pmartz@undies.dsd.es.com (Paul Martz) (04/17/91)
I have been developing some code on a Personal Iris GT which draws primarily vectors. Using the v3f function, I was getting about 135K vecs/sec, so I tried move/draw instead and started getting about 165K vecs/sec. From this I deduced that move/draw is optimized for the Personal Iris. Then I sent the compiled executable to a friend who has a VGX. He wrote back that vector performance was the worst he'd ever seen: 155K vecs/sec! (Still using move/draw.) So I have two questions: 1) Is it normal to see move/draw perform better on a Personal Iris GT than a VGX? I really should be using v3f for VGX machines, right? 2) Is the libgl_s.a library shipped with a Personal Iris different from what's shipped with a VGX? I.e., could I have my friend recompile/relink on the VGX and get different results? -- -paul pmartz@dsd.es.com Evans & Sutherland
kurt@cashew.asd.sgi.com (Kurt Akeley) (04/22/91)
In article <1991Apr17.144853.17441@dsd.es.com>, pmartz@undies.dsd.es.com (Paul Martz) writes: |> I have been developing some code on a Personal Iris GT which draws |> primarily vectors. Using the v3f function, I was getting about 135K |> vecs/sec, so I tried move/draw instead and started getting about 165K |> vecs/sec. From this I deduced that move/draw is optimized for the |> Personal Iris. |> |> Then I sent the compiled executable to a friend who has a VGX. He |> wrote back that vector performance was the worst he'd ever seen: 155K |> vecs/sec! (Still using move/draw.) |> |> So I have two questions: |> |> 1) Is it normal to see move/draw perform better on a Personal Iris GT |> than a VGX? I really should be using v3f for VGX machines, right? Yes, you really should use v3f on GT, GTX, and VGX machines. The VGX in particular cannot parallelize move/draw syntax commands, and therefore performs very poorly on them. The Personal Iris is unfortunately optimized for the soon-to-be-obsolete move/draw interface. |> 2) Is the libgl_s.a library shipped with a Personal Iris different |> from what's shipped with a VGX? I.e., could I have my friend |> recompile/relink on the VGX and get different results? Mixing libraries will not help, the slowdown is fundamental to the internals of the VGX. |> -- |> |> -paul pmartz@dsd.es.com |> Evans & Sutherland v3f calls are the future of the Iris Graphics Library. If you choose to take advantage of the performance advantage of move/draw calls on current Personal Iris equipment, you should consider writing your code to use v3f on all other platforms. -- Kurt
zombie@voodoo.UUCP (Mike York) (04/24/91)
In article <1991Apr22.163320.8755@odin.corp.sgi.com> kurt@cashew.asd.sgi.com (Kurt Akeley) writes: >In article <1991Apr17.144853.17441@dsd.es.com>, pmartz@undies.dsd.es.com (Paul Martz) writes: >|> 1) Is it normal to see move/draw perform better on a Personal Iris GT >|> than a VGX? I really should be using v3f for VGX machines, right? > >Yes, you really should use v3f on GT, GTX, and VGX machines. The VGX in >particular cannot parallelize move/draw syntax commands, and therefore >performs very poorly on them. The Personal Iris is unfortunately optimized >for the soon-to-be-obsolete move/draw interface. Does this mean that our PI's are soon-to-be-obsolete? >v3f calls are the future of the Iris Graphics Library. If you choose to >take advantage of the performance advantage of move/draw calls on current >Personal Iris equipment, you should consider writing your code to use >v3f on all other platforms. Here's an excerpt from the man page for move on 4D/35 we've got here: NOTE move should not be used in new development. Rather, lines should be drawn using the high-performance v commands, surrounded by calls to bgnline and endline. We've been following this advice. All of our production machines are currently 4D/25TG's (considering 4D/35 upgrades). Should we change back to move/draw commands, or is there hope for "high-performance v commands" to be high performance on the PI's? -- Mike York | Boeing Computer Services | Support your local nanobrewer -- (206) 865-6577 | No twist-off bottle caps. zombie@voodoo.boeing.com |