August 10, 2020 How to use LLVM intrinsics on a UEFI x64 target? | ||||
|---|---|---|---|---|
| ||||
I am currently writing an UEFI application and ran into the problem that using the llvm_cos and llvm_sin intrinsics results in an undefined symbol during linking as they are just replaced with call to C's cosf and sinf.
Reduced test:
import ldc.intrinsics;
extern(C) export __gshared int _fltused = 0;
extern(Windows) void efi_main() {
float a = llvm_cos(0.5f);
double b = llvm_sin(0.5);
}
compiled with
-mtriple=x86_64-windows-coff -defaultlib= -debuglib= -code-model=large -betterC
-L/entry:efi_main -L/dll -L/subsystem:EFI_APPLICATION -L/nodefaultlib
results in
lld-link: error: undefined symbol: cosf
referenced by D_Test\source\app.d:5
.dub\build\application-debug-windows-x86_64-ldc_2091-8D06C0136AE698EE35D2211A74530569\d-test.obj:(efi_main)
lld-link: error: undefined symbol: sin
referenced by D_Test\source\app.d:6
.dub\build\application-debug-windows-x86_64-ldc_2091-8D06C0136AE698EE35D2211A74530569\d-test.obj:(efi_main)
Error: linking with LLD failed
ldc2 failed with exit code 1.
How can i tell LDC that my target does not have a C standard library? I tried to use the target triple x86_64-none-coff but that is not accepted by LLVM:
LLVM ERROR: Cannot initialize MC for non-Windows COFF object files.
ldc2 failed with exit code 1.
Interestingly sqrt works fine.
| ||||
August 10, 2020 Re: How to use LLVM intrinsics on a UEFI x64 target? | ||||
|---|---|---|---|---|
| ||||
Posted in reply to KytoDragon | On Monday, 10 August 2020 at 08:30:52 UTC, KytoDragon wrote: > How can i tell LDC that my target does not have a C standard library? I don't think there's a way, plus not much LLVM can do - relying on an external library implementation for non-trivial transcendental math functions seems to make sense. Unfortunately, LLVM also doesn't seem to provide these functions in the 'builtins' compiler-rt library (ldc_rt.builtins.lib). Phobos' std.math has some implementations, but unfortunately none for cos/sin yet. > Interestingly sqrt works fine. Yes, because there's a x64/SSE instruction for it. | |||
Copyright © 1999-2021 by the D Language Foundation
Permalink
Reply