int ensure_dev_mounted(const char * devPath,const char * mountedPoint){
int ret;
if(devPath == NULL || mountedPoint == NULL){
return -1;
}
mkdir(mountedPoint, 0755); //in case it doesn't already exist
startTiming();
ret = mount(devPath, mountedPoint, "vfat",
MS_NOATIME | MS_NODEV | MS_NODIRATIME, "");
endTimming();
if(ret == 0){
LOGD("mount %s with fs 'vfat' success\n", devPath);
return 0;
}else{
startTiming();
ret = mount(devPath, mountedPoint, "ntfs",
MS_NOATIME | MS_NODEV | MS_NODIRATIME, "");
endTimming();
if(ret == 0){
LOGD("mount %s with fs 'ntfs' success\n", devPath);
return 0;
}else{
startTiming();
ret = mount(devPath, mountedPoint, "ext4",
MS_NOATIME | MS_NODEV | MS_NODIRATIME, "");
endTimming();
if(ret == 0){
LOGD("mount %s with fs 'ext4' success\n", devPath);
return 0;
}
}
LOGD("failed to mount %s (%s)\n", devPath, strerror(errno));
return -1;
}
}
void runTest(StateMachine *machine, int isTableMachine) {
FILE *file;
int buffSize, len;
char *buff;
double rate;
Timer t;
MachineStats stats;
int is_heavy, last_idx_in_root;
double uncommonRate;
stats.totalFailures = 0;
stats.totalGotos = 0;
file = fopen(TEST_INPUT, "rb");
if (!file) {
fprintf(stderr, "Error opening file for reading\n");
exit(1);
}
fseek(file, 0L, SEEK_END);
buffSize = ftell(file);
fseek(file, 0L, SEEK_SET);
buff = (char*)malloc(sizeof(char) * buffSize);
if (buff == NULL) {
fprintf(stderr, "Error allocating memory for buffer\n");
exit(1);
}
len = fread(buff, sizeof(char), buffSize, file);
if (len != buffSize) {
fprintf(stderr, "Error reading data from file\n");
exit(1);
}
t.micros = 0;
if (isTableMachine) {
startTiming(&t);
matchTableMachine((TableStateMachine*)machine, NULL, FALSE, buff, buffSize, 1, NULL, NULL, NULL, NULL, &is_heavy, &last_idx_in_root, &uncommonRate);
endTiming(&t);
} else {
startTiming(&t);
match(machine, buff, buffSize, 0, &stats, 0, 0);
endTiming(&t);
}
rate = GET_TRANSFER_RATE(buffSize, &t);
printf("Time(micros)\tData(No H)\tData(w/ H)\tRate(No H) Mb/s\tRate (w/ H) Mb/s\n");
printf("%8ld\t%9d\t%9d\t%5.4f\t%5.4f\n", t.micros, buffSize, buffSize, rate, rate);
free(buff);
fclose(file);
}
void render()
{
// Render frame
if (prepared)
{
startTiming();
if (animating)
{
if (animStart > 0.0f)
{
animStart -= 0.15f * (1.0f / frameTimer);
}
if ((animate) & (animStart <= 0.0f))
{
timer += 0.5f * (1.0f / frameTimer);
if (timer > 1.0)
{
timer -= 1.0f;
}
}
updateUniformBuffers();
}
draw();
endTiming();
}
}
bool Bilateral::runHalideCPU(Image input, Image output, const Params& params)
{
#if ENABLE_HALIDE
// Create halide buffers
buffer_t inputBuffer = createHalideBuffer(input);
buffer_t outputBuffer = createHalideBuffer(output);
reportStatus("Running Halide CPU filter");
// Warm-up run
halide_bilateral_cpu(&inputBuffer, &outputBuffer);
// Timed runs
startTiming();
for (int i = 0; i < params.iterations; i++)
{
halide_bilateral_cpu(&inputBuffer, &outputBuffer);
}
stopTiming();
halide_release(NULL);
return outputResults(input, output, params);
#else
reportStatus("Halide not enabled during build.");
return false;
#endif
}
/*========================================================================
initReadYUV DEFINITION
======================================================================*/
void initReadYUV(int id, int xSize, int ySize) {
int fsize;
if((ptfile[id] = fopen(path[id], "rb")) == NULL )
{
fprintf(stderr,"ERROR: Task read cannot open yuv_file '%s'\n", path[id]);
system("PAUSE");
return;
}
#ifdef VERBOSE
printf("Opened file '%s'\n", PATH);
#endif
// Obtain file size:
fseek (ptfile[id] , 0 , SEEK_END);
fsize = ftell (ptfile[id]);
rewind (ptfile[id]);
if(fsize < NB_FRAME*(xSize*ySize + xSize*ySize/2))
{
fprintf(stderr,"ERROR: Task read yuv_file incorrect size");
//system("PAUSE");
//return;
}
#ifdef VERBOSE
printf("Correct size for yuv_file '%s'\n", PATH);
#endif
// Set initial clock
startTiming(0);
}
double RFWTimer::start() {
double current = getTime();
base_time = 0.0;
startTiming();
running = true;
return current;
}
bool Sobel::runHalideGPU(Image input, Image output, const Params& params)
{
#if ENABLE_HALIDE
// Create halide buffers
buffer_t inputBuffer = createHalideBuffer(input);
buffer_t outputBuffer = createHalideBuffer(output);
reportStatus("Running Halide GPU filter");
// Warm-up run
inputBuffer.host_dirty = true;
halide_sobel_gpu(&inputBuffer, &outputBuffer);
halide_dev_sync(NULL);
// Timed runs
startTiming();
for (int i = 0; i < params.iterations; i++)
{
halide_sobel_gpu(&inputBuffer, &outputBuffer);
}
halide_dev_sync(NULL);
stopTiming();
halide_copy_to_host(NULL, &outputBuffer);
halide_release(NULL);
return outputResults(input, output, params);
#else
reportStatus("Halide not enabled during build.");
return false;
#endif
}
void run()
{
while (!_done)
{
if (_process)
{
startTiming(_viewer, otherThreadTimeName);
//------------------------------------------------------------
// Your processing goes here.
