mirrors/scylladb
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge 'toolchain: support building an optimized clang' from Takuya ASADA

Browse Source 
This is complete version of #12786, since I take over the issue from @mykaul.
Update from original version are:
- Support ARM64 build (disable BOLT for now since it doesn't functioning)
- Changed toolchain settings to get current scylla able to build (support WASM, etc)
- Stop git clone scylladb repo, create git-archive of current scylla directory and import it
-  Update Clang to 17.0.6
- Save entire clang directory for BUILD mode, not just /usr/bin/clang binary
- Implemented INSTALL_PREBUILT mode to install prebuilt image which built in BUILD mode

Note that this patch drops cross-build support of frozen toolchain, since building clang and scylla multiple time in qemu-user-static will very slow, it's not usable.
Instead, we should build the image for each architecture natively.

----

This is a different way attempting to combine building an optimized clang (using LTO, PGO and BOLT, based on compiling ScyllaDB) to dbuild. Per Avi's request, there are 3 options: skip this phase (which is the current default), build it and build + install it to the default path.

Fixes: #10985
Fixes: scylladb/scylla-enterprise#2539

Closes scylladb/scylladb#17196

* github.com:scylladb/scylladb:
  toolchain: support building an optimized clang
  configure.py: add --build-dir option
This commit is contained in:
Avi Kivity
2024-04-18 16:27:38 +00:00
parent a3481a4566 bd34f2fe46
commit 6e487a49aa
7 changed files with 259 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@@ -30,3 +30,4 @@ compile_commands.json
.ccls-cache/
.mypy_cache
.envrc
clang_build

