Skip to content
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

Generalize before packing propagation #947

Closed
wants to merge 1 commit into from
Closed

Generalize before packing propagation #947

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 3 additions & 1 deletion lib/TPP/PassBundles/TppMapping.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,11 +67,13 @@ struct TppMapping : public tpp::impl::TppMappingBase<TppMapping>,
// Run only canonicalizer at this stage as full cleanup (mostly CSE) can
// mess up tensor producer-consumer chains used for analysis in the
// following passes.
// Generalize named ops to allow packing propagation.
// TODO: Remove the generalization when upstream propagation is improved.
pm.addNestedPass<func::FuncOp>(createLinalgGeneralizeNamedOpsPass());
pm.addPass(createPropagatePackUnPack());
pm.addPass(createConstantFoldPack());
pm.addPass(createSimplifyAndCanonicalizePack());

pm.addNestedPass<func::FuncOp>(createLinalgGeneralizeNamedOpsPass());
pm.addPass(createCleanup());
pm.addNestedPass<func::FuncOp>(
createLinalgConvertCompareSelectToMaximumfPass());
Expand Down
22 changes: 22 additions & 0 deletions test/Passes/tpp-mapping.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,3 +207,25 @@ func.func @tile_and_fuse(%arg0: tensor<64x64xf32>, %arg1: tensor<64x64xf32>,
// CHECK-SAME:{{.*}}outs(%{{.+}} : tensor<32x32xf32>)
// CHECK: linalg.generic{{.*}}outs(%{{.+}} : tensor<32x32xf32>)
// CHECK: arith.maximumf

// -----

func.func @tile_and_fuse_named(%arg0: tensor<64x64xf32>, %arg1: tensor<64x64xf32>,
%arg2: tensor<64x64xf32>, %arg3: tensor<64x64xf32>) -> tensor<64x64xf32> {
%e = tensor.empty() : tensor<64x64xf32>
%0 = linalg.matmul ins(%arg0, %arg1 : tensor<64x64xf32>, tensor<64x64xf32>)
outs(%arg2 : tensor<64x64xf32>) -> tensor<64x64xf32>
%1 = linalg.add ins(%0, %arg3 : tensor<64x64xf32>, tensor<64x64xf32>)
outs(%e : tensor<64x64xf32>) -> tensor<64x64xf32>
return %1 : tensor<64x64xf32>
}

// CHECK-LABEL: tile_and_fuse_named(
// CHECK-COUNT-3: tensor.pack
// Fused matmul and relu
// CHECK: scf.forall
// CHECK: linalg.batch_reduce_matmul{{.*}}ins(%{{.+}}, %{{.+}} : tensor<2x32x32xf32>, tensor<2x32x32xf32>)
// CHECK-SAME:{{.*}}outs(%{{.+}} : tensor<32x32xf32>)
// CHECK: linalg.generic{{.*}}outs(%{{.+}} : tensor<32x32xf32>)
// CHECK: arith.addf
// CHECK: tensor.unpack