Merge pull request rust-lang#278 from rust-lang/feature/lto_2023-05-12

Add support for Link-Time Optimization

Author: antoyo

.github/workflows/ci.yml

@@ -57,8 +57,8 @@ jobs:
       uses: dawidd6/action-download-artifact@v2
       with:
           workflow: main.yml
-          name: ${{ matrix.libgccjit_version.gcc }}
-          path: gcc-build
+          name: gcc-13
+          path: gcc-13
           repo: antoyo/gcc
           branch: ${{ matrix.libgccjit_version.artifacts_branch }}
           event: push
@@ -71,9 +71,8 @@ jobs:
     - name: Setup path to libgccjit
       if: matrix.libgccjit_version.gcc != 'libgccjit12.so'
       run: |
-          echo $(readlink -f gcc-build) > gcc_path
-          # NOTE: the filename is still libgccjit.so even when the artifact name is different.
-          ln gcc-build/libgccjit.so gcc-build/libgccjit.so.0
+          sudo dpkg --force-overwrite -i gcc-13/gcc-13.deb
+          echo /usr/lib/ > gcc_path
 
     - name: Set env
       run: |
@@ -141,6 +140,9 @@ jobs:
       if: ${{ matrix.libgccjit_version.gcc == 'libgccjit12.so' }}
       run: cat failing-ui-tests12.txt >> failing-ui-tests.txt
 
+    - name: Add more failing tests because the sysroot is not compiled with LTO
+      run: cat failing-non-lto-tests.txt >> failing-ui-tests.txt
+
     - name: Run tests
       run: |
         ${{ matrix.libgccjit_version.env_extra }} ./test.sh --release --clean --build-sysroot ${{ matrix.commands }} ${{ matrix.libgccjit_version.extra }}

.github/workflows/release.yml

@@ -18,8 +18,6 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        libgccjit_version:
-          - { gcc: "libgccjit.so", artifacts_branch: "master" }
         commands: [
           "--test-successful-rustc --nb-parts 2 --current-part 0",
           "--test-successful-rustc --nb-parts 2 --current-part 1",
@@ -40,18 +38,17 @@ jobs:
       uses: dawidd6/action-download-artifact@v2
       with:
           workflow: main.yml
-          name: ${{ matrix.libgccjit_version.gcc }}
-          path: gcc-build
+          name: gcc-13
+          path: gcc-13
           repo: antoyo/gcc
-          branch: ${{ matrix.libgccjit_version.artifacts_branch }}
+          branch: "master"
           event: push
           search_artifacts: true # Because, instead, the action only check the last job ran and that won't work since we want multiple artifacts.
 
     - name: Setup path to libgccjit
       run: |
-          echo $(readlink -f gcc-build) > gcc_path
-          # NOTE: the filename is still libgccjit.so even when the artifact name is different.
-          ln gcc-build/libgccjit.so gcc-build/libgccjit.so.0
+          sudo dpkg --force-overwrite -i gcc-13/gcc-13.deb
+          echo /usr/lib/ > gcc_path
 
     - name: Set env
       run: |
@@ -89,7 +86,7 @@ jobs:
     - name: Build
       run: |
         ./y.sh prepare --only-libcore
-        ./build.sh --release --release-sysroot
+        EMBED_LTO_BITCODE=1 ./build.sh --release --release-sysroot
         cargo test
         ./clean_all.sh
 
@@ -98,6 +95,8 @@ jobs:
         git config --global user.email "[email protected]"
         git config --global user.name "User"
         ./y.sh prepare
+        # FIXME(antoyo): we cannot enable LTO for stdarch tests currently because of some failing LTO tests using proc-macros.
+        echo -n 'lto = "fat"' >> build_sysroot/Cargo.toml
 
     # Compile is a separate step, as the actions-rs/cargo action supports error annotations
     - name: Compile
@@ -106,6 +105,9 @@ jobs:
         command: build
         args: --release
 
+    - name: Add more failing tests because of undefined symbol errors (FIXME)
+      run: cat failing-lto-tests.txt >> failing-ui-tests.txt
+
     - name: Run tests
       run: |
-        ./test.sh --release --clean --release-sysroot --build-sysroot ${{ matrix.commands }}
+        EMBED_LTO_BITCODE=1 ./test.sh --release --clean --release-sysroot --build-sysroot ${{ matrix.commands }}

