bkietz commented on code in PR #401:
URL: https://github.com/apache/arrow-nanoarrow/pull/401#discussion_r1526651573
##########
dev/benchmarks/c/array_benchmark.cc:
##########
@@ -44,93 +55,364 @@ ArrowErrorCode InitSchemaAndArrayPrimitive(ArrowSchema*
schema, ArrowArray* arra
ArrowBitmapAppendInt8Unsafe(validity_bitmap, validity.data(),
validity.size());
}
+ // Set the length
+ switch (type) {
+ case NANOARROW_TYPE_STRING:
+ case NANOARROW_TYPE_LARGE_STRING:
+ case NANOARROW_TYPE_BINARY:
+ case NANOARROW_TYPE_LARGE_BINARY:
+ if (buffer1.size() > 0) {
+ array->length = buffer1.size() - 1;
+ } else {
+ array->length = 0;
+ }
+ break;
+
+ default:
+ array->length = buffer1.size();
+ break;
+ }
+
+ // Set the null count
+ if (validity.size() > 0) {
+ array->null_count = array->length -
ArrowBitCountSet(ArrowArrayBuffer(array, 0)->data,
+ 0, array->length);
+ } else {
+ array->null_count = 0;
+ }
+
NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ NANOARROW_RETURN_NOT_OK(
+ ArrowArrayViewInitFromSchema(array_view, schema.get(), nullptr));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayViewSetArray(array_view, array, nullptr));
return NANOARROW_OK;
}
template <typename CType, ArrowType type>
-static void BaseArrayViewGetIntUnsafe(benchmark::State& state, double
prop_null = 0.0) {
- nanoarrow::UniqueSchema schema;
+static void BaseArrayViewGetInt(benchmark::State& state) {
nanoarrow::UniqueArray array;
nanoarrow::UniqueArrayView array_view;
- int64_t n_values = 1000000;
+ int64_t n_values = kNumItemsPrettyBig;
std::vector<CType> values(n_values);
for (int64_t i = 0; i < n_values; i++) {
values[i] = i % std::numeric_limits<CType>::max();
}
- std::vector<int8_t> validity;
+ NANOARROW_THROW_NOT_OK(
+ InitArrayViewFromBuffers(type, array.get(), array_view.get(), {},
values));
- if (prop_null > 0) {
- int64_t num_nulls = n_values * prop_null;
- int64_t null_spacing = n_values / num_nulls;
- validity.resize(n_values);
+ std::vector<CType> values_out(n_values);
+ for (auto _ : state) {
for (int64_t i = 0; i < n_values; i++) {
- validity[i] = i % null_spacing != 0;
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
+ benchmark::DoNotOptimize(values_out);
}
- int code = InitSchemaAndArrayPrimitive<CType, type>(
- schema.get(), array.get(), std::move(values), std::move(validity));
- NANOARROW_THROW_NOT_OK(code);
- NANOARROW_THROW_NOT_OK(
- ArrowArrayViewInitFromSchema(array_view.get(), schema.get(), nullptr));
- NANOARROW_THROW_NOT_OK(ArrowArrayViewSetArray(array_view.get(), array.get(),
nullptr));
+ state.SetItemsProcessed(n_values * state.iterations());
+}
- std::vector<CType> values_out(n_values);
+/// \brief Use ArrowArrayViewGet() to consume an int8 array
+static void BenchmarkArrayViewGetInt8(benchmark::State& state) {
+ BaseArrayViewGetInt<int8_t, NANOARROW_TYPE_INT8>(state);
+}
- if (prop_null > 0) {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
- if (ArrowArrayViewIsNull(array_view.get(), i)) {
- values_out[i] = 0;
- } else {
- values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
- }
+/// \brief Use ArrowArrayViewGet() to consume an int16 array
+static void BenchmarkArrayViewGetInt16(benchmark::State& state) {
+ BaseArrayViewGetInt<int16_t, NANOARROW_TYPE_INT16>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int32 array
+static void BenchmarkArrayViewGetInt32(benchmark::State& state) {
+ BaseArrayViewGetInt<int32_t, NANOARROW_TYPE_INT32>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int64 array
+static void BenchmarkArrayViewGetInt64(benchmark::State& state) {
+ BaseArrayViewGetInt<int64_t, NANOARROW_TYPE_INT64>(state);
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that does not contain a validity buffer.
