Mercurial > hg > Game > Cerium
changeset 1544:5c4e3f0d372a draft
many_task add task array
| author | Yuhi TOMARI <yuhi@cr.ie.u-ryukyu.ac.jp> |
|---|---|
| date | Thu, 31 Jan 2013 18:47:13 +0900 |
| parents | 890cb39418ca |
| children | d9eb89610733 |
| files | TaskManager/Gpu/GpuScheduler.cc example/flip/main.cc example/flip/task_init.cc example/flip/twice.cl example/many_task/Makefile.gpu example/many_task/Makefile.macosx example/many_task/gpu/QuickSort.cl example/many_task/main.cc example/many_task/sort.cc example/many_task/sort.h example/many_task/sort_ta.cc example/many_task/task_init.cc example/word_count/a.txt example/word_count/gpu/Exec.cl example/word_count/main.cc example/word_count/ppe/Print.cc |
| diffstat | 16 files changed, 338 insertions(+), 91 deletions(-) [+] |
line wrap: on
line diff
--- a/TaskManager/Gpu/GpuScheduler.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/TaskManager/Gpu/GpuScheduler.cc Thu Jan 31 18:47:13 2013 +0900 @@ -115,15 +115,15 @@ const char *msg=convert_error_status(ret); error(msg); } - + ret = clSetKernelArg(kernel, param, sizeof(memaddr),(void *)&memparam); if (ret<0) { const char *msg=convert_error_status(ret); error(msg); } - + param++; - + cl_mem_flags mem_flag = CL_MEM_READ_ONLY; cl_mem *memin = new cl_mem[nextTask->inData_count]; if (!flag.flip) { // set input data when not flip @@ -146,8 +146,15 @@ } } - cl_mem *memout = new cl_mem[nextTask->outData_count]; - cl_mem_flags out_mem_flag = flag.flip? CL_MEM_READ_WRITE : CL_MEM_WRITE_ONLY; + cl_mem *memout; + cl_mem_flags out_mem_flag; + if (flag.flip) { + memout = new cl_mem[nextTask->inData_count]; + out_mem_flag = CL_MEM_READ_WRITE; + } else { + memout = new cl_mem[nextTask->outData_count]; + out_mem_flag = CL_MEM_WRITE_ONLY; + } for(int i=0;i<nextTask->outData_count;i++) { // set output data
--- a/example/flip/main.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/flip/main.cc Thu Jan 31 18:47:13 2013 +0900 @@ -83,7 +83,7 @@ twice->set_param(0, (memaddr)length); twice->set_inData(0, indata, sizeof (int)*length); - twice->set_outData(0, indata, sizeof (int)*length); + // twice->set_outData(0, indata, sizeof (int)*length); twice->set_cpu(GPU_0); twice->flip();
--- a/example/flip/task_init.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/flip/task_init.cc Thu Jan 31 18:47:13 2013 +0900 @@ -4,9 +4,5 @@ void task_init(void) { - int cmd = Twice; - int dim = 2; - size_t *l_work_size = new size_t(dim); - GpuNDRangeRegister(cmd, dim, l_work_size); - GpuSchedRegister(cmd, "twice.cl", "twice"); + GpuSchedRegister(Twice, "twice.cl", "twice"); }
--- a/example/flip/twice.cl Sun Jan 13 22:53:50 2013 +0900 +++ b/example/flip/twice.cl Thu Jan 31 18:47:13 2013 +0900 @@ -1,12 +1,12 @@ __kernel void twice(__constant int *data_count, - __global int *input_data) - //__global int *output_data) + __global int *input_data, + __global int *output_data) { long count = (long)data_count[0]; - for (int i = 0; i<count; i++) { - // output_data[i] = 2*input_data[i]; - input_data[i] *= 2; - } + for (int i = 0; i<count; i++) { + output_data[i] = 2*input_data[i]; + //input_data[i] *= 2; + } }
--- a/example/many_task/Makefile.gpu Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/Makefile.gpu Thu Jan 31 18:47:13 2013 +0900 @@ -1,7 +1,7 @@ include ./Makefile.def SRCS_TMP = $(wildcard *.cc) -SRCS_EXCLUDE = sort_test.cc gpu/gpu_task_init.cc # 除外するファイルを書く +SRCS_EXCLUDE = sort-compat.cc sort_test.cc gpu/gpu_task_init.cc # 除外するファイルを書く SRCS = $(filter-out $(SRCS_EXCLUDE),$(SRCS_TMP)) OBJS = $(SRCS:.cc=.o)
