int main()
{
//counters for row and colum, cored id and loop counter
unsigned row_loop,col_loop;
// this will contain the epiphany platform configura tion
e_platform_t epiphany;
e_epiphany_t dev;
e_mem_t memory;
e_mem_t memory2;
int message;
int core1;
int core2;
int message2;
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
for(row_loop=0; row_loop <4; row_loop ++)
{
for(col_loop=0; col_loop <3; col_loop = col_loop+2)
{
//one core within the parallella work group is 1 x 2 i.e dual core
e_open(&dev,row_loop,col_loop,1,2);
//reset the group
e_reset_group(&dev);
//load the group
result = e_load("hello_world.srec",&dev,0,0,E_FALSE);
if (result != E_OK) {
fprintf(stderr,"Error Loading the Epiphany Application 1 %i\n", result);
}
result = e_load("hello_world2.srec",&dev,0,1,E_FALSE);
if (result != E_OK) {
fprintf(stderr,"Error Loading the Epiphany Application 2 %i\n", result);
}
e_start_group(&dev);
usleep(10000);
e_read(&dev,0,0,0x3000, &message, sizeof(int));
e_read(&dev,0,0,0x3004, &core1,sizeof(int));
e_read(&dev,0,1,0x3000, &message2, sizeof(int));
e_read(&dev,0,1,0x3004, &core2,sizeof(int));
fprintf(stderr,"message from core %d ", core1);
fprintf(stderr,"core id = 0x%03x \n", message);
fprintf(stderr,"message from core %d ", core2);
fprintf(stderr,"core id = 0x%03x \n", message2);
e_close(&dev);
}
}
e_free(&memory);
e_finalize();
return 0;
}
int main(int argc, char *argv[]) {
unsigned row, col, coreid, i, j;
e_platform_t platform;
e_epiphany_t dev;
// Initialize the Epiphany HAL and connect to the chip
e_init(NULL);
// Reset the system
e_reset_system();
// Get the platform information
e_get_platform_info(&platform);
// Create a workgroup using all of the cores
e_open(&dev, 0, 0, platform.rows, platform.cols);
e_reset_group(&dev);
// Load the device code into each core of the chip, and don't start it yet
e_load_group("e-acc.elf", &dev, 0, 0, platform.rows, platform.cols, E_FALSE);
e_start_group(&dev);
uint64_t iterations;
while (MAX_ITERATIONS*16>iterations) {
for(row=0;row<platform.rows;row++)
{
for(col=0;col<platform.cols;col++)
{
uint64_t timestep;
uint64_t iter_num;
float loc;
float vel;
if(e_read(&dev, row, col, 0x7000, ×tep, sizeof(uint64_t)) != sizeof(uint64_t)){
fprintf(stderr, "Failed to read\n");
}
if(e_read(&dev, row, col, 0x7008, &iter_num, sizeof(uint64_t)) != sizeof(uint64_t)){
fprintf(stderr, "Failed to read\n");
}
if(e_read(&dev, row, col, 0x7010, &loc, sizeof(float)) != sizeof(float)){
fprintf(stderr, "Failed to read\n");
}
if(e_read(&dev, row, col, 0x7010 + 4*16, &vel, sizeof(float)) != sizeof(float)){
fprintf(stderr, "Failed to read\n");
}
fprintf(stderr, "The timestep is %d, the iterations*16=%d, the position is %f, and the vel is %f\n", timestep, iter_num, loc, vel);
iterations += iter_num;
}
}
}
}
int main()
{
//counters for row and colum, cored id and loop counter
unsigned row_loop,col_loop;
// this will contain the epiphany platform configuration
e_platform_t epiphany;
e_epiphany_t dev;
e_return_stat_t result;
int message;
int message2;
int loop;
int addr;;
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
//one core within the parallella work group is 1 x 2 i.e dual core
e_open(&dev,0,0,1,2);
//reset the group
e_reset_group(&dev);
//load the group
result = e_load("hello_world.srec",&dev,0,0,E_FALSE);
if (result != E_OK) {
fprintf(stderr,"Error Loading the Epiphany Application 1 %i\n", result);
}
result = e_load("hello_world2.srec",&dev,0,1,E_FALSE);
if (result != E_OK) {
fprintf(stderr,"Error Loading the Epiphany Application 2 %i\n", result);
}
e_start_group(&dev);
usleep(10000);
fprintf(stderr,"extracting from core memory \n ");
addr = 0x3000;
for(loop = 0; loop <20; loop ++) {
e_read(&dev,0,1,addr, &message, sizeof(int));
if (loop == 0)
fprintf(stderr,"message from core 0x%03x \n ", message);
else
