[mlir][sve][nfc] Merge the integration tests for linalg.matmul #74059
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Instead of duplicating the TD logic to tile and vectorize `linalg.matmul` in multiple test files, this patch uses `transform.foreach` to apply the same sequence to multiple functions within the same test file (e.g. `matmul_f32` and `matmul_mixed_ty` as defined in the original files).
|
@llvm/pr-subscribers-mlir-linalg @llvm/pr-subscribers-mlir Author: Andrzej Warzyński (banach-space) ChangesInstead of duplicating the TD logic to tile and vectorize Full diff: https://github.com/llvm/llvm-project/pull/74059.diff 2 Files Affected:
diff --git a/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul.mlir b/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul.mlir
index d771d32d548bbe2..17393412badf359 100644
--- a/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul.mlir
+++ b/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul.mlir
@@ -8,7 +8,10 @@
// RUN: %{compile}
-// RUN: %{run} | FileCheck %s
+// RUN: %{run} | FileCheck %s --check-prefix=F32
+
+// REDEFINE: %{entry_point} = matmul_mixed_ty
+// RUN: %{run} | FileCheck %s --check-prefix=MIXED
func.func @matmul_f32() {
// Matrix dimensions
@@ -32,37 +35,75 @@ func.func @matmul_f32() {
%C_out = linalg.matmul ins(%A, %B: tensor<?x?xf32>, tensor<?x?xf32>) outs(%C_in: tensor<?x?xf32>) -> tensor<?x?xf32>
// Print and verify the output
- // CHECK-LABEL: SVE: START OF TEST OUTPUT
+ // F32-LABEL: SVE: START OF TEST OUTPUT
vector.print str "SVE: START OF TEST OUTPUT"
- // CHECK-NEXT: Unranked Memref {{.*}} rank = 2 offset = 0 sizes = [5, 15] strides = [15, 1] data =
- // CHECK-COUNT-5: [29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788]
+ // F32-NEXT: Unranked Memref {{.*}} rank = 2 offset = 0 sizes = [5, 15] strides = [15, 1] data =
+ // F32-COUNT-5: [29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788, 29.5788]
%xf = tensor.cast %C_out : tensor<?x?xf32> to tensor<*xf32>
call @printMemrefF32(%xf) : (tensor<*xf32>) -> ()
- // CHECK-NEXT: SVE: END OF TEST OUTPUT
+ // F32-NEXT: SVE: END OF TEST OUTPUT
+ vector.print str "SVE: END OF TEST OUTPUT"
+
+ return
+}
+
+func.func @matmul_mixed_ty() {
+ // Matrix dimensions
+ %K = arith.constant 3 : index
+ %M = arith.constant 5 : index
+ %N = arith.constant 15 : index
+ %c0_i8 = arith.constant 0 : i8
+ %c0_i32 = arith.constant 0 : i32
+
+ // Allocate the matrices
+ %A_alloc = bufferization.alloc_tensor(%M, %K) : tensor<?x?xi8>
+ %B_alloc = bufferization.alloc_tensor(%K, %N) : tensor<?x?xi8>
+ %C_alloc = bufferization.alloc_tensor(%M, %N) : tensor<?x?xi32>
+
+ // Initialise the matrices
+ %pi = arith.constant 123 : i8
+ %A = linalg.fill ins(%pi : i8) outs(%A_alloc : tensor<?x?xi8>) -> tensor<?x?xi8>
+ %B = linalg.fill ins(%pi : i8) outs(%B_alloc : tensor<?x?xi8>) -> tensor<?x?xi8>
+ %C_in = linalg.fill ins(%c0_i32 : i32) outs(%C_alloc : tensor<?x?xi32>) -> tensor<?x?xi32>
+
+ // Matmul
+ %C_out = linalg.matmul ins(%A, %B: tensor<?x?xi8>, tensor<?x?xi8>) outs(%C_in: tensor<?x?xi32>) -> tensor<?x?xi32>
+
+ // Print and verify the output
+ // MIXED-LABEL: SVE: START OF TEST OUTPUT
+ vector.print str "SVE: START OF TEST OUTPUT"
+
+ // MIXED-NEXT: Unranked Memref {{.*}} rank = 2 offset = 0 sizes = [5, 15] strides = [15, 1] data =
+ // MIXED-COUNT-5: [45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387]
