Mercurial > hg > Game > Cerium
changeset 1579:7418c7aef534 draft
fft
| author | Shohei KOKUBO <e105744@ie.u-ryukyu.ac.jp> |
|---|---|
| date | Mon, 25 Mar 2013 23:27:04 +0900 |
| parents | a039ff7b0a45 |
| children | 806b4658ced6 |
| files | example/fft/main.cc example/fft/ppe/bitReverse.cc example/fft/ppe/butterfly.cc example/fft/ppe/setid.cc example/fft/ppe/spinFact.cc example/multiply/main.cc example/word_count/main.cc |
| diffstat | 7 files changed, 81 insertions(+), 100 deletions(-) [+] |
line wrap: on
line diff
--- a/example/fft/main.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/fft/main.cc Mon Mar 25 23:27:04 2013 +0900 @@ -26,7 +26,6 @@ static double ed_time; void TMend(TaskManager *); int ndrange_flag; - cl_device_id device_id = NULL; cl_context context = NULL; cl_command_queue queue = NULL; @@ -73,7 +72,7 @@ int fftCore(TaskManager *manager,cl_float2 *dst, cl_float2 *src, cl_float2 *spin, int m_, enum Mode direction) { - int *iter = new int[1]; + int iter; unsigned int* flag = new unsigned int[1]; switch (direction) { case forward:flag[0] = 0x00000000; break; @@ -86,54 +85,48 @@ size_t gws[2],lws[2]; int length_dst = n[0]*n[0]; int length_src = n[0]*n[0]; - HTask *brev = manager->create_task(BIT_REVERSE); + cl_uint dimension = 2; + HTask* brev; + int i,j; setWorkSize(gws,lws,n[0],n[0]); - cl_uint dimension = 2; - - brev->set_param(0,(memaddr)length_src); - brev->set_param(1,(memaddr)dimension); - brev->set_param(2,(memaddr)gws[0]); - brev->set_param(3,(memaddr)gws[1]); - brev->set_param(4,(memaddr)lws[0]); - brev->set_param(5,(memaddr)lws[1]); - brev->set_inData(0, src, length_src*sizeof(cl_float2)); - brev->set_inData(1, m,sizeof(int)); - brev->set_inData(2, n,sizeof(int)); - brev->set_outData(0, dst, length_dst*sizeof(cl_float2)); - brev->set_cpu(spe_cpu); - brev->nd_range(); + for(i=0;i<gws[0];i++){ + for(j=0;j<gws[1];j++){ + brev = manager->create_task(BIT_REVERSE); + brev->set_param(0,(memaddr)length_src); + brev->set_param(1,(memaddr)i); + brev->set_param(2,(memaddr)j); + brev->set_inData(0, src, length_src*sizeof(cl_float2)); + brev->set_inData(1, m,sizeof(int)); + brev->set_inData(2, n,sizeof(int)); + brev->set_outData(0, dst, length_dst*sizeof(cl_float2)); + brev->set_cpu(spe_cpu); + brev->spawn(); + } + } - HTaskPtr *bfly = (HTask**)manager->allocate(sizeof(HTask*)*m_); + HTask* bfly; setWorkSize(gws,lws,n[0]/2,n[0]); - for (iter[0]=1; iter[0]<=m_;iter[0]++) { - bfly[iter[0]-1] = manager->create_task(BUTTERFLY); - bfly[iter[0]-1]->set_param(0,(memaddr)length_dst); - bfly[iter[0]-1]->set_param(1,(memaddr)dimension); - bfly[iter[0]-1]->set_param(2,(memaddr)gws[0]); - bfly[iter[0]-1]->set_param(3,(memaddr)gws[1]); - bfly[iter[0]-1]->set_param(4,(memaddr)lws[0]); - bfly[iter[0]-1]->set_param(5,(memaddr)lws[1]); - bfly[iter[0]-1]->set_inData(0, dst, length_dst*sizeof(cl_float2)); - bfly[iter[0]-1]->set_inData(1, spin, sizeof(cl_float2)*(n[0]/2)); - bfly[iter[0]-1]->set_inData(2, m,sizeof(int)); - bfly[iter[0]-1]->set_inData(3, n,sizeof(int)); - bfly[iter[0]-1]->set_inData(4, iter,sizeof(int)); - bfly[iter[0]-1]->set_inData(5, flag,sizeof(int)); - bfly[iter[0]-1]->set_outData(0,dst,length_dst*sizeof(cl_float2)); - bfly[iter[0]-1]->set_cpu(spe_cpu); - bfly[iter[0]-1]->nd_range(); - bfly[iter[0]-1]->flip(); - if (iter[0]==1) { - bfly[iter[0]-1]->wait_for(brev); - } else { - bfly[iter[0]-1]->wait_for(bfly[iter[0]-2]); + for (iter=1; iter<=m_;iter++) { + for(i=0;i<gws[0];i++){ + for(j=0;i<gws[1];j++){ + bfly = manager->create_task(BUTTERFLY); + bfly->set_param(0,(memaddr)length_dst); + bfly->set_param(1,(memaddr)iter); + bfly->set_param(2,(memaddr)i); + bfly->set_param(3,(memaddr)j); + bfly->set_inData(0, dst, length_dst*sizeof(cl_float2)); + bfly->set_inData(1, spin, sizeof(cl_float2)*(n[0]/2)); + bfly->set_inData(2, m,sizeof(int)); + bfly->set_inData(3, n,sizeof(int)); + bfly->set_inData(4, flag,sizeof(int)); + bfly->set_outData(0,dst,length_dst*sizeof(cl_float2)); + bfly->set_cpu(spe_cpu); + bfly->wait_for(brev); + bfly->spawn(); + } } } - brev->spawn(); - for (int i=0;i<m_;i++) { - bfly[i]->spawn(); - } if (direction == inverse) { HTask *norm = manager->create_task(NORMALIZATION); @@ -149,7 +142,7 @@ norm->set_cpu(spe_cpu); norm->nd_range(); norm->flip(); - norm->wait_for(bfly[m[0]-1]); + norm->wait_for(bfly); norm->spawn(); } return 0; @@ -189,6 +182,7 @@ cl_float2 *xm = (cl_float2 *)malloc(n[0] * n[0] * sizeof(cl_float2)); cl_float2 *rm = (cl_float2 *)malloc(n[0] * n[0] * sizeof(cl_float2)); cl_float2 *wm = (cl_float2 *)malloc(n[0] / 2 * sizeof(cl_float2)); + int i,j; /* * [cl_float2] * typedef union @@ -204,39 +198,27 @@ * #endif * } cl_float2; */ - int i, j; - for (i=0; i < n[0]; i++) { + for (i=0; i<n[0]; i++) { for (j=0; j < n[0]; j++) { ((float*)xm)[(2*n[0]*j)+2*i+0] = (float)ipgm.buf[n[0]*j+i]; ((float*)xm)[(2*n[0]*j)+2*i+1] = (float)0; } } - - HTask* setid = manager->create_task(SET_ID); - setid->set_cpu(spe_cpu); - setid->spawn(); - - // Create spin factor - HTask *sfac = manager->create_task(SPIN_FACT); - int length_w = n[0] / 2; + + // Create spin factor setWorkSize(gws,lws,n[0]/2,1); // Todo:setWorkSize(ndr,n[0]/2,1);でできるように - - ND_RANGE_T *ndr = new ND_RANGE_T; - ndr->dimension = 1; - ndr->gws[0] = gws[0]; - ndr->gws[1] = gws[1]; - ndr->lws[0] = lws[0]; - ndr->lws[1] = lws[1]; - - sfac->set_param(0,ndr); - sfac->set_param(1, (memaddr)length_w); - sfac->set_inData(0, n, sizeof(int)); - sfac->set_outData(0, wm, length_w*sizeof(cl_float2)); - sfac->set_cpu(spe_cpu); - sfac->nd_range(); - sfac->wait_for(setid); - sfac->spawn(); - + int length_w = n[0] / 2; + HTask* sfac; + for(i=0;i<gws[0];i++){ + sfac = manager->create_task(SPIN_FACT); + sfac->set_param(0, (memaddr)length_w); + sfac->set_param(1,(memaddr)i); + sfac->set_inData(0, n, sizeof(int)); + sfac->set_outData(0, wm, length_w*sizeof(cl_float2)); + sfac->set_cpu(spe_cpu); + sfac->nd_range(); + sfac->spawn(); + } // Butterfly Operation fftCore(manager, rm, xm, wm, m[0], forward);
--- a/example/fft/ppe/bitReverse.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/fft/ppe/bitReverse.cc Mon Mar 25 23:27:04 2013 +0900 @@ -14,13 +14,12 @@ cl_float2* src = (cl_float2*)s->get_inputAddr(0); int* m = (int*)s->get_inputAddr(1); int* n = (int*)s->get_inputAddr(2); - cl_float2* dst = (cl_float2*)s->get_outputAddr(0); - - unsigned int* gid = (unsigned int*)s->global_get(FIRSTID); - unsigned int* nid = (unsigned int*)s->global_get(SECONDID); - - unsigned int j = gid[0]; + + unsigned long gid = (unsigned long)s->get_param(1); + unsigned long nid = (unsigned long)s->get_param(2); + unsigned int j = gid; + j = (j & 0x55555555) << 1 | (j & 0xAAAAAAAA) >> 1; j = (j & 0x33333333) << 2 | (j & 0xCCCCCCCC) >> 2; j = (j & 0x0F0F0F0F) << 4 | (j & 0xF0F0F0F0) >> 4; @@ -29,7 +28,7 @@ j >>= (32-m[0]); - dst[nid[0]*n[0]+j] = src[nid[0]*n[0]+gid[0]]; + dst[nid*n[0]+j] = src[nid*n[0]+gid]; return 0; }
