Michael Abrash doubled Quake framerste
fabiensanglard.net
Michael Abrash doubled Quake framerste
1–10 of 10 posts
Re: Michael Abrash doubled Quake framerste
#2Re: Michael Abrash doubled Quake framerste
#3Just curious, given that Pentium arch might still be relevant, than 486/487. How much of this optimization is still relevant today?
Re: Michael Abrash doubled Quake framerste
#4Just curious, given that Pentium arch might still be relevant, than 486/487. How much of this optimization is still relevant today?
Thinking about fpu pipeline parallelism/stalls still matter, though more often by hinting an optimizing compiler + verifying the output by inspection, than by inlining asm itself.
Re: Michael Abrash doubled Quake framerste
#5Re: Michael Abrash doubled Quake framerste
#6Re: Michael Abrash doubled Quake framerste
#7By going to godbolt Just realized how hard it is to generate x87 FPU code these days. Almost all supported compilers are outputing XMM? SSE instructions these days.
Re: Michael Abrash doubled Quake framerste
#8Title has a typo, should say “framerate”.
Re: Michael Abrash doubled Quake framerste
#9By going to godbolt Just realized how hard it is to generate x87 FPU code these days. Almost all supported compilers are outputing XMM? SSE instructions these days.
GCC: -mfpmath=387
Clang: -mfpmath=387 -mno-sse
Re: Michael Abrash doubled Quake framerste
#10Just curious, given that Pentium arch might still be relevant, than 486/487. How much of this optimization is still relevant today?
There are actually several flavors of x87 ISA: 8087, 80287, and 80387. Every 386+ chip supports the 387 ISA. The 287 is almost the same as the 387. The 8087 is almost a strict subset with some differences. There were also Weitek math coprocessors that were entirely different. The IIT 3C87 was an odd bird that supported the 387 ISA and some other instructions.