+ %xf = tensor.cast %C_out : tensor<?x?xi32> to tensor<*xi32>
+ call @printMemrefI32(%xf) : (tensor<*xi32>) -> ()
+
+ // MIXED-NEXT: SVE: END OF TEST OUTPUT
vector.print str "SVE: END OF TEST OUTPUT"
return
}
module attributes {transform.with_named_sequence} {
-transform.named_sequence @__transform_main(%module: !transform.any_op {transform.readonly}) {
- %matmul = transform.structured.match ops{["linalg.matmul"]} in %module
- : (!transform.any_op) -> !transform.any_op
+ // A sequence that will tile and vectorise a Matmul Op
+ transform.named_sequence @tile_and_vectorize_matmul(%func
+ : !transform.op<"func.func"> {transform.readonly}) {
+
+ // Step 0: Get a handle to the matmul Op
+ %matmul = transform.structured.match ops{["linalg.matmul"]} in %func
+ : (!transform.op<"func.func">) -> !transform.any_op
// Step 1: Tile
- %module_with_tiled_loops, %loops:3 = transform.structured.tile_using_for %matmul [2, [4], 1]
+ %tiled_matmul, %loops:3 = transform.structured.tile_using_for %matmul [2, [4], 1]
: (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)
+ transform.print %tiled_matmul {name = "matmul lal"}: !transform.any_op
// Step 2: Vectorize
- %tiled_matmul = transform.structured.match ops{["linalg.matmul"]} in %module_with_tiled_loops
- : (!transform.any_op) -> !transform.any_op
transform.structured.vectorize %tiled_matmul vector_sizes [2, [4], 1] : !transform.any_op
// Step 3: Lower vector.multi_reduction to vector.contract (+ some helpful patterns)
- %func = transform.structured.match ops{["func.func"]} in %module
- : (!transform.any_op) -> !transform.op<"func.func">
transform.apply_patterns to %func {
transform.apply_patterns.vector.reduction_to_contract
transform.apply_patterns.vector.transfer_permutation_patterns
@@ -77,6 +118,21 @@ transform.named_sequence @__transform_main(%module: !transform.any_op {transform
transform.yield
}
+
+ // A sequence that goes over all functions in tis module and applies
+ // "tile_and_vectorize_matmul"
+ transform.named_sequence @__transform_main(%module: !transform.any_op {transform.readonly}) {
+ %funcs = transform.structured.match ops{["func.func"]} in %module
+ : (!transform.any_op) -> !transform.op<"func.func">
+
+ transform.foreach %funcs : !transform.op<"func.func"> {
+ ^bb2(%func : !transform.op<"func.func">):
+ transform.include @tile_and_vectorize_matmul failures(propagate)
+ (%func) : (!transform.op<"func.func">) -> ()
+ }
+ transform.yield
+ }
}
func.func private @printMemrefF32(%ptr : tensor<*xf32>)
+func.func private @printMemrefI32(%ptr : tensor<*xi32>)
diff --git a/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul_mixed_ty.mlir b/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul_mixed_ty.mlir
deleted file mode 100644
index f4f2d87b4d0b42c..000000000000000
--- a/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/matmul_mixed_ty.mlir
+++ /dev/null
@@ -1,83 +0,0 @@
-// DEFINE: %{compile} = mlir-opt %s \
-// DEFINE: -transform-interpreter -test-transform-dialect-erase-schedule \
-// DEFINE: -one-shot-bufferize -func-bufferize -cse -canonicalize -convert-vector-to-scf -arm-sve-legalize-vector-storage \
-// DEFINE: -convert-vector-to-llvm="enable-arm-sve" -test-lower-to-llvm -o %t
-// DEFINE: %{entry_point} = matmul_mixed_ty
-// DEFINE: %{run} = %mcr_aarch64_cmd %t -e %{entry_point} -entry-point-result=void --march=aarch64 --mattr="+sve"\