.github/workflows/stdarch.yml

@@ -18,8 +18,6 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        libgccjit_version:
-          - { gcc: "libgccjit.so", artifacts_branch: "master" }
         cargo_runner: [
           "sde -future -rtm_mode full --",
           "",
@@ -54,18 +52,17 @@ jobs:
       uses: dawidd6/action-download-artifact@v2
       with:
           workflow: main.yml
-          name: ${{ matrix.libgccjit_version.gcc }}
-          path: gcc-build
+          name: gcc-13
+          path: gcc-13
           repo: antoyo/gcc
-          branch: ${{ matrix.libgccjit_version.artifacts_branch }}
+          branch: "master"
           event: push
           search_artifacts: true # Because, instead, the action only check the last job ran and that won't work since we want multiple artifacts.
 
     - name: Setup path to libgccjit
       run: |
-          echo $(readlink -f gcc-build) > gcc_path
-          # NOTE: the filename is still libgccjit.so even when the artifact name is different.
-          ln gcc-build/libgccjit.so gcc-build/libgccjit.so.0
+          sudo dpkg --force-overwrite -i gcc-13/gcc-13.deb
+          echo /usr/lib/ > gcc_path
 
     - name: Set env
       run: |
@@ -139,4 +136,5 @@ jobs:
       if: ${{ matrix.cargo_runner }}
       run: |
         cd build_sysroot/sysroot_src/library/stdarch/
+        # FIXME: these tests fail when the sysroot is compiled with LTO because of a missing symbol in proc-macro.
         STDARCH_TEST_EVERYTHING=1 CHANNEL=release CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="${{ matrix.cargo_runner }}" TARGET=x86_64-unknown-linux-gnu CG_RUSTFLAGS="-Ainternal_features" ../../../../cargo.sh test -- --skip rtm --skip tbm --skip sse4a

Cargo.lock

@@ -27,6 +27,10 @@ gccjit = { git = "https://github.com/antoyo/gccjit.rs" }
 # Local copy.
 #gccjit = { path = "../gccjit.rs" }
 
+object = { version = "0.30.1", default-features = false, features = [
+    "std",
+    "read",
+] }
 smallvec = { version = "1.6.1", features = ["union", "may_dangle"] }
 # TODO(antoyo): make tempfile optional.
 tempfile = "3.7.1"

Cargo.toml

@@ -91,6 +91,9 @@ $ CHANNEL="release" $CG_GCCJIT_DIR/cargo.sh run
 
 If you compiled cg_gccjit in debug mode (aka you didn't pass `--release` to `./test.sh`) you should use `CHANNEL="debug"` instead or omit `CHANNEL="release"` completely.
 
+To use LTO, you need to set the variable `FAT_LTO=1` and `EMBED_LTO_BITCODE=1` in addition to setting `lto = "fat"` in the `Cargo.toml`.
+Don't set `FAT_LTO` when compiling the sysroot, though: only set `EMBED_LTO_BITCODE=1`.
+
 ### Rustc
 
 > You should prefer using the Cargo method.
@@ -191,6 +194,48 @@ set substitute-path /usr/src/debug/gcc /path/to/gcc-repo/gcc
 
 TODO(antoyo): but that's not what I remember I was doing.
 
+### `failed to build archive` error
+
+When you get this error:
+
+```
+error: failed to build archive: failed to open object file: No such file or directory (os error 2)
+```
+
+That can be caused by the fact that you try to compile with `lto = "fat"`, but you didn't compile the sysroot with LTO.
+(Not sure if that's the reason since I cannot reproduce anymore. Maybe it happened when forgetting setting `FAT_LTO`.)
+
+### How to debug GCC LTO
+
+Run do the command with `-v -save-temps` and then extract the `lto1` line from the output and run that under the debugger.
+
+### How to send arguments to the GCC linker
+
+```
+CG_RUSTFLAGS="-Clink-args=-save-temps -v" ../cargo.sh build
+```
+
+### How to see the personality functions in the asm dump
+
+```
+CG_RUSTFLAGS="-Clink-arg=-save-temps -v -Clink-arg=-dA" ../cargo.sh build
+```
+
+### How to see the LLVM IR for a sysroot crate
+
+```
+cargo build -v --target x86_64-unknown-linux-gnu -Zbuild-std
+# Take the command from the output and add --emit=llvm-ir
+```
+
+### To prevent the linker from unmangling symbols
+
+Run with:
+
+```
+COLLECT_NO_DEMANGLE=1
+```
+
 ### How to use a custom-build rustc
 
