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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
|
#include "mkql_block_way.h"
#include <yql/essentials/minikql/arrow/arrow_util.h>
#include <yql/essentials/minikql/computation/mkql_block_impl.h>
#include <yql/essentials/minikql/computation/mkql_computation_node_holders.h>
#include <yql/essentials/minikql/mkql_node_builder.h>
#include <yql/essentials/minikql/mkql_node_cast.h>
#include <yql/essentials/public/udf/arrow/block_builder.h>
#include <yql/essentials/public/udf/arrow/dense_union_scalar.h>
#include <arrow/util/bitmap_ops.h>
#include <arrow/util/checked_cast.h>
namespace NKikimr::NMiniKQL {
namespace {
template <bool IsOptional>
class TWayBlockExecBase {
public:
explicit TWayBlockExecBase(TType* resultItemType)
: ResultItemType_(resultItemType)
{
}
virtual ~TWayBlockExecBase() = default;
arrow::Status Exec(arrow::compute::KernelContext* ctx, const arrow::compute::ExecBatch& batch, arrow::Datum* res) const {
const arrow::Datum& variantDatum = batch.values[0];
if (variantDatum.is_scalar()) {
*res = ConvertScalar(ResultItemType_, ComputeScalarKey(*variantDatum.scalar()), *ctx->memory_pool());
return arrow::Status::OK();
}
MKQL_ENSURE(variantDatum.is_array(), "Expected array datum");
const arrow::ArrayData& inputArray = *variantDatum.array();
const i8* typeCodes = GetUnionArray(inputArray).GetValues<i8>(1);
return ExecArray(ctx, inputArray, typeCodes, res);
}
protected:
static const arrow::ArrayData& GetUnionArray(const arrow::ArrayData& inputArray) {
if constexpr (IsOptional) {
return *inputArray.child_data[0];
} else {
return inputArray;
}
}
private:
virtual NUdf::TBlockItem MakeKey(ui32 alternativeIndex) const = 0;
virtual arrow::Status ExecArray(arrow::compute::KernelContext* ctx, const arrow::ArrayData& inputArray,
const i8* typeCodes, arrow::Datum* res) const = 0;
NUdf::TBlockItem ComputeScalarKey(const arrow::Scalar& scalar) const {
if constexpr (IsOptional) {
if (!scalar.is_valid) {
return NUdf::TBlockItem{};
}
const auto& structScalar = arrow::internal::checked_cast<const arrow::StructScalar&>(scalar);
const auto& unionScalar = arrow::internal::checked_cast<const NYql::NUdf::TDenseUnionScalar&>(*structScalar.value.front());
return MakeKey(unionScalar.Index);
} else {
const auto& unionScalar = arrow::internal::checked_cast<const NYql::NUdf::TDenseUnionScalar&>(scalar);
return MakeKey(unionScalar.Index);
}
}
TType* const ResultItemType_;
};
template <bool IsOptional>
class TWayBlockExecTuple final: public TWayBlockExecBase<IsOptional> {
using TBase = TWayBlockExecBase<IsOptional>;
public:
using TBase::TBase;
private:
NUdf::TBlockItem MakeKey(ui32 alternativeIndex) const final {
return NUdf::TBlockItem(alternativeIndex);
}
arrow::Status ExecArray(arrow::compute::KernelContext* ctx, const arrow::ArrayData& inputArray,
const i8* typeCodes, arrow::Datum* res) const final {
auto* pool = ctx->memory_pool();
const i64 length = inputArray.length;
std::shared_ptr<arrow::Buffer> values = ARROW_RESULT(arrow::AllocateBuffer(length * sizeof(ui32), pool));
ui32* outValues = reinterpret_cast<ui32*>(values->mutable_data());
for (i64 i = 0; i < length; ++i) {
outValues[i] = static_cast<ui32>(typeCodes[i]);
}
std::shared_ptr<arrow::Buffer> mask;
if (IsOptional && inputArray.buffers[0]) {
mask = MakeDenseBitmapCopyIfOffsetDiffers(inputArray.buffers[0], length, inputArray.offset, 0, pool);
}
*res = arrow::ArrayData::Make(arrow::uint32(), length, {std::move(mask), std::move(values)});
return arrow::Status::OK();
}
};
template <bool IsOptional>
class TWayBlockExecStruct final: public TWayBlockExecBase<IsOptional> {
using TBase = TWayBlockExecBase<IsOptional>;
public:
TWayBlockExecStruct(TType* resultItemType, TVector<TString>&& alternativeNames)
: TBase(resultItemType)
, AlternativeNames_(std::move(alternativeNames))
{
}
private:
NUdf::TBlockItem MakeKey(ui32 alternativeIndex) const final {
return NUdf::TBlockItem(NUdf::TStringRef(AlternativeNames_[alternativeIndex]));
}
arrow::Status ExecArray(arrow::compute::KernelContext* ctx, const arrow::ArrayData& inputArray,
const i8* typeCodes, arrow::Datum* res) const final {
auto* pool = ctx->memory_pool();
arrow::Datum built = BuildNameArray(pool, inputArray, typeCodes);
