Skip to content

Allocate registers for data_mov on the same tile #115

New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

Merged
tancheng merged 2 commits into from
Aug 19, 2025
Merged

Allocate registers for data_mov on the same tile #115

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a02df66
allocate registers for no-mov data_mov
ShangkunLi Aug 19, 2025
1605ed6
rename end_time -> exclusive_end_time
ShangkunLi Aug 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 38 additions & 0 deletions lib/NeuraDialect/Mapping/mapping_util.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -417,6 +417,44 @@ bool mlir::neura::tryRouteDataMove(Operation *mov_op, MappingLoc src_loc,
std::vector<MappingLoc> &path_out) {
// Specially handles the case where src and dst are the same tile.
if (src_loc.resource == dst_loc.resource) {
// When the source and destination are on the same tile, we still need to
// allocate registers for data movement.
Tile *tile = dyn_cast<Tile>(src_loc.resource);
if (!tile) {
llvm::errs() << "[tryRouteDataMove] Source is not a tile\n";
return false;
}

// Defines the time window for the register allocation.
int start_time = src_loc.time_step;
int exclusive_end_time = dst_loc.time_step;

// For backward moves, we need to adjust the end time.
if (is_backward_move) {
exclusive_end_time += state.getII();
}

// Finds a register that is available during start_time to
// exclusive_end_time.
Register *reg =
getAvailableRegister(state, tile, start_time, exclusive_end_time);
if (!reg) {
llvm::errs() << "[tryRouteDataMove] No available register found for "
<< "tile: " << tile->getId()
<< " from time step: " << start_time
<< " to time step: " << exclusive_end_time << "\n";
return false;
}

// Adds the register locations to the path.
for (int t = start_time; t < exclusive_end_time; ++t) {
MappingLoc reg_loc{reg, t};
path_out.push_back(reg_loc);
}

