mirror of
https://github.com/zama-ai/concrete.git
synced 2026-02-17 08:01:20 -05:00
319 lines
11 KiB
C++
319 lines
11 KiB
C++
// Part of the Concrete Compiler Project, under the BSD3 License with Zama
|
|
// Exceptions. See
|
|
// https://github.com/zama-ai/concrete/blob/main/LICENSE.txt
|
|
// for license information.
|
|
|
|
#include "concretelang/Common/Values.h"
|
|
#include "capnp/common.h"
|
|
#include "capnp/list.h"
|
|
#include "concrete-protocol.capnp.h"
|
|
#include "concretelang/Common/Error.h"
|
|
#include "concretelang/Common/Protocol.h"
|
|
#include <cstddef>
|
|
#include <cstdint>
|
|
#include <stdlib.h>
|
|
#include <string>
|
|
|
|
using concretelang::error::Result;
|
|
using concretelang::error::StringError;
|
|
using concretelang::protocol::dimensionsToProtoShape;
|
|
using concretelang::protocol::Message;
|
|
using concretelang::protocol::protoPayloadToVector;
|
|
using concretelang::protocol::protoShapeToDimensions;
|
|
using concretelang::protocol::vectorToProtoPayload;
|
|
|
|
namespace concretelang {
|
|
namespace values {
|
|
|
|
Value Value::fromRawTransportValue(TransportValue transportVal) {
|
|
Value output;
|
|
auto integerPrecision =
|
|
transportVal.asReader().getRawInfo().getIntegerPrecision();
|
|
auto isSigned = transportVal.asReader().getRawInfo().getIsSigned();
|
|
auto dimensions =
|
|
protoShapeToDimensions(transportVal.asReader().getRawInfo().getShape());
|
|
auto data = transportVal.asReader().getPayload();
|
|
if (integerPrecision == 8 && isSigned) {
|
|
auto values = protoPayloadToVector<int8_t>(data);
|
|
output.inner = Tensor<int8_t>{values, dimensions};
|
|
} else if (integerPrecision == 16 && isSigned) {
|
|
auto values = protoPayloadToVector<int16_t>(data);
|
|
output.inner = Tensor<int16_t>{values, dimensions};
|
|
} else if (integerPrecision == 32 && isSigned) {
|
|
auto values = protoPayloadToVector<int32_t>(data);
|
|
output.inner = Tensor<int32_t>{values, dimensions};
|
|
} else if (integerPrecision == 64 && isSigned) {
|
|
auto values = protoPayloadToVector<int64_t>(data);
|
|
output.inner = Tensor<int64_t>{values, dimensions};
|
|
} else if (integerPrecision == 8 && !isSigned) {
|
|
auto values = protoPayloadToVector<uint8_t>(data);
|
|
output.inner = Tensor<uint8_t>{values, dimensions};
|
|
} else if (integerPrecision == 16 && !isSigned) {
|
|
auto values = protoPayloadToVector<uint16_t>(data);
|
|
output.inner = Tensor<uint16_t>{values, dimensions};
|
|
} else if (integerPrecision == 32 && !isSigned) {
|
|
auto values = protoPayloadToVector<uint32_t>(data);
|
|
output.inner = Tensor<uint32_t>{values, dimensions};
|
|
} else if (integerPrecision == 64 && !isSigned) {
|
|
auto values = protoPayloadToVector<uint64_t>(data);
|
|
output.inner = Tensor<uint64_t>{values, dimensions};
|
|
} else {
|
|
assert(false);
|
|
}
|
|
|
|
return output;
|
|
}
|
|
|
|
TransportValue Value::intoRawTransportValue() const {
|
|
auto output = Message<concreteprotocol::Value>();
|
|
auto rawInfo = output.asBuilder().initRawInfo();
|
|
rawInfo.setShape(intoProtoShape().asReader());
|
|
rawInfo.setIntegerPrecision(getIntegerPrecision());
|
|
rawInfo.setIsSigned(isSigned());
|
|
output.asBuilder().setPayload(intoProtoPayload().asReader());
|
|
return output;
|
|
}
|
|
|
|
uint32_t Value::getIntegerPrecision() const {
|
|
if (hasElementType<uint8_t>() || hasElementType<int8_t>()) {
|
|
return 8;
|
|
} else if (hasElementType<uint16_t>() || hasElementType<int16_t>()) {
|
|
return 16;
|
|
} else if (hasElementType<uint32_t>() || hasElementType<int32_t>()) {
|
|
return 32;
|
|
} else if (hasElementType<uint64_t>() || hasElementType<int64_t>()) {
|
|
return 64;
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
bool Value::isSigned() const {
|
|
|
|
if (hasElementType<uint8_t>() || hasElementType<uint16_t>() ||
|
|
hasElementType<uint32_t>() || hasElementType<uint64_t>()) {
|
|
return false;
|
|
} else if (hasElementType<int8_t>() || hasElementType<int16_t>() ||
|
|
hasElementType<int32_t>() || hasElementType<int64_t>()) {
|
|
return true;
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
Message<concreteprotocol::Payload> Value::intoProtoPayload() const {
|
|
if (hasElementType<uint8_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<uint8_t>>(inner).values);
|
|
} else if (hasElementType<uint16_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<uint16_t>>(inner).values);
|
|
} else if (hasElementType<uint32_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<uint32_t>>(inner).values);
|
|
} else if (hasElementType<uint64_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<uint64_t>>(inner).values);
|
|
} else if (hasElementType<int8_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<int8_t>>(inner).values);
|
|
} else if (hasElementType<int16_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<int16_t>>(inner).values);
|
|
} else if (hasElementType<int32_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<int32_t>>(inner).values);
