int main(int argc, char *argv[])
{
unsigned row, col, coreid, i, j, m, n, k;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
char emsg[_BufSize];
unsigned num;
unsigned counter = 0;
srand(1);
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
//e_set_host_verbosity(H_D2);
//e_set_loader_verbosity(L_D1);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
// Open a workgroup
e_open(&dev, 0, 0, platform.rows, platform.cols);
// Load the device program onto core (0,0)
e_load("e_mutex_test0.srec", &dev, 0, 0, E_TRUE);
usleep(10000);
// Load the device program onto all the other eCores
e_load_group("e_mutex_test.srec", &dev, 0, 1, 1, 3, E_TRUE);
e_load_group("e_mutex_test.srec", &dev, 1, 0, 3, 4, E_TRUE);
usleep(100000);
// Wait for core program execution to finish
// Read message from shared buffer
e_read(&dev, 0, 0, 0x6200, &num, sizeof(num));
e_read(&dev, 0, 0, 0x6300, &counter, sizeof(counter));
// Print the message and close the workgroup.
fprintf(stderr, "The counter now is %d!\n", counter);
fprintf(stderr, "The clock cycle is %d!\n", num);
// Close the workgroup
e_close(&dev);
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
const uint32_t zero = 0;
uint32_t result = 0;
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
if (e_alloc(&emem, 0, 128) != E_OK)
exit(EXIT_FAILURE);
e_write(&emem, 0, 0, 0, &zero, sizeof(zero));
e_open(&dev, 0, 0, 1, 1);
if (e_load("emain.elf", &dev, 0, 0, E_TRUE) != E_OK)
exit(EXIT_FAILURE);
do {
e_read(&emem, 0, 0, 0, &result, sizeof(result));
} while (!result);
if (result != 1) {
fprintf(stderr,
"Test failed. elink RX remapping configured incorrectly. Faulty programs can\n"
"likely access the system's first 1MB memory region. Go patch your kernel!\n");
exit(EXIT_FAILURE);
}
fprintf(stderr, "Test passed\n");
return 0;
}
int main(int argc, char *argv[])
{
int result, fail;
fd = stderr;
pEpiphany = &Epiphany;
pERAM = &ERAM;
e_set_host_verbosity(H_D0);
if ( E_OK != e_init(NULL) ) {
fprintf(stderr, "\nERROR: epiphinay initialization failed!\n\n");
exit(1);
}
if (E_OK != e_reset_system() ) {
fprintf(stderr, "\nWARNING: epiphinay system rest failed!\n\n");
}
// prepare ERAM
if (E_OK != e_alloc(pERAM, 0x00000000, e_platform.emem[0].size))
{
fprintf(stderr, "\nERROR: Can't allocate Epiphany DRAM!\n\n");
exit(1);
}
e_set_host_verbosity(H_D0);
if (E_OK != e_open(pEpiphany, 0, 0, e_platform.rows, e_platform.cols))
{
fprintf(stderr, "\nERROR: Can't establish connection to Epiphany device!\n\n");
exit(1);
}
fail = 0;
//////////////////////////////
// Test Host-Device throughput
SRAM_speed();
ERAM_speed();
DRAM_speed();
/////////////////////////////
// Test eCore-ERAM throughput
result = EPI_speed();
//Finalize
e_close(pEpiphany);
e_free(pERAM);
e_finalize();
/////////////////////////////
//For now, always pass
return EXIT_SUCCESS;
}
int main(int argc, char *argv[])
{
unsigned row, col, coreid, i;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
char emsg[_BufSize];
srand(1);
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
for (i=0; i<_SeqLen; i++)
{
// Draw a random core
row = rand() % platform.rows;
col = rand() % platform.cols;
coreid = (row + platform.row) * 64 + col + platform.col;
fprintf(stderr, "%3d: Message from eCore 0x%03x (%2d,%2d): ", i, coreid, row, col);
// Open the single-core workgroup and reset the core, in
// case a previous process is running. Note that we used
// core coordinates relative to the workgroup.
e_open(&dev, row, col, 1, 1);
e_reset_group(&dev);
// Load the device program onto the selected eCore
// and launch after loading.
e_load("e_hello_world.srec", &dev, 0, 0, E_TRUE);
// Wait for core program execution to finish, then
// read message from shared buffer.
usleep(100000000);
e_read(&emem, 0, 0, 0x0, emsg, _BufSize);
// Print the message and close the workgroup.
fprintf(stderr, "\"%s\"\n", emsg);
e_close(&dev);
}
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return 0;
}
int main()
{
//counters for row and colum, cored id and loop counter
unsigned row, col, id, row_loop,col_loop;
// this will contain the epiphany platform configuration
e_platform_t epiphany;
e_epiphany_t dev;
e_mem_t memory;
char message[_BufSize];
e_return_stat_t result;
e_init(NULL); // initialise the system establish connection to the Device
e_reset_system(); // reset the epiphnay chip
e_get_platform_info(&epiphany);//gets the configuration info for the parallella platofrm
// allocatethe shared memory for recieivng the message from the core
e_alloc(&memory, _BufOffset, _BufSize);
row = 0;
col = 0;
for(row_loop=0; row_loop <4; row_loop ++)
{
for(col_loop=0; col_loop <4; col_loop ++)
{
//one core within the parallella work group is 1 x 1 i.e single core
e_open(&dev,row_loop,col_loop,1,1);
//reset the group
e_reset_group(&dev);
//load the group
result = e_load("hello_world.srec",&dev,0,0,E_TRUE);
if (result != E_OK) {
fprintf(stderr,"Error Loading the Epiphany Application %i\n", result);
}
usleep(10000);
e_read(&memory,0,0,0x0, message, _BufSize);
fprintf(stderr,"message from core = %s\n", message);
e_close(&dev);
}
}
e_free(&memory);
e_finalize();
return 0;
}
/**
* Loads up the code onto the appropriate Epiphany cores, sets up the state (Python bytecode, symbol table, data area etc)
* and then starts the cores running
*/
struct shared_basic * loadCodeOntoEpiphany(struct interpreterconfiguration* configuration) {
struct shared_basic * basicCode;
int i, result, codeOnCore=0;
e_set_host_verbosity(H_D0);
result = e_init(NULL);
if (result == E_ERR) fprintf(stderr, "Error on initialisation\n");
result = e_reset_system();
if (result == E_ERR) fprintf(stderr, "Error on system reset\n");
