calyx_opt/analysis/static_tree.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607
use super::{FSMEncoding, StaticFSM};
use calyx_ir::{self as ir};
use calyx_ir::{Nothing, build_assignments};
use calyx_ir::{guard, structure};
use itertools::Itertools;
use std::collections::{BTreeMap, HashMap};
use std::ops::Not;
use std::rc::Rc;
use super::GraphColoring;
/// Optional Rc of a RefCell of a StaticFSM object.
type OptionalStaticFSM = Option<ir::RRC<StaticFSM>>;
/// Query (i ,(j,k)) that corresponds to:
/// Am I in iteration i, and between cylces j and k within
/// that query?
type SingleIterQuery = (u64, (u64, u64));
/// Query (i,j) that corresponds to:
/// Am I between iterations i and j, inclusive?
type ItersQuery = (u64, u64);
/// Helpful for translating queries for the FSMTree structure.
/// Because of the tree structure, %[i:j] is no longer is always equal to i <= fsm < j.
/// Offload(i) means the FSM is offloading when fsm == i: so if the fsm == i,
/// we need to look at the children to know what cycle we are in exactly.
/// Normal(i,j) means the FSM is outputing (i..j), incrementing each cycle (i.e.,
/// like normal) and not offloading. Note that even though the FSM is outputting
/// i..j each cycle, that does not necesarily mean we are in cycles i..j (due
/// to offloading performed in the past.)
#[derive(Debug)]
pub enum StateType {
Normal((u64, u64)),
Offload(u64),
}
/// Node can either be a SingleNode (i.e., a single node) or ParNodes (i.e., a group of
/// nodes that are executing in parallel).
/// Most methods in `Node` simply call the equivalent methods for each
/// of the two possible variants.
/// Perhaps could be more compactly implemented as a Trait.
pub enum Node {
Single(SingleNode),
Par(ParNodes),
}
// The following methods are used to actually instantiate the FSMTree structure
// and compile static groups/control to dynamic groups/control.
impl Node {
/// Instantiate the necessary registers.
/// The equivalent methods for the two variants contain more implementation
/// details.
/// `coloring`, `colors_to_max_values`, and `colors_to_fsm` are necessary
/// to know whether we actually need to instantiate a new FSM, or we can
/// juse use another node's FSM.
pub fn instantiate_fsms(
&mut self,
builder: &mut ir::Builder,
coloring: &HashMap<ir::Id, ir::Id>,
colors_to_max_values: &HashMap<ir::Id, (u64, u64)>,
colors_to_fsm: &mut HashMap<
ir::Id,
(OptionalStaticFSM, OptionalStaticFSM),
>,
one_hot_cutoff: u64,
) {
match self {
Node::Single(single_node) => single_node.instantiate_fsms(
builder,
coloring,
colors_to_max_values,
colors_to_fsm,
one_hot_cutoff,
),
Node::Par(par_nodes) => par_nodes.instantiate_fsms(
builder,
coloring,
colors_to_max_values,
colors_to_fsm,
one_hot_cutoff,
),
}
}
/// Count to n. Need to call `instantiate_fsms` before calling `count_to_n`.
/// The equivalent methods for the two variants contain more implementation
/// details.
/// `incr_start_cond` can optionally guard the 0->1 transition.
pub fn count_to_n(
&mut self,
builder: &mut ir::Builder,
incr_start_cond: Option<ir::Guard<Nothing>>,
) {
match self {
Node::Single(single_node) => {
single_node.count_to_n(builder, incr_start_cond)
}
Node::Par(par_nodes) => {
par_nodes.count_to_n(builder, incr_start_cond)
}
}
}
/// "Realize" the static groups into dynamic groups.
/// The main challenge is converting %[i:j] into fsm guards.
/// Need to call `instantiate_fsms` and
/// `count_to_n` before calling `realize`.
/// The equivalent methods for the two variants contain more implementation
/// details.
/// `reset_early_map`, `fsm_info_map`, and `group_rewrites` are just metadata
/// to make it easier to rewrite control, add wrappers, etc.
pub fn realize(
&mut self,
ignore_timing_guards: bool,
static_groups: &Vec<ir::RRC<ir::StaticGroup>>,
reset_early_map: &mut HashMap<ir::Id, ir::Id>,
fsm_info_map: &mut HashMap<
ir::Id,
(ir::Id, ir::Guard<Nothing>, ir::Guard<Nothing>),
>,
group_rewrites: &mut ir::rewriter::PortRewriteMap,
builder: &mut ir::Builder,
) {
match self {
Node::Single(single_node) => single_node.realize(
ignore_timing_guards,
static_groups,
reset_early_map,
fsm_info_map,
group_rewrites,
builder,
),
Node::Par(par_nodes) => par_nodes.realize(
ignore_timing_guards,
static_groups,
reset_early_map,
fsm_info_map,
group_rewrites,
builder,
),
}
}
/// Get the equivalent fsm guard when the tree is between cycles i and j, i.e.,
/// when i <= cycle_count < j.
/// The equivalent methods for the two variants contain more implementation
/// details.
pub fn query_between(
&mut self,
query: (u64, u64),
builder: &mut ir::Builder,
) -> ir::Guard<Nothing> {
match self {
Node::Single(single_node) => {
single_node.query_between(query, builder)
}
Node::Par(par_nodes) => par_nodes.query_between(query, builder),
}
}
}
/// The following methods are used to help build the conflict graph for coloring
/// to share FSMs
impl Node {
/// Get the names of all nodes (i.e., the names of the groups for each node
/// in the tree).
pub fn get_all_nodes(&self) -> Vec<ir::Id> {
match self {
Node::Single(single_node) => single_node.get_all_nodes(),
Node::Par(par_nodes) => par_nodes.get_all_nodes(),
}
}
/// Adds conflicts between nodes in the tree that execute at the same time.
pub fn add_conflicts(&self, conflict_graph: &mut GraphColoring<ir::Id>) {
match self {
Node::Single(single_node) => {
single_node.add_conflicts(conflict_graph)
}
Node::Par(par_nodes) => par_nodes.add_conflicts(conflict_graph),
}
}
/// Get max value of all nodes in the tree, according to some function f.