fprintf(stderr,"message from core %d \n ", message);
addr = addr+0x04;
}
e_close(&dev);
e_finalize();
fprintf(stderr,"demo complete \n ");
return 0;
}
/*
* Main entry
*/
int main(int argc, char * argv[]) {
// Arguments handling
switch(argc) {
case 4: nb_cores = atoi(argv[3]);
case 3: sub_iteration = atoi(argv[2]);
case 2: main_iteration = atoi(argv[1]);
case 1: break;
default:
printf("Wrong number of args\nUsage: ./main.elf [main iterations] [sub iteration] [nb cores]\n");
return 0;
}
// Init the epiphany platform
e_platform_t platform;
e_epiphany_t dev;
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
e_open(&dev, 0, 0, 4, 4);
e_load_group("emain.srec", &dev, 0, 0, 4, 4, E_FALSE); // don't start immediately
e_start_group(&dev); // Start workgroup
unsigned ncores = nb_cores; if(ncores>16) exit(0);
unsigned k;
for(k = 0; k < ncores; k++) {
e_write(&dev, k/4, k%4, 0x400c, &sub_iteration, sizeof(unsigned));
}
// >>>>> Begin benchamrk
float start_t = second();
unsigned i,j;
float res = 0;
unsigned go = 1;
unsigned free_not_found = 1;
i = j = 0;
for(; i < main_iteration; i++) {
free_not_found = 1;
while(free_not_found) {
unsigned state;
e_read(&dev, j/4, j%4, 0x4008, &state, sizeof(unsigned));
if(state == 0) { // 1 busy, 0 free
float temp_res;
e_read(&dev, j/4, j%4, 0x4004, &temp_res, sizeof(float));
res += temp_res;
unsigned instruction = i; // copy i
e_write(&dev, j/4, j%4, 0x4000, &instruction, sizeof(unsigned));
e_write(&dev, j/4, j%4, 0x4008, &go, sizeof(unsigned));
free_not_found = 0;
}
j = (++j)%ncores;
}
}
// Be sure not to leave a core still working
for(j = 0; j < ncores; j++) {
unsigned state;
e_read(&dev, j/4, j%4, 0x4008, &state, sizeof(unsigned));
while(state == 1) {
e_read(&dev, j/4, j%4, 0x4008, &state, sizeof(unsigned));
}
float temp_res;
e_read(&dev, j/4, j%4, 0x4004, &temp_res, sizeof(float));
res += temp_res;
}
res *= 4;
float end_t = second();
// <<<<< End benchmark
float spent_t = end_t - start_t;
#ifdef STAT
printf("%i,\t%i,\t%f, \t%f\n", main_iteration, sub_iteration, spent_t, res);
#else
printf("PI = %f\ttime spent %fs\n", res, spent_t);
#endif
return 0;
}
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[]){
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(int argc, char *argv[])
{
unsigned row, col, coreid, i, j, m, n, k;
e_platform_t platform;
e_epiphany_t dev;
/* Assume 600 Mhz clock frequency. */
unsigned clk_max = 5600; /* 10 stddev */
unsigned clk_min = 1500;
unsigned num;
unsigned counter = 0;
const uint32_t one = 1;
const uint32_t zero = 0;
int err = 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);
// 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_FALSE);
// Load the device program onto all the other eCores
e_load_group("e_mutex_test.srec", &dev, 0, 1, 1, 3, E_FALSE);
e_load_group("e_mutex_test.srec", &dev, 1, 0, 3, 4, E_FALSE);
usleep(1000);
/* Clear the go flag */
e_write(&dev, 0, 0, 0x6400, &zero, sizeof(zero));
usleep(1000);
/* Start all cores */
e_start_group(&dev);
/* Give core0 plenty of time to initialize mutex */
usleep(10000);
/* Go! */
e_write(&dev, 0, 0, 0x6400, &one, sizeof(one));
// Wait for core program execution to finish
usleep(10000);
/* Read results from core0 */
e_read(&dev, 0, 0, 0x6200, &num, sizeof(num));
e_read(&dev, 0, 0, 0x6300, &counter, sizeof(counter));
/* Clear go flag */
e_write(&dev, 0, 0, 0x6400, &zero, sizeof(zero));
// Print the message
fprintf(stderr, "The counter now is %d\n", counter);
fprintf(stderr, "The clock cycle is %d\n", num);
if((num < clk_max)&&(num > clk_min))
{
fprintf(stderr, "Clock: PASS\n");
} else {
fprintf(stderr, "Clock: FAIL\n");
err = 1;
}
if (counter == 16) {
fprintf(stderr, "Counter: PASS\n");
} else {
fprintf(stderr, "Counter: FAIL\n");
err = 1;
}
// Close the workgroup
e_close(&dev);
// Finalize the e-platform connection.