73
configure.py
View File

@@ -20,9 +20,6 @@ import textwrap
from shutil import which
from typing import NamedTuple
outdir = 'build'
tempfile.tempdir = f"{outdir}/tmp"
configure_args = str.join(' ', [shlex.quote(x) for x in sys.argv[1:] if not x.startswith('--out=')])
@@ -287,7 +284,7 @@ def generate_compdb(compdb, ninja, buildfile, modes):
mode_out = outdir + '/' + mode
submodule_compdbs = [mode_out + '/' + submodule + '/' + compdb for submodule in ['seastar']]
with open(mode_out + '/' + compdb, 'w+b') as combined_mode_specific_compdb:
subprocess.run(['./scripts/merge-compdb.py', 'build/' + mode,
subprocess.run(['./scripts/merge-compdb.py', outdir + '/' + mode,
ninja_compdb.name] + submodule_compdbs, stdout=combined_mode_specific_compdb)
# sort modes by supposed indexing speed
@@ -780,10 +777,16 @@ arg_parser.add_argument('--date-stamp', dest='date_stamp', type=str,
help='Set datestamp for SCYLLA-VERSION-GEN')
arg_parser.add_argument('--use-cmake', action='store_true', help='Use CMake as the build system')
arg_parser.add_argument('--coverage', action = 'store_true', help = 'Compile scylla with coverage instrumentation')
arg_parser.add_argument('--build-dir', action='store', default='build',
help='Build directory path')
args = arg_parser.parse_args()
PROFILES_LIST_FILE_NAME = "coverage_sources.list"
outdir = os.path.abspath(args.build_dir)
tempfile.tempdir = f"{outdir}/tmp"
if args.list_artifacts:
for artifact in sorted(all_artifacts):
print(artifact)
@@ -1613,7 +1616,7 @@ def generate_version(date_stamp):
date_stamp_opt = ''
if date_stamp:
date_stamp_opt = f'--date-stamp {date_stamp}'
status = subprocess.call(f"./SCYLLA-VERSION-GEN {date_stamp_opt}", shell=True)
status = subprocess.call(f"./SCYLLA-VERSION-GEN --output-dir {outdir} {date_stamp_opt}", shell=True)
if status != 0:
print('Version file generation failed')
sys.exit(1)
@@ -1885,17 +1888,17 @@ def write_build_file(f,
rule strip
command = scripts/strip.sh $in
rule package
command = scripts/create-relocatable-package.py --build-dir build/$mode $out
command = scripts/create-relocatable-package.py --build-dir $builddir/$mode --node-exporter-dir $builddir/node_exporter --debian-dir $builddir/debian/debian $out
rule stripped_package
command = scripts/create-relocatable-package.py --stripped --build-dir build/$mode $out
command = scripts/create-relocatable-package.py --stripped --build-dir $builddir/$mode --node-exporter-dir $builddir/node_exporter --debian-dir $builddir/debian/debian $out
rule debuginfo_package
command = dist/debuginfo/scripts/create-relocatable-package.py --build-dir build/$mode $out
command = dist/debuginfo/scripts/create-relocatable-package.py --build-dir $builddir/$mode --node-exporter-dir $builddir/node_exporter $out
rule rpmbuild
command = reloc/build_rpm.sh --reloc-pkg $in --builddir $out
rule debbuild
command = reloc/build_deb.sh --reloc-pkg $in --builddir $out
rule unified
command = unified/build_unified.sh --build-dir build/$mode --unified-pkg $out
command = unified/build_unified.sh --build-dir $builddir/$mode --unified-pkg $out
rule rust_header
command = cxxbridge --include rust/cxx.h --header $in > $out
description = RUST_HEADER $out
@@ -2127,6 +2130,17 @@ def write_build_file(f,
mode=mode,
)
)
compiler_training_artifacts=[]
if mode == 'dev':
compiler_training_artifacts.append(f'$builddir/{mode}/scylla')
elif mode == 'release' or mode == 'debug':
compiler_training_artifacts.append(f'$builddir/{mode}/service/storage_proxy.o')
f.write(
'build {mode}-compiler-training: phony {artifacts}\n'.format(
mode=mode,
artifacts=str.join(' ', compiler_training_artifacts)
)
)
gen_dir = '$builddir/{}/gen'.format(mode)
gen_headers = []
@@ -2266,6 +2280,9 @@ def write_build_file(f,
f.write(
'build wasm: phony {}\n'.format(' '.join([f'$builddir/{binary}' for binary in sorted(wasms)]))
)
f.write(
'build compiler-training: phony {}\n'.format(' '.join(['{mode}-compiler-training'.format(mode=mode) for mode in default_modes]))
)
f.write(textwrap.dedent(f'''\