-// DEFINE: -shared-libs=%mlir_runner_utils,%mlir_c_runner_utils
-
-// RUN: %{compile}
-
-// RUN: %{run} | FileCheck %s
-
-func.func @matmul_mixed_ty() {
- // Matrix dimensions
- %K = arith.constant 3 : index
- %M = arith.constant 5 : index
- %N = arith.constant 15 : index
- %c0_i8 = arith.constant 0 : i8
- %c0_i32 = arith.constant 0 : i32
-
- // Allocate the matrices
- %A_alloc = bufferization.alloc_tensor(%M, %K) : tensor<?x?xi8>
- %B_alloc = bufferization.alloc_tensor(%K, %N) : tensor<?x?xi8>
- %C_alloc = bufferization.alloc_tensor(%M, %N) : tensor<?x?xi32>
-
- // Initialise the matrices
- %pi = arith.constant 123 : i8
- %A = linalg.fill ins(%pi : i8) outs(%A_alloc : tensor<?x?xi8>) -> tensor<?x?xi8>
- %B = linalg.fill ins(%pi : i8) outs(%B_alloc : tensor<?x?xi8>) -> tensor<?x?xi8>
- %C_in = linalg.fill ins(%c0_i32 : i32) outs(%C_alloc : tensor<?x?xi32>) -> tensor<?x?xi32>
-
- // Matmul
- %C_out = linalg.matmul ins(%A, %B: tensor<?x?xi8>, tensor<?x?xi8>) outs(%C_in: tensor<?x?xi32>) -> tensor<?x?xi32>
-
- // Print and verify the output
- // CHECK-LABEL: SVE: START OF TEST OUTPUT
- vector.print str "SVE: START OF TEST OUTPUT"
-
- // CHECK-NEXT: Unranked Memref {{.*}} rank = 2 offset = 0 sizes = [5, 15] strides = [15, 1] data =
- // CHECK-COUNT-5: [45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387, 45387]
- %xf = tensor.cast %C_out : tensor<?x?xi32> to tensor<*xi32>
- call @printMemrefI32(%xf) : (tensor<*xi32>) -> ()
-
- // CHECK-NEXT: SVE: END OF TEST OUTPUT
- vector.print str "SVE: END OF TEST OUTPUT"
-
- return
-}
-
-module attributes {transform.with_named_sequence} {
-transform.named_sequence @__transform_main(%module: !transform.any_op {transform.readonly}) {
- %matmul = transform.structured.match ops{["linalg.matmul"]} in %module
- : (!transform.any_op) -> !transform.any_op
-
- // Step 1: Tile
- %module_with_tiled_loops, %loops:3 = transform.structured.tile_using_for %matmul [2, [4], 1]
- : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)
-
- // Step 2: Vectorize
- %tiled_matmul = transform.structured.match ops{["linalg.matmul"]} in %module_with_tiled_loops
- : (!transform.any_op) -> !transform.any_op
- transform.structured.vectorize %tiled_matmul vector_sizes [2, [4], 1] : !transform.any_op
-
- // Step 3: Lower vector.multi_reduction to vector.contract (+ some helpful patterns)
- %func = transform.structured.match ops{["func.func"]} in %module
- : (!transform.any_op) -> !transform.op<"func.func">
- transform.apply_patterns to %func {
- transform.apply_patterns.vector.reduction_to_contract
- transform.apply_patterns.vector.transfer_permutation_patterns
- transform.apply_patterns.vector.lower_masked_transfers
- } : !transform.op<"func.func">
-
- // Step 4: Lower vector.contract to vector.fma
- transform.apply_patterns to %func {
- transform.apply_patterns.vector.lower_contraction lowering_strategy = "outerproduct"
- transform.apply_patterns.vector.lower_outerproduct
- } : !transform.op<"func.func">
-
- transform.yield
- }
-}
-
-func.func private @printMemrefI32(%ptr : tensor<*xi32>)
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
At the moment the logic to tile and vectorize
linalg.matmulis duplicated in multiple test files:Instead, this patch uses
transform.foreachto apply the same sequenceto multiple functions within the same test file (e.g.
matmul_f32andmatmul_mixed_tyasdefined in the original files).