Skip to content

[MLIR][Vector] Allow any shaped type to be distributed for vector.wa… #114215

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

Closed
wants to merge 2 commits into from
Closed

[MLIR][Vector] Allow any shaped type to be distributed for vector.wa… #114215

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
7d5b1b4
[MLIR][Vector] Allow any shaped typed to be distributed for vector.wa…
kurapov-peter Oct 17, 2024
834094f
Update warp_execute_on_lane_0 description
kurapov-peter Oct 30, 2024
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
3 changes: 2 additions & 1 deletion mlir/include/mlir/Dialect/Vector/IR/VectorOps.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3000,7 +3000,8 @@ def Vector_WarpExecuteOnLane0Op : Vector_Op<"warp_execute_on_lane_0",

Return values are distributed on all lanes using laneId as index. The
vector is distributed based on the shape ratio between the vector type of
the yield and the result type.
the yield and the result type. Any ShapedType return value is allowed to be
distributed to support other vector-like types (e.g., xegpu.tensor_desc).
If the shapes are the same this means the value is broadcasted to all lanes.
In the future the distribution can be made more explicit using affine_maps
and will support having multiple Ids.
Expand Down
12 changes: 6 additions & 6 deletions mlir/lib/Dialect/Vector/IR/VectorOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6558,14 +6558,14 @@ static LogicalResult verifyDistributedType(Type expanded, Type distributed,
// If the types matches there is no distribution.
if (expanded == distributed)
return success();
auto expandedVecType = llvm::dyn_cast<VectorType>(expanded);
auto distributedVecType = llvm::dyn_cast<VectorType>(distributed);
auto expandedVecType = llvm::dyn_cast<ShapedType>(expanded);
auto distributedVecType = llvm::dyn_cast<ShapedType>(distributed);
if (!expandedVecType || !distributedVecType)
return op->emitOpError("expected vector type for distributed operands.");
return op->emitOpError("expected shaped type for distributed operands.");
if (expandedVecType.getRank() != distributedVecType.getRank() ||
expandedVecType.getElementType() != distributedVecType.getElementType())
return op->emitOpError(
"expected distributed vectors to have same rank and element type.");
"expected distributed types to have same rank and element type.");

SmallVector<int64_t> scales(expandedVecType.getRank(), 1);
for (int64_t i = 0, e = expandedVecType.getRank(); i < e; i++) {
Expand All @@ -6575,8 +6575,8 @@ static LogicalResult verifyDistributedType(Type expanded, Type distributed,
continue;
if (eDim % dDim != 0)
return op->emitOpError()
<< "expected expanded vector dimension #" << i << " (" << eDim
<< ") to be a multipler of the distributed vector dimension ("
<< "expected expanded type dimension #" << i << " (" << eDim
<< ") to be a multipler of the distributed type dimension ("
<< dDim << ")";
scales[i] = eDim / dDim;
}
Expand Down
6 changes: 3 additions & 3 deletions mlir/test/Dialect/Vector/invalid.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1665,7 +1665,7 @@ func.func @warp_2_distributed_dims(%laneid: index) {
// -----

func.func @warp_2_distributed_dims(%laneid: index) {
// expected-error@+1 {{expected expanded vector dimension #1 (8) to be a multipler of the distributed vector dimension (3)}}
// expected-error@+1 {{expected expanded type dimension #1 (8) to be a multipler of the distributed type dimension (3)}}
%2 = vector.warp_execute_on_lane_0(%laneid)[32] -> (vector<1x3xi32>) {
%0 = arith.constant dense<2>: vector<4x8xi32>
vector.yield %0 : vector<4x8xi32>
Expand All @@ -1676,7 +1676,7 @@ func.func @warp_2_distributed_dims(%laneid: index) {
// -----

func.func @warp_mismatch_rank(%laneid: index) {
// expected-error@+1 {{'vector.warp_execute_on_lane_0' op expected distributed vectors to have same rank and element type.}}
// expected-error@+1 {{'vector.warp_execute_on_lane_0' op expected distributed types to have same rank and element type.}}
%2 = vector.warp_execute_on_lane_0(%laneid)[32] -> (vector<4x4xi32>) {
%0 = arith.constant dense<2>: vector<128xi32>
vector.yield %0 : vector<128xi32>
Expand All @@ -1687,7 +1687,7 @@ func.func @warp_mismatch_rank(%laneid: index) {
// -----

func.func @warp_mismatch_rank(%laneid: index) {
// expected-error@+1 {{'vector.warp_execute_on_lane_0' op expected vector type for distributed operands.}}
// expected-error@+1 {{'vector.warp_execute_on_lane_0' op expected shaped type for distributed operands.}}
%2 = vector.warp_execute_on_lane_0(%laneid)[32] -> (i32) {
%0 = arith.constant dense<2>: vector<128xi32>
vector.yield %0 : vector<128xi32>
Expand Down
Loading