Update to LLVM 10 by nikic · Pull Request #67759 · rust-lang/rust

nikic · 2019-12-31T14:59:38Z

LLVM 10 is going to be branched soon, so it's a good time to start finding all those tasty new miscompiles and performance regressions ;)

Status:

Preparation split off into Prepare for LLVM 10 upgrade #67900.
Optimization regressions:
- https://bugs.llvm.org/show_bug.cgi?id=44419 => https://reviews.llvm.org/D72048 has landed.
- https://bugs.llvm.org/show_bug.cgi?id=44423 => https://reviews.llvm.org/D72060 has landed.
- https://reviews.llvm.org/D72169 submitted.
- https://bugs.llvm.org/show_bug.cgi?id=44461 reported. https://reviews.llvm.org/D72420 submitted, but unlikely eligible for LLVM 10.
Compile-time regressions:
- GlobalOpt regression identified. ~~fhahn proposed https://reviews.llvm.org/D72214.~~ fhahn has reverted the patch.
- Even with the revert, there are large regressions.
Assertion failures / infinite loops:
- https://bugs.llvm.org/show_bug.cgi?id=44600 => https://reviews.llvm.org/D73135, https://reviews.llvm.org/D73854 and https://reviews.llvm.org/D73908 have landed and been cherry-picked to the 10.x branch.
- https://bugs.llvm.org/show_bug.cgi?id=44835 => https://reviews.llvm.org/D74278 has landed and been cherry-picked.

rust-highfive · 2019-12-31T15:20:49Z

The job x86_64-gnu-llvm-7 of your PR failed (pretty log, raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