result = e_open(&epiphany, 0, 0, e_platform.chip[0].rows, e_platform.chip[0].cols);
if (result != E_OK) fprintf(stderr, "Error opening Epiphany\n");
result = e_alloc(&management_DRAM, EXTERNAL_MEM_ABSOLUTE_START, SHARED_DATA_SIZE);
if (result == E_ERR) fprintf(stderr, "Error allocating memory\n");
basicCode=(void*) management_DRAM.base;
basicCode->length=getMemoryFilledSize();
if (configuration->forceCodeOnCore) {
codeOnCore=1;
} else if (configuration->forceCodeOnShared) {
codeOnCore=0;
} else {
codeOnCore=basicCode->length <= CORE_CODE_MAX_SIZE;
if (!codeOnCore) {
printf("Warning: Your code size of %d bytes exceeds the %d byte limit for placement on cores so storing in shared memory\n", basicCode->length, CORE_CODE_MAX_SIZE);
}
}
basicCode->symbol_size=getNumberEntriesInSymbolTable();
basicCode->allInSharedMemory=configuration->forceDataOnShared;
basicCode->codeOnCores=codeOnCore==1;
basicCode->num_procs=configuration->coreProcs+configuration->hostProcs;
basicCode->baseHostPid=configuration->coreProcs;
initialiseCores(basicCode, codeOnCore, configuration);
placeByteCode(basicCode, codeOnCore, configuration->intentActive);
startApplicableCores(basicCode, configuration);
pb=(unsigned int*) malloc(sizeof(unsigned int) * TOTAL_CORES);
for (i=0;i<TOTAL_CORES;i++) {
pb[i]=1;
}
return basicCode;
}
static bool epiphany_thread_prepare(struct thr_info *thr)
{
e_epiphany_t *dev = &thr->cgpu->epiphany_dev;
e_mem_t *emem = &thr->cgpu->epiphany_emem;
unsigned rows = thr->cgpu->epiphany_rows;
unsigned cols = thr->cgpu->epiphany_cols;
char *fullpath = alloca(PATH_MAX);
if (e_alloc(emem, _BufOffset, rows * cols * sizeof(shared_buf_t)) == E_ERR) {
applog(LOG_ERR, "Error: Could not alloc shared Epiphany memory.");
return false;
}
if (e_open(dev, 0, 0, rows, cols) == E_ERR) {
applog(LOG_ERR, "Error: Could not start Epiphany cores.");
return false;
}
strcpy(fullpath, cgminer_path);
strcat(fullpath, "epiphany-scrypt.srec");
FILE* checkf = fopen(fullpath, "r");
if (!checkf) {
thr->cgpu->status = LIFE_SICK;
applog(LOG_ERR, "Error: Could not find epiphany-scrypt.srec.");
applog(LOG_ERR, " Is epiphany-scrypt.srec in cgminer directory?.");
return false;
}
fclose(checkf);
if (e_load_group(fullpath, dev, 0, 0, rows, cols, E_FALSE) == E_ERR) {
applog(LOG_ERR, "Error: Could not load epiphany-scrypt.srec on Epiphany.");
return false;
}
thread_reportin(thr);
return true;
}
int main(int argc, char *argv[])
{
unsigned int row, col;
unsigned int data;
int i,j;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
char emsg[_BufSize];
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_set_host_verbosity(H_D0);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
// Open a workgroup
e_open(&dev, 0, 0, platform.rows, platform.cols);
//Turn off the LVDS Links from the a core program
e_load("e_link_lowpower_mode.srec", &dev, 0, 0, E_TRUE);
// Close the workgroup
e_close(&dev);
// Release the allocated buffer and finalize the
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
e_epiphany_t Epiphany, *pEpiphany;
e_mem_t DRAM, *pDRAM;
unsigned int msize;
int row, col, cnum;
ILuint ImgId;
// ILenum Error;
ILubyte *imdata;
ILuint imsize, imBpp;
unsigned int addr;
size_t sz;
struct timespec timer[4];
uint32_t time_p[TIMERS];
uint32_t time_d[TIMERS];
FILE *fo;
// FILE *fi;
int result;
pEpiphany = &Epiphany;
pDRAM = &DRAM;
msize = 0x00400000;
//get_args(argc, argv);
strcpy(ar.ifname,argv[1]);
strcpy(ar.elfFile,argv[2]);
strcpy(ar.ofname, ar.ifname);
printf("------------------------------------------------------------\n");
fo = fopen("matprt.m", "w");
if ((fo == NULL)) // || (fi == NULL))
{
fprintf(stderr, "Could not open Octave file \"%s\" ...exiting.\n", "matprt.m");
exit(4);
}
// fo = stderr;
// Connect to device for communicating with the Epiphany system
// Prepare device
e_set_host_verbosity(ar.verbose);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
if (e_open(pEpiphany, 0, 0, platform.rows, platform.cols))
{
fprintf(fo, "\nERROR: Can't establish connection to Epiphany device!\n\n");
exit(1);
}
if (e_alloc(pDRAM, 0x00000000, msize))
{
fprintf(fo, "\nERROR: Can't allocate Epiphany DRAM!\n\n");
exit(1);
}
// Initialize Epiphany "Ready" state
addr = offsetof(shared_buf_t, core.ready);
Mailbox.core.ready = 0;
e_write(pDRAM, 0, 0, addr, (void *) &(Mailbox.core.ready), sizeof(Mailbox.core.ready));
result = e_load_group(ar.elfFile, pEpiphany, 0, 0, platform.rows, platform.cols, (e_bool_t) (ar.run_target));
if (result == E_ERR) {
printf("Error loading Epiphany program.\n");
exit(1);
}
// Check if the DevIL shared lib's version matches the executable's version.
if (ilGetInteger(IL_VERSION_NUM) < IL_VERSION)
{
fprintf(stderr, "DevIL version is different ...exiting!\n");
exit(2);
}
// Initialize DevIL.
ilInit();
#ifdef ILU_ENABLED
iluInit();
#endif
// create the coreID list
init_coreID(pEpiphany, coreID, _Nside, _Nside, 0x808);
// Generate the main image name to use, bind it and load the image file.
ilGenImages(1, &ImgId);
ilBindImage(ImgId);
if (!ilLoadImage(ar.ifname))//ar.ifname
{
fprintf(stderr, "Could not open input image file \"%s\" ...exiting.\n", ar.ifname);
exit(3);
}
// Display the image's dimensions to the end user.
/*
printf("Width: %d Height: %d Depth: %d Bpp: %d\n\n",
ilGetInteger(IL_IMAGE_WIDTH),
ilGetInteger(IL_IMAGE_HEIGHT),
ilGetInteger(IL_IMAGE_DEPTH),
ilGetInteger(IL_IMAGE_BITS_PER_PIXEL));
*/
imdata = ilGetData();
imsize = ilGetInteger(IL_IMAGE_WIDTH) * ilGetInteger(IL_IMAGE_HEIGHT);
imBpp = ilGetInteger(IL_IMAGE_BYTES_PER_PIXEL);
if (imsize != (_Sfft * _Sfft))
{
printf("Image file size is different from %dx%d ...exiting.\n", _Sfft, _Sfft);
exit(5);
}
// Extract image data into the A matrix.