+static void BenchmarkArrayViewIsNullNonNullable(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), {},
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
- } else {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that contains 20% nulls.
+static void BenchmarkArrayViewIsNull(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ // Create validity buffer
+ double prop_null = 0.2;
+ int64_t num_nulls = n_values * prop_null;
+ int64_t null_spacing = n_values / num_nulls;
+
+ std::vector<int8_t> validity(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ validity[i] = i % null_spacing != 0;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), validity,
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewGetStringUnsafe() to consume a string array
+static void BenchmarkArrayViewGetString(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ // Create an array of relatively small strings
+ int64_t n_values = kNumItemsPrettyBig;
+ int64_t value_size = 7;
+ std::string alphabet =
"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+
+ std::vector<int32_t> offsets(n_values + 1);
+ for (int64_t i = 0; i < n_values; i++) {
+ offsets[i + 1] = i * value_size;
+ }
+
+ int64_t n_alphabets = n_values / alphabet.size() + 1;
+ std::vector<char> data(alphabet.size() * n_alphabets);
+ for (int64_t data_pos = 0; data_pos < data.size(); data_pos +=
alphabet.size()) {
+ memcpy(data.data() + data_pos, alphabet.data(), alphabet.size());
+ }
+
+ // Read the array as non-copying views. Possibly less realistic than
+ // what somebody might actually do, but also is a more direct benchmark
+ // of the overhead associated with calling it.
+ std::vector<ArrowStringView> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ values_out[i] = ArrowArrayViewGetStringUnsafe(array_view.get(), i);
+ }
+ benchmark::DoNotOptimize(values_out);
+ }
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// @}
+
+/// \defgroup nanoarrow-benchmark-array ArrowArray-related benchmarks
+///
+/// Benchmarks for producing ArrowArrays using the `ArrowArrayXXX()` functions.
+///
+/// @{
+
+template <typename CType, ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayInt(ArrowArray* array,
+ const std::vector<CType>&
values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ for (int64_t i = 0; i < values.size(); i++) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendInt(array, values[i]));
+ }
+
+ NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ return NANOARROW_OK;
+}
+
+template <ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayString(
+ ArrowArray* array, const std::vector<std::string>& values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ ArrowStringView view;
+ for (int64_t i = 0; i < values.size(); i++) {
+ const std::string& item = values[i];
+ view.data = item.data();
+ view.size_bytes = item.size();
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendString(array, view));
+ }
Review Comment:
```suggestion
for (const std::string& s : values) {
NANOARROW_RETURN_NOT_OK(ArrowArrayAppendString(array, {s.data(),
s.size()}));
}
```
##########
dev/benchmarks/c/array_benchmark.cc:
##########
@@ -44,93 +55,364 @@ ArrowErrorCode InitSchemaAndArrayPrimitive(ArrowSchema*
schema, ArrowArray* arra
ArrowBitmapAppendInt8Unsafe(validity_bitmap, validity.data(),
validity.size());
}
+ // Set the length
+ switch (type) {
+ case NANOARROW_TYPE_STRING:
+ case NANOARROW_TYPE_LARGE_STRING:
+ case NANOARROW_TYPE_BINARY:
+ case NANOARROW_TYPE_LARGE_BINARY:
+ if (buffer1.size() > 0) {
+ array->length = buffer1.size() - 1;
+ } else {
+ array->length = 0;
+ }
+ break;
+
+ default:
+ array->length = buffer1.size();
+ break;
+ }
+
+ // Set the null count
+ if (validity.size() > 0) {
+ array->null_count = array->length -
ArrowBitCountSet(ArrowArrayBuffer(array, 0)->data,
+ 0, array->length);
+ } else {
+ array->null_count = 0;
+ }
+
NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ NANOARROW_RETURN_NOT_OK(
+ ArrowArrayViewInitFromSchema(array_view, schema.get(), nullptr));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayViewSetArray(array_view, array, nullptr));
return NANOARROW_OK;
}
template <typename CType, ArrowType type>