--- a/example/fft/ppe/butterfly.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/fft/ppe/butterfly.cc Mon Mar 25 23:27:04 2013 +0900 @@ -13,21 +13,21 @@ cl_float2* x_in = (cl_float2*)s->get_inputAddr(0); cl_float2* w = (cl_float2*)s->get_inputAddr(1); int* n = (int*)s->get_inputAddr(3); - int* iter = (int*)s->get_inputAddr(4); - unsigned int* flag = (unsigned int*)s->get_inputAddr(5); + unsigned int* flag = (unsigned int*)s->get_inputAddr(4); cl_float2* x_out = (cl_float2*)s->get_outputAddr(0); - unsigned int* gid = (unsigned int*)s->global_get(FIRSTID); - unsigned int* nid = (unsigned int*)s->global_get(SECONDID); + long iter = (long)s->get_inputAddr(1); + unsigned long gid = (unsigned long)s->get_param(2); + unsigned long nid = (unsigned long)s->get_param(3); - int butterflySize = 1 << (iter[0]-1); - int butterflyGrpDist = 1 << iter[0]; - int butterflyGrpNum = n[0] >> iter[0]; - int butterflyGrpBase = (gid[0] >> (iter[0]-1))*(butterflyGrpDist); - int butterflyGrpOffset = gid[0] & (butterflySize-1); + int butterflySize = 1 << (iter-1); + int butterflyGrpDist = 1 << iter; + int butterflyGrpNum = n[0] >> iter; + int butterflyGrpBase = (gid >> (iter-1))*(butterflyGrpDist); + int butterflyGrpOffset = gid & (butterflySize-1); - int a = nid[0] * n[0] + butterflyGrpBase + butterflyGrpOffset; + int a = nid * n[0] + butterflyGrpBase + butterflyGrpOffset; int b = a + butterflySize; int l = butterflyGrpNum * butterflyGrpOffset;
--- a/example/fft/ppe/setid.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/fft/ppe/setid.cc Mon Mar 25 23:27:04 2013 +0900 @@ -7,7 +7,8 @@ setid(SchedTask* s,void* rbuf,void* wbuf) { s->global_alloc(FIRSTID,sizeof(int)); + s->global_alloc(SECONDID,sizeof(int)); - + return 0; }
--- a/example/fft/ppe/spinFact.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/fft/ppe/spinFact.cc Mon Mar 25 23:27:04 2013 +0900 @@ -19,14 +19,13 @@ cl_float2* w = (cl_float2*)s->get_output(wbuf,0); - unsigned int* i = (unsigned int*)s->global_get(FIRSTID); + unsigned long i = (unsigned long)s->get_param(1); cl_float2 angle; - printf("%d\n",((unsigned int*)s->global_get(SECONDID))[0]); - angle.x = (float)(2*i[0]*PI/(float)n[0]); - angle.y = (float)((2*i[0]*PI/(float)n[0]) + PI_2); + angle.x = (float)(2*i*PI/(float)n[0]); + angle.y = (float)((2*i*PI/(float)n[0]) + PI_2); - w[i[0]].x = cos(angle.x); - w[i[0]].y = cos(angle.y); + w[i].x = cos(angle.x); + w[i].y = cos(angle.y); return 0; }
--- a/example/multiply/main.cc Sun Mar 24 21:13:10 2013 +0900 +++ b/example/multiply/main.cc Mon Mar 25 23:27:04 2013 +0900 @@ -75,10 +75,10 @@ // set ND_RANGE parameter // 初期値は全部1になっている - ndr = new ND_RANGE_T; + // ndr = new ND_RANGE_T; - ndr->gws[0] = length; - manager->set_NDRange(ndr); + // ndr->gws[0] = length; + // manager->set_NDRange(ndr); multiply = manager->create_task(MULTIPLY_TASK); multiply->nd_range();