  * Build the stage2 compiler (`rustup toolchain link debug-current build/x86_64-unknown-linux-gnu/stage2`).
@@ -253,4 +298,4 @@ generate it in [gimple.md](./doc/gimple.md).
  * Set `linker='-Clinker=m68k-linux-gcc'`.
  * Set the path to the cross-compiling libgccjit in `gcc_path`.
  * Comment the line: `context.add_command_line_option("-masm=intel");` in src/base.rs.
- * (might not be necessary) Disable the compilation of libstd.so (and possibly libcore.so?).
+ * (might not be necessary) Disable the compilation of libstd.so (and possibly libcore.so?): Remove dylib from build_sysroot/sysroot_src/library/std/Cargo.toml.

Readme.md

@@ -19,3 +19,4 @@ rustc-std-workspace-std = { path = "./sysroot_src/library/rustc-std-workspace-st
 
 [profile.release]
 debug = true
+#lto = "fat" # TODO(antoyo): re-enable when the failing LTO tests regarding proc-macros are fixed.

build_sysroot/Cargo.toml

@@ -38,7 +38,14 @@ if [[ "$HOST_TRIPLE" != "$TARGET_TRIPLE" ]]; then
    fi
 fi
 
-export RUSTFLAGS="$CG_RUSTFLAGS $linker -Csymbol-mangling-version=v0 -Cdebuginfo=2 -Clto=off -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot $TEST_FLAGS"
+# Since we don't support ThinLTO, disable LTO completely when not trying to do LTO.
+# TODO(antoyo): remove when we can handle ThinLTO.
+disable_lto_flags=''
+if [[ ! -v FAT_LTO ]]; then
+    disable_lto_flags='-Clto=off'
+fi
+
+export RUSTFLAGS="$CG_RUSTFLAGS $linker -Csymbol-mangling-version=v0 -Cdebuginfo=2 $disable_lto_flags -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot $TEST_FLAGS"
 
 # FIXME(antoyo): remove once the atomic shim is gone
 if [[ `uname` == 'Darwin' ]]; then
@@ -50,3 +57,7 @@ export RUSTC_LOG=warn # display metadata load errors
 
 export LD_LIBRARY_PATH="$(pwd)/target/out:$(pwd)/build_sysroot/sysroot/lib/rustlib/$TARGET_TRIPLE/lib:$GCC_PATH"
 export DYLD_LIBRARY_PATH=$LD_LIBRARY_PATH
+# NOTE: To avoid the -fno-inline errors, use /opt/gcc/bin/gcc instead of cc.
+# To do so, add a symlink for cc to /opt/gcc/bin/gcc in our PATH.
+# Another option would be to add the following Rust flag: -Clinker=/opt/gcc/bin/gcc
+export PATH="/opt/gcc/bin:$PATH"

config.sh

@@ -0,0 +1,23 @@
+tests/ui/lint/unsafe_code/forge_unsafe_block.rs
+tests/ui/lint/unused-qualification-in-derive-expansion.rs
+tests/ui/macro-quote-test.rs
+tests/ui/macros/proc_macro.rs
+tests/ui/panic-runtime/lto-unwind.rs
+tests/ui/resolve/derive-macro-1.rs
+tests/ui/resolve/derive-macro-2.rs
+tests/ui/rfcs/rfc-2565-param-attrs/param-attrs-pretty.rs
+tests/ui/rfcs/rfc-2565-param-attrs/issue-64682-dropping-first-attrs-in-impl-fns.rs
+tests/ui/rfcs/rfc-3348-c-string-literals/edition-spans.rs
+tests/ui/rust-2018/suggestions-not-always-applicable.rs
+tests/ui/rust-2021/reserved-prefixes-via-macro.rs
+tests/ui/underscore-imports/duplicate.rs
+tests/ui/async-await/issues/issue-60674.rs
+tests/ui/attributes/main-removed-2/main.rs
+tests/ui/cfg/assume-incomplete-release/assume-incomplete.rs
+tests/ui/crate-loading/cross-compiled-proc-macro.rs
+tests/ui/derives/derive-marker-tricky.rs
+tests/ui/diagnostic_namespace/existing_proc_macros.rs
+tests/ui/fmt/format-args-capture-issue-106408.rs
+tests/ui/fmt/indoc-issue-106408.rs
+tests/ui/hygiene/issue-77523-def-site-async-await.rs
+tests/ui/inherent-impls-overlap-check/no-overlap.rs