/// `f` takes in a Tree (i.e., a node type) and returns a `u64`.
pub fn get_max_value<F>(&self, name: &ir::Id, f: &F) -> u64
where
F: Fn(&SingleNode) -> u64,
{
match self {
Node::Single(single_node) => single_node.get_max_value(name, f),
Node::Par(par_nodes) => par_nodes.get_max_value(name, f),
}
}
}
// Used to compile static component interface. This is really annoying to do, since
// for static components, they only need to be guarded for %0, while for static
// groups, they need to be guarded for %[0:n]. This creates some annoying `if`
// statements.
impl Node {
// Helper to `preprocess_static_interface_assigns`
// Looks recursively thru guard to transform %[0:n] into %0 | %[1:n].
fn preprocess_static_interface_guard(
guard: ir::Guard<ir::StaticTiming>,
comp_sig: ir::RRC<ir::Cell>,
) -> ir::Guard<ir::StaticTiming> {
match guard {
ir::Guard::Info(st) => {
let (beg, end) = st.get_interval();
if beg == 0 {
// Replace %[0:n] -> (%0 & comp.go) | %[1:n]
// Cannot just do comp.go | %[1:n] because we want
// clients to be able to assert `comp.go` even after the first
// cycle w/o affecting correctness.
let first_cycle =
ir::Guard::Info(ir::StaticTiming::new((0, 1)));
let comp_go = guard!(comp_sig["go"]);
let first_and_go = ir::Guard::and(comp_go, first_cycle);
if end == 1 {
return first_and_go;
} else {
let after =
ir::Guard::Info(ir::StaticTiming::new((1, end)));
let cong = ir::Guard::or(first_and_go, after);
return cong;
}
}
guard
}
ir::Guard::And(l, r) => {
let left = Self::preprocess_static_interface_guard(
*l,
Rc::clone(&comp_sig),
);
let right =
Self::preprocess_static_interface_guard(*r, comp_sig);
ir::Guard::and(left, right)
}
ir::Guard::Or(l, r) => {
let left = Self::preprocess_static_interface_guard(
*l,
Rc::clone(&comp_sig),
);
let right =
Self::preprocess_static_interface_guard(*r, comp_sig);
ir::Guard::or(left, right)
}
ir::Guard::Not(g) => {
let a = Self::preprocess_static_interface_guard(*g, comp_sig);
ir::Guard::Not(Box::new(a))
}
_ => guard,
}
}
// Looks recursively thru assignment's guard to %[0:n] into %0 | %[1:n].
pub fn preprocess_static_interface_assigns(
assign: &mut ir::Assignment<ir::StaticTiming>,
comp_sig: ir::RRC<ir::Cell>,
) {
assign
.guard
.update(|g| Self::preprocess_static_interface_guard(g, comp_sig));
}
}
// The following are just standard `getter` methods.
impl Node {
/// Take the assignments of the root of the tree and return them.
/// This only works on a single node (i.e., the `Tree`` variant).
pub fn take_root_assigns(&mut self) -> Vec<ir::Assignment<Nothing>> {
match self {
Node::Single(single_node) => {
std::mem::take(&mut single_node.root.1)
}
Node::Par(_) => {
unreachable!(
"Cannot take root assignments of Node::Par variant"
)
}
}
}
/// Get the name of the root of the tree and return them.
/// This only works on a single node (i.e., the `Tree`` variant).
pub fn get_root_name(&mut self) -> ir::Id {
match self {
Node::Single(single_node) => single_node.root.0,
Node::Par(_) => {
unreachable!("Cannot take root name of Node::Par variant")
}
}
}
/// Get the name of the group at the root of the tree (if a `Tree` variant) or
/// of the equivalent `par` group (i.e., the name of the group that triggers
/// execution of all the trees) if a `Par` variant.
pub fn get_group_name(&self) -> ir::Id {
match self {
Node::Single(single_node) => single_node.root.0,
Node::Par(par_nodes) => par_nodes.group_name,
}
}
/// Gets latency of the overall tree.
pub fn get_latency(&self) -> u64 {
match self {
Node::Single(single_node) => single_node.latency,
Node::Par(par_nodes) => par_nodes.latency,
}
}
/// Gets the children of root of the tree (if a `Tree` variant) or
/// of the threads (i.e., trees) that are scheduled to execute (if a `Par`
/// variant.)
pub fn get_children(&mut self) -> &mut Vec<(Node, (u64, u64))> {
match self {
Node::Single(single_node) => &mut single_node.children,
Node::Par(par_nodes) => &mut par_nodes.threads,
}
}
/// Get number of repeats.
fn get_num_repeats(&self) -> u64 {
match self {
Node::Single(single_node) => single_node.num_repeats,
Node::Par(par_nodes) => par_nodes.num_repeats,
}
}
}
/// `SingleNode` struct.
pub struct SingleNode {
/// latency of one iteration.
pub latency: u64,
/// number of repeats. (So "total" latency = `latency` x `num_repeats`)
pub num_repeats: u64,
/// number of states in this node
pub num_states: u64,
/// (name of static group, assignments to build a corresponding dynamic group)
pub root: (ir::Id, Vec<ir::Assignment<Nothing>>),
/// maps cycles (i,j) -> fsm state type.
/// Here is an example FSM schedule:
/// Cycles FSM State (i.e., `fsm.out`)
/// (0..10) -> Normal[0,10)
/// (10..30) -> Offload(10) // Offloading to child
/// (30..40) -> Normal[11, 21)
/// (40,80) -> Offload(21)
/// (80,100)-> Normal[22, 42)
pub fsm_schedule: BTreeMap<(u64, u64), StateType>,
/// vec of (Node Object, cycles for which that child is executing).
/// Note that you can build `fsm_schedule` from just this information,
/// but it's convenient to have `fsm_schedule` avaialable.
pub children: Vec<(Node, (u64, u64))>,
/// Keep track of where we are within a single iteration.
/// If `latency` == 1, then we don't need an `fsm_cell`.
pub fsm_cell: Option<ir::RRC<StaticFSM>>,
/// Keep track of which iteration we are on. If iteration count == 1, then
/// we don't need an `iter_count_cell`.
pub iter_count_cell: Option<ir::RRC<StaticFSM>>,
}
impl SingleNode {
/// Instantiates the necessary registers.
/// Because we share FSM registers, it's possible that this register has already
/// been instantiated.
/// Therefore we take in a bunch of data structures to keep track of coloring:
/// - `coloring` that maps group names -> colors,
/// - `colors_to_max_values` which maps colors -> (max latency, max_num_repeats)
/// (we need to make sure that when we instantiate a color,
/// we give enough bits to support the maximum latency/num_repeats that will be
/// used for that color)
/// - `colors_to_fsm`
/// which maps colors to (fsm_register, iter_count_register): fsm_register counts
/// up for a single iteration, iter_count_register counts the number of iterations
/// that have passed.
///
/// Note that it is not always necessary to instantiate one or both registers (e.g.,
/// if num_repeats == 1 then you don't need an iter_count_register).
///
/// `one_hot_cutoff` is the cutoff to choose between binary and one hot encoding.
