Regression in 7f2a5b89b30d143014bc0363b99dc3d942457ae7

[mgerhardy]

template<typename T, qualifier Q>
GLM_FUNC_QUALIFIER GLM_CONSTEXPR vec<3, T, Q> operator/(vec<3, T, Q> const& v, T scalar)
{
  return vec<3, T, Q>(v) *= 1/scalar;
}

7f2a5b8

this change breaks stuff like

glm::ivec3 center = _mins + _width / 2;

this basically means that center is now _mins because the division leads to *=0 for any scalar > 1

[dimitre]

I've noticed some template issues also when using the define
#define GLM_FORCE_NEON

[christophe-lunarg]

Ok, I'll reverse it. :/ @laurentcau

[laurentcau]

Strange to not fix the issue instead of revert. Never mind.
What is the template issues with GLM_FORCE_NEON ?

[laurentcau]

@dimitre

[remixer-dec]

Hello!

Just wanted to let you know that PR #1278 causes major performance degradation, I tested it on m1 mac.
I tested on this project: https://github.com/benikabocha/saba
I replaced the library in external/glm with the latest one and found that it performs significantly worse than previously released version (37 FPS -> 24 FPS). After rolling back the commits I found that #1278 is what is causing it.

All this with base config build with SIMD off.

Next, I decided to turn it on to see if it helps.

    add_compile_options(-march=armv8-a+simd -mtune=apple-m1)
    add_definitions(
        -DGLM_FORCE_NEON
        -D__ARM_ARCH=8
    )

After manually enabling it with I faced multiple errors like

saba/external/glm/include/glm/./ext/../detail/type_vec_simd.inl:942:30: error: template argument for non-type template parameter must be an expression
                        return !compute_vec_equal<int, Q, false, 32, true>::call(v1, v2);

type_vec_simd.inl meaning that manually enabling SIMD/Neon worked, but compiler was not compatible with this syntax
after dealing with it and with errors like

./saba/src/Saba/Model/MMD/MMDModel.cpp:145:20: error: no viable conversion from 'mat<4, 4, [2 * ...]>' to 'const mat<3, 3, [2 * ...]>'
                        const glm::mat3 invZ = glm::scale(glm::mat4(1.0f), glm::vec3(1, 1, -1));
                                        ^      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.saba/src/Saba/Model/MMD/PMXModel.cpp:508:18: error: no viable overloaded '='

mat.m_diffuse = pmxMat.m_diffuse;

~~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~

./saba/external/glm/include/glm/./ext/../detail/type_vec3.hpp:185:55: note: candidate function not viable: no known conversion from 'const vec<4, [2 * ...]>' to 'const vec<3, [2 * ...]>' for 1st argument

GLM_DEFAULTED_FUNC_DECL GLM_CONSTEXPR vec<3, T, Q>& operator=(vec<3, T, Q> const& v) GLM_DEFAULT;
./saba/src/Saba/Model/MMD/MMDModel.cpp:145:46: error: no matching conversion for functional-style cast from 'glm::vec3' (aka 'vec<3, float, defaultp>') to 'glm::mat3' (aka 'mat<3, 3, float, defaultp>')

const glm::mat3 invZ = glm::mat3(1.0f) * glm::mat3(glm::vec3(1, 1, -1));

I managed to successfully build the app. To my disappointment, the performance remained the same at 24FPS as with SIMD off, maybe I missed something, I'm not really a low-level developer, correct me if I'm wrong.

[dimitre]

I've just updated to latest commit in this runners here (building for multiple platforms)
and there are some template errors for some platforms.
using commit 18feaec
https://github.com/dimitre/ofLibs/actions/runs/12768319570/job/35588448387
using release tag 1.0.2
https://github.com/dimitre/ofLibs/actions/runs/12735527744

and here the CMake parameters for each platform
https://github.com/dimitre/ofLibs/blob/main/glm/chalet.yaml