and time */
/* NOTE: If a module is running that uses only one process (e.g.
* memcpy), we assume that it will always have the args.tr flag
* set. Thus we make some special allowances in the transmit
* section that are not in the receive section.
*/
if( args.tr || args.bidir )
{
/*
This is the transmitter: send the block TRIALS times, and
if we are not streaming, expect the receiver to return each
block.
*/
for (i = 0; i < (integCheck ? 1 : TRIALS); i++)
{
if(args.preburst && asyncReceive && !streamopt)
{
/* We need to save the value of the recv ptr so
* we can reset it after we do the preposts, in case
* the module needs to use the same ptr values again
* so it can wait on the last byte to change to indicate
* the recv is finished.
*/
SaveRecvPtr(&args);
for(j=0; j<nrepeat; j++)
{
PrepareToReceive(&args);
if(!args.cache)
AdvanceRecvPtr(&args, len_buf_align);
}
ResetRecvPtr(&args);
}
/* Flush the cache using the dummy buffer */
if (!args.cache)
flushcache(memcache, MEMSIZE/sizeof(int));
Sync(&args);
t0 = When();
for (j = 0; j < nrepeat; j++)
{
if (!args.preburst && asyncReceive && !streamopt)
{
PrepareToReceive(&args);
}
if (integCheck) SetIntegrityData(&args);
SendData(&args);
if (!streamopt)
{
RecvData(&args);
if (integCheck) VerifyIntegrity(&args);
if(!args.cache)
AdvanceRecvPtr(&args, len_buf_align);
}
/* Wait to advance send pointer in case RecvData uses
* it (e.g. memcpy module).
*/
if (!args.cache)
AdvanceSendPtr(&args, len_buf_align);
}
/* t is the 1-directional trasmission time */
t = (When() - t0)/ nrepeat;
if( !streamopt && !args.bidir) t /= 2; /* Normal ping-pong */
Reset(&args);
/* NOTE: NetPIPE does each data point TRIALS times, bouncing the message
* nrepeats times for each trial, then reports the lowest of the TRIALS
* times. -Dave Turner
*/
bwdata[n].t = MIN(bwdata[n].t, t);
/* t1 += t;*/
/* t2 += t*t;*/
}
if (streamopt){ /* Get time info from Recv node */
RecvTime(&args, &bwdata[n].t);
/* RecvTime(&args, &t1);*/
/* RecvTime(&args, &t2);*/
}
/* Calculate variance after completing this set of trials */
/* bwdata[n].variance = t2/TRIALS - t1/TRIALS * t1/TRIALS;*/
}
else if( args.rcv )
{
/*
This is the receiver: receive the block TRIALS times, and
if we are not streaming, send the block back to the
sender.
*/
for (i = 0; i < (integCheck ? 1 : TRIALS); i++)
{
if (asyncReceive)
{
if (args.preburst)
{
/* We need to save the value of the recv ptr so
* we can reset it after we do the preposts, in case
* the module needs to use the same ptr values again
* so it can wait on the last byte to change to
* indicate the recv is finished.
*/
SaveRecvPtr(&args);
for (j=0; j < nrepeat; j++)
{
PrepareToReceive(&args);
if (!args.cache)
AdvanceRecvPtr(&args, len_buf_align);
}
ResetRecvPtr(&args);
}
else
{
PrepareToReceive(&args);
}
}
/* Flush the cache using the dummy buffer */
if (!args.cache)
flushcache(memcache, MEMSIZE/sizeof(int));
Sync(&args);
t0 = When();
for (j = 0; j < nrepeat; j++)
{
RecvData(&args);
if (integCheck) VerifyIntegrity(&args);
if (!args.cache)
{
AdvanceRecvPtr(&args, len_buf_align);
}
if (!args.preburst && asyncReceive && (j < nrepeat-1))
{
PrepareToReceive(&args);
}
if (!streamopt)
{
if (integCheck) SetIntegrityData(&args);
SendData(&args);
if(!args.cache)
AdvanceSendPtr(&args, len_buf_align);
}
}
t = (When() - t0)/ nrepeat;
if( !streamopt && !args.bidir) t /= 2; /* Normal ping-pong */
Reset(&args);
bwdata[n].t = MIN(bwdata[n].t, t);
/* t1 += t;*/
/* t2 += t*t;*/
}
if (streamopt){ /* Recv proc calcs time and sends to Trans */
SendTime(&args, &bwdata[n].t);
/* SendTime(&args, &t1);*/
/* SendTime(&args, &t2);*/
}
}
else /* Just going along for the ride */
{
for (i = 0; i < (integCheck ? 1 : TRIALS); i++)
{
Sync(&args);
}
}
/* Streaming mode doesn't really calculate correct latencies
* for small message sizes, and on some nics we can get
* zero second latency after doing the math. Protect against
* this.