// Do nothing for the specified number of milliseconds, just so we can
// see it in the stats.
osg::Timer_t startTick = osg::Timer::instance()->tick();
while (osg::Timer::instance()->delta_m(startTick, osg::Timer::instance()->tick()) < _timeToRun)
{
OpenThreads::Thread::YieldCurrentThread();
}
//------------------------------------------------------------
endTiming(_viewer, otherThreadTimeName);
_process = false;
}
else
{
OpenThreads::Thread::microSleep(50);
}
}
}
bool Sobel::runOpenCL(Image input, Image output, const Params& params)
{
if (!initCL(params, sobel_kernel, "-cl-fast-relaxed-math"))
{
return false;
}
cl_int err;
cl_kernel kernel;
cl_mem d_input, d_output;
cl_image_format format = {CL_RGBA, CL_UNORM_INT8};
kernel = clCreateKernel(m_program, "sobel", &err);
CHECK_ERROR_OCL(err, "creating kernel", return false);
d_input = clCreateImage2D(
m_context, CL_MEM_READ_ONLY, &format,
input.width, input.height, 0, NULL, &err);
CHECK_ERROR_OCL(err, "creating input image", return false);
d_output = clCreateImage2D(
m_context, CL_MEM_WRITE_ONLY, &format,
input.width, input.height, 0, NULL, &err);
CHECK_ERROR_OCL(err, "creating output image", return false);
size_t origin[3] = {0, 0, 0};
size_t region[3] = {input.width, input.height, 1};
err = clEnqueueWriteImage(
m_queue, d_input, CL_TRUE,
origin, region, 0, 0, input.data, 0, NULL, NULL);
CHECK_ERROR_OCL(err, "writing image data", return false);
err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &d_input);
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_output);
CHECK_ERROR_OCL(err, "setting kernel arguments", return false);
reportStatus("Running OpenCL kernel");
const size_t global[2] = {output.width, output.height};
const size_t *local = NULL;
if (params.wgsize[0] && params.wgsize[1])
{
local = params.wgsize;
}
// Timed runs
for (int i = 0; i < params.iterations + 1; i++)
{
err = clEnqueueNDRangeKernel(
m_queue, kernel, 2, NULL, global, local, 0, NULL, NULL);
CHECK_ERROR_OCL(err, "enqueuing kernel", return false);
// Start timing after warm-up run
if (i == 0)
{
err = clFinish(m_queue);
CHECK_ERROR_OCL(err, "running kernel", return false);
startTiming();
}
}
double RFWTimer::resume() {
if (running) return getTime();
else {
running = true;
startTiming();
return base_time;
}
}
void FunctionBot::runArm()
{
int state;
int pnulvl;
float lim;
state=getState();
pnulvl=pnuState(state);
if(state==5||state==6)
{
lim=MAX_POS_0/300.0f;
}
else
lim=MAX_POS_1/300.0f;
if(pnulvl>lpnulvl)
{
if(!inTransit)
{
updatePNU(pnulvl);
inTransit=true;
startTiming(1);
}
else if(timeExpired())
{
inTransit=false;
lpnulvl=pnulvl;
if(desiredAngle()<lim)
{
updateElbow(desiredAngle());
}
else
updateElbow(lim);
}
else
{
if(desiredAngle()<MAX_POS_0/300.0f)
updateElbow(desiredAngle());
else
updateElbow(MAX_POS_0/300.0f);
}
}
else
{
if(currentAngle()>lim)
{
updateElbow(lim);
}
else
{
updatePNU(pnulvl);
lpnulvl=pnulvl;
if(desiredAngle()<lim)
updateElbow(desiredAngle());
else
updateElbow(lim);
}
}
updateClaw();
}
void minorCheneyCopyGC (GC_state s) {
size_t bytesAllocated;
size_t bytesCopied;
struct rusage ru_start;
if (DEBUG_GENERATIONAL)
fprintf (stderr, "minorGC nursery = "FMTPTR" frontier = "FMTPTR"\n",
(uintptr_t)s->heap.nursery, (uintptr_t)s->frontier);
assert (invariantForGC (s));
bytesAllocated = s->frontier - s->heap.nursery;
if (bytesAllocated == 0)
return;
s->cumulativeStatistics.bytesAllocated += bytesAllocated;
if (not s->canMinor) {
s->heap.oldGenSize += bytesAllocated;
bytesCopied = 0;
} else {
if (detailedGCTime (s))
startTiming (&ru_start);
s->cumulativeStatistics.numMinorGCs++;
s->forwardState.amInMinorGC = TRUE;
if (DEBUG_GENERATIONAL or s->controls.messages) {
fprintf (stderr,
"[GC: Starting minor Cheney-copy;]\n");
fprintf (stderr,
"[GC:\tfrom nursery at "FMTPTR" of size %s bytes.]\n",
(uintptr_t)(s->heap.nursery),
uintmaxToCommaString(bytesAllocated));
}
s->forwardState.toStart = s->heap.start + s->heap.oldGenSize;
assert (isFrontierAligned (s, s->forwardState.toStart));
s->forwardState.toLimit = s->forwardState.toStart + bytesAllocated;
assert (invariantForGC (s));
s->forwardState.back = s->forwardState.toStart;
/* Forward all globals. Would like to avoid doing this once all
* the globals have been assigned.
*/
foreachGlobalObjptr (s, forwardObjptrIfInNursery);
forwardInterGenerationalObjptrs (s);
foreachObjptrInRange (s, s->forwardState.toStart, &s->forwardState.back,
forwardObjptrIfInNursery, TRUE);
updateWeaksForCheneyCopy (s);
bytesCopied = s->forwardState.back - s->forwardState.toStart;
s->cumulativeStatistics.bytesCopiedMinor += bytesCopied;
s->heap.oldGenSize += bytesCopied;
s->lastMajorStatistics.numMinorGCs++;
if (detailedGCTime (s))
stopTiming (&ru_start, &s->cumulativeStatistics.ru_gcMinor);
if (DEBUG_GENERATIONAL or s->controls.messages)
fprintf (stderr,
"[GC: Finished minor Cheney-copy; copied %s bytes.]\n",
uintmaxToCommaString(bytesCopied));
}
}
void majorCheneyCopyGC (GC_state s) {
size_t bytesCopied;
struct rusage ru_start;
pointer toStart;
assert (s->secondaryHeap.size >= s->heap.oldGenSize);
if (detailedGCTime (s))
startTiming (&ru_start);
s->cumulativeStatistics.numCopyingGCs++;
s->forwardState.amInMinorGC = FALSE;
if (DEBUG or s->controls.messages) {
fprintf (stderr,
"[GC: Starting major Cheney-copy;]\n");
fprintf (stderr,
"[GC:\tfrom heap at "FMTPTR" of size %s bytes,]\n",
(uintptr_t)(s->heap.start),
uintmaxToCommaString(s->heap.size));
fprintf (stderr,
"[GC:\tto heap at "FMTPTR" of size %s bytes.]\n",
(uintptr_t)(s->secondaryHeap.start),
uintmaxToCommaString(s->secondaryHeap.size));
}
s->forwardState.toStart = s->secondaryHeap.start;
s->forwardState.toLimit = s->secondaryHeap.start + s->secondaryHeap.size;
assert (s->secondaryHeap.start != (pointer)NULL);
/* The next assert ensures there is enough space for the copy to
* succeed. It does not assert
* (s->secondaryHeap.size >= s->heap.size)
* because that is too strong.
