[mlir][sparse] Add arith-expand pass to the sparse-compiler pipeline.

[bixia]

Summary

Modify an existing test to test the situation.

[Peiming]

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir

58

Is the patch to address issues related to the use of this instruction?

[bixia]

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir

58

Yes.

[aartbik]

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir

46–47

Just curious, why did you not simply add another test for this but instead modified an existing one?
Most tests seems better in general ;-)

mlir/lib/Dialect/SparseTensor/Pipelines/SparseTensorPipelines.cpp

 //===- SparseTensorPipelines.cpp - Pipelines for sparse tensor code -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/SparseTensor/Pipelines/Passes.h"
 
 #include "mlir/Conversion/Passes.h"
+#include "mlir/Dialect/Arithmetic/Transforms/Passes.h"
 #include "mlir/Dialect/Bufferization/Transforms/Bufferize.h"
 #include "mlir/Dialect/Bufferization/Transforms/OneShotAnalysis.h"
 #include "mlir/Dialect/Bufferization/Transforms/Passes.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/Dialect/Linalg/Passes.h"
 #include "mlir/Dialect/SparseTensor/IR/SparseTensor.h"
 #include "mlir/Dialect/SparseTensor/Transforms/Passes.h"
 #include "mlir/Pass/PassManager.h"
void mlir::sparse_tensor::buildSparseCompiler(
   // it to this pipeline.
   pm.addNestedPass<func::FuncOp>(createConvertLinalgToLoopsPass());
   pm.addNestedPass<func::FuncOp>(createConvertVectorToSCFPass());
   pm.addNestedPass<func::FuncOp>(createConvertSCFToCFPass());
   pm.addPass(createLowerAffinePass());
   pm.addPass(createConvertVectorToLLVMPass(options.lowerVectorToLLVMOptions()));
   pm.addPass(createMemRefToLLVMPass());
   pm.addNestedPass<func::FuncOp>(createConvertComplexToStandardPass());
+  pm.addNestedPass<mlir::func::FuncOp>(
+      mlir::arith::createArithmeticExpandOpsPass());
   pm.addNestedPass<func::FuncOp>(createConvertMathToLLVMPass());
   pm.addPass(createConvertMathToLibmPass());
   pm.addPass(createConvertComplexToLibmPass());
   pm.addPass(createConvertComplexToLLVMPass());
   pm.addPass(createConvertFuncToLLVMPass());
   pm.addPass(createReconcileUnrealizedCastsPass());
 }
 

mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir

 //
 // Contains test cases for the sparse_tensor.binary operator (different cases when left/right/overlap
 // is empty/identity, etc).
 //
 
 module {
   // Creates a new sparse vector using the minimum values from two input sparse vectors.
   // When there is no overlap, include the present value in the output.
-  func.func @vector_min(%arga: tensor<?xf64, #SparseVector>,
-                        %argb: tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector> {
+  func.func @vector_min(%arga: tensor<?xi32, #SparseVector>,
+                        %argb: tensor<?xi32, #SparseVector>) -> tensor<?xi32, #SparseVector> {
     %c = arith.constant 0 : index
-    %d = tensor.dim %arga, %c : tensor<?xf64, #SparseVector>
-    %xv = bufferization.alloc_tensor(%d) : tensor<?xf64, #SparseVector>
+    %d = tensor.dim %arga, %c : tensor<?xi32, #SparseVector>
+    %xv = bufferization.alloc_tensor(%d) : tensor<?xi32, #SparseVector>
     %0 = linalg.generic #trait_vec_op
-       ins(%arga, %argb: tensor<?xf64, #SparseVector>, tensor<?xf64, #SparseVector>)
-        outs(%xv: tensor<?xf64, #SparseVector>) {
-        ^bb(%a: f64, %b: f64, %x: f64):
-          %1 = sparse_tensor.binary %a, %b : f64, f64 to f64
+       ins(%arga, %argb: tensor<?xi32, #SparseVector>, tensor<?xi32, #SparseVector>)
+        outs(%xv: tensor<?xi32, #SparseVector>) {
+        ^bb(%a: i32, %b: i32, %x: i32):
+          %1 = sparse_tensor.binary %a, %b : i32, i32 to i32
             overlap={
-              ^bb0(%a0: f64, %b0: f64):
-                %cmp = arith.cmpf "olt", %a0, %b0 : f64
-                %2 = arith.select %cmp, %a0, %b0: f64
-                sparse_tensor.yield %2 : f64
+              ^bb0(%a0: i32, %b0: i32):
+                %2 = arith.minsi %a0, %b0: i32
+                sparse_tensor.yield %2 : i32
             }
             left=identity
             right=identity
-          linalg.yield %1 : f64
-    } -> tensor<?xf64, #SparseVector>
-    return %0 : tensor<?xf64, #SparseVector>
+          linalg.yield %1 : i32
+    } -> tensor<?xi32, #SparseVector>
+    return %0 : tensor<?xi32, #SparseVector>
   }
 