*/
if(bwdata[n].t == 0.0) {
bwdata[n].t = 0.000001;
}
tlast = bwdata[n].t;
bwdata[n].bits = args.bufflen * CHARSIZE * (1+args.bidir);
bwdata[n].bps = bwdata[n].bits / (bwdata[n].t * 1024 * 1024);
bwdata[n].repeat = nrepeat;
if (args.tr)
{
if(integCheck) {
fprintf(out,"%8d %d", bwdata[n].bits / 8, nrepeat);
} else {
fprintf(out,"%8d %lf %12.8lf",
bwdata[n].bits / 8, bwdata[n].bps, bwdata[n].t);
}
fprintf(out, "\n");
fflush(out);
}
/* Free using original buffer addresses since we may have aligned
r_buff and s_buff */
if (args.cache)
FreeBuff(args.r_buff_orig, NULL);
if ( args.tr ) {
if(integCheck) {
fprintf(stderr, " Integrity check passed\n");
} else {
fprintf(stderr," %8.2lf Mbps in %10.2lf usec\n",
bwdata[n].bps, tlast*1.0e6);
}
}
} /* End of perturbation loop */
} /* End of main loop */
/* Free using original buffer addresses since we may have aligned
r_buff and s_buff */
if (!args.cache) {
FreeBuff(args.s_buff_orig, args.r_buff_orig);
}
if (args.tr) fclose(out);
CleanUp(&args);
return 0;
}
void CPlainGraphicalTextEditor::DeleteLine(void)
{
mcEditor.DeleteLine();
Sync();
}
void CPlainGraphicalTextEditor::HomeEnter(void)
{
mcEditor.HomeEnter();
Sync();
}
void CPlainGraphicalTextEditor::Paste(char* szSource)
{
mcEditor.Paste(szSource);
Sync();
}
void CPlainGraphicalTextEditor::Duplicate(void)
{
mcEditor.Duplicate();
Sync();
}
void CPlainGraphicalTextEditor::SelectWholeWord(void)
{
mcEditor.SelectWholeWord();
Sync();
}
void CPlainGraphicalTextEditor::ClearSelection(void)
{
mcEditor.ClearSelection();
Sync();
}
void RecvRepeat(ArgStruct *p, int *rpt)
{
/* Only child calls RecvRepeat */
Sync(p);
*rpt = my_shared->repeat;
}
void Setup(ArgStruct *p)
{
pid_t pid;
struct knem_cmd_info info;
char *filename = (char*) "NPknem.mmap";
size_t pagesize = sysconf(_SC_PAGESIZE);
int need_fork = 0, am_parent = 0, flags = O_RDWR;
/* If no filename was specified, then we're the parent until we
fork */
if (NULL == p->host) {
am_parent = 1;
need_fork = 1;
unlink(filename);
}
/* If the file was specified, see if the file exists. If it does,
then we're the parent. Otherwise, we're the "child". */
else {
int ret;
struct stat sbuf;
filename = p->host;
need_fork = 0;
ret = stat(filename, &sbuf);
if (ret < 0 && errno == ENOENT) {
am_parent = 1;
}
}
/* Open a file so that we can mmap it */
if (am_parent) {
flags |= O_CREAT;
printf("NPknem: creating shared file: %s\n", filename);
} else {
printf("NPknem: attaching to shared file: %s\n", filename);
}
shared_fd = open(filename, flags, 0600);
if (shared_fd < 0) {
fprintf(stderr, "Netpipe failed to create a file to mmap. Aborting in despair.\n");
exit(1);
}
shared_len = pagesize * 2;
if (0 != ftruncate(shared_fd, shared_len)) {
fprintf(stderr, "Netpipe failed to set the size of a shared file. Aborting in despair.\n");
exit(1);
}
/* Now fork the child and reset transmitter / receiver flags as
appropriate */
if (need_fork) {
pid = fork();