*/
assert (s->secondaryHeap.size >= s->heap.oldGenSize);
toStart = alignFrontier (s, s->secondaryHeap.start);
s->forwardState.back = toStart;
foreachGlobalObjptr (s, forwardObjptr);
foreachObjptrInRange (s, toStart, &s->forwardState.back, forwardObjptr, TRUE);
updateWeaksForCheneyCopy (s);
s->secondaryHeap.oldGenSize = s->forwardState.back - s->secondaryHeap.start;
bytesCopied = s->secondaryHeap.oldGenSize;
s->cumulativeStatistics.bytesCopied += bytesCopied;
swapHeapsForCheneyCopy (s);
s->lastMajorStatistics.kind = GC_COPYING;
if (detailedGCTime (s))
stopTiming (&ru_start, &s->cumulativeStatistics.ru_gcCopying);
if (DEBUG or s->controls.messages)
fprintf (stderr,
"[GC: Finished major Cheney-copy; copied %s bytes.]\n",
uintmaxToCommaString(bytesCopied));
}
/*========================================================================
readYUV DEFINITION
======================================================================*/
void readYUV(int id, int xSize, int ySize, unsigned char *y, unsigned char *u, unsigned char *v) {
if( ftell(ptfile[id])/(xSize*ySize + xSize*ySize/2) >=NB_FRAME){
rewind(ptfile[id]);
}
if(id == 1 && ftell(ptfile[id])%(FPS*(xSize*ySize + xSize*ySize/2)) == 0){
unsigned int time = 0;
time = stopTiming(0);
printf("\nMain: %d frames in %d us - %f fps\n", FPS ,time, ((float)FPS)/(float)time*1000000);
startTiming(0);
}
fread(y, sizeof(char), xSize * ySize, ptfile[id]);
fread(u, sizeof(char), xSize * ySize / 4, ptfile[id]);
fread(v, sizeof(char), xSize * ySize / 4, ptfile[id]);
}
/// Will just sleep for the given number of milliseconds in the same thread
/// as the caller, recording the time taken in the viewer's stats.
void doSomethingAndTimeIt(osgViewer::Viewer& viewer, const std::string& name, double milliseconds)
{
startTiming(viewer, name);
//------------------------------------------------------------
// Your processing goes here.
// Do nothing for the specified number of milliseconds, just so we can
// see it in the stats.
osg::Timer_t startTick = osg::Timer::instance()->tick();
while (osg::Timer::instance()->delta_m(startTick, osg::Timer::instance()->tick()) < milliseconds)
{
OpenThreads::Thread::YieldCurrentThread();
}
//------------------------------------------------------------
endTiming(viewer, name);
}
void runBench(HASH_FUNCTION hash_fn, int numReplicas, int numNodes, int numKeys, int keySize) {
char *hash = NULL;
if(hash_fn == HASH_FUNCTION_MD5) hash = "MD5";
else if(hash_fn == HASH_FUNCTION_SHA1) hash = "SHA1";
printf("----------------------------------------------------\n");
printf("bench (%s): replicas = %d, nodes = %d, keys: %d, ring size: %d\n", hash, numReplicas, numNodes, numKeys, numReplicas * numNodes);
printf("----------------------------------------------------\n");
hash_ring_t *ring = hash_ring_create(numReplicas, hash_fn);
addNodes(ring, numNodes);
uint8_t *keys = (uint8_t*)malloc(keySize * numKeys);
generateKeys(keys, numKeys, keySize);
printf("running...\r");
uint64_t min = 0;
uint64_t max = 0;
uint64_t total = 0;
int times = 100;
int x, y;
for(y = 0; y < times; y++) {
startTiming();
for(x = 0; x < numKeys; x++) {
assert(hash_ring_find_node(ring, keys + (keySize * x), keySize) != NULL);
}
uint64_t result = endTiming();
if(result > max) max = result;
if(min == 0 || result < min) min = result;
total += result;
}
printf("stats: total = %.5fs, avg/lookup: %.5fus, min: %.5fus, max: %.5fus, ops/sec: %.0f\n",
(double)total / 1000000000,
(((double)(total / numKeys)) / 1000) / times,
(double)min / numKeys / 1000,
(double)max / numKeys / 1000,
1000000000 / ((double)(total / (numKeys * times))));
free(keys);
hash_ring_free(ring);
}
void render()
{
if (prepared)
{
startTiming();
if (animating)
{
// Update rotation
state.rotation.y += 0.05f * frameTimer;
if (state.rotation.y > 360.0f)
{
state.rotation.y -= 360.0f;
}
updateUniformBuffers();
}
draw();
endTiming();
}
}
//单击 ”定时“ 处理
void ToolGlobal::pbn_clock_clicked()
{
if (! clockState) //如果 clockState 是 关闭 状态
{
pbn_clock_state->move(gbx_clock->x()+73, pbn_clock_state->y()); //设置 按钮位置
pbn_clock->move(gbx_clock->pos()); //...........
pbn_clock_state ->setText(tr("OFF")); //设置 按钮文字
pbn_clock ->setText(tr("关闭")); //...........
pbn_clock_state ->setToolTip(tr("关闭")); //设置 按钮提示
pbn_clock ->setToolTip(tr("关闭")); //...........
clockState = true; //设置为 “开启” 状态
// timing = new Timing(parentWidget()); //创建 “定时器设置” 窗口 并显示
timing = new Timing(gbx_clock->x()-42, 113, 225, 135, parentWidget()); //创建 “定时器设置” 窗口 并显示
timing ->show(); //.........................
connect(timing, SIGNAL(timing_pbnOk_click()), this, SLOT(startTiming())); //关联 定时器 “确定” 按钮
connect(timing, SIGNAL(timing_pbnCancle_click()), this, SLOT(stopTiming())); //关联 定时器 “取消” 按钮
}
else //如果 clockState 是 开启 状态
{
if (timing) //如果存在 “定时器设置” 窗口
{
timing ->close(); //关闭 “定时器设置” 窗口
}
if (lcdNumber) //如果存在 “LCD显示器”
{
delete lcdNumber; //销毁 “LCD显示器”
lcdNumber = 0; //指针赋0值
timer_default ->stop(); //终止 “1秒倒计时"
timer_target ->stop(); //终止 "自定义倒计时”
}
pbn_clock_state->move(gbx_clock->x()+13, pbn_clock_state->y()); //设置 按钮位置
pbn_clock->move(gbx_clock->x()+46, pbn_clock->y()); //...........
pbn_clock_state ->setText(tr("ON")); //设置 按钮文字
pbn_clock ->setText(tr("定时")); //...........
pbn_clock_state ->setToolTip(tr("开启魔音")); //设置 按钮提示
pbn_clock ->setToolTip(tr("开启魔音")); //...........
