riscv
/
OpenFPGA
mirror of https://github.com/lnis-uofu/OpenFPGA.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
OpenFPGA/vpr7_x2p/vpr/SRC/route/route_common.c

1494 lines
50 KiB
C
Executable File
Raw Blame History

#include <math.h>
#include <stdio.h>
#include <assert.h>
#include <time.h>
#include "util.h"
#include "vpr_types.h"
#include "vpr_utils.h"
#include "globals.h"
#include "route_export.h"
#include "route_common.h"
#include "route_tree_timing.h"
#include "route_timing.h"
#include "route_breadth_first.h"
#include "place_and_route.h"
#include "rr_graph.h"
#include "read_xml_arch_file.h"
#include "ReadOptions.h"
/* mrFPGA */
#include "mrfpga_globals.h"
#include "buffer_insertion.h"
/* end */
#include "rr_graph_builder_utils.h"
/* Xifan TANG: useful functions for pb_pin_eq_auto_detect */
void reassign_rr_node_net_num_from_scratch();
/***************** Variables shared only by route modules *******************/
t_rr_node_route_inf *rr_node_route_inf = NULL; /* [0..num_rr_nodes-1] */
struct s_bb *route_bb = NULL; /* [0..num_nets-1]. Limits area in which each */
/* net must be routed. */
/**************** Static variables local to route_common.c ******************/
static struct s_heap **heap; /* Indexed from [1..heap_size] */
static int heap_size; /* Number of slots in the heap array */
static int heap_tail; /* Index of first unused slot in the heap array */
/* For managing my own list of currently free heap data structures. */
static struct s_heap *heap_free_head = NULL;
/* For keeping track of the sudo malloc memory for the heap*/
static t_chunk heap_ch = {NULL, 0, NULL};
/* For managing my own list of currently free trace data structures. */
static struct s_trace *trace_free_head = NULL;
/* For keeping track of the sudo malloc memory for the trace*/
static t_chunk trace_ch = {NULL, 0, NULL};
#ifdef DEBUG
static int num_trace_allocated = 0; /* To watch for memory leaks. */
static int num_heap_allocated = 0;
static int num_linked_f_pointer_allocated = 0;
#endif
static struct s_linked_f_pointer *rr_modified_head = NULL;
static struct s_linked_f_pointer *linked_f_pointer_free_head = NULL;
static t_chunk linked_f_pointer_ch = {NULL, 0, NULL};
/* The numbering relation between the channels and clbs is: *
* *
* | IO | chan_ | CLB | chan_ | CLB | *
* |grid[0][2] | y[0][2] |grid[1][2] | y[1][2] | grid[2][2]| *
* +-----------+ +-----------+ +-----------+ *
* } capacity in *
* No channel chan_x[1][1] chan_x[2][1] } chan_width *
* } _x[1] *
* +-----------+ +-----------+ +-----------+ *
* | | chan_ | | chan_ | | *
* | IO | y[0][1] | CLB | y[1][1] | CLB | *
* |grid[0][1] | |grid[1][1] | |grid[2][1] | *
* | | | | | | *
* +-----------+ +-----------+ +-----------+ *
* } capacity in *
* chan_x[1][0] chan_x[2][0] } chan_width *
* } _x[0] *
* +-----------+ +-----------+ *
* No | | No | | *
* Channel | IO | Channel | IO | *
* |grid[1][0] | |grid[2][0] | *
* | | | | *
* +-----------+ +-----------+ *
* *
* {=======} {=======} *
* Capacity in Capacity in *
* chan_width_y[0] chan_width_y[1] *
* */
/******************** Subroutines local to route_common.c *******************/
static void free_trace_data(struct s_trace *tptr);
static void load_route_bb(int bb_factor);
static struct s_trace *alloc_trace_data(void);
static void add_to_heap(struct s_heap *hptr);
static struct s_heap *alloc_heap_data(void);
static struct s_linked_f_pointer *alloc_linked_f_pointer(void);
static t_ivec **alloc_and_load_clb_opins_used_locally(void);
static void adjust_one_rr_occ_and_pcost(int inode, int add_or_sub,
float pres_fac);
/************************** Subroutine definitions ***************************/
void save_routing(struct s_trace **best_routing,
t_ivec ** clb_opins_used_locally,
t_ivec ** saved_clb_opins_used_locally) {
/* This routing frees any routing currently held in best routing, *
* then copies over the current routing (held in trace_head), and *
* finally sets trace_head and trace_tail to all NULLs so that the *
* connection to the saved routing is broken. This is necessary so *
* that the next iteration of the router does not free the saved *
* routing elements. Also saves any data about locally used clb_opins, *
* since this is also part of the routing. */
int inet, iblk, iclass, ipin, num_local_opins;
struct s_trace *tptr, *tempptr;
t_type_ptr type;
for (inet = 0; inet < num_nets; inet++) {
/* Free any previously saved routing. It is no longer best. */
tptr = best_routing[inet];
while (tptr != NULL) {
tempptr = tptr->next;
free_trace_data(tptr);
tptr = tempptr;
}
/* Save a pointer to the current routing in best_routing. */
best_routing[inet] = trace_head[inet];
/* Set the current (working) routing to NULL so the current trace *
* elements won't be reused by the memory allocator. */
trace_head[inet] = NULL;
trace_tail[inet] = NULL;
}
/* Save which OPINs are locally used. */
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
/* Xifan TANG: Bypass those with pin_equivalence auto-detect */
if (TRUE == type->output_ports_eq_auto_detect) {
continue;
}
/* Xifan TANG: By pass IO */
if (IO_TYPE == type) {
continue;
}
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opins = clb_opins_used_locally[iblk][iclass].nelem;
/* clb_opins_used_locally may be changed.
