flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb invpcid_single pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp md_clear flush_l1dSee AMD's SSE5 vs Intel's SSE4 conflict in x86 - both have instructions that the other doesn't have, and some that both have, and, from what I understand, even the instruction formats conflict.
One cannot easily make an x86 extension because there are approximately ~3 entities on the planet allowed to make x86 extensions.
I wouldn't be surprised if Sony/Microsoft did too for their consoles.
It's doable if you have enough money to get AMD or VIA to make you a one-off.
There is the idea of profiles, i.e. a single nice name for a group of extensions so that there's something one could provide a precompiled binary for, but I don't know what, if any, are RISC-V's plans for that.
Profiles. See RVA22.
It'll likely work much the same way for RISC-V, eg. in Fedora we require that the C (compressed) extension is supported, otherwise the distro won't boot, and things like the vector extension will probably be used only by specialized math libraries and the like.
It's a problem that already exists right now.
If you compile code on a machine with SSE or AVX and run it on an old Pentium, does it crash?
I had a fun bug once running ARM code that had some NEON in it and it would crashed on Scaleway's VPS bc their servers ran chips that didn't have NEON implemented
This doesn’t make sense, though. If you have something in your instruction stream that you don’t understand, it would be insane to do anything but fault immediately, unless the instruction is some sort of optional feature such as a hint or security mechanism where there is no real behavior associated with it. I mean ARM has this “capability” too by virtue of a fixed length instruction set but it doesn’t do anything with this like you’re describing.
Yes. Or it leaves them unused on the machine with avx.
For short runs of instructions, the overhead of fallback selection is larger than the overhead of using the least common denominator. For large functions, fallback selection is done by selecting specific functions after benchmarking.
Things like OpenJDK, pixman, nss, ffmpeg.
Edit: I am mistaken, thanks everyone for educating me.