1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "dense_xw_product_function.h"
#include <vespa/vespalib/objects/objectvisitor.h>
#include <vespa/eval/eval/value.h>
#include <vespa/eval/eval/operation.h>
#include <cassert>
#include <cblas.h>
namespace vespalib::eval {
using namespace tensor_function;
using namespace operation;
namespace {
template <typename LCT, typename RCT, typename OCT, bool common_inner>
OCT my_dot_product(const LCT *lhs, const RCT *rhs, size_t vector_size, size_t result_size) {
OCT result = 0.0;
for (size_t i = 0; i < vector_size; ++i) {
result += ((*lhs) * (*rhs));
++lhs;
rhs += (common_inner ? 1 : result_size);
}
return result;
}
template <typename LCT, typename RCT, typename OCT, bool common_inner>
void my_xw_product_op(InterpretedFunction::State &state, uint64_t param) {
const DenseXWProductFunction::Self &self = unwrap_param<DenseXWProductFunction::Self>(param);
auto vector_cells = state.peek(1).cells().typify<LCT>();
auto matrix_cells = state.peek(0).cells().typify<RCT>();
auto dst_cells = state.stash.create_uninitialized_array<OCT>(self.result_size);
OCT *dst = dst_cells.begin();
const RCT *matrix = matrix_cells.cbegin();
for (size_t i = 0; i < self.result_size; ++i) {
*dst++ = my_dot_product<LCT,RCT,OCT,common_inner>(vector_cells.cbegin(), matrix, self.vector_size, self.result_size);
matrix += (common_inner ? self.vector_size : 1);
}
state.pop_pop_push(state.stash.create<DenseValueView>(self.result_type, TypedCells(dst_cells)));
}
template <bool common_inner>
void my_cblas_double_xw_product_op(InterpretedFunction::State &state, uint64_t param) {
const DenseXWProductFunction::Self &self = unwrap_param<DenseXWProductFunction::Self>(param);
auto vector_cells = state.peek(1).cells().typify<double>();
auto matrix_cells = state.peek(0).cells().typify<double>();
auto dst_cells = state.stash.create_array<double>(self.result_size);
cblas_dgemv(CblasRowMajor, common_inner ? CblasNoTrans : CblasTrans,
common_inner ? self.result_size : self.vector_size,
common_inner ? self.vector_size : self.result_size,
1.0, matrix_cells.cbegin(), common_inner ? self.vector_size : self.result_size, vector_cells.cbegin(), 1,
0.0, dst_cells.begin(), 1);
state.pop_pop_push(state.stash.create<DenseValueView>(self.result_type, TypedCells(dst_cells)));
}
template <bool common_inner>
void my_cblas_float_xw_product_op(InterpretedFunction::State &state, uint64_t param) {
const DenseXWProductFunction::Self &self = unwrap_param<DenseXWProductFunction::Self>(param);
auto vector_cells = state.peek(1).cells().typify<float>();
auto matrix_cells = state.peek(0).cells().typify<float>();
auto dst_cells = state.stash.create_array<float>(self.result_size);
cblas_sgemv(CblasRowMajor, common_inner ? CblasNoTrans : CblasTrans,
common_inner ? self.result_size : self.vector_size,
common_inner ? self.vector_size : self.result_size,
1.0, matrix_cells.cbegin(), common_inner ? self.vector_size : self.result_size, vector_cells.cbegin(), 1,
0.0, dst_cells.begin(), 1);
state.pop_pop_push(state.stash.create<DenseValueView>(self.result_type, TypedCells(dst_cells)));
}
bool isDenseTensor(const ValueType &type, size_t d) {
return (type.is_dense() && (type.dimensions().size() == d));
}
bool isDenseXWProduct(const ValueType &res, const ValueType &vec, const ValueType &mat) {
if (isDenseTensor(res, 1) &&
isDenseTensor(vec, 1) &&
isDenseTensor(mat, 2))
{
size_t res_idx = mat.dimension_index(res.dimensions()[0].name);
size_t vec_idx = mat.dimension_index(vec.dimensions()[0].name);
