Skip to content

Commit 7b1f4ef

Browse files
banach-spaceaadeshps-mcw
banach-space
authored and
aadeshps-mcw
committed
[mlir][Vector] Add support for scalable vectors to ScanToArithOps (llvm#123117)
Note, scalable reductions dims are left as a TODO.
1 parent fa78495 commit 7b1f4ef

File tree

2 files changed

+105
-3
lines changed

2 files changed

+105
-3
lines changed

‎mlir/lib/Dialect/Vector/Transforms/LowerVectorScan.cpp‎

Lines changed: 12 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -111,7 +111,7 @@ struct ScanToArithOps : public OpRewritePattern<vector::ScanOp> {
111111
if (!isValidKind(isInt, scanOp.getKind()))
112112
return failure();
113113

114-
VectorType resType = VectorType::get(destShape, elType);
114+
VectorType resType = destType;
115115
Value result = arith::ConstantOp::create(rewriter, loc, resType,
116116
rewriter.getZeroAttr(resType));
117117
int64_t reductionDim = scanOp.getReductionDim();
@@ -121,8 +121,18 @@ struct ScanToArithOps : public OpRewritePattern<vector::ScanOp> {
121121
int64_t initialValueRank = initialValueType.getRank();
122122

123123
SmallVector<int64_t> reductionShape(destShape);
124+
SmallVector<bool> reductionScalableDims(destType.getScalableDims());
125+
126+
if (reductionScalableDims[reductionDim])
127+
return rewriter.notifyMatchFailure(
128+
scanOp, "Trying to reduce scalable dimension - not yet supported!");
129+
130+
// The reduction dimension, after reducing, becomes 1. It's a fixed-width
131+
// dimension - no need to touch the scalability flag.
124132
reductionShape[reductionDim] = 1;
125-
VectorType reductionType = VectorType::get(reductionShape, elType);
133+
VectorType reductionType =
134+
VectorType::get(reductionShape, elType, reductionScalableDims);
135+
126136
SmallVector<int64_t> offsets(destRank, 0);
127137
SmallVector<int64_t> strides(destRank, 1);
128138
SmallVector<int64_t> sizes(destShape);

‎mlir/test/Dialect/Vector/vector-scan-transforms.mlir‎

Lines changed: 93 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
// RUN: mlir-opt %s --test-vector-scan-lowering | FileCheck %s
1+
// RUN: mlir-opt %s -split-input-file --test-vector-scan-lowering | FileCheck %s
22

33
// CHECK-LABEL: func @scan1d_inc
44
// CHECK-SAME: %[[ARG0:.*]]: vector<2xi32>,
@@ -18,6 +18,20 @@ func.func @scan1d_inc(%arg0 : vector<2xi32>, %arg1 : vector<i32>) -> (vector<2xi
1818
return %0#0, %0#1 : vector<2xi32>, vector<i32>
1919
}
2020

21+
// -----
22+
23+
// Reducing scalable dims is not yet supported!
24+
25+
// CHECK-LABEL: func @scan1d_inc_scalable
26+
// CHECK: vector.scan
27+
func.func @scan1d_inc_scalable(%arg0 : vector<[2]xi32>, %arg1 : vector<i32>) -> (vector<[2]xi32>, vector<i32>) {
28+
%0:2 = vector.scan <add>, %arg0, %arg1 {inclusive = true, reduction_dim = 0} :
29+
vector<[2]xi32>, vector<i32>
30+
return %0#0, %0#1 : vector<[2]xi32>, vector<i32>
31+
}
32+
33+
// -----
34+
2135
// CHECK-LABEL: func @scan1d_exc
2236
// CHECK-SAME: %[[ARG0:.*]]: vector<2xi32>,
2337
// CHECK-SAME: %[[ARG1:.*]]: vector<i32>
@@ -36,6 +50,20 @@ func.func @scan1d_exc(%arg0 : vector<2xi32>, %arg1 : vector<i32>) -> (vector<2xi
3650
return %0#0, %0#1 : vector<2xi32>, vector<i32>
3751
}
3852

53+
// -----
54+
55+
// Rducing scalable dims is not yet supported!
56+
57+
// CHECK-LABEL: func @scan1d_exc_scalable
58+
// CHECK: vector.scan
59+
func.func @scan1d_exc_scalable(%arg0 : vector<[2]xi32>, %arg1 : vector<i32>) -> (vector<[2]xi32>, vector<i32>) {
60+
%0:2 = vector.scan <add>, %arg0, %arg1 {inclusive = false, reduction_dim = 0} :
61+
vector<[2]xi32>, vector<i32>
62+
return %0#0, %0#1 : vector<[2]xi32>, vector<i32>
63+
}
64+
65+
// -----
66+
3967
// CHECK-LABEL: func @scan2d_mul_dim0
4068
// CHECK-SAME: %[[ARG0:.*]]: vector<2x3xi32>,
4169
// CHECK-SAME: %[[ARG1:.*]]: vector<3xi32>
@@ -53,6 +81,27 @@ func.func @scan2d_mul_dim0(%arg0 : vector<2x3xi32>, %arg1 : vector<3xi32>) -> (v
5381
return %0#0, %0#1 : vector<2x3xi32>, vector<3xi32>
5482
}
5583

84+
// -----
85+
86+
// CHECK-LABEL: func @scan2d_mul_dim0_scalable
87+
// CHECK-SAME: %[[ARG0:.*]]: vector<2x[3]xi32>,
88+
// CHECK-SAME: %[[ARG1:.*]]: vector<[3]xi32>
89+
// CHECK: %[[A:.*]] = arith.constant dense<0> : vector<2x[3]xi32>
90+
// CHECK: %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0], sizes = [1, 3], strides = [1, 1]} : vector<2x[3]xi32> to vector<1x[3]xi32>
91+
// CHECK: %[[C:.*]] = vector.insert_strided_slice %[[B]], %[[A]] {offsets = [0, 0], strides = [1, 1]} : vector<1x[3]xi32> into vector<2x[3]xi32>
92+
// CHECK: %[[D:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [1, 0], sizes = [1, 3], strides = [1, 1]} : vector<2x[3]xi32> to vector<1x[3]xi32>
93+
// CHECK: %[[E:.*]] = arith.muli %[[B]], %[[D]] : vector<1x[3]xi32>
94+
// CHECK: %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[C]] {offsets = [1, 0], strides = [1, 1]} : vector<1x[3]xi32> into vector<2x[3]xi32>
95+
// CHECK: %[[G:.*]] = vector.shape_cast %[[E]] : vector<1x[3]xi32> to vector<[3]xi32>
96+
// CHECK: return %[[F]], %[[G]] : vector<2x[3]xi32>, vector<[3]xi32>
97+
func.func @scan2d_mul_dim0_scalable(%arg0 : vector<2x[3]xi32>, %arg1 : vector<[3]xi32>) -> (vector<2x[3]xi32>, vector<[3]xi32>) {
98+
%0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = true, reduction_dim = 0} :
99+
vector<2x[3]xi32>, vector<[3]xi32>
100+
return %0#0, %0#1 : vector<2x[3]xi32>, vector<[3]xi32>
101+
}
102+
103+
// -----
104+
56105
// CHECK-LABEL: func @scan2d_mul_dim1
57106
// CHECK-SAME: %[[ARG0:.*]]: vector<2x3xi32>,
58107
// CHECK-SAME: %[[ARG1:.*]]: vector<2xi32>
@@ -73,6 +122,30 @@ func.func @scan2d_mul_dim1(%arg0 : vector<2x3xi32>, %arg1 : vector<2xi32>) -> (v
73122
return %0#0, %0#1 : vector<2x3xi32>, vector<2xi32>
74123
}
75124

125+
// -----
126+
127+
// CHECK-LABEL: func @scan2d_mul_dim1_scalable
128+
// CHECK-SAME: %[[ARG0:.*]]: vector<[2]x3xi32>,
129+
// CHECK-SAME: %[[ARG1:.*]]: vector<[2]xi32>
130+
// CHECK: %[[A:.*]] = arith.constant dense<0> : vector<[2]x3xi32>
131+
// CHECK: %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0], sizes = [2, 1], strides = [1, 1]} : vector<[2]x3xi32> to vector<[2]x1xi32>
132+
// CHECK: %[[C:.*]] = vector.insert_strided_slice %[[B]], %[[A]] {offsets = [0, 0], strides = [1, 1]} : vector<[2]x1xi32> into vector<[2]x3xi32>
133+
// CHECK: %[[D:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 1], sizes = [2, 1], strides = [1, 1]} : vector<[2]x3xi32> to vector<[2]x1xi32>
134+
// CHECK: %[[E:.*]] = arith.muli %[[B]], %[[D]] : vector<[2]x1xi32>
135+
// CHECK: %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[C]] {offsets = [0, 1], strides = [1, 1]} : vector<[2]x1xi32> into vector<[2]x3xi32>
136+
// CHECK: %[[G:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 2], sizes = [2, 1], strides = [1, 1]} : vector<[2]x3xi32> to vector<[2]x1xi32>
137+
// CHECK: %[[H:.*]] = arith.muli %[[E]], %[[G]] : vector<[2]x1xi32>
138+
// CHECK: %[[I:.*]] = vector.insert_strided_slice %[[H]], %[[F]] {offsets = [0, 2], strides = [1, 1]} : vector<[2]x1xi32> into vector<[2]x3xi32>
139+
// CHECK: %[[J:.*]] = vector.shape_cast %[[H]] : vector<[2]x1xi32> to vector<[2]xi32>
140+
// CHECK: return %[[I]], %[[J]] : vector<[2]x3xi32>, vector<[2]xi32>
141+
func.func @scan2d_mul_dim1_scalable(%arg0 : vector<[2]x3xi32>, %arg1 : vector<[2]xi32>) -> (vector<[2]x3xi32>, vector<[2]xi32>) {
142+
%0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = true, reduction_dim = 1} :
143+
vector<[2]x3xi32>, vector<[2]xi32>
144+
return %0#0, %0#1 : vector<[2]x3xi32>, vector<[2]xi32>
145+
}
146+
147+
// -----
148+
76149
// CHECK-LABEL: func @scan3d_mul_dim1
77150
// CHECK-SAME: %[[ARG0:.*]]: vector<4x2x3xf32>,
78151
// CHECK-SAME: %[[ARG1:.*]]: vector<4x3xf32>
@@ -89,3 +162,22 @@ func.func @scan3d_mul_dim1(%arg0 : vector<4x2x3xf32>, %arg1 : vector<4x3xf32>) -
89162
vector<4x2x3xf32>, vector<4x3xf32>
90163
return %0#0, %0#1 : vector<4x2x3xf32>, vector<4x3xf32>
91164
}
165+
166+
// -----
167+
168+
// CHECK-LABEL: func @scan3d_mul_dim1_scalable
169+
// CHECK-SAME: %[[ARG0:.*]]: vector<4x2x[3]xf32>,
170+
// CHECK-SAME: %[[ARG1:.*]]: vector<4x[3]xf32>
171+
// CHECK: %[[A:.*]] = arith.constant dense<0.000000e+00> : vector<4x2x[3]xf32>
172+
// CHECK: %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x[3]xf32> to vector<4x1x[3]xf32>
173+
// CHECK: %[[C:.*]] = vector.shape_cast %[[ARG1]] : vector<4x[3]xf32> to vector<4x1x[3]xf32>
174+
// CHECK: %[[D:.*]] = vector.insert_strided_slice %[[C]], %[[A]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x[3]xf32> into vector<4x2x[3]xf32>
175+
// CHECK: %[[E:.*]] = arith.mulf %[[C]], %[[B]] : vector<4x1x[3]xf32>
176+
// CHECK: %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[D]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x[3]xf32> into vector<4x2x[3]xf32>
177+
// CHECK: %[[G:.*]] = vector.shape_cast %[[E]] : vector<4x1x[3]xf32> to vector<4x[3]xf32>
178+
// CHECK: return %[[F]], %[[G]] : vector<4x2x[3]xf32>, vector<4x[3]xf32>
179+
func.func @scan3d_mul_dim1_scalable(%arg0 : vector<4x2x[3]xf32>, %arg1 : vector<4x[3]xf32>) -> (vector<4x2x[3]xf32>, vector<4x[3]xf32>) {
180+
%0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = false, reduction_dim = 1} :
181+
vector<4x2x[3]xf32>, vector<4x[3]xf32>
182+
return %0#0, %0#1 : vector<4x2x[3]xf32>, vector<4x[3]xf32>
183+
}

0 commit comments

Comments
 (0)