build dist-unified-tar: phony {' '.join([f'$builddir/{mode}/dist/tar/{scylla_product}-unified-{scylla_version}-{scylla_release}.{arch}.tar.gz' for mode in default_modes])}
@@ -2278,54 +2295,54 @@ def write_build_file(f,
build dist-server: phony dist-server-tar dist-server-debuginfo dist-server-rpm dist-server-deb
rule build-submodule-reloc
command = cd $reloc_dir && ./reloc/build_reloc.sh --version $$(<../../build/SCYLLA-PRODUCT-FILE)-$$(sed 's/-/~/' <../../build/SCYLLA-VERSION-FILE)-$$(<../../build/SCYLLA-RELEASE-FILE) --nodeps $args
command = cd $reloc_dir && ./reloc/build_reloc.sh --version $$(<$builddir/SCYLLA-PRODUCT-FILE)-$$(sed 's/-/~/' <$builddir/SCYLLA-VERSION-FILE)-$$(<$builddir/SCYLLA-RELEASE-FILE) --nodeps $args
rule build-submodule-rpm
command = cd $dir && ./reloc/build_rpm.sh --reloc-pkg $artifact
rule build-submodule-deb
command = cd $dir && ./reloc/build_deb.sh --reloc-pkg $artifact
build tools/jmx/build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | build/SCYLLA-PRODUCT-FILE build/SCYLLA-VERSION-FILE build/SCYLLA-RELEASE-FILE
build tools/jmx/build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE
reloc_dir = tools/jmx
build dist-jmx-rpm: build-submodule-rpm tools/jmx/build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/jmx
artifact = $builddir/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-jmx-deb: build-submodule-deb tools/jmx/build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/jmx
artifact = $builddir/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-jmx-tar: phony {' '.join(['$builddir/{mode}/dist/tar/{scylla_product}-jmx-{scylla_version}-{scylla_release}.noarch.tar.gz'.format(mode=mode, scylla_product=scylla_product, scylla_version=scylla_version, scylla_release=scylla_release) for mode in default_modes])}
build dist-jmx: phony dist-jmx-tar dist-jmx-rpm dist-jmx-deb
build tools/java/build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | build/SCYLLA-PRODUCT-FILE build/SCYLLA-VERSION-FILE build/SCYLLA-RELEASE-FILE
build tools/java/build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE
reloc_dir = tools/java
build dist-tools-rpm: build-submodule-rpm tools/java/build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/java
artifact = $builddir/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-tools-deb: build-submodule-deb tools/java/build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/java
artifact = $builddir/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-tools-tar: phony {' '.join(['$builddir/{mode}/dist/tar/{scylla_product}-tools-{scylla_version}-{scylla_release}.noarch.tar.gz'.format(mode=mode, scylla_product=scylla_product, scylla_version=scylla_version, scylla_release=scylla_release) for mode in default_modes])}
build dist-tools: phony dist-tools-tar dist-tools-rpm dist-tools-deb
build tools/cqlsh/build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | build/SCYLLA-PRODUCT-FILE build/SCYLLA-VERSION-FILE build/SCYLLA-RELEASE-FILE
build tools/cqlsh/build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz: build-submodule-reloc | $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE
reloc_dir = tools/cqlsh
build dist-cqlsh-rpm: build-submodule-rpm tools/cqlsh/build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/cqlsh
artifact = $builddir/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-cqlsh-deb: build-submodule-deb tools/cqlsh/build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
dir = tools/cqlsh
artifact = $builddir/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
artifact = build/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz
build dist-cqlsh-tar: phony {' '.join(['$builddir/{mode}/dist/tar/{scylla_product}-cqlsh-{scylla_version}-{scylla_release}.noarch.tar.gz'.format(mode=mode, scylla_product=scylla_product, scylla_version=scylla_version, scylla_release=scylla_release) for mode in default_modes])}
build dist-cqlsh: phony dist-cqlsh-tar dist-cqlsh-rpm dist-cqlsh-deb
