vuron/adept: Платформа для обучения и вывода нейронных сетей, с API приближенным к PyTorch.

tests/broadcast_tests.cpp
192 строки6 KB
kolkir
Update tests remove int8 type
28 фев 2025, 14:13
28 фев 2025, 14:134016935
#include <adept/tensor.hpp>
#include <adept/tensor_print.hpp>
#include <adept/types.hpp>

#include "catch.hpp"
#include "test_utils.hpp"

using namespace adept;
using namespace adept::test;

// Scalar vector tests
TEMPLATE_TEST_CASE("Tensor add scalar vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_scalar_test_data<TestType, std::plus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x + y;
  equals(dz, z);

  x += y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor sub scalar vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_scalar_test_data<TestType, std::minus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x - y;
  equals(dz, z);

  x -= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor mul scalar vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_scalar_test_data<TestType, std::multiplies>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x * y;
  equals(dz, z);

  x *= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor div scalar vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_scalar_test_data<TestType, std::divides>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x / y;
  equals(dz, z);

  x /= y;
  equals(dz, x);
}

// Row vector tests
TEMPLATE_TEST_CASE("Tensor add row vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_row_broadcast_test_data<TestType, std::plus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x + y;
  equals(dz, z);

  x += y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor sub row vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_row_broadcast_test_data<TestType, std::minus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x - y;
  equals(dz, z);

  x -= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor mul row vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_row_broadcast_test_data<TestType, std::multiplies>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x * y;
  equals(dz, z);

  x *= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor div row vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_row_broadcast_test_data<TestType, std::divides>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {1, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x / y;
  equals(dz, z);

  x /= y;
  equals(dz, x);
}

// Column vector tests
TEMPLATE_TEST_CASE("Tensor add col vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_col_broadcast_test_data<TestType, std::plus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {3, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x + y;
  equals(dz, z);

  x += y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor sub col vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_col_broadcast_test_data<TestType, std::minus>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {3, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x - y;
  equals(dz, z);

  x -= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor mul col vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_col_broadcast_test_data<TestType, std::multiplies>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {3, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x * y;
  equals(dz, z);

  x *= y;
  equals(dz, x);
}

TEMPLATE_TEST_CASE("Tensor div col vector", "[broadcast]", float32_t, float64_t, int32_t) {
  auto [dx, dy, dz] = make_col_broadcast_test_data<TestType, std::divides>();
  auto x = Tensor::from_blob(
      dx.data(), {.shape = {3, 3}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto y = Tensor::from_blob(
      dy.data(), {.shape = {3, 1}, .device = device_t::CPU, .dtype = to_dtype<TestType>()});

  auto z = x / y;
  equals(dz, z);

  x /= y;
  equals(dz, x);
}