clockState = false; //设置为 “关闭” 状态
}
}
void EmuSampler::setMode(EmuMode mod, FlowID fid)
/* Stop and start statistics for a given mode */
{
//printf("Thread: %u, set to mode: %u\n", fid, mod);
nSwitches->inc();
stop();
switch(mod) {
case EmuRabbit:
startRabbit(fid);
break;
case EmuWarmup:
startWarmup(fid);
break;
case EmuDetail:
startDetail(fid);
break;
case EmuTiming:
startTiming(fid);
break;
default:
I(0);
}
}
InitAndQuit(bool shadersOn)
{
if(SDL_Init(SDL_INIT_VIDEO | SDL_INIT_AUDIO | SDL_INIT_TIMER) != 0)
{
std::ostringstream error;
error << "Failed on SDL_Init: " << SDL_GetError() << "\n";
throw std::runtime_error(error.str());
}
graphics::setupGraphics(shadersOn);
SDL_WM_SetCaption("Sousaphone Hero", "Sousaphone Hero");
if(Mix_OpenAudio(AUDIO_RATE, AUDIO_FORMAT, AUDIO_CHANNELS, AUDIO_BUFFERS) != 0)
{
SDL_Quit();
std::ostringstream error;
error << "Failed on Mix_OpenAudio: " << Mix_GetError() << "\n";
throw std::runtime_error(error.str());
}
std::srand(time(NULL));
Mix_ChannelFinished(doChangedNotes); // set callback function
startTiming();
}
int main (int argc, char **argv) {
int N;
double * values;
char type = argc > 2 ? argv[2][0] : 'A';
if (argc < 2) {
fprintf (stderr, "Usage: csort SIZE [ TYPE [SEED]] \n");
exit (1);
}
N = atoi (argv[1]);
srand (argc > 3 ? atol (argv[3]) : 42);
if (type == 'A')
values = randomVector(N, 0.0, 1.0);
else
values = randomVector(N, 1.0, 20.0);
startTiming();
sort (values, N);
stopTiming();
if (N < 20)
printArr(values, N);
return 0;
}
void SamplerGPUSim::queue(uint32_t insn, uint64_t pc, uint64_t addr, uint32_t fid, char op)
/* main qemu/gpu/tracer/... entry point {{{1 */
{
if(!execute(fid,icount))
return; // QEMU can still send a few additional instructions (emul should stop soon)
I(mode!=EmuInit);
I(insn!=0);
I(icount!=0);
if (doPower){
uint64_t ti = 0;
bool callpwr = callPowerModel(ti, fid);
if (callpwr){
I(ti > 0);
//printf("totalnInst:%ld, nPassedInst:%ld, interval:%ld\n", totalnInst, nPassedInst, interval);
bool dummy = false;
//std::cout<<"mode "<<mode<<" Timeinterval "<<ti<<" last time "<<lastTime<<"\n";
int simt = 0;
if (ti > 0){
setMode(EmuTiming, fid);
simt = BootLoader::pwrmodel.calcStats(ti,
!(mode == EmuTiming), static_cast<float>(freq), dummy, dummy, dummy, dummy);
endSimSiged = (simt==90)?1:0;
BootLoader::pwrmodel.sescThermWrapper->sesctherm.updateMetrics();
}
}
}// doPower
if (nInstMax < totalnInst || endSimSiged) {
markDone();
return;
}
if (nInstSkip>totalnInst) {
I(mode==EmuRabbit);
return;
}
I(nInstSkip<=totalnInst);
if (mode == EmuRabbit) {
stop();
startTiming(fid);
}
#if 0
static std::set<AddrType> seenPC;
static Time_t seenPC_last = 0;
static Time_t seenPC_active = 0;
static Time_t seenPC_total = 0;
seenPC_total++;
if (seenPC.find(pc^insn) == seenPC.end()) {
seenPC.insert(pc^insn);
seenPC_last = seenPC_total;
}
if ((seenPC_last+1000) > seenPC_total)
seenPC_active++;
/*
if ((seenPC_total & 1048575) == 1)
MSG("%5.3f",(100.0*seenPC_active)/(1.0+seenPC_total)); */
#endif
emul->queueInstruction(insn,pc,addr, (op&0x80) /* thumb */, fid);
}
void performGC (GC_state s,
size_t oldGenBytesRequested,
size_t nurseryBytesRequested,
bool forceMajor,
bool mayResize) {
uintmax_t gcTime;
bool stackTopOk;
size_t stackBytesRequested;
struct rusage ru_start;
size_t totalBytesRequested;
enterGC (s);
s->cumulativeStatistics.numGCs++;
if (DEBUG or s->controls.messages) {
size_t nurserySize = s->heap.size - ((size_t)(s->heap.nursery - s->heap.start));
size_t nurseryUsed = (size_t)(s->frontier - s->heap.nursery);
fprintf (stderr,
"[GC: Starting gc #%s; requesting %s nursery bytes and %s old-gen bytes,]\n",
uintmaxToCommaString(s->cumulativeStatistics.numGCs),
uintmaxToCommaString(nurseryBytesRequested),
uintmaxToCommaString(oldGenBytesRequested));
fprintf (stderr,
"[GC:\theap at "FMTPTR" of size %s bytes (+ %s bytes card/cross map),]\n",
(uintptr_t)(s->heap.start),
uintmaxToCommaString(s->heap.size),
uintmaxToCommaString(s->heap.withMapsSize - s->heap.size));
fprintf (stderr,
"[GC:\twith old-gen of size %s bytes (%.1f%% of heap),]\n",
uintmaxToCommaString(s->heap.oldGenSize),
100.0 * ((double)(s->heap.oldGenSize) / (double)(s->heap.size)));
fprintf (stderr,
"[GC:\tand nursery of size %s bytes (%.1f%% of heap),]\n",
uintmaxToCommaString(nurserySize),
100.0 * ((double)(nurserySize) / (double)(s->heap.size)));
fprintf (stderr,
"[GC:\tand nursery using %s bytes (%.1f%% of heap, %.1f%% of nursery).]\n",
uintmaxToCommaString(nurseryUsed),
100.0 * ((double)(nurseryUsed) / (double)(s->heap.size)),
100.0 * ((double)(nurseryUsed) / (double)(nurserySize)));
}
assert (invariantForGC (s));
if (needGCTime (s))
startTiming (&ru_start);
minorGC (s);
stackTopOk = invariantForMutatorStack (s);
stackBytesRequested =
stackTopOk
? 0
: sizeofStackWithHeader (s, sizeofStackGrowReserved (s, getStackCurrent (s)));
totalBytesRequested =
oldGenBytesRequested
+ nurseryBytesRequested
+ stackBytesRequested;
if (forceMajor
or totalBytesRequested > s->heap.size - s->heap.oldGenSize)
majorGC (s, totalBytesRequested, mayResize);
setGCStateCurrentHeap (s, oldGenBytesRequested + stackBytesRequested,
nurseryBytesRequested);
assert (hasHeapBytesFree (s, oldGenBytesRequested + stackBytesRequested,
nurseryBytesRequested));
unless (stackTopOk)
growStackCurrent (s);
setGCStateCurrentThreadAndStack (s);
if (needGCTime (s)) {
gcTime = stopTiming (&ru_start, &s->cumulativeStatistics.ru_gc);
s->cumulativeStatistics.maxPauseTime =
max (s->cumulativeStatistics.maxPauseTime, gcTime);
} else
gcTime = 0; /* Assign gcTime to quell gcc warning. */
if (DEBUG or s->controls.messages) {
size_t nurserySize = s->heap.size - (size_t)(s->heap.nursery - s->heap.start);
fprintf (stderr,
"[GC: Finished gc #%s; time %s ms,]\n",
uintmaxToCommaString(s->cumulativeStatistics.numGCs),
uintmaxToCommaString(gcTime));
fprintf (stderr,
"[GC:\theap at "FMTPTR" of size %s bytes (+ %s bytes card/cross map),]\n",
(uintptr_t)(s->heap.start),
uintmaxToCommaString(s->heap.size),
uintmaxToCommaString(s->heap.withMapsSize - s->heap.size));
fprintf (stderr,
"[GC:\twith old-gen of size %s bytes (%.1f%% of heap),]\n",
