Mercurial > hg > Game > Cerium

changeset 1519:9a5f87f4b60f draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
add spe_cpu for gpu
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Sat, 10 Nov 2012 18:21:16 +0900
parents 940ffd32e5bd
children 031f26b15ae6
files TaskManager/Gpu/GpuScheduler.cc example/many_task/Makefile.gpu example/many_task/main.cc example/many_task/sort-compat.cc example/many_task/sort.cc example/many_task/task_init.cc
diffstat 6 files changed, 33 insertions(+), 16 deletions(-) [+]
line wrap: on
line diff
--- a/TaskManager/Gpu/GpuScheduler.cc	Sat Nov 10 11:15:00 2012 +0900
+++ b/TaskManager/Gpu/GpuScheduler.cc	Sat Nov 10 18:21:16 2012 +0900
@@ -175,29 +175,29 @@
     const char *filename = (const char *)task_list[cmd].gputask->kernel;
     const char *functionname = task_list[cmd].name;
 
-    int fp;
+    int fd;
     char *source_str;
     size_t source_size;
 
-    fp = open(filename, O_RDONLY);
+    fd = open(filename, O_RDONLY);
 
-    if (!fp) {
-        fprintf(stderr, "Failed to load kernel.\n");
+    if (fd<0) {
+        fprintf(stderr, "Failed to load kernel %s.\n",filename);
         exit(1);
     }
 
     struct stat stats;
-    fstat(fp,&stats);
+    fstat(fd,&stats);
     off_t size = stats.st_size;
 
-    if (!size) {
+    if (size<=0) {
         fprintf(stderr, "Failed to load kernel.\n");
         exit(1);
     }
 
     source_str = (char*)alloca(size);
-    source_size = read(fp, source_str, size);
-    close(fp);
+    source_size = read(fd, source_str, size);
+    close(fd);
 
     cl_program program =
         clCreateProgramWithSource(context, 1,
--- a/example/many_task/Makefile.gpu	Sat Nov 10 11:15:00 2012 +0900
+++ b/example/many_task/Makefile.gpu	Sat Nov 10 18:21:16 2012 +0900
@@ -11,6 +11,8 @@
 TASK_SRCS = $(filter-out $(TASK_DIR)/$(TASK_SRCS_EXCLUDE),$(TASK_SRCS_TMP))
 TASK_OBJS = $(TASK_SRCS:.cc=.o)
 
+CFLAGS += -D__CERIUM_GPU__
+
 LIBS += -lGpuManager -framework opencl `sdl-config --libs`
 
 .SUFFIXES: .cc .o
--- a/example/many_task/main.cc	Sat Nov 10 11:15:00 2012 +0900
+++ b/example/many_task/main.cc	Sat Nov 10 18:21:16 2012 +0900
@@ -25,6 +25,7 @@
 static double ed_time;
 
 static int length = 1200;
+CPU_TYPE spe_cpu = SPE_ANY;
 
 // prototype
 void TMend(TaskManager *);
@@ -53,6 +54,9 @@
         if (strcmp(argv[i], "-a") == 0 ) {
             all = 1;
         }
+        if (strcmp(argv[i], "-g") == 0 ) {
+            spe_cpu = GPU_0;
+        }
         if (strcmp(argv[i], "-c") == 0 ) {
             sort_task = SortCompat;
         }
--- a/example/many_task/sort-compat.cc	Sat Nov 10 11:15:00 2012 +0900
+++ b/example/many_task/sort-compat.cc	Sat Nov 10 18:21:16 2012 +0900
@@ -6,6 +6,7 @@
 
 extern void check_data();
 extern int all;  // allocate task at once
+extern CPU_TYPE spe_cpu; 
 
 SchedDefineTask1(SortCompat, sort_start_compat );
 
@@ -37,7 +38,7 @@
 	if (i<s->split_num-2 && s->bsort[i]) {
 	    s->fsort[i]->wait_for(s->bsort[i]);
 	}
-	s->fsort[i]->set_cpu(SPE_ANY);
+	s->fsort[i]->set_cpu(spe_cpu);
     }
 
