/** =========================================================================
* Soft reset
*
**/
void chip_soft_reset(void)
{
unsigned reg_value;
read_sswitch_reg(get_core_id(), 6, ®_value);
write_sswitch_reg(0, 6, reg_value);
write_sswitch_reg(get_core_id(), 6, reg_value);
}
/////////////////////////////////////////////////////////
// main testing function
/////////////////////////////////////////////////////////
int main(int argc, const char * const argv[])
{
(void)argc;
(void)argv;
boolean_T pass;
int coreid, k;
float filt[200];
float tmp[2];
/////////////////////////////////////////////////////////
// main test loop
// each core loops over a kernel instance
/////////////////////////////////////////////////////////
coreid = get_core_id();
printf("starting %d kernel iterations... (coreid = %d)\n",KERNEL_ITS,coreid);
if (coreid>3)
coreid=coreid-4;
synch_barrier();
perf_begin();
for(k = 0; k < getKernelIts(); k++)
{
// matlab kernel
mlButter(fv1, *(float (*)[200])&fv0[200 * coreid], filt);
}
synch_barrier();
perf_end();
/////////////////////////////////////////////////////////
// check results
/////////////////////////////////////////////////////////
synch_barrier();
tmp[0] = sum(filt);
tmp[1] = var(filt);
pass = checkRes(tmp, *(float (*)[4])&fv2[coreid << 2]);
flagPassFail(pass, get_core_id());
/////////////////////////////////////////////////////////
// synchronize and exit
/////////////////////////////////////////////////////////
return !pass;
}
/**
* Prepare one shrink image IPC message.
*/
void register_shrink_on_core(const uint8_T *Img, const uint32_T SubArea[4], const uint32_t yStart, const uint32_t yEnd,
const uint32_t yEvenOdd, const uint32_t yHeight, const uint32_t xWidth, const uint32_t xWidthSmall, uint8_T *ImgSmall, const uint32_T CoreNo)
{
#ifdef _NO_IPC_TEST_
shrinkImage_on_core(Img, SubArea, yStart, yEnd, yEvenOdd, yHeight, xWidth, xWidthSmall, ImgSmall);
#else
//To see which entries have to be placed to which parameter look at the union declarations in struct processing_info.
//Very ugly, I know. But this software is only a scientific proof of concept and it's near its end, hence this kind of code can be written now during the last days ...
process_message_t * p_msg = 0;
p_msg = p_queue_msg[CoreNo];
p_msg->core_id = get_core_id(CoreNo);
p_msg->info.processing_type = pt_shrink;
memcpy(p_msg->info.Shr_SubArea, SubArea, sizeof(uint32_T) * 4);
p_msg->info.Shr_Img=Img;
p_msg->info.Shr_ImgSmall=ImgSmall;
p_msg->info.Shr_yStart=yStart;
p_msg->info.Shr_yEnd=yEnd;
p_msg->info.Shr_yEvenOdd=yEvenOdd;
p_msg->info.Shr_yHeight=yHeight;
p_msg->info.Shr_xWidth=xWidth;
p_msg->info.Shr_xWidthSmall=xWidthSmall;
#endif
}
/**
* tries to determine the physical package, a cpu belongs to
*/
int get_pkg(int cpu)
{
int pkg=-1;
char buffer[10];
if (cpu == -1) { cpu = get_cpu(); }
if (cpu != -1)
{
sprintf(path, "/sys/devices/system/cpu/cpu%i/topology/physical_package_id", cpu);
if( read_file(path, buffer, sizeof(buffer)) ) pkg = atoi(buffer);
/* fallbacks if sysfs is not working */
if (pkg == -1)
{
/* assume 0 if there is only one CPU or only one package */
if ((num_cpus() == 1) || (num_packages() == 1)) { pkg = 0; }
/* get the physical package id from /proc/cpuinfo */
else if(!get_proc_cpuinfo_data("physical id", buffer, cpu)) { pkg = atoi(buffer); }
/* if the number of cpus equals the number of packages assume pkg_id = cpu_id*/
else if (num_cpus() == num_packages()) { pkg = cpu; }
/* if there is only one core per package assume pkg_id = core_id */
else if (num_cores_per_package() == 1) { pkg = get_core_id(cpu); }