uintmaxToCommaString(s->heap.oldGenSize),
100.0 * ((double)(s->heap.oldGenSize) / (double)(s->heap.size)));
fprintf (stderr,
"[GC:\tand nursery of size %s bytes (%.1f%% of heap).]\n",
uintmaxToCommaString(nurserySize),
100.0 * ((double)(nurserySize) / (double)(s->heap.size)));
}
/* Send a GC signal. */
if (s->signalsInfo.gcSignalHandled
and s->signalHandlerThread != BOGUS_OBJPTR) {
if (DEBUG_SIGNALS)
fprintf (stderr, "GC Signal pending.\n");
s->signalsInfo.gcSignalPending = TRUE;
unless (s->signalsInfo.amInSignalHandler)
s->signalsInfo.signalIsPending = TRUE;
}
if (DEBUG)
displayGCState (s, stderr);
assert (hasHeapBytesFree (s, oldGenBytesRequested, nurseryBytesRequested));
assert (invariantForGC (s));
leaveGC (s);
}
void inspectDumpFile(const char *path, int repeat, StateMachine *machine, TableStateMachine *tableMachine, int isTableMachine,
int verbose, int timing, int threads, int packets_to_steal, int dedicated_use_compressed,
int work_group_size, int max_wgs, double *thresholds, int drop) {
#else
void inspectDumpFile(const char *path, int repeat, StateMachine *machine, int isTableMachine, int verbose, int timing, int threads) {
#endif
double /*rate,*/ combinedRate, threadRate;//, rateWithHeaders;
Timer t;
long size;//, sizeWithHeaders;
int i, cpuid;
#ifdef GLOBAL_TIMING
GlobalTimerResult gtimer_result;
int j;
#ifdef PRINT_GLOBAL_TIMER_EVENTS
GlobalTimerEvent **events;
#endif
#endif
ScannerData *scanners;
PacketReaderData reader;
LinkedList *packet_queues;
MulticoreManager manager;
#ifdef COUNT_FAIL_PERCENT
long totalFailures, totalGotos;
#endif
#ifdef PAPI
if (PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT) {
fprintf(stderr, "Cannot init PAPI\n");
exit(1);
}
if (PAPI_thread_init((unsigned long (*)(void))pthread_self) != PAPI_OK) {
fprintf(stderr, "Cannot init PAPI for threads\n");
exit(1);
}
#endif
packet_queues = (LinkedList*)malloc(sizeof(LinkedList) * threads);
scanners = (ScannerData*)malloc(sizeof(ScannerData) * threads);
for (i = 0; i < threads; i++) {
list_init(&packet_queues[i]);
}
packetreader_init(&reader, path, repeat, packet_queues, threads);
for (i = 0; i < threads; i++) {
#ifdef HYBRID_SCANNER
scanner_init(&(scanners[i]), i, &manager, machine, tableMachine, isTableMachine, &packet_queues[i], verbose, drop);
#else
scanner_init(&(scanners[i]), i, &manager, machine, isTableMachine, &packet_queues[i], verbose);
#endif
}
#ifdef HYBRID_SCANNER
multicore_manager_init(&manager, scanners, threads, work_group_size, max_wgs, packets_to_steal, dedicated_use_compressed);
multicore_manager_set_thresholds(&manager, thresholds);
#else
multicore_manager_init(&manager, scanners, threads, 1, threads, 0, 0);
#endif
packetreader_start(&reader);
packetreader_join(&reader);
#ifdef HYBRID_SCANNER
multicore_manager_start(&manager);
#endif
#ifdef GLOBAL_TIMING
#ifdef PRINT_GLOBAL_TIMER_EVENTS
events = NULL;
#endif
global_timer_start(&(manager.gtimer));
#endif
if (timing) {
startTiming(&t);
}
for (i = 0; i < threads; i++) {
// If CPUs are ordered [core0,core0,...,core0,core1,core1,...,core1,...]
//cpuid = i;
// If CPUs are ordered [core0,core1,...,coreN,core0,core1,...,coreN,...]
cpuid = (i % 2 == 0) ? i / 2 : (threads + i) / 2;
scanner_start_with_affinity(&(scanners[i]), cpuid);
// If you use the next line, comment out the pthread_attr_destroy call in scanner_join!!!
//scanner_start(&(scanners[i]));
}
for (i = 0; i < threads; i++) {
scanner_join(&(scanners[i]));
}
// scanner_start(&(scanners[0]));
// scanner_start(&(scanners[1]));
// scanner_join(&(scanners[0]));
// scanner_join(&(scanners[1]));
if (timing) {
endTiming(&t);
}
#ifdef GLOBAL_TIMING
global_timer_end(&(manager.gtimer));
#endif
#ifdef HYBRID_SCANNER
multicore_manager_stop(&manager);
multicore_manager_join(&manager);
#endif
#ifdef GLOBAL_TIMING
global_timer_join(&(manager.gtimer));
global_timer_get_results(&(manager.gtimer), >imer_result);
#endif
if (timing) {
//endTiming(&t);
size = reader.size;
//sizeWithHeaders = reader.sizeWithHeaders;
//rate = GET_TRANSFER_RATE(size, &t);
//rateWithHeaders = GET_TRANSFER_RATE(sizeWithHeaders, &t);
// printf("Time(micros)\tData(No H)\tData(w/ H)\tRate(No H) Mb/s\tRate (w/ H) Mb/s\n");
//printf("%8ld\t%9ld\t%9ld\t%5.4f\t%5.4f\n", t.micros, size, sizeWithHeaders, rate, rateWithHeaders);
printf("TOTAL_BYTES\tTotal data scanned: %ld bytes\n", size);
//printf("TOTAL_TIME\tTotal time: %ld ms\n", t.micros);
//printf("TOTAL_THRPT\tTotal throughput: %5.4f Mbps\n", rate);
combinedRate = 0;
printf("Alert mode timer: %ld us\n", manager.alert_mode_timer.micros);
for (i = 0; i < threads; i++) {
if (0 && manager.alert_mode_used) {
threadRate = GET_TRANSFER_RATE(scanners[i].bytes_scanned_since_alert, &(manager.alert_mode_timer));
} else {
threadRate = GET_SCANNER_TRANSFER_RATE(&(scanners[i]));
}
combinedRate += threadRate;
printf("T_%2d_THRPT\t%5.4f\tMbps\t%lu\tB\t%lu\tB\t%ld\tus\n", i, threadRate, scanners[i].bytes_scanned, scanners[i].bytes_scanned_since_alert, scanners[i].timer.micros);
}
printf("COMB_THRPT\t%5.4f\tMbps\n", combinedRate);
#ifdef GLOBAL_TIMING
//printf("\nGlobal timing:\n");
/*
printf("Time\t");
for (j = 0; j < manager.gtimer.intervals; j++) {
printf("%6ld\t", gtimer_result.times[j]);
}
printf("\n");
*/
for (i = 0; i < manager.gtimer.num_scanners; i++) {
printf("T_%2d_GTIME\t", i);
for (j = 0; j < manager.gtimer.intervals; j++) {
printf("%5.3f\t", gtimer_result.results[gtimer_result.intervals * i + j]);
}
printf("\n");
}
#ifdef PRINT_GLOBAL_TIMER_EVENTS
j = global_timer_get_events(&(manager.gtimer), &events);
if (j > 0) {
printf("\nEvents:\n");
for (i = 0; i < j; i++) {
printf("Event %d: %s [Time: %d, Source: %s]\n", i, events[i]->text, events[i]->interval, events[i]->source);
}
}
#endif
#endif
}
#ifdef COUNT_FAIL_PERCENT
totalFailures = totalGotos = 0;
for (i = 0; i < threads; i++) {
totalFailures += scanners[i].stats.totalFailures;
totalGotos += scanners[i].stats.totalGotos;
}
printf("Fail percent: %f\n", ((double)totalFailures) / (totalFailures + totalGotos));
printf("Total failures: %ld, Total gotos: %ld\n", totalFailures, totalGotos);