build tools/python3/build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz: build-submodule-reloc | build/SCYLLA-PRODUCT-FILE build/SCYLLA-VERSION-FILE build/SCYLLA-RELEASE-FILE
build tools/python3/build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz: build-submodule-reloc | $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE
reloc_dir = tools/python3
args = --packages "{python3_dependencies}" --pip-packages "{pip_dependencies}" --pip-symlinks "{pip_symlinks}"
build dist-python3-rpm: build-submodule-rpm tools/python3/build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
dir = tools/python3
artifact = $builddir/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
artifact = build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
build dist-python3-deb: build-submodule-deb tools/python3/build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
dir = tools/python3
artifact = $builddir/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
artifact = build/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz
build dist-python3-tar: phony {' '.join(['$builddir/{mode}/dist/tar/{scylla_product}-python3-{scylla_version}-{scylla_release}.{arch}.tar.gz'.format(mode=mode, scylla_product=scylla_product, arch=arch, scylla_version=scylla_version, scylla_release=scylla_release) for mode in default_modes])}
build dist-python3: phony dist-python3-tar dist-python3-rpm dist-python3-deb
build dist-deb: phony dist-server-deb dist-python3-deb dist-jmx-deb dist-tools-deb dist-cqlsh-deb
@@ -2360,10 +2377,10 @@ def write_build_file(f,
f.write(textwrap.dedent('''\
rule configure
command = {python} configure.py --out=build.ninja.new $configure_args && mv build.ninja.new build.ninja
command = {python} configure.py --out={buildfile}.new $configure_args && mv {buildfile}.new {buildfile}
generator = 1
description = CONFIGURE $configure_args
build build.ninja {build_ninja_list}: configure | configure.py SCYLLA-VERSION-GEN $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE {args.seastar_path}/CMakeLists.txt
build {buildfile} {build_ninja_list}: configure | configure.py SCYLLA-VERSION-GEN $builddir/SCYLLA-PRODUCT-FILE $builddir/SCYLLA-VERSION-FILE $builddir/SCYLLA-RELEASE-FILE {args.seastar_path}/CMakeLists.txt
rule cscope
command = find -name '*.[chS]' -o -name "*.cc" -o -name "*.hh" | cscope -bq -i-
description = CSCOPE
@@ -2377,7 +2394,7 @@ def write_build_file(f,
description = List configured modes
build mode_list: mode_list
default {modes_list}
''').format(modes_list=' '.join(default_modes), build_ninja_list=' '.join([f'build/{mode}/{dir}/build.ninja' for mode in build_modes for dir in ['seastar']]), **globals()))
''').format(modes_list=' '.join(default_modes), build_ninja_list=' '.join([f'{outdir}/{mode}/{dir}/build.ninja' for mode in build_modes for dir in ['seastar']]), **globals()))
unit_test_list = set(test for test in build_artifacts if test in set(tests))
f.write(textwrap.dedent('''\
rule unit_test_list
@@ -2388,14 +2405,14 @@ def write_build_file(f,
f.write(textwrap.dedent('''\
build always: phony
rule scylla_version_gen
command = ./SCYLLA-VERSION-GEN
command = ./SCYLLA-VERSION-GEN --output-dir $builddir
restat = 1
build $builddir/SCYLLA-RELEASE-FILE $builddir/SCYLLA-VERSION-FILE: scylla_version_gen | always
rule debian_files_gen
command = ./dist/debian/debian_files_gen.py
command = ./dist/debian/debian_files_gen.py --build-dir $builddir
build $builddir/debian/debian: debian_files_gen | always
rule extract_node_exporter
command = tar -C build -xvpf {node_exporter_filename} --no-same-owner && rm -rfv build/node_exporter && mv -v build/{node_exporter_dirname} build/node_exporter
command = tar -C $builddir -xvpf {node_exporter_filename} --no-same-owner && rm -rfv $builddir/node_exporter && mv -v $builddir/{node_exporter_dirname} $builddir/node_exporter
build $builddir/node_exporter/node_exporter: extract_node_exporter | always
build $builddir/node_exporter/node_exporter.stripped: strip $builddir/node_exporter/node_exporter
build $builddir/node_exporter/node_exporter.debug: phony $builddir/node_exporter/node_exporter.stripped

