developing switch block pattern for tileable routing architecture

This commit is contained in:
tangxifan 2019-06-18 16:52:42 -06:00
parent 352c97302b
commit 9ca1b42f4c
3 changed files with 193 additions and 44 deletions

View File

@ -47,12 +47,16 @@
#include <vector>
#include <algorithm>
#include "vtr_ndmatrix.h"
#include "vpr_types.h"
#include "globals.h"
#include "vpr_utils.h"
#include "rr_graph_util.h"
#include "ReadOptions.h"
#include "rr_graph.h"
#include "rr_graph2.h"
#include "rr_graph_sbox.h"
#include "route_common.h"
#include "fpga_x2p_types.h"
#include "rr_graph_tileable_builder.h"
@ -892,9 +896,9 @@ void alloc_rr_graph_fast_lookup(const DeviceCoordinator& device_size,
/* For OPINs, IPINs, SOURCE, SINKs, CHANX and CHANY */
for (int type = 0; type < NUM_RR_TYPES; ++type) {
/* Skip SOURCE and OPIN, they will share with SOURCE and SINK
* SOURCE and SINK have unique ptc values so their data can be shared.
* IPIN and OPIN have unique ptc values so their data can be shared.
*/
* SOURCE and SINK have unique ptc values so their data can be shared.
* IPIN and OPIN have unique ptc values so their data can be shared.
*/
if ((SOURCE == type) || (OPIN == type) ) {
continue;
}
@ -1337,8 +1341,9 @@ void add_one_edge_for_two_rr_nodes(t_rr_graph* rr_graph,
***********************************************************************/
static
void build_edges_for_one_tileable_rr_gsb(t_rr_graph* rr_graph, RRGSB* rr_gsb,
int** Fc_in, int** Fc_out,
enum e_switch_block_type sb_type, int Fs,
std::vector < std::vector< std::vector<int> > > track2ipin_lookup,
std::vector < std::vector< std::vector<int> > > opin2track_map,
std::vector < std::vector< std::vector<int> > > sb_conn,
int num_directs, t_clb_to_clb_directs* clb_to_clb_directs,
int num_switches, int delayless_switch) {
/* Check rr_gsb */
@ -1362,16 +1367,19 @@ void build_edges_for_one_tileable_rr_gsb(t_rr_graph* rr_graph, RRGSB* rr_gsb,
/* Find IPINs */
for (size_t inode = 0; inode < rr_gsb->get_num_ipin_nodes(gsb_side); ++inode) {
t_rr_node* ipin_node = rr_gsb->get_ipin_node(gsb_side, inode);
/* 1. create edges between SOURCE and OPINs */
/* 2. create edges between IPINs and SINKs */
int sink_node_id = get_rr_node_index(ipin_node->xlow, ipin_node->ylow,
SINK, ipin_node->ptc_num,
rr_graph->rr_node_indices);
/* add edges to the src_node */
/* add edges to connect the IPIN node to SINK nodes */
add_one_edge_for_two_rr_nodes(rr_graph, ipin_node - rr_graph->rr_node, sink_node_id,
delayless_switch);
}
}
/* 3. create edges between CHANX | CHANY and IPINs (connections inside connection blocks) */
/* For TOP and BOTTOM */
return;
}
@ -1393,8 +1401,9 @@ void build_rr_graph_edges(t_rr_graph* rr_graph,
const DeviceCoordinator& device_size,
std::vector<size_t> device_chan_width,
std::vector<t_segment_inf> segment_inf,
int** Fc_in, int** Fc_out,
enum e_switch_block_type sb_type, int Fs,
int L_num_types, t_type_ptr types,
struct s_ivec**** track_to_ipin_lookup, int***** opin_to_track_map,
vtr::NdMatrix<std::vector<int>,3> switch_block_conn,
int num_directs, t_clb_to_clb_directs* clb_to_clb_directs,
int num_switches, int delayless_switch) {
DeviceCoordinator device_range(device_size.get_x() - 1, device_size.get_y() - 1);