   // Creates a new sparse vector by multiplying a sparse vector with a dense vector.
   // When there is no overlap, leave the result empty.
   func.func @vector_mul(%arga: tensor<?xf64, #SparseVector>,
                         %argb: tensor<?xf64>) -> tensor<?xf64, #SparseVector> {
     %c = arith.constant 0 : index
     %d = tensor.dim %arga, %c : tensor<?xf64, #SparseVector>
func.func @entry() {
     %v2 = arith.constant sparse<
        [ [1], [3], [4], [10], [16], [18], [21], [28], [29], [31] ],
          [11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0 ]
     > : tensor<32xf64>
     %v3 = arith.constant dense<
       [0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 0., 1., 2., 3., 4., 5., 6., 7., 8., 9.,
        0., 1., 2., 3., 4., 5., 6., 7., 8., 9., 0., 1.]
     > : tensor<32xf64>
+    %v1_si = arith.fptosi %v1 : tensor<32xf64> to tensor<32xi32>
+    %v2_si = arith.fptosi %v2 : tensor<32xf64> to tensor<32xi32>
+
     %sv1 = sparse_tensor.convert %v1 : tensor<32xf64> to tensor<?xf64, #SparseVector>
     %sv2 = sparse_tensor.convert %v2 : tensor<32xf64> to tensor<?xf64, #SparseVector>
+    %sv1_si = sparse_tensor.convert %v1_si : tensor<32xi32> to tensor<?xi32, #SparseVector>
+    %sv2_si = sparse_tensor.convert %v2_si : tensor<32xi32> to tensor<?xi32, #SparseVector>
     %dv3 = tensor.cast %v3 : tensor<32xf64> to tensor<?xf64>
 
     // Setup sparse matrices.
     %m1 = arith.constant sparse<
        [ [0,0], [0,1], [1,7], [2,2], [2,4], [2,7], [3,0], [3,2], [3,3] ],
          [ 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0 ]
     > : tensor<4x8xf64>
     %m2 = arith.constant sparse<
%m4 = arith.constant dense<
         [ 0.0, 0.5, 0.0, 0.0],
         [ 1.0, 5.0, 2.0, 0.0],
         [ 2.0, 0.0, 0.0, 0.0] ]> : tensor<4x4xf64>
 
     %sm3 = sparse_tensor.convert %m3 : tensor<4x4xf64> to tensor<4x4xf64, #DCSR>
     %sm4 = sparse_tensor.convert %m4 : tensor<4x4xf64> to tensor<4x4xf64, #DCSR>
 
     // Call sparse vector kernels.
-    %0 = call @vector_min(%sv1, %sv2)
-       : (tensor<?xf64, #SparseVector>,
-          tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector>
+    %0 = call @vector_min(%sv1_si, %sv2_si)
+       : (tensor<?xi32, #SparseVector>,
+          tensor<?xi32, #SparseVector>) -> tensor<?xi32, #SparseVector>
     %1 = call @vector_mul(%sv1, %dv3)
       : (tensor<?xf64, #SparseVector>,
          tensor<?xf64>) -> tensor<?xf64, #SparseVector>
     %2 = call @vector_setdiff(%sv1, %sv2)
        : (tensor<?xf64, #SparseVector>,
           tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector>
     %3 = call @vector_index(%sv1)
        : (tensor<?xf64, #SparseVector>) -> tensor<?xi32, #SparseVector>
func.func @entry() {
 
     //
     // Verify the results.
     //
     // CHECK:      ( 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 1, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 4, 0, 0, 5, 6, 0, 0, 0, 0, 0, 0, 7, 8, 0, 9 )
     // CHECK-NEXT: ( 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 0, 11, 0, 12, 13, 0, 0, 0, 0, 0, 14, 0, 0, 0, 0, 0, 15, 0, 16, 0, 0, 17, 0, 0, 0, 0, 0, 0, 18, 19, 0, 20 )
-    // CHECK-NEXT: ( 1, 11, 2, 13, 14, 3, 15, 4, 16, 5, 6, 7, 8, 9, -1, -1 )
+    // CHECK-NEXT: ( 1, 11, 2, 13, 14, 3, 15, 4, 16, 5, 6, 7, 8, 9, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 1, 11, 0, 2, 13, 0, 0, 0, 0, 0, 14, 3, 0, 0, 0, 0, 15, 4, 16, 0, 5, 6, 0, 0, 0, 0, 0, 0, 7, 8, 0, 9 )
     // CHECK-NEXT: ( 0, 6, 3, 28, 0, 6, 56, 72, 9, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 0, 0, 0, 6, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 28, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0, 56, 72, 0, 9 )
     // CHECK-NEXT: ( 1, 3, 4, 5, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 4, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
     // CHECK-NEXT: ( 0, 3, 11, 17, 20, 21, 28, 29, 31, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 11, 0, 0, 0, 0, 0, 17, 0, 0, 20, 21, 0, 0, 0, 0, 0, 0, 28, 29, 0, 31 )
     // CHECK-NEXT: ( ( 7, 0, 0, 0, 0, 0, 0, -5 ), ( -4, 0, 0, 0, 0, 0, -3, 0 ), ( 0, -2, 0, 0, 0, 0, 0, 7 ), ( 0, 0, 0, 0, 0, 0, 0, 0 ) )
     // CHECK-NEXT: ( ( 2, 0, 4, 1 ), ( 0, 2.5, 0, 0 ), ( 1, 5, 2, 4 ), ( 5, 4, 0, 0 ) )
     // CHECK-NEXT:   ( 2, 4, 1, 2.5, 1, 5, 2, 4, 5, 4, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( ( 2, 0, 4, 1 ), ( 0, 2.5, 0, 0 ), ( 1, 5, 2, 4 ), ( 5, 4, 0, 0 ) )
     // CHECK-NEXT:   ( 2, 4, 1, 2.5, 1, 5, 2, 4, 5, 4, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( ( 2, 0, 4, 1 ), ( 0, 2.5, 0, 0 ), ( -1, -5, 2, 4 ), ( 1, 4, 0, 0 ) )
     // CHECK-NEXT:   ( 2, 4, 1, 2.5, -1, -5, 2, 4, 1, 4, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( ( 0, 0, 1, -1 ), ( 0, 1, 0, 0 ), ( -1, -2, -2, 2 ), ( 1, 2, 0, 0 ) )
     // CHECK-NEXT:   ( 0, 1, -1, 1, -1, -2, -2, 2, 1, 2, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( ( 1, 0, 0, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ), ( 0, 0, 0, 0 ) )
     // CHECK-NEXT:   ( 1, 0, 0, 0, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 )
     // CHECK-NEXT: ( ( 0, 0, 0, -1 ), ( 0, 0, 0, 0 ), ( -1, -5, -2, 4 ), ( 0, 4, 0, 0 ) )
     // CHECK-NEXT:   ( -1, -1, -5, -2, 4, 4, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 )
     //
     call @dump_vec(%sv1) : (tensor<?xf64, #SparseVector>) -> ()
     call @dump_vec(%sv2) : (tensor<?xf64, #SparseVector>) -> ()
-    call @dump_vec(%0) : (tensor<?xf64, #SparseVector>) -> ()
+    call @dump_vec_i32(%0) : (tensor<?xi32, #SparseVector>) -> ()
     call @dump_vec(%1) : (tensor<?xf64, #SparseVector>) -> ()
     call @dump_vec(%2) : (tensor<?xf64, #SparseVector>) -> ()
     call @dump_vec_i32(%3) : (tensor<?xi32, #SparseVector>) -> ()
     call @dump_mat(%5) : (tensor<?x?xf64, #DCSR>) -> ()
     call @dump_mat_4x4(%6) : (tensor<4x4xf64, #DCSR>) -> ()
     call @dump_mat_4x4(%7) : (tensor<4x4xf64, #DCSR>) -> ()
     call @dump_mat_4x4(%8) : (tensor<4x4xf64, #DCSR>) -> ()
     call @dump_mat_4x4(%9) : (tensor<4x4xf64, #DCSR>) -> ()
     call @dump_mat_4x4_i8(%10) : (tensor<4x4xi8, #DCSR>) -> ()
     call @dump_mat_4x4(%11) : (tensor<4x4xf64, #DCSR>) -> ()
 
     // Release the resources.
     bufferization.dealloc_tensor %sv1 : tensor<?xf64, #SparseVector>
     bufferization.dealloc_tensor %sv2 : tensor<?xf64, #SparseVector>
     bufferization.dealloc_tensor %sm1 : tensor<?x?xf64, #DCSR>
     bufferization.dealloc_tensor %sm2 : tensor<?x?xf64, #DCSR>
     bufferization.dealloc_tensor %sm3 : tensor<4x4xf64, #DCSR>
     bufferization.dealloc_tensor %sm4 : tensor<4x4xf64, #DCSR>
-    bufferization.dealloc_tensor %0 : tensor<?xf64, #SparseVector>
+    bufferization.dealloc_tensor %0 : tensor<?xi32, #SparseVector>
     bufferization.dealloc_tensor %1 : tensor<?xf64, #SparseVector>
     bufferization.dealloc_tensor %2 : tensor<?xf64, #SparseVector>
     bufferization.dealloc_tensor %3 : tensor<?xi32, #SparseVector>
     bufferization.dealloc_tensor %5 : tensor<?x?xf64, #DCSR>
     bufferization.dealloc_tensor %6 : tensor<4x4xf64, #DCSR>
     bufferization.dealloc_tensor %7 : tensor<4x4xf64, #DCSR>
     bufferization.dealloc_tensor %8 : tensor<4x4xf64, #DCSR>
     bufferization.dealloc_tensor %9 : tensor<4x4xf64, #DCSR>
     bufferization.dealloc_tensor %10 : tensor<4x4xi8, #DCSR>
     bufferization.dealloc_tensor %11 : tensor<4x4xf64, #DCSR>
     return
   }
 }