/// Any number of states greater than the cutoff will get binary encoding.
fn instantiate_fsms(
&mut self,
builder: &mut ir::Builder,
coloring: &HashMap<ir::Id, ir::Id>,
colors_to_max_values: &HashMap<ir::Id, (u64, u64)>,
colors_to_fsm: &mut HashMap<
ir::Id,
(OptionalStaticFSM, OptionalStaticFSM),
>,
one_hot_cutoff: u64,
) {
// Get color assigned to this node.
let color = coloring.get(&self.root.0).expect("couldn't find group");
// Check if we've already instantiated the registers for this color.
match colors_to_fsm.get(color) {
// We need to create the registers for the colors.
None => {
// First we get the maximum num_states and num_repeats
// for this color so we know how many bits each register needs.
let (num_states, num_repeats) = colors_to_max_values
.get(color)
.expect("Couldn't find color");
// Only need a `self.fsm_cell` if num_states > 1.
if *num_states != 1 {
// Choose encoding based on one_hot_cutoff.
let encoding = if *num_states > one_hot_cutoff {
FSMEncoding::Binary
} else {
FSMEncoding::OneHot
};
let fsm_cell = ir::rrc(StaticFSM::from_basic_info(
*num_states,
encoding,
builder,
));
self.fsm_cell = Some(fsm_cell);
}
// Only need a `self.iter_count_cell` if num_states > 1.
if *num_repeats != 1 {
let encoding = if *num_repeats > one_hot_cutoff {
FSMEncoding::Binary
} else {
FSMEncoding::OneHot
};
let repeat_counter = ir::rrc(StaticFSM::from_basic_info(
*num_repeats,
encoding,
builder,
));
self.iter_count_cell = Some(repeat_counter);
}
// Insert into `colors_to_fsms` so the next time we call this method
// we see we've already instantiated the registers.
colors_to_fsm.insert(
*color,
(
self.fsm_cell.as_ref().map(Rc::clone),
self.iter_count_cell.as_ref().map(Rc::clone),
),
);
}
Some((fsm_option, repeat_option)) => {
// Trivially assign to `self.fsm_cell` and `self.iter_count_cell` since
// we've already created it.
self.fsm_cell = fsm_option.as_ref().map(Rc::clone);
self.iter_count_cell = repeat_option.as_ref().map(Rc::clone);
}
}
// Recursively instantiate fsms for all the children.
for (child, _) in &mut self.children {
child.instantiate_fsms(
builder,
coloring,
colors_to_max_values,
colors_to_fsm,
one_hot_cutoff,
);
}
}
/// Counts to n.
/// If `incr_start_cond.is_some()`, then we will add it as an extra
/// guard guarding the 0->1 transition.
fn count_to_n(
&mut self,
builder: &mut ir::Builder,
incr_start_cond: Option<ir::Guard<Nothing>>,
) {
// res_vec will contain the assignments that count to n.
let mut res_vec: Vec<ir::Assignment<Nothing>> = Vec::new();
// Only need to count up to n if self.num_states > 1.
// If self.num_states == 1, then either a) latency is 1 cycle or b)
// we're just offloading the entire time (so the child will count).
// Either way, there's no need to instantiate a self.fsm_cell to count.
if self.num_states > 1 {
// `offload_states` are the fsm_states that last >1 cycles (i.e., states
// where children are executing, unless the child only lasts one cycle---
// then we can discount it as an "offload" state).
let offload_states: Vec<u64> = self
.fsm_schedule
.iter()
.filter_map(|((beg, end), state_type)| match state_type {
StateType::Normal(_) => None,
StateType::Offload(offload_state) => {
// Only need to include the children that last more than one cycle.
if beg + 1 == *end {
None
} else {
Some(*offload_state)
}
}
})
.collect();
// There are two conditions under which we increment the FSM.
// 1) Increment when we are NOT in an offload state
// 2) Increment when we ARE in an offload state, but the child being offloaded
// is in its final state. (intuitively, we need to increment because
// we know the control is being passed back to parent in the next cycle).
// (when we are in the final state, we obviously should not increment:
// we should reset back to 0.)
let parent_fsm = Rc::clone(
self.fsm_cell
.as_mut()
.expect("should have set self.fsm_cell"),
);
// Build an adder to increment the parent fsm.
let (adder_asssigns, adder) =
parent_fsm.borrow_mut().build_incrementer(builder);
res_vec.extend(adder_asssigns);
// Handle situation 1). Increment when we are NOT in an offload state
res_vec.extend(self.increment_if_not_offloading(
incr_start_cond.clone(),
&offload_states,
Rc::clone(&adder),
Rc::clone(&parent_fsm),
builder,
));
// Handle situation 2): Increment when we ARE in an offload state
// but the child being offloaded is in its final state.
res_vec.extend(self.increment_if_child_final_state(
&offload_states,
adder,
Rc::clone(&parent_fsm),
builder,
));
// Reset the FSM when it is at its final fsm_state.
let final_fsm_state =
self.get_fsm_query((self.latency - 1, self.latency), builder);
res_vec.extend(
parent_fsm
.borrow_mut()
.conditional_reset(final_fsm_state, builder),
);
}
// If self.num_states > 1, then it's guaranteed that self.latency > 1.
// However, even if self.num_states == 1, self.latency might still be
// greater than 1 if we're just offloading the computation for the entire time.
// In this case, we still need the children to count to n.
if self.latency > 1 {
for (child, (beg, end)) in self.children.iter_mut() {
// If beg == 0 and end > 1 then we need to "transfer" the incr_start_condition
// to the child so it guards the 0->1 transition.
let cond = if *beg == 0 && *end > 1 {
incr_start_cond.clone()
} else {
None
};
// Recursively call `count_to_n`
child.count_to_n(builder, cond);
}
}
// Handle repeats (i.e., make sure we actually interate `self.num_repeats` times).
if self.num_repeats != 1 {
// If self.latency == 10, then we should increment the self.iter_count_cell
// each time fsm == 9, i.e., `final_fsm_state`.
let final_fsm_state =
self.get_fsm_query((self.latency - 1, self.latency), builder);
// `repeat_fsm` store number of iterations.
let repeat_fsm = Rc::clone(
self.iter_count_cell
.as_mut()
.expect("should have set self.iter_count_cell"),
);
// Build an incrementer to increment `self.iter_count_cell`.
let (repeat_adder_assigns, repeat_adder) =
repeat_fsm.borrow_mut().build_incrementer(builder);
// We shouldn't increment `self.iter_count_cell` if we are in the final iteration:
// we should reset it instead.
let final_repeat_state = *repeat_fsm.borrow_mut().query_between(
builder,
(self.num_repeats - 1, self.num_repeats),
);
let not_final_repeat_state = final_repeat_state.clone().not();
res_vec.extend(repeat_adder_assigns);
// Incrementing self.iter_count_cell when appropriate.
res_vec.extend(repeat_fsm.borrow_mut().conditional_increment(
final_fsm_state.clone().and(not_final_repeat_state),
repeat_adder,
builder,
));
// Resetting self.iter_count_cell when appropriate.
res_vec.extend(repeat_fsm.borrow_mut().conditional_reset(
final_fsm_state.clone().and(final_repeat_state),
builder,
));
}
// Extend root assigns to include `res_vec` (which counts to n).
self.root.1.extend(res_vec);
}
/// Helper to `count_to_n`
/// Increment when we are NOT in an offload state
/// e.g., if `offload_states` == [2,4,6] then
/// We should increment when !(fsm == 2 | fsm == 4 | fsm == 6).