* Reallocate saved_clb_opins_used_locally if needed
*/
if (0 == num_local_opins) {
if (NULL != saved_clb_opins_used_locally[iblk][iclass].list) {
free(saved_clb_opins_used_locally[iblk][iclass].list);
}
saved_clb_opins_used_locally[iblk][iclass].list = NULL;
} else {
saved_clb_opins_used_locally[iblk][iclass].list = (int*)my_realloc(saved_clb_opins_used_locally[iblk][iclass].list,
clb_opins_used_locally[iblk][iclass].nelem * sizeof(int));
}
/* Fill the list of saved_clb_opins_used_locally */
for (ipin = 0; ipin < num_local_opins; ipin++) {
saved_clb_opins_used_locally[iblk][iclass].list[ipin] =
clb_opins_used_locally[iblk][iclass].list[ipin];
}
}
}
}
void restore_routing(struct s_trace **best_routing,
t_ivec ** clb_opins_used_locally,
t_ivec ** saved_clb_opins_used_locally) {
/* Deallocates any current routing in trace_head, and replaces it with *
* the routing in best_routing. Best_routing is set to NULL to show that *
* it no longer points to a valid routing. NOTE: trace_tail is not *
* restored -- it is set to all NULLs since it is only used in *
* update_traceback. If you need trace_tail restored, modify this *
* routine. Also restores the locally used opin data. */
int inet, iblk, ipin, iclass, num_local_opins;
t_type_ptr type;
/* mrFPGA : Xifan TANG*/
if (is_mrFPGA && is_wire_buffer) {
load_best_buffer_list();
}
/* end */
for (inet = 0; inet < num_nets; inet++) {
/* Free any current routing. */
free_traceback(inet);
/* Set the current routing to the saved one. */
trace_head[inet] = best_routing[inet];
best_routing[inet] = NULL; /* No stored routing. */
}
/* Save which OPINs are locally used. */
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opins = clb_opins_used_locally[iblk][iclass].nelem;
for (ipin = 0; ipin < num_local_opins; ipin++) {
clb_opins_used_locally[iblk][iclass].list[ipin] =
saved_clb_opins_used_locally[iblk][iclass].list[ipin];
}
}
}
}
void get_serial_num(void) {
/* This routine finds a "magic cookie" for the routing and prints it. *
* Use this number as a routing serial number to ensure that programming *
* changes do not break the router. */
int inet, serial_num, inode;
struct s_trace *tptr;
serial_num = 0;
for (inet = 0; inet < num_nets; inet++) {
/* Global nets will have null trace_heads (never routed) so they *
* are not included in the serial number calculation. */
tptr = trace_head[inet];
while (tptr != NULL) {
inode = tptr->index;
serial_num += (inet + 1)
* (rr_node[inode].xlow * (nx + 1) - rr_node[inode].yhigh);
serial_num -= rr_node[inode].ptc_num * (inet + 1) * 10;
serial_num -= rr_node[inode].type * (inet + 1) * 100;
serial_num %= 2000000000; /* Prevent overflow */
tptr = tptr->next;
}
}
vpr_printf(TIO_MESSAGE_INFO, "Serial number (magic cookie) for the routing is: %d\n", serial_num);
}
boolean try_route(int width_fac, struct s_router_opts router_opts,
struct s_det_routing_arch det_routing_arch, t_segment_inf * segment_inf,
t_timing_inf timing_inf, float **net_delay, t_slack * slacks,
t_chan_width_dist chan_width_dist, t_ivec ** clb_opins_used_locally,
boolean * Fc_clipped, t_direct_inf *directs, int num_directs,
/*Xifan TANG: Switch Segment Pattern Support*/
t_swseg_pattern_inf* swseg_patterns) {
/* Attempts a routing via an iterated maze router algorithm. Width_fac *
* specifies the relative width of the channels, while the members of *
* router_opts determine the value of the costs assigned to routing *
* resource node, etc. det_routing_arch describes the detailed routing *
* architecture (connection and switch boxes) of the FPGA; it is used *
* only if a DETAILED routing has been selected. */
int tmp;
clock_t begin, end;
boolean success;
t_graph_type graph_type;
if (router_opts.route_type == GLOBAL) {
graph_type = GRAPH_GLOBAL;
/* Xifan Tang: tileable undirectional rr_graph support */
} else if (BI_DIRECTIONAL == det_routing_arch.directionality) {
graph_type = GRAPH_BIDIR;
} else if (UNI_DIRECTIONAL == det_routing_arch.directionality) {
if (true == det_routing_arch.tileable) {
graph_type = GRAPH_UNIDIR_TILEABLE;
} else {
graph_type = GRAPH_UNIDIR;
}
}
/* Set the channel widths */
init_chan(width_fac, chan_width_dist);
/* Free any old routing graph, if one exists. */
free_rr_graph();
begin = clock();
/* Set up the routing resource graph defined by this FPGA architecture. */
build_rr_graph(graph_type, num_types, type_descriptors, nx, ny, grid,
chan_width_x[0], NULL,
det_routing_arch.switch_block_type, det_routing_arch.Fs,
det_routing_arch.switch_block_sub_type, det_routing_arch.sub_Fs,
det_routing_arch.wire_opposite_side,
det_routing_arch.num_segment,
det_routing_arch.num_switch, segment_inf,
det_routing_arch.global_route_switch,
det_routing_arch.delayless_switch, timing_inf,
det_routing_arch.wire_to_ipin_switch, router_opts.base_cost_type,
directs, num_directs, FALSE,
&tmp,
// Xifan TANG: Add Switch Segment Pattern Support
det_routing_arch.num_swseg_pattern, swseg_patterns, FALSE, TRUE);
end = clock();
#ifdef CLOCKS_PER_SEC
vpr_printf(TIO_MESSAGE_INFO, "Build rr_graph took %g seconds.\n", (float)(end - begin) / CLOCKS_PER_SEC);
#else
vpr_printf(TIO_MESSAGE_INFO, "Build rr_graph took %g seconds.\n", (float)(end - begin) / CLK_PER_SEC);
#endif
/* Allocate and load some additional rr_graph information needed only by *
* the router. */
alloc_and_load_rr_node_route_structs();
init_route_structs(router_opts.bb_factor);
if (router_opts.router_algorithm == BREADTH_FIRST) {
vpr_printf(TIO_MESSAGE_INFO, "Confirming Router Algorithm: BREADTH_FIRST.\n");
success = try_breadth_first_route(router_opts, clb_opins_used_locally,
width_fac);
} else { /* TIMING_DRIVEN route */
vpr_printf(TIO_MESSAGE_INFO, "Confirming Router Algorithm: TIMING_DRIVEN.\n");
assert(router_opts.route_type != GLOBAL);
success = try_timing_driven_route(router_opts, net_delay, slacks,
clb_opins_used_locally,timing_inf.timing_analysis_enabled);
}
free_rr_node_route_structs();
return (success);
}
boolean feasible_routing(void) {