2019-12-31T15:00:13.5244805Z ##[command]git remote add origin https://github.com/rust-lang/rust
2019-12-31T15:00:14.1661198Z ##[command]git config gc.auto 0
2019-12-31T15:00:14.1666994Z ##[command]git config --get-all http.https://github.com/rust-lang/rust.extraheader
2019-12-31T15:00:14.1671697Z ##[command]git config --get-all http.proxy
2019-12-31T15:00:14.1677709Z ##[command]git -c http.extraheader="AUTHORIZATION: basic ***" fetch --force --tags --prune --progress --no-recurse-submodules --depth=2 origin +refs/heads/*:refs/remotes/origin/* +refs/pull/67759/merge:refs/remotes/pull/67759/merge
---
2019-12-31T15:07:48.9476380Z Compiling memoffset v0.5.1
2019-12-31T15:07:49.9020375Z error: failed to run custom build command for `rustc_llvm v0.0.0 (/checkout/src/librustc_llvm)`
2019-12-31T15:07:49.9032390Z 2019-12-31T15:07:49.9032496Z Caused by:
2019-12-31T15:07:49.9033044Z process didn't exit successfully: `/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-rustc/release/build/rustc_llvm-12ceb52d5dc2c5b7/build-script-build` (exit code: 1)
2019-12-31T15:07:49.9033287Z --- stdout
2019-12-31T15:07:49.9033654Z cargo:rerun-if-env-changed=REAL_LIBRARY_PATH_VAR
2019-12-31T15:07:49.9033853Z cargo:rerun-if-env-changed=REAL_LIBRARY_PATH
2019-12-31T15:07:49.9034087Z cargo:rerun-if-changed=/usr/lib/llvm-7/bin/llvm-config
2019-12-31T15:07:49.9034275Z cargo:rerun-if-env-changed=LLVM_CONFIG
2019-12-31T15:07:49.9034616Z cargo:rustc-cfg=llvm_component="aarch64"
2019-12-31T15:07:49.9034969Z cargo:rustc-cfg=llvm_component="amdgpu"
2019-12-31T15:07:49.9035492Z cargo:rustc-cfg=llvm_component="arm"
2019-12-31T15:07:49.9035914Z cargo:rustc-cfg=llvm_component="asmparser"
2019-12-31T15:07:49.9036624Z cargo:rustc-cfg=llvm_component="bitreader"
2019-12-31T15:07:49.9036814Z cargo:rustc-cfg=llvm_component="bitwriter"
2019-12-31T15:07:49.9037179Z cargo:rustc-cfg=llvm_component="hexagon"
2019-12-31T15:07:49.9037729Z cargo:rustc-cfg=llvm_component="instrumentation"
2019-12-31T15:07:49.9038138Z cargo:rustc-cfg=llvm_component="ipo"
2019-12-31T15:07:49.9038510Z cargo:rustc-cfg=llvm_component="linker"
2019-12-31T15:07:49.9038868Z cargo:rustc-cfg=llvm_component="lto"
2019-12-31T15:07:49.9039205Z cargo:rustc-cfg=llvm_component="mips"
2019-12-31T15:07:49.9039567Z cargo:rustc-cfg=llvm_component="msp430"
2019-12-31T15:07:49.9039905Z cargo:rustc-cfg=llvm_component="nvptx"
2019-12-31T15:07:49.9040241Z cargo:rustc-cfg=llvm_component="powerpc"
2019-12-31T15:07:49.9040605Z cargo:rustc-cfg=llvm_component="sparc"
2019-12-31T15:07:49.9040944Z cargo:rustc-cfg=llvm_component="systemz"
2019-12-31T15:07:49.9041442Z cargo:rustc-cfg=llvm_component="webassembly"
2019-12-31T15:07:49.9041968Z cargo:rustc-cfg=llvm_component="x86"
2019-12-31T15:07:49.9042296Z cargo:rerun-if-changed-env=LLVM_RUSTLLVM
2019-12-31T15:07:49.9042650Z cargo:rerun-if-changed=../rustllvm/ArchiveWrapper.cpp
2019-12-31T15:07:49.9042998Z cargo:rerun-if-changed=../rustllvm/.editorconfig
2019-12-31T15:07:49.9043339Z cargo:rerun-if-changed=../rustllvm/PassWrapper.cpp
2019-12-31T15:07:49.9043696Z cargo:rerun-if-changed=../rustllvm/Linker.cpp
2019-12-31T15:07:49.9044378Z cargo:rerun-if-changed=../rustllvm/rustllvm.h
2019-12-31T15:07:49.9044717Z cargo:rerun-if-changed=../rustllvm/README
2019-12-31T15:07:49.9045335Z cargo:rerun-if-changed=../rustllvm/RustWrapper.cpp
2019-12-31T15:07:49.9045691Z TARGET = Some("x86_64-unknown-linux-gnu")
2019-12-31T15:07:49.9045872Z OPT_LEVEL = Some("2")
2019-12-31T15:07:49.9046141Z HOST = Some("x86_64-unknown-linux-gnu")
2019-12-31T15:07:49.9046336Z CXX_x86_64-unknown-linux-gnu = Some("sccache c++")
2019-12-31T15:07:49.9046578Z CXXFLAGS_x86_64-unknown-linux-gnu = Some("-ffunction-sections -fdata-sections -fPIC -m64")
2019-12-31T15:07:49.9046625Z CRATE_CC_NO_DEFAULTS = None
2019-12-31T15:07:49.9046666Z DEBUG = Some("false")
2019-12-31T15:07:49.9046722Z CARGO_CFG_TARGET_FEATURE = Some("fxsr,mmx,sse,sse2")
2019-12-31T15:07:49.9048439Z running: "sccache" "sccache" "c++" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-I/usr/lib/llvm-7/include" "-std=c++0x" "-fuse-ld=gold" "-Wl,--no-keep-files-mapped" "-Wl,--no-map-whole-files" "-fPIC" "-fvisibility-inlines-hidden" "-Werror=date-time" "-std=c++11" "-Wall" "-Wextra" "-Wno-unused-parameter" "-Wwrite-strings" "-Wcast-qual" "-Wno-missing-field-initializers" "-pedantic" "-Wno-long-long" "-Wno-maybe-uninitialized" "-Wdelete-non-virtual-dtor" "-Wno-comment" "-ffunction-sections" "-fdata-sections" "-O2" "-DNDEBUG" "-fno-exceptions" "-D_GNU_SOURCE" "-D__STDC_CONSTANT_MACROS" "-D__STDC_FORMAT_MACROS" "-D__STDC_LIMIT_MACROS" "-DLLVM_COMPONENT_AARCH64" "-DLLVM_COMPONENT_AMDGPU" "-DLLVM_COMPONENT_ARM" "-DLLVM_COMPONENT_ASMPARSER" "-DLLVM_COMPONENT_BITREADER" "-DLLVM_COMPONENT_BITWRITER" "-DLLVM_COMPONENT_HEXAGON" "-DLLVM_COMPONENT_INSTRUMENTATION" "-DLLVM_COMPONENT_IPO" "-DLLVM_COMPONENT_LINKER" "-DLLVM_COMPONENT_LTO" "-DLLVM_COMPONENT_MIPS" "-DLLVM_COMPONENT_MSP430" "-DLLVM_COMPONENT_NVPTX" "-DLLVM_COMPONENT_POWERPC" "-DLLVM_COMPONENT_SPARC" "-DLLVM_COMPONENT_SYSTEMZ" "-DLLVM_COMPONENT_WEBASSEMBLY" "-DLLVM_COMPONENT_X86" "-DNDEBUG" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/rustc_llvm-20b5b3b7a5d2cc79/out/../rustllvm/PassWrapper.o" "-c" "../rustllvm/PassWrapper.cpp"
2019-12-31T15:07:49.9049232Z cargo:warning=../rustllvm/PassWrapper.cpp: In function 'LLVMRustThinLTOData* LLVMRustCreateThinLTOData(LLVMRustThinLTOModule*, int, const char**, int)':