/// There are a couple corner cases we need to think about (in particular,
/// we should guard the 0->1 transition differently if `incr_start_cond` is
/// some(), and we should reset rather than increment when we are in the final
/// fsm state).
fn increment_if_not_offloading(
&mut self,
incr_start_cond: Option<ir::Guard<Nothing>>,
offload_states: &[u64],
adder: ir::RRC<ir::Cell>,
parent_fsm: ir::RRC<StaticFSM>,
builder: &mut ir::Builder,
) -> Vec<ir::Assignment<Nothing>> {
let mut res_vec = vec![];
let mut offload_state_guard: ir::Guard<Nothing> =
ir::Guard::Not(Box::new(ir::Guard::True));
for offload_state in offload_states {
// Creating a guard that checks whether the parent fsm is
// in an offload state.
offload_state_guard.update(|g| {
g.or(*parent_fsm.borrow_mut().query_between(
builder,
(*offload_state, offload_state + 1),
))
});
}
let not_offload_state = offload_state_guard.not();
let mut incr_guard = not_offload_state;
// If incr_start_cond.is_some(), then we have to specially take into
// account this scenario when incrementing the FSM.
if let Some(g) = incr_start_cond.clone() {
// If we offload during the transition from cycle 0->1 transition
// then we don't need a special first transition guard.
// (we will make sure the child will add this guard when
// it is counting to n.)
if let Some(((beg, end), state_type)) =
self.fsm_schedule.iter().next()
{
if !(matches!(state_type, StateType::Offload(_))
&& *beg == 0
&& *end > 1)
{
let first_state = self.get_fsm_query((0, 1), builder);
// We must handle the 0->1 transition separately.
// fsm.in = fsm == 0 & incr_start_cond ? fsm + 1;
// fsm.write_en = fsm == 0 & incr_start_cond ? 1'd1;
res_vec.extend(
parent_fsm.borrow_mut().conditional_increment(
first_state.clone().and(g),
Rc::clone(&adder),
builder,
),
);
// We also have to add fsm != 0 to incr_guard since
// we have just added assignments to handle this situation
// separately
incr_guard = incr_guard.and(first_state.not())
}
}
};
// We shouldn't increment when we are in the final state
// (we should be resetting instead).
// So we need to `& !in_final_state` to the guard.
let final_fsm_state =
self.get_fsm_query((self.latency - 1, self.latency), builder);
let not_final_state = final_fsm_state.not();
// However, if the final state is an offload state, then there's no need
// to make this extra check of not being in the last state.
if let Some((_, (_, end_final_child))) = self.children.last() {
// If the final state is not an offload state, then
// we need to add this check.
if *end_final_child != self.latency {
incr_guard = incr_guard.and(not_final_state);
}
} else {
// Also, if there is just no offloading, then we need to add this check.
incr_guard = incr_guard.and(not_final_state);
};
// Conditionally increment based on `incr_guard`
res_vec.extend(parent_fsm.borrow_mut().conditional_increment(
incr_guard,
Rc::clone(&adder),
builder,
));
res_vec
}
/// Helper to `count_to_n`
/// Increment when we ARE in an offload state, but the child being
/// offloaded is in its final state.
fn increment_if_child_final_state(
&mut self,
offload_states: &[u64],
adder: ir::RRC<ir::Cell>,
parent_fsm: ir::RRC<StaticFSM>,
builder: &mut ir::Builder,
) -> Vec<ir::Assignment<Nothing>> {
let mut res_vec = vec![];
for (i, (child, (_, end))) in self
.children
.iter_mut()
// If child only lasts a single cycle, then we can just unconditionally increment.
// We handle that case above (since `offload_states` only includes children that
// last more than one cycle).
.filter(|(_, (beg, end))| beg + 1 != *end)
.enumerate()
{
// We need to increment parent when child is in final state.
// For example, if the parent is offloading to `child_x` when it
// is in state 5, the guard would look like
// fsm.in = fsm == 5 && child_x_fsm_in_final_state ? fsm + 1;
// fsm.write_en == 5 && child_x_fsm_in_final_state ? 1'd1;
// The one exception:
// If the offload state is the last state (end == self.latency) then we don't
// increment, we need to reset to 0 (which we handle separately).
if *end != self.latency {
// We know each offload state corresponds to exactly one child.
let child_state = offload_states[i];
// Checking that we are in child state, e.g., `(fsm == 5)`
// in the above example.
let in_child_state = parent_fsm
.borrow_mut()
.query_between(builder, (child_state, child_state + 1));
// now we need to check `child_fsm_in_final_state`
let total_child_latency =
child.get_latency() * child.get_num_repeats();
let child_final_state = child.query_between(
(total_child_latency - 1, total_child_latency),
builder,
);
// Conditionally increment when `fsm==5 & child_final_state`
let parent_fsm_incr =
parent_fsm.borrow_mut().conditional_increment(
in_child_state.and(child_final_state),
Rc::clone(&adder),
builder,
);
res_vec.extend(parent_fsm_incr);
}
}
res_vec
}
/// `Realize` each static group in the tree into a dynamic group.
/// In particular, this involves converting %[i:j] guards into actual
/// fsm register queries (which can get complicated with out tree structure:
/// it's not just i <= fsm < j anymore).
///
/// `reset_early_map`, `fsm_info_map`, and `group_rewrites` are all
/// metadata to make it more easier later on when we are rewriting control,
/// adding wrapper groups when necessary, etc.
fn realize(
&mut self,
ignore_timing_guards: bool,
static_groups: &Vec<ir::RRC<ir::StaticGroup>>,
reset_early_map: &mut HashMap<ir::Id, ir::Id>,
fsm_info_map: &mut HashMap<
ir::Id,
(ir::Id, ir::Guard<Nothing>, ir::Guard<Nothing>),
>,
group_rewrites: &mut ir::rewriter::PortRewriteMap,
builder: &mut ir::Builder,
) {
// Get static group we are "realizing".
let static_group = Rc::clone(
static_groups
.iter()
.find(|sgroup| sgroup.borrow().name() == self.root.0)
.expect("couldn't find static group"),
);
// Create the dynamic "early reset group" that will replace the static group.
let static_group_name = static_group.borrow().name();
let mut early_reset_name = static_group_name.to_string();
early_reset_name.insert_str(0, "early_reset_");
let early_reset_group = builder.add_group(early_reset_name);
// Realize the static %[i:j] guards to fsm queries.
