117         template <int nblocks> // Number of histogram blocks processed by single GPU thread block
131             float* final_hist = smem + cnbins * 48 * nblocks;  in compute_hists_kernel_many_blocks()
147                     hist[bin_id * 48 * nblocks] = 0.f;  in compute_hists_kernel_many_blocks()
168                     hist[bin.x * 48 * nblocks] += gaussian * interp_weight * vote.x;  in compute_hists_kernel_many_blocks()
169                     hist[bin.y * 48 * nblocks] += gaussian * interp_weight * vote.y;  in compute_hists_kernel_many_blocks()
173                 for (int bin_id = 0; bin_id < cnbins; ++bin_id, hist_ += 48 * nblocks)  in compute_hists_kernel_many_blocks()
199             const int nblocks = 1;  in compute_hists()  local
206             dim3 grid(divUp(img_block_width, nblocks), img_block_height);  in compute_hists()
207             dim3 threads(32, 2, nblocks);  in compute_hists()
209             cudaSafeCall(cudaFuncSetCacheConfig(compute_hists_kernel_many_blocks<nblocks>,  in compute_hists()
215 …   int hists_size = (nbins * CELLS_PER_BLOCK_X * CELLS_PER_BLOCK_Y * 12 * nblocks) * sizeof(float);  in compute_hists()
216 …  int final_hists_size = (nbins * CELLS_PER_BLOCK_X * CELLS_PER_BLOCK_Y * nblocks) * sizeof(float);  in compute_hists()
218             compute_hists_kernel_many_blocks<nblocks><<<grid, threads, smem>>>(  in compute_hists()
262                   int nblocks> // Number of block hisograms processed by one GPU thread block
274             __shared__ float sh_squares[nthreads * nblocks];  in normalize_hists_kernel_many_blocks()
298             const int nblocks = 1;  in normalize_hists()  local
302             dim3 threads(nthreads, 1, nblocks);  in normalize_hists()
306             dim3 grid(divUp(img_block_width, nblocks), img_block_height);  in normalize_hists()
309 …normalize_hists_kernel_many_blocks<32, nblocks><<<grid, threads>>>(block_hist_size, img_block_widt…  in normalize_hists()
311 …normalize_hists_kernel_many_blocks<64, nblocks><<<grid, threads>>>(block_hist_size, img_block_widt…  in normalize_hists()
313 …normalize_hists_kernel_many_blocks<64, nblocks><<<grid, threads>>>(block_hist_size, img_block_widt…  in normalize_hists()
315 …normalize_hists_kernel_many_blocks<256, nblocks><<<grid, threads>>>(block_hist_size, img_block_wid…  in normalize_hists()
317 …normalize_hists_kernel_many_blocks<512, nblocks><<<grid, threads>>>(block_hist_size, img_block_wid…  in normalize_hists()
333                  int nblocks>  // Number of histogram block processed by single GPU thread block
355            __shared__ float products[nthreads * nblocks];  in compute_confidence_hists_kernel_many_blocks()
371            const int nblocks = 1;  in compute_confidence_hists()  local
378            dim3 threads(nthreads, 1, nblocks);  in compute_confidence_hists()
379            dim3 grid(divUp(img_win_width, nblocks), img_win_height);  in compute_confidence_hists()
381 …cudaSafeCall(cudaFuncSetCacheConfig(compute_confidence_hists_kernel_many_blocks<nthreads, nblocks>,  in compute_confidence_hists()
386            compute_confidence_hists_kernel_many_blocks<nthreads, nblocks><<<grid, threads>>>(  in compute_confidence_hists()
395                   int nblocks>  // Number of histogram block processed by single GPU thread block
417             __shared__ float products[nthreads * nblocks];  in classify_hists_kernel_many_blocks()
433             const int nblocks = 1;  in classify_hists()  local
440             dim3 threads(nthreads, 1, nblocks);  in classify_hists()
441             dim3 grid(divUp(img_win_width, nblocks), img_win_height);  in classify_hists()
443 …cudaSafeCall(cudaFuncSetCacheConfig(classify_hists_kernel_many_blocks<nthreads, nblocks>, cudaFunc…  in classify_hists()
446             classify_hists_kernel_many_blocks<nthreads, nblocks><<<grid, threads>>>(  in classify_hists()