failing-lto-tests.txt

@@ -0,0 +1,11 @@
+tests/ui/issues/issue-44056.rs
+tests/ui/lto/fat-lto.rs
+tests/ui/lto/debuginfo-lto.rs
+tests/ui/lto/lto-many-codegen-units.rs
+tests/ui/lto/issue-100772.rs
+tests/ui/lto/lto-rustc-loads-linker-plugin.rs
+tests/ui/panic-runtime/lto-unwind.rs
+tests/ui/sanitize/issue-111184-generator-witness.rs
+tests/ui/sepcomp/sepcomp-lib-lto.rs
+tests/ui/lto/lto-opt-level-s.rs
+tests/ui/lto/lto-opt-level-z.rs

failing-non-lto-tests.txt

@@ -1,11 +1,5 @@
-tests/ui/allocator/custom-in-block.rs
-tests/ui/allocator/custom-in-submodule.rs
-tests/ui/allocator/custom.rs
-tests/ui/allocator/hygiene.rs
 tests/ui/allocator/no_std-alloc-error-handler-custom.rs
 tests/ui/allocator/no_std-alloc-error-handler-default.rs
-tests/ui/allocator/xcrate-use.rs
-tests/ui/allocator/xcrate-use2.rs
 tests/ui/asm/may_unwind.rs
 tests/ui/asm/x86_64/multiple-clobber-abi.rs
 tests/ui/debuginfo/debuginfo-emit-llvm-ir-and-split-debuginfo.rs
@@ -14,7 +8,6 @@ tests/ui/linkage-attr/linkage1.rs
 tests/ui/lto/dylib-works.rs
 tests/ui/numbers-arithmetic/saturating-float-casts.rs
 tests/ui/polymorphization/promoted-function.rs
-tests/ui/process/nofile-limit.rs
 tests/ui/sepcomp/sepcomp-cci.rs
 tests/ui/sepcomp/sepcomp-extern.rs
 tests/ui/sepcomp/sepcomp-fns-backwards.rs
@@ -53,8 +46,8 @@ tests/ui/issues/issue-40883.rs
 tests/ui/issues/issue-43853.rs
 tests/ui/issues/issue-47364.rs
 tests/ui/macros/rfc-2011-nicer-assert-messages/assert-without-captures-does-not-create-unnecessary-code.rs
-tests/ui/rfcs/rfc-2091-track-caller/std-panic-locations.rs
 tests/ui/rfcs/rfc-1857-stabilize-drop-order/drop-order.rs
+tests/ui/rfcs/rfc-2091-track-caller/std-panic-locations.rs
 tests/ui/simd/issue-17170.rs
 tests/ui/simd/issue-39720.rs
 tests/ui/simd/issue-89193.rs
@@ -64,9 +57,15 @@ tests/ui/alloc-error/default-alloc-error-hook.rs
 tests/ui/generator/panic-safe.rs
 tests/ui/issues/issue-14875.rs
 tests/ui/issues/issue-29948.rs
-tests/ui/panic-while-printing.rs
-tests/ui/enum-discriminant/get_discr.rs
 tests/ui/panics/nested_panic_caught.rs
 tests/ui/simd/intrinsic/generic-bswap-byte.rs
 tests/ui/const_prop/ice-issue-111353.rs
 tests/ui/process/println-with-broken-pipe.rs
+tests/ui/panic-runtime/lto-abort.rs
+tests/ui/lto/thin-lto-inlines2.rs
+tests/ui/lto/weak-works.rs
+tests/ui/lto/thin-lto-inlines.rs
+tests/ui/lto/thin-lto-global-allocator.rs
+tests/ui/lto/msvc-imp-present.rs
+tests/ui/lto/lto-thin-rustc-loads-linker-plugin.rs
+tests/ui/lto/all-crates.rs