for (unsigned int i=0; i<imsize; i++)
{
Mailbox.A[i] = (float) imdata[i*imBpp] + 0.0 * I;
}
fprintf(fo, "\n");
// Generate operand matrices based on a provided seed
matrix_init(0);
#ifdef _USE_DRAM_
// Copy operand matrices to Epiphany system
addr = DRAM_BASE + offsetof(shared_buf_t, A[0]);
sz = sizeof(Mailbox.A);
fprintf(fo, "%% Writing A[%ldB] to address %08x...\n", sz, addr);
e_write(addr, (void *) Mailbox.A, sz);
addr = DRAM_BASE + offsetof(shared_buf_t, B[0]);
sz = sizeof(Mailbox.B);
fprintf(fo, "%% Writing B[%ldB] to address %08x...\n", sz, addr);
e_write(addr, (void *) Mailbox.B, sz);
#else
// Copy operand matrices to Epiphany cores' memory
fprintf(fo, "%% Writing image to Epiphany\n");
sz = sizeof(Mailbox.A) / _Ncores;
for (row=0; row<(int) platform.rows; row++)
for (col=0; col<(int) platform.cols; col++)
{
addr = BankA_addr;
fflush(stdout);
cnum = e_get_num_from_coords(pEpiphany, row, col);
// printf( "Writing A[%uB] to address %08x...\n", sz, addr);
fprintf(fo, "%% Writing A[%uB] to address %08x...\n", sz, (coreID[cnum] << 20) | addr);
fflush(fo);
e_write(pEpiphany, row, col, addr, (void *) &Mailbox.A[cnum * _Score * _Sfft], sz);
}
#endif
// Call the Epiphany fft2d() function
fprintf(fo, "%% GO!\n");
fflush(stdout);
fflush(fo);
clock_gettime(CLOCK_MONOTONIC, &timer[0]);
fft2d_go(pDRAM);
clock_gettime(CLOCK_MONOTONIC, &timer[1]);
fprintf(fo, "%% Done!\n\n");
fflush(stdout);
fflush(fo);
// Read time counters
// printf( "Reading time count...\n");
fprintf(fo, "%% Reading time count...\n");
addr = 0x7128+0x4*2 + offsetof(core_t, time_p[0]);
sz = TIMERS * sizeof(uint32_t);
e_read(pEpiphany, 0, 0, addr, (void *) (&time_p[0]), sz);
// for (int i=0; i<TIMERS; i++)
// printf("time_p[%d] = %u\n", i, time_p[i]);
time_d[2] = time_p[7] - time_p[2]; // FFT setup
time_d[3] = time_p[2] - time_p[3]; // bitrev (x8)
time_d[4] = time_p[3] - time_p[4]; // FFT-1D (x8)
time_d[5] = time_p[4] - time_p[5]; // corner-turn
time_d[6] = time_p[7] - time_p[8]; // FFT-2D
time_d[7] = time_p[6] - time_p[7]; // LPF
time_d[9] = time_p[0] - time_p[9]; // Total cycles
fprintf(fo, "%% Finished calculation in %u cycles (%5.3f msec @ %3.0f MHz)\n\n", time_d[9], (time_d[9] * 1000.0 / eMHz), (eMHz / 1e6));
printf( "FFT2D - %7u cycles (%5.3f msec)\n", time_d[6], (time_d[6] * 1000.0 / eMHz));
printf( " FFT Setup - %7u cycles (%5.3f msec)\n", time_d[2], (time_d[2] * 1000.0 / eMHz));
printf( " BITREV - %7u cycles (%5.3f msec)\n", time_d[3], (time_d[3] * 1000.0 / eMHz));
printf( " FFT1D - %7u cycles (%5.3f msec x2)\n", time_d[4], (time_d[4] * 1000.0 / eMHz));
printf( " Corner Turn - %7u cycles (%5.3f msec)\n", time_d[5], (time_d[5] * 1000.0 / eMHz));
printf( "LPF - %7u cycles (%5.3f msec)\n", time_d[7], (time_d[7] * 1000.0 / eMHz));
fprintf(fo, "%% Reading processed image back to host\n");
// Read result matrix
#ifdef _USE_DRAM_
addr = DRAM_BASE + offsetof(shared_buf_t, B[0]);
sz = sizeof(Mailbox.B);
printf( "Reading B[%ldB] from address %08x...\n", sz, addr);
fprintf(fo, "%% Reading B[%ldB] from address %08x...\n", sz, addr);
blknum = sz / RdBlkSz;
remndr = sz % RdBlkSz;
for (i=0; i<blknum; i++)
{
fflush(stdout);
e_read(addr+i*RdBlkSz, (void *) ((long unsigned)(Mailbox.B)+i*RdBlkSz), RdBlkSz);
}
fflush(stdout);
e_read(addr+i*RdBlkSz, (void *) ((long unsigned)(Mailbox.B)+i*RdBlkSz), remndr);
#else
// Read result matrix from Epiphany cores' memory
sz = sizeof(Mailbox.A) / _Ncores;
for (row=0; row<(int) platform.rows; row++)
for (col=0; col<(int) platform.cols; col++)
{
addr = BankA_addr;
fflush(stdout);
cnum = e_get_num_from_coords(pEpiphany, row, col);
// printf( "Reading A[%uB] from address %08x...\n", sz, addr);
fprintf(fo, "%% Reading A[%uB] from address %08x...\n", sz, (coreID[cnum] << 20) | addr);
fflush(fo);
e_read(pEpiphany, row, col, addr, (void *) &Mailbox.B[cnum * _Score * _Sfft], sz);
}
#endif
// Convert processed image matrix B into the image file date.
for (unsigned int i=0; i<imsize; i++)
{
for (unsigned int j=0; j<imBpp; j++)
imdata[i*imBpp+j] = cabs(Mailbox.B[i]);
}
// Save processed image to the output file.
ilEnable(IL_FILE_OVERWRITE);
if (!ilSaveImage(ar.ofname))
{
fprintf(stderr, "Could not open output image file \"%s\" ...exiting.\n", ar.ofname);
exit(7);
}
// We're done with the image, so let's delete it.
ilDeleteImages(1, &ImgId);
// Simple Error detection loop that displays the Error to the user in a human-readable form.
// while ((Error = ilGetError()))
// PRINT_ERROR_MACRO;
// Close connection to device
if (e_close(pEpiphany))
{
fprintf(fo, "\nERROR: Can't close connection to Epiphany device!\n\n");
exit(1);
}
if (e_free(pDRAM))
{
fprintf(fo, "\nERROR: Can't release Epiphany DRAM!\n\n");
exit(1);
}
fflush(fo);
fclose(fo);
//Returnin success if test runs expected number of clock cycles
//Need to add comparison with golden reference image!
printf("------------------------------------------------------------\n");
if(time_d[9]>50000) {
printf( "TEST \"fft2d\" PASSED\n");
return EXIT_SUCCESS;
}
else {
printf( "TEST \"fft2d\" FAILED\n");
return EXIT_FAILURE;
}
}
int main(int argc, char *argv[])
{
unsigned rows, cols, ncores, coreid, i, j;
const uint32_t one = 1, zero = 0;
int result[_MAX_CORES];
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
int fault, highest;
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
//open the workgroup
rows = platform.rows;
cols = platform.cols;
e_open(&dev, 0, 0, rows, cols);
//load the device program on the board
e_load_group("emain.srec", &dev, 0, 0, rows, cols, E_FALSE);
e_start_group(&dev);
usleep(100000);
ncores = rows * cols;
printf("num-cores = %4d\n", ncores);
fault = 0x0;
for (i=0; i < 0x10000; i++) {
/* Pause leader core so we can read without races */
e_write(&dev, 0, 0, 0x7000, &one, sizeof(one));
/* Lazily assume cores will be paused and writes from all cores
* to ERAM have propagated after below sleep. This must be
* calibrated w.r.t Epiphany chip clock frequency and delay
* cycles in the device code */
usleep(1000);
/* read the results */
e_read(&emem, 0, 0, 0x0, &result, ncores*sizeof(int));
/* Resume */
e_write(&dev, 0, 0, 0x7000, &zero, sizeof(zero));
highest = result[0];
for (j=0; j<ncores; j++) {
if (result[j] != result[0]) {
fault++;
if (highest < result[j])
highest = result[j];
}
}
/* Don't print every iteration */
if (i % 0x10)
continue;
printf("[%03x] ", i);
for (j=0; j<ncores; j++)
printf("%04x ", result[j]);
printf("\n");
if (highest >= NBARRIERS)
break;
/* Do a small wait so it is easy to see that the E cores are
* running independently. */
usleep(10000);
}
//print the success/error message duel to the number of fault
if (fault == 0)
printf("\ntest #20: Hardware Barrier Passed!\n");
else
printf("\ntest #20: Hardware Barrier Failed! Fault is 0x%08x!\n", fault);
e_close(&dev);
e_free(&emem);
e_finalize();
return fault != 0;
}
int main(int argc, char *argv[])
{
unsigned row, col, coreid, i;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
Mailbox mailbox;
mailbox.flag = 0;
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, 1024);
row = 0;
col = 0;
coreid = (row + platform.row) * 64 + col + platform.col;
fprintf(stderr,"\n\nMultiplying A[%d][%d] x B[%d][%d] = C[%d][%d]\n",N,ROWS,N,ROWS,N,ROWS);
// Open the single-core workgroup Note that we used
// core coordinates relative to the workgroup.
e_open(&dev, row, col, 1, 1);
// Load the device program onto the selected eCore
// and launch after loading.
e_load("e_dev_main.elf", &dev, 0, 0, E_FALSE);
e_start_group(&dev);
// Wait for core program execution to finish, then
// read mailbox from shared buffer.
unsigned int addr = offsetof(Mailbox, flag);
while (mailbox.flag != 1) {
e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
}
// Wait for core program execution to finish, then
// read message from shared buffer.
