Skip to content

[AP] General Fixed/Unfixed Blocks Cleanup #2990

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

Merged
merged 1 commit into from
Apr 22, 2025
Merged

[AP] General Fixed/Unfixed Blocks Cleanup #2990

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
19 changes: 19 additions & 0 deletions vpr/src/analytical_place/analytical_solver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -446,6 +446,25 @@ void B2BSolver::solve(unsigned iteration, PartialPlacement& p_placement) {
// Store an initial placement into the p_placement object as a starting point
// for the B2B solver.
if (iteration == 0) {
// If there are no fixed blocks, running bound2bound will always yield
// the trivial solution (all blocks on top of each other anywhere on the
// device). Skip having to solve for this by just putting all the blocks
// at the center of the device.
// TODO: This can be further improved by using the average compatible
// tile location for each AP block. The center is just an
// approximation.
if (num_fixed_blocks_ == 0) {
for (size_t row_id_idx = 0; row_id_idx < num_moveable_blocks_; row_id_idx++) {
APRowId row_id = APRowId(row_id_idx);
APBlockId blk_id = row_id_to_blk_id_[row_id];
p_placement.block_x_locs[blk_id] = device_grid_width_ / 2.0;
p_placement.block_y_locs[blk_id] = device_grid_height_ / 2.0;
}
block_x_locs_solved = p_placement.block_x_locs;
block_y_locs_solved = p_placement.block_y_locs;
return;
}

// In the first iteration, we have no prior information.
// Run the intial placer to get a first guess.
switch (initial_placement_ty_) {
Expand Down
6 changes: 3 additions & 3 deletions vpr/src/analytical_place/ap_netlist.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,10 +35,10 @@
*/
struct APFixedBlockLoc {
// Value that represents an unfixed dimension.
static constexpr int UNFIXED_DIM = -1;
static constexpr float UNFIXED_DIM = -1;
// The dimensions to fix.
int x = UNFIXED_DIM;
int y = UNFIXED_DIM;
float x = UNFIXED_DIM;
float y = UNFIXED_DIM;
int layer_num = UNFIXED_DIM;
int sub_tile = UNFIXED_DIM;
};
Expand Down
9 changes: 7 additions & 2 deletions vpr/src/analytical_place/gen_ap_netlist_from_atoms.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,8 +89,13 @@ APNetlist gen_ap_netlist_from_atoms(const AtomNetlist& atom_netlist,
const vtr::Rect<int>& region_rect = region.get_rect();
VTR_ASSERT(region_rect.xmin() == region_rect.xmax() && "AP: Expect each region to be a single point in x!");
VTR_ASSERT(region_rect.ymin() == region_rect.ymax() && "AP: Expect each region to be a single point in y!");
int blk_x_loc = region_rect.xmin();
int blk_y_loc = region_rect.ymin();
// Here we offset by 0.5 to put the fixed point in the center of the
// tile (assuming the tile is 1x1).
// TODO: Think about what to do when the user fixes blocks to large
// tiles. However, this solution will at least keep the atoms
// away from the edge of tiles.
float blk_x_loc = region_rect.xmin() + 0.5f;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you look up pin offsets (i.e. where a pin is on a large block) when computing where a net is attracted to a fixed pin? (It will be elsewhere in the code.) I think that should be done.
Same question for floating block too -- do you take pin offsets into account?
For 1x1 tiles a (0.5,0.5) offset seems right. When pin_offsets are taken into account for large blocks it would be good to check though; I think a (0.5,0.5) offset would still make sense but am not sure -- it depends how we set the offsets for pins on the periphery of a large block.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That is a good point! We currently do absolutely nothing with pin offsets. All blocks in the AP Netlist are currently treated as points of zero area which have mass. I will raise an issue to track this as a strong TODO item to look into. This would be a good long-term fix for the 0.5 offset and it would improve the mass-legalized solution.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Issue raised: #2995

