Initial investigation shows that the RayTracer with float4 has only 20% of the performance of the RayTracer with float (MacBook Pro, MacOS 10.6, Flash Player Release build from fr-float, 32-bit and 64-bit appear to behave the same though 32-bit is canonical, remember to force 32-bit-only if testing that as the default is 64-bit). This is a tracker bug for that problem.
finddef work blocks resolution of this bug (details to be provided).
Depends on: 705756
MacBook Pro ca 2008, 2.6 GHz Core 2 Duo, 4GB RAM, Release Standalone 32-bit, code synced ca 10:30 CET 8 December. RayTracer-float-V: 38fps RayTracer-float4-V: 34fps When I unroll the float4.dot3 calls I get 42fps - a significant improvement over the pure float code, yay. And the difference (42/34=1.24) is all in the finddef overhead, which it is now critical that we fix. (In 64-bit the difference is much greater, with the float code reaching 39fps and the float4 code petering out at about 26fps, but here we don't yet have inlining of float4 primitives so that's understandable.)
A hack proposed in bug #705756 works around the finddef problem somewhat effectively, getting us up to 42 fps.
changeset: 7052:25acba717e3c tag: tip user: Lars T Hansen <firstname.lastname@example.org> date: Thu Dec 08 16:18:50 2011 +0100 summary: Fix 708204 - RayTracer-float4-V slower than RayTracer-float-V
Status: NEW → RESOLVED
Last Resolved: 7 years ago
Resolution: --- → FIXED
Also in fr-float: CL #1011234.
You need to log in before you can comment on or make changes to this bug.