/* if the number of packages equals the number of numa nodes assume pkg_id = numa node */
else if (num_numa_nodes() == num_packages()) { pkg = get_numa_node(cpu); }
/* NOTE pkg_id in UMA Systems with multiple sockets and more than 1 Core per socket can't be determined
without correct topology information in sysfs*/
}
}
return pkg;
}
void mp_barrier(cycles_t *measurement)
{
coreid_t tid = get_core_id();
#ifdef QRM_DBG_ENABLED
++_num_barrier;
uint32_t _num_barrier_recv = _num_barrier;
#endif
debug_printfff(DBG__REDUCE, "barrier enter #%d\n", _num_barrier);
// Recution
// --------------------------------------------------
#ifdef QRM_DBG_ENABLED
uint32_t _tmp =
#endif
mp_reduce(_num_barrier);
#ifdef QRM_DBG_ENABLED
// Sanity check
if (tid==get_sequentializer()) {
assert (_tmp == get_num_threads()*_num_barrier);
}
if (measurement)
*measurement = bench_tsc();
#endif
// Broadcast
// --------------------------------------------------
if (tid == get_sequentializer()) {
mp_send_ab(_num_barrier);
} else {
#ifdef QRM_DBG_ENABLED
_num_barrier_recv =
#endif
mp_receive_forward(0);
}
#ifdef QRM_DBG_ENABLED
if (_num_barrier_recv != _num_barrier) {
debug_printf("ASSERTION fail %d != %d\n", _num_barrier_recv, _num_barrier);
}
assert (_num_barrier_recv == _num_barrier);
// Add a shared memory barrier to absolutely make sure that
// everybody finished the barrier before leaving - this simplifies
// debugging, as the programm will get stuck if barriers are
// broken, rather than some threads (wrongly) continuing and
// causing problems somewhere else
#if 0 // Enable separately
debug_printfff(DBG_REDUCE, "finished barrier .. waiting for others\n");
shl_barrier_shm(get_num_threads());
#endif
#endif
debug_printfff(DBG__REDUCE, "barrier complete #%d\n", _num_barrier);
}
int main()
{
if(get_core_id() == 0) {
run_suite(testcases);
}
return 0;
}
void __init smp_init_cpus(void)
{
unsigned i;
unsigned int ncpus = get_core_count();
unsigned int core_id = get_core_id();
pr_info("%s: Core Count = %d\n", __func__, ncpus);
pr_info("%s: Core Id = %d\n", __func__, core_id);
for (i = 0; i < ncpus; ++i)
set_cpu_possible(i, true);
}
int main()
{
int coreid, i, error = 0;
coreid = get_core_id();
// set start value of jrand function
next = 1;
if (coreid == 0)
{
int f=0;
initialize_aes();
// 1 iterations of enc+dec
for (f=0;f<1;f++){
compute_aes();
//check output
for (i = 0; i < 16; i++){
if (encoutbuf[i] != check_encoutbuf[i]) {
error+=1;
/* printf("Error occured in encryption\n",0,0,0,0); */
//printf("encrypted: %d, expected: %d\n",encoutbuf[i],check_encoutbuf[i],0,0);
}
if (decoutbuf[i] != check_decoutbuf[i]) {
error+=1;
/* printf("Error occured in decryption\n",0,0,0,0); */
//printf("decrypted: %d, expected: %d\n",decoutbuf[i],check_decoutbuf[i],0,0);
}
}
}
int *DEFAULT_RESULT;
if (error == 0) {
//printf ("OOOOOOK!!!!!!\n",0,0,0,0);
DEFAULT_RESULT = (int*)0x10003ffc;
*(DEFAULT_RESULT) = 1;
}
else {
//printf ("Not OK!! %d\n",error,0,0,0);
DEFAULT_RESULT = (int*)0x10003ffc;
*(DEFAULT_RESULT) = error;
}
}
synch_barrier();
eoc(0);
}
void __init smp_init_cpus(void)
{
unsigned i;
unsigned int ncpus = get_core_count();
unsigned int core_id = get_core_id();
pr_info("%s: Core Count = %d\n", __func__, ncpus);
pr_info("%s: Core Id = %d\n", __func__, core_id);
if (ncpus > NR_CPUS) {
ncpus = NR_CPUS;
pr_info("%s: limiting core count by %d\n", __func__, ncpus);
}
for (i = 0; i < ncpus; ++i)
set_cpu_possible(i, true);
}
/**
* Brief: Adds task to a chosen core and if the core isn't running,
* resets it's thread to start the execution.