/* This routine checks to see if this is a resource-feasible routing. *
* That is, are all rr_node capacity limitations respected? It assumes *
* that the occupancy arrays are up to date when it is called. */
int inode;
for (inode = 0; inode < num_rr_nodes; inode++) {
if (rr_node[inode].occ > rr_node[inode].capacity) {
/*
vpr_printf(TIO_MESSAGE_ERROR, "(File:%s,[LINE%d]rr_node[%d] occupancy(%d) exceeds its capacity(%d)!\n",
__FILE__, __LINE__, inode, rr_node[inode].occ, rr_node[inode].capacity);
*/
return (FALSE);
}
}
return (TRUE);
}
void pathfinder_update_one_cost(struct s_trace *route_segment_start,
int add_or_sub, float pres_fac) {
/* This routine updates the occupancy and pres_cost of the rr_nodes that are *
* affected by the portion of the routing of one net that starts at *
* route_segment_start. If route_segment_start is trace_head[inet], the *
* cost of all the nodes in the routing of net inet are updated. If *
* add_or_sub is -1 the net (or net portion) is ripped up, if it is 1 the *
* net is added to the routing. The size of pres_fac determines how severly *
* oversubscribed rr_nodes are penalized. */
struct s_trace *tptr;
int inode, occ, capacity;
tptr = route_segment_start;
if (tptr == NULL) /* No routing yet. */
return;
for (;;) {
inode = tptr->index;
occ = rr_node[inode].occ + add_or_sub;
capacity = rr_node[inode].capacity;
rr_node[inode].occ = occ;
/* pres_cost is Pn in the Pathfinder paper. I set my pres_cost according to *
* the overuse that would result from having ONE MORE net use this routing *
* node. */
if (occ < capacity) {
rr_node_route_inf[inode].pres_cost = 1.;
} else {
rr_node_route_inf[inode].pres_cost = 1.
+ (occ + 1 - capacity) * pres_fac;
}
if (rr_node[inode].type == SINK) {
tptr = tptr->next; /* Skip next segment. */
if (tptr == NULL)
break;
}
tptr = tptr->next;
} /* End while loop -- did an entire traceback. */
}
void pathfinder_update_cost(float pres_fac, float acc_fac) {
/* This routine recomputes the pres_cost and acc_cost of each routing *
* resource for the pathfinder algorithm after all nets have been routed. *
* It updates the accumulated cost to by adding in the number of extra *
* signals sharing a resource right now (i.e. after each complete iteration) *
* times acc_fac. It also updates pres_cost, since pres_fac may have *
* changed. THIS ROUTINE ASSUMES THE OCCUPANCY VALUES IN RR_NODE ARE UP TO *
* DATE. */
int inode, occ, capacity;
for (inode = 0; inode < num_rr_nodes; inode++) {
occ = rr_node[inode].occ;
capacity = rr_node[inode].capacity;
if (occ > capacity) {
rr_node_route_inf[inode].acc_cost += (occ - capacity) * acc_fac;
rr_node_route_inf[inode].pres_cost = 1.
+ (occ + 1 - capacity) * pres_fac;
}
/* If occ == capacity, we don't need to increase acc_cost, but a change *
* in pres_fac could have made it necessary to recompute the cost anyway. */
else if (occ == capacity) {
rr_node_route_inf[inode].pres_cost = 1. + pres_fac;
}
}
}
void init_route_structs(int bb_factor) {
/* Call this before you route any nets. It frees any old traceback and *
* sets the list of rr_nodes touched to empty. */
int i;
for (i = 0; i < num_nets; i++)
free_traceback(i);
load_route_bb(bb_factor);
/* Check that things that should have been emptied after the last routing *
* really were. */
if (rr_modified_head != NULL) {
vpr_printf(TIO_MESSAGE_ERROR, "in init_route_structs. List of modified rr nodes is not empty.\n");
exit(1);
}
if (heap_tail != 1) {
vpr_printf(TIO_MESSAGE_ERROR, "in init_route_structs. Heap is not empty.\n");
exit(1);
}
}
struct s_trace *
update_traceback(struct s_heap *hptr, int inet) {
/* This routine adds the most recently finished wire segment to the *
* traceback linked list. The first connection starts with the net SOURCE *
* and begins at the structure pointed to by trace_head[inet]. Each *
* connection ends with a SINK. After each SINK, the next connection *
* begins (if the net has more than 2 pins). The first element after the *
* SINK gives the routing node on a previous piece of the routing, which is *
* the link from the existing net to this new piece of the net. *
* In each traceback I start at the end of a path and trace back through *
* its predecessors to the beginning. I have stored information on the *
* predecesser of each node to make traceback easy -- this sacrificies some *
* memory for easier code maintenance. This routine returns a pointer to *
* the first "new" node in the traceback (node not previously in trace). */
struct s_trace *tptr, *prevptr, *temptail, *ret_ptr;
int inode;
short iedge;
#ifdef DEBUG
t_rr_type rr_type;
#endif
inode = hptr->index;
#ifdef DEBUG
rr_type = rr_node[inode].type;
if (rr_type != SINK) {
vpr_printf(TIO_MESSAGE_ERROR, "in update_traceback. Expected type = SINK (%d).\n", SINK);
vpr_printf(TIO_MESSAGE_ERROR, "\tGot type = %d while tracing back net %d.\n", rr_type, inet);
exit(1);
}
#endif
tptr = alloc_trace_data(); /* SINK on the end of the connection */
tptr->index = inode;
tptr->iswitch = OPEN;
tptr->next = NULL;
temptail = tptr; /* This will become the new tail at the end */
/* of the routine. */
/* Now do it's predecessor. */
inode = hptr->u.prev_node;
iedge = hptr->prev_edge;
while (inode != NO_PREVIOUS) {
prevptr = alloc_trace_data();
prevptr->index = inode;
prevptr->iswitch = rr_node[inode].switches[iedge];
prevptr->next = tptr;
tptr = prevptr;
iedge = rr_node_route_inf[inode].prev_edge;
inode = rr_node_route_inf[inode].prev_node;
}
if (trace_tail[inet] != NULL) {
trace_tail[inet]->next = tptr; /* Traceback ends with tptr */
ret_ptr = tptr->next; /* First new segment. */
} else { /* This was the first "chunk" of the net's routing */
trace_head[inet] = tptr;
ret_ptr = tptr; /* Whole traceback is new. */
}
trace_tail[inet] = temptail;
return (ret_ptr);
}
void reset_path_costs(void) {
/* The routine sets the path_cost to HUGE_POSITIVE_FLOAT for all channel segments *
* touched by previous routing phases. */