if (0 == pid) {
am_parent = 0;
} else {
am_parent = 1;
}
}
if (am_parent) {
p->rcv = 0;
p->tr = 1;
} else {
p->rcv = 1;
p->tr = 0;
}
/* Map an anonymous shared memory segment between the two
processes for passing cookies, etc. */
shared_map = mmap(NULL, shared_len, (PROT_READ | PROT_WRITE), MAP_SHARED,
shared_fd, 0);
if (MAP_FAILED == shared_map) {
if (am_parent || !need_fork) {
perror("mmap");
fprintf(stderr, "Netpipe failed to mmap a shared file. Aborting in despair.\n");
}
exit(1);
}
/* Once we have both mmaped the file, parent unlinks the file so
that we don't deal with filesystem synchronization. Parent's
data is the first page; child's data is the second. This
way we can pin them down to our local memory via "first touch"
method. */
if (am_parent) {
my_shared = (shared_data_t*) shared_map;
peer_shared = (shared_data_t*) (shared_map + pagesize);
} else {
peer_shared = (shared_data_t*) shared_map;
my_shared = (shared_data_t*) (shared_map + pagesize);
}
memset((void*) my_shared, 0, sizeof(*my_shared));
Sync(p);
if (am_parent){
unlink(filename);
}
/* See if we can find /dev/knem */
knem_fd = open(KNEM_DEVICE_FILENAME, O_RDWR);
if (knem_fd < 0) {
if (am_parent || !need_fork) {
fprintf(stderr, "Netpipe cannot open %s on this host. Aborting in despair.\n", KNEM_DEVICE_FILENAME);
}
exit(1);
}
/* Check that we're compiled against the same ABI version that
we're running with */
if (ioctl(knem_fd, KNEM_CMD_GET_INFO, &info) < 0) {
if (am_parent || !need_fork) {
fprintf(stderr, "Netpipe failed to get ABI version from %s. Aborting in despair.\n", KNEM_DEVICE_FILENAME);
}
exit(1);
}
#if KNEM_ABI_VERSION < 0x0000000c
#error KNEM is too old, pleasse upgrade to 0.7 or later
#endif
if (KNEM_ABI_VERSION != info.abi) {
if (am_parent || !need_fork) {
fprintf(stderr, "The knem ABI that Netpipe was compiled against is different than that of %s. Aborting in despair.\n", KNEM_DEVICE_FILENAME);
}
exit(1);
}
/* If DMA was asked for, ensure that it's available */
if (0 != p->prot.flags && 0 == (info.features & KNEM_FEATURE_DMA)) {
if (am_parent || !need_fork) {
fprintf(stderr, "DMA copy mode was requested but is not supported in the running kernel. Aborting in despair.\n");
}
exit(1);
}
}
static DECLFW(M179WriteLo)
{
if(A==0x5ff1) reg[1]=V;
Sync();
}
void SendData(ArgStruct *p)
{
struct knem_cmd_param_iovec knem_iov = {
.base = (uintptr_t) p->s_ptr,
.len = p->bufflen
};
struct knem_cmd_create_region createregioncmd = {
.iovec_array = (uintptr_t) &knem_iov,
.iovec_nr = 1,
.flags = KNEM_FLAG_SINGLEUSE,
.protection = PROT_READ,
};
/* Create the region */
if (0 != ioctl(knem_fd, KNEM_CMD_CREATE_REGION, &createregioncmd)) {
fprintf(stderr, "Netpipe failed to KNEM_CMD_CREATE_REGION. Aborting in despair.\n");
exit(1);
}
/* Send the cookie to the peer */
peer_shared->cookies[peer_cookie_index] = createregioncmd.cookie;
peer_cookie_index = (0 == peer_cookie_index) ? 1 : 0;
}
void RecvData(ArgStruct *p)
