Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

ARIMA - Kalman loop rewrite: single megakernel instead of host loop #4006

Merged
merged 17 commits into from
Jul 12, 2021
Merged

ARIMA - Kalman loop rewrite: single megakernel instead of host loop #4006

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
8edacaf
Initial rewrite of the Kalman filter
Nyrio Jun 16, 2021
1f1bce5
Remove unused sparse matrices
Nyrio Jun 16, 2021
8a0ac86
More efficient block-level gemv
Nyrio Jun 17, 2021
a4f8fa2
Shared mem Z and alpha
Nyrio Jun 17, 2021
c21b0a0
Shared mem K
Nyrio Jun 18, 2021
b1a27f6
Rewrite loads for block gemm with support for vectorization
Nyrio Jun 21, 2021
ffdb182
Tune Kalman loop gemm policy
Nyrio Jun 22, 2021
9c7200a
Clean-up, add docs and add test
Nyrio Jun 23, 2021
eeada6f
Merge branch 'branch-21.08' into opt-kalman-loop
Nyrio Jun 24, 2021
1aa9b85
Merge branch 'branch-21.08' into opt-kalman-loop
Nyrio Jun 30, 2021
c07967c
Test improvements + small fixes
Nyrio Jul 5, 2021
5f048b4
Test multiple policies for block gemm and gemv
Nyrio Jul 5, 2021
5e84fcf
Overlap host and device computations in block prims tests
Nyrio Jul 5, 2021
3073f8a
Test with and without preload for gemv and xAx'
Nyrio Jul 5, 2021
c4685f8
Include style fix
Nyrio Jul 5, 2021
e1e5e01
Merge branch 'branch-21.08' into opt-kalman-loop
Nyrio Jul 9, 2021
b07d5b1
Format block linalg prims and corresponding unit test according to ne…
Nyrio Jul 12, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 3 additions & 6 deletions cpp/include/cuml/tsa/arima_common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,14 +210,14 @@ struct ARIMAMemory {
T *params_mu, *params_ar, *params_ma, *params_sar, *params_sma, *params_sigma2, *Tparams_mu,
*Tparams_ar, *Tparams_ma, *Tparams_sar, *Tparams_sma, *Tparams_sigma2, *d_params, *d_Tparams,
*Z_dense, *R_dense, *T_dense, *RQR_dense, *RQ_dense, *P_dense, *alpha_dense, *ImT_dense,
*ImT_inv_dense, *T_values, *v_tmp_dense, *m_tmp_dense, *K_dense, *TP_dense, *vs, *y_diff,
*loglike, *loglike_base, *loglike_pert, *x_pert, *F_buffer, *sumLogF_buffer, *sigma2_buffer,
*ImT_inv_dense, *v_tmp_dense, *m_tmp_dense, *K_dense, *TP_dense, *vs, *y_diff, *loglike,
*loglike_base, *loglike_pert, *x_pert, *F_buffer, *sumLogF_buffer, *sigma2_buffer,
*I_m_AxA_dense, *I_m_AxA_inv_dense, *Ts_dense, *RQRs_dense, *Ps_dense;
T **Z_batches, **R_batches, **T_batches, **RQR_batches, **RQ_batches, **P_batches,
**alpha_batches, **ImT_batches, **ImT_inv_batches, **v_tmp_batches, **m_tmp_batches,
**K_batches, **TP_batches, **I_m_AxA_batches, **I_m_AxA_inv_batches, **Ts_batches,
**RQRs_batches, **Ps_batches;
int *T_col_index, *T_row_index, *ImT_inv_P, *ImT_inv_info, *I_m_AxA_P, *I_m_AxA_info;
int *ImT_inv_P, *ImT_inv_info, *I_m_AxA_P, *I_m_AxA_info;

size_t size;

Expand Down Expand Up @@ -281,9 +281,6 @@ struct ARIMAMemory {
append_buffer<assign>(ImT_inv_batches, batch_size);
append_buffer<assign>(ImT_inv_P, r * batch_size);
append_buffer<assign>(ImT_inv_info, batch_size);
append_buffer<assign>(T_values, rd * rd * batch_size);
append_buffer<assign>(T_col_index, rd * rd);
append_buffer<assign>(T_row_index, rd + 1);
append_buffer<assign>(v_tmp_dense, rd * batch_size);
append_buffer<assign>(v_tmp_batches, batch_size);
append_buffer<assign>(m_tmp_dense, rd * rd * batch_size);
Expand Down
Loading