llvm::errs() << "[tryRouteDataMove] Allocated register " << reg->getId()
<< " for same-tile data movement from t=" << start_time
<< " to t=" << exclusive_end_time << "\n";
return true;
}
struct QueueEntry {
Expand Down
10 changes: 5 additions & 5 deletions test/controflow_fuse/simple_loop/simple_loop.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,11 +193,11 @@ module attributes {} {
// FUSE-MAPPING-NEXT: %6 = "neura.grant_once"() <{constant_value = 0 : i64}> {mapping_locs = [{id = 6 : i32, resource = "tile", time_step = 0 : i32, x = 2 : i32, y = 1 : i32}]} : () -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %7 = "neura.grant_always"() <{constant_value = true}> {mapping_locs = [{id = 4 : i32, resource = "tile", time_step = 0 : i32, x = 0 : i32, y = 1 : i32}]} : () -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %8 = "neura.data_mov"(%7) {mapping_locs = [{id = 10 : i32, resource = "link", time_step = 0 : i32}, {id = 14 : i32, resource = "link", time_step = 1 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %9 = "neura.data_mov"(%6) {mapping_locs = []} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %9 = "neura.data_mov"(%6) {mapping_locs = [{id = 24 : i32, resource = "register", time_step = 0 : i32}, {id = 24 : i32, resource = "register", time_step = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %10 = "neura.data_mov"(%3) {mapping_locs = [{id = 28 : i32, resource = "link", time_step = 0 : i32}, {id = 33 : i32, resource = "link", time_step = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %11 = "neura.data_mov"(%2) {mapping_locs = [{id = 36 : i32, resource = "link", time_step = 0 : i32}, {id = 21 : i32, resource = "link", time_step = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %nextindex, %valid = neura.loop_control(parent_valid = %8, start = %9, end = %10, step = %11) {iterationType = "increment", mapping_locs = [{id = 6 : i32, resource = "tile", time_step = 2 : i32, x = 2 : i32, y = 1 : i32}]} : !neura.data<i1, i1>, !neura.data<i64, i1>, !neura.data<i64, i1>, !neura.data<i64, i1> -> !neura.data<i64, i1>, !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %12 = "neura.data_mov"(%valid) {mapping_locs = []} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %12 = "neura.data_mov"(%valid) {mapping_locs = [{id = 25 : i32, resource = "register", time_step = 2 : i32}, {id = 25 : i32, resource = "register", time_step = 3 : i32}, {id = 25 : i32, resource = "register", time_step = 4 : i32}, {id = 25 : i32, resource = "register", time_step = 5 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %13 = "neura.not"(%12) {mapping_locs = [{id = 6 : i32, resource = "tile", time_step = 6 : i32, x = 2 : i32, y = 1 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %14 = "neura.data_mov"(%0) {mapping_locs = [{id = 36 : i32, resource = "link", time_step = 1 : i32}, {id = 28 : i32, resource = "register", time_step = 2 : i32}]} : (!neura.data<memref<?xi32>, i1>) -> !neura.data<memref<?xi32>, i1>
// FUSE-MAPPING-NEXT: %15 = "neura.data_mov"(%valid) {mapping_locs = [{id = 18 : i32, resource = "link", time_step = 2 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
Expand All @@ -212,7 +212,7 @@ module attributes {} {
// FUSE-MAPPING-NEXT: %24 = "neura.data_mov"(%valid) {mapping_locs = [{id = 19 : i32, resource = "link", time_step = 2 : i32}, {id = 6 : i32, resource = "link", time_step = 3 : i32}, {id = 9 : i32, resource = "link", time_step = 4 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %25 = neura.grant_predicate %23, %24 {mapping_locs = [{id = 7 : i32, resource = "tile", time_step = 5 : i32, x = 3 : i32, y = 1 : i32}]} : !neura.data<memref<?xi32>, i1>, !neura.data<i1, i1> -> !neura.data<memref<?xi32>, i1>
// FUSE-MAPPING-NEXT: %26 = "neura.data_mov"(%16) {mapping_locs = [{id = 21 : i32, resource = "link", time_step = 3 : i32}]} : (!neura.data<memref<?xi32>, i1>) -> !neura.data<memref<?xi32>, i1>
// FUSE-MAPPING-NEXT: %27 = "neura.data_mov"(%nextindex) {mapping_locs = []} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %27 = "neura.data_mov"(%nextindex) {mapping_locs = [{id = 24 : i32, resource = "register", time_step = 2 : i32}, {id = 24 : i32, resource = "register", time_step = 3 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %28 = neura.load_indexed %26[%27 : !neura.data<i64, i1>] !neura.data<memref<?xi32>, i1> {mapping_locs = [{id = 6 : i32, resource = "tile", time_step = 4 : i32, x = 2 : i32, y = 1 : i32}]} : !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %29 = "neura.data_mov"(%28) {mapping_locs = [{id = 17 : i32, resource = "link", time_step = 4 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %30 = "neura.data_mov"(%19) {mapping_locs = [{id = 4 : i32, resource = "link", time_step = 4 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
Expand All @@ -221,8 +221,8 @@ module attributes {} {