// *This is the most of the difficult thing the function does*.
// This is significantly more complicated with a tree structure.
let mut assigns = static_group
.borrow()
.assignments
.clone()
.into_iter()
.map(|assign| {
self.make_assign_dyn(
assign,
false,
ignore_timing_guards,
builder,
)
})
.collect_vec();
// Add assignment `group[done] = ud.out`` to the new group.
structure!( builder; let ud = prim undef(1););
let early_reset_done_assign = build_assignments!(
builder;
early_reset_group["done"] = ? ud["out"];
);
assigns.extend(early_reset_done_assign);
// Adding the assignments of `self.root` (mainly the `count_to_n`
// assignments).
assigns.extend(std::mem::take(&mut self.root.1));
self.root.1 = assigns.clone();
early_reset_group.borrow_mut().assignments = assigns;
early_reset_group.borrow_mut().attributes =
static_group.borrow().attributes.clone();
// Now we have to update the fields with a bunch of information.
// This makes it easier when we have to build wrappers, rewrite ports, etc.
// Map the static group name -> early reset group name.
reset_early_map
.insert(static_group_name, early_reset_group.borrow().name());
// self.group_rewrite_map helps write static_group[go] to early_reset_group[go]
// Technically we could do this w/ early_reset_map but is easier w/
// group_rewrite, which is explicitly of type `PortRewriterMap`
group_rewrites.insert(
ir::Canonical::new(static_group_name, ir::Id::from("go")),
early_reset_group.borrow().find("go").unwrap_or_else(|| {
unreachable!(
"group {} has no go port",
early_reset_group.borrow().name()
)
}),
);
let fsm_identifier = match self.fsm_cell.as_ref() {
// If the tree does not have an fsm cell, then we can err on the
// side of giving it its own unique identifier.
None => self.root.0,
Some(fsm_rc) => fsm_rc.borrow().get_unique_id(),
};
let total_latency = self.latency * self.num_repeats;
fsm_info_map.insert(
early_reset_group.borrow().name(),
(
fsm_identifier,
self.query_between((0, 1), builder),
self.query_between((total_latency - 1, total_latency), builder),
),
);
// Recursively realize each child.
self.children.iter_mut().for_each(|(child, _)| {
child.realize(
ignore_timing_guards,
static_groups,
reset_early_map,
fsm_info_map,
group_rewrites,
builder,
)
})
}
// Rephrasing an (i,j) query: this breaks up the guard and makes it easier
// to figure out what logic we need to instantiate to perform the query.
// Restructure an (i,j) query into:
// (beg, middle, end) query.
// This is best explained by example.
// Suppose latency = 5, num repeats = 10.
// Suppose we query %[3:32].
// beg = Some(0, 3-5). 0 bc we are on the 0th iteration,
// and only cycles 3-5 of that iteration.
// middle = Some([1,6)). These are the iterations for which the query is true
// throughout the entire iteration.
// end = Some(6,0-2). 6 bc 6th iteration, 0-2 because only cycles 0-2 of that
// iteration.
fn restructure_query(
&self,
query: (u64, u64),
) -> (
Option<SingleIterQuery>,
Option<ItersQuery>,
Option<SingleIterQuery>,
) {
// Splitting the query into an fsm query and and iteration query.
// (beg_iter_query, end_iter_query) is an inclusive (both sides) query
// on the iterations we are active for.
// (beg_fsm_query, end_fsm_query) is the fsm query we should be supporting.
let (beg_query, end_query) = query;
let (beg_iter_query, beg_fsm_query) =
(beg_query / self.latency, beg_query % self.latency);
let (end_iter_query, mut end_fsm_query) =
((end_query - 1) / self.latency, (end_query) % self.latency);
if end_fsm_query == 0 {
end_fsm_query = self.latency;
}
// Scenario 1: the query spans only a single iteration.
// In this case, we set beg_query to
// `Some(<that single iteration>, (beg_fsm_query->end_fsm_query))``
// and set middle and end to None.
if beg_iter_query == end_iter_query {
let repeat_query = beg_iter_query;
let fsm_query = (beg_fsm_query, end_fsm_query);
let res = Some((repeat_query, fsm_query));
(res, None, None)
}
// Scenario 2: the query spans only 2 iterations.
// In this case, we only need a beg_query and an end_query, but no
// middle query.
else if beg_iter_query + 1 == end_iter_query {
let middle_res = None;
let repeat_query0 = beg_iter_query;
// We know the beg_query stretches into the next iteration,
// so we can end it at self.latency.
let fsm_query0 = (beg_fsm_query, self.latency);
let beg_res = Some((repeat_query0, fsm_query0));
let repeat_query1 = end_iter_query;
// We know the end_query stretches backwards into the previous iteration,
// so we can start it at 0.
let fsm_query1 = (0, end_fsm_query);
let end_res = Some((repeat_query1, fsm_query1));
(beg_res, middle_res, end_res)
}
// Scenario 3: the query spans 3 or more iterations.
// In this case, we need the middle_query for the middle iterations,
// and the beg and end queries for (parts of) the
// first and last iterations for this query.
else {
let mut unconditional_repeat_query =
(beg_iter_query + 1, end_iter_query);
let repeat_query0 = beg_iter_query;
// We know the beg_query stretches into the next iteration,
// so we can end it at self.latency.
let fsm_query0 = (beg_fsm_query, self.latency);
let mut beg_res = Some((repeat_query0, fsm_query0));
// if beg_fsm_query == 0, then beg_query spans the entire iterations,
// so we can just add it the unconditional_repeat_query (i.e., the middle_query).
if beg_fsm_query == 0 {
beg_res = None;
unconditional_repeat_query.0 -= 1;
}
let repeat_query1 = end_iter_query;
// We know the end_query stretches backwards into the previous iteration,
// so we can start it at 0.
let fsm_query1 = (0, end_fsm_query);
let mut end_res = Some((repeat_query1, fsm_query1));
// If end_fsm_query == self.latency, then end_res spans the entire iterations,
// so we can just add it the unconditional_repeat_query (i.e., the middle_query).
if end_fsm_query == self.latency {
end_res = None;
unconditional_repeat_query.1 += 1;
}
(beg_res, Some(unconditional_repeat_query), end_res)
}
}
// Given query (i,j), get the fsm query for cycles (i,j).
// Does NOT check the iteration number.
// This is greatly complicated by the offloading to children.