float blk_y_loc = region_rect.ymin() + 0.5f;
// Get the layer.
VTR_ASSERT(region.get_layer_range().first == region.get_layer_range().second && "AP: Expect each region to be a single point in layer!");
int blk_layer_num = region.get_layer_range().first;
Expand Down
1 change: 1 addition & 0 deletions vtr_flow/parse/qor_config/qor_ap_fixed_chan_width.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ vpr_status;output.txt;vpr_status=(.*)
crit_path_delay;vpr.out;Critical path: (.*) ns
post_gp_hpwl;vpr.out;\s*Placement HPWL: (.*)
post_fl_hpwl;vpr.out;Initial placement BB estimate of wirelength: (.*)
post_dp_hpwl;vpr.out;BB estimate of min-dist \(placement\) wire length: (.*)
total_wirelength;vpr.out;\s*Total wirelength: (\d+)
post_gp_overfilled_bins;vpr.out;\s*Number of overfilled bins: (\d+)
post_gp_avg_overfill;vpr.out;\s*Average overfill magnitude: (.*)
Expand Down
8 changes: 4 additions & 4 deletions ...ression_tests/vtr_reg_strong/strong_ap/annealer_detailed_placer/config/golden_results.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
arch circuit script_params vtr_flow_elapsed_time vtr_max_mem_stage vtr_max_mem error odin_synth_time max_odin_mem parmys_synth_time max_parmys_mem abc_depth abc_synth_time abc_cec_time abc_sec_time max_abc_mem ace_time max_ace_mem num_clb num_io num_memories num_mult vpr_status vpr_revision vpr_build_info vpr_compiler vpr_compiled hostname rundir max_vpr_mem num_primary_inputs num_primary_outputs num_pre_packed_nets num_pre_packed_blocks num_netlist_clocks num_post_packed_nets num_post_packed_blocks device_width device_height device_grid_tiles device_limiting_resources device_name pack_mem pack_time initial_placed_wirelength_est placed_wirelength_est total_swap accepted_swap rejected_swap aborted_swap place_mem place_time place_quench_time initial_placed_CPD_est placed_CPD_est placed_setup_TNS_est placed_setup_WNS_est placed_geomean_nonvirtual_intradomain_critical_path_delay_est place_delay_matrix_lookup_time place_quench_timing_analysis_time place_quench_sta_time place_total_timing_analysis_time place_total_sta_time ap_mem ap_time ap_full_legalizer_mem ap_full_legalizer_time routed_wirelength avg_routed_wirelength routed_wiresegment avg_routed_wiresegment total_nets_routed total_connections_routed total_heap_pushes total_heap_pops logic_block_area_total logic_block_area_used routing_area_total routing_area_per_tile crit_path_route_success_iteration num_rr_graph_nodes num_rr_graph_edges collapsed_nodes critical_path_delay geomean_nonvirtual_intradomain_critical_path_delay setup_TNS setup_WNS hold_TNS hold_WNS create_rr_graph_time create_intra_cluster_rr_graph_time adding_internal_edges route_mem crit_path_route_time crit_path_total_timing_analysis_time crit_path_total_sta_time router_lookahead_mem tile_lookahead_computation_time router_lookahead_computation_time
k6_frac_N10_40nm.xml apex4.pre-vpr.blif common 3.59 vpr 75.35 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 81 9 -1 -1 success v8.0.0-12398-g63e0de52b-dirty release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-10T16:22:34 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 77156 9 19 897 28 0 562 109 16 16 256 -1 mcnc_medium -1 -1 9742 6119 10249 1760 5821 2668 75.3 MiB 2.81 0.01 6.60912 5.26997 -81.789 -5.26997 nan 0.05 0.00179692 0.00145997 0.0947522 0.0791883 75.3 MiB 2.81 75.3 MiB 1.73 9723 17.3316 2505 4.46524 4006 19859 666446 110600 1.05632e+07 4.36541e+06 1.26944e+06 4958.75 17 28900 206586 -1 5.51321 nan -87.5291 -5.51321 0 0 0.34 -1 -1 75.3 MiB 0.25 0.210289 0.182458 75.3 MiB -1 0.05
k6_frac_N10_40nm.xml des.pre-vpr.blif common 1.99 vpr 76.11 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 52 256 -1 -1 success v8.0.0-12398-g63e0de52b-dirty release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-10T16:22:34 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 77936 256 245 954 501 0 582 553 22 22 484 -1 mcnc_large -1 -1 8094 7501 39126 389 7567 31170 76.1 MiB 1.14 0.02 5.73036 4.09857 -784.452 -4.09857 nan 0.12 0.0036209 0.00320967 0.0974498 0.0783511 76.1 MiB 1.14 76.1 MiB 0.58 10270 17.6460 2810 4.82818 2238 5018 286471 63337 2.15576e+07 2.80249e+06 1.49107e+06 3080.73 14 47664 245996 -1 4.45761 nan -846.555 -4.45761 0 0 0.37 -1 -1 76.1 MiB 0.25 0.241683 0.211802 76.1 MiB -1 0.12