struct s_linked_f_pointer *mod_ptr;
#ifdef DEBUG
int num_mod_ptrs;
#endif
/* The traversal method below is slightly painful to make it faster. */
if (rr_modified_head != NULL) {
mod_ptr = rr_modified_head;
#ifdef DEBUG
num_mod_ptrs = 1;
#endif
while (mod_ptr->next != NULL) {
*(mod_ptr->fptr) = HUGE_POSITIVE_FLOAT;
mod_ptr = mod_ptr->next;
#ifdef DEBUG
num_mod_ptrs++;
#endif
}
*(mod_ptr->fptr) = HUGE_POSITIVE_FLOAT; /* Do last one. */
/* Reset the modified list and put all the elements back in the free *
* list. */
mod_ptr->next = linked_f_pointer_free_head;
linked_f_pointer_free_head = rr_modified_head;
rr_modified_head = NULL;
#ifdef DEBUG
num_linked_f_pointer_allocated -= num_mod_ptrs;
#endif
}
}
float get_rr_cong_cost(int inode) {
/* Returns the *congestion* cost of using this rr_node. */
short cost_index;
float cost;
cost_index = rr_node[inode].cost_index;
cost = rr_indexed_data[cost_index].base_cost
* rr_node_route_inf[inode].acc_cost
* rr_node_route_inf[inode].pres_cost;
return (cost);
}
void mark_ends(int inet) {
/* Mark all the SINKs of this net as targets by setting their target flags *
* to the number of times the net must connect to each SINK. Note that *
* this number can occassionally be greater than 1 -- think of connecting *
* the same net to two inputs of an and-gate (and-gate inputs are logically *
* equivalent, so both will connect to the same SINK). */
int ipin, inode;
for (ipin = 1; ipin <= clb_net[inet].num_sinks; ipin++) {
inode = net_rr_terminals[inet][ipin];
rr_node_route_inf[inode].target_flag++;
}
}
void node_to_heap(int inode, float cost, int prev_node, int prev_edge,
float backward_path_cost, float R_upstream) {
/* Puts an rr_node on the heap, if the new cost given is lower than the *
* current path_cost to this channel segment. The index of its predecessor *
* is stored to make traceback easy. The index of the edge used to get *
* from its predecessor to it is also stored to make timing analysis, etc. *
* easy. The backward_path_cost and R_upstream values are used only by the *
* timing-driven router -- the breadth-first router ignores them. */
struct s_heap *hptr;
if (cost >= rr_node_route_inf[inode].path_cost)
return;
hptr = alloc_heap_data();
hptr->index = inode;
hptr->cost = cost;
hptr->u.prev_node = prev_node;
hptr->prev_edge = prev_edge;
hptr->backward_path_cost = backward_path_cost;
hptr->R_upstream = R_upstream;
add_to_heap(hptr);
}
void free_traceback(int inet) {
/* Puts the entire traceback (old routing) for this net on the free list *
* and sets the trace_head pointers etc. for the net to NULL. */
struct s_trace *tptr, *tempptr;
if(trace_head == NULL) {
return;
}
tptr = trace_head[inet];
while (tptr != NULL) {
tempptr = tptr->next;
free_trace_data(tptr);
tptr = tempptr;
}
trace_head[inet] = NULL;
trace_tail[inet] = NULL;
}
t_ivec **
alloc_route_structs(void) {
/* Allocates the data structures needed for routing. */
t_ivec **clb_opins_used_locally;
alloc_route_static_structs();
clb_opins_used_locally = alloc_and_load_clb_opins_used_locally();
return (clb_opins_used_locally);
}
void alloc_route_static_structs(void) {
trace_head = (struct s_trace **) my_calloc(num_nets,
sizeof(struct s_trace *));
trace_tail = (struct s_trace **) my_malloc(
num_nets * sizeof(struct s_trace *));
heap_size = nx * ny;
heap = (struct s_heap **) my_malloc(heap_size * sizeof(struct s_heap *));
heap--; /* heap stores from [1..heap_size] */
heap_tail = 1;
route_bb = (struct s_bb *) my_malloc(num_nets * sizeof(struct s_bb));
}
struct s_trace **
alloc_saved_routing(t_ivec ** clb_opins_used_locally,
t_ivec *** saved_clb_opins_used_locally_ptr) {
/* Allocates data structures into which the key routing data can be saved, *
* allowing the routing to be recovered later (e.g. after a another routing *
* is attempted). */
struct s_trace **best_routing;
t_ivec **saved_clb_opins_used_locally;
int iblk, iclass, num_local_opins;
t_type_ptr type;
best_routing = (struct s_trace **) my_calloc(num_nets,
sizeof(struct s_trace *));
saved_clb_opins_used_locally = (t_ivec **) my_malloc(
num_blocks * sizeof(t_ivec *));
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
saved_clb_opins_used_locally[iblk] = (t_ivec *) my_malloc(
type->num_class * sizeof(t_ivec));
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opins = clb_opins_used_locally[iblk][iclass].nelem;
saved_clb_opins_used_locally[iblk][iclass].nelem = num_local_opins;
if (num_local_opins == 0) {
saved_clb_opins_used_locally[iblk][iclass].list = NULL;
} else {
saved_clb_opins_used_locally[iblk][iclass].list =
(int *) my_malloc(num_local_opins * sizeof(int));
}
}
}
*saved_clb_opins_used_locally_ptr = saved_clb_opins_used_locally;
return (best_routing);
}
/* TODO: super hacky, jluu comment, I need to rethink this whole function, without it, logically equivalent output pins incorrectly use more pins than needed. I force that CLB output pin uses at most one output pin */
static t_ivec **
alloc_and_load_clb_opins_used_locally(void) {
/* Allocates and loads the data needed to make the router reserve some CLB *
* output pins for connections made locally within a CLB (if the netlist *
* specifies that this is necessary). */
t_ivec **clb_opins_used_locally;
int iblk, clb_pin, iclass, num_local_opins;
int class_low, class_high;
t_type_ptr type;
clb_opins_used_locally = (t_ivec **) my_malloc(
num_blocks * sizeof(t_ivec *));
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
get_class_range_for_block(iblk, &class_low, &class_high);
clb_opins_used_locally[iblk] = (t_ivec *) my_malloc(
type->num_class * sizeof(t_ivec));
for (iclass = 0; iclass < type->num_class; iclass++)
clb_opins_used_locally[iblk][iclass].nelem = 0;
for (clb_pin = 0; clb_pin < type->num_pins; clb_pin++) {
// another hack to avoid I/Os, whole function needs a rethink
if(type == IO_TYPE) {
continue;
}
/* Comment by Xifan TANG: count the number of unused clb OPINs ? Those pins may be used locally...