// We use a resturcturing that organizes the query into (beg, middle, end),
// similar to (but not the same as) self.restructure_query().
fn get_fsm_query(
&mut self,
query: (u64, u64),
builder: &mut ir::Builder,
) -> ir::Guard<Nothing> {
// If guard is true the entire execution, then return `true`.
if 0 == query.0 && self.latency == query.1 {
return ir::Guard::True;
}
let fsm_cell_opt = self.fsm_cell.as_ref();
if fsm_cell_opt.is_none() {
// If there is no fsm cell even though latency > 1, then we must
// have offloaded the entire latency. Therefore we just need
// to query the child.
assert!(self.offload_entire_latency());
let (only_child, _) = self.children.iter_mut().next().unwrap();
return only_child.query_between(query, builder);
}
let fsm_cell: Rc<std::cell::RefCell<StaticFSM>> =
Rc::clone(fsm_cell_opt.expect("just checked if None"));
let (query_beg, query_end) = query;
let mut beg_interval = ir::Guard::True.not();
let mut end_interval = ir::Guard::True.not();
let mut middle_interval = None;
let mut child_index = 0;
// Suppose fsm_schedule = Cycles FSM State (i.e., `fsm.out`)
// (0..10) -> Normal[0,10)
// (10..30) -> Offload(10) // Offloading to child
// (30..40) -> Normal[11, 21)
// (40,80) -> Offload(21)
// (80,100)-> Normal[22, 42)
// And query = (15,95).
// Then at the end of the following `for` loop we want:
// `beg_interval` should be fsm == 10 && <child.query_between(5,20)>
// `middle_interval` should be (11, 22)
// `end_interval` should be 22 <= fsm < 37
for ((beg, end), state_type) in self.fsm_schedule.iter() {
// Check if the query encompasses the entire interval.
// If so, we add it to the "middle" interval.
if query_beg <= *beg && *end <= query_end {
// Get the interval we have to add, based on `state_type`.
let interval_to_add = match state_type {
StateType::Normal(fsm_interval) => *fsm_interval,
StateType::Offload(offload_state) => {
(*offload_state, offload_state + 1)
}
};
// Add `interval_to_add` to `middle_interval`.
match middle_interval {
None => middle_interval = Some(interval_to_add),
Some((cur_start, cur_end)) => {
assert!(cur_end == interval_to_add.0);
middle_interval = Some((cur_start, interval_to_add.1));
}
}
}
// Otherwise check if the beginning of the query lies within the
// interval (This should only happen once). Add to `beg_interval`.
else if *beg <= query_beg && query_beg < *end {
assert!(beg_interval.is_false());
// This is the query, but relativized to the start of the current interval.
let relative_query = (query_beg - beg, query_end - beg);
match state_type {
// If we are not offloading, then we can just produce a normal
// query.
StateType::Normal((beg_fsm_interval, end_fsm_interval)) => {
let translated_query = (
beg_fsm_interval + relative_query.0,
// This query either stretches into the next interval, or
// ends within the interval: we want to capture both of these choices.
std::cmp::min(
beg_fsm_interval + relative_query.1,
*end_fsm_interval,
),
);
beg_interval = *fsm_cell
.borrow_mut()
.query_between(builder, translated_query);
}
// If we are not offloading, then we first check the state,
// then we must query the corresponding child.
StateType::Offload(offload_state) => {
let in_offload_state =
*fsm_cell.borrow_mut().query_between(
builder,
(*offload_state, offload_state + 1),
);
let (child, _) =
self.children.get_mut(child_index).unwrap();
let child_query = child.query_between(
(
relative_query.0,
// This query either stretches into another interval, or
// ends within the interval: we want to capture both of these choices.
std::cmp::min(
relative_query.1,
child.get_latency()
* child.get_num_repeats(),
),
),
builder,
);
beg_interval = in_offload_state.and(child_query);
}
};
}
// Check if the end of the query lies within the
// interval (This should only happen once) Add to `end_interval`.
else if *beg < query_end && query_end <= *end {
// We only need the end of the relative query.
// If we try to get the beginning then we could get overflow error.
let relative_query_end = query_end - beg;
assert!(end_interval.is_false());
match state_type {
StateType::Normal((beg_fsm_interval, _)) => {
end_interval = *fsm_cell.borrow_mut().query_between(
builder,
// This query must stretch backwards into a preiouvs interval
// Otherwise it would have been caught by the
// So beg_fsm_interval is a safe start.
(
*beg_fsm_interval,
beg_fsm_interval + relative_query_end,
),
);
}
StateType::Offload(offload_state) => {
let in_offload_state =
*fsm_cell.borrow_mut().query_between(
builder,
(*offload_state, offload_state + 1),
);
let (child, _) =
self.children.get_mut(child_index).unwrap();
// We know this must stretch backwards
// into a previous interval: otherwise, it
// would have been caught by the previous elif condition.
// therefore, we can start the child query at 0.
let child_query = child
.query_between((0, relative_query_end), builder);
end_interval = in_offload_state.and(child_query);
}
};
}
if matches!(state_type, StateType::Offload(_)) {
child_index += 1;
}
}
// Turn `middle_interval` into an actual `ir::Guard`.
let middle_query = match middle_interval {
None => Box::new(ir::Guard::True.not()),
Some((i, j)) => self
.fsm_cell
.as_mut()
.unwrap()
.borrow_mut()
.query_between(builder, (i, j)),
};
beg_interval.or(end_interval.or(*middle_query))
}
// Produces a guard that checks whether query.0 <= self.iter_count_cell < query.1
fn get_repeat_query(
&mut self,
query: (u64, u64),
builder: &mut ir::Builder,
) -> Box<ir::Guard<Nothing>> {
// If self.num_repeats == 1, then no need for a complicated query.
match self.num_repeats {
1 => {
assert!(query.0 == 0 && query.1 == 1);
Box::new(ir::Guard::True)
}
_ => self
.iter_count_cell
.as_mut()
.expect("querying repeat implies cell exists")
.borrow_mut()
.query_between(builder, (query.0, query.1)),
}
}
// Produce a guard that checks:
// - whether iteration == repeat_query AND
// - whether %[fsm_query.0:fsm_query.1]
fn check_iteration_and_fsm_state(
&mut self,
(repeat_query, fsm_query): (u64, (u64, u64)),
builder: &mut ir::Builder,
) -> ir::Guard<Nothing> {
let fsm_guard = self.get_fsm_query(fsm_query, builder);
// Checks `self.iter_count_cell`.
let counter_guard =
self.get_repeat_query((repeat_query, repeat_query + 1), builder);
ir::Guard::And(Box::new(fsm_guard), counter_guard)
}
// Converts a %[i:j] query into a query of `self`'s and its childrens
// iteration registers.
fn query_between(
&mut self,
query: (u64, u64),
builder: &mut ir::Builder,
) -> ir::Guard<Nothing> {
// See `restructure_query` to see what we're doing.