k6_frac_N10_40nm.xml ex1010.pre-vpr.blif common 12.92 vpr 106.02 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 290 10 -1 -1 success v8.0.0-12398-g63e0de52b-dirty release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-10T16:22:34 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 108560 10 10 2659 20 0 1285 310 22 22 484 -1 mcnc_large -1 -1 35642 24564 56902 15498 35924 5480 106.0 MiB 10.38 0.02 9.2649 6.61089 -64.2928 -6.61089 nan 0.15 0.00607931 0.0048535 0.385282 0.31707 106.0 MiB 10.38 106.0 MiB 5.44 37798 29.4148 9576 7.45214 8278 52602 2224659 293057 2.15576e+07 1.56293e+07 3.51389e+06 7260.09 17 64568 594370 -1 6.83554 nan -66.0686 -6.83554 0 0 1.10 -1 -1 106.0 MiB 0.85 0.754783 0.64805 106.0 MiB -1 0.15
k6_frac_N10_40nm.xml seq.pre-vpr.blif common 4.71 vpr 76.63 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 82 41 -1 -1 success v8.0.0-12398-g63e0de52b-dirty release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-10T16:22:34 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 78472 41 35 1006 76 0 578 158 16 16 256 -1 mcnc_medium -1 -1 10490 6475 10406 884 4940 4582 76.6 MiB 3.58 0.01 6.29657 4.68967 -140.848 -4.68967 nan 0.09 0.00393615 0.00326892 0.130272 0.112932 76.6 MiB 3.58 76.6 MiB 2.21 10487 18.1436 2747 4.75259 4099 21208 684107 115230 1.05632e+07 4.41931e+06 1.26944e+06 4958.75 18 28900 206586 -1 4.97595 nan -146.106 -4.97595 0 0 0.33 -1 -1 76.6 MiB 0.51 0.372342 0.333335 76.6 MiB -1 0.09
k6_frac_N10_40nm.xml apex4.pre-vpr.blif common 4.63 vpr 75.62 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 83 9 -1 -1 success v8.0.0-12490-ge99a5ee8c release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-21T16:53:01 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 77436 9 19 897 28 0 565 111 16 16 256 -1 mcnc_medium -1 -1 6985 6104 4899 468 3178 1253 75.6 MiB 3.56 0.01 5.45737 4.88762 -81.315 -4.88762 nan 0.09 0.00332256 0.00272635 0.0920151 0.0793873 75.6 MiB 3.56 75.6 MiB 2.42 9609 17.0372 2580 4.57447 4604 23759 790399 131608 1.05632e+07 4.4732e+06 1.26944e+06 4958.75 18 28900 206586 -1 5.07748 nan -84.5559 -5.07748 0 0 0.33 -1 -1 75.6 MiB 0.47 0.288973 0.25613 75.6 MiB -1 0.09
k6_frac_N10_40nm.xml des.pre-vpr.blif common 1.81 vpr 76.21 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 58 256 -1 -1 success v8.0.0-12490-ge99a5ee8c release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-21T16:53:01 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 78036 256 245 954 501 0 584 559 22 22 484 -1 mcnc_large -1 -1 7428 7369 18967 157 2360 16450 76.2 MiB 0.93 0.02 4.87092 4.07054 -789.645 -4.07054 nan 0.13 0.00365979 0.00331889 0.047619 0.043593 76.2 MiB 0.93 76.2 MiB 0.59 10168 17.4110 2803 4.79966 2303 5353 301769 65803 2.15576e+07 3.12585e+06 1.49107e+06 3080.73 13 47664 245996 -1 4.58117 nan -866.844 -4.58117 0 0 0.37 -1 -1 76.2 MiB 0.25 0.188446 0.174077 76.2 MiB -1 0.13
k6_frac_N10_40nm.xml ex1010.pre-vpr.blif common 11.93 vpr 105.82 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 289 10 -1 -1 success v8.0.0-12490-ge99a5ee8c release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-21T16:53:01 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 108360 10 10 2659 20 0 1312 309 22 22 484 -1 mcnc_large -1 -1 30891 24822 37893 7171 26674 4048 105.8 MiB 9.23 0.02 9.04847 6.65923 -63.8387 -6.65923 nan 0.16 0.00660438 0.00519941 0.309592 0.257376 105.8 MiB 9.23 105.8 MiB 4.50 37153 28.3178 9567 7.29192 8219 52828 2194741 285748 2.15576e+07 1.55754e+07 3.51389e+06 7260.09 18 64568 594370 -1 7.35737 nan -66.9799 -7.35737 0 0 1.18 -1 -1 105.8 MiB 0.93 0.709043 0.615207 105.8 MiB -1 0.16
k6_frac_N10_40nm.xml seq.pre-vpr.blif common 4.73 vpr 76.79 MiB -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 84 41 -1 -1 success v8.0.0-12490-ge99a5ee8c release VTR_ASSERT_LEVEL=3 GNU 13.3.0 on Linux-6.8.0-57-generic x86_64 2025-04-21T16:53:01 srivatsan-Precision-Tower-5810 /home/alex/vtr-verilog-to-routing 78628 41 35 1006 76 0 573 160 16 16 256 -1 mcnc_medium -1 -1 7311 6476 7538 345 3855 3338 76.8 MiB 3.59 0.01 5.73065 5.08486 -143.975 -5.08486 nan 0.09 0.0040653 0.00347069 0.103188 0.0899686 76.8 MiB 3.59 76.8 MiB 2.31 10018 17.4834 2728 4.76091 4171 21468 688416 120931 1.05632e+07 4.5271e+06 1.26944e+06 4958.75 18 28900 206586 -1 5.29306 nan -148.307 -5.29306 0 0 0.37 -1 -1 76.8 MiB 0.47 0.321818 0.287107 76.8 MiB -1 0.09
Loading