* Param: The task to be executed.
*/
std::size_t add_task(TASK task)
{
std::size_t core_id = get_core_id(); // Get a suitable core.
std::size_t task_id = get_task_id(); // Find the smallest id.
task_pack<TASK> tmp_pack;
tmp_pack.id = task_id;
tmp_pack.task = std::move(task);
set_result(task_id, T(), false); // Create a record for this task.
cores_[core_id].add_task(std::move(tmp_pack));
if(!cores_[core_id].running)
{ // Re-start the core thread if it's not running.
cores_[core_id].running = true;
core_threads_[core_id].reset(
new std::thread(&Scheduler<T, TASK>::run_core, this, core_id)
);
core_threads_[core_id]->detach();
}
return task_id;
}
std::size_t add_task(TASK task)
{
std::size_t core_id = get_core_id(); // Get a suitable core.
std::size_t task_id = get_task_id(); // Find the smallest id.
task_pack<TASK> tmp_pack;
tmp_pack.id = task_id;
tmp_pack.task = std::move(task);
set_result(task_id, 0, false); // Create a record for this task.
cores_[core_id].add_task(std::move(tmp_pack));
if(!cores_[core_id].running)
{ // If the core is available, execute the task immedietly.
// The mutex will be unlocked at the end of the run_core method.
cores_[core_id].running = true;
core_threads_[core_id].reset(
new std::thread(&Scheduler<void, TASK>::run_core, this, core_id)
);
core_threads_[core_id]->detach();
}
return task_id;
}
/**
* Cachec invalidate zeroized memory on all cores (happens between the calculations / after a calculation result and at boot time)
*/
void cacheinval_on_core(const uint8_T *TBuf, const uint32_T TSize, const uint32_T number_of_cores)
{
#ifndef _NO_IPC_TEST_
int32_t i=0;
//To see which entries have to be placed to which parameter look at the union declarations in struct processing_info.
//Very ugly, I know. But this software is only a scientific proof of concept and it's near its end, hence this kind of code can be written now during the last days ...
process_message_t * p_msg = 0;
for (i = CORE_AMOUNT-1; i >= (int)(CORE_AMOUNT-number_of_cores); i-- )
{
p_msg = p_queue_msg[i];
p_msg->core_id = get_core_id(i);
p_msg->info.processing_type = pt_cacheinval;
p_msg->info.Tvec = TBuf;
p_msg->info.Tsize = TSize;
}
send_to_cores(pt_cacheinval, number_of_cores, NULL, NULL, NULL);
#endif
}
/**
* Prepare one ssd() and/or jacobian() IPC message.
