Optimize mold-git

- use LTO+PGO+BOLT

Author: ms178

packages/mold-git/PKGBUILD

@@ -1,19 +1,18 @@
-# Maintainer: Peter Jung ptr1337 <[email protected]>
-
 pkgname=mold-git
-pkgver=v2.31.0_32_g08be7ff2
+pkgver=v2.36.0_2_gbe087071
 pkgrel=1
 pkgdesc="A Modern Linker"
 arch=(x86_64)
 url="https://github.com/rui314/mold"
 license=("AGPL3")
 # xxhash is bundled
 depends=('gcc-libs' 'mimalloc' 'openssl' 'zlib' 'onetbb')
-makedepends=('git' 'python' 'cmake' 'mold')
+makedepends=('git' 'python' 'cmake' 'mold' 'llvm')
 checkdepends=('clang' 'libdwarf')
 source=("mold::git+https://github.com/rui314/mold")
 sha256sums=('SKIP')
 provides=("mold=$pkgver")
+options=(!lto !strip)
 conflicts=("mold")
 reponame="mold"
 
@@ -22,24 +21,92 @@ pkgver() {
     git describe --long --tags | sed "s/-/_/g"
 }
 
-
 build() {
+  # First build - Generate profile data
   cmake \
-  -S "$reponame" \
-  -B build \
-  -DCMAKE_BUILD_TYPE='Release' \
-  -DCMAKE_INSTALL_PREFIX='/usr' \
-  -DCMAKE_INSTALL_LIBEXECDIR='lib' \
-  -DMOLD_USE_SYSTEM_MIMALLOC=OFF \
-  -DMOLD_USE_SYSTEM_TBB=OFF \
-  -DCMAKE_C_STANDARD=23 \
-  -DCMAKE_CXX_STANDARD=23 \
-  -DMOLD_USE_MOLD=ON \
-  -DMOLD_LTO=ON
+    -S "$reponame" \
+    -B build-pgo \
+    -DCMAKE_BUILD_TYPE='Debug' \
+    -DCMAKE_INSTALL_PREFIX='/usr' \
+    -DCMAKE_INSTALL_LIBEXECDIR='lib' \
+    -DMOLD_USE_SYSTEM_MIMALLOC=OFF \
+    -DMOLD_USE_SYSTEM_TBB=OFF \
+    -DCMAKE_C_STANDARD=23 \
+    -DCMAKE_CXX_STANDARD=23 \
+    -DMOLD_USE_MOLD=OFF \
+    -DMOLD_LTO=OFF \
+    -DCMAKE_C_FLAGS="-fprofile-generate -fno-omit-frame-pointer -g3 -mllvm -vp-counters-per-site=6" \
+    -DCMAKE_CXX_FLAGS="-fprofile-generate -fno-omit-frame-pointer -g3 -mllvm -vp-counters-per-site=6"
+
+  cmake --build build-pgo
+
+  # Run tests to generate profile data
+  cd build-pgo
+  ctest --output-on-failure
+  cd ..
+
+  # Merge profile data
+  llvm-profdata merge -output=build-pgo/default.profdata build-pgo/default*.profraw
+
+  # Second build - Use profile data with debug info for BOLT
+  cmake \
+    -S "$reponame" \
+    -B build \
+    -DCMAKE_BUILD_TYPE='Release' \
+    -DCMAKE_INSTALL_PREFIX='/usr' \
+    -DCMAKE_INSTALL_LIBEXECDIR='lib' \
+    -DMOLD_USE_SYSTEM_MIMALLOC=OFF \
+    -DMOLD_USE_SYSTEM_TBB=OFF \
+    -DCMAKE_C_STANDARD=23 \
+    -DCMAKE_CXX_STANDARD=23 \
+    -DMOLD_USE_MOLD=OFF \
+    -DMOLD_LTO=ON \
+    -DCMAKE_C_FLAGS="-fprofile-use=/tmp/makepkg/mold-git/src/build-pgo/default.profdata" \
+    -DCMAKE_CXX_FLAGS="-fprofile-use=/tmp/makepkg/mold-git/src/build-pgo/default.profdata" \
+    -DCMAKE_EXE_LINKER_FLAGS="-Wl,--emit-relocs" \
+    -DCMAKE_MODULE_LINKER_FLAGS="-Wl,--emit-relocs" \
+    -DCMAKE_SHARED_LINKER_FLAGS="-Wl,--emit-relocs"
 
   cmake --build build
-}
 
+  # Create perf data for BOLT
+  cd build
+  perf record -e cycles:u -j any,u -o perf.data -- ctest --output-on-failure
+  cd ..
+
+  # Process with BOLT
+  llvm-bolt build/mold \
+    --data=build/perf.data \
+    --dyno-stats \
+    --cu-processing-batch-size=64 \
+    --eliminate-unreachable\
+    --frame-opt=all \
+    --icf=all \
+    --jump-tables=aggressive \
+    --min-branch-clusters \
+    --stoke \
+    --sctc-mode=always \
+    --plt=all \
+    --hot-data \
+    --hot-text \
+    --frame-opt-rm-stores \
+    --peepholes=all \
+    --infer-stale-profile=1 \
+    --x86-strip-redundant-address-size \
+    --indirect-call-promotion=all \
+    --reg-reassign \
+    --use-aggr-reg-reassign \
+    --reorder-blocks=ext-tsp \
+    --reorder-functions=cdsort \
+    --split-all-cold \
+    --split-eh \
+    --split-functions \
+    --split-strategy=cdsplit \
+    -o build/mold.bolt
+
+  # Replace original binary with BOLTed one
+  mv build/mold.bolt build/mold
+}
 
 check() {
   ctest --test-dir build --output-on-failure