Entry tags:

benchmark, c, c++, d language, java, ldc, raytracing

I have improved the code of the Sphereflake raytracer and thanks to Tomas Lindquist Olsen I have fixed my LDC installation, so there I have redone the benchmarks on Pubuntu:

Timings on Pubuntu, w=h=1024, lvl=6, best of 3, seconds:
(66_430 spheres, WITH_SHADOWS=true, FASTER_LDC=true)
  D:    4.33  (claiming 4.6 MB) (first fast version + LTO+I)
  D:    4.51  (claiming 2.5 MB) (second fast version + LTO+I)
  D:    4.55  (claiming 4.6 MB) (basic version + LTO+I)
  D:    4.70  (claiming 2.5 MB) (second fast version)
  D:    4.72  (claiming 4.6 MB) (first fast version)
  D:    4.97  (claiming 4.6 MB) (basic version)
  C++:  5.04  (claiming 4.6 MB) (+4 bytes padding)
  C++:  5.75  (claiming 4.3 MB) (original version)

Timings on Pubuntu, w=h=1024, lvl=7, best of 3, seconds:
(597_871 spheres, WITH_SHADOWS=true, FASTER_LDC=true)
  D:    5.51  (claiming 23 MB) (second fast version + LTO+I)
  D:    5.74  (claiming 41 MB) (first fast version + LTO+I)

Timings on Pubuntu, w=h=1024, lvl=8, best of 3, seconds:
  D:   10.84  (claiming 205 MB) (second fast version + LTO+I)
  D:   11.21  (claiming 205 MB) (second fast version)
  D:   23.34  (claiming 369 MB) (first fast version + LTO+I)
  D:   23.84  (claiming 369 MB) (basic version + LTO+I)
  D:   24.06  (claiming 369 MB) (first fast version)
  D:   24.55  (claiming 369 MB) (basic version)
  C++: 24.61  (claiming 369 MB) (+4 bytes padding)

For more information, and for the updated Sphereflake code:
http://www.fantascienza.net/leonardo/js/index.html#flake

----------------

I have added a new benchmark, a sparse matrix multiplication, this time the performance of D code (with LDC) isn't much good. Info and timings inside the zip:
http://www.fantascienza.net/leonardo/js/index.html#jaspa