* It seems that jluu wants to force all these pins are used, even though there is no net mapped!
* Then router will not route with these unused clb_pins!!!
*/
if ((block[iblk].nets[clb_pin] != OPEN
&& clb_net[block[iblk].nets[clb_pin]].num_sinks == 0) || block[iblk].nets[clb_pin] == OPEN
) {
iclass = type->pin_class[clb_pin];
if(type->class_inf[iclass].type == DRIVER) {
/* Check to make sure class is in same range as that assigned to block */
assert(iclass >= class_low && iclass <= class_high);
clb_opins_used_locally[iblk][iclass].nelem++;
}
}
}
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opins = clb_opins_used_locally[iblk][iclass].nelem;
if (num_local_opins == 0)
clb_opins_used_locally[iblk][iclass].list = NULL;
else
clb_opins_used_locally[iblk][iclass].list = (int *) my_malloc(
num_local_opins * sizeof(int));
}
}
return (clb_opins_used_locally);
}
void free_trace_structs(void) {
/*the trace lists are only freed after use by the timing-driven placer */
/*Do not free them after use by the router, since stats, and draw */
/*routines use the trace values */
int i;
for (i = 0; i < num_nets; i++)
free_traceback(i);
if(trace_head) {
free(trace_head);
free(trace_tail);
}
trace_head = NULL;
trace_tail = NULL;
}
void free_route_structs() {
/* Frees the temporary storage needed only during the routing. The *
* final routing result is not freed. */
if(heap != NULL) {
free(heap + 1);
}
if(route_bb != NULL) {
free(route_bb);
}
heap = NULL; /* Defensive coding: crash hard if I use these. */
route_bb = NULL;
/*free the memory chunks that were used by heap and linked f pointer */
free_chunk_memory(&heap_ch);
free_chunk_memory(&linked_f_pointer_ch);
heap_free_head = NULL;
linked_f_pointer_free_head = NULL;
}
void free_saved_routing(struct s_trace **best_routing,
t_ivec ** saved_clb_opins_used_locally) {
/* Frees the data structures needed to save a routing. */
int i;
free(best_routing);
for (i = 0; i < num_blocks; i++) {
free_ivec_vector(saved_clb_opins_used_locally[i], 0,
block[i].type->num_class - 1);
}
free(saved_clb_opins_used_locally);
}
void alloc_and_load_rr_node_route_structs(void) {
/* Allocates some extra information about each rr_node that is used only *
* during routing. */
int inode;
if (rr_node_route_inf != NULL) {
vpr_printf(TIO_MESSAGE_ERROR, "in alloc_and_load_rr_node_route_structs: old rr_node_route_inf array exists.\n");
exit(1);
}
rr_node_route_inf = (t_rr_node_route_inf *) my_malloc(num_rr_nodes * sizeof(t_rr_node_route_inf));
for (inode = 0; inode < num_rr_nodes; inode++) {
rr_node_route_inf[inode].prev_node = NO_PREVIOUS;
rr_node_route_inf[inode].prev_edge = NO_PREVIOUS;
rr_node_route_inf[inode].pres_cost = 1.;
rr_node_route_inf[inode].acc_cost = 1.;
rr_node_route_inf[inode].path_cost = HUGE_POSITIVE_FLOAT;
rr_node_route_inf[inode].target_flag = 0;
}
}
void reset_rr_node_route_structs(void) {
/* Allocates some extra information about each rr_node that is used only *
* during routing. */
int inode;
assert(rr_node_route_inf != NULL);
for (inode = 0; inode < num_rr_nodes; inode++) {
rr_node_route_inf[inode].prev_node = NO_PREVIOUS;
rr_node_route_inf[inode].prev_edge = NO_PREVIOUS;
rr_node_route_inf[inode].pres_cost = 1.;
rr_node_route_inf[inode].acc_cost = 1.;
rr_node_route_inf[inode].path_cost = HUGE_POSITIVE_FLOAT;
rr_node_route_inf[inode].target_flag = 0;
}
}
void free_rr_node_route_structs(void) {
/* Frees the extra information about each rr_node that is needed only *
* during routing. */
free(rr_node_route_inf);
rr_node_route_inf = NULL; /* Mark as free */
}
/* RESEARCH TODO: Bounding box heuristic needs to be redone for heterogeneous blocks */
static void load_route_bb(int bb_factor) {
/* This routine loads the bounding box arrays used to limit the space *
* searched by the maze router when routing each net. The search is *
* limited to channels contained with the net bounding box expanded *
* by bb_factor channels on each side. For example, if bb_factor is *
* 0, the maze router must route each net within its bounding box. *
* If bb_factor = nx, the maze router will search every channel in *
* the FPGA if necessary. The bounding boxes returned by this routine *
* are different from the ones used by the placer in that they are *
* clipped to lie within (0,0) and (nx+1,ny+1) rather than (1,1) and *
* (nx,ny). */
int k, xmax, ymax, xmin, ymin, x, y, inet;
for (inet = 0; inet < num_nets; inet++) {
x = block[clb_net[inet].node_block[0]].x;
y =
block[clb_net[inet].node_block[0]].y
+ block[clb_net[inet].node_block[0]].type->pin_height[clb_net[inet].node_block_pin[0]];
xmin = x;
ymin = y;
xmax = x;
ymax = y;
for (k = 1; k <= clb_net[inet].num_sinks; k++) {
x = block[clb_net[inet].node_block[k]].x;
y =
block[clb_net[inet].node_block[k]].y
+ block[clb_net[inet].node_block[k]].type->pin_height[clb_net[inet].node_block_pin[k]];
if (x < xmin) {
xmin = x;
} else if (x > xmax) {
xmax = x;
}
if (y < ymin) {
ymin = y;
} else if (y > ymax) {
