How do you see Intel’s complete adoption of LLVM for its C/C++/Fortran compilers

see blog: Intel C/C++ compilers complete adoption of LLVM

We will only have ifx based on LLVM, no more ifort classic?


Here is the corresponding Twitter announcement:

Here is the current implementation status of Fortran standard for ifx. (ifort classic Fortran compiler has full support of F18)

1 Like

It sounds like they’re going to maintain ifort at least until there is feature parity. Personally, though, I think the move to LLVM is probably a good one, and I am looking forward to switching to ifx once Coarray support is added.

Does anyone know how they achieve better performance via LLVM in their C++ compilers than Clang can do for floating point (see the second graph in the blog post with title “Relative Floating Point Rate Performance (est.)”, when you click on it, it goes to another page with more details)?

Do they have some proprietary patches on top of LLVM implementing some low level optimization passes, or does their compiler simply generate better LLVM IR code, but uses stock LLVM?

I think we’re close to the final loop of the “Race To Exacsale”: The shift to LLVM-based compilers is certainly also to support others to build EDGE-compilers in the (near) future ( and thus, to support compilation for the upcoming Intel CSAccelerator: Intel’s Exascale Dataflow Engine Drops X86 And Von Neumann . (Searching in this article for ‘LLVM’ will bring you to the relevant paragraph).

1 Like