--- a/example/many_task/Makefile.macosx Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/Makefile.macosx Thu Jan 31 18:47:13 2013 +0900 @@ -2,7 +2,7 @@ SRCS_TMP = $(wildcard *.cc) -SRCS_EXCLUDE = sort_test.cc # 除外するファイルを書く +SRCS_EXCLUDE = sort-compat.cc sort_test.cc # 除外するファイルを書く SRCS = $(filter-out $(SRCS_EXCLUDE),$(SRCS_TMP)) OBJS = $(SRCS:.cc=.o)
--- a/example/many_task/gpu/QuickSort.cl Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/gpu/QuickSort.cl Thu Jan 31 18:47:13 2013 +0900 @@ -35,7 +35,7 @@ int p; while (1) { - while (beegin < end) { + while (begin < end) { /* * if (end-begin <= 50) { * bubble_sort(data,begin,end);
--- a/example/many_task/main.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/main.cc Thu Jan 31 18:47:13 2013 +0900 @@ -109,8 +109,8 @@ sorter.split_num = get_split_num(sorter.data_length, sorter.cpuNum); // data の分割数 int half_num = sorter.split_num-1; - sorter.fsort = (HTaskPtr*)manager->allocate(sizeof(HTaskPtr)*sorter.split_num); - sorter.bsort = (HTaskPtr*)manager->allocate(sizeof(HTaskPtr)*half_num); + sorter.fsort = (Task**)manager->allocate(sizeof(Task*)*sorter.split_num); + sorter.bsort = (Task**)manager->allocate(sizeof(Task*)*half_num); memset((void*)sorter.bsort,0, sizeof(HTaskPtr)*half_num); for (int i = 0; i < length; i++) {
--- a/example/many_task/sort.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/sort.cc Thu Jan 31 18:47:13 2013 +0900 @@ -58,88 +58,86 @@ return 0; } + HTask **task_array_f = (HTask**)manager->allocate(sizeof(HTask*)*ta); + HTask **task_array_b = (HTask**)manager->allocate(sizeof(HTask*)*half_num); - + for (int i = 0; i < s->split_num;i++) { + task_array_f[i] = manager->create_task_array(QUICK_SORT, task_array_num,1,1,1); + s->fsort[i]=0; + } + for (int i = 0; i<half_num;i++) { + task_array_b[i] = manager->create_task_array(QUICK_SORT, task_array_num,1,1,1); + s->bsort[i]=0; + } for (int i = 0; i < s->split_num-1; i++) { - if (task_array_num) { - s->fsort[i] = manager->create_task_array(QUICK_SORT, task_array_num, block_num+1, block_num, block_num); - } else { - s->fsort[i] = manager->create_task(QUICK_SORT, - (memaddr)&s->data[i*block_num], sizeof(Data)*block_num, - (memaddr)&s->data[i*block_num], sizeof(Data)*block_num); - } - - s->fsort[i]->flip(); - + s->fsort[i] = task_array_f[i]->next_task_array(QUICK_SORT,s->fsort[i]); + s->fsort[i]->set_param(0,(memaddr)block_num); + s->fsort[i]->set_inData(0,(memaddr)&s->data[i*block_num], sizeof(Data)*block_num); if (i>0 && s->bsort[i-1]) { - s->fsort[i]->wait_for(s->bsort[i-1]); + task_array_f[i]->wait_for(task_array_b[i-1]); } if (i<s->split_num-2 && s->bsort[i]) { - s->fsort[i]->wait_for(s->bsort[i]); + task_array_f[i]->wait_for(task_array_b[i]); } - s->fsort[i]->set_cpu(spe_cpu); - s->fsort[i]->set_param(0,(memaddr)block_num); } // 最後の block は端数なので last_block_num を使う { + int i = s->split_num-1; - s->fsort[i] = manager->create_task(QUICK_SORT, - (memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num, - (memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num); - s->fsort[i]->flip(); + s->fsort[i] = task_array_f[i]->next_task_array(QUICK_SORT,s->fsort[i]); + s->fsort[i]->set_param(0,(memaddr)last_block_num); + s->fsort[i]->set_inData(0,(memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num); if (i>0 && s->bsort[i-1]) { - s->fsort[i]->wait_for(s->bsort[i-1]); + task_array_f[i]->wait_for(task_array_b[i-1]); } - s->fsort[i]->set_cpu(spe_cpu); - s->fsort[i]->set_param(0,(memaddr)last_block_num); } if (s->split_num > 1) { for (int i = 0; i < half_num-1; i++) { - if (s->bsort[i]) manager->free_htask(s->bsort[i]); - s->bsort[i] = manager->create_task(QUICK_SORT, - (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num, - (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num); - s->bsort[i]->flip(); - s->bsort[i]->set_cpu(spe_cpu); + if (s->bsort[i]) s->bsort[i]=0; + s->bsort[i] = task_array_b[i]->next_task_array(QUICK_SORT,s->bsort[i]); + s->bsort[i]->set_inData(0,(memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num); s->bsort[i]->set_param(0,(memaddr)block_num); } { int i = half_num-1; - if (s->bsort[i]) manager->free_htask(s->bsort[i]); - s->bsort[i] = manager->create_task(QUICK_SORT, - (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num, - (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num); - s->bsort[i]->flip(); - s->bsort[i]->set_cpu(spe_cpu); + if (s->bsort[i]) s->bsort[i]=0; + s->bsort[i] = task_array_b[i]->next_task_array(QUICK_SORT,s->bsort[i]); + s->bsort[i]->set_inData(0,(memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num); s->bsort[i]->set_param(0,(memaddr)last_half_block_num); } for (int i = 0; i < half_num; i++) { - s->bsort[i]->wait_for(s->fsort[i]); - s->bsort[i]->wait_for(s->fsort[i+1]); - s->bsort[i]->no_auto_free(); - s->bsort[i]->spawn(); + task_array_b[i]->wait_for(task_array_f[i]); + task_array_b[i]->wait_for(task_array_f[i+1]); + task_array_b[i]->no_auto_free(); + task_array_b[i]->spawn_task_array(s->bsort[i]->next()); + task_array_b[i]->set_cpu(spe_cpu); + task_array_b[i]->flip(); + task_array_b[i]->spawn(); } } HTaskPtr restart = manager->create_task(SortSimple,0,0,0,0); restart->set_param(0,(memaddr)s); - if (!all) restart->wait_for(s->fsort[0]); + if (!all) restart->wait_for(task_array_f[0]); for (int i = 0; i < s->split_num; i++) { - s->fsort[i]->spawn(); + task_array_f[i]->spawn_task_array(s->fsort[i]->next()); + task_array_f[i]->set_cpu(spe_cpu); + task_array_f[i]->flip(); + task_array_f[i]->spawn(); } if (sort_count == 1) { // last loop wait for all task // we should not need this? for (int i = 0; i < half_num; i++) { - restart->wait_for(s->bsort[i]); - s->bsort[i]->auto_free(); + restart->wait_for(task_array_b[i]); + task_array_b[i]->auto_free(); } } restart->spawn();
--- a/example/many_task/sort.h Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/sort.h Thu Jan 31 18:47:13 2013 +0900 @@ -10,8 +10,8 @@ int data_length; int split_num; long cpuNum; - HTaskPtr *fsort; - HTaskPtr *bsort; + Task **fsort; + Task **bsort; DataPtr data; } Sort, *SortPtr;
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/example/many_task/sort_ta.cc Thu Jan 31 18:47:13 2013 +0900 @@ -0,0 +1,232 @@ +#include "TaskManager.h" +#include "SchedTask.h" +#include "sort.h" +#include "Func.h" +#include <string.h> + +extern int get_split_num(int len, int num); +extern int all; // allocate task at once +extern CPU_TYPE spe_cpu ; +extern int task_array_num; + +/** + * 一つの block にある data の数が MAX_BLOCK_SIZE 超えないような + * len の分割数を返す + * + * @param len sort する data の総数 + * @param num 使用する SPE の数 + * + * @return data の分割数 + * + * TODO: + * len が num 以下とか考えてません + */ +int +get_split_num(int len, int num) +{ + if (len / num < MAX_BLOCK_SIZE) { + return num; + } else { + // 切り上げ + return (len + MAX_BLOCK_SIZE - 1) / MAX_BLOCK_SIZE; + } +} + + +/** + * btask が全て終了したら、再び sort_start を実行する + * @param d 生成された btask の数 + */ + +SchedDefineTask1(SortSimple, sort_start ); + +static int +sort_start(SchedTask *manager, void *d, void *e) +{ + Sort *s = (Sort*)manager->get_param(0); + int half_num = s->split_num-1; + static int sort_count = s->split_num; // sort 完了に必要な回数 + + // 一つのタスクで sort する data 数 + int block_num = (s->data_length + s->split_num -1)/s->split_num; + int half_block_num = block_num/2; + + int last_block_num = s->data_length - (s->split_num-1)*block_num; + int last_half_block_num = half_block_num+(last_block_num/2); + + if (--sort_count < 0) { + return 0; + } + + if (task_array_num > 0) { + // run task array + HTask **task_array_f = (HTask**)manager->allocate(sizeof(HTask*)*s->split_num); + HTask **task_array_b = (HTask**)manager->allocate(sizeof(HTask*)*s->split_num); + + for (int i = 0; i < s->split_num;i++) { + task_array_f[i] = manager->create_task_array(QUICK_SORT, task_array_num,1,1,1); + s->fsort[i]=0; + } + for (int i = 0; i<half_num;i++) { + task_array_b[i] = manager->create_task_array(QUICK_SORT, task_array_num,1,1,1); + s->bsort[i]=0; + } + for (int i = 0; i < s->split_num-1; i++) { + s->fsort[i] = task_array_f[i]->next_task_array(QUICK_SORT,s->fsort[i]); + s->fsort[i]->set_param(0,(memaddr)block_num); + s->fsort[i]->set_inData(0,(memaddr)&s->data[i*block_num], sizeof(Data)*block_num); + if (i>0 && s->bsort[i-1]) { + task_array_f[i]->wait_for(task_array_b[i-1]); + } + if (i<s->split_num-2 && s->bsort[i]) { + task_array_f[i]->wait_for(task_array_b[i]); + } + } + + // 最後の block は端数なので last_block_num を使う + { + + int i = s->split_num-1; + + s->fsort[i] = task_array_f[i]->next_task_array(QUICK_SORT,s->fsort[i]); + s->fsort[i]->set_param(0,(memaddr)last_block_num); + s->fsort[i]->set_inData(0,(memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num); + if (i>0 && s->bsort[i-1]) { + task_array_f[i]->wait_for(task_array_b[i-1]); + } + } + + if (s->split_num > 1) { + + for (int i = 0; i < half_num-1; i++) { + if (s->bsort[i]) s->bsort[i]=0; + s->bsort[i] = task_array_b[i]->next_task_array(QUICK_SORT,s->bsort[i]); + s->bsort[i]->set_inData(0,(memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num); + s->bsort[i]->set_param(0,(memaddr)block_num); + } + + { + int i = half_num-1; + + if (s->bsort[i]) s->bsort[i]=0; + s->bsort[i] = task_array_b[i]->next_task_array(QUICK_SORT,s->bsort[i]); + s->bsort[i]->set_inData(0,(memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num); + s->bsort[i]->set_param(0,(memaddr)last_half_block_num); + } + + for (int i = 0; i < half_num; i++) { + task_array_b[i]->wait_for(task_array_f[i]); + task_array_b[i]->wait_for(task_array_f[i+1]); + task_array_b[i]->no_auto_free(); + task_array_b[i]->spawn_task_array(s->bsort[i]->next()); + task_array_b[i]->set_cpu(spe_cpu); + task_array_b[i]->flip(); + task_array_b[i]->spawn(); + } + } + + HTaskPtr restart = manager->create_task(SortSimple,0,0,0,0); + restart->set_param(0,(memaddr)s); + if (!all) restart->wait_for(task_array_f[0]); + for (int i = 0; i < s->split_num; i++) { + task_array_f[i]->spawn_task_array(s->fsort[i]->next()); + task_array_f[i]->set_cpu(spe_cpu); + task_array_f[i]->flip(); + task_array_f[i]->spawn(); + } + if (sort_count == 1) { + // last loop wait for all task + // we should