ymax = y;
}
}
/* Want the channels on all 4 sides to be usuable, even if bb_factor = 0. */
xmin -= 1;
ymin -= 1;
/* Expand the net bounding box by bb_factor, then clip to the physical *
* chip area. */
route_bb[inet].xmin = std::max(xmin - bb_factor, 0);
route_bb[inet].xmax = std::min(xmax + bb_factor, nx + 1);
route_bb[inet].ymin = std::max(ymin - bb_factor, 0);
route_bb[inet].ymax = std::min(ymax + bb_factor, ny + 1);
}
}
void add_to_mod_list(float *fptr) {
/* This routine adds the floating point pointer (fptr) into a *
* linked list that indicates all the pathcosts that have been *
* modified thus far. */
struct s_linked_f_pointer *mod_ptr;
mod_ptr = alloc_linked_f_pointer();
/* Add this element to the start of the modified list. */
mod_ptr->next = rr_modified_head;
mod_ptr->fptr = fptr;
rr_modified_head = mod_ptr;
}
static void add_to_heap(struct s_heap *hptr) {
/* Adds an item to the heap, expanding the heap if necessary. */
int ito, ifrom;
struct s_heap *temp_ptr;
if (heap_tail > heap_size) { /* Heap is full */
heap_size *= 2;
heap = (struct s_heap **) my_realloc((void *) (heap + 1),
heap_size * sizeof(struct s_heap *));
heap--; /* heap goes from [1..heap_size] */
}
heap[heap_tail] = hptr;
ifrom = heap_tail;
ito = ifrom / 2;
heap_tail++;
while ((ito >= 1) && (heap[ifrom]->cost < heap[ito]->cost)) {
temp_ptr = heap[ito];
heap[ito] = heap[ifrom];
heap[ifrom] = temp_ptr;
ifrom = ito;
ito = ifrom / 2;
}
}
/*WMF: peeking accessor :) */
boolean is_empty_heap(void) {
return (boolean)(heap_tail == 1);
}
struct s_heap *
get_heap_head(void) {
/* Returns a pointer to the smallest element on the heap, or NULL if the *
* heap is empty. Invalid (index == OPEN) entries on the heap are never *
* returned -- they are just skipped over. */
int ito, ifrom;
struct s_heap *heap_head, *temp_ptr;
do {
if (heap_tail == 1) { /* Empty heap. */
vpr_printf(TIO_MESSAGE_WARNING, "Empty heap occurred in get_heap_head.\n");
vpr_printf(TIO_MESSAGE_WARNING, "Some blocks are impossible to connect in this architecture.\n");
return (NULL);
}
heap_head = heap[1]; /* Smallest element. */
/* Now fix up the heap */
heap_tail--;
heap[1] = heap[heap_tail];
ifrom = 1;
ito = 2 * ifrom;
while (ito < heap_tail) {
if (heap[ito + 1]->cost < heap[ito]->cost)
ito++;
if (heap[ito]->cost > heap[ifrom]->cost)
break;
temp_ptr = heap[ito];
heap[ito] = heap[ifrom];
heap[ifrom] = temp_ptr;
ifrom = ito;
ito = 2 * ifrom;
}
} while (heap_head->index == OPEN); /* Get another one if invalid entry. */
return (heap_head);
}
void empty_heap(void) {
int i;
for (i = 1; i < heap_tail; i++)
free_heap_data(heap[i]);
heap_tail = 1;
}
static struct s_heap *
alloc_heap_data(void) {
struct s_heap *temp_ptr;
if (heap_free_head == NULL) { /* No elements on the free list */
heap_free_head = (struct s_heap *) my_chunk_malloc(sizeof(struct s_heap),&heap_ch);
heap_free_head->u.next = NULL;
}
temp_ptr = heap_free_head;
heap_free_head = heap_free_head->u.next;
#ifdef DEBUG
num_heap_allocated++;
#endif
return (temp_ptr);
}
void free_heap_data(struct s_heap *hptr) {
hptr->u.next = heap_free_head;
heap_free_head = hptr;
#ifdef DEBUG
num_heap_allocated--;
#endif
}
void invalidate_heap_entries(int sink_node, int ipin_node) {
/* Marks all the heap entries consisting of sink_node, where it was reached *
* via ipin_node, as invalid (OPEN). Used only by the breadth_first router *
* and even then only in rare circumstances. */
int i;
for (i = 1; i < heap_tail; i++) {
if (heap[i]->index == sink_node && heap[i]->u.prev_node == ipin_node)
heap[i]->index = OPEN; /* Invalid. */
}
}
static struct s_trace *
alloc_trace_data(void) {
struct s_trace *temp_ptr;
if (trace_free_head == NULL) { /* No elements on the free list */
trace_free_head = (struct s_trace *) my_chunk_malloc(sizeof(struct s_trace),&trace_ch);
trace_free_head->next = NULL;
}
temp_ptr = trace_free_head;
trace_free_head = trace_free_head->next;
#ifdef DEBUG
num_trace_allocated++;
#endif
return (temp_ptr);
}
static void free_trace_data(struct s_trace *tptr) {
/* Puts the traceback structure pointed to by tptr on the free list. */
tptr->next = trace_free_head;
trace_free_head = tptr;
#ifdef DEBUG
num_trace_allocated--;
#endif
}
static struct s_linked_f_pointer *
alloc_linked_f_pointer(void) {
/* This routine returns a linked list element with a float pointer as *
* the node data. */
/*int i;*/
struct s_linked_f_pointer *temp_ptr;
if (linked_f_pointer_free_head == NULL) {
/* No elements on the free list */
linked_f_pointer_free_head = (struct s_linked_f_pointer *) my_chunk_malloc(sizeof(struct s_linked_f_pointer),&linked_f_pointer_ch);
linked_f_pointer_free_head->next = NULL;
}
temp_ptr = linked_f_pointer_free_head;
linked_f_pointer_free_head = linked_f_pointer_free_head->next;
#ifdef DEBUG
num_linked_f_pointer_allocated++;
#endif
return (temp_ptr);
}
void print_route(char *route_file) {
/* Prints out the routing to file route_file. */
int inet, inode, ipin, bnum, ilow, jlow, node_block_pin, iclass;