*/
void prepare_ipc_message(const processing_type_e ProcessingType, const real32_T w[3], const uint32_T BoundBox[4], const uint32_T MarginAddon[3], const real32_T
DSPRange[4], const emxArray_uint8_T *Tvec, const uint32_T TOffset, const emxArray_uint8_T *Rvec, const uint32_T ROffset,
const uint32_T d, const uint32_T CoreNo, const uint32_T i_from, const uint32_T i_to)
{
process_message_t * p_msg = 0;
p_msg = p_queue_msg[CoreNo];
p_msg->core_id = get_core_id(CoreNo);
p_msg->info.processing_type = ProcessingType;
memcpy(p_msg->info.w, w, sizeof(real32_T) * 3);
memcpy(p_msg->info.BoundBox, BoundBox, sizeof(uint32_T) * 4);
memcpy(p_msg->info.MarginAddon, MarginAddon, sizeof(uint32_T) * 3);
memcpy(p_msg->info.DSPRange, DSPRange, sizeof(real32_T) * 4);
p_msg->info.Tvec = &Tvec->data[0];
p_msg->info.Tsize = Tvec->allocatedSize;
p_msg->info.TOffset = TOffset;
p_msg->info.Rvec = &Rvec->data[0];
p_msg->info.Rsize = Rvec->allocatedSize;
p_msg->info.ROffset = ROffset;
p_msg->info.d = d;
p_msg->info.i_from = i_from;
p_msg->info.i_to = i_to;
p_msg->info.NewImageDataArrived = g_NewImageDataArrived;
}
/////////////////////////////////////////////////////////
// main testing function
/////////////////////////////////////////////////////////
int main(int argc, const char * const argv[])
{
(void)argc;
(void)argv;
int coreid;
int it;
int k;
boolean_T pass, flag;
float y[100];
int ix;
float b_y;
float xbar;
float r;
float c_y;
float tmp[2];
float golden[4];
/////////////////////////////////////////////////////////
// main test loop
// each core loops over a kernel instance
/////////////////////////////////////////////////////////
coreid = get_core_id();
printf("starting %d kernel iterations... (coreid = %d)\n",KERNEL_ITS,coreid);
if (coreid>3)
coreid=coreid-4;
synch_barrier();
perf_begin();
for(it = 0; it < KERNEL_ITS; it++)
{
// matlab kernel
for (ix = 0; ix < 100; ix++) {
y[ix] = (real32_T)fLog(fv0[ix + 100 * coreid]);
}
}
synch_barrier();
perf_end();
synch_barrier();
/////////////////////////////////////////////////////////
// check results
/////////////////////////////////////////////////////////
pass = true;
b_y = y[0];
ix = 0;
xbar = y[0];
for (k = 0; k < 99; k++) {
b_y += y[k + 1];
ix++;
xbar += y[ix];
}
xbar *= 1.0F/100.0F;
ix = 0;
r = y[0] - xbar;
c_y = r * r;
for (k = 0; k < 99; k++) {
ix++;
r = y[ix] - xbar;
c_y += r * r;
}
c_y *= 1.0F/99.0F;
tmp[0] = b_y;
tmp[1] = c_y;
pass = true;
for (ix = 0; ix < 2; ix++) {
for (k = 0; k < 2; k++) {
golden[k + (ix << 1)] = fv1[(k + (ix << 1)) + (coreid << 2)];
}
flag = true;
flag = flag && (tmp[ix] <= golden[ix << 1]);
flag = flag && (tmp[ix] >= golden[1 + (ix << 1)]);
printErrors(!flag, ix, tmp[ix] ,golden[(ix << 1)] ,golden[1 + (ix << 1)]);
pass = pass && flag;
}
flagPassFail(pass, get_core_id());
synch_barrier();
/////////////////////////////////////////////////////////
// synchronize and exit
/////////////////////////////////////////////////////////
return !pass;
}
#include "memTester.h"
sl_def(memTester,, sl_shparm(sl_place_t, syscall_gateway)) {
