[Mlir-commits] [mlir] [mlir][linalg][nfc] Split hoisting tests into dedicated test functions (PR #145234)

[Han-Chung Wang]

================
@@ -1,76 +1,210 @@
 // RUN: mlir-opt  -transform-interpreter -canonicalize --split-input-file --allow-unregistered-dialect %s | FileCheck %s
 
-// CHECK-LABEL: func @hoist_vector_transfer_pairs(
-//  CHECK-SAME:   %[[MEMREF0:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[MEMREF1:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[MEMREF2:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[MEMREF3:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[MEMREF4:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[MEMREF5:[a-zA-Z0-9]*]]: memref<?x?xf32>,
-//  CHECK-SAME:   %[[VAL:[a-zA-Z0-9]*]]: index,
-//  CHECK-SAME:   %[[LB:[a-zA-Z0-9]*]]: index,
-//  CHECK-SAME:   %[[UB:[a-zA-Z0-9]*]]: index,
-//  CHECK-SAME:   %[[STEP:[a-zA-Z0-9]*]]: index,
-//  CHECK-SAME:   %[[CMP:[a-zA-Z0-9]*]]: i1
-func.func @hoist_vector_transfer_pairs(
-    %memref0: memref<?x?xf32>, %memref1: memref<?x?xf32>, %memref2: memref<?x?xf32>,
-    %memref3: memref<?x?xf32>, %memref4: memref<?x?xf32>, %memref5: memref<?x?xf32>,
-    %val: index, %lb : index, %ub : index, %step: index, %cmp: i1) {
+///----------------------------------------------------------------------------------------
+/// Tests for vector.transfer_read + vector.transfer_write pairs
+///
+/// * Nested in double loops
+//  * Indices depend on induction variables
+///----------------------------------------------------------------------------------------
+
+// CHECK-LABEL: func @mem_use_outside
+// CHECK-SAME:      %[[MEM:[a-zA-Z0-9]+]]: memref<?x?xf32>,
+// CHECK-SAME:      %[[LB:[a-zA-Z0-9]+]]: index,
+// CHECK-SAME:      %[[UB:[a-zA-Z0-9]+]]: index,
+// CHECK-SAME:      %[[STEP:[a-zA-Z0-9]+]]: index)
+func.func @mem_use_outside(%mem: memref<?x?xf32>, %lb : index, %ub : index, %step: index) {
+  %pad = arith.constant 0.0 : f32
+
+// CHECK:           %[[PAD:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK:           scf.for %[[I:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] {
+// CHECK:             %[[READ:.*]] = vector.transfer_read %[[MEM]][%[[I]], %[[I]]], %[[PAD]] : memref<?x?xf32>, vector<1xf32>
+// CHECK:             %[[SCF:.*]] = scf.for %[[J:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] iter_args(%[[VAL_5:.*]] = %[[READ]]) -> (vector<1xf32>) {
+// CHECK:               %[[USE:.*]] = "val_use"(%[[VAL_5]]) : (vector<1xf32>) -> vector<1xf32>
+// CHECK:               scf.yield %[[USE]] : vector<1xf32>
+// CHECK:             }
----------------
hanhanW wrote:

I think it is okay, I just wanna point out that it is not necessary, and it depends on if you want to have fewer checks. 

https://github.com/llvm/llvm-project/pull/145234