// But basically:
// beg_iter_query = Option(iteration number, cycles during that iteration the query is true).
// middle_iter_query = Option(iterations during which the query is true the entire iteration).
// end_iter_query = Option(iteration number, cycles during that iteration the query is true).
let (beg_iter_query, middle_iter_query, end_iter_query) =
self.restructure_query(query);
// Call `check_iteration_and_fsm_state` for beg and end queries.
let g0 = match beg_iter_query {
None => ir::Guard::True.not(),
Some(q0) => self.check_iteration_and_fsm_state(q0, builder),
};
let g1 = match end_iter_query {
None => ir::Guard::True.not(),
Some(q1) => self.check_iteration_and_fsm_state(q1, builder),
};
// Call `get_repeat_query` for middle_iter_queries.
let rep_query = match middle_iter_query {
None => Box::new(ir::Guard::True.not()),
Some(rq) => self.get_repeat_query(rq, builder),
};
g0.or(g1.or(*rep_query))
}
// Takes in a static guard `guard`, and returns equivalent dynamic guard
// The only thing that actually changes is the Guard::Info case
// We need to turn static_timing to dynamic guards using `fsm`.
// See `make_assign_dyn` for explanations of `global_view` and `ignore_timing`
// variable.
fn make_guard_dyn(
&mut self,
guard: ir::Guard<ir::StaticTiming>,
global_view: bool,
ignore_timing: bool,
builder: &mut ir::Builder,
) -> Box<ir::Guard<Nothing>> {
match guard {
ir::Guard::Or(l, r) => Box::new(ir::Guard::Or(
self.make_guard_dyn(*l, global_view, ignore_timing, builder),
self.make_guard_dyn(*r, global_view, ignore_timing, builder),
)),
ir::Guard::And(l, r) => Box::new(ir::Guard::And(
self.make_guard_dyn(*l, global_view, ignore_timing, builder),
self.make_guard_dyn(*r, global_view, ignore_timing, builder),
)),
ir::Guard::Not(g) => Box::new(ir::Guard::Not(self.make_guard_dyn(
*g,
global_view,
ignore_timing,
builder,
))),
ir::Guard::CompOp(op, l, r) => {
Box::new(ir::Guard::CompOp(op, l, r))
}
ir::Guard::Port(p) => Box::new(ir::Guard::Port(p)),
ir::Guard::True => Box::new(ir::Guard::True),
ir::Guard::Info(static_timing) => {
// If `ignore_timing` is true, then just return a true guard.
if ignore_timing {
assert!(static_timing.get_interval() == (0, 1));
return Box::new(ir::Guard::True);
}
if global_view {
// For global_view we call `query_between`
Box::new(
self.query_between(
static_timing.get_interval(),
builder,
),
)
} else {
// For local_view we call `get_fsm_query`
Box::new(
self.get_fsm_query(
static_timing.get_interval(),
builder,
),
)
}
}
}
}
/// Takes in static assignment `assign` and returns a dynamic assignments
/// For example, it could transform the guard %[2:3] -> fsm.out >= 2 & fsm.out <= 3
/// `global_view`: are you just querying for a given iteration, or are
/// you querying for the entire tree's execution?
/// - if `global_view` is true, then you have to include the iteration
/// count register in the assignment's guard.
/// - if `global_view` is false, then you dont' have to include it
///
/// `ignore_timing`: remove static timing guards instead of transforming them
/// into an FSM query. Note that in order to do this, the timing guard must
/// equal %[0:1], otherwise we will throw an error. This option is here
/// mainly to save resource usage.
pub fn make_assign_dyn(
&mut self,
assign: ir::Assignment<ir::StaticTiming>,
global_view: bool,
ignore_timing: bool,
builder: &mut ir::Builder,
) -> ir::Assignment<Nothing> {
ir::Assignment {
src: assign.src,
dst: assign.dst,
attributes: assign.attributes,
guard: self.make_guard_dyn(
*assign.guard,
global_view,
ignore_timing,
builder,
),
}
}
// Helper function: checks
// whether the tree offloads its entire latency, and returns the
// appropriate `bool`.
fn offload_entire_latency(&self) -> bool {
self.children.len() == 1
&& self
.children
.iter()
.any(|(_, (beg, end))| *beg == 0 && *end == self.latency)
// This last check is prob unnecessary since it follows from the first two.
&& self.num_states == 1
}
}
/// These methods handle adding conflicts to the tree (to help coloring for
/// sharing FSMs)
impl SingleNode {
// Get names of groups corresponding to all nodes
pub fn get_all_nodes(&self) -> Vec<ir::Id> {
let mut res = vec![self.root.0];
for (child, _) in &self.children {
res.extend(child.get_all_nodes())
}
res
}
// Adds conflicts between children and any descendents.
// Also add conflicts between any overlapping children. XXX(Caleb): normally
// there shouldn't be overlapping children, but when we are doing the traditional
// method in we don't offload (and therefore don't need this tree structure)
// I have created dummy trees for the sole purpose of drawing conflicts
pub fn add_conflicts(&self, conflict_graph: &mut GraphColoring<ir::Id>) {
let root_name = self.root.0;
for (child, _) in &self.children {
for sgroup in &child.get_all_nodes() {
conflict_graph.insert_conflict(&root_name, sgroup);
}
child.add_conflicts(conflict_graph);
}
// Adding conflicts between overlapping children.
for ((child_a, (beg_a, end_a)), (child_b, (beg_b, end_b))) in
self.children.iter().tuple_combinations()
{
// Checking if children overlap: either b begins within a, it
// ends within a, or it encompasses a's entire interval.
if ((beg_a <= beg_b) & (beg_b < end_a))
| ((beg_a < end_b) & (end_b <= end_a))
| (beg_b <= beg_a && end_a <= end_b)
{
// Adding conflicts between all nodes of the children if
// the children overlap.
for a_node in child_a.get_all_nodes() {
for b_node in child_b.get_all_nodes() {
conflict_graph.insert_conflict(&a_node, &b_node);
}
}
}
}
}
// Gets max value according to some function f.
pub fn get_max_value<F>(&self, name: &ir::Id, f: &F) -> u64
where
F: Fn(&SingleNode) -> u64,
{
let mut cur_max = 0;
if self.root.0 == name {
cur_max = std::cmp::max(cur_max, f(self));
}
for (child, _) in &self.children {
cur_max = std::cmp::max(cur_max, child.get_max_value(name, f));
}
cur_max
}
}
/// Represents a group of `Nodes` that execute in parallel.
pub struct ParNodes {
/// Name of the `par_group` that fires off the threads
pub group_name: ir::Id,
/// Latency
pub latency: u64,
/// Num Repeats
pub num_repeats: u64,
/// (Thread, interval thread is active).