//usleep(100000);
e_read(&emem, 0, 0, 0x0, &mailbox, sizeof(mailbox));
float c_o[N][N];
addr = (unsigned int)mailbox.c_o;
e_read(&dev, 0, 0, addr, &c_o[0][0], N * N * sizeof(float));
print_to_file("../output/optresult",c_o);
// Print the message and close the workgroup.
e_close(&dev);
printf("\nOptimized MATMUL time = %d cycles\tTime: %9.3f msec\n", mailbox.clocks1,clock_to_time(mailbox.clocks1));
float gflops = ((2 * N * ROWS * N)/(clock_to_time(mailbox.clocks1)/1000))/1000/1000/1000;
printf("\nGFlops: %9.6f\tPerformance = %9.4f %% of peak\n",gflops,gflops/1.2*100);
fprintf(stderr,"\n");
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
unsigned rows, cols, coreid, i, j;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
//open the workgroup
rows = platform.rows;
cols = platform.cols;
e_open(&dev, 0, 0, rows, cols);
//load the device program on the board
e_load_group("emain.srec", &dev, 0, 0, rows, cols, E_FALSE);
//set up the event list table
strcpy(event[0], "CLK");
strcpy(event[1], "IDLE");
strcpy(event[2], "IALU_INST");
strcpy(event[3], "FPU_INST");
strcpy(event[4], "DUAL_INST");
strcpy(event[5], "E1_STALLS");
strcpy(event[6], "RA_STALLS");
strcpy(event[7], "EXT_FETCH_STALLS");
strcpy(event[8], "EXT_LOAD_STALLS");
strcpy(event[9], "IALU_INST");
for (i=0; i<rows; i++)
{
for (j=0; j<cols; j++)
{
coreid = (i + platform.row) * 64 + j + platform.col;
fprintf(stderr, "Message from eCore 0x%03x (%2d,%2d): \n", coreid, i, j);
e_start(&dev, i, j);
//wait for core to execute the program
usleep(100000);
e_read(&emem, 0, 0, 0x0, &result, sizeof(unsigned)*10);
check();
}
}
e_close(&dev);
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
unsigned row, col, coreid, i, j, m, n, k;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
char emsg[_BufSize];
srand(1);
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
//e_set_host_verbosity(H_D2);
//e_set_host_verbosity(H_D1);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
// Open a workgroup
e_open(&dev, 0, 0, platform.rows, platform.cols);
// Reset the workgroup
for (m=0; m<platform.rows; m++)
{ for(n=0; n<platform.cols;n++)
{
ee_reset_core(&dev, m, n);
}
}
// Load the device program onto all the eCores
e_load_group("e_nested_test.srec", &dev, 0, 0, platform.rows, platform.cols, E_FALSE);
// Select one core to work
for (i=0; i<platform.rows; i++)
{
for (j=0; j<platform.cols; j++)
{
// Draw to a certain core
row=i;
col=j;
coreid = (row + platform.row) * 64 + col + platform.col;
fprintf(stderr,"%d: Message from eCore 0x%03x (%2d,%2d): \n",(i*platform.cols+j),coreid,row,col);
e_start(&dev, i, j);
usleep(1000000);
// Wait for core program execution to finish
// Read message from shared buffer
e_read(&emem, 0, 0, 0x0, &emsg, _BufSize);
// Print the message and close the workgroup.
fprintf(stderr, "%s\n", emsg);
}
}
// Close the workgroup
e_close(&dev);
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
unsigned rows, cols, coreid, i, j;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
int result;
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
//open the workgroup
rows = platform.rows;
cols = platform.cols;
e_open(&dev, 0, 0, rows, cols);
//load the device program on the board
e_load_group("emainorigin.srec", &dev, 0, 0, rows, cols, E_FALSE);
// for (i=0; i<rows; i++)
// {
// for (j=0; j<cols; j++)
// {
// i = 3; j = 3;
for (i=0; i<1; i++)
{
for (j=0; j<2; j++)
{
coreid = (i + platform.row) * 64 + j + platform.col;
fprintf(stderr, "Message from eCore 0x%03x (%2d,%2d): \n", coreid, i, j);
e_start(&dev, i, j);
usleep(100000);
e_read(&dev, i, j, 0x5200, &result, sizeof(int));
if(result == 0)
fprintf(stderr, "\"test MULTICAST passed!\"\n");
else
fprintf(stderr, "\"test MULTICAST failed!\t\t\tWarnning, test failed! Num of fualt is %d!\"\n", result);
usleep(100000);
}
}
e_close(&dev);
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
e_epiphany_t Epiphany, *pEpiphany;
e_mem_t DRAM, *pDRAM;
unsigned int msize;
float seed;
unsigned int addr; //, clocks;
size_t sz;
double tdiff[4];
int result, rerval;
pEpiphany = &Epiphany;
pDRAM = &DRAM;
msize = 0x00400000;
get_args(argc, argv);
fo = stderr;
fi = stdin;
printf("\nMatrix: C[%d][%d] = A[%d][%d] * B[%d][%d]\n\n", _Smtx, _Smtx, _Smtx, _Smtx, _Smtx, _Smtx);
printf("Using %d x %d cores\n\n", _Nside, _Nside);
seed = 0.0;
printf("Seed = %f\n", seed);
// Connect to device for communicating with the Epiphany system
// Prepare device
e_set_host_verbosity(H_D0);
e_init(NULL);
e_reset_system();
if (e_alloc(pDRAM, 0x00000000, msize))
{
printf("\nERROR: Can't allocate Epiphany DRAM!\n\n");
exit(1);
}
if (e_open(pEpiphany, 0, 0, e_platform.chip[0].rows, e_platform.chip[0].cols))
{
printf("\nERROR: Can't establish connection to Epiphany device!\n\n");
exit(1);
}
// Initialize Epiphany "Ready" state
addr = offsetof(shared_buf_t, core.ready);
Mailbox.core.ready = 0;
e_write(pDRAM, 0, 0, addr, &Mailbox.core.ready, sizeof(Mailbox.core.ready));
printf("Loading program on Epiphany chip...\n");
e_set_loader_verbosity(ar.verbose);
result = e_load_group(ar.srecFile, pEpiphany, 0, 0, pEpiphany->rows, pEpiphany->cols, ar.run_target);
if (result == E_ERR) {
printf("Error loading Epiphany program.\n");
exit(1);
}
// Generate operand matrices based on a provided seed
matrix_init(seed);
#ifdef __WIPE_OUT_RESULT_MATRIX__
// Wipe-out any previous remains in result matrix (for verification)
addr = offsetof(shared_buf_t, C[0]);
sz = sizeof(Mailbox.C);
printf("Writing C[%uB] to address %08x...\n", sz, addr);
e_write(pDRAM, 0, 0, addr, (void *) Mailbox.C, sz);
#endif
clock_gettime(CLOCK_MONOTONIC, &timer[0]);
// Copy operand matrices to Epiphany system
addr = offsetof(shared_buf_t, A[0]);
sz = sizeof(Mailbox.A);
printf("Writing A[%uB] to address %08x...\n", sz, addr);
e_write(pDRAM, 0, 0, addr, (void *) Mailbox.A, sz);
addr = offsetof(shared_buf_t, B[0]);
sz = sizeof(Mailbox.B);