sl_place_t syscall_gateway = sl_getp(syscall_gateway);
syscall_target(&syscall_gateway);
sl_index(i);
unsigned pid = get_current_place();
unsigned core_id = get_core_id();
output_string("MemTester (thread ", 2);
output_uint((unsigned int)i, 2);
output_string(") now running on core ", 2);
output_uint(core_id, 2);
output_string(", place_id ", 2);
output_hex(pid, 2);
output_char('\n', 2);
output_char('\n', 2);
output_char('\n', 2);
run(0);
//Let the compiler know we care
sl_setp(syscall_gateway, sl_getp(syscall_gateway));
}
sl_enddef
// Space reserved for testing small pages (4KiB):
// 0x440000 - 0x550000 inclusive
// Smallest page = 4KB = 0x440000 : 0x440FFF
//
/////////////////////////////////////////////////////////
// main testing function
/////////////////////////////////////////////////////////
int main(int argc, const char * const argv[])
{
(void)argc;
(void)argv;
boolean_T pass, flag;
int coreid;
float omega, ampl, runningPhase;
float sig[200];
int k;
int i;
float y;
int ix;
float xbar;
float r;
float b_y;
float tmp[2];
float golden[4];
boolean_T c_y;
boolean_T exitg1;
/////////////////////////////////////////////////////////
// main test loop
// each core loops over a kernel instance
/////////////////////////////////////////////////////////
coreid = get_core_id();
printf("starting %d kernel iterations... (coreid = %d)\n",KERNEL_ITS,coreid);
if (coreid>3)
coreid=coreid-4;
synch_barrier();
perf_begin();
omega = fv1[coreid];
ampl = fv2[coreid];
for(k = 0; k < getKernelIts(); k++)
{
runningPhase = omega;
// matlab kernel
for (i = 0; i < 200; i++) {
sig[i] = ampl * fSin(runningPhase);
runningPhase += omega;
if(runningPhase > pi2[0])
{
runningPhase -= pi2[0];
}
}
}
synch_barrier();
perf_end();
/////////////////////////////////////////////////////////
// check results
/////////////////////////////////////////////////////////
synch_barrier();
y = sig[0];
ix = 0;
xbar = sig[0];
for (k = 0; k < 199; k++) {
y += sig[k + 1];
ix++;
xbar += sig[ix];
}
xbar = fDiv(xbar,200.0F);
ix = 0;
r = sig[0] - xbar;
b_y = r * r;
for (k = 0; k < 199; k++) {
ix++;
r = sig[ix] - xbar;
b_y += r * r;
}
b_y = fDiv(b_y,199.0F);
tmp[0] = y;
tmp[1] = b_y;
pass = true;
for (k = 0; k < 2; k++) {
for (ix = 0; ix < 2; ix++) {
golden[ix + (k << 1)] = fv0[(ix + (k << 1)) + (coreid << 2)];
}
flag = true;
flag = flag && (tmp[k] <= golden[k << 1]);
flag = flag && (tmp[k] >= golden[1 + (k << 1)]);
printErrors(!flag, k, tmp[k] ,golden[k << 1], golden[1 + (k << 1)]);
pass = pass && flag;
}
flagPassFail(pass, get_core_id());
/////////////////////////////////////////////////////////
// synchronize and exit
/////////////////////////////////////////////////////////
return !pass;
}
int main() {
/* Variable Definition */
int coreid;
int i,j;
int start_frame;
int index;
int time;
coreid = get_core_id();
if (coreid == 0) {
// initialization
NB_BLOB = 0 ;
start_frame = 0;
for(i=0;i<NFRAME;i++){ // for each frame
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
//%%%%%%%%%%%%%%% * DATA TRANSFER FROM L2 TO L1 * %%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