2019-12-31T15:07:49.9049623Z cargo:warning=../rustllvm/PassWrapper.cpp:866:19: error: 'make_unique' is not a member of 'std'
2019-12-31T15:07:49.9049810Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOData>();
2019-12-31T15:07:49.9049886Z cargo:warning= ^~~~~~~~~~~
2019-12-31T15:07:49.9050168Z cargo:warning=../rustllvm/PassWrapper.cpp:866:19: note: suggested alternative: '__unique'
2019-12-31T15:07:49.9050452Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOData>();
2019-12-31T15:07:49.9050543Z cargo:warning= ^~~~~~~~~~~
2019-12-31T15:07:49.9050602Z cargo:warning= __unique
2019-12-31T15:07:49.9050959Z cargo:warning=../rustllvm/PassWrapper.cpp:866:50: error: expected primary-expression before '>' token
2019-12-31T15:07:49.9051140Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOData>();
2019-12-31T15:07:49.9051821Z cargo:warning= ^
2019-12-31T15:07:49.9052189Z cargo:warning=../rustllvm/PassWrapper.cpp:866:52: error: expected primary-expression before ')' token
2019-12-31T15:07:49.9052365Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOData>();
2019-12-31T15:07:49.9052461Z cargo:warning= ^
2019-12-31T15:07:49.9052780Z cargo:warning=../rustllvm/PassWrapper.cpp: In function 'LLVMRustThinLTOBuffer* LLVMRustThinLTOBufferCreate(LLVMModuleRef)':
2019-12-31T15:07:49.9053209Z cargo:warning=../rustllvm/PassWrapper.cpp:1108:19: error: 'make_unique' is not a member of 'std'
2019-12-31T15:07:49.9053555Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOBuffer>();
2019-12-31T15:07:49.9053845Z cargo:warning= ^~~~~~~~~~~
2019-12-31T15:07:49.9054388Z cargo:warning=../rustllvm/PassWrapper.cpp:1108:19: note: suggested alternative: '__unique'
2019-12-31T15:07:49.9054442Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOBuffer>();
2019-12-31T15:07:49.9054634Z cargo:warning= ^~~~~~~~~~~
2019-12-31T15:07:49.9054713Z cargo:warning= __unique
2019-12-31T15:07:49.9055027Z cargo:warning=../rustllvm/PassWrapper.cpp:1108:52: error: expected primary-expression before '>' token
2019-12-31T15:07:49.9055229Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOBuffer>();
2019-12-31T15:07:49.9055309Z cargo:warning= ^
2019-12-31T15:07:49.9055655Z cargo:warning=../rustllvm/PassWrapper.cpp:1108:54: error: expected primary-expression before ')' token
2019-12-31T15:07:49.9055856Z cargo:warning= auto Ret = std::make_unique<LLVMRustThinLTOBuffer>();
2019-12-31T15:07:49.9055947Z cargo:warning= ^
2019-12-31T15:07:49.9056032Z cargo:warning=../rustllvm/PassWrapper.cpp: At global scope:
2019-12-31T15:07:49.9056455Z cargo:warning=../rustllvm/PassWrapper.cpp:838:1: warning: 'const llvm::GlobalValueSummary* getFirstDefinitionForLinker(const GlobalValueSummaryList&)' defined but not used [-Wunused-function]
2019-12-31T15:07:49.9056690Z cargo:warning= getFirstDefinitionForLinker(const GlobalValueSummaryList &GVSummaryList) {
2019-12-31T15:07:49.9057032Z cargo:warning= ^~~~~~~~~~~~~~~~~~~~~~~~~~~
2019-12-31T15:07:49.9057149Z 2019-12-31T15:07:49.9057374Z --- stderr
2019-12-31T15:07:49.9057401Z 2019-12-31T15:07:49.9058329Z 2019-12-31T15:07:49.9058329Z 2019-12-31T15:07:49.9060058Z error occurred: Command "sccache" "sccache" "c++" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-I/usr/lib/llvm-7/include" "-std=c++0x" "-fuse-ld=gold" "-Wl,--no-keep-files-mapped" "-Wl,--no-map-whole-files" "-fPIC" "-fvisibility-inlines-hidden" "-Werror=date-time" "-std=c++11" "-Wall" "-Wextra" "-Wno-unused-parameter" "-Wwrite-strings" "-Wcast-qual" "-Wno-missing-field-initializers" "-pedantic" "-Wno-long-long" "-Wno-maybe-uninitialized" "-Wdelete-non-virtual-dtor" "-Wno-comment" "-ffunction-sections" "-fdata-sections" "-O2" "-DNDEBUG" "-fno-exceptions" "-D_GNU_SOURCE" "-D__STDC_CONSTANT_MACROS" "-D__STDC_FORMAT_MACROS" "-D__STDC_LIMIT_MACROS" "-DLLVM_COMPONENT_AARCH64" "-DLLVM_COMPONENT_AMDGPU" "-DLLVM_COMPONENT_ARM" "-DLLVM_COMPONENT_ASMPARSER" "-DLLVM_COMPONENT_BITREADER" "-DLLVM_COMPONENT_BITWRITER" "-DLLVM_COMPONENT_HEXAGON" "-DLLVM_COMPONENT_INSTRUMENTATION" "-DLLVM_COMPONENT_IPO" "-DLLVM_COMPONENT_LINKER" "-DLLVM_COMPONENT_LTO" "-DLLVM_COMPONENT_MIPS" "-DLLVM_COMPONENT_MSP430" "-DLLVM_COMPONENT_NVPTX" "-DLLVM_COMPONENT_POWERPC" "-DLLVM_COMPONENT_SPARC" "-DLLVM_COMPONENT_SYSTEMZ" "-DLLVM_COMPONENT_WEBASSEMBLY" "-DLLVM_COMPONENT_X86" "-DNDEBUG" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/rustc_llvm-20b5b3b7a5d2cc79/out/../rustllvm/PassWrapper.o" "-c" "../rustllvm/PassWrapper.cpp" with args "c++" did not execute successfully (status code exit code: 1).
2019-12-31T15:07:49.9060298Z 2019-12-31T15:07:49.9060320Z 2019-12-31T15:07:49.9061220Z warning: build failed, waiting for other jobs to finish...
2019-12-31T15:07:50.0699244Z error: build failed
---
2019-12-31T15:07:50.8023825Z local time: Tue Dec 31 15:07:50 UTC 2019
2019-12-31T15:07:50.8023889Z network time: Tue, 31 Dec 2019 15:07:50 GMT
2019-12-31T15:07:50.8023940Z == end clock drift check ==
2019-12-31T15:07:52.2523357Z 2019-12-31T15:07:52.2583057Z ##[error]Bash exited with code '1'.
2019-12-31T15:07:52.2606973Z ##[section]Starting: Checkout
2019-12-31T15:07:52.2608333Z ==============================================================================
2019-12-31T15:07:52.2608377Z Task : Get sources
2019-12-31T15:07:52.2608427Z Description : Get sources from a repository. Supports Git, TfsVC, and SVN repositories.

