#include <mpp/shmem.h>
#include <omp.h>
#include <cl.h>
#include <malloc.h>
float e;
// kernel
#define BLOCK (512)
const char *source =
"__kernel void sub1(__global float* fx,\
__global const float* fy,\
__local float* se, __global float* fe) {\
const unsigned int t = get_global_id(0);\
const unsigned int b = get_group_id(0);\
const unsigned block = 512;\
const unsigned int i = block*b+t;\
float e;\
/* do computation */\
fx[t] += ( fy[t+2] + fy[t] )*.5;\
e = fy[t+1] * fy[t+1];\
/* reduction */\
se[t] = e;\
barrier(CLK_LOCAL_MEM_FENCE);\
if (t<256) {\
se[t] += se[t+256];\
barrier(CLK_LOCAL_MEM_FENCE);\
}\
if (t<128) {\
se[t] += se[t+128];\
barrier(CLK_LOCAL_MEM_FENCE);\
}\
if (t<64) {\
se[t] += se[t+64];\
barrier(CLK_LOCAL_MEM_FENCE);\
}\
if (t<32) {\
se[t] += se[t+32];\
se[t] += se[t+16];\
se[t] += se[t+8];\
se[t] += se[t+4];\
se[t] += se[t+2];\
se[t] += se[t+1];\
}\
if (t==0)\
fe[b] = se[0];\
}";
int main(int argc, char *argv[]) {
int n = ...;
start_pes(0);
int nn = (n-1) / _num_pes();
int n_local0 = 1 + _my_pe() * nn;
int n_local1 = 1 + (_my_pe()+1) * nn;
// allocate only local part + ghost zone of the arrays x,y
float *x, *y;
x = (float*)shmalloc((n_local1 - n_local0 + 2)*sizeof(float));
y = (float*)shmalloc((n_local1 - n_local0 + 2)*sizeof(float));
x -= (n_local0 - 1);
y -= (n_local0 - 1);
shmem_barrier_all();
... // fill x, y
// fill ghost zone
if (_my_pe() > 0)
shmem_float_put(&y[n_local1], &y[n_local0], 1, _my_pe()-1);
if (_my_pe() < _num_pes()-1)
shmem_float_put(&y[n_local0-1], &y[n_local1-1], 1, _my_pe()+1);
shmem_barrier_all();
e = 0;
#pragma omp parallel
{
int p = omp_get_thread_num();
int num = omp_get_num_threads();
// allocate GPU
cl_context ct = clCreateContextFromType(0, CL_DEVICE_TYPE_GPU, 0, 0, 0);
size_t ctsize;
clGetContextInfo(ct, CL_CONTEXT_DEVICES, 0, 0, &ctsize);
cl_device_id *aDevices = (cl_device_id*)malloc(ctsize);
clGetContextInfo(ct, CL_CONTEXT_DEVICES, ctsize, aDevices, 0);
// compile kernel
cl_program prog = clCreateProgramWithSource(ct, 1, &source, 0, 0);
clBuildProgram(prog, 0, 0, 0, 0, 0);
cl_kernel kern = clCreateKernel(prog, "sub1", 0);
int n0 = 1+((n_local1-n_local0)*p)/num;
int n1 = 1+((n_local1-n_local0)*(p+1))/num;
// pick GPU
cl_command_queue queue = clCreateCommandQueue(ct, aDevices[p], 0, 0);
// allocate GPU memory
cl_mem fx = clCreateBuffer(ct, CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,
(n1-n0)*sizeof(cl_float), &x[n0], 0);
cl_mem fy = clCreateBuffer(ct, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
(n1-n0+2)*sizeof(cl_float), &y[n0-1], 0);
cl_mem se = clCreateBuffer(ct, CL_MEM_READ_WRITE,
BLOCK*sizeof(cl_float), 0, 0);
cl_mem fe = clCreateBuffer(ct, CL_MEM_WRITE_ONLY,
(n1-n0)/BLOCK*sizeof(cl_float), 0, 0);
clSetKernelArg(kern, 0, sizeof(cl_mem), (void *)&fx);
clSetKernelArg(kern, 1, sizeof(cl_mem), (void *)&fx);
clSetKernelArg(kern, 2, sizeof(cl_mem), (void *)&se);
clSetKernelArg(kern, 3, sizeof(cl_mem), (void *)&fe);
float *d = new float[(n1-n0)/BLOCK];
// call GPU
const unsigned int size = BLOCK;
const unsigned int dim = n1-n0+2;
clEnqueueNDRangeKernel(queue, kern, 1, 0, &dim, &size, 0, 0, 0);
// copy to host memory
clEnqueueReadBuffer(queue, fx, CL_TRUE, 0,
(n1-n0) * sizeof(cl_float), &x[n0], 0, 0, 0);
clEnqueueReadBuffer(queue, fe, CL_TRUE, 0,
(n1-n0) * sizeof(cl_float), d, 0, 0, 0);
float ee = 0;
for (int i=0; i<(n1-n0)/BLOCK; ++i)
ee += d[i];
#pragma omp atomic
e += ee;
delete[] d;
// release GPU memory
clReleaseMemObject(fx);
clReleaseMemObject(fy);
clReleaseMemObject(se);
clReleaseMemObject(fe);
}
static float work[_SHMEM_REDUCE_MIN_WRKDATA_SIZE];
static long sync[_SHMEM_REDUCE_MIN_WRKDATA_SIZE];
static float el, es;
el = e;
shmem_float_sum_to_all(&es, &el, 1,
0, 0, _num_pes(), work, sync);
e = es;
... // output x, e
x += (n_local0 - 1);
y += (n_local0 - 1);
shfree(x);
shfree(y);
return 0;
}
|