printf("Writing B[%uB] to address %08x...\n", sz, addr);
e_write(pDRAM, 0, 0, addr, (void *) Mailbox.B, sz);
// Call the Epiphany matmul() function
printf("GO Epiphany! ... ");
clock_gettime(CLOCK_MONOTONIC, &timer[1]);
matmul_go(pDRAM);
clock_gettime(CLOCK_MONOTONIC, &timer[2]);
printf("Finished calculating Epiphany result.\n");
// Read result matrix and timing
addr = offsetof(shared_buf_t, C[0]);
sz = sizeof(Mailbox.C);
printf("Reading result from address %08x...\n", addr);
e_read(pDRAM, 0, 0, addr, (void *) Mailbox.C, sz);
clock_gettime(CLOCK_MONOTONIC, &timer[3]);
// Calculate a reference result
printf("Calculating result on Host ... ");
clock_gettime(CLOCK_THREAD_CPUTIME_ID, &timer[4]);
#ifndef __DO_STRASSEN__
matmul(Mailbox.A, Mailbox.B, Cref, _Smtx);
#else
matmul_strassen(Mailbox.A, Mailbox.B, Cref, _Smtx);
#endif
clock_gettime(CLOCK_THREAD_CPUTIME_ID, &timer[5]);
printf("Finished calculating Host result.\n");
addr = offsetof(shared_buf_t, core.clocks);
sz = sizeof(Mailbox.core.clocks);
printf("Reading time from address %08x...\n", addr);
e_read(pDRAM,0, 0, addr, &Mailbox.core.clocks, sizeof(Mailbox.core.clocks));
// clocks = Mailbox.core.clocks;
// Calculate the difference between the Epiphany result and the reference result
printf("\n*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***\n");
printf("Verifying result correctness ... ");
matsub(Mailbox.C, Cref, Cdiff, _Smtx);
tdiff[0] = (timer[2].tv_sec - timer[1].tv_sec) * 1000 + ((double) (timer[2].tv_nsec - timer[1].tv_nsec) / 1000000.0);//total
tdiff[1] = (timer[1].tv_sec - timer[0].tv_sec) * 1000 + ((double) (timer[1].tv_nsec - timer[0].tv_nsec) / 1000000.0);//write
tdiff[2] = (timer[3].tv_sec - timer[2].tv_sec) * 1000 + ((double) (timer[3].tv_nsec - timer[2].tv_nsec) / 1000000.0);//read
tdiff[3] = (timer[5].tv_sec - timer[4].tv_sec) * 1000 + ((double) (timer[5].tv_nsec - timer[4].tv_nsec) / 1000000.0);//ref
// If the difference is 0, then the matrices are identical and the
// calculation was correct
if (iszero(Cdiff, _Smtx))
{
printf("C_epiphany == C_host\n");
rerval = 0;
} else {
printf("\n\nERROR: C_epiphany is different from C_host !!!\n");
rerval = 1;
}
printf("*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***\n");
printf("\n");
printf("Epiphany (compute): %9.1f msec (@ %03d MHz)\n" , tdiff[0], eMHz);
printf(" (write) : %9.1f msec \n" , tdiff[1]);
printf(" (read) : %9.1f msec\n" , tdiff[2]);
printf(" (*total*): %9.1f msec\n\n" , tdiff[2]+tdiff[1]+tdiff[0]);
printf("Host (*total*): %9.1f msec (@ %03d MHz)\n" , tdiff[3], aMHz);
#ifdef __DUMP_MATRICES__
printf("\n\n\n");
printf("A[][] = \n");
matprt(Mailbox.A, _Smtx);
printf("B[][] = \n");
matprt(Mailbox.B, _Smtx);
printf("C[][] = \n");
matprt(Mailbox.C, _Smtx);
printf("Cref[][] = \n");
matprt(Cref, _Smtx);
int i, j;
for (i=0; i<_Nside; i++)
for (j=0; j<_Nside; j++)
{
e_read(pEpiphany, i, j, 0x2000+0*sizeof(float), &Aepi[(i*_Score+0)*_Smtx + j*_Score], 2*sizeof(float));
e_read(pEpiphany, i, j, 0x2000+2*sizeof(float), &Aepi[(i*_Score+1)*_Smtx + j*_Score], 2*sizeof(float));
e_read(pEpiphany, i, j, 0x4000+0*sizeof(float), &Bepi[(i*_Score+0)*_Smtx + j*_Score], 2*sizeof(float));
e_read(pEpiphany, i, j, 0x4000+2*sizeof(float), &Bepi[(i*_Score+1)*_Smtx + j*_Score], 2*sizeof(float));
}
printf("Aepi[][] = \n");
matprt(Aepi, _Smtx);
printf("Bepi[][] = \n");
matprt(Bepi, _Smtx);
#endif
printf("\n* * * EPIPHANY FTW !!! * * *\n");
// Close connection to device
if (e_close(pEpiphany))
{
printf("\nERROR: Can't close connection to Epiphany device!\n\n");
exit(1);
}
if (e_free(pDRAM))
{
printf("\nERROR: Can't release Epiphany DRAM!\n\n");
exit(1);
}
e_finalize();
return rerval;
}
int main(int argc, char *argv[]){
e_loader_diag_t e_verbose;
e_platform_t platform;
e_epiphany_t dev, *pdev;
e_mem_t dram, *pdram;
size_t size;
int status=1;//pass
char elfFile[4096];
pdev = &dev;
pdram = &dram;
int a,b;
int i,j;
unsigned result[N];
unsigned data = 0xDEADBEEF;
unsigned tmp,fail;
int idelay[TAPS]={0x00000000,0x00000000,//0
0x11111111,0x00000001,//1
0x22222222,0x00000002,//2
0x33333333,0x00000003,//3
0x44444444,0x00000004,//4
0x55555555,0x00000005,//5
0x66666666,0x00000006,//6
0x77777777,0x00000007,//7
0x88888888,0x00000008,//8
0x99999999,0x00000009,//9
0xaaaaaaaa,0x0000000a,//10
0xbbbbbbbb,0x0000000b,//11
0xcccccccc,0x0000000c,//12
0xdddddddd,0x0000000d,//13
0xeeeeeeee,0x0000000e,//14
0xffffffff,0x0000000f,//15
0x00000000,0x00000010,//16
0x11111111,0x00000011,//17
0x22222222,0x00000012,//18
0x33333333,0x00000013,//29
0x44444444,0x00000014,//20
0x55555555,0x00000015,//21
0x66666666,0x00000016,//22
0x77777777,0x00000017,//23
0x88888888,0x00000018,//24
0x99999999,0x00000019,//25
0xaaaaaaaa,0x0000001a,//26
0xbbbbbbbb,0x0000001b,//27
0xcccccccc,0x0000001c,//28
0xdddddddd,0x0000001d,//29
0xeeeeeeee,0x0000001e,//30
0xffffffff,0x0000001f};//31
//Gets ELF file name from command line
strcpy(elfFile, "./bin/e-task.elf");
//Initalize Epiphany device
e_set_host_verbosity(H_D0);
e_init(NULL);
my_reset_system();
e_get_platform_info(&platform);
e_open(&dev, 0, 0, 1, 1); //open core 0,0
e_alloc(pdram, 0x00000000, 0x00400000);
//Set Idelay
ee_write_esys(0xF0310, idelay[2*7]);
ee_write_esys(0xF0314, idelay[2*7+1]);
//Start Program
e_load_group(elfFile, &dev, 0, 0, 1, 1, E_FALSE);
e_start_group(&dev);
usleep(1000000);
//Check status
int pre_stat,mbox_lo,mbox_hi,post_stat;
int ddata;
for(i=0;i<32;i++){
e_read(pdram,0,0, i, &ddata, sizeof(ddata));
pre_stat = ee_read_esys(0xF0738);
mbox_lo = ee_read_esys(0xF0730);
//mbox_hi = ee_read_esys(0xF0734);
post_stat = ee_read_esys(0xF0738);
printf ("PRE_STAT=%08x POST_STAT=%08x LO=%08x HI=%08x DDATA=%04x\n", pre_stat, post_stat, mbox_lo, mbox_hi,ddata);
}
for(i=0;i<16;i++){
e_read(pdram,0,0, i*4, &ddata, sizeof(ddata));
printf ("DDATA=%04x\n", ddata);
}
//Close down Epiphany device
e_close(&dev);
e_finalize();
//self check
