私は OpenCL を学習しようとしています。これは非常に興味深いですが、学習曲線が急勾配です (少なくとも私にとっては)。
本に続いて: OpenCl in Action
私が持っているopenclデバイスに関する情報を提供する簡単なプログラムを書きましたが、まだコンテキストもカーネルも何もありません。
2 つの構造体を定義しました (オンラインのブログからのアイデア)。
注：ファイルの最後にある通知にジャンプしてください
デバイスの説明
typedef struct DeviceDesc{
cl_device_id deviceID;
cl_device_type Type;
char* TypeString;
char* Name;
char* Ext;
cl_bool Available;
cl_ulong GlobalMemSize;
cl_ulong LocalMemSize;
cl_uint Clock;
cl_uint ComputeUnits;
} DeviceDesc;
PlatformDesc
typedef struct PlatformDesc{
cl_platform_id pID;
char* pName;
char* pExt;
char* pVersion;
char* pVendor;
} PlatformDesc;
また、2つの列挙型を定義しました
ピン情報
enum pINFO{
name = 0,
vendor = 1,
ext = 2,
profile = 3,
version = 4,
}pINFO;
DINFO
enum dINFO{
dname = 0, //char*
davail = 1, //clbool
dext = 2, //char*
dgmemsize = 3, //ulong
dlmemsize = 4, //ulong
dmaxclock = 5, //uint
dcompunits = 6,//uint
}dINFO;
これらの関数も定義しました:
void cl_error_check(cl_int err,char* msg) {
if (!err==CL_SUCCESS) {
printf("%s\n",msg);
exit(1);
}
}
void cl_PFINFO(PlatformDesc* platform,enum pINFO itype,int i) {
size_t datasize;
char* data;
cl_int err;
//name = 0, //char*
//avail = 1, //clbool
//ext = 2, //char*
//gmemsize = 3, //ulong
//lmemsize = 4, //ulong
//maxclock = 5, //uint
//compunits = 6,//uint
switch(itype) {
case 0: //name
err = clGetPlatformInfo(platform[i].pID,CL_PLATFORM_NAME,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Plaform Name!");
data = (char*)malloc(sizeof(datasize));
clGetPlatformInfo(platform[i].pID,CL_PLATFORM_NAME,datasize,data,NULL);
platform[i].pName = data;
break;
case 1: //avail
err = clGetPlatformInfo(platform[i].pID,CL_PLATFORM_VENDOR,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Plaform Vendor!");
data = (char*)malloc(sizeof(datasize));
clGetPlatformInfo(platform[i].pID,CL_PLATFORM_VENDOR,datasize,data,NULL);
platform[i].pVendor = data;
break;
case 2: //ext
err = clGetPlatformInfo(platform[i].pID,CL_PLATFORM_EXTENSIONS,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Plaform Extensions!");
data = (char*)malloc(sizeof(datasize));
clGetPlatformInfo(platform[i].pID,CL_PLATFORM_EXTENSIONS,datasize,data,NULL);
platform[i].pExt = data;
break;
case 3: //globalmemorysize
break;
case 4: //localmemorysize
err = clGetPlatformInfo(platform[i].pID,CL_PLATFORM_VERSION,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Plaform Version!");
data = (char*)malloc(sizeof(datasize));
clGetPlatformInfo(platform[i].pID,CL_PLATFORM_VERSION,datasize,data,NULL);
platform[i].pVersion = data;
break;
break;
default:
printf("Unknown Platform Info Type! %d",itype);
exit(1);
}
}
void cl_DVINFO(DeviceDesc* dev,enum dINFO itype,int i) {
size_t datasize;
char* data;
cl_bool av;
cl_int err;
cl_ulong gm,lm;
cl_uint clock,units;
//name = 0, //char*
//avail = 1, //clbool
//ext = 2, //char*
//gmemsize = 3, //ulong
//lmemsize = 4, //ulong
//maxclock = 5, //uint
//compunits = 6,//uint
switch(itype) {
case 0: //name
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_NAME,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Device Name!");
data = (char*)malloc(sizeof(datasize));
clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_NAME,datasize,data,NULL);
dev[i].Name = data;
printf("\nDevice Name: %s",dev[i].Name);
break;
case 1: //avail
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_AVAILABLE,sizeof(cl_bool),&av,NULL);
cl_error_check(err,"Couldn't Obtain Device Availability!");
dev[i].Available = av;
printf("\nDevice Available: %d",dev[i].Available);
break;
case 2: //ext
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_EXTENSIONS,0,NULL,&datasize);
cl_error_check(err,"Couldn't Obtain Device Extensions!");
data = (char*)malloc(sizeof(datasize));
clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_EXTENSIONS,datasize,data,NULL);
dev[i].Ext = data;
break;
case 3: //globalmemorysize
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_GLOBAL_MEM_SIZE,sizeof(cl_ulong),&gm,NULL);
cl_error_check(err,"Couldn't Obtain Device Global Memory Size!");
dev[i].GlobalMemSize = gm;
printf("\nDevice Global Memory Size: %lu",dev[i].GlobalMemSize);
