void internal_lessequal(long a, float b)
{
//printf("Inside internal add\n");
//b must be wrapped in a PyArrayObject
PyArrayObject *scalar = (PyArrayObject*)_malloc_align(sizeof(PyArrayObject),7);
scalar->blockData = (char**)_malloc_align(sizeof(char*),7);
scalar->blockData[0] = (char*)_malloc_align(sizeof(float),7);
((float*)(scalar->blockData[0]))[0] = b;
scalar->numberOfBlocks = 1;
//Setting the Operation object
Operation_t op1;
op1.shaderSize = arraylessequal_arrayscalar_size;
op1.EA_shader = arraylessequal_arrayscalar;
op1.obj[0] = a;
op1.obj[1] = scalar;
op1.obj[2] = a;
//printf("arraymultiply: %d\n",arraydivide);
//printf("&arraymultiply: %d\n",&arraydivide);
op1.num_SPES = speThreads;
unsigned int i = 0;
for(i = 0;i<speThreads;i++)
{
spe_pointer_addr[i][0] = &op1;
}
//printf("Adr of op1 sat. It is: %x\n",&op1);
unsigned int state = 1;
unsigned int y;
for ( y = 0 ; y < speThreads ; y++ )
{
//printf("Sending state to disp\n");
spe_in_mbox_write ( speData[y].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
//printf("TO THE DISP. THE SUM STATE: %d, has just been sent\n",&state);
//printf("This is the actual value: %d\n",state);
}
//printf("ADD Waiting for the SPE's\n");//ALSO, check that the structure used is correct!
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
unsigned int r;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
//printf("Something read on the inbox\n");
}
}
//printf("Done waiting on threads to finish in LESSEQUAL\n");
}
/*
* Run with three arrays, no upload
*/
void run30nu( unsigned int shaderindex )
{
unsigned int checked = 0;
unsigned int state = 300 + shaderindex, r;
unsigned int i;
unsigned int num_SPES = 1;
printf( "HERE2\n" );
for ( i = 0 ; i < num_SPES ; i++ )
{
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < num_SPES )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
}
/*
* Run with one array
*/
void run10( PyArrayObject *pyobj1, unsigned int shadersize, unsigned int *shader )
{
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
// Determining the number of SPEs
op1.num_SPES = _GetNumberOfSPES( pyobj1->numberOfBlocks );
Printf1( "Using %u SPEs\n", op1.num_SPES );
op1.obj[0] = pyobj1;
unsigned int checked = 0;
unsigned int state = 0, r;
unsigned int i;
for ( i = 0 ; i < op1.num_SPES ; i++ )
{
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < op1.num_SPES )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
}
int spe_get_status(int id)
{
if(spe_out_mbox_status(threads[id].speid) > 0)
{
spe_out_mbox_read(threads[id].speid, &(threads[id].status), 1);
}
return threads[id].status;
}
void *mailbox_pthread_function(void *thread_arg) {
spe_context_ptr_t ctx = * (spe_context_ptr_t*) thread_arg;
unsigned int mbox_data = 0;
int i=0; for (i=0;i<10;i++) { // citim zece mesaje de la SPE
/*TODO(2): cititi prin mailbox-ul out_intr etapa terminata de SPE-ul cu contextul ctx*/
while(spe_out_mbox_status(ctx) == 0);
spe_out_mbox_read(ctx, &mbox_data, 1);
printf("[PPU]: SPU %p a terminat etapa %d\n",ctx,mbox_data);
}
return NULL;
}
int SPE_WaitForMsg(_THIS, spu_data_t * spe_data, unsigned int msg)
{
deprintf(2, "[PS3->SPU] Waiting for message from %s\n", spe_data->program_name);
unsigned int out_messages[1];
while (!spe_out_mbox_status(spe_data->ctx));
int mbox_read = spe_out_mbox_read(spe_data->ctx, out_messages, 1);