if(status){
return EXIT_SUCCESS;
}
else{
return EXIT_FAILURE;
}
}
int main () {
unsigned int row, col, core, t;
e_platform_t platform;
e_epiphany_t device;
e_mem_t mem;
static msg_block_t msg;
memset(&msg, 0, sizeof(msg));
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
e_alloc(&mem, BUF_OFFSET, sizeof(msg_block_t));
/* Cómo sé que ^ pone el buffer en 0x8f00000000? */
/* Esta definido en el hdf por default. */
srand(SEED);
for (row = 0; row < platform.rows; row++) {
for (col = 0; col < platform.cols; col++) {
core = row*platform.cols + col;
msg.shared_msg[core].seed = SEED + core;
printf("A (%d,%d) le toco %d\n", row, col, msg.shared_msg[core].seed);
}
}
printf("\n---\n\n");
e_open(&device, 0, 0, platform.rows, platform.cols);
e_write(&mem, 0, 0, 0, &msg, sizeof(msg));
e_reset_group(&device);
e_load_group("epiphany.srec", &device, 0, 0, platform.rows, platform.cols, E_TRUE);
nano_wait(0, 10000000); /* Necesario para sincronizar? */
for (row = 0; row < platform.rows; row++) {
for (col = 0; col < platform.cols; col++) {
core = row*platform.cols + col;
t = 0;
while (E_TRUE) { /* espero hasta que cambie algo */
e_read(&mem,
0,
0,
(off_t) ((char *)&msg.shared_msg[core] - (char *)&msg),
&msg.shared_msg[core],
sizeof(msg_info_t));
if (msg.shared_msg[core].coreid != 0) {
printf("Termino %d\n", core);
break;
}
printf(".");
nano_wait(0, 1000000);
if (t++ == 10) {
printf("Colgo %d\n", core);
break;
}
}
}
}
/* Ya hice todo lo que tenia que hacer, falta updatear. */
nano_wait(0, 1000000);
for (row = 0; row < platform.rows; row++) {
for (col = 0; col < platform.cols; col++) {
core = row*platform.cols + col;
e_read(&mem,
0,
0,
(off_t) ((char *)&msg.shared_msg[core] - (char *)&msg),
&msg.shared_msg[core],
sizeof(msg_info_t));
}
}
for (row = 0; row < platform.rows; row++) {
for (col = 0; col < platform.cols; col++) {
core = row*platform.cols + col;
printf("Hola, soy %#03x [%u] (%-2d, %-2d)! Tengo el mensaje %#03x, "
"recibi el mensaje %u, y tarde %u ticks en procesar todo. "
"seed ahora vale %d.\n",
msg.shared_msg[core].coreid,
msg.shared_msg[core].coreid,
msg.shared_msg[core].coreid >> 6,
msg.shared_msg[core].coreid & 0x3f,
msg.shared_msg[core].msg,
msg.shared_msg[core].external,
msg.shared_msg[core].timer,
msg.shared_msg[core].seed);
}
}
e_close(&device);
e_free(&mem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
unsigned rows, cols, coreid, i, j, flag, fail = 0;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
unsigned time[sizeN];
unsigned result[sizeN];
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
e_alloc(&emem, 0x01800000, 0x4000);
//open the workgroup
rows = platform.rows;
cols = platform.cols;
e_open(&dev, 0, 0, rows, cols);
//load the device program on the board
e_load_group("emain.elf", &dev, 0, 0, rows, cols, E_FALSE);
for (i=0; i<rows; i++)
{
for (j=0; j<cols; j++)
{
coreid = (i + platform.row) * 64 + j + platform.col;
fprintf(stderr, "Message from eCore 0x%03x (%2d,%2d): \n", coreid, i, j);
flag = 0;
e_write(&emem, 0, 0, 0x3000, &flag, sizeof(flag));
e_start(&dev, i, j);
//wait for core to execute the program
while (!flag) {
e_read(&emem, 0, 0, 0x3000, &flag, sizeof(flag));
usleep(1000);
}
//check results
e_read(&emem, 0, 0, 0x1000, &result[0], sizeN*sizeof(unsigned));
e_read(&emem, 0, 0, 0x2000, &time[0], sizeN*sizeof(unsigned));
if ((result[1] == result[0]) && (result[1] == result[2]) && (time[1]<time[0]) && (time[1]<time[2]))
fprintf(stderr, "\ntest hardware_loop passed!\n\n");
else
{
fprintf(stderr, "\ntest hardware_loop failed!\n");
fprintf(stderr, "result:\tauto = %10d hw = %10d sf = %10d \n", result[0],result[1],result[2]);
fprintf(stderr, "time: \tauto = %5d cycles hw = %5d cycles sf = %5d cycles \n\n", time[0],time[1],time[2]);
fail++;
}
}
}
// Release the allocated buffer and finalize the
// e-platform connection.
e_close(&dev);
e_free(&emem);
e_finalize();
return !(fail == 0);
}
int main(int argc, char *argv[])
{
int i,j;
unsigned row, col, coreid;
unsigned di, ci, go, go_all;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
// Initialize progress state in mailbox
for(i=0;i<corenum;i++)
for(j=0;j<40;j++)
M[i][j] ='\0';
// initialize system, read platform params from
// default HDF. Then, reset the platform.
e_init(NULL);
fprintf(stderr, "finished init\n");
e_reset_connected_system();
fprintf(stderr, "finished reset\n");
e_get_platform_info(&platform);
fprintf(stderr, "platform info: num_chips =0x%x ,emems = 0x%x\n",platform.num_chips,platform.num_emems);
fprintf(stderr, "hdf_ platform info: core (0,0) id =%x\n ", (0 + platform.row) * 64 + 0 + platform.col );
fprintf(stderr, "platform info: rows =0x%x ,cols = 0x%x\n",platform.rows,platform.cols );
corenum=platform.rows*platform.cols;
// Open the first and second cores for master and slave programs, resp.
fprintf(stderr, "starting e_open\n");
usleep(1e6);
e_open(&dev, 0, 0, platform.rows, platform.cols);
fprintf(stderr, "finished e_open\n");
// Allocate the ext. mem. mailbox
e_alloc(&emem, _BufOffset, sizeof(M));
// Load programs on cores.
fprintf(stderr, "starting e_load\n");
e_load_group("e-int-test.master.srec", &dev, 0, 0, platform.rows, platform.cols, E_FALSE);
fprintf(stderr, "starting e_load\n");
//e_load("e-int-test.slave.srec", &dev, 0, 1, E_FALSE);
// clear mailbox.
e_write(&emem, 0, 0, (off_t) (0x0000), (void *) &(M[0]), sizeof(M));
usleep(500e3);
// Print mbox status.
print_mbox(&dev, &emem, "1. Clearing mbox:");
// start the master program
e_start_group(&dev);
printf("started :\n");
usleep(2e6);
print_mbox(&dev, &emem, "2. started:");
//usleep(4e6);
usleep(3e6);
print_mbox(&dev, &emem, "3. started:");
// At this point, the mailbox should contain all of the progress
// indicators, and look like the following:
//
// 0x808 0x809 0x22222222 0x33333333 0x44444444
//
// If there is a "0xdeadbeef" state in one of the slots, it means
// that something went wrong.