break;
case 4: //localmemorysize
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_LOCAL_MEM_SIZE,sizeof(cl_ulong),&lm,NULL);
cl_error_check(err,"Couldn't Obtain Device Local Memory Size!");
dev[i].LocalMemSize = lm;
printf("\nDevice Local Memory Size: %lu",dev[i].LocalMemSize);
break;
case 5: //Max Clock
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_MAX_CLOCK_FREQUENCY,sizeof(cl_uint),&clock,NULL);
cl_error_check(err,"Couldn't Obtain info on Device Max Clock Frequency!");
dev[i].Clock = clock;
printf("\nDevice Clock: %u",dev[i].Clock);
break;
case 6://Max Compte
err = clGetDeviceInfo(dev[i].deviceID,CL_DEVICE_MAX_COMPUTE_UNITS,sizeof(cl_uint),&units,NULL);
cl_error_check(err,"Couldn't Obtain info on Device Max Computing Units!");
dev[i].ComputeUnits = units;
printf("\nDevice Compute Units: %u",dev[i].ComputeUnits);
break;
default:
printf("Unknown Device Info Type! %d",itype);
exit(1);
}
}
私はこれがたくさんあることを知っていますが、コメントで ==> とマークされたコードの部分に到達するまで、私のプログラムは正常に実行されます。ここがメインです
int main(int argc, char** argv) {
/*
OPEN CL INIT AND STUFF
*/
cl_platform_id platformID;
cl_device_id *CPUID,*GPUID;
cl_uint num_devices,addr_data,num_platforms = 1;
cl_uint num_CPU,num_GPU;
cl_int err, i;
char dname_data[48],dext_data[4096];
int k;
PlatformDesc* platform = (PlatformDesc*)malloc(sizeof(PlatformDesc));
DeviceDesc* CPUS = NULL;
DeviceDesc* GPUS = NULL;
err = clGetPlatformIDs(num_platforms,&platformID,NULL);
cl_error_check(err,"Couldn't find any platforms!");
platform[0].pID = platformID;
//Get CPUs
err = clGetDeviceIDs(platform[0].pID,CL_DEVICE_TYPE_CPU,NULL,NULL,&num_CPU);
cl_error_check(err,"Error in finding CPUS!");
CPUS = (DeviceDesc*)malloc(num_CPU*sizeof(DeviceDesc));
CPUID = (cl_device_id*)malloc(num_CPU*sizeof(cl_device_id));
err = clGetDeviceIDs(platform[0].pID,CL_DEVICE_TYPE_CPU,num_CPU,CPUID,NULL);\
cl_error_check(err,"Error in Getting CPUS!");
for (k = 0;k<num_CPU;k++) {
CPUS[k].deviceID = CPUID[k];
CPUS[k].Type = CL_DEVICE_TYPE_CPU;
CPUS[k].TypeString = "CPU";
//get device info
int l;
for (l=0;l<7;l++)
cl_DVINFO(CPUS,l,k);
}
//Get GPUs
err = clGetDeviceIDs(platform[0].pID,CL_DEVICE_TYPE_GPU,NULL,NULL,&num_GPU);
cl_error_check(err,"Error in finding GPUS!");
GPUS = (DeviceDesc*)malloc(num_GPU*sizeof(DeviceDesc));
GPUID = (cl_device_id*)malloc(num_GPU*sizeof(cl_device_id));
err = clGetDeviceIDs(platform[0].pID,CL_DEVICE_TYPE_GPU,num_GPU,GPUID,NULL);\
cl_error_check(err,"Error in Getting GPUS!");
for (k = 0;k<num_GPU;k++) {
//==> Most of the crashing occurs here... if they occur
GPUS[k].deviceID = GPUID[k];
GPUS[k].Type = CL_DEVICE_TYPE_GPU;
GPUS[k].TypeString = "GPU";
int l;
for (l=0;l<7;l++)
cl_DVINFO(GPUS,l,k); //Here, i am getting the info "L" for the GPU K
}
//Ok Now get Platform info
cl_PFINFO(platform,name,0);
cl_PFINFO(platform,vendor,0);
cl_PFINFO(platform,ext,0);
cl_PFINFO(platform,version,0);
scanf("%d",&k); //THIS SCANF is just to see if it reaches this part or not, useless
//CLEANUP
free(GPUID);
free(CPUID);
free(CPUS);
free(GPUS);
return 0;
}
ここに私のマシンから得たサンプル出力があります
Device Name: AMD FX(tm)-8120 Eight-Core Processor
Device Available: 1
Device Global Memory Size: 2147483648
Device Local Memory Size: 32768
Device Clock: 4100
Device Compute Units: 8
Device Name: Tahiti
Device Available: 1
Device Global Memory Size: 2147483648
Device Local Memory Size: 32768
Device Clock: 950
Device Compute Units: 28
Device Name: Tahiti
Device Available: 1
Device Global Memory Size: 2147483648
Device Local Memory Size: 32768
Device Clock: 950
Device Compute Units: 28
Process returned -1073741819 (0xC0000005) execution time : 2.206 s
Press any key to continue.
2 番目のタヒチが特定されない場合があります。とにかく、私は0xC0000005を取得します（これはアクセス違反だと思います、別名ポインターと関係があります）
私の質問は、どこで間違ったのでしょうか:)。つまり、これができない場合、どうすればカーネルを処理できるのでしょうか :P
知らせ
リクエストにより、コードが失敗するセグメントを投稿します。
コードはここで失敗します
for (k = 0;k<num_CPU;k++) {
CPUS[k].deviceID = CPUID[k];
CPUS[k].Type = CL_DEVICE_TYPE_CPU;
CPUS[k].TypeString = "CPU";
//get device info
int l;
for (l=0;l<7;l++)
cl_DVINFO(CPUS,l,k);
}