e_finalize();
return err;
}
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 device;
w_mapper_t mapper;
w_sa_config_t sa_config;
w_list_t task1;
w_core_id_t first_id, last_id;
Mailbox mailbox;
memset(&mailbox, 0, sizeof(mailbox));
w_init_list(&task1, 0);
printf("=== Initializing system\n");
e_set_host_verbosity(H_D0);
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
printf("=== Creating workgroup\n");
e_open(&device, 0, 0, platform.rows, platform.cols);
printf("=== Mapping device program\n");
e_reset_group(&device);
// Initialize the mapping system: we will use this to automatically map our
// application, to assign a device program to each core.
mapper = w_create_mapper(platform.rows, platform.cols);
// Tell the mapper about our application. It needs to know about allocated
// tasks, constraints and faults are optional. By default the mapper assumes
// no faults and no constraints.
w_set_allocation_matrix(&mapper, allocation_matrix);
w_set_constraint_matrix(&mapper, constraint_matrix);
w_set_fault_matrix(&mapper, fault_matrix);
// Find the ID of all cores allocated for task 1, and create a link between
// each core. Links are used to indicate which tasks that communicate with
// each other, and to minimize the distance between communicating tasks.
w_find_in_matrix(&mapper.allocation_matrix, &task1, 1);
connect_task1(&mapper, &task1);
// Map the application using simulated annealing. This is will optimize our
// poor allocation.
sa_config = w_create_sa_config();
if (w_assign_sa(&mapper, &sa_config) != E_OK) {
printf("ERROR: Assignment failed.\n");
return 1;
}
w_print_matrix(&mapper.assignment_matrix, "Assignment");
w_print_matrix(&mapper.mapping_matrix, "Mapping");
// Find the ID of all cores assigned to task 1.
w_find_in_matrix(&mapper.assignment_matrix, &task1, 1);
first_id = task1.elements[0];
last_id = task1.elements[task1.size - 1];
printf("=== Setting initial conditions\n");
init_task1(&device, &task1);
printf("=== Loading device program\n");
// Load the device program onto all cores assigned to task 1.
if (w_load(&device, &task1, "e_main.srec") != E_OK) {
printf("ERROR: Unable to load device program.\n");
return 1;
}
printf("=== Starting device\n");
e_start_group(&device);
printf("=== Starting counter\n");
mailbox.go = 1;
w_write(&device, first_id, _MAILBOX_ADDRESS, &mailbox, sizeof(mailbox));
printf("=== Waiting for last core\n");
do {
msleep(100);
w_read(&device, last_id, _MAILBOX_ADDRESS, &mailbox, sizeof(mailbox));
} while (mailbox.done == 0);
printf("Counted to %i (expected %i)\n", mailbox.counter, task1.size - 1);
printf("=== Finalizing\n");
w_free_mapper(&mapper);
w_free_list(&task1);
e_close(&device);
e_finalize();
printf("=== Done\n");
return 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(int argc, char *argv[])
{
unsigned row, col, core;
e_platform_t platform;
e_epiphany_t dev;
int all_done, core_done;
int i, x, y;
int res;
unsigned int *fp;
fb = open(FB, O_RDWR);
if (fb == -1) {
perror("Unable to open fb " FB);
return 1;
}
// rest here
res = ioctl(fb, FBIOGET_FSCREENINFO, &fix);
if (res != 0) {
perror("getscreeninfo failed");