{
uint64_t cookie;
struct knem_cmd_param_iovec knem_iov = {
.base = (uintptr_t) p->r_ptr,
.len = p->bufflen
};
struct knem_cmd_inline_copy icopycmd = {
.local_iovec_array = (uintptr_t) &knem_iov,
.local_iovec_nr = 1,
.write = 0,
.async_status_index = 0,
.flags = p->prot.flags
};
/* Wait for the sender to set my cookie */
while (0 == (cookie = my_shared->cookies[my_cookie_index])) {
continue;
}
icopycmd.remote_cookie = cookie;
icopycmd.remote_offset = 0;
my_shared->cookies[my_cookie_index] = 0;
my_cookie_index = (0 == my_cookie_index) ? 1 : 0;
/* Initiate the receive (synchronous mode) */
if (0 != ioctl(knem_fd, KNEM_CMD_INLINE_COPY, &icopycmd)) {
fprintf(stderr, "Netpipe failed to KNEM_CMD_INLINE_COPY. Aborting in despair.\n");
exit(1);
}
if (icopycmd.current_status != KNEM_STATUS_SUCCESS) {
fprintf(stderr, "Netpipe failed to complete KNEM_CMD_INLINE_COPY. Aborting in despair.\n");
exit(1);
}
}
void SendTime(ArgStruct *p, double *t)
{
/* Only child calls SendTime */
peer_shared->time = *t;
Sync(p);
}
void RecvTime(ArgStruct *p, double *t)
{
/* Only parent calls RecvTime */
Sync(p);
*t = my_shared->time;
}
void SendRepeat(ArgStruct *p, int rpt)
{
/* Only parent calls SendRepeat */
peer_shared->repeat = rpt;
Sync(p);
}
static DECLFW(M179Write)
{
if(A==0xa000) reg[0]=V;
Sync();
}
static DECLFW(UNLDSOUNDV1Write) {
reg[A & 3] = V;
// FCEU_printf("cmd %04x:%02x\n", A, V);
Sync();
}
static DECLFW(WriteHi)
{
PRGBank16 = V;
Sync();
}
void CPlainGraphicalTextEditor::SelectDocumentEnd(void)
{
mcEditor.SelectDocumentEnd();
Sync();
}
static DECLFW(M178Write) {
reg[A & 3] = V;
Sync();
}
void CPlainGraphicalTextEditor::SelectPageDown(void)
{
mcEditor.SelectPageDown();
Sync();
}
static void StateRestore(int version) {
Sync();
}
void CPlainGraphicalTextEditor::SelectAll(void)
{
mcEditor.SelectAll();
Sync();
}
void CPlainGraphicalTextEditor::Backspace(void)
{
mcEditor.Backspace();
Sync();
}
void CPlainGraphicalTextEditor::Copy(CChars* pszDest)
{
mcEditor.Copy(pszDest);
Sync();
}
void CPlainGraphicalTextEditor::Printable(char c, BOOL bInsert)
{
mcEditor.Printable(c, bInsert);
Sync();
}
void CPlainGraphicalTextEditor::CutLine(CChars* pszDest)
{
mcEditor.CutLine(pszDest);
Sync();
}
void CPlainGraphicalTextEditor::DocumentHome(void)
{
mcEditor.DocumentHome();
Sync();
}
void CPlainGraphicalTextEditor::DeleteWordRemainingLeft(void)
{
mcEditor.DeleteWordRemainingLeft();
Sync();
}
void CPlainGraphicalTextEditor::PageDown(void)
{
mcEditor.PageDown();
Sync();
}
void CPlainGraphicalTextEditor::EndEnter(void)
{
mcEditor.EndEnter();
Sync();
}
void CPlainGraphicalTextEditor::SelectWordRight(void)
{
mcEditor.SelectWordRight();
Sync();
}
void CPlainGraphicalTextEditor::Space(BOOL bInsert)
{
mcEditor.Space(bInsert);
Sync();
}
void AsyncRenderer::SetGameMap(GameMap *gm) {
SPADES_MARK_FUNCTION();
rcmds::SetGameMap *cmd = generator->AllocCommand<rcmds::SetGameMap>();
cmd->map = gm;
Sync();
}