deprintf(2, "[PS3->SPU] Got message from %s, message was %u\n", spe_data->program_name, out_messages[0]);
if (out_messages[0] == msg)
return 0;
else
return -1;
}
void internal_div(long a, long b)
{
//printf("Inside internal add\n");
//Setting the Operation object
Operation_t op1;
op1.shaderSize = arraydivide_size;
op1.EA_shader = arraydivide;
op1.obj[0] = a;
op1.obj[1] = b;
op1.obj[2] = a;
//printf("arraymultiply: %d\n",arraydivide);
//printf("&arraymultiply: %d\n",&arraydivide);
op1.num_SPES = speThreads;
unsigned int i = 0;
for(i = 0;i<speThreads;i++)
{
spe_pointer_addr[i][0] = &op1;
}
//printf("Adr of op1 sat. It is: %x\n",&op1);
unsigned int state = 1;
unsigned int y;
for ( y = 0 ; y < speThreads ; y++ )
{
//printf("Sending state to disp\n");
spe_in_mbox_write ( speData[y].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
//printf("TO THE DISP. THE SUM STATE: %d, has just been sent\n",&state);
//printf("This is the actual value: %d\n",state);
}
//printf("ADD Waiting for the SPE's\n");//ALSO, check that the structure used is correct!
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
unsigned int r;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
//printf("Something read on the inbox\n");
}
}
//printf("Done waiting on threads to finish in DIV\n");
}
/*
* Run with one array, with return value
*/
float run10r( PyArrayObject *pyobj1, unsigned int shadersize, unsigned int *shader )
{
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
// Determining the number of SPEs
op1.num_SPES = _GetNumberOfSPES( pyobj1->numberOfBlocks );
op1.obj[0] = pyobj1;
//float *results[6];
unsigned int checked = 0;
unsigned int state = 100, r;
unsigned int i;
for ( i = 0 ; i < op1.num_SPES ; i++ )
{
//results[i] = (float *)memalign( 128, ( 4 + 127 ) & ~127 );
r = (unsigned int)resultFloats[i];
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
spe_in_mbox_write ( speData[i].spe_ctx, &r, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < op1.num_SPES )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
float r1 = 0.0f;
for ( i = 0 ; i < op1.num_SPES ; i++ )
{
r1 += resultFloats[i][0];
// Cleanup
// free( results[i] );
}
return r1;
}
double SNRM2( PyArrayObject *pyobj1, unsigned int shadersize, unsigned int *shader )
{
Tic();
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
op1.obj[0] = pyobj1;
op1.num_SPES = speThreads;
float *results[6];
unsigned int state = 2, r;
unsigned int i;
//printf( "Sending states to SPEs\n" );
for ( i = 0 ; i < speThreads ; i++ )
{
results[i] = (float *)memalign( 128, ( 4 + 127 ) & ~127 );
r = (unsigned int)results[i];
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
spe_in_mbox_write ( speData[i].spe_ctx, &r, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
float r1 = 0.0f;
for ( i = 0 ; i < speThreads ; i++ )
{
r1 += results[i][0];
// Cleanup
free( results[i] );
}
r1 = sqrtf( r1 );
double time = Toc_d();
PrintTicToc( "Finished at ", Toc() );
printf( "Result=%f\n", r1 );
return time;
}
double SSCAL( PyArrayObject *pyobj1, PyArrayObject *pyscalar1, unsigned int shadersize, unsigned int *shader )
{
Tic();
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
op1.obj[0] = pyobj1;
op1.scalar[0] = pyscalar1;
op1.num_SPES = speThreads;
unsigned int state = 1, r;
unsigned int i;
//printf( "Sending states to SPEs\n" );
for ( i = 0 ; i < speThreads ; i++ )
{