1
install-dependencies.sh
View File

@@ -113,6 +113,7 @@ fedora_packages=(
wabt
binaryen
lcov
llvm-bolt
)
# lld is not available on s390x, see

13
tools/toolchain/Dockerfile
View File

@@ -1,17 +1,18 @@
FROM docker.io/fedora:38
ADD ./install-dependencies.sh ./
ADD ./seastar/install-dependencies.sh ./seastar/
ADD ./tools/jmx/install-dependencies.sh ./tools/jmx/
ADD ./tools/java/install-dependencies.sh ./tools/java/
ADD ./tools/toolchain/system-auth ./
ARG CLANG_BUILD="SKIP"
ARG CLANG_ARCHIVE
WORKDIR /mnt
RUN dnf -y update \
&& dnf -y install 'dnf-command(copr)' \
&& dnf -y install ccache \
&& dnf -y install devscripts debhelper fakeroot file rpm-build \
&& ./install-dependencies.sh && dnf clean all \
&& echo 'ALL ALL=(ALL:ALL) NOPASSWD: ALL' >> /etc/sudoers \
&& cp system-auth /etc/pam.d \
&& cp ./tools/toolchain/system-auth /etc/pam.d \
&& echo 'Defaults !requiretty' >> /etc/sudoers
RUN mkdir -p /root/.m2/repository
ENV JAVA8_HOME=/usr/lib/jvm/java-1.8.0-openjdk
RUN ./tools/toolchain/optimized_clang.sh
CMD /bin/bash