size_t npos = ValueType::Dimension::npos;
if ((res_idx != npos) && (vec_idx != npos) && (res_idx != vec_idx)) {
assert(mat.dimensions()[res_idx].size == res.dimensions()[0].size);
assert(mat.dimensions()[vec_idx].size == vec.dimensions()[0].size);
return true;
}
}
return false;
}
const TensorFunction &createDenseXWProduct(const ValueType &res, const TensorFunction &vec, const TensorFunction &mat, Stash &stash) {
bool common_inner = (mat.result_type().dimension_index(vec.result_type().dimensions()[0].name) == 1);
return stash.create<DenseXWProductFunction>(res, vec, mat,
vec.result_type().dimensions()[0].size,
res.dimensions()[0].size,
common_inner);
}
struct MyXWProductOp {
template<typename LCM, typename RCM, typename CommonInner> static auto invoke() {
constexpr CellMeta ocm = CellMeta::join(LCM::value, RCM::value).reduce(false);
using LCT = CellValueType<LCM::value.cell_type>;
using RCT = CellValueType<RCM::value.cell_type>;
using OCT = CellValueType<ocm.cell_type>;
if (std::is_same_v<LCT,double> && std::is_same_v<RCT,double>) {
assert((std::is_same_v<OCT,double>));
return my_cblas_double_xw_product_op<CommonInner::value>;
} else if (std::is_same_v<LCT,float> && std::is_same_v<RCT,float>) {
assert((std::is_same_v<OCT,float>));
return my_cblas_float_xw_product_op<CommonInner::value>;
} else {
return my_xw_product_op<LCT, RCT, OCT, CommonInner::value>;
}
}
};
} // namespace <unnamed>
DenseXWProductFunction::Self::Self(const ValueType &result_type_in,
size_t vector_size_in, size_t result_size_in)
: result_type(result_type_in),
vector_size(vector_size_in),
result_size(result_size_in)
{
}
DenseXWProductFunction::Self::~Self() = default;
DenseXWProductFunction::DenseXWProductFunction(const ValueType &result_type,
const TensorFunction &vector_in,
const TensorFunction &matrix_in,
size_t vector_size,
size_t result_size,
bool common_inner)
: tensor_function::Op2(result_type, vector_in, matrix_in),
_vector_size(vector_size),
_result_size(result_size),
_common_inner(common_inner)
{
}
InterpretedFunction::Instruction
DenseXWProductFunction::compile_self(const CTFContext &ctx) const
{
Self &self = ctx.stash.create<Self>(result_type(), _vector_size, _result_size);
assert(self.result_type.cell_meta().is_scalar == false);
using MyTypify = TypifyValue<TypifyCellMeta,vespalib::TypifyBool>;
auto op = typify_invoke<3,MyTypify,MyXWProductOp>(lhs().result_type().cell_meta().not_scalar(),
rhs().result_type().cell_meta().not_scalar(),
_common_inner);
return InterpretedFunction::Instruction(op, wrap_param<DenseXWProductFunction::Self>(self));
}
void
DenseXWProductFunction::visit_self(vespalib::ObjectVisitor &visitor) const
{
Super::visit_self(visitor);
visitor.visitInt("vector_size", _vector_size);
visitor.visitInt("result_size", _result_size);
visitor.visitBool("common_inner", _common_inner);
}
const TensorFunction &
DenseXWProductFunction::optimize(const TensorFunction &expr, Stash &stash)
{
const Reduce *reduce = as<Reduce>(expr);
if (reduce && (reduce->aggr() == Aggr::SUM)) {
const ValueType &result_type = reduce->result_type();
const Join *join = as<Join>(reduce->child());
if (join && (join->function() == Mul::f)) {
const TensorFunction &lhs = join->lhs();
const TensorFunction &rhs = join->rhs();
if (isDenseXWProduct(result_type, lhs.result_type(), rhs.result_type())) {
return createDenseXWProduct(result_type, lhs, rhs, stash);
}
if (isDenseXWProduct(result_type, rhs.result_type(), lhs.result_type())) {
return createDenseXWProduct(result_type, rhs, lhs, stash);
}
}
}
return expr;
}
} // namespace
|