// Finalize
e_close(&dev);
e_free(&emem);
e_finalize();
return 0;
}
int main()
{
e_epiphany_t group0;
e_mem_t shm1;
host_chan_t chan2;
e_mem_t shm3;
pthread_t t5;
bool r12;
e_init(0);
e_reset_system();
e_open(&group0, 0, 0, 4, 4);
e_reset_group(&group0);
setup_queues();
e_alloc(&shm1, sa2, 2048);
init_host_chan(&chan2, &group0, 0, 0, &shm1, la0, la1);
init_core_chan(&group0, 0, 1, la3, la4);
e_load("core0.srec", &group0, 0, 0, 1);
init_core_chan(&group0, 0, 2, la6, la7);
e_load("core1.srec", &group0, 0, 1, 1);
init_core_chan(&group0, 0, 3, la9, la10);
e_load("core2.srec", &group0, 0, 2, 1);
init_core_chan(&group0, 1, 3, la12, la13);
e_load("core3.srec", &group0, 0, 3, 1);
init_core_chan(&group0, 1, 2, la15, la16);
e_load("core7.srec", &group0, 1, 3, 1);
init_core_chan(&group0, 1, 1, la18, la19);
e_load("core6.srec", &group0, 1, 2, 1);
init_core_chan(&group0, 1, 0, la21, la22);
e_load("core5.srec", &group0, 1, 1, 1);
init_core_chan(&group0, 2, 0, la24, la25);
e_load("core4.srec", &group0, 1, 0, 1);
init_core_chan(&group0, 2, 1, la27, la28);
e_load("core8.srec", &group0, 2, 0, 1);
init_core_chan(&group0, 2, 2, la30, la31);
e_load("core9.srec", &group0, 2, 1, 1);
init_core_chan(&group0, 2, 3, la33, la34);
e_load("core10.srec", &group0, 2, 2, 1);
init_core_chan(&group0, 3, 3, la36, la37);
e_load("core11.srec", &group0, 2, 3, 1);
init_core_chan(&group0, 3, 2, la39, la40);
e_load("core15.srec", &group0, 3, 3, 1);
init_core_chan(&group0, 3, 1, la42, la43);
e_load("core14.srec", &group0, 3, 2, 1);
init_core_chan(&group0, 3, 0, la45, la46);
e_load("core13.srec", &group0, 3, 1, 1);
e_alloc(&shm3, sa50, 2048);
init_host_chan(&chan4, &group0, 3, 0, &shm3, la48, la49);
e_load("core12.srec", &group0, 3, 0, 1);
pthread_create(&t5, NULL, thread_t5, NULL);
r12 = true;
while (1) {
bool v13;
float _a14[512];
float *a14 = _a14;
bool v15;
v13 = r12;
if (!v13)
break;
v15 = receive_samples(a14);
if (v15) {
uint32_t r16;
float _a17[512];
float *a17 = _a17;
uint32_t v18;
bool v19;
r16 = 512;
r16 = 512;
for (v18 = 0; v18 < 512; v18++) {
a17[v18] = a14[v18];
}
v19 = host_write_h2c(chan2, a17, 0, r16);
r12 = v19;
} else {
r12 = false;
}
}
host_close_chan(chan2);
pthread_join(t5, NULL);
teardown_queues();
e_free(&shm1);
e_free(&shm3);
e_reset_group(&group0);
e_close(&group0);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
unsigned row, col, coreid, i, j;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
double tdiff;
mailbox.flag = -1;
matrix_init(0.0);
//matrix_init(54.0);
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, sizeof(Mailbox_t));
row = 0;
col = 0;
coreid = (row + platform.row) * 64 + col + platform.col;
fprintf(stderr,"\n\nMultiplying A[%d][%d] x B[%d][%d] = C[%d][%d]\n",_Smtx,_Smtx,_Smtx,_Smtx,_Smtx,_Smtx);
fprintf(stderr, "\nGroup rows: %d Group_cols: %d. Starting row: %d col : %d\n",group_rows_num,group_cols_num,row,col);
// Open the single-core workgroup. Note that we used
// core coordinates relative to the workgroup.
e_open(&dev, row, col, group_rows_num, group_cols_num);
// Load the device program onto the selected eCore
// and launch after loading.
int load_err=e_load_group("matmul_multi.elf", &dev, 0, 0, group_rows_num, group_cols_num, E_FALSE);
char load_result[5];
if (load_err == E_OK) strcpy(load_result,"E_OK");
if (load_err == E_ERR) strcpy(load_result,"E_ERR");
if (load_err == E_WARN) strcpy(load_result,"E_WARN");
fprintf(stderr,"Load result: %s\n",load_result);
//gettimeofday(&timer[0], NULL);
e_start_group(&dev);
unsigned int addr = offsetof(Mailbox_t, flag);
while (mailbox.flag != 0) {
e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
}
//fprintf(stderr,"\nReceived Ready signal from Epiphany: %d\n",mailbox.flag);
//Initialize and write the matrix to shared memory
addr = offsetof(Mailbox_t, A[0]);
e_write(&emem, 0, 0, addr, (void *)mailbox.A, sizeof(mailbox.A));
addr = offsetof(Mailbox_t, B[0]);
e_write(&emem, 0, 0, addr, (void *)mailbox.B, sizeof(mailbox.B));
addr = offsetof(Mailbox_t, C[0]);
e_write(&emem, 0, 0, addr, (void *)mailbox.C, sizeof(mailbox.C));
mailbox.flag = 1;
addr = offsetof(Mailbox_t, flag);
//fprintf(stderr,"\nSending Ready signal to Epiphany: %d\n",mailbox.flag);
e_write(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
while (mailbox.flag != 2)
e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
gettimeofday(&timer[0], NULL);
while (mailbox.flag != 3)
e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
gettimeofday(&timer[1], NULL);
//usleep(3000000);
//e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
// Wait for core program execution to finish, then
// read mailbox from shared buffer.
while (mailbox.flag != 4) {
e_read(&emem, 0, 0, addr, &mailbox.flag, sizeof(mailbox.flag));
}
// Wait for core program execution to finish, then
// read message from shared buffer.
//usleep(100000);
addr = offsetof(Mailbox_t, output);
e_read(&emem, 0, 0, addr, &mailbox.output, sizeof(mailbox.output));
addr = offsetof(Mailbox_t, C[0]);
e_read(&emem, 0, 0, addr, (void *)mailbox.C, sizeof(mailbox.C));
print_to_file("../output/optresult",(void *)mailbox.C);
// Print the message and close the workgroup.
