mirror of
https://github.com/g-truc/glm.git
synced 2024-11-22 17:04:35 +00:00
4137519418
- Add simd aligned_vec3 (and sse aligned_dvec3 - 2 x xmm) - Fast packed_vec3 <=> aligned_vec3 and packed_vec4 <=> aligned_vec4 conversion - Fast aligned_vec3 <=> aligned_vec4 conversion - Optimized aligned_mat x aligned_mat and aligned_mat x aligned_vec - Inverse aligned_mat3 simd version (actually slower than ssid on my computer even it has 30% less instruction ?)
203 lines
5.5 KiB
C++
203 lines
5.5 KiB
C++
#define GLM_FORCE_INLINE
|
|
#include <glm/ext/matrix_float2x2.hpp>
|
|
#include <glm/ext/matrix_double2x2.hpp>
|
|
#include <glm/ext/matrix_float3x3.hpp>
|
|
#include <glm/ext/matrix_double3x3.hpp>
|
|
#include <glm/ext/matrix_float4x4.hpp>
|
|
#include <glm/ext/matrix_double4x4.hpp>
|
|
#include <glm/ext/matrix_transform.hpp>
|
|
#include <glm/ext/matrix_relational.hpp>
|
|
#include <glm/ext/vector_float4.hpp>
|
|
#if GLM_CONFIG_SIMD == GLM_ENABLE
|
|
#include <glm/gtc/type_aligned.hpp>
|
|
#include <vector>
|
|
#include <chrono>
|
|
#include <cstdio>
|
|
|
|
|
|
inline bool
|
|
is_aligned(const void* ptr, std::uintptr_t alignment) noexcept {
|
|
auto iptr = reinterpret_cast<std::uintptr_t>(ptr);
|
|
return !(iptr % alignment);
|
|
}
|
|
|
|
template <typename matType>
|
|
static void align_check(matType const& M, std::vector<matType> const& I, std::vector<matType>& O)
|
|
{
|
|
if (matType::col_type::is_aligned::value)
|
|
{
|
|
if (!is_aligned(&M, 16))
|
|
abort();
|
|
for (std::size_t i = 0, n = I.size(); i < n; ++i)
|
|
{
|
|
if (!is_aligned(&I[i], 16))
|
|
abort();
|
|
|
|
if (!is_aligned(&O[i], 16))
|
|
abort();
|
|
}
|
|
}
|
|
}
|
|
|
|
template <typename matType>
|
|
static void test_mat_mul_mat(matType const& M, std::vector<matType> const& I, std::vector<matType>& O)
|
|
{
|
|
for (std::size_t i = 0, n = I.size(); i < n; ++i)
|
|
O[i] = M * I[i];
|
|
}
|
|
|
|
template <typename matType>
|
|
static int launch_mat_mul_mat(std::vector<matType>& O, matType const& Transform, matType const& Scale, std::size_t Samples)
|
|
{
|
|
typedef typename matType::value_type T;
|
|
|
|
std::vector<matType> I(Samples);
|
|
O.resize(Samples);
|
|
|
|
for(std::size_t i = 0; i < Samples; ++i)
|
|
I[i] = Scale * static_cast<T>(i);
|
|
|
|
align_check<matType>(Transform, I, O);
|
|
|
|
std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
|
|
test_mat_mul_mat<matType>(Transform, I, O);
|
|
std::chrono::high_resolution_clock::time_point t2 = std::chrono::high_resolution_clock::now();
|
|
|
|
return static_cast<int>(std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count());
|
|
}
|
|
|
|
template <typename packedMatType, typename alignedMatType>
|
|
static int comp_mat2_mul_mat2(std::size_t Samples)
|
|
{
|
|
typedef typename packedMatType::value_type T;
|
|
|
|
int Error = 0;
|
|
|
|
packedMatType const Transform(1, 2, 3, 4);
|
|
packedMatType const Scale(0.01, 0.02, 0.03, 0.05);
|
|
|
|
std::vector<packedMatType> SISD;
|
|
std::printf("- SISD: %d us\n", launch_mat_mul_mat<packedMatType>(SISD, Transform, Scale, Samples));
|
|
|
|
std::vector<alignedMatType> SIMD;
|
|
std::printf("- SIMD: %d us\n", launch_mat_mul_mat<alignedMatType>(SIMD, Transform, Scale, Samples));
|
|
|
|
for(std::size_t i = 0; i < Samples; ++i)
|
|
{
|
|
packedMatType const A = SISD[i];
|
|
packedMatType const B = SIMD[i];