24
tools/toolchain/README.md
View File

@@ -85,45 +85,33 @@ there is a need to update a toolchain, the branch designation is added to
the tag to avoid ambiguity.
Publishing an image is complicated since multiple architectures are supported.
There are two procedures, one using emulation (can run on any x86 machine) and
another using native systems, which requires access to aarch64 and s390x machines.
You will need credentials to push images. You either need to setup permanent
credentials with `podman login`, or you can just append
`--creds=yourname@scylladb.com` to the `podman manifest push` command and then
type in your password.
## Emulated publishing procedure (slow)
## Publishing procedure
1. Pick a new name for the image (in `tools/toolchain/image`) and
commit it. The commit updating install-dependencies.sh should
include the toolchain change, for atomicity. Do not push the commit
to `next` yet.
2. Run `tools/toolchain/prepare` and wait. It requires `buildah` and
`qemu-user-static` to be installed (and will complain if they are not).
3. Publish the image using the instructions printed by the previous step.
4. Push the `next` branch that refers to the new toolchain.
## Native publishing procedure (complicated)
1. Pick a new name for the image (in `tools/toolchain/image`) and
commit it. The commit updating install-dependencies.sh should
include the toolchain change, for atomicity. Do not push the commit
to `next` yet.
2. Push the commit to a personal repository/branch.
3. Perform the following on an x86 and an ARM machine:
2. Run `tools/toolchain/prepare --clang-build-mode INSTALL` and wait. It requires `buildah` and to be installed (and will complain if they are not).
3. Push the commit to a personal repository/branch.
4. Perform the following on an x86 and an ARM machine:
1. check out the branch containing the new toolchain name
2. Run `git submodule update --init --recursive` to make sure
all the submodules are synchronized
3. Run `podman build --no-cache --pull --tag mytag-arch -f tools/toolchain/Dockerfile .`, where mytag-arch is a new, unique tag that is different for x86 and ARM.
4. Push the resulting images to a personal docker repository.
4. Now, create a multiarch image with the following:
5. Now, create a multiarch image with the following:
1. Pull the two images with `podman pull`. Let's call the two tags
`mytag-x86` and `mytag-arm`.
2. Create the new toolchain manifest with `podman manifest create $(<tools/toolchain/image)`
3. Add each image with `podman manifest add --all $(<tools/toolchain/image) mytag-x86` and `podman manifest add --all $(<tools/toolchain/image) mytag-arm`
4. Push the image with `podman manifest push --all $(<tools/toolchain/image) docker://$(<tools/toolchain/image)`
5. Now push the commit that updated the toolchain with `git push`.
6. Now push the commit that updated the toolchain with `git push`.
## Troubleshooting