|
|
} else if (hasElementType<int64_t>()) {
|
|
return vectorToProtoPayload(std::get<Tensor<int64_t>>(inner).values);
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
Message<concreteprotocol::Shape> Value::intoProtoShape() const {
|
|
return dimensionsToProtoShape(getDimensions());
|
|
}
|
|
|
|
std::vector<size_t> Value::getDimensions() const {
|
|
if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return tensor.value().dimensions;
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
size_t Value::getLength() const {
|
|
if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return tensor.value().values.size();
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
bool Value::isCompatibleWithShape(
|
|
const Message<concreteprotocol::Shape> &shape) const {
|
|
auto dimensions = getDimensions();
|
|
if ((uint32_t)shape.asReader().getDimensions().size() != dimensions.size()) {
|
|
return false;
|
|
}
|
|
for (uint32_t i = 0; i < dimensions.size(); i++) {
|
|
if (shape.asReader().getDimensions()[i] != dimensions[i]) {
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool Value::operator==(const Value &b) const {
|
|
if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return tensor == b.getTensor<uint8_t>();
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return tensor == b.getTensor<uint16_t>();
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return tensor == b.getTensor<uint32_t>();
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return tensor == b.getTensor<uint64_t>();
|
|
} else if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return tensor == b.getTensor<int8_t>();
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return tensor == b.getTensor<int16_t>();
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return tensor == b.getTensor<int32_t>();
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return tensor == b.getTensor<int64_t>();
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
bool Value::isScalar() const {
|
|
if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return tensor.value().isScalar();
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
Value Value::toUnsigned() const {
|
|
if (!this->isSigned()) {
|
|
return *this;
|
|
} else if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return Value((Tensor<uint8_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return Value((Tensor<uint16_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return Value((Tensor<uint32_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return Value((Tensor<uint8_t>)tensor.value());
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
Value Value::toSigned() const {
|
|
if (!this->isSigned()) {
|
|
return *this;
|
|
} else if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return Value((Tensor<int8_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return Value((Tensor<int16_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return Value((Tensor<int32_t>)tensor.value());
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return Value((Tensor<int8_t>)tensor.value());
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
template <typename T>
|
|
std::string printTypeWithScalarTensor(std::string type, Tensor<T> tensor) {
|
|
std::stringstream str;
|
|
if (tensor.isScalar()) {
|
|
str << type << "(" << tensor.values[0] << ")";
|
|
} else {
|
|
str << type << "[](";
|
|
for (auto v : tensor.values) {
|
|
str << v << ",";
|
|
}
|
|
str << ")";
|
|
}
|
|
return str.str();
|
|
}
|
|
|
|
std::string Value::toString() const {
|
|
if (auto tensor = getTensor<int8_t>(); tensor) {
|
|
return printTypeWithScalarTensor("int8_t", *tensor);
|
|
} else if (auto tensor = getTensor<int16_t>(); tensor) {
|
|
return printTypeWithScalarTensor("int16_t", *tensor);
|
|
} else if (auto tensor = getTensor<int32_t>(); tensor) {
|
|
return printTypeWithScalarTensor("int32_t", *tensor);
|
|
} else if (auto tensor = getTensor<int64_t>(); tensor) {
|
|
return printTypeWithScalarTensor("int64_t", *tensor);
|
|
} else if (auto tensor = getTensor<uint8_t>(); tensor) {
|
|
return printTypeWithScalarTensor("uint8_t", *tensor);
|
|
} else if (auto tensor = getTensor<uint16_t>(); tensor) {
|
|
return printTypeWithScalarTensor("uint16_t", *tensor);
|
|
} else if (auto tensor = getTensor<uint32_t>(); tensor) {
|
|
return printTypeWithScalarTensor("uint32_t", *tensor);
|
|
} else if (auto tensor = getTensor<uint64_t>(); tensor) {
|
|
return printTypeWithScalarTensor("uint64_t", *tensor);
|
|
} else {
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
size_t getCorrespondingPrecision(size_t originalPrecision) {
|
|
if (originalPrecision <= 8) {
|
|
return 8;
|
|
}
|
|
if (originalPrecision <= 16) {
|
|
return 16;
|
|
}
|
|
if (originalPrecision <= 32) {
|
|
return 32;
|
|
}
|
|
if (originalPrecision <= 64) {
|
|
return 64;
|
|
}
|
|
assert(false);
|
|
}
|
|
|
|
} // namespace values
|
|
} // namespace concretelang
|