Dan Luu was born feeling adventurous ;-) But seriously, if you're writing x86 assembly inside of LLVM IR, why not just write x86 inside of C and call the C using ccall? Seems easier to me.
> On Dec 26, 2013, at 12:34 AM, Isaiah Norton <[email protected]> wrote: > > There is no way to do this right now. The most pragmatic thing would be to > ignore the rest of my email and do one of the other suggestions :) > > But if you are feeling adventurous you might be able to cobble it together > starting from: > > - https://github.com/JuliaLang/julia/pull/5046 (inline IR support for Julia) > - Another PR adding MCJIT support. AFAIK the legacy JIT doesn't support > inlining. > > The final issue would be that rdpmc and related instructions are not > supported by LLVM (at least as of Sep., per the todo-list; haven't checked > further - there is a chance the list is out of sync with the current > codebase). So you would need to write inline assembly, inside of inline IR, > inside of a Julia function. Have fun! > > >> On Tue, Dec 24, 2013 at 4:19 PM, Dan Luu <[email protected]> wrote: >> Is there way to write inline assembly or otherwise emit the rdpmc >> instruction? >> >> There are some simple TLB/cache performance experiments I want to run. >> Julia seems as good as C for this, except that I want to be able to >> look at perf counters and probably the TSC, too, and I can't find >> documentation on how to do this. >> >> >> Thanks, >> Dan >