117
tools/toolchain/optimized_clang.sh Executable file
View File

@@ -0,0 +1,117 @@
#!/bin/bash -uex
case "${CLANG_BUILD}" in
"SKIP")
echo "CLANG_BUILD: ${CLANG_BUILD}"
exit 0
;;
"INSTALL" | "INSTALL_FROM")
echo "CLANG_BUILD: ${CLANG_BUILD}"
if [[ -z "${CLANG_ARCHIVE}" ]]; then
echo "CLANG_ARCHIVE not specified"
fi
;;
"")
echo "CLANG_BUILD not specified"
exit 1
;;
*)
echo "Invalid mode specified on CLANG_BUILD: ${CLANG_BUILD}"
exit 1
;;
esac
ARCH="$(arch)"
if [[ "${ARCH}" = "x86_64" ]]; then
LLVM_TARGET_ARCH=X86
elif [[ "${ARCH}" = "aarch64" ]]; then
LLVM_TARGET_ARCH=AArch64
else
echo "Unsupported architecture: ${ARCH}"
exit 1
fi
SCYLLA_DIR=/mnt
CLANG_ROOT_DIR="${SCYLLA_DIR}"/clang_build
CLANG_CHECKOUT_NAME=llvm-project
CLANG_BUILD_DIR="${CLANG_ROOT_DIR}"/"${CLANG_CHECKOUT_NAME}"
SCYLLA_BUILD_DIR="${SCYLLA_DIR}"/build_profile
SCYLLA_NINJA_FILE="${SCYLLA_DIR}"/build_profile.ninja
# Which LLVM release to build in order to compile Scylla
LLVM_CLANG_TAG=16.0.6
CLANG_SUFFIX=16
CLANG_ARCHIVE=$(cd "${SCYLLA_DIR}" && realpath -m "${CLANG_ARCHIVE}")
CLANG_OPTS=(-DCMAKE_C_COMPILER="/usr/bin/clang" -DCMAKE_CXX_COMPILER="/usr/bin/clang++" -DLLVM_USE_LINKER="/usr/bin/ld.lld" -DCMAKE_BUILD_TYPE=Release -DLLVM_TARGETS_TO_BUILD="${LLVM_TARGET_ARCH};WebAssembly" -DLLVM_TARGET_ARCH="${LLVM_TARGET_ARCH}" -G Ninja -DLLVM_INCLUDE_BENCHMARKS=OFF -DLLVM_INCLUDE_EXAMPLES=OFF -DLLVM_INCLUDE_TESTS=OFF -DLLVM_ENABLE_BINDINGS=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" -DLLVM_ENABLE_LTO=Thin -DCLANG_DEFAULT_PIE_ON_LINUX=OFF -DLLVM_BUILD_TOOLS=OFF -DLLVM_VP_COUNTERS_PER_SITE=6)
SCYLLA_OPTS=(--date-stamp "$(date "+%Y%m%d")" --debuginfo 1 --tests-debuginfo 1 --c-compiler="${CLANG_BUILD_DIR}/build/bin/clang" --compiler="${CLANG_BUILD_DIR}/build/bin/clang++" --build-dir="${SCYLLA_BUILD_DIR}" --out="${SCYLLA_NINJA_FILE}")
if [[ "${CLANG_BUILD}" = "INSTALL" ]]; then
# Build a PGO-optimized compiler using the boostrapped compiler.
rm -rf "${CLANG_BUILD_DIR}"
git clone https://github.com/llvm/llvm-project --branch llvmorg-"${LLVM_CLANG_TAG}" --depth=1 "${CLANG_BUILD_DIR}"
cd "${CLANG_BUILD_DIR}"
cmake -B build -S llvm "${CLANG_OPTS[@]}" -DLLVM_BUILD_INSTRUMENTED=IR
ninja -C build
rm -rf "${SCYLLA_BUILD_DIR}" "${SCYLLA_NINJA_FILE}"
cd "${SCYLLA_DIR}"
./configure.py "${SCYLLA_OPTS[@]}"
LLVM_PROFILE_FILE="${CLANG_BUILD_DIR}"/build/profiles/default_%p-%m.profraw ninja -f "${SCYLLA_NINJA_FILE}" compiler-training
cd "${CLANG_BUILD_DIR}"
llvm-profdata merge "${CLANG_BUILD_DIR}"/build/profiles/default_*.profraw -output=ir.prof
rm -rf build
cmake -B build -S llvm "${CLANG_OPTS[@]}" -DLLVM_BUILD_INSTRUMENTED=CSIR -DLLVM_PROFDATA_FILE="$(realpath ir.prof)"
ninja -C build
# 2nd compilation: gathering a clang profile for CSPGO
rm -rf "${SCYLLA_BUILD_DIR}" "${SCYLLA_NINJA_FILE}"
cd "${SCYLLA_DIR}"
./configure.py "${SCYLLA_OPTS[@]}"
LLVM_PROFILE_FILE="${CLANG_BUILD_DIR}"/build/profiles/csir-%p-%m.profraw ninja -f "${SCYLLA_NINJA_FILE}" compiler-training
cd "${CLANG_BUILD_DIR}"
llvm-profdata merge build/csprofiles/default_*.profraw -output=csir.prof
llvm-profdata merge ir.prof csir.prof -output=combined.prof
rm -rf build
# -DLLVM_LIBDIR_SUFFIX=64 for Fedora compatibility
cmake -B build -S llvm "${CLANG_OPTS[@]}" -DLLVM_PROFDATA_FILE="$(realpath combined.prof)" -DCMAKE_EXE_LINKER_FLAGS="-Wl,--emit-relocs" -DCMAKE_INSTALL_PREFIX=/usr/local -DLLVM_LIBDIR_SUFFIX=64
ninja -C build
#TODO: skipping BOLT for aarch64 for now, since it causes segfault
if [[ "${ARCH}" != "aarch64" ]]; then
# BOLT phase
mv build/bin/clang-"${CLANG_SUFFIX}" build/bin/clang-"${CLANG_SUFFIX}".prebolt
mkdir -p build/profiles
llvm-bolt build/bin/clang-"${CLANG_SUFFIX}".prebolt -o build/bin/clang-"${CLANG_SUFFIX}" --instrument --instrumentation-file="${CLANG_BUILD_DIR}"/build/profiles/prof --instrumentation-file-append-pid --conservative-instrumentation
# 3rd ScyllaDB compilation: gathering a clang profile for BOLT
rm -rf "${SCYLLA_BUILD_DIR}" "${SCYLLA_NINJA_FILE}"
cd "${SCYLLA_DIR}"
./configure.py "${SCYLLA_OPTS[@]}"
ninja -f "${SCYLLA_NINJA_FILE}" compiler-training
cd "${CLANG_BUILD_DIR}"
merge-fdata build/profiles/*.fdata > prof.fdata
llvm-bolt build/bin/clang-"${CLANG_SUFFIX}".prebolt -o build/bin/clang-"${CLANG_SUFFIX}" --data=prof.fdata --reorder-functions=hfsort --reorder-blocks=ext-tsp --split-functions --split-all-cold --split-eh --dyno-stats
fi
cd "${CLANG_ROOT_DIR}"
rm -rf "${CLANG_BUILD_DIR}"/{build/profiles,*.prof,prof.fdata}
tar -cpzf "${CLANG_ARCHIVE}" "${CLANG_CHECKOUT_NAME}"
elif [[ "${CLANG_BUILD}" = "INSTALL_FROM" ]]; then
mkdir -p "${CLANG_ROOT_DIR}"
tar -C "${CLANG_ROOT_DIR}" -xpzf "${CLANG_ARCHIVE}"
fi
cd "${CLANG_BUILD_DIR}"
mv /usr/bin/clang-"${CLANG_SUFFIX}" /usr/bin/clang-"${CLANG_SUFFIX}".orig
mv /usr/bin/lld /usr/bin/lld.orig
mv /usr/lib64/libLTO.so."${CLANG_SUFFIX}" /usr/lib64/libLTO.so."${CLANG_SUFFIX}".orig
install -Z -m755 "${CLANG_BUILD_DIR}"/build/bin/clang-"${CLANG_SUFFIX}" /usr/bin/clang-"${CLANG_SUFFIX}"
install -Z -m755 "${CLANG_BUILD_DIR}"/build/bin/lld /usr/bin/lld
install -Z -m755 "${CLANG_BUILD_DIR}"/build/lib64/libLTO.so."${CLANG_SUFFIX}" /usr/lib64/libLTO.so."${CLANG_SUFFIX}"
rm -rf "${CLANG_BUILD_DIR}" "${SCYLLA_BUILD_DIR}" "${SCYLLA_NINJA_FILE}"