t_rr_type rr_type;
struct s_trace *tptr;
const char *name_type[] = { "SOURCE", "SINK", "IPIN", "OPIN", "CHANX", "CHANY",
"INTRA_CLUSTER_EDGE" };
FILE *fp;
fp = fopen(route_file, "w");
fprintf(fp, "Array size: %d x %d logic blocks.\n", nx, ny);
fprintf(fp, "\nRouting:");
for (inet = 0; inet < num_nets; inet++) {
if (clb_net[inet].is_global == FALSE) {
if (clb_net[inet].num_sinks == FALSE) {
fprintf(fp, "\n\nNet %d (%s)\n\n", inet, clb_net[inet].name);
fprintf(fp, "\n\nUsed in local cluster only, reserved one CLB pin\n\n");
} else {
fprintf(fp, "\n\nNet %d (%s)\n\n", inet, clb_net[inet].name);
tptr = trace_head[inet];
while (tptr != NULL) {
inode = tptr->index;
rr_type = rr_node[inode].type;
ilow = rr_node[inode].xlow;
jlow = rr_node[inode].ylow;
fprintf(fp, "Node:\t%d\t%6s (%d,%d) ", inode, name_type[rr_type], ilow, jlow);
if ((ilow != rr_node[inode].xhigh)
|| (jlow != rr_node[inode].yhigh))
fprintf(fp, "to (%d,%d) ", rr_node[inode].xhigh,
rr_node[inode].yhigh);
switch (rr_type) {
case IPIN:
case OPIN:
if (grid[ilow][jlow].type == IO_TYPE) {
fprintf(fp, " Pad: ");
} else { /* IO Pad. */
fprintf(fp, " Pin: ");
}
break;
case CHANX:
case CHANY:
fprintf(fp, " Track: ");
break;
case SOURCE:
case SINK:
if (grid[ilow][jlow].type == IO_TYPE) {
fprintf(fp, " Pad: ");
} else { /* IO Pad. */
fprintf(fp, " Class: ");
}
break;
default:
vpr_printf(TIO_MESSAGE_ERROR, "in print_route: Unexpected traceback element type: %d (%s).\n",
rr_type, name_type[rr_type]);
exit(1);
break;
}
/* A kind of dirty fix for tileable routing,
* the track_ids is allocated by tileable routing.
* If the vector is not empty, it means tileable routing is enabled
* we need another function to get the track_id rather than ptc_num
*/
if (0 == rr_node[inode].track_ids.size()) {
fprintf(fp, "%d ", rr_node[inode].ptc_num);
} else {
/* Xifan Tang: for routing tracks, get the actual track ids */
DeviceCoordinator cur_coord(ilow, jlow);
fprintf(fp, "%d ", get_rr_node_actual_track_id(&(rr_node[inode]), cur_coord));
}
/* Uncomment line below if you're debugging and want to see the switch types *
* used in the routing. */
/* fprintf (fp, "Switch: %d", tptr->iswitch); */
fprintf(fp, "\n");
tptr = tptr->next;
}
}
}
else { /* Global net. Never routed. */
fprintf(fp, "\n\nNet %d (%s): global net connecting:\n\n", inet,
clb_net[inet].name);
for (ipin = 0; ipin <= clb_net[inet].num_sinks; ipin++) {
bnum = clb_net[inet].node_block[ipin];
node_block_pin = clb_net[inet].node_block_pin[ipin];
iclass = block[bnum].type->pin_class[node_block_pin];
fprintf(fp, "Block %s (#%d) at (%d, %d), Pin class %d.\n",
block[bnum].name, bnum, block[bnum].x, block[bnum].y,
iclass);
}
}
}
fclose(fp);
if (getEchoEnabled() && isEchoFileEnabled(E_ECHO_MEM)) {
fp = my_fopen(getEchoFileName(E_ECHO_MEM), "w", 0);
fprintf(fp, "\nNum_heap_allocated: %d Num_trace_allocated: %d\n",
num_heap_allocated, num_trace_allocated);
fprintf(fp, "Num_linked_f_pointer_allocated: %d\n",
num_linked_f_pointer_allocated);
fclose(fp);
}
}
/* TODO: check if this is still necessary for speed */
void reserve_locally_used_opins(float pres_fac, boolean rip_up_local_opins,
t_ivec ** clb_opins_used_locally) {
/* In the past, this function implicitly allowed LUT duplication when there are free LUTs.
This was especially important for logical equivalence; however, now that we have a very general
logic cluster, it does not make sense to allow LUT duplication implicitly. we'll need to look into how we want to handle this case
*/
int iblk, num_local_opin, inode, from_node, iconn, num_edges, to_node;
int iclass, ipin;
float cost;
struct s_heap *heap_head_ptr;
t_type_ptr type;
if (rip_up_local_opins) {
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opin = clb_opins_used_locally[iblk][iclass].nelem;
/* Always 0 for pads and for RECEIVER (IPIN) classes */
for (ipin = 0; ipin < num_local_opin; ipin++) {
inode = clb_opins_used_locally[iblk][iclass].list[ipin];
adjust_one_rr_occ_and_pcost(inode, -1, pres_fac);
}
}
}
}
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
/* By pass type_descriptors that turns on pin equivalence auto_detect */
//if (TRUE == type->output_ports_eq_auto_detect) {
// continue;
//}
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opin = clb_opins_used_locally[iblk][iclass].nelem;
/* Always 0 for pads and for RECEIVER (IPIN) classes */
if (num_local_opin != 0) { /* Have to reserve (use) some OPINs */
from_node = rr_blk_source[iblk][iclass];
num_edges = rr_node[from_node].num_edges;
for (iconn = 0; iconn < num_edges; iconn++) {
to_node = rr_node[from_node].edges[iconn];
/* Xifan TANG: the to_node may not be the one should be reserved
* Need double check if the ptc_num of this node matches class_id
*/
//if (type->pin_class[rr_node[to_node].ptc_num] != iclass) {
// continue;
//}
/* Original VPR */
cost = get_rr_cong_cost(to_node);
/* Push nodes to heap:
* Xifan TANG:
* Need to check we do not push a node twice into the heap!
