Don’t expect high floating point performance from the Delphi for Linux compiler
Posted by jpluimers on 2018/07/31
Don’t expect high floating point performance of the Delphi for Linux compiler as this thread proves it is not: [WayBack] Based on a recent post here, it looks like the new Linux compiler uses the x87 unit for floating point. Naturally this won’t perform well. If it is true… – David Heffernan – Google+
–jeroen
KMorwath said
Maybe because otherwise they should provide a decent math library for functions the FPU provides but SIMD doesn’t? Or did they resurrected the old Kylix compiler?
Thaddy Koning said
Isn’t this a kind of “Hey, we write compilers”, no you don’t, you write llvm front-ends,”But look, we wrote a compiler?” Yes, good first time effort. Your grand pa did better… You still use his code, do you?
sglienke said
If you write the llvm frontend properly and enable all the optimizations for the backend you get amazing performance.
People should stop blaming llvm for bad delphi compiler codegen. clang is the proof that it can emit well optimized code. Go check for yourself on https://godbolt.org/
rvelthuis said
@Thaddy Koning: Sure, as Stefan said, the intermediate code generation seems to be deficient for now. But it is silly to say they can‘t write compilers. They already wrote quite a number of them, for Win32, Win64, .NET, etc.
thaddy said
I am not bashing LLVM, I use it myself for several projects (with Clang indeed, but also experimal setups, like a Pascal2js transpiler toolchain to llvm and FPC Java bytecode to llvm toolchain:fun but not productive yet).
And Rudy: the boys and girls that wrote the original compilers are mostly -of not all – gone. I see some positives lately, though. I have a bit more nuance than you think.