not need this? + for (int i = 0; i < half_num; i++) { + restart->wait_for(task_array_b[i]); + task_array_b[i]->auto_free(); + } + } + restart->spawn(); + } else { + + for (int i = 0; i < s->split_num-1; i++) { + s->fsort[i] = manager->create_task(QUICK_SORT, + (memaddr)&s->data[i*block_num], sizeof(Data)*block_num, + (memaddr)&s->data[i*block_num], sizeof(Data)*block_num); + + s->fsort[i]->flip(); + + if (i>0 && s->bsort[i-1]) { + s->fsort[i]->wait_for(s->bsort[i-1]); + } + if (i<s->split_num-2 && s->bsort[i]) { + s->fsort[i]->wait_for(s->bsort[i]); + } + s->fsort[i]->set_cpu(spe_cpu); + s->fsort[i]->set_param(0,(memaddr)block_num); + } + + // 最後の block は端数なので last_block_num を使う + { + int i = s->split_num-1; + + s->fsort[i] = manager->create_task(QUICK_SORT, + (memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num, + (memaddr)&s->data[i*block_num], sizeof(Data)*last_block_num); + s->fsort[i]->flip(); + if (i>0 && s->bsort[i-1]) { + s->fsort[i]->wait_for(s->bsort[i-1]); + } + s->fsort[i]->set_cpu(spe_cpu); + s->fsort[i]->set_param(0,(memaddr)last_block_num); + } + + if (s->split_num > 1) { + + for (int i = 0; i < half_num-1; i++) { + if (s->bsort[i]) manager->free_htask(s->bsort[i]); + s->bsort[i] = manager->create_task(QUICK_SORT, + (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num, + (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*block_num); + s->bsort[i]->flip(); + s->bsort[i]->set_cpu(spe_cpu); + s->bsort[i]->set_param(0,(memaddr)block_num); + } + + { + int i = half_num-1; + + if (s->bsort[i]) manager->free_htask(s->bsort[i]); + s->bsort[i] = manager->create_task(QUICK_SORT, + (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num, + (memaddr)&s->data[i*block_num+half_block_num], sizeof(Data)*last_half_block_num); + s->bsort[i]->flip(); + s->bsort[i]->set_cpu(spe_cpu); + s->bsort[i]->set_param(0,(memaddr)last_half_block_num); + } + + for (int i = 0; i < half_num; i++) { + s->bsort[i]->wait_for(s->fsort[i]); + s->bsort[i]->wait_for(s->fsort[i+1]); + s->bsort[i]->no_auto_free(); + s->bsort[i]->spawn(); + } + } + + HTaskPtr restart = manager->create_task(SortSimple,0,0,0,0); + restart->set_param(0,(memaddr)s); + if (!all) restart->wait_for(s->fsort[0]); + for (int i = 0; i < s->split_num; i++) { + s->fsort[i]->spawn(); + } + if (sort_count == 1) { + // last loop wait for all task + // we should not need this? + for (int i = 0; i < half_num; i++) { + restart->wait_for(s->bsort[i]); + s->bsort[i]->auto_free(); + } + } + restart->spawn(); + } + return 0; +} + + +/* end */
--- a/example/many_task/task_init.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/many_task/task_init.cc Thu Jan 31 18:47:13 2013 +0900 @@ -19,5 +19,5 @@ SchedRegister(SortSimple); - SchedRegister(SortCompat); + // SchedRegister(SortCompat); }
--- a/example/word_count/a.txt Sun Jan 13 22:53:50 2013 +0900 +++ b/example/word_count/a.txt Thu Jan 31 18:47:13 2013 +0900 @@ -1,5 +1,6 @@ aaa bbb -aaa bbb ccc +aaa bbb cc +aaa dd aaa aaa aaa
--- a/example/word_count/gpu/Exec.cl Sun Jan 13 22:53:50 2013 +0900 +++ b/example/word_count/gpu/Exec.cl Thu Jan 31 18:47:13 2013 +0900 @@ -1,15 +1,13 @@ __kernel void run(__global int *data_count, - __global char *i_data, - __global unsigned long long *o_data) + __global void *rbuf, + __global void *wbuf) { /* * Todo:ここの書式もCerium側に合わせる。 - * 第2、第3引数をvoid *r_buf, void *w_bufにして、以下のような感じ - * __global char *i_data = (char *)r_buf; - * __global unsigned long long *o_data = (unsigned long long*)w_buf; - * __global unsigned long long *head_tail_flag = o_data +2; - */ + * 第2、第3引数をvoid *r_buf, void *w_bufにして、以下のような感じ*/ + __global char *i_data = (__global char *)rbuf; + __global unsigned long long *o_data = (__global unsigned long long*)wbuf; __global unsigned long long *head_tail_flag = o_data +2; int length = data_count[0]; int word_flag = 0;
