Commandlines

From Manta

Hi!

These are some commandlines used to benchmark pieces of manta:

Table of contents

Revision 1234

  • macbook, gcc 4.0 (RayPacket::MaxSize = 64, Fragment::MaxSize = 256)
bin/manta -np 2 -res 1024x1024 \
-scene "lib/libscene_iwviewer.dylib( -file /Users/boulos/models/iw_models/erw6.iw -flat )" \
-camera "pinhole( -eye -2.9 1.5 -2.2 -lookat 0 1 0 -up 0 1 0 -normalizeRays)" \
-imagetraverser "tiled(-tilesize 16x16 -square)" \
-ui null -imagedisplay null \
-bench 100 25 \
SSE: 8.84 fps


Revision 1175

  • macbook, gcc 4.0 (RayPacket::MaxSize = 16, Fragment::MaxSize = 16)
bin/manta -np 2 -scene "lib/libscene_iwviewer.dylib( -file /Users/boulos/models/iw_models/erw6.iw )" \
-camera "pinhole( -eye -2.9 1.5 -2.2 -lookat 0 1 0 -up 0 1 0)" -imagetraverser "tiled(-tilesize 16x16 -square)" \
-shadow null -bench 100 25
SSE: 24.1 fps


Revision 1170

  • macbook, gcc 4.0 (RayPacket::MaxSize = 16, Fragment::MaxSize = 16)
bin/manta -np 2 -scene "lib/libscene_iwviewer.dylib( -file /Users/boulos/models/iw_models/erw6.iw )" \
-camera "pinhole( -eye -2.9 1.5 -2.2 -lookat 0 1 0 -up 0 1 0)" -imagetraverser "tiled(-tilesize 16x16 -square)" \
-shadow null -bench 100 25
SSE: 20.5 fps

Revision 984

  • muzzle, gcc 3.3.5
bin/manta -scene "lib/libscene_perf.so( -Wald )" -imagetraverser "hard(-tilesize 64x64)" -ui null -imagedisplay null -imagetype rgbfloat -pixelsampler fast -bench 100 25
C: 22.34
SSE: 24.80
SSE+ConstantOrigin: 33.97
  • acheron, gcc 4.0
bin/manta -scene "lib/libscene_perf.dylib(-Wald)" -imagetraverser "hard(-tilesize 64x64)" -ui null -imagedisplay null -imagetype rgbfloat -pixelsampler fast -bench 100 25
C: 48.95
Altivec: 50.19
Altivec+ConstantOrigin: 62.63
  • g4 laptop, gcc 4.0
bin/manta -scene "lib/libscene_perf.dylib(-Wald)" -imagetraverser "hard(-tilesize 64x64)" -ui null -imagedisplay null -imagetype rgbfloat -pixelsampler fast -bench 100 25
C: 13.20 (Intersect is 40% profile)
Altivec: 15.90 (Intersect is 30% profile)
Altivec+ConstantOrigin: 18.08 (Intersect is 20% profile)

Note the above profiling reveals that around .030, .019, and .011 spf are spent in triangle intersection on the laptop respectively. The C version of the code had the correct ConstantOrigin test, so really the altivec speedup should be considered as going from .030 to .011 (almost 3x). Until I can get on muzzle and acheron with a profiler, I won't know the similar speedup factors for those machines.

Revision 966

  • muzzle, gcc 3.3.5 (DynRT flags) Fragment Size 256, Ray Packet Max Size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 1000 5
Benchmark completed in 3.59536 seconds (1000 frames, 278.136 frames per second)
  • muzzle, gcc 3.3.5 (DynRT flags) Fragment Size 4096, Ray Packet Max Size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 1000 5
Benchmark completed in 3.60532 seconds (1000 frames, 277.368 frames per second)


  • muzzle, icc 9.0.030 (DynRT flags) Fragment Size 256, Ray Packet Max Size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 1000 5
Benchmark completed in 5.07379 seconds (1000 frames, 197.091 frames per second)
  • muzzle, icc 9.0.030 (DynRT flags) Fragment Size 4096, Ray Packet Max Size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 1000 5
Benchmark completed in 5.05893 seconds (1000 frames, 197.67 frames per second)

Revision 965

  • muzzle, gcc 3.3.5 (-O3) Fragment size 256, ray packet max size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 10 5
Benchmark completed in 0.085992 seconds (10 frames, 116.29 frames per second)
  • muzzle, gcc 3.3.5 (-O3) Fragment size 4096, ray packet max size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 10 5
Benchmark completed in 0.974155 seconds (10 frames, 10.2653 frames per second)
  • muzzle, ICC 9.0.030 (DynRT flags) Fragment size 256, ray packet max size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 10 5
Benchmark completed in 0.460043 seconds (10 frames, 21.7371 frames per second)
  • muzzle, ICC 9.0.030 (DynRT flags) Fragment size 4096, ray packet max size 256, Real == float
bin/manta -res 1024x1024 -renderer raygen -imagetype null -ui null -imagedisplay null -pixelsampler null -imagetraverser "hard( -tilesize 64x64 )" -loadbalancer simple -bench 10 5
Benchmark completed in 6.86858 seconds (10 frames, 1.4559 frames per second)