-static void BaseArrayViewGetIntUnsafe(benchmark::State& state, double
prop_null = 0.0) {
- nanoarrow::UniqueSchema schema;
+static void BaseArrayViewGetInt(benchmark::State& state) {
nanoarrow::UniqueArray array;
nanoarrow::UniqueArrayView array_view;
- int64_t n_values = 1000000;
+ int64_t n_values = kNumItemsPrettyBig;
std::vector<CType> values(n_values);
for (int64_t i = 0; i < n_values; i++) {
values[i] = i % std::numeric_limits<CType>::max();
}
- std::vector<int8_t> validity;
+ NANOARROW_THROW_NOT_OK(
+ InitArrayViewFromBuffers(type, array.get(), array_view.get(), {},
values));
- if (prop_null > 0) {
- int64_t num_nulls = n_values * prop_null;
- int64_t null_spacing = n_values / num_nulls;
- validity.resize(n_values);
+ std::vector<CType> values_out(n_values);
+ for (auto _ : state) {
for (int64_t i = 0; i < n_values; i++) {
- validity[i] = i % null_spacing != 0;
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
+ benchmark::DoNotOptimize(values_out);
}
- int code = InitSchemaAndArrayPrimitive<CType, type>(
- schema.get(), array.get(), std::move(values), std::move(validity));
- NANOARROW_THROW_NOT_OK(code);
- NANOARROW_THROW_NOT_OK(
- ArrowArrayViewInitFromSchema(array_view.get(), schema.get(), nullptr));
- NANOARROW_THROW_NOT_OK(ArrowArrayViewSetArray(array_view.get(), array.get(),
nullptr));
+ state.SetItemsProcessed(n_values * state.iterations());
+}
- std::vector<CType> values_out(n_values);
+/// \brief Use ArrowArrayViewGet() to consume an int8 array
+static void BenchmarkArrayViewGetInt8(benchmark::State& state) {
+ BaseArrayViewGetInt<int8_t, NANOARROW_TYPE_INT8>(state);
+}
- if (prop_null > 0) {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
- if (ArrowArrayViewIsNull(array_view.get(), i)) {
- values_out[i] = 0;
- } else {
- values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
- }
+/// \brief Use ArrowArrayViewGet() to consume an int16 array
+static void BenchmarkArrayViewGetInt16(benchmark::State& state) {
+ BaseArrayViewGetInt<int16_t, NANOARROW_TYPE_INT16>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int32 array
+static void BenchmarkArrayViewGetInt32(benchmark::State& state) {
+ BaseArrayViewGetInt<int32_t, NANOARROW_TYPE_INT32>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int64 array
+static void BenchmarkArrayViewGetInt64(benchmark::State& state) {
+ BaseArrayViewGetInt<int64_t, NANOARROW_TYPE_INT64>(state);
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that does not contain a validity buffer.
+static void BenchmarkArrayViewIsNullNonNullable(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), {},
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
- } else {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that contains 20% nulls.
+static void BenchmarkArrayViewIsNull(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ // Create validity buffer
+ double prop_null = 0.2;
+ int64_t num_nulls = n_values * prop_null;
+ int64_t null_spacing = n_values / num_nulls;
+
+ std::vector<int8_t> validity(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ validity[i] = i % null_spacing != 0;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), validity,
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewGetStringUnsafe() to consume a string array
+static void BenchmarkArrayViewGetString(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ // Create an array of relatively small strings
+ int64_t n_values = kNumItemsPrettyBig;
+ int64_t value_size = 7;
+ std::string alphabet =
"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+
+ std::vector<int32_t> offsets(n_values + 1);
+ for (int64_t i = 0; i < n_values; i++) {
+ offsets[i + 1] = i * value_size;
+ }
+
+ int64_t n_alphabets = n_values / alphabet.size() + 1;
+ std::vector<char> data(alphabet.size() * n_alphabets);
+ for (int64_t data_pos = 0; data_pos < data.size(); data_pos +=
alphabet.size()) {
+ memcpy(data.data() + data_pos, alphabet.data(), alphabet.size());
+ }
+
+ // Read the array as non-copying views. Possibly less realistic than
+ // what somebody might actually do, but also is a more direct benchmark
+ // of the overhead associated with calling it.