close(fb);
return 1;
}
//printf("framebuffer size %d @ %08x\n", fix.smem_len, fix.smem_start);
res = ioctl(fb, FBIOGET_VSCREENINFO, &var);
if (res != 0) {
perror("getscreeninfo failed");
close(fb);
return 1;
}
//printf("size %dx%d @ %d bits per pixel\n", var.xres_virtual, var.yres_virtual, var.bits_per_pixel);
fp = mmap(NULL, fix.smem_len, O_RDWR, MAP_SHARED, fb, 0);
if (fp == (void *)-1) {
perror("mmap failed");
close(fb);
return 1;
}
//printf("virtual @ %p\n", fp);
int stride = var.xres_virtual;
srand(time(NULL));
int nBodies = 30000;
int startFlag = 0x00000001;
int iters = 5000; // simulation iterations
const float dt = 0.05f; // time step
if (argc > 1) nBodies = atoi(argv[1]);
if (argc > 2) iters = atoi(argv[2]);
Body *buf = (Body *) malloc(sizeof(Body) * nBodies);
Body *bufOutput = (Body *) malloc(sizeof(Body) * nBodies);
randomizeBodies(buf, nBodies); // Init pos / vel data
e_init(NULL);
e_reset_system();
e_get_platform_info(&platform);
e_open(&dev, 0, 0, 4, 4);
e_reset_group(&dev);
e_load_group("e_rob_nbody.srec", &dev, 0, 0, 4, 4, E_FALSE);
for (row = 0; row < platform.rows; row++){
for (col = 0; col < platform.cols; col++){
e_write(&dev, row, col, 0x00000004, &nBodies, sizeof(int));
}
}
e_write(&dev, 0, 0, 0x1000, (Body *) buf, sizeof(Body) * nBodies);
x = 0;
// for(x = 0; x < iters; x++){
while(1){
//fprintf(stderr, "Iter %d\n", x);
for (row = 0; row < platform.rows; row++){
for (col = 0; col < platform.cols; col++){
e_write(&dev, row, col, 0x00000008, &startFlag, sizeof(int));
}
}
e_start_group(&dev);
//Check if all cores are done
while(1){
all_done = 0;
for (row = 0; row < platform.rows; row++){
for (col = 0; col < platform.cols; col++){
e_read(&dev, row, col, 0x00000008, &core_done, sizeof(int));
all_done += core_done;
}
}
if(all_done == 0){
break;
}
}
e_read(&dev, 0, 0, 0x1000, (Body *) bufOutput, sizeof(Body) * nBodies);
if(x != 0){
draw_stars(buf, nBodies, fp, stride, 0x00000000);
}
else{
x = 1;
}
draw_stars(bufOutput, nBodies, fp, stride, 0x00ffffff);
memcpy(buf, bufOutput, sizeof(Body) * nBodies);
}
e_close(&dev);
// Release the allocated buffer and finalize the
// e-platform connection.
e_finalize();
return 0;
}
static bool epiphany_scrypt(struct thr_info *thr, const unsigned char __maybe_unused *pmidstate,
unsigned char *pdata, unsigned char __maybe_unused *phash1,
unsigned char __maybe_unused *phash, const unsigned char *ptarget,
uint32_t max_nonce, uint32_t *last_nonce, uint32_t n)
{
uint32_t i;
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;
uint8_t *core_working = calloc(rows*cols, sizeof(uint8_t));
uint32_t *core_nonce = calloc(rows*cols, sizeof(uint32_t));
uint32_t cores_working = 0;
uint32_t *nonce = (uint32_t *)(pdata + 76);
uint32_t ostate;
uint32_t data[20];
const uint8_t core_go = 1;
uint32_t tmp_hash7;
uint32_t Htarg = ((const uint32_t *)ptarget)[7];
bool ret = false;
be32enc_vect(data, (const uint32_t *)pdata, 19);
if (e_start_group(dev) == E_ERR) {
applog(LOG_ERR, "Error: Could not start Epiphany cores.");
return false;
}
off_t offdata = offsetof(shared_buf_t, data);
off_t offostate = offsetof(shared_buf_t, ostate);