     // 最後の block は端数なので last_block_num を使う
@@ -50,7 +51,7 @@
 	if (i>0 && s->bsort[i-1]) {
 	    s->fsort[i]->wait_for(s->bsort[i-1]);
 	}
-	s->fsort[i]->set_cpu(SPE_ANY);
+	s->fsort[i]->set_cpu(spe_cpu);
    }
 
     if (s->split_num > 1) {
@@ -62,7 +63,7 @@
 				 sizeof(Data)*block_num);
 	    s->bsort[i]->set_outData(0,&s->data[i*block_num+half_block_num],
 				  sizeof(Data)*block_num);
-	    s->bsort[i]->set_cpu(SPE_ANY);
+	    s->bsort[i]->set_cpu(spe_cpu);
 	}
 
 	{
@@ -74,7 +75,7 @@
 				 sizeof(Data)*last_half_block_num);
 	    s->bsort[i]->set_outData(0,&s->data[i*block_num+half_block_num],
 				  sizeof(Data)*last_half_block_num);
-	    s->bsort[i]->set_cpu(SPE_ANY);	
+	    s->bsort[i]->set_cpu(spe_cpu);	
 	}
 	
 	for (int i = 0; i < half_num; i++) {
--- a/example/many_task/sort.cc	Sat Nov 10 11:15:00 2012 +0900
+++ b/example/many_task/sort.cc	Sat Nov 10 18:21:16 2012 +0900
@@ -6,6 +6,7 @@
 
 extern int get_split_num(int len, int num);
 extern int all;  // allocate task at once
+extern CPU_TYPE spe_cpu ;
 
 /**
  * 一つの block にある data の数が MAX_BLOCK_SIZE 超えないような
@@ -71,7 +72,7 @@
         if (i<s->split_num-2 && s->bsort[i]) {
             s->fsort[i]->wait_for(s->bsort[i]);
         }
-        s->fsort[i]->set_cpu(SPE_ANY);
+        s->fsort[i]->set_cpu(spe_cpu);
         s->fsort[i]->set_param(0,(memaddr)block_num);
     }
 
@@ -86,7 +87,7 @@
         if (i>0 && s->bsort[i-1]) {
             s->fsort[i]->wait_for(s->bsort[i-1]);
         }
-        s->fsort[i]->set_cpu(SPE_ANY);
+        s->fsort[i]->set_cpu(spe_cpu);
         s->fsort[i]->set_param(0,(memaddr)last_block_num);
    }
 
@@ -98,7 +99,7 @@
                 (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_ANY);
+            s->bsort[i]->set_cpu(spe_cpu);
             s->bsort[i]->set_param(0,(memaddr)block_num);
         }
 
@@ -110,7 +111,7 @@
                 (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_ANY);
+            s->bsort[i]->set_cpu(spe_cpu);
             s->bsort[i]->set_param(0,(memaddr)last_half_block_num);
         }
 
--- a/example/many_task/task_init.cc	Sat Nov 10 11:15:00 2012 +0900
+++ b/example/many_task/task_init.cc	Sat Nov 10 18:21:16 2012 +0900
@@ -1,14 +1,23 @@
 #include "Func.h"
 #include "Scheduler.h"
+#include "GpuScheduler.h"
 
+#ifndef __CERIUM_GPU__
 SchedExternTask(QuickSort);
+#endif // __CERIUM_GPU__
 SchedExternTask(SortSimple);
 SchedExternTask(SortCompat);
 
 void
 task_init(void)
 {
+#ifdef __CERIUM_GPU__
+    GpuSchedRegister(QUICK_SORT, "sort.cl", "sort");
+#else
     SchedRegisterTask(QUICK_SORT, QuickSort);
+#endif
+
+
     SchedRegister(SortSimple);
     SchedRegister(SortCompat);
 }