@ -1404,9 +1413,15 @@ void build_rr_graph_edges(t_rr_graph* rr_graph,
DeviceCoordinator gsb_coordinator(ix, iy);
/* Create a GSB object */
RRGSB rr_gsb = build_one_tileable_rr_gsb(device_range, device_chan_width, segment_inf, gsb_coordinator, rr_graph);
/* adapt the track_to_ipin_lookup for the GSB nodes */
std::vector < std::vector< std::vector<int> > > track2ipin_lookup; /* [0..gsb_side][0..num_tracks][0..Fc] */
/* adapt the opin_to_track_map for the GSB nodes */
std::vector < std::vector< std::vector<int> > > opin2track_map; /* [0..gsb_side][0..num_opin_node][0..Fc] */
/* adapt the switch_block_conn for the GSB nodes */
std::vector < std::vector< std::vector<int> > > sb_conn; /* [0..gsb_side][0..chan_width][0..Fc] */
/* Build edges for a GSB */
build_edges_for_one_tileable_rr_gsb(rr_graph, &rr_gsb, Fc_in, Fc_out,
sb_type, Fs,
build_edges_for_one_tileable_rr_gsb(rr_graph, &rr_gsb,
track2ipin_lookup, opin2track_map, sb_conn,
num_directs, clb_to_clb_directs,
num_switches, delayless_switch);
/* Finish this GSB, go to the next*/
@ -1416,6 +1431,76 @@ void build_rr_graph_edges(t_rr_graph* rr_graph,
return;
}
/************************************************************************
* Build internal connection pattern for a switch block
* This function is adapt to fit the tileable routing context from
* rr_graph_sbox.c : alloc_and_load_switch_block_conn
* Switch box: *
* TOP (CHANY) *
* | | | | | | *
* +-----------+ *
* --| |-- *
* --| |-- *
* LEFT --| |-- RIGHT *
* (CHANX)--| |--(CHANX) *
* --| |-- *
* --| |-- *
* +-----------+ *
* | | | | | | *
* BOTTOM (CHANY)
*
* [0..3][0..3][0..nodes_per_chan-1]. Structure below is indexed as: *
* [from_side][to_side][from_track]. That yields an integer vector (ivec) *
* of the tracks to which from_track connects in the proper to_location. *
* For simple switch boxes this is overkill, but it will allow complicated *
* switch boxes with Fs > 3, etc. without trouble.
***********************************************************************/
static
vtr::NdMatrix<std::vector<int>,3> alloc_and_load_tileable_switch_block_conn(size_t chan_width,
enum e_switch_block_type switch_block_type,
int Fs) {
/* Currently Fs must be 3 since each track maps once to each other side */
VTR_ASSERT(3 == Fs);
vtr::NdMatrix<std::vector<int>,3> switch_block_conn({4, 4, chan_width});
for (e_side from_side : {TOP, RIGHT, BOTTOM, LEFT}) {
for (e_side to_side : {TOP, RIGHT, BOTTOM, LEFT}) {
for (size_t from_track = 0; from_track < chan_width; from_track++) {
if (from_side != to_side) {
switch_block_conn[from_side][to_side][from_track].resize(1);
switch_block_conn[from_side][to_side][from_track][0] = get_simple_switch_block_track(from_side, to_side,
from_track, switch_block_type,
chan_width);
} else { /* from_side == to_side -> no connection. */
switch_block_conn[from_side][to_side][from_track].clear();
}
}
}
}
if (getEchoEnabled()) {
FILE *out = fopen("switch_block_conn.echo", "w");
for (int l = 0; l < 4; ++l) {
for (int k = 0; k < 4; ++k) {
fprintf(out, "Side %d to %d\n", l, k);
for (size_t j = 0; j < chan_width; ++j) {
fprintf(out, "%zu: ", j);
for (unsigned i = 0; i < switch_block_conn[l][k][j].size(); ++i) {
fprintf(out, "%d ", switch_block_conn[l][k][j][i]);
}
fprintf(out, "\n");
}
fprintf(out, "\n");