index = 0;
N_pixel = N_pixelL2[i];
reset_timer();
start_timer();
for(j=start_frame; j<start_frame + N_pixel*2; j++){
pixel[index++]=pixelL2[j];
}
stop_timer();
printf("FRAME: %d (%d-%d) Transfer Time: %d\n",i,start_frame,j,get_time());
start_frame = j;
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
//%%%%%%%%%%%%%%% * PROCESSING * %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
init_data();
reset_timer();
start_timer();
blob_formation();
stop_timer();
printf("Blob Formation Time: %d\n",get_time(),0,0,0);
reset_timer();
start_timer();
prevBlob_filter();
stop_timer();
printf("Filtering prev Blob List Time: %d\n",get_time(),0,0,0);
reset_timer();
start_timer();
newBlob_filter();
stop_timer();
printf("Filtering new Blob List Time: %d\n",get_time(),0,0,0);
reset_timer();
start_timer();
blob_merge();
stop_timer();
printf("Blob Merging Time: %d\n",get_time(),0,0,0);
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
//%%%%%%%%%%%%%%% * CHECKSUM * %%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
printf("FRAME = %d\n",i,0,0,0);
printf("NB_BLOB = %d\n",NB_BLOB,0,0,0);
for(j=0;j<NB_BLOB;j++){
printf("Blob %d: centroid = (%d,%d), weight = %d, ",j,BLOB_LIST[j].xc,BLOB_LIST[j].yc,BLOB_LIST[j].W);
printf("xmax = %d, xmin = %d, ymax = %d, ymin = %d\n",BLOB_LIST[j].xmax, BLOB_LIST[j].xmin, BLOB_LIST[j].ymax, BLOB_LIST[j].ymin);
if(BLOB_LIST[j].xc == results[(i*B_MAX+j)*6 ]) printf("OK xc!\t",0,0,0,0); else printf("FAIL xc!\t",0,0,0,0);
if(BLOB_LIST[j].yc == results[(i*B_MAX+j)*6+1 ]) printf("OK yc!\t",0,0,0,0); else printf("FAIL yc!\t",0,0,0,0);
if(BLOB_LIST[j].xmax == results[(i*B_MAX+j)*6+2 ]) printf("OK xmax!\t",0,0,0,0); else printf("FAIL xmax!\t",0,0,0,0);
if(BLOB_LIST[j].xmin == results[(i*B_MAX+j)*6+3 ]) printf("OK xmin!\t",0,0,0,0); else printf("FAIL xmin!\t",0,0,0,0);
if(BLOB_LIST[j].ymax == results[(i*B_MAX+j)*6+4 ]) printf("OK ymax!\t",0,0,0,0); else printf("FAIL ymax!\t",0,0,0,0);
if(BLOB_LIST[j].ymin == results[(i*B_MAX+j)*6+5 ]) printf("OK ymin!\n",0,0,0,0); else printf("FAIL ymin!\n",0,0,0,0);
}
printf("\n\n",0,0,0,0);
//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%//
}
eoc(0);
}
}
void init_rng(unsigned *p) {
for (unsigned i = 0; i < get_core_id(); i++) rand_r(p);
}
int mm_ctx_dup(mm_context *dst_ctx, mm_context *src_ctx)
{
/* Copy the src ctx to the dst creating mappings as we go
*
* Return -1 if there is a problem
*/
struct mm_mapping *pmap;
int err;
kdebug("MM_CTX_DUP core = %d\n", get_core_id());
mm_context_dump(src_ctx);
mm_context_dump(dst_ctx);
rwlock_wrlock(&src_ctx->lock);
pmap = first_mapping(src_ctx);
while (pmap)
{
struct mm_object *new_obj1, *new_obj2;
struct mm_mapping *new_map;
if (!(pmap->prot & PROT_WRITE))
{
/* This map is read-only - share the existing object */
new_obj1 = pmap->object;
++new_obj1->refcnt;
/* Create a new mapping */
if ((new_map = mapping_new(pmap->start_addr, pmap->length, pmap->prot,
pmap->fd, new_obj1)) == NULL)
{
rwlock_unlock(&src_ctx->lock);
return -1;