98
tools/toolchain/prepare
View File

@@ -32,29 +32,95 @@ if reg digest $(<tools/toolchain/image) > /dev/null; then
exit 1
fi
archs=(amd64 arm64)
# docker arch has a diffrent spelling than uname arch
declare -A arch_unames=(
[amd64]=x86_64
[arm64]=aarch64
declare -A docker_arch=(
[x86_64]=amd64
[aarch64]=arm64
[s390x]=s390x
)
for arch in "${archs[@]}"; do
# translate from docker arch to uname arch
arch_uname="${arch_unames[$arch]}"
if [[ "$(uname -m)" == "${arch_uname}" ]]; then
continue
fi
if [[ ! -f /proc/sys/fs/binfmt_misc/qemu-"${arch_uname}" ]]; then
echo install qemu-user-static
exit 1
fi
usage() {
cat <<EOF
Options:
--clang-build-mode <mode> specify one of following build modes:
SKIP: skip building optimized clang
INSTALL: build and install an optimized clang compiler binary, and save it as a tarball.
INSTALL_FROM: install optimized clang from a tarball.
--clang-archive <file> specify optimized clang tarball path
--help this help snippet
EOF
}
CLANG_BUILD="SKIP"
while [[ $# -gt 0 ]]; do
case "$1" in
"--clang-build-mode")
if [[ -z "$2" ]]; then
usage
exit 1
fi
CLANG_BUILD="$2"
shift 2
;;
"--clang-archive")
if [[ -z "$2" ]]; then
usage
exit 1
fi
CLANG_ARCHIVE="$(realpath -m --relative-to=. "$2")"
shift 2
;;
"--help")
usage
exit 0
;;
*)
usage
exit 1
;;
esac
done
buildah bud "${archs[@]/#/--platform=linux/}" --jobs 0 --squash --no-cache --pull -f tools/toolchain/Dockerfile --manifest "$(<tools/toolchain/image)"
case "${CLANG_BUILD}" in
"SKIP" | "INSTALL" | "INSTALL_FROM")
;;
*)
echo "Invalid mode specified on CLANG_BUILD: ${CLANG_BUILD}"
exit 1
;;
esac
# clang archive file need to be under scylla directory, since we only share
# scylla directory with the container
if [[ "${CLANG_ARCHIVE}" = "../"* ]]; then
echo "clang archive file need to be under scylla directory"
exit 1
fi
# set default archive path if not specified
if [[ -z "${CLANG_ARCHIVE}" ]]; then
CURDIR="$(dirname "$0")"
LLVM_CLANG_TAG="$(sed -n -e 's/^LLVM_CLANG_TAG=\(.*\)/\1/p' "${CURDIR}"/optimized_clang.sh)"
IMAGE_ID="$(sed -e 's#docker.io/scylladb/scylla-toolchain:##' "${CURDIR}"/image)"
CLANG_ARCHIVE="clang_build/optimized_clang_${LLVM_CLANG_TAG}_${IMAGE_ID}.$(arch).tar.gz"
fi
if [[ "${CLANG_BUILD}" = "INSTALL_FROM" ]] && [[ ! -f "${CLANG_ARCHIVE}" ]]; then
echo "${CLANG_ARCHIVE}" does not exist.
echo "Please specify vaild file with --clang-archive"
exit 1
fi
buildah bud --platform="linux/${docker_arch[$(arch)]}" --jobs 0 --squash --no-cache --pull -f tools/toolchain/Dockerfile --manifest "$(<tools/toolchain/image)" -v "$(realpath ./):/mnt:Z" --build-arg CLANG_BUILD="${CLANG_BUILD}" --build-arg CLANG_ARCHIVE="${CLANG_ARCHIVE}"
echo "Done building $(<tools/toolchain/image). You can now test it, and push with"
echo ""
echo " podman manifest push --all $(<tools/toolchain/image) docker://$(<tools/toolchain/image)"
if [[ "${CLANG_BUILD}" = "INSTALL" ]]; then
echo ""
echo "Optimized clang archive saved at:"
echo " ${CLANG_ARCHIVE}"
echo "You can rebuild frozen toolchain without rebuilding optimized clang by:"
echo " ./tools/toolchain/prepare --clang-build-mode INSTALL_FROM --clang-archive ${CLANG_ARCHIVE}"
fi