}
}
fclose(out);
}
return switch_block_conn;
}
/************************************************************************
* Main function of this file
* Builder for a detailed uni-directional tileable rr_graph
@ -1484,7 +1569,7 @@ void build_tileable_unidir_rr_graph(INP int L_num_types,
for (int ix = 0; ix < (L_nx + 2); ++ix) {
grids[ix].resize(L_ny + 2);
for (int iy = 0; ix < (L_ny + 2); ++iy) {
grid[ix][iy] = L_grid[ix][iy];
grids[ix][iy] = L_grid[ix][iy];
}
}
/* Create a vector of channel width, we support X-direction and Y-direction has different W */
@ -1532,9 +1617,13 @@ void build_tileable_unidir_rr_graph(INP int L_num_types,
load_rr_nodes_basic_info(&rr_graph, device_size, grids, device_chan_width, segment_infs);
/************************************************************************
* 3. Create the connectivity of OPINs
* a. Evenly assign connections to OPINs to routing tracks
* b. the connection pattern should be same across the fabric
* 3.1 Create the connectivity of OPINs
* a. Evenly assign connections to OPINs to routing tracks
* b. the connection pattern should be same across the fabric
*
* 3.2 Create the connectivity of IPINs
* a. Evenly assign connections from routing tracks to IPINs
* b. the connection pattern should be same across the fabric
***********************************************************************/
int **Fc_in = NULL; /* [0..num_types-1][0..num_pins-1] */
boolean Fc_clipped;
@ -1545,18 +1634,42 @@ void build_tileable_unidir_rr_graph(INP int L_num_types,
*Warnings |= RR_GRAPH_WARN_FC_CLIPPED;
}
/************************************************************************
* 4. Create the connectivity of IPINs
* a. Evenly assign connections from routing tracks to IPINs
* b. the connection pattern should be same across the fabric
***********************************************************************/
int **Fc_out = NULL; /* [0..num_types-1][0..num_pins-1] */
Fc_clipped = FALSE;
Fc_out = alloc_and_load_actual_fc(L_num_types, types, chan_width,
TRUE, UNI_DIRECTIONAL, &Fc_clipped, ignore_Fc_0);
if (Fc_clipped) {
*Warnings |= RR_GRAPH_WARN_FC_CLIPPED;
}
/* START IPINP MAP */
/* Create ipin map lookups */
int***** ipin_to_track_map = (int*****) my_calloc(L_num_types, sizeof(int****));
struct s_ivec**** track_to_ipin_lookup = (struct s_ivec****) my_calloc(L_num_types, sizeof(struct s_ivec***));
boolean* perturb_ipins = alloc_and_load_perturb_ipins(chan_width, L_num_types, Fc_in, Fc_out, UNI_DIRECTIONAL);
for (int i = 0; i < L_num_types; ++i) {
ipin_to_track_map[i] = alloc_and_load_pin_to_track_map(RECEIVER, chan_width, Fc_in[i], &types[i],
perturb_ipins[i], UNI_DIRECTIONAL);
track_to_ipin_lookup[i] = alloc_and_load_track_to_pin_lookup(ipin_to_track_map[i], Fc_in[i],
types[i].height, types[i].num_pins, chan_width);
}
/* END IPINP MAP */
/* START OPINP MAP */
/* Create opin map lookups */
int***** opin_to_track_map = (int*****) my_calloc(L_num_types, sizeof(int****));
for (int i = 0; i < L_num_types; ++i) {
opin_to_track_map[i] = alloc_and_load_pin_to_track_map(DRIVER, chan_width, Fc_out[i], &types[i], FALSE, UNI_DIRECTIONAL);
}
/************************************************************************
* 5. Build the connections tile by tile:
* 4. Build switch block connection matrix
***********************************************************************/
vtr::NdMatrix<std::vector<int>,3> switch_block_conn = alloc_and_load_tileable_switch_block_conn(chan_width, sb_type, Fs);
/************************************************************************
* 6. Build the connections tile by tile:
* We classify rr_nodes into a general switch block (GSB) data structure
* where we create edges to each rr_nodes in the GSB with respect to
* Fc_in and Fc_out, switch block patterns
@ -1571,7 +1684,7 @@ void build_tileable_unidir_rr_graph(INP int L_num_types,
/* Create edges for a tileable rr_graph */
build_rr_graph_edges(&rr_graph, device_size, device_chan_width, segment_infs,
Fc_in, Fc_out, sb_type, Fs,
L_num_types, types, track_to_ipin_lookup, opin_to_track_map, switch_block_conn,
num_directs, clb_to_clb_directs, num_switches, delayless_switch);
/************************************************************************
@ -1592,6 +1705,37 @@ void build_tileable_unidir_rr_graph(INP int L_num_types,
rr_node = rr_graph.rr_node;
rr_node_indices = rr_graph.rr_node_indices;
/************************************************************************
* 9. Free all temp stucts
***********************************************************************/
/* Free all temp structs */
if (Fc_in) {
free_matrix(Fc_in,0, L_num_types, 0, sizeof(int));
Fc_in = NULL;
}
if (Fc_out) {
free_matrix(Fc_out,0, L_num_types, 0, sizeof(int));
Fc_out = NULL;
}
if (perturb_ipins) {
free(perturb_ipins);
perturb_ipins = NULL;
}
if (opin_to_track_map) {
for (int i = 0; i < L_num_types; ++i) {
free_matrix4(opin_to_track_map[i], 0, types[i].num_pins - 1, 0,
types[i].height - 1, 0, 3, 0, sizeof(int));
}
free(opin_to_track_map);
}
free_type_pin_to_track_map(ipin_to_track_map, types);
free_type_track_to_ipin_map(track_to_ipin_lookup, types, chan_width);
if(clb_to_clb_directs != NULL) {
free(clb_to_clb_directs);
}
return;
}

View File

@ -69,14 +69,6 @@ static t_chunk rr_mem_ch = {NULL, 0, NULL};
/* Status of current chunk being dished out by calls to my_chunk_malloc. */
/********************* Subroutines local to this module. *******************/
static int ****alloc_and_load_pin_to_track_map(INP enum e_pin_type pin_type,
INP int nodes_per_chan, INP int *Fc, INP t_type_ptr Type,
INP boolean perturb_switch_pattern,
INP enum e_directionality directionality);
static struct s_ivec ***alloc_and_load_track_to_pin_lookup(
INP int ****pin_to_track_map, INP int *Fc, INP int height,
INP int num_pins, INP int nodes_per_chan);
static void build_bidir_rr_opins(INP int i, INP int j,
INOUTP t_rr_node * L_rr_node, INP t_ivec *** L_rr_node_indices,
@ -124,10 +116,6 @@ static void check_all_tracks_reach_pins(t_type_ptr type,
int ****tracks_connected_to_pin, int nodes_per_chan, int Fc,
enum e_pin_type ipin_or_opin);
static boolean *alloc_and_load_perturb_ipins(INP int nodes_per_chan,
INP int L_num_types, INP int **Fc_in, INP int **Fc_out,
INP enum e_directionality directionality);
static void build_rr_sinks_sources(INP int i, INP int j,
INP t_rr_node * L_rr_node, INP t_ivec *** L_rr_node_indices,
INP int delayless_switch, INP struct s_grid_tile **L_grid);
@ -188,11 +176,6 @@ static void print_distribution(FILE * fptr,
