Add tests for sin

[AuroraPerego]

to run the tests:

cd test
make all
make runAll

[fwyzard]

Suggested change

	#define CATCH_CONFIG_MAIN
	#include <catch.hpp>
	#include "math.h"
	#include <cuda_runtime.h>
	#include <limits>
	#include <vector>
	template <typename T> __global__ void sinKernel(double *result, T input) {
	result[0] = static_cast<double>(xtd::sin(input));
	}
	template <typename T> __global__ void sinfKernel(double *result, T input) {
	result[0] = static_cast<double>(xtd::sinf(input));
	}
	TEST_CASE("sinCuda", "[sin]") {
	int deviceCount;
	cudaError_t cudaStatus = cudaGetDeviceCount(&deviceCount);
	if (cudaStatus != cudaSuccess || deviceCount == 0) {
	exit(EXIT_SUCCESS);
	}
	cudaSetDevice(0);
	cudaStream_t q;
	cudaStreamCreate(&q);
	std::vector<double> values{-1., 0., M_PI / 2, M_PI, 42.};
	double *result;
	int constexpr N = 6;
	cudaMallocAsync(&result, N * sizeof(double), q);
	for (auto v : values) {
	cudaMemsetAsync(&result, 0x00, N * sizeof(double), q);
	sinKernel<<<1, 1, 0, q>>>(&result[0], static_cast<int>(v));
	sinKernel<<<1, 1, 0, q>>>(&result[1], static_cast<float>(v));
	sinKernel<<<1, 1, 0, q>>>(&result[2], static_cast<double>(v));
	sinfKernel<<<1, 1, 0, q>>>(&result[3], static_cast<int>(v));
	sinfKernel<<<1, 1, 0, q>>>(&result[4], static_cast<float>(v));
	sinfKernel<<<1, 1, 0, q>>>(&result[5], static_cast<double>(v));
	double resultHost[N];
	cudaMemcpyAsync(resultHost, result, N * sizeof(double),
	cudaMemcpyDeviceToHost, q);
	cudaStreamSynchronize(q);
	auto const epsilon = std::numeric_limits<double>::epsilon();
	auto const epsilon_f = std::numeric_limits<float>::epsilon();
	REQUIRE_THAT(resultHost[0], Catch::Matchers::WithinAbs(
	std::sin(static_cast<int>(v)), epsilon));
	REQUIRE_THAT(resultHost[1],
	Catch::Matchers::WithinAbs(std::sin(v), epsilon_f));
	REQUIRE_THAT(resultHost[2],
	Catch::Matchers::WithinAbs(std::sin(v), epsilon));
	REQUIRE_THAT(resultHost[3], Catch::Matchers::WithinAbs(
	sinf(static_cast<int>(v)), epsilon_f));
	REQUIRE_THAT(resultHost[4], Catch::Matchers::WithinAbs(sinf(v), epsilon_f));
	REQUIRE_THAT(resultHost[5], Catch::Matchers::WithinAbs(sinf(v), epsilon_f));
	}
	cudaFreeAsync(result, q);
	}
	#include <limits>
	#include <vector>
	#include <cuda_runtime.h>
	#define CATCH_CONFIG_MAIN
	#include <catch.hpp>
	#include "math.h"
	#include "cuda_check.h"
	template <typename T> __global__ void sinKernel(double *result, T input) {
	*result = static_cast<double>(xtd::sin(input));
	}
	template <typename T> __global__ void sinfKernel(double *result, T input) {
	*result = static_cast<double>(xtd::sinf(input));
	}
	TEST_CASE("sinCuda", "[sin]") {
	int deviceCount;
	cudaError_t cudaStatus = cudaGetDeviceCount(&deviceCount);
	if (cudaStatus != cudaSuccess || deviceCount == 0) {
	exit(EXIT_SUCCESS);
	}
	CUDA_CHECK(cudaSetDevice(0));
	cudaStream_t q;
	CUDA_CHECK(cudaStreamCreate(&q));