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

nikic · 2019-12-31T17:13:13Z

First optimization regression, eventually leading to failure to eliminate bounds checks: https://bugs.llvm.org/show_bug.cgi?id=44419

nikic · 2020-01-01T17:08:19Z

Second optimization regression: https://bugs.llvm.org/show_bug.cgi?id=44423

nikic · 2020-01-02T21:57:38Z

src/test/codegen/issue-45222.rs shows a regression in optimization of the notoriously fickle RangeInclusive.

While previously

bb10: ; preds = %bb3.i.i, %bb6
%iter1.sroa.5.0 = phi i64 [ 100000, %bb6 ], [ %spec.select, %bb3.i.i ]
%iter1.sroa.9.0 = phi i8 [ 2, %bb6 ], [ %6, %bb3.i.i ]
%count.1 = phi i64 [ %count.0, %bb6 ], [ %8, %bb3.i.i ]
%2 = icmp eq i8 %iter1.sroa.9.0, 2
%3 = and i8 %iter1.sroa.9.0, 1
%phitmp.i.i.i = icmp eq i8 %3, 0
%or.cond = or i1 %2, %phitmp.i.i.i
br i1 %or.cond, label %bb3.i.i, label %bb2.loopexit
bb3.i.i: ; preds = %bb10
%4 = icmp ne i64 %iter1.sroa.5.0, 0
%5 = xor i1 %4, true
%6 = zext i1 %5 to i8
%7 = add i64 %iter1.sroa.5.0, -1
%spec.select = select i1 %4, i64 %7, i64 0
%8 = add i64 %count.1, %iter1.sroa.5.0
br label %bb10