t_mux_size_distribution * distr_struct);
#endif
static void free_type_pin_to_track_map(int***** ipin_to_track_map,
t_type_ptr types);
static void free_type_track_to_ipin_map(struct s_ivec**** track_to_pin_map,
t_type_ptr types, int nodes_per_chan);
static t_seg_details *alloc_and_load_global_route_seg_details(
INP int nodes_per_chan, INP int global_route_switch);
@ -543,7 +526,7 @@ void rr_graph_externals(t_timing_inf timing_inf,
alloc_and_load_rr_clb_source(rr_node_indices);
}
static boolean *
boolean *
alloc_and_load_perturb_ipins(INP int nodes_per_chan, INP int L_num_types,
INP int **Fc_in, INP int **Fc_out, INP enum e_directionality directionality) {
int i;
@ -675,7 +658,7 @@ alloc_and_load_actual_fc(INP int L_num_types, INP t_type_ptr types,
}
/* frees the track to ipin mapping for each physical grid type */
static void free_type_track_to_ipin_map(struct s_ivec**** track_to_pin_map,
void free_type_track_to_ipin_map(struct s_ivec**** track_to_pin_map,
t_type_ptr types, int nodes_per_chan) {
int i, itrack, ioff, iside;
for (i = 0; i < num_types; i++) {
@ -698,7 +681,7 @@ static void free_type_track_to_ipin_map(struct s_ivec**** track_to_pin_map,
}
/* frees the ipin to track mapping for each physical grid type */
static void free_type_pin_to_track_map(int***** ipin_to_track_map,
void free_type_pin_to_track_map(int***** ipin_to_track_map,
t_type_ptr types) {
int i;
for (i = 0; i < num_types; i++) {
@ -1532,7 +1515,7 @@ void alloc_and_load_edges_and_switches(INP t_rr_node * L_rr_node, INP int inode,
assert(i == num_edges);
}
static int ****
int ****
alloc_and_load_pin_to_track_map(INP enum e_pin_type pin_type,
INP int nodes_per_chan, INP int *Fc, INP t_type_ptr Type,
INP boolean perturb_switch_pattern,
@ -1873,7 +1856,7 @@ static void check_all_tracks_reach_pins(t_type_ptr type,
/* Allocates and loads the track to ipin lookup for each physical grid type. This
* is the same information as the ipin_to_track map but accessed in a different way. */
static struct s_ivec ***
struct s_ivec ***
alloc_and_load_track_to_pin_lookup(INP int ****pin_to_track_map, INP int *Fc,
INP int height, INP int num_pins, INP int nodes_per_chan) {
int ipin, iside, itrack, iconn, ioff, pin_counter;

View File

@ -63,5 +63,27 @@ void rr_graph_externals(t_timing_inf timing_inf,
t_segment_inf * segment_inf, int num_seg_types, int nodes_per_chan,
int wire_to_ipin_switch, enum e_base_cost_type base_cost_type);
/* Xifan Tang: Functions shared by tileable rr_graph generator */
int ****alloc_and_load_pin_to_track_map(INP enum e_pin_type pin_type,
INP int nodes_per_chan, INP int *Fc, INP t_type_ptr Type,
INP boolean perturb_switch_pattern,
INP enum e_directionality directionality);
struct s_ivec ***alloc_and_load_track_to_pin_lookup(
INP int ****pin_to_track_map, INP int *Fc, INP int height,
INP int num_pins, INP int nodes_per_chan);
boolean *
alloc_and_load_perturb_ipins(INP int nodes_per_chan, INP int L_num_types,
INP int **Fc_in, INP int **Fc_out, INP enum e_directionality directionality);
void free_type_pin_to_track_map(int***** ipin_to_track_map,
t_type_ptr types);
void free_type_track_to_ipin_map(struct s_ivec**** track_to_pin_map,
t_type_ptr types, int nodes_per_chan);
#endif