+ std::vector<ArrowStringView> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ values_out[i] = ArrowArrayViewGetStringUnsafe(array_view.get(), i);
+ }
+ benchmark::DoNotOptimize(values_out);
+ }
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// @}
+
+/// \defgroup nanoarrow-benchmark-array ArrowArray-related benchmarks
+///
+/// Benchmarks for producing ArrowArrays using the `ArrowArrayXXX()` functions.
+///
+/// @{
+
+template <typename CType, ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayInt(ArrowArray* array,
+ const std::vector<CType>&
values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ for (int64_t i = 0; i < values.size(); i++) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendInt(array, values[i]));
+ }
+
+ NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ return NANOARROW_OK;
+}
+
+template <ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayString(
+ ArrowArray* array, const std::vector<std::string>& values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ ArrowStringView view;
+ for (int64_t i = 0; i < values.size(); i++) {
+ const std::string& item = values[i];
+ view.data = item.data();
+ view.size_bytes = item.size();
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendString(array, view));
+ }
+
+ NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ return NANOARROW_OK;
+}
+
+/// \brief Use ArrowArrayAppendString() to build a string array
+static void BenchmarkArrayAppendString(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+
+ int64_t n_values = kNumItemsPrettyBig;
+ int64_t value_size = 7;
+ std::string alphabet =
"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+
+ std::vector<std::string> values(n_values);
+ int64_t alphabet_pos = 0;
+ for (int64_t i = 0; i < n_values; i++) {
+ if ((alphabet_pos + value_size) >= alphabet.size()) {
+ alphabet_pos = 0;
+ }
+
+ std::string value(alphabet.data() + alphabet_pos, value_size);
+ alphabet_pos += value_size;
+ values[i] = value;
+ }
+
+ for (auto _ : state) {
+ array.reset();
+ NANOARROW_THROW_NOT_OK(
+ CreateAndAppendToArrayString<NANOARROW_TYPE_STRING>(array.get(),
values));
+ benchmark::DoNotOptimize(array);
}
state.SetItemsProcessed(n_values * state.iterations());
}
-/// \brief Use ArrowArrayViewGetIntUnsafe() to consume an int8 array
-static void BenchmarkArrayViewGetIntUnsafeInt8(benchmark::State& state) {
- BaseArrayViewGetIntUnsafe<int8_t, NANOARROW_TYPE_INT8>(state);
+template <typename CType, ArrowType type>
+static void BaseBenchmarkArrayAppendInt(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ std::vector<CType> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % std::numeric_limits<CType>::max();
+ }
+
+ for (auto _ : state) {
+ array.reset();
+ int code = CreateAndAppendToArrayInt<CType, type>(array.get(), values);
+ NANOARROW_THROW_NOT_OK(code);
+ benchmark::DoNotOptimize(array);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayAppendInt() to build an int8 array
+static void BenchmarkArrayAppendInt8(benchmark::State& state) {
+ BaseBenchmarkArrayAppendInt<int8_t, NANOARROW_TYPE_INT8>(state);
+}
+
Review Comment:
FWIW, you can use BENCHMARK with template instantiations:
```c++
BENCHMARK(BenchmarkArrayAppend<int8_t, NANOARROW_TYPE_INT8>);
```
##########
dev/benchmarks/c/array_benchmark.cc:
##########
@@ -44,93 +55,364 @@ ArrowErrorCode InitSchemaAndArrayPrimitive(ArrowSchema*
schema, ArrowArray* arra
ArrowBitmapAppendInt8Unsafe(validity_bitmap, validity.data(),
validity.size());
}
+ // Set the length
+ switch (type) {
+ case NANOARROW_TYPE_STRING:
+ case NANOARROW_TYPE_LARGE_STRING:
+ case NANOARROW_TYPE_BINARY:
+ case NANOARROW_TYPE_LARGE_BINARY:
+ if (buffer1.size() > 0) {
+ array->length = buffer1.size() - 1;
+ } else {
+ array->length = 0;
+ }
+ break;
+
+ default:
+ array->length = buffer1.size();
+ break;
+ }
+
+ // Set the null count
+ if (validity.size() > 0) {
+ array->null_count = array->length -
ArrowBitCountSet(ArrowArrayBuffer(array, 0)->data,
+ 0, array->length);
+ } else {
+ array->null_count = 0;
+ }
+
NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ NANOARROW_RETURN_NOT_OK(
+ ArrowArrayViewInitFromSchema(array_view, schema.get(), nullptr));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayViewSetArray(array_view, array, nullptr));
return NANOARROW_OK;
}
template <typename CType, ArrowType type>
-static void BaseArrayViewGetIntUnsafe(benchmark::State& state, double
prop_null = 0.0) {
- nanoarrow::UniqueSchema schema;
+static void BaseArrayViewGetInt(benchmark::State& state) {
nanoarrow::UniqueArray array;
nanoarrow::UniqueArrayView array_view;
- int64_t n_values = 1000000;
+ int64_t n_values = kNumItemsPrettyBig;
std::vector<CType> values(n_values);
for (int64_t i = 0; i < n_values; i++) {
values[i] = i % std::numeric_limits<CType>::max();
}
- std::vector<int8_t> validity;
+ NANOARROW_THROW_NOT_OK(
+ InitArrayViewFromBuffers(type, array.get(), array_view.get(), {},
values));
- if (prop_null > 0) {
- int64_t num_nulls = n_values * prop_null;
- int64_t null_spacing = n_values / num_nulls;
- validity.resize(n_values);
+ std::vector<CType> values_out(n_values);
+ for (auto _ : state) {
for (int64_t i = 0; i < n_values; i++) {
- validity[i] = i % null_spacing != 0;
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
+ benchmark::DoNotOptimize(values_out);
}
- int code = InitSchemaAndArrayPrimitive<CType, type>(
- schema.get(), array.get(), std::move(values), std::move(validity));
- NANOARROW_THROW_NOT_OK(code);
- NANOARROW_THROW_NOT_OK(
- ArrowArrayViewInitFromSchema(array_view.get(), schema.get(), nullptr));
- NANOARROW_THROW_NOT_OK(ArrowArrayViewSetArray(array_view.get(), array.get(),
nullptr));
+ state.SetItemsProcessed(n_values * state.iterations());
+}
- std::vector<CType> values_out(n_values);
+/// \brief Use ArrowArrayViewGet() to consume an int8 array
+static void BenchmarkArrayViewGetInt8(benchmark::State& state) {
+ BaseArrayViewGetInt<int8_t, NANOARROW_TYPE_INT8>(state);
+}
- if (prop_null > 0) {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
- if (ArrowArrayViewIsNull(array_view.get(), i)) {
- values_out[i] = 0;
- } else {
- values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
- }
+/// \brief Use ArrowArrayViewGet() to consume an int16 array
+static void BenchmarkArrayViewGetInt16(benchmark::State& state) {
+ BaseArrayViewGetInt<int16_t, NANOARROW_TYPE_INT16>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int32 array
+static void BenchmarkArrayViewGetInt32(benchmark::State& state) {
+ BaseArrayViewGetInt<int32_t, NANOARROW_TYPE_INT32>(state);
+}
+
+/// \brief Use ArrowArrayViewGet() to consume an int64 array
+static void BenchmarkArrayViewGetInt64(benchmark::State& state) {
+ BaseArrayViewGetInt<int64_t, NANOARROW_TYPE_INT64>(state);
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that does not contain a validity buffer.
+static void BenchmarkArrayViewIsNullNonNullable(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), {},
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
+ values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
- } else {
- for (auto _ : state) {
- for (int64_t i = 0; i < n_values; i++) {
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewIsNull() to check for nulls while consuming an
int32 array
+/// that contains 20% nulls.
