Simd improvement - Githubissues

g-truc / glm

OpenGL Mathematics (GLM)

https://glm.g-truc.net

Other

9.05k stars 2.1k forks source link

Simd improvement #1278

Closed laurentcau closed 5 months ago

laurentcau commented 5 months ago

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 ?)

laurentcau commented 5 months ago

Note: I fixed the integer div issue reported there: https://github.com/g-truc/glm/issues/1255 I also added a test to check there will be no regression in the future. For the template issues with GLM_FORCE_NEON, since there is no detail, I can't fix it. All tests are compiling fine on my side with clang + neon. So that should be something not covered by tests but what ? @dimitre

laurentcau commented 5 months ago

Tests passed https://github.com/laurentcau/glm/actions/runs/8326367556 @christophe-lunarg

christophe-lunarg commented 5 months ago

Considering you didn't receive any answer from the ARM NEON issue after multiple query, I think it's fair to break it and have been for who that matters submit a PR.

christophe-lunarg commented 5 months ago

Thanks for contributing!