}
}
else
{
/* This map is writable - make a new object */
/* Create an mm_object */
/* TODO: COW */
/*
* Create 2 new objects
* Point them at the existing obj
* Create new map
* Point each map at an obj
*/
if ((new_obj1 = object_new()) == NULL)
{
rwlock_unlock(&src_ctx->lock);
return -1;
}
if ((new_obj2 = object_new()) == NULL)
{
rwlock_unlock(&src_ctx->lock);
return -1;
}
new_obj1->chain = pmap->object;
new_obj1->share_type = share_private;
new_obj2->chain = pmap->object;
new_obj2->share_type = share_private;
//pmap->object->share_type = cow;
/* Point the original map at new_obj1 */
pmap->object = new_obj1;
/* Create a new mapping to point at new_obj2 */
if ((new_map = mapping_new(pmap->start_addr, pmap->length, pmap->prot,
pmap->fd, new_obj2)) == NULL)
{
rwlock_unlock(&src_ctx->lock);
return -1;
}
mm_change_commit(src_ctx, pmap, PM_USER);
}
dst_ctx->mapping_list.add_tail(new_map);
mm_check_uncommit(dst_ctx, new_map);
/* Go to the next parent mapping */
pmap = next_mapping(src_ctx, pmap);
}
cpu_tlb_flush_global();
rwlock_unlock(&src_ctx->lock);
// kdebug("MM_CTX_DUP done\n");
// mm_context_dump(src_ctx);
// mm_context_dump(dst_ctx);
// pagestruct_audit();
return 0;
}
/////////////////////////////////////////////////////////
// main testing function
/////////////////////////////////////////////////////////
int main(int argc, const char * const argv[])
{
(void)argc;
(void)argv;
int coreid, k;
boolean_T pass;
float V[25];
float s[5];
float U[25];
int b_k;
float y[25];
float b_y;
float c_y;
float d_y;
float tmp[3];
init_fp_regs();
/////////////////////////////////////////////////////////
// main test loop
// each core loops over a kernel instance
/////////////////////////////////////////////////////////
coreid = get_core_id();
printf("starting %d kernel iterations... (coreid = %d)\n",KERNEL_ITS,coreid);
if (coreid>3)
coreid=coreid-4;
synch_barrier();
perf_begin();
for(k = 0; k < getKernelIts(); k++)
{
// call matlab kernel
eml_xgesvd(*(float (*)[25])&fv0[25 * coreid], U, s, V);
}
synch_barrier();
perf_end();
/////////////////////////////////////////////////////////
// check results
/////////////////////////////////////////////////////////
synch_barrier();
for (b_k = 0; b_k < 25; b_k++) {
y[b_k] = fAbs(U[b_k]);
}
b_y = y[0];
c_y = s[0];
for (b_k = 0; b_k < 4; b_k++) {
c_y += s[b_k + 1];
}
for (b_k = 0; b_k < 24; b_k++) {
b_y += y[b_k + 1];
}
for (b_k = 0; b_k < 25; b_k++) {
y[b_k] = fAbs(V[b_k]);
}
d_y = y[0];
for (b_k = 0; b_k < 24; b_k++) {
d_y += y[b_k + 1];
}
tmp[0] = b_y;
tmp[1] = c_y;
tmp[2] = d_y;
pass = true;
for (b_k = 0; b_k < 3; b_k++) {
pass = pass && (tmp[b_k] <= fv1[(0 + (b_k << 1)) + 6 * coreid]);
pass = pass && (tmp[b_k] >= fv1[(1 + (b_k << 1)) + 6 * coreid]);
}
flagPassFail(pass, get_core_id());
synch_barrier();
/////////////////////////////////////////////////////////
// synchronize and exit
/////////////////////////////////////////////////////////
return !pass;
}
int main() {
/* Initialize the RNG */
unsigned rng, i, trials = TRIALS / get_num_cores(), x, y, active;