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
double time = Toc_d();
PrintTicToc( "Finished at ", Toc() );
for( i = 0 ; i < 3 ; i++ )
{
printf( "%u=%f\n", i, ((float*)pyobj1->blockData[0])[i] );
}
return time;
}
/*
* Run with three arrays, no run
*/
void run30nr( PyArrayObject *pyobj1, PyArrayObject *pyobj2, PyArrayObject *pyobj3, unsigned int shadersize, unsigned int *shader, unsigned int shaderindex )
{
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
// Determining the number of SPEs
//op1.num_SPES = _GetNumberOfSPES( pyobj1->numberOfBlocks );
op1.num_SPES = 1;
op1.obj[0] = pyobj1;
op1.obj[1] = pyobj2;
op1.obj[2] = pyobj3;
unsigned int checked = 0;
unsigned int state = 200 + shaderindex, r;
unsigned int i;
// printf( "HERE %u!\n", state );
// printf( "SIZE=%u, ADDR=%#x\n", op1.shaderSize, op1.EA_shader );
for ( i = 0 ; i < op1.num_SPES ; i++ )
{
//printf( "-->%u\n",i );
spe_pointer_addr[i][shaderindex] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < op1.num_SPES )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
}
/*
* Run with two arrays and two scalars
*/
void run22( PyArrayObject *A, PyArrayObject *x, PyArrayObject *alpha, PyArrayObject *j, unsigned int shadersize, unsigned int *shader )
{
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
// Determining the number of SPEs
op1.num_SPES = _GetNumberOfSPES( A->numberOfBlocks );
op1.obj[0] = A;
op1.obj[1] = x;
op1.scalar[0] = alpha;
op1.scalar[1] = j;
unsigned int checked = 0;
unsigned int state = 0, r;
unsigned int i;
for ( i = 0 ; i < op1.num_SPES ; i++ )
{
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < op1.num_SPES )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
}
unsigned int internal_sum(long a)
{
//printf("Inside internal add\n");
//Setting the Operation object
Operation_t op1;
op1.shaderSize = arraysum_size;
op1.EA_shader = arraysum;
op1.obj[0] = a;
//printf("arraymultiply: %d\n",arraydivide);
//printf("&arraymultiply: %d\n",&arraydivide);
op1.num_SPES = speThreads;
unsigned int i = 0;
for(i = 0;i<speThreads;i++)
{
spe_pointer_addr[i][0] = &op1;
}
//printf("Adr of op1 sat. It is: %x\n",&op1);
unsigned int state = 2;
unsigned int y;
for ( y = 0 ; y < speThreads ; y++ )
{
//printf("Sending state to disp\n");
spe_in_mbox_write ( speData[y].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
//printf("TO THE DISP. THE SUM STATE: %d, has just been sent\n",&state);
//printf("This is the actual value: %d\n",state);
}
//printf("ADD Waiting for the SPE's\n");//ALSO, check that the structure used is correct!
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
unsigned int signal;
//float r;
//printf("1. r[0]: %f\n",resultFloats[0][0]);
//printf("1. adr is: %x\n",&(resultFloats[0][0]));
unsigned int pointersToResultFloats[speThreads];
unsigned int e;
for(e=0;e<speThreads;e++)
{
pointersToResultFloats[e] = &(resultFloats[e][0]);
}
for ( y = 0 ; y < speThreads ; y++ )
{
//printf("Sending res pointers to disp\n");
spe_in_mbox_write ( speData[y].spe_ctx, &pointersToResultFloats[y], 1, SPE_MBOX_ALL_BLOCKING );
//printf("TO THE DISP. THE SUM STATE: %d, has just been sent\n",&state);
//printf("This is the actual value: %d\n",state);
}
float finalRes = 0;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &signal, 1 );
checked++;
//printf("Something read on the inbox\n");
}
}