+static void BenchmarkArrayViewIsNull(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ int64_t n_values = kNumItemsPrettyBig;
+
+ // Create values
+ std::vector<int32_t> values(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ values[i] = i % 1000;
+ }
+
+ // Create validity buffer
+ double prop_null = 0.2;
+ int64_t num_nulls = n_values * prop_null;
+ int64_t null_spacing = n_values / num_nulls;
+
+ std::vector<int8_t> validity(n_values);
+ for (int64_t i = 0; i < n_values; i++) {
+ validity[i] = i % null_spacing != 0;
+ }
+
+ NANOARROW_THROW_NOT_OK(InitArrayViewFromBuffers(NANOARROW_TYPE_INT32,
array.get(),
+ array_view.get(), validity,
values));
+
+ // Read the array
+ std::vector<int32_t> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ if (ArrowArrayViewIsNull(array_view.get(), i)) {
+ values_out[i] = 0;
+ } else {
values_out[i] = ArrowArrayViewGetIntUnsafe(array_view.get(), i);
}
- benchmark::DoNotOptimize(values_out);
}
+ benchmark::DoNotOptimize(values_out);
+ }
+
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// \brief Use ArrowArrayViewGetStringUnsafe() to consume a string array
+static void BenchmarkArrayViewGetString(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+ nanoarrow::UniqueArrayView array_view;
+
+ // Create an array of relatively small strings
+ int64_t n_values = kNumItemsPrettyBig;
+ int64_t value_size = 7;
+ std::string alphabet =
"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+
+ std::vector<int32_t> offsets(n_values + 1);
+ for (int64_t i = 0; i < n_values; i++) {
+ offsets[i + 1] = i * value_size;
+ }
+
+ int64_t n_alphabets = n_values / alphabet.size() + 1;
+ std::vector<char> data(alphabet.size() * n_alphabets);
+ for (int64_t data_pos = 0; data_pos < data.size(); data_pos +=
alphabet.size()) {
+ memcpy(data.data() + data_pos, alphabet.data(), alphabet.size());
+ }
+
+ // Read the array as non-copying views. Possibly less realistic than
+ // what somebody might actually do, but also is a more direct benchmark
+ // of the overhead associated with calling it.
+ std::vector<ArrowStringView> values_out(n_values);
+ for (auto _ : state) {
+ for (int64_t i = 0; i < n_values; i++) {
+ values_out[i] = ArrowArrayViewGetStringUnsafe(array_view.get(), i);
+ }
+ benchmark::DoNotOptimize(values_out);
+ }
+ state.SetItemsProcessed(n_values * state.iterations());
+}
+
+/// @}
+
+/// \defgroup nanoarrow-benchmark-array ArrowArray-related benchmarks
+///
+/// Benchmarks for producing ArrowArrays using the `ArrowArrayXXX()` functions.
+///
+/// @{
+
+template <typename CType, ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayInt(ArrowArray* array,
+ const std::vector<CType>&
values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ for (int64_t i = 0; i < values.size(); i++) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendInt(array, values[i]));
+ }
+
+ NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ return NANOARROW_OK;
+}
+
+template <ArrowType type>
+static ArrowErrorCode CreateAndAppendToArrayString(
+ ArrowArray* array, const std::vector<std::string>& values) {
+ NANOARROW_RETURN_NOT_OK(ArrowArrayInitFromType(array, type));
+ NANOARROW_RETURN_NOT_OK(ArrowArrayStartAppending(array));
+
+ ArrowStringView view;
+ for (int64_t i = 0; i < values.size(); i++) {
+ const std::string& item = values[i];
+ view.data = item.data();
+ view.size_bytes = item.size();
+ NANOARROW_RETURN_NOT_OK(ArrowArrayAppendString(array, view));
+ }
+
+ NANOARROW_RETURN_NOT_OK(ArrowArrayFinishBuildingDefault(array, nullptr));
+ return NANOARROW_OK;
+}
+
+/// \brief Use ArrowArrayAppendString() to build a string array
+static void BenchmarkArrayAppendString(benchmark::State& state) {
+ nanoarrow::UniqueArray array;
+
+ int64_t n_values = kNumItemsPrettyBig;
+ int64_t value_size = 7;
+ std::string alphabet =
"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+
+ std::vector<std::string> values(n_values);
+ int64_t alphabet_pos = 0;
+ for (int64_t i = 0; i < n_values; i++) {
+ if ((alphabet_pos + value_size) >= alphabet.size()) {
+ alphabet_pos = 0;
+ }
+
+ std::string value(alphabet.data() + alphabet_pos, value_size);
+ alphabet_pos += value_size;
+ values[i] = value;
+ }
Review Comment:
```suggestion
std::vector<std::string> values(kNumItemsPrettyBig);
int64_t alphabet_pos = 0;
for (std::string& value : values) {
if ((alphabet_pos + value_size) >= alphabet.size()) {
alphabet_pos = 0;
}
value.assign(alphabet.data() + alphabet_pos, value_size);
alphabet_pos += value_size;
}
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