/// Interval thread is active should always start at 0.
pub threads: Vec<(Node, (u64, u64))>,
}
impl ParNodes {
/// Instantiates FSMs by recursively instantiating FSM for each thread.
pub fn instantiate_fsms(
&mut self,
builder: &mut ir::Builder,
coloring: &HashMap<ir::Id, ir::Id>,
colors_to_max_values: &HashMap<ir::Id, (u64, u64)>,
colors_to_fsm: &mut HashMap<
ir::Id,
(OptionalStaticFSM, OptionalStaticFSM),
>,
one_hot_cutoff: u64,
) {
for (thread, _) in &mut self.threads {
thread.instantiate_fsms(
builder,
coloring,
colors_to_max_values,
colors_to_fsm,
one_hot_cutoff,
);
}
}
/// Counts to N by recursively calling `count_to_n` on each thread.
pub fn count_to_n(
&mut self,
builder: &mut ir::Builder,
incr_start_cond: Option<ir::Guard<Nothing>>,
) {
for (thread, _) in &mut self.threads {
thread.count_to_n(builder, incr_start_cond.clone());
}
}
/// Realizes static groups into dynamic group.
pub fn realize(
&mut self,
ignore_timing_guards: bool,
static_groups: &Vec<ir::RRC<ir::StaticGroup>>,
reset_early_map: &mut HashMap<ir::Id, ir::Id>,
fsm_info_map: &mut HashMap<
ir::Id,
(ir::Id, ir::Guard<Nothing>, ir::Guard<Nothing>),
>,
group_rewrites: &mut ir::rewriter::PortRewriteMap,
builder: &mut ir::Builder,
) {
// Get static grouo we are "realizing".
let static_group = Rc::clone(
static_groups
.iter()
.find(|sgroup| sgroup.borrow().name() == self.group_name)
.expect("couldn't find static group"),
);
// Create the dynamic "early reset group" that will replace the static group.
let static_group_name = static_group.borrow().name();
let mut early_reset_name = static_group_name.to_string();
early_reset_name.insert_str(0, "early_reset_");
let early_reset_group = builder.add_group(early_reset_name);
// Get the longest node.
let longest_node = self.get_longest_node();
// If one thread lasts 10 cycles, and another lasts 5 cycles, then the par_group
// will look like this:
// static<10> group par_group {
// thread1[go] = 1'd1;
// thread2[go] = %[0:5] ? 1'd1;
// }
// Therefore the %[0:5] needs to be realized using the FSMs from thread1 (the
// longest FSM).
let mut assigns = static_group
.borrow()
.assignments
.clone()
.into_iter()
.map(|assign| {
longest_node.make_assign_dyn(
assign,
true,
ignore_timing_guards,
builder,
)
})
.collect_vec();
// Add assignment `group[done] = ud.out`` to the new group.
structure!( builder; let ud = prim undef(1););
let early_reset_done_assign = build_assignments!(
builder;
early_reset_group["done"] = ? ud["out"];
);
assigns.extend(early_reset_done_assign);
early_reset_group.borrow_mut().assignments = assigns;
early_reset_group.borrow_mut().attributes =
static_group.borrow().attributes.clone();
// Now we have to update the fields with a bunch of information.
// This makes it easier when we have to build wrappers, rewrite ports, etc.
// Map the static group name -> early reset group name.
// This is helpful for rewriting control
reset_early_map
.insert(static_group_name, early_reset_group.borrow().name());
// self.group_rewrite_map helps write static_group[go] to early_reset_group[go]
// Technically we could do this w/ early_reset_map but is easier w/
// group_rewrite, which is explicitly of type `PortRewriterMap`
group_rewrites.insert(
ir::Canonical::new(static_group_name, ir::Id::from("go")),
early_reset_group.borrow().find("go").unwrap_or_else(|| {
unreachable!(
"group {} has no go port",
early_reset_group.borrow().name()
)
}),
);
let fsm_identifier = match longest_node.fsm_cell.as_ref() {
// If the tree does not have an fsm cell, then we can err on the
// side of giving it its own unique identifier.
None => longest_node.root.0,
Some(fsm_rc) => fsm_rc.borrow().get_unique_id(),
};
let total_latency = self.latency * self.num_repeats;
fsm_info_map.insert(
early_reset_group.borrow().name(),
(
fsm_identifier,
self.query_between((0, 1), builder),
self.query_between((total_latency - 1, total_latency), builder),
),
);
// Recursively realize each child.
self.threads.iter_mut().for_each(|(child, _)| {
child.realize(
ignore_timing_guards,
static_groups,
reset_early_map,
fsm_info_map,
group_rewrites,
builder,
)
})
}
/// Recursively searches each thread to get the longest (in terms of
/// cycle counts) SingleNode.
pub fn get_longest_node(&mut self) -> &mut SingleNode {
let max = self.threads.iter_mut().max_by_key(|(child, _)| {
(child.get_latency() * child.get_num_repeats()) as i64
});
if let Some((max_child, _)) = max {
match max_child {
Node::Par(par_nodes) => par_nodes.get_longest_node(),
Node::Single(single_node) => single_node,
}
} else {
unreachable!("self.children is empty/no maximum value found");
}
}
/// Use the longest node to query between.
pub fn query_between(
&mut self,
query: (u64, u64),
builder: &mut ir::Builder,
) -> ir::Guard<Nothing> {
let longest_node = self.get_longest_node();
longest_node.query_between(query, builder)
}
}
/// Used to add conflicts for graph coloring for sharing FSMs.
/// See the equivalent SingleNode implementation for more details.
impl ParNodes {
pub fn get_all_nodes(&self) -> Vec<ir::Id> {
let mut res = vec![];
for (thread, _) in &self.threads {
res.extend(thread.get_all_nodes())
}
res
}
pub fn add_conflicts(&self, conflict_graph: &mut GraphColoring<ir::Id>) {
for ((thread1, _), (thread2, _)) in
self.threads.iter().tuple_combinations()
{
for sgroup1 in thread1.get_all_nodes() {
for sgroup2 in thread2.get_all_nodes() {
conflict_graph.insert_conflict(&sgroup1, &sgroup2);
}
}
thread1.add_conflicts(conflict_graph);
thread2.add_conflicts(conflict_graph);
}
}
pub fn get_max_value<F>(&self, name: &ir::Id, f: &F) -> u64
where
F: Fn(&SingleNode) -> u64,
{
let mut cur_max = 0;
for (thread, _) in &self.threads {
cur_max = std::cmp::max(cur_max, thread.get_max_value(name, f));
}
cur_max
}
}