Skip to content

[mlir][sparse] introduce sparse_tensor.extract_value operation. #101219

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.

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
merged 2 commits into from
Jul 30, 2024

Conversation

PeimingLiu
Copy link
Member

No description provided.

@llvmbot
Copy link
Member

llvmbot commented Jul 30, 2024

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-sparse

Author: Peiming Liu (PeimingLiu)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/101219.diff

4 Files Affected:

  • (modified) mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td (+25)
  • (modified) mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp (+13)
  • (modified) mlir/test/Dialect/SparseTensor/invalid.mlir (+36)
  • (modified) mlir/test/Dialect/SparseTensor/roundtrip.mlir (+21)
diff --git a/mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td b/mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td
index f31df080d7811..14fb471939f11 100644
--- a/mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td
+++ b/mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td
@@ -1531,6 +1531,31 @@ def ExtractIterSpaceOp : SparseTensor_Op<"extract_iteration_space",
   let hasVerifier = 1;
 }
 
+def ExtractValOp : SparseTensor_Op<"extract_value", [
+    Pure,
+    TypesMatchWith<"result type matches element type of tensor",
+                   "tensor", "result",
+                   "::llvm::cast<TensorType>($_self).getElementType()">]> {
+  let summary = "Extracts a value from a sparse tensor.";
+  let description = [{
+      The `sparse_tensor.extract_value` operation extract the value
+      pointed by a sparse iterator from a sparse tensor.
+
+      Example:
+
+      ```mlir
+      %val = sparse_tensor.extract_value %sp at %it
+           : tensor<?x?xf32, #CSR>, !sparse_tensor.iterator<#CSR, lvl = 1>
+      ```
+  }];
+
+  let arguments = (ins AnySparseTensor:$tensor, AnySparseIterator:$iterator);
+  let results = (outs AnyType:$result);
+
+  let assemblyFormat = "$tensor `at` $iterator attr-dict `:` type($tensor)`,` qualified(type($iterator))";
+  let hasVerifier = 1;
+}
+
 def IterateOp : SparseTensor_Op<"iterate",
     [RecursiveMemoryEffects, RecursivelySpeculatable,
      DeclareOpInterfaceMethods<LoopLikeOpInterface,
diff --git a/mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp b/mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp
index 616e91ae04055..0a276d87f3bca 100644
--- a/mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp
+++ b/mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp
@@ -2267,6 +2267,19 @@ LogicalResult ExtractIterSpaceOp::verify() {
   return success();
 }
 
+LogicalResult ExtractValOp::verify() {
+  auto stt = getSparseTensorType(getTensor());
+  auto itTp = getIterator().getType();
+
+  if (stt.getEncoding() != itTp.getEncoding())
+    return emitOpError("mismatch in tensor encoding and iterator encoding.");
+
+  if (stt.getLvlRank() != itTp.getHiLvl())
+    return emitOpError("must use last-level iterator to extract values. ");
+
+  return success();
+}
+
 struct RemoveUnusedLvlCrds : public OpRewritePattern<IterateOp> {
   using OpRewritePattern::OpRewritePattern;
 
diff --git a/mlir/test/Dialect/SparseTensor/invalid.mlir b/mlir/test/Dialect/SparseTensor/invalid.mlir
index eb0dc01be25b9..61cc9be88685c 100644
--- a/mlir/test/Dialect/SparseTensor/invalid.mlir
+++ b/mlir/test/Dialect/SparseTensor/invalid.mlir
@@ -1099,6 +1099,42 @@ func.func @sparse_extract_iter_space(%sp : tensor<4x8xf32, #COO>, %it1 : !sparse
   return
 }
 
+// -----
+
+#COO = #sparse_tensor.encoding<{
+  map = (i, j) -> (
+    i : compressed(nonunique),
+    j : singleton(soa)
+  )
+}>
+
+#CSR = #sparse_tensor.encoding<{
+  map = (i, j) -> (
+    i : dense,
+    j : compressed
+  )
+}>
+
+func.func @sparse_extract_value(%sp : tensor<4x8xf32, #COO>, %it1 : !sparse_tensor.iterator<#CSR, lvls = 1>) -> f32 {
+  // expected-error@+1 {{'sparse_tensor.extract_value' op mismatch in tensor encoding and iterator encoding.}}
+  %f = sparse_tensor.extract_value %sp at %it1 : tensor<4x8xf32, #COO>, !sparse_tensor.iterator<#CSR, lvls = 1>
+  return %f : f32
+}
+
+// -----
+
+#COO = #sparse_tensor.encoding<{
+  map = (i, j) -> (
+    i : compressed(nonunique),
+    j : singleton(soa)
+  )
+}>
+
+func.func @sparse_extract_value(%sp : tensor<4x8xf32, #COO>, %it1 : !sparse_tensor.iterator<#COO, lvls = 0>) -> f32 {
+  // expected-error@+1 {{'sparse_tensor.extract_value' op must use last-level iterator to extract values.}}
+  %f = sparse_tensor.extract_value %sp at %it1 : tensor<4x8xf32, #COO>, !sparse_tensor.iterator<#COO, lvls = 0>
+  return %f : f32
+}
 
 // -----
 
diff --git a/mlir/test/Dialect/SparseTensor/roundtrip.mlir b/mlir/test/Dialect/SparseTensor/roundtrip.mlir
index bce0b41a99828..055709ee69eb7 100644
--- a/mlir/test/Dialect/SparseTensor/roundtrip.mlir
+++ b/mlir/test/Dialect/SparseTensor/roundtrip.mlir
@@ -739,6 +739,27 @@ func.func @sparse_has_runtime() -> i1 {
   return %has_runtime : i1
 }
 
+// -----
+
+#COO = #sparse_tensor.encoding<{
+  map = (i, j) -> (
+    i : compressed(nonunique),
+    j : singleton(soa)
+  )
+}>
+
+// CHECK-LABEL:   func.func @sparse_extract_value(
+// CHECK-SAME:      %[[VAL_0:.*]]: tensor<4x8xf32, #sparse>,
+// CHECK-SAME:      %[[VAL_1:.*]]: !sparse_tensor.iterator<#sparse, lvls = 1>) -> f32 {
+// CHECK:           %[[VAL_2:.*]] = sparse_tensor.extract_value %[[VAL_0]] at %[[VAL_1]] : tensor<4x8xf32, #sparse>, !sparse_tensor.iterator<#sparse, lvls = 1>
+// CHECK:           return %[[VAL_2]] : f32
+// CHECK:         }
+func.func @sparse_extract_value(%sp : tensor<4x8xf32, #COO>, %it1 : !sparse_tensor.iterator<#COO, lvls = 1>) -> f32 {
+  %f = sparse_tensor.extract_value %sp at %it1 : tensor<4x8xf32, #COO>, !sparse_tensor.iterator<#COO, lvls = 1>
+  return %f : f32
+}
+
+
 // -----
 
 #COO = #sparse_tensor.encoding<{

@PeimingLiu PeimingLiu merged commit 12189f8 into llvm:main Jul 30, 2024
4 of 6 checks passed
@PeimingLiu PeimingLiu deleted the extract-val-intro branch July 30, 2024 23:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
mlir:sparse Sparse compiler in MLIR mlir
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants