maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[mlir][sparse] add sparse kernels test to sparse compiler test suite 

This test makes sure kernels map to efficient sparse code, i.e. all
compressed for-loops, no co-iterating while loops.  In addition, this
revision removes the special constant folding inside the sparse
compiler in favor of Mahesh' new generic linalg folding. Thanks!

NOTE: relies on Mahesh fix, which needs to be rebased first

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D110001
This commit is contained in:
Aart Bik 2021-09-20 11:15:13 -07:00
parent 1552179ac0
commit a924fcc7c3
5 changed files with 160 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
mlir/include/mlir/Dialect/SparseTensor/Utils/Merger.h
View File

@ -230,7 +230,6 @@ public:
Value v1);
private:
bool isZero(unsigned e) const;
bool maybeZero(unsigned e) const;
bool isInvariant(unsigned e) const;
Type inferType(unsigned e, Value src);

16
mlir/lib/Dialect/SparseTensor/Utils/Merger.cpp
View File

@ -489,11 +489,6 @@ unsigned Merger::buildLattices(unsigned e, unsigned i) {
// ---+---+---+ ---+---+---+
// !x | 0 | y | !x | 0 |-y |
// x | x |x+y| x | x |x-y|
//
// TODO: remove this zero "folding" in favor of external pass into linalg
//
if (isZero(tensorExps[e].children.e1))
return buildLattices(tensorExps[e].children.e0, i);
return takeDisj(kind, // take binary disjunction
buildLattices(tensorExps[e].children.e0, i),
buildLattices(tensorExps[e].children.e1, i));
@ -516,17 +511,6 @@ Optional<unsigned> Merger::buildTensorExpFromLinalg(linalg::GenericOp op) {
return buildTensorExp(op, yield->getOperand(0));
}
/// Only returns true if we are certain this is a zero.
bool Merger::isZero(unsigned e) const {
if (tensorExps[e].kind == kInvariant) {
if (auto c = tensorExps[e].val.getDefiningOp<ConstantIntOp>())
return c.getValue() == 0;
if (auto c = tensorExps[e].val.getDefiningOp<ConstantFloatOp>())
return c.getValue().isZero();
}
return false;
}
/// Only returns false if we are certain this is a nonzero.
bool Merger::maybeZero(unsigned e) const {
if (tensorExps[e].kind == kInvariant) {

157
mlir/test/Dialect/SparseTensor/sparse_kernels.mlir Normal file
View File

@ -0,0 +1,157 @@
// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
// RUN: mlir-opt %s \
// RUN: --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
// RUN: --sparsification | FileCheck %s
#DCSR = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed" ] }>
// CHECK-LABEL: func @matmul(
// CHECK-SAME: %[[VAL_0:.*]]: tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>,
// CHECK-SAME: %[[VAL_1:.*]]: tensor<20x30xf32>,
// CHECK-SAME: %[[VAL_2:.*]]: tensor<10x30xf32>) -> tensor<10x30xf32> {
// CHECK-DAG: %[[VAL_3:.*]] = constant 0 : index
// CHECK-DAG: %[[VAL_4:.*]] = constant 1 : index
// CHECK-DAG: %[[VAL_5:.*]] = constant 30 : index
// CHECK: %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_7:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20xf32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_11:.*]] = memref.buffer_cast %[[VAL_1]] : memref<20x30xf32>
// CHECK: %[[VAL_12:.*]] = memref.buffer_cast %[[VAL_2]] : memref<10x30xf32>
// CHECK: %[[VAL_13:.*]] = memref.alloc() : memref<10x30xf32>
// CHECK: memref.copy %[[VAL_12]], %[[VAL_13]] : memref<10x30xf32> to memref<10x30xf32>
// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_3]]] : memref<?xindex>
// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_4]] {
// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xindex>
// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex>
// CHECK: %[[VAL_19:.*]] = addi %[[VAL_16]], %[[VAL_4]] : index
// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_4]] {
// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex>
// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xf32>
// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_24]]] : memref<10x30xf32>
// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_22]], %[[VAL_24]]] : memref<20x30xf32>
// CHECK: %[[VAL_27:.*]] = mulf %[[VAL_23]], %[[VAL_26]] : f32
// CHECK: %[[VAL_28:.*]] = addf %[[VAL_25]], %[[VAL_27]] : f32
// CHECK: memref.store %[[VAL_28]], %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_24]]] : memref<10x30xf32>
// CHECK: }
// CHECK: }
// CHECK: }
// CHECK: %[[VAL_29:.*]] = memref.tensor_load %[[VAL_13]] : memref<10x30xf32>
// CHECK: return %[[VAL_29]] : tensor<10x30xf32>
// CHECK: }
func @matmul(%a: tensor<10x20xf32, #DCSR>,
%b: tensor<20x30xf32>,
%c: tensor<10x30xf32>) -> tensor<10x30xf32> {
%0 = linalg.matmul
ins(%a, %b: tensor<10x20xf32, #DCSR>, tensor<20x30xf32>)
outs(%c: tensor<10x30xf32>) -> tensor<10x30xf32>
return %0 : tensor<10x30xf32>
}
// CHECK-LABEL: func @conv2d(
// CHECK-SAME: %[[VAL_0:.*]]: tensor<8x8xi32>,
// CHECK-SAME: %[[VAL_1:.*]]: tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>,
// CHECK-SAME: %[[VAL_2:.*]]: tensor<6x6xi32>) -> tensor<6x6xi32> {
// CHECK-DAG: %[[VAL_3:.*]] = constant 0 : index
// CHECK-DAG: %[[VAL_4:.*]] = constant 1 : index
// CHECK-DAG: %[[VAL_5:.*]] = constant 6 : index
// CHECK: %[[VAL_6:.*]] = memref.buffer_cast %[[VAL_0]] : memref<8x8xi32>
// CHECK: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_3]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_3]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<3x3xi32, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_12:.*]] = memref.buffer_cast %[[VAL_2]] : memref<6x6xi32>
// CHECK: %[[VAL_13:.*]] = memref.alloc() : memref<6x6xi32>
// CHECK: memref.copy %[[VAL_12]], %[[VAL_13]] : memref<6x6xi32> to memref<6x6xi32>
// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_3]]] : memref<?xindex>
// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_4]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_4]] {
// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex>
// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_16]]] : memref<?xindex>
// CHECK: %[[VAL_19:.*]] = addi %[[VAL_16]], %[[VAL_4]] : index
// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_19]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_4]] {
// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex>
// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_21]]] : memref<?xi32>
// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_3]] to %[[VAL_5]] step %[[VAL_4]] {
// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_25]], %[[VAL_24]]] : memref<6x6xi32>
// CHECK: %[[VAL_27:.*]] = addi %[[VAL_25]], %[[VAL_17]] : index
// CHECK: %[[VAL_28:.*]] = addi %[[VAL_24]], %[[VAL_22]] : index
// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_27]], %[[VAL_28]]] : memref<8x8xi32>
// CHECK: %[[VAL_30:.*]] = muli %[[VAL_29]], %[[VAL_23]] : i32
// CHECK: %[[VAL_31:.*]] = addi %[[VAL_26]], %[[VAL_30]] : i32
// CHECK: memref.store %[[VAL_31]], %[[VAL_13]]{{\[}}%[[VAL_25]], %[[VAL_24]]] : memref<6x6xi32>
// CHECK: }
// CHECK: }
// CHECK: }
// CHECK: }
// CHECK: %[[VAL_32:.*]] = memref.tensor_load %[[VAL_13]] : memref<6x6xi32>
// CHECK: return %[[VAL_32]] : tensor<6x6xi32>
// CHECK: }
func @conv2d(%input: tensor<8x8xi32>,
%filter: tensor<3x3xi32, #DCSR>,
%output: tensor<6x6xi32>) -> tensor<6x6xi32> {
%0 = linalg.conv_2d
ins (%input, %filter: tensor<8x8xi32>, tensor<3x3xi32, #DCSR>)
outs (%output: tensor<6x6xi32>) -> tensor<6x6xi32>
return %0 : tensor<6x6xi32>
}
// CHECK-LABEL: func @quantized_matmul(
// CHECK-SAME: %[[VAL_0:.*]]: tensor<5x3xi8>,
// CHECK-SAME: %[[VAL_1:.*]]: tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>,
// CHECK-SAME: %[[VAL_2:.*]]: tensor<5x6xi64>) -> tensor<5x6xi64> {
// CHECK-DAG: %[[VAL_3:.*]] = constant 2 : i64
// CHECK-DAG: %[[VAL_4:.*]] = constant 0 : index
// CHECK-DAG: %[[VAL_5:.*]] = constant 1 : index
// CHECK-DAG: %[[VAL_6:.*]] = constant 5 : index
// CHECK: %[[VAL_7:.*]] = memref.buffer_cast %[[VAL_0]] : memref<5x3xi8>
// CHECK: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_5]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_5]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<3x6xi8, #sparse_tensor.encoding<{{{.*}}}>>
// CHECK: %[[VAL_13:.*]] = memref.buffer_cast %[[VAL_2]] : memref<5x6xi64>
// CHECK: %[[VAL_14:.*]] = memref.alloc() : memref<5x6xi64>
// CHECK: memref.copy %[[VAL_13]], %[[VAL_14]] : memref<5x6xi64> to memref<5x6xi64>
// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_4]]] : memref<?xindex>
// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_5]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_15]] to %[[VAL_16]] step %[[VAL_5]] {
// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_17]]] : memref<?xindex>
// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_17]]] : memref<?xindex>
// CHECK: %[[VAL_20:.*]] = addi %[[VAL_17]], %[[VAL_5]] : index
// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_20]]] : memref<?xindex>
// CHECK: scf.for %[[VAL_22:.*]] = %[[VAL_19]] to %[[VAL_21]] step %[[VAL_5]] {
// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_22]]] : memref<?xindex>
// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]]] : memref<?xi8>
// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_4]] to %[[VAL_6]] step %[[VAL_5]] {
// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_25]], %[[VAL_23]]] : memref<5x6xi64>
// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_25]], %[[VAL_18]]] : memref<5x3xi8>
// CHECK: %[[VAL_28:.*]] = sexti %[[VAL_27]] : i8 to i64
// CHECK: %[[VAL_29:.*]] = subi %[[VAL_28]], %[[VAL_3]] : i64
// CHECK: %[[VAL_30:.*]] = sexti %[[VAL_24]] : i8 to i64
// CHECK: %[[VAL_31:.*]] = muli %[[VAL_29]], %[[VAL_30]] : i64
// CHECK: %[[VAL_32:.*]] = addi %[[VAL_26]], %[[VAL_31]] : i64
// CHECK: memref.store %[[VAL_32]], %[[VAL_14]]{{\[}}%[[VAL_25]], %[[VAL_23]]] : memref<5x6xi64>
// CHECK: }
// CHECK: }
// CHECK: }
// CHECK: %[[VAL_33:.*]] = memref.tensor_load %[[VAL_14]] : memref<5x6xi64>
// CHECK: return %[[VAL_33]] : tensor<5x6xi64>
// CHECK: }
func @quantized_matmul(%input1: tensor<5x3xi8>,
%input2: tensor<3x6xi8, #DCSR>,
%output: tensor<5x6xi64>) -> tensor<5x6xi64> {
%c0 = constant 0 : i32
%c2 = constant 2 : i32
%0 = linalg.quantized_matmul
ins(%input1, %input2, %c2, %c0 : tensor<5x3xi8>, tensor<3x6xi8, #DCSR>, i32, i32)
outs(%output : tensor<5x6xi64>) -> tensor<5x6xi64>
return %0: tensor<5x6xi64>
}

4
mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_filter_conv2d.mlir
View File

@ -1,5 +1,5 @@
// RUN: mlir-opt %s \
// RUN: --linalg-generalize-named-ops \
// RUN: --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
// RUN: --sparsification --sparse-tensor-conversion \
// RUN: --convert-vector-to-scf --convert-scf-to-std \
// RUN: --func-bufferize --tensor-constant-bufferize --tensor-bufferize \
@ -14,7 +14,7 @@
// Do the same run, but now with SIMDization as well. This should not change the outcome.
//
// RUN: mlir-opt %s \
// RUN: --linalg-generalize-named-ops \
// RUN: --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
// RUN: --sparsification="vectorization-strategy=2 vl=2" --sparse-tensor-conversion \
// RUN: --convert-vector-to-scf --convert-scf-to-std \
// RUN: --func-bufferize --tensor-constant-bufferize --tensor-bufferize \

2
mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_quantized_matmul.mlir
View File

@ -1,5 +1,5 @@
// RUN: mlir-opt %s \
// RUN: --linalg-generalize-named-ops \
// RUN: --linalg-generalize-named-ops --linalg-fuse-elementwise-ops \
// RUN: --sparsification --sparse-tensor-conversion \
// RUN: --convert-vector-to-scf --convert-scf-to-std \
// RUN: --func-bufferize --tensor-constant-bufferize --tensor-bufferize \