#endif
multicore_manager_destroy(&manager);
#ifdef GLOBAL_TIMING
global_timer_destroy(&(manager.gtimer));
global_timer_destroy_result(>imer_result);
#endif
free(scanners);
for (i = 0; i < threads; i++) {
//printf("Status of input-queue of thread %d: in=%d, out=%d\n", i, packet_queues[i].in, packet_queues[i].out);
list_destroy(&(packet_queues[i]), 1);
}
free(packet_queues);
}
// ----------------------------------------------------------------------------
// gpuNUFFT_gpu: NUFFT
//
// Inverse gpuNUFFT implementation - interpolation from uniform grid data onto
// nonuniform k-space data based on optimized
// gpuNUFFT kernel with minimal oversampling
// ratio (see Beatty et al.)
//
// Basic steps: - apodization correction
// - zero padding with osf
// - FFT
// - convolution and resampling
//
// parameters:
// * data : output kspace data
// * data_count : number of samples on trajectory
// * n_coils : number of channels or coils
// * crds : coordinates on trajectory, passed as SoA
// * imdata : input image data
// * imdata_count : number of image data points
// * grid_width : size of grid
// * kernel : precomputed convolution kernel as lookup table
// * kernel_count : number of kernel lookup table entries
// * sectors : mapping of data indices according to each sector
// * sector_count : number of sectors
// * sector_centers: coordinates (x,y,z) of sector centers
// * sector_width : width of sector
// * im_width : dimension of image
// * osr : oversampling ratio
// * gpuNUFFT_out : enum indicating how far gpuNUFFT has to be processed
//
void gpuNUFFT::GpuNUFFTOperator::performForwardGpuNUFFT(gpuNUFFT::Array<DType2> imgData,gpuNUFFT::Array<CufftType>& kspaceData, GpuNUFFTOutput gpuNUFFTOut)
{
if (DEBUG)
{
std::cout << "performing forward gpuNUFFT!!!" << std::endl;
std::cout << "dataCount: " << kspaceData.count() << " chnCount: " << kspaceData.dim.channels << std::endl;
std::cout << "imgCount: " << imgData.count() << " gridWidth: " << this->getGridWidth() << std::endl;
}
showMemoryInfo();
if (debugTiming)
startTiming();
int data_count = (int)this->kSpaceTraj.count();
int n_coils = (int)kspaceData.dim.channels;
IndType imdata_count = this->imgDims.count();
int sector_count = (int)this->gridSectorDims.count();
//cuda mem allocation
DType2 *imdata_d;
CufftType *data_d;
if (DEBUG)
printf("allocate and copy imdata of size %d...\n",imdata_count);
allocateDeviceMem<DType2>(&imdata_d,imdata_count);
if (DEBUG)
printf("allocate and copy data of size %d...\n",data_count);
allocateDeviceMem<CufftType>(&data_d,data_count);
initDeviceMemory(n_coils);
if (debugTiming)
printf("Memory allocation: %.2f ms\n",stopTiming());
int err;
//iterate over coils and compute result
for (int coil_it = 0; coil_it < n_coils; coil_it++)
{
int data_coil_offset = coil_it * data_count;
int im_coil_offset = coil_it * (int)imdata_count;
if (this->applySensData())
// perform automatically "repeating" of input image in case
// of existing sensitivity data
copyToDevice(imgData.data,imdata_d,imdata_count);
else
copyToDevice(imgData.data + im_coil_offset,imdata_d,imdata_count);
//reset temp arrays
cudaMemset(gdata_d,0, sizeof(CufftType)*gi_host->grid_width_dim);
cudaMemset(data_d,0, sizeof(CufftType)*data_count);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 1: %s\n",cudaGetErrorString(cudaGetLastError()));
if (this->applySensData())
{
copyToDevice(this->sens.data + im_coil_offset, sens_d,imdata_count);
performSensMul(imdata_d,sens_d,gi_host,false);
}
// apodization Correction
if (n_coils > 1 && deapo_d != NULL)
performForwardDeapodization(imdata_d,deapo_d,gi_host);
else
performForwardDeapodization(imdata_d,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 2: %s\n",cudaGetErrorString(cudaGetLastError()));
// resize by oversampling factor and zero pad
performPadding(imdata_d,gdata_d,gi_host);
if (debugTiming)
startTiming();
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 3: %s\n",cudaGetErrorString(cudaGetLastError()));
// shift image to get correct zero frequency position
performFFTShift(gdata_d,INVERSE,getGridDims(),gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 4: %s\n",cudaGetErrorString(cudaGetLastError()));
// eventually free imdata_d
// Forward FFT to kspace domain
if (err=pt2CufftExec(fft_plan, gdata_d, gdata_d, CUFFT_FORWARD) != CUFFT_SUCCESS)
{
fprintf(stderr,"cufft has failed with err %i \n",err);
showMemoryInfo(true,stderr);
}
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 5: %s\n",cudaGetErrorString(cudaGetLastError()));
performFFTShift(gdata_d,FORWARD,getGridDims(),gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 6: %s\n",cudaGetErrorString(cudaGetLastError()));
if (debugTiming)
printf("FFT (incl. shift): %.2f ms\n",stopTiming());
if (debugTiming)
startTiming();
// convolution and resampling to non-standard trajectory
forwardConvolution(data_d,crds_d,gdata_d,NULL,sectors_d,sector_centers_d,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at thread synchronization 7: %s\n",cudaGetErrorString(cudaGetLastError()));
if (debugTiming)
printf("Forward Convolution: %.2f ms\n",stopTiming());
performFFTScaling(data_d,gi_host->data_count,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error: at thread synchronization 8: %s\n",cudaGetErrorString(cudaGetLastError()));
//write result in correct order back into output array
writeOrderedGPU(data_sorted_d,data_indices_d,data_d,(int)this->kSpaceTraj.count());
copyFromDevice(data_sorted_d,kspaceData.data + data_coil_offset,data_count);
}//iterate over coils
freeTotalDeviceMemory(data_d,imdata_d,NULL);
freeDeviceMemory(n_coils);
if ((cudaThreadSynchronize() != cudaSuccess))
fprintf(stderr,"error in performForwardGpuNUFFT function: %s\n",cudaGetErrorString(cudaGetLastError()));
free(gi_host);
}
// ----------------------------------------------------------------------------
// performGpuNUFFTAdj: NUFFT^H
//
// GpuNUFFT implementation - interpolation from nonuniform k-space data onto
// oversampled grid based on optimized gpuNUFFT kernel
// with minimal oversampling ratio (see Beatty et al.)