was rotated to

bb3.i.i: ; preds = %bb6, %bb3.i.i
%count.159 = phi i64 [ %count.0, %bb6 ], [ %6, %bb3.i.i ]
%iter1.sroa.5.058 = phi i64 [ 100000, %bb6 ], [ %spec.select, %bb3.i.i ]
%2 = icmp ne i64 %iter1.sroa.5.058, 0
%3 = xor i1 %2, true
%4 = zext i1 %3 to i8
%5 = add i64 %iter1.sroa.5.058, -1
%spec.select = select i1 %2, i64 %5, i64 0
%6 = add i64 %count.159, %iter1.sroa.5.058
%7 = icmp eq i8 %4, 2
%8 = and i8 %4, 1
%phitmp.i.i.i = icmp eq i8 %8, 0
%or.cond = or i1 %7, %phitmp.i.i.i
br i1 %or.cond, label %bb3.i.i, label %bb2.loopexit

we now get

bb10: ; preds = %bb3.i.i, %bb6
%iter1.sroa.5.0 = phi i64 [ 100000, %bb6 ], [ %spec.select, %bb3.i.i ]
%iter1.sroa.9.0 = phi i8 [ 2, %bb6 ], [ %3, %bb3.i.i ]
%count.1 = phi i64 [ %count.0, %bb6 ], [ %4, %bb3.i.i ]
switch i8 %iter1.sroa.9.0, label %bb2.loopexit [
i8 2, label %bb3.i.i
i8 0, label %bb3.i.i
]
bb3.i.i: ; preds = %bb10, %bb10
%2 = icmp ne i64 %iter1.sroa.5.0, 0
%_10.i.i = xor i1 %2, true
%3 = zext i1 %_10.i.i to i8
%_5.0.i.i.i.i = add i64 %iter1.sroa.5.0, -1
%spec.select = select i1 %2, i64 %_5.0.i.i.i.i, i64 0
%4 = add i64 %count.1, %iter1.sroa.5.0
br label %bb10

which does not get rotated, and thus not simplified subsequently.

The reason why we get a switch here (created by SimplifyCFG) is that CVP eliminated a masking operation, namely instruction %5 in https://llvm.godbolt.org/z/cCuRtG. The original optimization does get applied at -C opt-level=3 -- ironically, the reason is that argument promotion results in a loss of !range metadata, preventing the CVP optimization...

I'm not yet sure how to address this issue.

nikic · 2020-01-02T22:27:18Z

Actually, CVP might be both the problem and the solution here ... if we relax https://github.com/llvm/llvm-project/blob/a58da1a2ff039dd3bb4c43db3919995cf4a74cc7/llvm/lib/Transforms/Scalar/CorrelatedValuePropagation.cpp#L315 to allow phis, then condition %4 can be folded as well, which will likely avoid the followup issues.

nikic · 2020-01-03T22:22:31Z

@bors try @rust-timer queue

rust-timer · 2020-01-03T22:22:32Z

Awaiting bors try build completion

bors · 2020-01-03T22:22:44Z

⌛ Trying commit d6773bf with merge d5ca0b366ee2e90c658ca0334740b4907ac1ce69...

rust-highfive · 2020-01-03T22:41:32Z

The job dist-x86_64-linux-alt of your PR failed (pretty log, raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