e_close(&dev);
tdiff = (timer[1].tv_sec - timer[0].tv_sec) * 1000 + ((double) (timer[1].tv_usec - timer[0].tv_usec) / 1000.0);
//printf("Optimized MATMUL time: %d cycles\tTime: %9.9f msec\n", mailbox.output.clocks,clock_to_time(mailbox.output.clocks));
//printf("Optimized MATMUL Exec time: %d cycles\tTime: %9.9f msec\n", mailbox.output.dummy1,clock_to_time(mailbox.output.dummy1));
//printf("Optimized MATMUL Shared memory Comms time: %d cycles\tTime: %9.9f msec\n", (mailbox.output.clocks-mailbox.output.dummy1),clock_to_time(mailbox.output.clocks-mailbox.output.dummy1));
//
//printf("\nTime from host: %9.6f msec\n",tdiff);
float gflops = ((2.0 * _Smtx * _Smtx * _Smtx)/(clock_to_time(mailbox.output.clocks)/1000))/1000/1000/1000;
float gflops2 = ((2.0 * _Smtx * _Smtx * _Smtx)/(clock_to_time(mailbox.output.dummy1)/1000))/1000/1000/1000;
printf("\nGFlops (On chip): %9.6f\tPerformance = %9.4f %% of peak\n",gflops2,gflops2/(1.2*group_rows_num*group_cols_num)*100);
printf("\nGFlops (including off-chip transfers): %9.6f\tPerformance = %9.4f %% of peak\n",gflops,gflops/(1.2*group_rows_num*group_cols_num)*100);
fprintf(stderr,"\n");
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return 0;
}
int main(int argc, char *argv[])
{
e_platform_t platform;
e_epiphany_t Epiphany, *pEpiphany;
e_mem_t DRAM, *pDRAM;
unsigned int msize;
float seed;
unsigned int addr; //, clocks;
size_t sz;
int result, rerval;
pEpiphany = &Epiphany;
pDRAM = &DRAM;
msize = 0x00400000;
// load j1.bin into shared mem
uint8_t ROM[1<<16];
printf("load %s \n", argv[1]);
FILE* f = fopen("j1.bin","r");
int i = 0;
uint16_t op;
while(fread(&op, 2, 1, f)) {
*(ROM+i*2+1) = op&0xff;
*(ROM+i*2) = op>>8;
i++;
}
fclose(f);
printf ("read %d words\n",i);
for(int i=0;i<50;i++)printf("%x ", (uint16_t)*(ROM+2*i));
printf("\n");
get_args(argc, argv);
fo = stdout;
fi = stdin;
// Connect to device for communicating with the Epiphany system
// Prepare device
e_set_host_verbosity(H_D0);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
if (e_alloc(pDRAM, 0x00000000, msize))
{
fprintf(fo, "\nERROR: Can't allocate Epiphany DRAM!\n\n");
exit(1);
}
if (e_open(pEpiphany, 0, 0, 1, 1))
{
fprintf(fo, "\nERROR: Can't establish connection to Epiphany device!\n\n");
exit(1);
}
e_reset_core(pEpiphany, 0, 0);
//fprintf(fo, "host base %x \n", pDRAM->base); fflush(fo);
// init all
for(int i=0;i<16;i++){
int n = 0;
addr = offsetof(shared_buf_t, core.seq);
e_mwrite_buf(pDRAM, addr, &n, sizeof(int));
addr = offsetof(shared_buf_t, core.go_out);
e_mwrite_buf(pDRAM, addr, &n, sizeof(int));
}
addr = offsetof(shared_buf_t, DRAM);
e_mwrite_buf(pDRAM, addr, ROM, sizeof(ROM));
printf("Loading program on Epiphany chip...\n");
//e_set_loader_verbosity(ar.verbose);
//result = e_load_group(ar.srecFile, pEpiphany, 0, 0, pEpiphany->rows, pEpiphany->cols, ar.run_target);
result = e_load(ar.srecFile, pEpiphany, 0, 0, ar.run_target);
if (result == E_ERR) {
printf("Error loading Epiphany program.\n");
exit(1);
}
// Generate operand matrices based on a provided seed
matrix_init(seed);
sleep(2);
struct sockaddr_in si_me, si_other;
int s, slen, n=0;
if ((s=socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP))==-1) {
printf("socket open error\n");
return -1;
}
memset((char *) &si_me, 0, sizeof(si_me));
si_me.sin_family = AF_INET;
si_me.sin_port = htons(27777);
si_me.sin_addr.s_addr = htonl(INADDR_ANY);
if (bind(s, (struct sockaddr *) &si_me, sizeof(si_me))==-1) {
printf("ERROR, bind failed\n");
return -1;
}
int ep_seq[16];
bzero(ep_seq, 16*sizeof(int) );
while (1) {
// read command from udp client
char buf[1024];
ssize_t len =
recvfrom(s, buf, 255, 0, (struct sockaddr *)&si_other, (socklen_t*)&slen);
if (len == -1) {
printf("socket error\n");
break;
}
printf("Received packet from %s:%d\nlen %d Data: %s\n",
inet_ntoa(si_other.sin_addr), ntohs(si_other.sin_port), len, buf);
// R then L load vm, G run, R reset
n = buf[0];
printf("seq %d core %d cmd %c\n", ep_seq[n], n, buf[1]);
if('s' == buf[1])printf("set debug mask %x \n", buf[2]);
else if('L' == buf[1]) {
pthread_create(&trace_reader_a, NULL, trace_reader, pDRAM);
}
char string[256] = "aaa bbb ccc ";
char s_out[256];
int addr_to = offsetof(shared_buf_t, core.go[n]);
int addr_out = offsetof(shared_buf_t, core.go_out[n]);
int addr_seq = offsetof(shared_buf_t, core.seq[n]);
int addr_core_seq = offsetof(shared_buf_t, core.core_seq[n]);
// send command
memcpy(string,buf+1,10);
sz= e_mwrite_buf(pDRAM, addr_to, string, 25 );
sz= e_mwrite_buf(pDRAM, addr_seq, &ep_seq[n], sizeof(uint32_t) );
ep_seq[n] ++;
sleep(1);
// read from core
uint32_t core_out;
sz = e_mread_buf(pDRAM, addr_out, s_out, 25);
result = e_mread_buf(pDRAM, addr_core_seq, &core_out,sizeof(uint32_t));
printf("check seq %x and output %s\n", core_out, s_out);
}
// Close connection to device
if (e_close(pEpiphany))
{
fprintf(fo, "\nERROR: Can't close connection to Epiphany device!\n\n");
exit(1);
}
if (e_free(pDRAM))
{
fprintf(fo, "\nERROR: Can't release Epiphany DRAM!\n\n");
exit(1);
}
return rerval;
}
int main(int argc, char *argv[])
{
unsigned row, col, coreid, i,j,m,n,k,l;
unsigned int acc = 0;
uint32_t flag[41];
const uint32_t zero = 0;
uint32_t done = 0;
e_platform_t platform;
e_epiphany_t dev;
e_mem_t emem;
char emsg[_BufSize];
int errors = 0;
srand(1);
// initialize system, read platform params from
// default HDF. Then, reset the platform and
// get the actual system parameters.
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
// Allocate a buffer in shared external memory
// for message passing from eCore to host.
e_alloc(&emem, _BufOffset, _BufSize);
// Open a workgroup
e_open(&dev, 0, 0, platform.rows, platform.cols);
// Load the device program onto all the eCores
// To get the verified values
//e_load_group("e_math_test.elf", &dev, 0, 0, platform.rows, platform.cols, E_FALSE);
// To test
e_load_group("e_math_test", &dev, 0, 0, platform.rows, platform.cols, E_FALSE);
for (i=0; i<platform.rows ; i++)
{
for (j=0; j<platform.cols; j++)
{
//Draw to a certain core
row=i;
col=j;
coreid = (row + platform.row) * 64 + col + platform.col;
//fprintf(stderr,"%3d: Message from eCore 0x%03x (%2d,%2d) : \n",(i*platform.cols+j),coreid,row,col);
e_write(&dev, i, j, 0x5ffc, &zero, sizeof(zero));
e_start(&dev, i, j);
done = 0;
while (!done) {
e_read(&dev, i, j, 0x5ffc, &done, sizeof(done));
usleep(1000);
}
// Wait for core program execution to finish
// Read message from shared buffer
//usleep(100000);
e_read(&emem, 0, 0, 0x0, emsg, _BufSize);
e_read(&dev, i, j, 0x6000, &flag, sizeof(flag));
// Print the message and close the workgroup.
if(flag[40] == 40) {
for (l = 0; l < 40; l+=4) {
fprintf(stdout, "%d ", flag[l]);
fprintf(stdout, "%d ", flag[l + 1]);
fprintf(stdout, "%d ", flag[l + 2]);
fprintf(stdout, "%d | ", flag[l + 3]);
}
acc = 0;
for (l = 0; l < 40; l++) {
acc += flag[l];
acc += flag[l + 1];
acc += flag[l + 2];
acc += flag[l + 3];
}
fprintf(stdout, "total: %d\n", acc);
} else {
fprintf(stderr,"FAIL! %d\n", flag[20]);
errors++;
}
//Only print out messages on core 0
if(i==0 & j==0){
fprintf(stderr, "%s\n", emsg);
}
}
}
// Close the workgroup
e_close(&dev);
// Release the allocated buffer and finalize the
// e-platform connection.
e_free(&emem);
e_finalize();
return errors;
}