//
// Basic steps: - density compensation
// - convolution with interpolation function
// - iFFT
// - cropping due to oversampling ratio
// - apodization correction
//
// parameters:
// * data : input kspace data
// * data_count : number of samples on trajectory
// * n_coils : number of channels or coils
// * crds : coordinate array on trajectory
// * imdata : output image data
// * imdata_count : number of image data points
// * grid_width : size of grid
// * kernel : precomputed convolution kernel as lookup table
// * kernel_count : number of kernel lookup table entries
// * sectors : mapping of start and end points of each sector
// * sector_count : number of sectors
// * sector_centers: coordinates (x,y,z) of sector centers
// * sector_width : width of sector
// * im_width : dimension of image
// * osr : oversampling ratio
// * do_comp : true, if density compensation has to be done
// * density_comp : densiy compensation array
// * gpuNUFFT_out : enum indicating how far gpuNUFFT has to be processed
//
void gpuNUFFT::GpuNUFFTOperator::performGpuNUFFTAdj(gpuNUFFT::Array<DType2> kspaceData, gpuNUFFT::Array<CufftType>& imgData, GpuNUFFTOutput gpuNUFFTOut)
{
if (DEBUG)
{
std::cout << "performing gpuNUFFT adjoint!!!" << std::endl;
std::cout << "dataCount: " << kspaceData.count() << " chnCount: " << kspaceData.dim.channels << std::endl;
std::cout << "imgCount: " << imgData.count() << " gridWidth: " << this->getGridWidth() << std::endl;
std::cout << "apply density comp: " << this->applyDensComp() << std::endl;
std::cout << "apply sens data: " << this->applySensData() << std::endl;
}
if (debugTiming)
startTiming();
showMemoryInfo();
int data_count = (int)this->kSpaceTraj.count();
int n_coils = (int)kspaceData.dim.channels;
IndType imdata_count = this->imgDims.count();
int sector_count = (int)this->gridSectorDims.count();
// select data ordered and leave it on gpu
DType2* data_d;
if (DEBUG)
printf("allocate data of size %d...\n",data_count);
allocateDeviceMem<DType2>(&data_d,data_count);
CufftType *imdata_d, *imdata_sum_d = NULL;
if (DEBUG)
printf("allocate and copy imdata of size %d...\n",imdata_count);
allocateDeviceMem<CufftType>(&imdata_d,imdata_count);
if (this->applySensData())
{
if (DEBUG)
printf("allocate and copy temp imdata of size %d...\n",imdata_count);
allocateDeviceMem<CufftType>(&imdata_sum_d,imdata_count);
cudaMemset(imdata_sum_d,0,imdata_count*sizeof(CufftType));
}
initDeviceMemory(n_coils);
int err;
if (debugTiming)
printf("Memory allocation: %.2f ms\n",stopTiming());
//iterate over coils and compute result
for (int coil_it = 0; coil_it < n_coils; coil_it++)
{
int data_coil_offset = coil_it * data_count;
int im_coil_offset = coil_it * (int)imdata_count;//gi_host->width_dim;
cudaMemset(gdata_d,0, sizeof(CufftType)*gi_host->grid_width_dim);
//copy coil data to device and select ordered
copyToDevice(kspaceData.data + data_coil_offset,data_d,data_count);
selectOrderedGPU(data_d,data_indices_d,data_sorted_d,data_count);
if (this->applyDensComp())
performDensityCompensation(data_sorted_d,density_comp_d,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at adj thread synchronization 1: %s\n",cudaGetErrorString(cudaGetLastError()));
if (debugTiming)
startTiming();
adjConvolution(data_sorted_d,crds_d,gdata_d,NULL,sectors_d,sector_centers_d,gi_host);
if (debugTiming)
printf("Adjoint convolution: %.2f ms\n",stopTiming());
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
fprintf(stderr,"error at adj thread synchronization 2: %s\n",cudaGetErrorString(cudaGetLastError()));
if (gpuNUFFTOut == CONVOLUTION)
{
if (DEBUG)
printf("stopping output after CONVOLUTION step\n");
//get output
copyFromDevice<CufftType>(gdata_d,imgData.data,gi_host->grid_width_dim);
if (DEBUG)
printf("test value at point zero: %f\n",(imgData.data)[0].x);
free(gi_host);
freeTotalDeviceMemory(data_d,imdata_d,imdata_sum_d,NULL);
freeDeviceMemory(n_coils);
return;
}
if ((cudaThreadSynchronize() != cudaSuccess))
fprintf(stderr,"error at adj thread synchronization 3: %s\n",cudaGetErrorString(cudaGetLastError()));
if (debugTiming)
startTiming();
performFFTShift(gdata_d,INVERSE,getGridDims(),gi_host);
//Inverse FFT
if (err=pt2CufftExec(fft_plan, gdata_d, gdata_d, CUFFT_INVERSE) != CUFFT_SUCCESS)
{
fprintf(stderr,"cufft has failed at adj with err %i \n",err);
showMemoryInfo(true,stderr);
}
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
fprintf(stderr,"error at adj thread synchronization 4: %s\n",cudaGetErrorString(cudaGetLastError()));
if (gpuNUFFTOut == FFT)
{
if (DEBUG)
printf("stopping output after FFT step\n");
//get output
copyFromDevice<CufftType>(gdata_d,imgData.data,gi_host->grid_width_dim);
free(gi_host);
freeTotalDeviceMemory(data_d,imdata_d,imdata_sum_d,NULL);
freeDeviceMemory(n_coils);
printf("last cuda error: %s\n", cudaGetErrorString(cudaGetLastError()));
return;
}
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at adj thread synchronization 5: %s\n",cudaGetErrorString(cudaGetLastError()));
performFFTShift(gdata_d,INVERSE,getGridDims(),gi_host);
if (debugTiming)
printf("iFFT (incl. shift) : %.2f ms\n",stopTiming());
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at adj thread synchronization 6: %s\n",cudaGetErrorString(cudaGetLastError()));
performCrop(gdata_d,imdata_d,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at adj thread synchronization 7: %s\n",cudaGetErrorString(cudaGetLastError()));
//check if precomputed deapo function can be used
if (n_coils > 1 && deapo_d != NULL)
performDeapodization(imdata_d,deapo_d,gi_host);
else
performDeapodization(imdata_d,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error at adj thread synchronization 8: %s\n",cudaGetErrorString(cudaGetLastError()));
performFFTScaling(imdata_d,gi_host->im_width_dim,gi_host);
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error: at adj thread synchronization 9: %s\n",cudaGetErrorString(cudaGetLastError()));
if (this->applySensData())
{
copyToDevice(this->sens.data + im_coil_offset, sens_d,imdata_count);
performSensMul(imdata_d,sens_d,gi_host,true);
performSensSum(imdata_d,imdata_sum_d,gi_host);
}
else
{
// get result per coil
// no summation is performed in absence of sensitity data
copyFromDevice<CufftType>(imdata_d,imgData.data+im_coil_offset,imdata_count);
}
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error: at adj thread synchronization 10: %s\n",cudaGetErrorString(cudaGetLastError()));
}//iterate over coils
if (this->applySensData())
{
// get result of combined coils
copyFromDevice<CufftType>(imdata_sum_d,imgData.data,imdata_count);
}
if (DEBUG && (cudaThreadSynchronize() != cudaSuccess))
printf("error: at adj thread synchronization 11: %s\n",cudaGetErrorString(cudaGetLastError()));
freeTotalDeviceMemory(data_d,imdata_d,imdata_sum_d,NULL);
freeDeviceMemory(n_coils);
if ((cudaThreadSynchronize() != cudaSuccess))
fprintf(stderr,"error in gpuNUFFT_gpu_adj function: %s\n",cudaGetErrorString(cudaGetLastError()));
free(gi_host);
}
/*
* Translate a method into native code.
