5 ms·

> As an example, in my desktop, it takes ~25min to compile a Debug version of LLVM, but ~23min to compile a Release version. oh I think I know what might cause

> As an example, in my desktop, it takes ~25min to compile a Debug version of LLVM, but ~23min to compile a Release version. oh I think I know what might cause this: TableGen. The `llvm-tblgen` run time accounts for a good chunk of LLVM build time. In a debug build `llvm-tblgen` is also unoptimized, hence the long run time generating those .inc / .def files. You can enable cmake variable `LLVM_OPTIMIZED_TABLEGEN` to build `llvm-tblgen` in release mode while leaving the rest of the project in debug build (or whatever `CMAKE_BUILD_TYPE` you choose).

baziotis 2y ago
TIL that LLVM_OPTIMIZED_TABLEGEN exists, thanks! Unfortunately, it didn't make the build any faster on my machine.
yshui 2y ago
i'd say it's mostly due to linking. the debug info generated is absolutely enormous.
- vient 2y ago
  LLVM_USE_SPLIT_DWARF may help with this, some recent measurements: https://www.tweag.io/blog/2023-11-23-debug-fission/#an-example-at-scale-building-llvm https://www.tweag.io/blog/2023-11-23-debug-fission/#an-examp...
delta_p_delta_x 2y ago
The biggest factor in improving LLVM compile/link times is sheer RAM capacity. LLVM compilation is extremely voracious with memory and the more you have the faster it is; otherwise, much time is spent swapping in and out of disk. I've been compiling it with MSVC on Windows and the swap file after an LLVM compile balloons to ~60 GB. I inserted a couple more sticks to bring my 32 GB to 64 GB and my compile times shortened significantly (I don't have exact stopwatch times right now).
- almostgotcaught 2y ago
  Use clang-cl instead - way better RAM usage profile and way faster too - I'm talking like 50% reduction.
almostgotcaught 2y ago
> good chunk of LLVM build time Nah no way it's more than 5%. 5% is definitely a chunk but not big enough that if you made it completely vanish I'd notice. I rebuild LLVM with no cache many times a day (debug, release, all targets, no targets, etc). Debug build takes longer because of all the symbols - LLVM has some enormous symbols due to templating hijinks. Protips for speeding up building LLVM: 1. LLVM_TARGETS_TO_BUILD=host 2. LLVM_ENABLE_PROJECTS only what you're interested in (do you really care about lld or flang whatever?) 3. Use clang/lld instead of gcc - way better RAM usage 4. Use ccache 4. Disable tests

baziotis 2y ago

TIL that LLVM_OPTIMIZED_TABLEGEN exists, thanks! Unfortunately, it didn't make the build any faster on my machine.

yshui 2y ago

i'd say it's mostly due to linking. the debug info generated is absolutely enormous.

vient 2y ago
LLVM_USE_SPLIT_DWARF may help with this, some recent measurements: https://www.tweag.io/blog/2023-11-23-debug-fission/#an-example-at-scale-building-llvm https://www.tweag.io/blog/2023-11-23-debug-fission/#an-examp...

delta_p_delta_x 2y ago

The biggest factor in improving LLVM compile/link times is sheer RAM capacity. LLVM compilation is extremely voracious with memory and the more you have the faster it is; otherwise, much time is spent swapping in and out of disk. I've been compiling it with MSVC on Windows and the swap file after an LLVM compile balloons to ~60 GB. I inserted a couple more sticks to bring my 32 GB to 64 GB and my compile times shortened significantly (I don't have exact stopwatch times right now).

almostgotcaught 2y ago
Use clang-cl instead - way better RAM usage profile and way faster too - I'm talking like 50% reduction.

almostgotcaught 2y ago

> good chunk of LLVM build time Nah no way it's more than 5%. 5% is definitely a chunk but not big enough that if you made it completely vanish I'd notice. I rebuild LLVM with no cache many times a day (debug, release, all targets, no targets, etc). Debug build takes longer because of all the symbols - LLVM has some enormous symbols due to templating hijinks. Protips for speeding up building LLVM: 1. LLVM_TARGETS_TO_BUILD=host 2. LLVM_ENABLE_PROJECTS only what you're interested in (do you really care about lld or flang whatever?) 3. Use clang/lld instead of gcc - way better RAM usage 4. Use ccache 4. Disable tests