*/
node_to_heap(to_node, cost, OPEN, OPEN, 0., 0.);
}
for (ipin = 0; ipin < num_local_opin; ipin++) {
heap_head_ptr = get_heap_head();
inode = heap_head_ptr->index;
/* Xifan TANG: we only modify occ for single driver OPIN ! */
//if (1 == rr_node[inode].fan_in) {
adjust_one_rr_occ_and_pcost(inode, 1, pres_fac);
//}
clb_opins_used_locally[iblk][iclass].list[ipin] = inode;
free_heap_data(heap_head_ptr);
}
empty_heap();
}
}
}
}
/* Xifan TANG: new function to auto detect and reserved locally used opins */
void auto_detect_and_reserve_locally_used_opins(float pres_fac, boolean rip_up_local_opins,
t_ivec ** clb_opins_used_locally) {
/* In the past, this function implicitly allowed LUT duplication when there are free LUTs.
This was especially important for logical equivalence; however, now that we have a very general
logic cluster, it does not make sense to allow LUT duplication implicitly. we'll need to look into how we want to handle this case
*/
int iblk, num_local_opin, inode, from_node, iconn, num_edges, to_node;
int iclass, ipin;
float cost;
struct s_heap *heap_head_ptr;
t_type_ptr type;
/* Xifan TANG: Update net_num for all the rr_nodes */
reassign_rr_node_net_num_from_scratch();
/* VPR original method */
if (rip_up_local_opins) {
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
/* Bypass those with pin_equivalence auto-detect */
if (TRUE == type->output_ports_eq_auto_detect) {
continue;
}
/* By pass IO */
if (IO_TYPE == type) {
continue;
}
for (iclass = 0; iclass < type->num_class; iclass++) {
num_local_opin = clb_opins_used_locally[iblk][iclass].nelem;
/* Always 0 for pads and for RECEIVER (IPIN) classes */
for (ipin = 0; ipin < num_local_opin; ipin++) {
inode = clb_opins_used_locally[iblk][iclass].list[ipin];
adjust_one_rr_occ_and_pcost(inode, -1, pres_fac);
}
}
}
}
for (iblk = 0; iblk < num_blocks; iblk++) {
type = block[iblk].type;
/* Bypass those with pin_equivalence auto-detect */
if (TRUE == type->output_ports_eq_auto_detect) {
continue;
}
/* By pass IO */
if (IO_TYPE == type) {
continue;
}
for (iclass = 0; iclass < type->num_class; iclass++) {
/* Bypass non driver class */
if (DRIVER != type->class_inf[iclass].type) {
continue;
}
/* Always 0 for pads and for RECEIVER (IPIN) classes */
num_local_opin = clb_opins_used_locally[iblk][iclass].nelem;
/* Have to reserve (use) some OPINs */
ipin = 0;
/* We push nodes into heap and then we can pop-up with a sort by cost */
from_node = rr_blk_source[iblk][iclass];
num_edges = rr_node[from_node].num_edges;
/* initialize rr_node element: is_in_heap */
for (iconn = 0; iconn < num_edges; iconn++) {
to_node = rr_node[from_node].edges[iconn];
rr_node[to_node].is_in_heap = FALSE;
}
/* Find unmapped pins and add to heap */
for (iconn = 0; iconn < num_edges; iconn++) {
to_node = rr_node[from_node].edges[iconn];
if (OPEN != rr_node[to_node].net_num) {
continue;
}
/* we search by net_num if this inode is used or not */
if (FALSE == rr_node[to_node].is_in_heap) {
rr_node[to_node].is_in_heap = TRUE;
/* Original VPR */
cost = get_rr_cong_cost(to_node);
/* Push nodes to heap:
* Xifan TANG:
* Need to check we do not push a node twice into the heap!
*/
node_to_heap(to_node, cost, OPEN, OPEN, 0., 0.);
ipin++;
}
}
/* Re-allocate the look-up table if needed */
if (num_local_opin != ipin) {
clb_opins_used_locally[iblk][iclass].nelem = ipin;
if (0 == clb_opins_used_locally[iblk][iclass].nelem) {
clb_opins_used_locally[iblk][iclass].list = NULL;
} else {
clb_opins_used_locally[iblk][iclass].list = (int *) my_realloc(clb_opins_used_locally[iblk][iclass].list,
clb_opins_used_locally[iblk][iclass].nelem * sizeof(int));
}
}
/* We want to re-build the list of locally used opins from lowest cost to highest */
for (iconn = 0; iconn < clb_opins_used_locally[iblk][iclass].nelem; iconn++) {
heap_head_ptr = get_heap_head();
inode = heap_head_ptr->index;
/* Only update used pins */
assert(OPEN == rr_node[inode].net_num);
adjust_one_rr_occ_and_pcost(inode, 1, pres_fac); /* Reserve the pin ? */
clb_opins_used_locally[iblk][iclass].list[iconn] = inode;
rr_node[inode].is_in_heap = FALSE; /* reset the flag */
free_heap_data(heap_head_ptr);
}
empty_heap();
}
}
}
static void adjust_one_rr_occ_and_pcost(int inode, int add_or_sub,
float pres_fac) {
/* Increments or decrements (depending on add_or_sub) the occupancy of *
* one rr_node, and adjusts the present cost of that node appropriately. */
int occ, capacity;
occ = rr_node[inode].occ + add_or_sub;
capacity = rr_node[inode].capacity;
rr_node[inode].occ = occ;
if (occ < capacity) {
rr_node_route_inf[inode].pres_cost = 1.;
} else {
rr_node_route_inf[inode].pres_cost = 1.
+ (occ + 1 - capacity) * pres_fac;
}
}
void free_chunk_memory_trace(void) {
if (trace_ch.chunk_ptr_head != NULL) {
free_chunk_memory(&trace_ch);
}
}
Reference in New Issue View Git Blame Copy Permalink