// FUSE-MAPPING-NEXT: %33 = "neura.data_mov"(%22) {mapping_locs = [{id = 31 : i32, resource = "link", time_step = 4 : i32}, {id = 36 : i32, resource = "register", time_step = 5 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %34 = "neura.add"(%32, %33) {mapping_locs = [{id = 9 : i32, resource = "tile", time_step = 6 : i32, x = 1 : i32, y = 2 : i32}]} : (!neura.data<i32, i1>, !neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %35 = "neura.data_mov"(%34) {mapping_locs = [{id = 28 : i32, resource = "link", time_step = 6 : i32}, {id = 33 : i32, resource = "link", time_step = 7 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %36 = "neura.data_mov"(%25) {mapping_locs = [{id = 21 : i32, resource = "link", time_step = 5 : i32}, {id = 24 : i32, resource = "register", time_step = 6 : i32}, {id = 24 : i32, resource = "register", time_step = 7 : i32}]} : (!neura.data<memref<?xi32>, i1>) -> !neura.data<memref<?xi32>, i1>
// FUSE-MAPPING-NEXT: %37 = "neura.data_mov"(%nextindex) {mapping_locs = []} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %36 = "neura.data_mov"(%25) {mapping_locs = [{id = 21 : i32, resource = "link", time_step = 5 : i32}, {id = 26 : i32, resource = "register", time_step = 6 : i32}, {id = 26 : i32, resource = "register", time_step = 7 : i32}]} : (!neura.data<memref<?xi32>, i1>) -> !neura.data<memref<?xi32>, i1>
// FUSE-MAPPING-NEXT: %37 = "neura.data_mov"(%nextindex) {mapping_locs = [{id = 27 : i32, resource = "register", time_step = 2 : i32}, {id = 27 : i32, resource = "register", time_step = 3 : i32}, {id = 27 : i32, resource = "register", time_step = 4 : i32}, {id = 27 : i32, resource = "register", time_step = 5 : i32}, {id = 27 : i32, resource = "register", time_step = 6 : i32}, {id = 27 : i32, resource = "register", time_step = 7 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: neura.store_indexed %35 to %36[%37 : !neura.data<i64, i1>] !neura.data<memref<?xi32>, i1> {mapping_locs = [{id = 6 : i32, resource = "tile", time_step = 8 : i32, x = 2 : i32, y = 1 : i32}]} : !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: "neura.return"() {mapping_locs = [{id = 0 : i32, resource = "tile", time_step = 6 : i32, x = 0 : i32, y = 0 : i32}]} : () -> ()
// FUSE-MAPPING-NEXT: }
4 changes: 2 additions & 2 deletions test/controflow_fuse/simple_loop_reduction/simple_loop_reduction.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,7 @@ module attributes {} {
// FUSE-MAPPING-NEXT: %7 = "neura.data_mov"(%1) {mapping_locs = [{id = 39 : i32, resource = "link", time_step = 0 : i32}, {id = 33 : i32, resource = "register", time_step = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %8 = "neura.data_mov"(%0) {mapping_locs = [{id = 12 : i32, resource = "link", time_step = 0 : i32}, {id = 34 : i32, resource = "register", time_step = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %nextindex, %valid = neura.loop_control(parent_valid = %5, start = %6, end = %7, step = %8) {iterationType = "increment", mapping_locs = [{id = 8 : i32, resource = "tile", time_step = 2 : i32, x = 0 : i32, y = 2 : i32}]} : !neura.data<i1, i1>, !neura.data<i64, i1>, !neura.data<i64, i1>, !neura.data<i64, i1> -> !neura.data<i64, i1>, !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %9 = "neura.data_mov"(%valid) {mapping_locs = []} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %9 = "neura.data_mov"(%valid) {mapping_locs = [{id = 32 : i32, resource = "register", time_step = 2 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %10 = "neura.not"(%9) {mapping_locs = [{id = 8 : i32, resource = "tile", time_step = 3 : i32, x = 0 : i32, y = 2 : i32}]} : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// FUSE-MAPPING-NEXT: %11 = neura.reserve : !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %12 = "neura.data_mov"(%2) {mapping_locs = [{id = 0 : i32, resource = "link", time_step = 1 : i32}, {id = 4 : i32, resource = "register", time_step = 2 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
Expand All @@ -185,7 +185,7 @@ module attributes {} {
// FUSE-MAPPING-NEXT: %19 = neura.grant_predicate %17, %18 {mapping_locs = [{id = 0 : i32, resource = "tile", time_step = 5 : i32, x = 0 : i32, y = 0 : i32}]} : !neura.data<i32, i1>, !neura.data<i1, i1> -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %20 = "neura.data_mov"(%nextindex) {mapping_locs = [{id = 25 : i32, resource = "link", time_step = 2 : i32}, {id = 16 : i32, resource = "register", time_step = 3 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i64, i1>
// FUSE-MAPPING-NEXT: %21 = "neura.cast"(%20) <{cast_type = "i64_to_i32"}> {mapping_locs = [{id = 4 : i32, resource = "tile", time_step = 4 : i32, x = 0 : i32, y = 1 : i32}]} : (!neura.data<i64, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %22 = "neura.data_mov"(%16) {mapping_locs = []} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %22 = "neura.data_mov"(%16) {mapping_locs = [{id = 20 : i32, resource = "register", time_step = 4 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %23 = "neura.data_mov"(%21) {mapping_locs = [{id = 10 : i32, resource = "link", time_step = 4 : i32}]} : (!neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: %24 = "neura.add"(%22, %23) {mapping_locs = [{id = 5 : i32, resource = "tile", time_step = 5 : i32, x = 1 : i32, y = 1 : i32}]} : (!neura.data<i32, i1>, !neura.data<i32, i1>) -> !neura.data<i32, i1>
// FUSE-MAPPING-NEXT: neura.ctrl_mov %24 -> %11 {mapping_locs = [{id = 15 : i32, resource = "link", time_step = 5 : i32}]} : !neura.data<i32, i1> !neura.data<i32, i1>
Expand Down
Loading