--- a/example/word_count/main.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/word_count/main.cc Thu Jan 31 18:47:13 2013 +0900 @@ -6,6 +6,7 @@ #include <sys/stat.h> #include <fcntl.h> #include <unistd.h> +#include <sys/time.h> #include "TaskManager.h" #include "SchedTask.h" #include "Func.h" @@ -16,7 +17,9 @@ */ extern void task_init(); - +void TMend(TaskManager *); +static double st_time; +static double ed_time; int all = 0; int use_task_array = 1; int use_task_creater = 0; @@ -26,6 +29,13 @@ CPU_TYPE spe_cpu = SPE_ANY; const char *usr_help_str = "Usage: ./word_count [-a -c -s] [-cpu spe_num] [-file filename]\n"; +static double +getTime() { + struct timeval tv; + gettimeofday(&tv, NULL); + return tv.tv_sec + (double)tv.tv_usec*1e-6; +} + typedef struct { caddr_t file_mmap; off_t size; @@ -35,10 +45,8 @@ int command, int in_data_size, int out_data_size, void *in_data, void *out_data, SchedTask *manager, HTask *wait_i, HTask *wait_me) { - int in_task_size = 0; int out_task_size = 0; - int length = in_data_size/sizeof(char); if (in_total_size != 0) { in_task_size = in_total_size / in_data_size; @@ -89,7 +97,7 @@ } - + int length = in_data_size/sizeof(char); for (int j = 0; j < array_length; j++) { for (int k = 0; k < array_num; k++) { @@ -235,7 +243,7 @@ } } - //ここから + // ここから HTask **task_array = (HTask**)manager->allocate(sizeof(HTask*)*spe_num); Task **t_exec = (Task**)manager->allocate(sizeof(Task*)*spe_num); @@ -249,15 +257,14 @@ } } - for (int j = 0; j < array_task_num; j++) { for (int k = 0; k < spe_num; k++) { int a = w->task_spwaned++; - + if (w->size < size) size = w->size; + int length = size/sizeof(char); - int length = size/sizeof(char); t_exec[k] = task_array[k]->next_task_array(TASK_EXEC,t_exec[k]); t_exec[k]->set_param(0,(memaddr)length); t_exec[k]->set_inData(0,w->file_mmap + a*w->division_size, size); @@ -494,8 +501,16 @@ task_init(); run_start(manager, filename); - + st_time = getTime(); + manager->set_TMend(TMend); return 0; } +void +TMend(TaskManager *manager) +{ + ed_time = getTime(); + printf("Time: %0.6f\n",ed_time-st_time); +} + /* end */
--- a/example/word_count/ppe/Print.cc Sun Jan 13 22:53:50 2013 +0900 +++ b/example/word_count/ppe/Print.cc Thu Jan 31 18:47:13 2013 +0900 @@ -63,24 +63,24 @@ s->printf("start sum\n"); for (int i = 0; i < status_num; i++) { - word_data[i] = 0; + word_data[i] = 0; } for (int i = 0; i < out_task_num ; i++) { - word_data[0] += idata[i*w->out_size+0]; - word_data[1] += idata[i*w->out_size+1]; - unsigned long long *head_tail_flag = - &idata[i*w->out_size+2]; + word_data[0] += idata[i*w->out_size+0]; + word_data[1] += idata[i*w->out_size+1]; + unsigned long long *head_tail_flag = + &idata[i*w->out_size+2]; if((i!=out_task_num-1)&& (head_tail_flag[1] == 1) && (head_tail_flag[4] == 0)) { - flag_cal_sum++; + flag_cal_sum++; } } word_data[0] += flag_cal_sum; for (int i = status_num-1; i >=0; i--) { - s->printf("%llu ",word_data[i]); + s->printf("%llu ",word_data[i]); } //printf("j = %d\n",j);