glm/test/perf/perf_matrix_mul_vector.cpp
Laurent Caumont 458882e46c Simd improvement
- 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 ?)
2023-12-25 11:35:42 +01:00

160 lines
4.9 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>
template <typename matType, typename vecType>
static void test_mat_mul_vec(matType const& M, std::vector<vecType> const& I, std::vector<vecType>& O)
{
for (std::size_t i = 0, n = I.size(); i < n; ++i)
O[i] = M * I[i];
}
template <typename matType, typename vecType>
static int launch_mat_mul_vec(std::vector<vecType>& O, matType const& Transform, vecType const& Scale, std::size_t Samples)
{
typedef typename matType::value_type T;
std::vector<vecType> I(Samples);
O.resize(Samples);
for(std::size_t i = 0; i < Samples; ++i)
I[i] = Scale * static_cast<T>(i);
std::chrono::high_resolution_clock::time_point t1 = std::chrono::high_resolution_clock::now();
test_mat_mul_vec<matType, vecType>(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 packedVecType, typename alignedMatType, typename alignedVecType>
static int comp_mat2_mul_vec2(std::size_t Samples)
{
typedef typename packedMatType::value_type T;
int Error = 0;
packedMatType const Transform(1, 2, 3, 4);
packedVecType const Scale(0.01, 0.02);
std::vector<packedVecType> SISD;
std::printf("- SISD: %d us\n", launch_mat_mul_vec<packedMatType, packedVecType>(SISD, Transform, Scale, Samples));
std::vector<alignedVecType> SIMD;
std::printf("- SIMD: %d us\n", launch_mat_mul_vec<alignedMatType, alignedVecType>(SIMD, Transform, Scale, Samples));
for(std::size_t i = 0; i < Samples; ++i)
{
packedVecType const A = SISD[i];
packedVecType const B = packedVecType(SIMD[i]);
Error += glm::all(glm::equal(A, B, static_cast<T>(0.001))) ? 0 : 1;
}
return Error;
}
template <typename packedMatType, typename packedVecType, typename alignedMatType, typename alignedVecType>
static int comp_mat3_mul_vec3(std::size_t Samples)
{
typedef typename packedMatType::value_type T;
int Error = 0;
std::vector<packedVecType> SISD;
{
packedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9);
packedVecType const Scale(0.01, 0.02, 0.05);
std::printf("- SISD: %d us\n", launch_mat_mul_vec<packedMatType, packedVecType>(SISD, Transform, Scale, Samples));
}
std::vector<alignedVecType> SIMD;
{
alignedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9);
alignedVecType const Scale(0.01, 0.02, 0.05);
std::printf("- SIMD: %d us\n", launch_mat_mul_vec<alignedMatType, alignedVecType>(SIMD, Transform, Scale, Samples));
}
for(std::size_t i = 0; i < Samples; ++i)
{
packedVecType const A = SISD[i];
packedVecType const B = SIMD[i];
Error += glm::all(glm::equal(A, B, static_cast<T>(0.001))) ? 0 : 1;
}
return Error;
}
template <typename packedMatType, typename packedVecType, typename alignedMatType, typename alignedVecType>
static int comp_mat4_mul_vec4(std::size_t Samples)
{
typedef typename packedMatType::value_type T;
int Error = 0;
packedMatType const Transform(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
packedVecType const Scale(0.01, 0.02, 0.03, 0.05);
std::vector<packedVecType> SISD;
std::printf("- SISD: %d us\n", launch_mat_mul_vec<packedMatType, packedVecType>(SISD, Transform, Scale, Samples));
std::vector<alignedVecType> SIMD;
std::printf("- SIMD: %d us\n", launch_mat_mul_vec<alignedMatType, alignedVecType>(SIMD, Transform, Scale, Samples));
for(std::size_t i = 0; i < Samples; ++i)
{
packedVecType const A = SISD[i];
packedVecType const B = SIMD[i];
Error += glm::all(glm::equal(A, B, static_cast<T>(0.001))) ? 0 : 1;
}
return Error;
}
int main()
{
std::size_t const Samples = 1000;
int Error = 0;
std::printf("mat2 * vec2:\n");
Error += comp_mat2_mul_vec2<glm::mat2, glm::vec2, glm::aligned_mat2, glm::aligned_vec2>(Samples);
std::printf("dmat2 * dvec2:\n");
Error += comp_mat2_mul_vec2<glm::dmat2, glm::dvec2, glm::aligned_dmat2, glm::aligned_dvec2>(Samples);
std::printf("mat3 * vec3:\n");
Error += comp_mat3_mul_vec3<glm::mat3, glm::vec3, glm::aligned_mat3, glm::aligned_vec3>(Samples);
std::printf("dmat3 * dvec3:\n");
Error += comp_mat3_mul_vec3<glm::dmat3, glm::dvec3, glm::aligned_dmat3, glm::aligned_dvec3>(Samples);
std::printf("mat4 * vec4:\n");
Error += comp_mat4_mul_vec4<glm::mat4, glm::vec4, glm::aligned_mat4, glm::aligned_vec4>(Samples);
std::printf("dmat4 * dvec4:\n");
Error += comp_mat4_mul_vec4<glm::dmat4, glm::dvec4, glm::aligned_dmat4, glm::aligned_dvec4>(Samples);
return Error;
}
#else
int main()
{
return 0;
}
#endif