off_t offcorego = offsetof(shared_buf_t, go);
off_t offcoreend = offsetof(shared_buf_t, working);
off_t offcore;
#ifdef EPIPHANY_DEBUG
#define SCRATCHBUF_SIZE (131584)
uint32_t ostate2[8];
char *scratchbuf = malloc(SCRATCHBUF_SIZE);
uint32_t *ostatearm = calloc(rows*cols, sizeof(uint32_t));
#endif
i = 0;
while(1) {
offcore = i * sizeof(shared_buf_t);
if ((!core_working[i]) && (n < max_nonce)) {
*nonce = ++n;
data[19] = n;
core_working[i] = 1;
cores_working++;
core_nonce[i] = n;
#ifdef EPIPHANY_DEBUG
scrypt_1024_1_1_256_sp(data, scratchbuf, ostate2);
ostatearm[i] = ostate2[7];
#endif
e_write(emem, 0, 0, offcore + offdata, (void *) data, sizeof(data));
e_write(emem, 0, 0, offcore + offcoreend, (void *) &core_working[i], sizeof(core_working[i]));
e_write(emem, 0, 0, offcore + offcorego, (void *) &core_go, sizeof(core_go));
}
e_read(emem, 0, 0, offcore + offcoreend, (void *) &(core_working[i]), sizeof(core_working[i]));
if (!core_working[i]) {
e_read(emem, 0, 0, offcore + offostate, (void *) &(ostate), sizeof(ostate));
#ifdef EPIPHANY_DEBUG
applog(LOG_DEBUG, "CORE %u - EPI HASH %u - ARM HASH %u - %s", i, ostate, ostatearm[i], ((ostate==ostatearm[i])? "OK":"FAIL"));
#endif
tmp_hash7 = be32toh(ostate);
cores_working--;
if (unlikely(tmp_hash7 <= Htarg)) {
((uint32_t *)pdata)[19] = htobe32(core_nonce[i]);
*last_nonce = core_nonce[i];
#ifdef EPIPHANY_DEBUG
free(scratchbuf);
#endif
return true;
}
}
if (unlikely(((n >= max_nonce) && !cores_working) || thr->work_restart)) {
*last_nonce = n;
#ifdef EPIPHANY_DEBUG
free(scratchbuf);
#endif
return false;
}
i++;
i %= rows * cols;
}
#ifdef EPIPHANY_DEBUG
free(scratchbuf);
#endif
return false;
}
int main(int argc, char *argv[])
{
e_platform_t platform;
unsigned success;
unsigned board_rows, board_cols;
unsigned ROWS, COLS;
unsigned MaxGroupRows = 2;
unsigned MaxGroupCols = 2;
unsigned MODE;
unsigned row, col;
unsigned i, j, k, limit;
unsigned a[3000], b[3000], c[3000];
FILE *fo = stdout;
e_init(NULL);
e_get_platform_info(&platform);
board_rows = platform.rows;
board_cols = platform.cols;
MODE = atoi(argv[1]);
limit = atoi(argv[2]);
srand(time(NULL));
// Init input vectors
for (k=0; k<limit; k++)
{
a[k] = k;
b[k] = k;
c[k] = 0;
}
while (1)
{
ROWS = rand() % MaxGroupRows + 1;
COLS = rand() % MaxGroupCols + 1;
// Resource manager
success = e_reserve(MODE, ROWS, COLS, &row, &col);
if (success == E_OK)
{
// Open workgroup
fprintf(fo, "\tOpen workgroup...\n");
e_open(&dev, row, col, ROWS, COLS);
// Load the program
fprintf(fo, "\tLoad program onto workgroup...\n");
e_load_group("e_demo.srec", &dev, 0, 0, ROWS, COLS, E_FALSE);
for (i=0; i<ROWS; i++)
{
for (j=0; j<COLS; j++)
{
e_write(&dev, i, j, 0x1e00, &limit, sizeof(limit));
e_write(&dev, i, j, 0x2000, a, sizeof(unsigned)*limit);
e_write(&dev, i, j, 0x4000, b, sizeof(unsigned)*limit);
}
}
e_start_group(&dev);
usleep(300000);
// Release the resource
fprintf(fo, "\tRelease workgroup resource...\n");
e_release(ROWS, COLS, row, col);
fprintf(fo, "Group was successfully released.\n");
} else
fprintf(fo, "Failed to reserve workgroup!\n");
e_close(&dev);
}
e_finalize();
return 0;
}