*
* Registers are allocated per basic block, using an LRU algorithm.
* Contents of registers are spilled at the end of basic block,
* depending on the edges in the CFG leaving the basic block:
*
* - If there is an edge from the basic block to an exception handler,
* local variables are spilled on the stack
*
* - If there is only one non-exception edge, and the target basic
* block is following the current block immediately, no spills are done
*
* - Otherwise, the local variables and the operand stack are spilled
* onto the stack
*/
jboolean
translate(Method* xmeth, errorInfo* einfo)
{
#if defined(KAFFE_VMDEBUG)
int i;
#endif /* defined(KAFFE_VMDEBUG) */
jint low;
jint high;
jvalue tmpl;
int idx;
SlotInfo* tmp;
SlotInfo* tmp2;
SlotInfo* mtable;
bytecode* base;
uint32 len;
callInfo cinfo;
fieldInfo finfo;
Hjava_lang_Class* crinfo;
codeinfo* mycodeInfo;
nativeCodeInfo ncode;
int64 tms = 0;
int64 tme;
static bool reinvoke = false;
jboolean success = true;
lockClass(xmeth->class);
if (METHOD_TRANSLATED(xmeth)) {
goto done3;
}
/* If this code block is native, then just set it up and return */
if (methodIsNative(xmeth)) {
void *func = native(xmeth, einfo);
if (func != NULL) {
engine_create_wrapper(xmeth, func);
KAFFEJIT_TO_NATIVE(xmeth);
} else {
success = false;
}
goto done3;
}
/* Scan the code and determine the basic blocks */
success = analyzeMethod(xmeth, &mycodeInfo, einfo);
if (success == false) {
/* It may happen that we already have translated it
* by implicit recursion in the verifier.
*/
if (METHOD_TRANSLATED(xmeth))
success = true;
goto done3;
}
#if defined(KAFFE_FEEDBACK)
if( kaffe_feedback_file )
lockMutex(kaffe_feedback_file);
#endif
/* Only one in the translator at once. Must check the translation
* hasn't been done by someone else once we get it.
*/
enterTranslator();
startTiming(&fulljit, "JIT translation");
if (Kaffe_JavaVMArgs.enableVerboseJIT) {
tms = currentTime();
}
DBG(MOREJIT,
dprintf("callinfo = %p\n", &cinfo);
);
int main(int argc, char** argv)
{
cout << endl<< endl << "********************** program start************************ " << endl << endl;
int nx = 256, ny = 256, nz = 256, nn = nx*ny*nz;
FLOAT dx = Lx/(FLOAT)nx, dy = Ly/(FLOAT)ny, dz = Lz/(FLOAT)nz;
FLOAT dt = 0.1*dx*dx/KAPPA;
int step = 1000;
double elaps=0.0;
double getElapsedTime();
int thread_num=1;
#ifdef _OPENMP
#pragma omp parallel
{
thread_num = omp_get_num_threads();
if(omp_get_thread_num()==0)cout<<"\nUsed Number of Threads : "<< thread_num <<endl<<endl;
}
#endif
// To avoid chaching effects for small message sizes //
int fact = 1; for(;fact*nn*sizeof(FLOAT)<100e6;++fact);
cout << "fact = " << fact << endl;
// FLOAT* f = (FLOAT *)scalable_aligned_malloc(sizeof(FLOAT) * nn, SIMDALIGN);
//FLOAT* fn = (FLOAT *)scalable_aligned_malloc(sizeof(FLOAT) * nn, SIMDALIGN);
FLOAT* f = (FLOAT *)_mm_malloc(sizeof(FLOAT) * nn, SIMDALIGN);
FLOAT* fn = (FLOAT *)_mm_malloc(sizeof(FLOAT) * nn, SIMDALIGN);
initArray(f ,nx, ny, nz);
initArray(fn,nx, ny, nz);
long data = 0;
FLOAT flops=0.0;
startTiming();
for(int n = 0;n<step;++n){
// flops += diffusion_simd(nx, ny, nz, nn, dx, dy, dz, dt, f, fn);
// flops += diffusion_peel(nx, ny, nz, nn, dx, dy, dz, dt, f, fn);
flops += diffusion_tiled(nx, ny, nz, nn, dx, dy, dz, dt, f, fn);
data+=nn*2*sizeof(FLOAT);
swap(&f, &fn);
elaps += dt;
}
endTiming();
cout<<"Buffer Size: " <<sizeof(FLOAT)*nn/(1000.0*1000.0) <<" [MB] Total Data: "<<data/(1000.0*1000.0*1000.0)<<" [GB]"<<endl;
cout<<"Bandwidth: " <<data/(1000.0*1000.0*1000.0*getElapsedTime())<<"[GB/s]"<<endl;
cout<<"FLOPS : " <<flops/(1000.0*1000.0*1000.0*getElapsedTime())<<"[GFLOPS]"<<endl;
cout<<"Elapsed Time: " <<getElapsedTime()<<endl<<endl;
error_func(nx, ny, nz, dx, f, elaps);
// scalable_aligned_free(f );
// scalable_aligned_free(fn);
_mm_free(f );
_mm_free(fn);
return 0;
}