|
|
Error += glm::all(glm::equal(A, B, static_cast<T>(0.001))) ? 0 : 1;
|
|
}
|
|
|
|
return Error;
|
|
}
|
|
|
|
template<typename T1, typename T2>
|
|
bool percent_error(const T1& a, const T2& b, float percentThreshold)
|
|
{
|
|
typedef typename T1::value_type value_type;
|
|
for (int i = 0; i < a.length(); ++i)
|
|
for (int j = 0; j < a[i].length(); ++j)
|
|
{
|
|
value_type v;
|
|
if (a[i][j] != value_type(0))
|
|
v = ((b[i][j] - a[i][j]) / a[i][j]) * value_type(100);
|
|
else
|
|
v = b[i][j] * value_type(100);
|
|
|
|
if (v > value_type(percentThreshold))
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
template <typename packedMatType, typename alignedMatType>
|
|
static int comp_mat3_mul_mat3(std::size_t Samples)
|
|
{
|
|
|
|
int Error = 0;
|
|
|
|
std::vector<packedMatType> SISD;
|
|
{
|
|
packedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9);
|
|
packedMatType const Scale(0.01, 0.02, 0.03, 0.05, 0.01, 0.02, 0.03, 0.05, 0.01);
|
|
std::printf("- SISD: %d us\n", launch_mat_mul_mat<packedMatType>(SISD, Transform, Scale, Samples));
|
|
}
|
|
|
|
std::vector<alignedMatType> SIMD;
|
|
{
|
|
alignedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9);
|
|
alignedMatType const Scale(0.01, 0.02, 0.03, 0.05, 0.01, 0.02, 0.03, 0.05, 0.01);
|
|
std::printf("- SIMD: %d us\n", launch_mat_mul_mat<alignedMatType>(SIMD, Transform, Scale, Samples));
|
|
}
|
|
for(std::size_t i = 0; i < Samples; ++i)
|
|
{
|
|
packedMatType const A = SISD[i];
|
|
packedMatType const B = SIMD[i];
|
|
Error += percent_error(A, B, 0.01f) ? 0 : 1;
|
|
}
|
|
|
|
return Error;
|
|
}
|
|
|
|
template <typename packedMatType, typename alignedMatType>
|
|
static int comp_mat4_mul_mat4(std::size_t Samples)
|
|
{
|
|
|
|
int Error = 0;
|
|
|
|
packedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
|
|
packedMatType const Scale(0.01, 0.02, 0.03, 0.05, 0.01, 0.02, 0.03, 0.05, 0.01, 0.02, 0.03, 0.05, 0.01, 0.02, 0.03, 0.05);
|
|
|
|
std::vector<packedMatType> SISD;
|
|
std::printf("- SISD: %d us\n", launch_mat_mul_mat<packedMatType>(SISD, Transform, Scale, Samples));
|
|
|
|
std::vector<alignedMatType> SIMD;
|
|
std::printf("- SIMD: %d us\n", launch_mat_mul_mat<alignedMatType>(SIMD, Transform, Scale, Samples));
|
|
|
|
for(std::size_t i = 0; i < Samples; ++i)
|
|
{
|
|
packedMatType const A = SISD[i];
|
|
packedMatType const B = SIMD[i];
|
|
Error += percent_error(A, B, 0.01f) ? 0 : 1;
|
|
}
|
|
|
|
return Error;
|
|
}
|
|
|
|
int main()
|
|
{
|
|
std::size_t const Samples = 1000;
|
|
|
|
int Error = 0;
|
|
|
|
std::printf("mat2 * mat2:\n");
|
|
Error += comp_mat2_mul_mat2<glm::mat2, glm::aligned_mat2>(Samples);
|
|
|
|
std::printf("dmat2 * dmat2:\n");
|
|
Error += comp_mat2_mul_mat2<glm::dmat2, glm::aligned_dmat2>(Samples);
|
|
|
|
std::printf("mat3 * mat3:\n");
|
|
Error += comp_mat3_mul_mat3<glm::mat3, glm::aligned_mat3>(Samples);
|
|
|
|
std::printf("dmat3 * dmat3:\n");
|
|
Error += comp_mat3_mul_mat3<glm::dmat3, glm::aligned_dmat3>(Samples);
|
|
|
|
std::printf("mat4 * mat4:\n");
|
|
Error += comp_mat4_mul_mat4<glm::mat4, glm::aligned_mat4>(Samples);
|
|
|
|
std::printf("dmat4 * dmat4:\n");
|
|
Error += comp_mat4_mul_mat4<glm::dmat4, glm::aligned_dmat4>(Samples);
|
|
|
|
return Error;
|
|
}
|
|
|
|
#else
|
|
|
|
int main()
|
|
{
|
|
return 0;
|
|
}
|
|
|
|
#endif
|