[Codegen] Implement getDistributedShape for NestedLayout (iree-org#16377

)
nithinsubbiah · Feb 12, 2024 · 9722440 · 9722440
1 parent 9aabcb3
commit 9722440
Show file tree

Hide file tree

Showing 2 changed files with 38 additions and 1 deletion.
diff --git a/compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_vector_distribution.mlir b/compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_vector_distribution.mlir
@@ -34,6 +34,37 @@ func.func @distribute_elementwise_i32(%a: vector<16x16xi32>, %b: vector<16x16xi3
   return %d : vector<16x16xi32>
 }
 
+#nested = #iree_vector_ext.nested_layout<
+  subgroups_per_workgroup = [2, 1, 1],
+  batches_per_subgroup    = [8, 2, 4],
+  outers_per_batch        = [1, 4, 4],
+  threads_per_outer       = [8, 2, 4],
+  elements_per_thread     = [1, 8, 2],
+
+  subgroup_order          = [0, 1, 2],
+  batch_order             = [0, 1, 2],
+  outer_order             = [0, 1, 2],
+  thread_order            = [0, 1, 2],
+  element_order           = [0, 2, 1]
+>
+
+// CHECK-LABEL: @distribute_elementwise_nested_layout_f16
+func.func @distribute_elementwise_nested_layout_f16(%a: vector<128x128x128xf16>, %b: vector<128x128x128xf16>) -> vector<128x128x128xf16> {
+  %c0 = arith.constant 0 : index
+  %cst_0 = arith.constant 0.0 : f16
+  // CHECK: %[[ROOT:.*]] = arith.constant dense<0.000000e+00> : vector<8x2x4x1x4x4x1x2x8xf16>
+  %root = arith.constant {"__vector_layout_test_anchor_result_0" = #nested} dense<0.0> : vector<128x128x128xf16>
+  // CHECK-DAG: %[[B:.*]] = iree_vector_ext.to_simt %{{.*}} : vector<128x128x128xf16> -> vector<8x2x4x1x4x4x1x2x8xf16>
+  // CHECK-DAG: %[[C:.*]] = arith.mulf %[[B]], %[[ROOT]] : vector<8x2x4x1x4x4x1x2x8xf16>
+  %c = arith.mulf %root, %b : vector<128x128x128xf16>
+  // CHECK-DAG: %[[A:.*]] = iree_vector_ext.to_simt %{{.*}} : vector<128x128x128xf16> -> vector<8x2x4x1x4x4x1x2x8xf16>
+  // CHECK-DAG: %[[D:.*]] = arith.addf %[[C]], %[[A]] fastmath<reassoc,nnan> : vector<8x2x4x1x4x4x1x2x8xf16>
+  %d = arith.addf %c, %a fastmath<reassoc,nnan> : vector<128x128x128xf16>
+  // CHECK: iree_vector_ext.to_simd %[[D]] : vector<8x2x4x1x4x4x1x2x8xf16> -> vector<128x128x128xf16>
+  return %d : vector<128x128x128xf16>
+}
+
+// CHECK-LABEL: @distribute_scf_for
 func.func @distribute_scf_for(%a: vector<16x16xi32>, %b: vector<16x16xi32>) -> vector<16x16xi32> {
   %c0 = arith.constant 0 : index
   %c1 = arith.constant 1 : index

diff --git a/llvm-external-projects/iree-dialects/lib/Dialect/VectorExt/IR/VectorExtAttrs.cpp b/llvm-external-projects/iree-dialects/lib/Dialect/VectorExt/IR/VectorExtAttrs.cpp
@@ -206,8 +206,14 @@ NestedLayoutAttr::permute(ArrayRef<int64_t> permutation) const {
   llvm_unreachable("Not yet implemented");
 }
 
+/// We distribute to:
+/// <BATCH x OUTER x ELEMENT>
 SmallVector<int64_t> NestedLayoutAttr::getDistributedShape() const {
-  llvm_unreachable("Not yet implemented");
+  SmallVector<int64_t> shape;
+  shape.append(applyPermutation(getBatchesPerSubgroup(), getBatchOrder()));
+  shape.append(applyPermutation(getOutersPerBatch(), getOuterOrder()));
+  shape.append(applyPermutation(getElementsPerThread(), getElementOrder()));
+  return shape;
 }
 
 bool NestedLayoutAttr::isValidLayout(ArrayRef<int64_t> shape) const {