const ui8* maskBits = (IsOptional && inputArray.buffers[0]) ? inputArray.buffers[0]->data() : nullptr;
*res = maskBits ? ReattachNullMask(pool, built, inputArray, maskBits) : std::move(built);
return arrow::Status::OK();
}
arrow::Datum BuildNameArray(arrow::MemoryPool* pool, const arrow::ArrayData& inputArray, const i8* typeCodes) const {
const i64 length = inputArray.length;
NYql::NUdf::TStringArrayBuilder<arrow::StringType, false> builder(
TTypeInfoHelper(), arrow::utf8(), *pool, static_cast<size_t>(length));
for (i64 i = 0; i < length; ++i) {
const bool isNull = IsOptional && NYql::NUdf::IsNull(inputArray, static_cast<size_t>(i));
const NUdf::TStringRef name = isNull
? NUdf::TStringRef("", 0)
: NUdf::TStringRef(AlternativeNames_[static_cast<ui32>(typeCodes[i])]);
builder.Add(NUdf::TBlockItem(name));
}
return builder.Build(true);
}
static arrow::Datum ReattachNullMask(arrow::MemoryPool* pool, const arrow::Datum& built,
const arrow::ArrayData& inputArray, const ui8* maskBits) {
TVector<std::shared_ptr<arrow::ArrayData>> chunks;
i64 processed = 0;
ForEachArrayData(built, [&](const std::shared_ptr<arrow::ArrayData>& chunk) {
auto withMask = chunk->Copy();
MKQL_ENSURE(chunk->offset == 0, "Expected offset after string builder to be 0");
auto chunkMask = AllocateBitmapWithReserve(chunk->length, pool);
arrow::internal::CopyBitmap(maskBits, inputArray.offset + processed, chunk->length, chunkMask->mutable_data(), 0);
withMask->buffers[0] = std::move(chunkMask);
withMask->SetNullCount(arrow::kUnknownNullCount);
processed += chunk->length;
chunks.push_back(std::move(withMask));
});
return MakeArray(chunks);
}
const TVector<TString> AlternativeNames_;
};
template <bool IsOptional>
std::shared_ptr<arrow::compute::ScalarKernel> MakeBlockWayKernel(const TVector<TType*>& argTypes, TType* resultType,
std::shared_ptr<const TWayBlockExecBase<IsOptional>> exec) {
auto kernel = std::make_shared<arrow::compute::ScalarKernel>(
ConvertToInputTypes(argTypes),
ConvertToOutputType(resultType),
[exec](arrow::compute::KernelContext* ctx, const arrow::compute::ExecBatch& batch, arrow::Datum* res) {
return exec->Exec(ctx, batch, res);
});
kernel->null_handling = arrow::compute::NullHandling::COMPUTED_NO_PREALLOCATE;
kernel->mem_allocation = arrow::compute::MemAllocation::NO_PREALLOCATE;
return kernel;
}
TVector<TString> CollectStructAlternativeNames(TStructType* structType) {
TVector<TString> names;
names.reserve(structType->GetMembersCount());
for (ui32 i = 0; i < structType->GetMembersCount(); ++i) {
names.emplace_back(structType->GetMemberName(i));
}
return names;
}
template <bool IsOptional>
std::shared_ptr<arrow::compute::ScalarKernel> MakeBlockWayKernel(const TVector<TType*>& argTypes, TType* resultType,
TType* underlyingType) {
TType* resultItemType = AS_TYPE(TBlockType, resultType)->GetItemType();
std::shared_ptr<const TWayBlockExecBase<IsOptional>> exec;
if (underlyingType->IsTuple()) {
exec = std::make_shared<TWayBlockExecTuple<IsOptional>>(resultItemType);
} else {
exec = std::make_shared<TWayBlockExecStruct<IsOptional>>(
resultItemType, CollectStructAlternativeNames(AS_TYPE(TStructType, underlyingType)));
}
return MakeBlockWayKernel<IsOptional>(argTypes, resultType, std::move(exec));
}
} // namespace
IComputationNode* WrapBlockWay(TCallable& callable, const TComputationNodeFactoryContext& ctx) {
MKQL_ENSURE(callable.GetInputsCount() == 1, "Expected 1 argument");
auto blockType = AS_TYPE(TBlockType, callable.GetInput(0).GetStaticType());
bool isOptional;
auto variantItemType = UnpackOptional(blockType->GetItemType(), isOptional);
auto underlyingType = AS_TYPE(TVariantType, variantItemType)->GetUnderlyingType();
TVector<TType*> argsTypes = {blockType};
auto resultType = callable.GetType()->GetReturnType();
auto kernel = isOptional
? MakeBlockWayKernel<true>(argsTypes, resultType, underlyingType)
: MakeBlockWayKernel<false>(argsTypes, resultType, underlyingType);
TComputationNodePtrVector argsNodes = {LocateNode(ctx.NodeLocator, callable, 0)};
return new TBlockFuncNode(ctx.Mutables,
ctx.RuntimeSettings->DatumValidation.Get(),
callable.GetType()->GetName(),
std::move(argsNodes),
argsTypes,
resultType,
*kernel,
kernel);
}
} // namespace NKikimr::NMiniKQL
|