//printf("Done waiting on threads to finish in DIV\n");
//printf("2. r[0]: %f\n",resultFloats[y][0]);
//Add all the results together
unsigned int u = 0;
for(u=0;u<speThreads;u++)
{
finalRes = finalRes+resultFloats[u][0];
}
return finalRes;
}
void internal_random(PyArrayObject *a)
{
//printf("adr of PyArrayObject: %x,%d\n",result,result);
//Setting the PyArrayObject
//result->numberOfBlocks = _numberOfBlocks*_numberOfBlocks;
//result->numberOfBlocksXDim = _numberOfBlocks;
//result->numberOfBlocksYDim = _numberOfBlocks;
//result->blockSize = _blockSize;//in elements
//result->nd = 1;
//result->blockData = (char**)_malloc_align(sizeof(char*)*_numberOfBlocks*_numberOfBlocks,7);
//unsigned int i = 0;
//for(i = 0;i<_numberOfBlocks*_numberOfBlocks;i++)
// {
// result->blockData[i] = (char*)_malloc_align(sizeof(char)*_blockSize*_blockSize*4,7);
// }
//printf("Adr of blockData: %x\n",result->blockData);
//printf("Adr of blockData: %d\n",result->blockData);
//Setting the Operation object
Operation_t op1;
op1.shaderSize = randomfiller_size;
op1.EA_shader = randomfiller;
op1.obj[0] = a;
//printf("randomfiller: %d\n",randomfiller);
//printf("&randomfiller: %d\n",&randomfiller);
op1.num_SPES = speThreads;
unsigned int i = 0;
for(i = 0;i<speThreads;i++)
{
spe_pointer_addr[i][0] = &op1;
}
//printf("Adr of op1 sat. It is: %x\n",&op1);
unsigned int state = 1;
unsigned int y;
for ( y = 0 ; y < speThreads ; y++ )
{
spe_in_mbox_write ( speData[y].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
//printf("TO THE DISP. THE SUM STATE: %d, has just been sent\n",&state);
//printf("This is the actual value: %d\n",state);
}
//printf("SUM Waiting for the SPE's\n");//ALSO, check that the structure used is correct!
// Waiting for SPEs!
//printf( "Waiting for SPEs\n" );
unsigned int checked = 0;
unsigned int r;
//printf("checked is: %d\n",checked);
//printf("speThreads is: %d\n",speThreads);
while( checked < speThreads )
{
//printf("checked is: %d\n",checked);
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
//printf("Something read on the inbox\n");
}
}
//printf("Done waiting on threads to finish\n");
}
double SDOT2( PyArrayObject *pyobj1, PyArrayObject *pyobj2, unsigned int shadersize, unsigned int *shader )
{
Operation_t op1;
op1.shaderSize = shadersize;
op1.EA_shader = shader;
op1.obj[0] = pyobj1;
op1.obj[1] = pyobj2;
op1.num_SPES = speThreads;
unsigned int state = 4;
unsigned int i, r;
unsigned int checked = 0;
float *results[6];
// Setup SPEs
for ( i = 0 ; i < speThreads ; i++ )
{
results[i] = (float *)memalign( 128, ( 4 + 127 ) & ~127 );
r = (unsigned int)results[i];
spe_pointer_addr[i][0] = (unsigned int)&op1;
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
spe_in_mbox_write ( speData[i].spe_ctx, &r, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
Tic();
// Make SPEs run
state = 6;
for ( i = 0 ; i < speThreads ; i++ )
{
spe_in_mbox_write ( speData[i].spe_ctx, &state, 1, SPE_MBOX_ALL_BLOCKING );
}
// Waiting for SPEs!
checked = 0;
while( checked < speThreads )
{
if ( spe_out_mbox_status( speData[checked].spe_ctx ) )
{
spe_out_mbox_read( speData[checked].spe_ctx, &r, 1 );
checked++;
}
}
// Get result
float r1 = 0.0f;
for ( i = 0 ; i < speThreads ; i++ )
{
r1 += results[i][0];
// Cleanup
free( results[i] );
}
double time = Toc_d();
PrintTicToc( "Finished at ", Toc() );
printf( "Result=%f\n", r1 );
return time;
}
