As mentioned in WASM performance, some big steps have been made for the performance of the WASM version. These also improved the performance using GCC-15 on x86_64. There is more good news: using -O3 for GCC now boosts performance another 15%. In older tests -O2 produced faster and smaller binaries. The standard benchmark is now 25% faster (Fedora 42 on AMD3950X) than 2 days ago 
That makes the gap with Clang really big. It turns out there was a mistake in the configuration why PGO optimization did not work. But, still Clang is not great. The best result comparing for Clang-20 is 25% slower than GCC-15 on Fedora on AMD3950X. For AppleClang (17) -O2 vs -O3 with or without PGO makes little difference. Still, SWI-Prolog performs about the same on M1 than on my AMD3950X desktop on single threaded workloads.