Commit 78be0d0, linux x86_64, g++ 12.3.0. I observe that the generated file wasm-to-js-tq-csa.cc takes > 20 minutes to compile on an otherwise fast system.
It's a fairly big file (391k, just under 7kLoC) but not that big (template heavy, however.) Other, much bigger torque files compile in seconds flat.
Possibly hits a pathological case in gcc. I wasn't really able to draw any conclusions from looking at perf top except that it was working hard. Perhaps tweaking compiler flags helps.
bnoordh+ 119949 98.4 7.0 1229736 1142472 pts/1 R+ 11:04 14:19 /usr/lib/gcc/x86_64-linux-gnu/12/cc1plus -quiet -I ../deps/v8 -I ../deps/v8/include -I /home/bnoordhuis/src/node/out/Release/obj/gen -I /home/bnoordhuis/src/node/out/Release/obj/gen/generate-bytecode-output-root -I ../deps/icu-small/source/i18n -I ../deps/icu-small/source/common -imultiarch x86_64-linux-gnu -MMD /home/bnoordhuis/src/node/out/Release/obj.target/v8_initializers/gen/torque-generated/src/builtins/wasm-to-js-tq-csa.d -MF /home/bnoordhuis/src/node/out/Release/.deps//home/bnoordhuis/src/node/out/Release/obj.target/v8_initializers/gen/torque-generated/src/builtins/wasm-to-js-tq-csa.o.d.raw -MQ /home/bnoordhuis/src/node/out/Release/obj.target/v8_initializers/gen/torque-generated/src/builtins/wasm-to-js-tq-csa.o -D_GNU_SOURCE -D_REENTRANT -D _GLIBCXX_USE_CXX11_ABI=1 -D NODE_OPENSSL_CONF_NAME=nodejs_conf -D NODE_OPENSSL_HAS_QUIC -D ICU_NO_USER_DATA_OVERRIDE -D V8_GYP_BUILD -D V8_TYPED_ARRAY_MAX_SIZE_IN_HEAP=64 -D __STDC_FORMAT_MACROS -D OPENSSL_NO_PINSHARED -D OPENSSL_THREADS -D V8_TARGET_ARCH_X64 -D V8_HAVE_TARGET_OS -D V8_TARGET_OS_LINUX -D V8_EMBEDDER_STRING="-node.13" -D ENABLE_DISASSEMBLER -D V8_PROMISE_INTERNAL_FIELD_COUNT=1 -D V8_ENABLE_PRIVATE_MAPPING_FORK_OPTIMIZATION -D V8_SHORT_BUILTIN_CALLS -D OBJECT_PRINT -D V8_INTL_SUPPORT -D V8_ATOMIC_OBJECT_FIELD_WRITES -D V8_ENABLE_LAZY_SOURCE_POSITIONS -D V8_USE_SIPHASH -D V8_SHARED_RO_HEAP -D V8_WIN64_UNWINDING_INFO -D V8_ENABLE_REGEXP_INTERPRETER_THREADED_DISPATCH -D V8_USE_ZLIB -D V8_ENABLE_MAGLEV -D V8_ENABLE_TURBOFAN -D V8_ENABLE_WEBASSEMBLY -D V8_ENABLE_JAVASCRIPT_PROMISE_HOOKS -D V8_ALLOCATION_FOLDING -D V8_ALLOCATION_SITE_TRACKING -D V8_ADVANCED_BIGINT_ALGORITHMS -D UCONFIG_NO_SERVICE=1 -D U_ENABLE_DYLOAD=0 -D U_STATIC_IMPLEMENTATION=1 -D U_HAVE_STD_STRING=1 -D UCONFIG_NO_BREAK_ITERATION=0 /home/bnoordhuis/src/node/out/Release/obj/gen/torque-generated/src/builtins/wasm-to-js-tq-csa.cc -quiet -dumpdir /home/bnoordhuis/src/node/out/Release/obj.target/v8_initializers/gen/torque-generated/src/builtins/ -dumpbase wasm-to-js-tq-csa.cc -dumpbase-ext .cc -m64 -mtune=generic -march=x86-64 -O3 -O3 -Wno-unused-parameter -Wno-return-type -Wno-invalid-offsetof -std=gnu++17 -flax-vector-conversions -fno-strict-aliasing -fno-omit-frame-pointer -fdata-sections -ffunction-sections -fno-rtti -fno-exceptions -fasynchronous-unwind-tables -fstack-protector-strong -Wformat -Wformat-security -fstack-clash-protection -fcf-protection -o /tmp/ccsXRDpk.s
Commit 78be0d0, linux x86_64, g++ 12.3.0. I observe that the generated file wasm-to-js-tq-csa.cc takes > 20 minutes to compile on an otherwise fast system.
It's a fairly big file (391k, just under 7kLoC) but not that big (template heavy, however.) Other, much bigger torque files compile in seconds flat.
Possibly hits a pathological case in gcc. I wasn't really able to draw any conclusions from looking at
perf topexcept that it was working hard. Perhaps tweaking compiler flags helps.