rng = 0;
init_rng(&rng);
barrier(&b0);
/* Do the simulation */
for (i = 0; i < trials; ++i) {
x = rand_mc(&rng) / ((1u<<31)/10000);
y = rand_mc(&rng) / ((1u<<31)/10000);
if (x * x + y * y <= 100000000) count[get_core_id()]++;
}
printf("core %u: %u\n", get_core_id(), count[get_core_id()]);
barrier(&b1); if (get_core_id() == 0) barrier_init(&b0);
barrier(&b2); if (get_core_id() == 0) barrier_init(&b1);
/* Do the final reduction */
for (active = get_num_cores()/2; active > 0; active /= 2) {
if (get_core_id() < active) {
unsigned idx0 = get_core_id(), idx1 = get_core_id() + active;
count[idx0] = count[idx0] + count[idx1];
printf("%u active cores, sum = %u\n", active, count[idx0]);
}
barrier(&b0); if (get_core_id() == 0) barrier_init(&b2);
barrier(&b1); if (get_core_id() == 0) barrier_init(&b0);
barrier(&b2); if (get_core_id() == 0) barrier_init(&b1);
}
if (get_core_id() == 0) {
unsigned pi_whole = count[0]*4/TRIALS,
pi_frac = count[0]*4%TRIALS/(TRIALS/100);
printf("pi is approximately %u.%02u\n", pi_whole, pi_frac);
}
return 0;
}
/////////////////////////////////////////////////////////
// main testing function
/////////////////////////////////////////////////////////
int main(int argc, const char * const argv[])
{
(void)argc;
(void)argv;
int coreid;
int it;
boolean_T pass;
boolean_T flag;
float y[10];
int ix;
float b_y;
int b_k;
float xbar;
float r;
float c_y;
float check[2];
float golden[4];
/////////////////////////////////////////////////////////
// main test loop
// each core loops over a kernel instance
/////////////////////////////////////////////////////////
coreid = get_core_id();
printf("starting %d kernel iterations... (coreid = %d)\n",KERNEL_ITS,coreid);
if (coreid>3)
coreid=coreid-4;
synch_barrier();
perf_begin();
for(it = 0; it < getKernelIts(); it++)
{
// matlab kernel
for (ix = 0; ix < 10; ix++) {
b_y = 0.0F;
for (b_k = 0; b_k < 10; b_k++) {
b_y += fv1[(ix + 10 * b_k) + 100 * coreid] * fv0[b_k + 10 * coreid];
}
y[ix] = b_y + fv3[coreid] * fv2[ix + 10 * coreid];
}
}
synch_barrier();
perf_end();
synch_barrier();
/////////////////////////////////////////////////////////
// check results
/////////////////////////////////////////////////////////
b_y = y[0];
ix = 0;
xbar = y[0];
for (b_k = 0; b_k < 9; b_k++) {
b_y += y[b_k + 1];
ix++;
xbar += y[ix];
}
xbar *= 1.0F/10.0F;
ix = 0;
r = y[0] - xbar;
c_y = r * r;
for (b_k = 0; b_k < 9; b_k++) {
ix++;
r = y[ix] - xbar;
c_y += r * r;
}
c_y *= 1.0F/9.0F;
check[0] = b_y;
check[1] = c_y;
pass = true;
for (ix = 0; ix < 2; ix++) {
for (b_k = 0; b_k < 2; b_k++) {
golden[b_k + (ix << 1)] = fv4[(b_k + (ix << 1)) + (coreid << 2)];
}
flag = true;
flag = pass && (check[ix] <= golden[ix << 1]);
flag = pass && (check[ix] >= golden[1 + (ix << 1)]);
printErrors(!flag, ix, check[ix], golden[ix<<1], golden[1+(ix<<1)]);
pass = pass && flag;
}
flagPassFail(pass, get_core_id());
/////////////////////////////////////////////////////////
// synchronize and exit
/////////////////////////////////////////////////////////
return !pass;
}