	std::vector<double> values{-1., 0., M_PI / 2, M_PI, 42.};
	double *result;
	int constexpr N = 6;
	CUDA_CHECK(cudaMallocAsync(&result, N * sizeof(double), q));
	for (auto v : values) {
	CUDA_CHECK(cudaMemsetAsync(result, 0x00, N * sizeof(double), q));
	sinKernel<<<1, 1, 0, q>>>(result + 0, static_cast<int>(v));
	CUDA_CHECK(cudaGetLastError());
	sinKernel<<<1, 1, 0, q>>>(result + 1, static_cast<float>(v));
	CUDA_CHECK(cudaGetLastError());
	sinKernel<<<1, 1, 0, q>>>(result + 2, static_cast<double>(v));
	CUDA_CHECK(cudaGetLastError());
	sinfKernel<<<1, 1, 0, q>>>(result + 3, static_cast<int>(v));
	CUDA_CHECK(cudaGetLastError());
	sinfKernel<<<1, 1, 0, q>>>(result + 4, static_cast<float>(v));
	CUDA_CHECK(cudaGetLastError());
	sinfKernel<<<1, 1, 0, q>>>(result + 5, static_cast<double>(v));
	CUDA_CHECK(cudaGetLastError());
	double resultHost[N];
	CUDA_CHECK(cudaMemcpyAsync(resultHost, result, N * sizeof(double), cudaMemcpyDeviceToHost, q));
	CUDA_CHECK(cudaStreamSynchronize(q));
	auto const epsilon = std::numeric_limits<double>::epsilon();
	auto const epsilon_f = std::numeric_limits<float>::epsilon();
	REQUIRE_THAT(resultHost[0], Catch::Matchers::WithinAbs(std::sin(static_cast<int>(v)), epsilon));
	REQUIRE_THAT(resultHost[1], Catch::Matchers::WithinAbs(std::sin(v), epsilon_f));
	REQUIRE_THAT(resultHost[2], Catch::Matchers::WithinAbs(std::sin(v), epsilon));
	REQUIRE_THAT(resultHost[3], Catch::Matchers::WithinAbs(sinf(static_cast<int>(v)), epsilon_f));
	REQUIRE_THAT(resultHost[4], Catch::Matchers::WithinAbs(sinf(v), epsilon_f));
	REQUIRE_THAT(resultHost[5], Catch::Matchers::WithinAbs(sinf(v), epsilon_f));
	}
	CUDA_CHECK(cudaFreeAsync(result, q));
	CUDA_CHECK(cudaStreamDestroy(q));
	}

[fwyzard]

On my notebook I get

fwyzard@fools:~/src/xtd/test$ make
Running /home/fwyzard/src/xtd/test/sin/bin/sin_t_cc
===============================================================================
All tests passed (30 assertions in 1 test case)


Running /home/fwyzard/src/xtd/test/sin/bin/sin_t_hip
No AMD GPUs found, the test will be skipped.

Running /home/fwyzard/src/xtd/test/sin/bin/sin_t_cuda
===============================================================================
All tests passed (30 assertions in 1 test case)


Running /home/fwyzard/src/xtd/test/sin/bin/sin_t_cpusycl
===============================================================================
All tests passed (30 assertions in 1 test case)


Running /home/fwyzard/src/xtd/test/sin/bin/sin_t_gpusycl
Native API failed. Native API returns: -2 (PI_ERROR_DEVICE_NOT_AVAILABLE) -2 (PI_ERROR_DEVICE_NOT_AVAILABLE)Exception caught at file:sin/sin_t.sycl.cc, line:60

[AuroraPerego]

Thanks for the improvements :D
The tests pass on:

• NVIDIA L4, V100 and T4
• Intel GPU Flex

The hip test hangs on the AMD GPU Radeon PRO WX 9100 (don't know yet if it's due to the GPU or the test)