2020-01-03T22:40:54.9106810Z Compiling rustc-std-workspace-core v1.99.0 (/checkout/src/tools/rustc-std-workspace-core)
2020-01-03T22:41:01.7062662Z error: failed to run custom build command for `profiler_builtins v0.0.0 (/checkout/src/libprofiler_builtins)`
2020-01-03T22:41:01.7062812Z 2020-01-03T22:41:01.7062911Z Caused by:
2020-01-03T22:41:01.7063662Z process didn't exit successfully: `/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-std/release/build/profiler_builtins-79d44dd7732723e3/build-script-build` (exit code: 1)
2020-01-03T22:41:01.7064278Z --- stdout
2020-01-03T22:41:01.7064543Z TARGET = Some("x86_64-unknown-linux-gnu")
2020-01-03T22:41:01.7064615Z OPT_LEVEL = Some("2")
2020-01-03T22:41:01.7064857Z HOST = Some("x86_64-unknown-linux-gnu")
2020-01-03T22:41:01.7065439Z CC_x86_64-unknown-linux-gnu = Some("sccache clang")
2020-01-03T22:41:01.7066383Z CFLAGS_x86_64-unknown-linux-gnu = Some("-ffunction-sections -fdata-sections -fPIC --target=x86_64-unknown-linux-gnu -fdebug-prefix-map=/checkout=/rustc/d5ca0b366ee2e90c658ca0334740b4907ac1ce69")
2020-01-03T22:41:01.7066528Z CRATE_CC_NO_DEFAULTS = None
2020-01-03T22:41:01.7066600Z DEBUG = Some("true")
2020-01-03T22:41:01.7067755Z running: "sccache" "sccache" "clang" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-g" "-fno-omit-frame-pointer" "--target=x86_64-unknown-linux-gnu" "-ffunction-sections" "-fdata-sections" "-fPIC" "--target=x86_64-unknown-linux-gnu" "-fdebug-prefix-map=/checkout=/rustc/d5ca0b366ee2e90c658ca0334740b4907ac1ce69" "-fno-builtin" "-fvisibility=hidden" "-fomit-frame-pointer" "-ffreestanding" "-DVISIBILITY_HIDDEN" "-DCOMPILER_RT_HAS_UNAME=1" "-DCOMPILER_RT_HAS_FCNTL_LCK=1" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-std/x86_64-unknown-linux-gnu/release/build/profiler_builtins-3fff06880a6b0050/out/GCDAProfiling.o" "-c" "/checkout/src/llvm-project/compiler-rt/lib/profile/GCDAProfiling.c"
2020-01-03T22:41:01.7068350Z cargo:warning=In file included from /checkout/src/llvm-project/compiler-rt/lib/profile/GCDAProfiling.c:62:
2020-01-03T22:41:01.7068756Z cargo:warning=/checkout/src/llvm-project/compiler-rt/lib/profile/InstrProfiling.h:16:10: fatal error: 'profile/InstrProfData.inc' file not found
2020-01-03T22:41:01.7068907Z cargo:warning=#include "profile/InstrProfData.inc"
2020-01-03T22:41:01.7069008Z cargo:warning= ^~~~~~~~~~~~~~~~~~~~~~~~~~~
2020-01-03T22:41:01.7069083Z cargo:warning=1 error generated.
2020-01-03T22:41:01.7069361Z 2020-01-03T22:41:01.7069562Z --- stderr
2020-01-03T22:41:01.7069596Z 2020-01-03T22:41:01.7069627Z 2020-01-03T22:41:01.7069627Z 2020-01-03T22:41:01.7071484Z error occurred: Command "sccache" "sccache" "clang" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-g" "-fno-omit-frame-pointer" "--target=x86_64-unknown-linux-gnu" "-ffunction-sections" "-fdata-sections" "-fPIC" "--target=x86_64-unknown-linux-gnu" "-fdebug-prefix-map=/checkout=/rustc/d5ca0b366ee2e90c658ca0334740b4907ac1ce69" "-fno-builtin" "-fvisibility=hidden" "-fomit-frame-pointer" "-ffreestanding" "-DVISIBILITY_HIDDEN" "-DCOMPILER_RT_HAS_UNAME=1" "-DCOMPILER_RT_HAS_FCNTL_LCK=1" "-o" "/checkout/obj/build/x86_64-unknown-linux-gnu/stage0-std/x86_64-unknown-linux-gnu/release/build/profiler_builtins-3fff06880a6b0050/out/GCDAProfiling.o" "-c" "/checkout/src/llvm-project/compiler-rt/lib/profile/GCDAProfiling.c" with args "clang" did not execute successfully (status code exit code: 1).
2020-01-03T22:41:01.7072062Z 2020-01-03T22:41:01.7072113Z 2020-01-03T22:41:01.7072467Z warning: build failed, waiting for other jobs to finish...
2020-01-03T22:41:03.4272429Z error: build failed
---
2020-01-03T22:41:03.4396848Z local time: Fri Jan 3 22:41:03 UTC 2020
2020-01-03T22:41:03.9739354Z network time: Fri, 03 Jan 2020 22:41:03 GMT
2020-01-03T22:41:03.9742186Z == end clock drift check ==
2020-01-03T22:41:07.4513425Z 2020-01-03T22:41:07.4632411Z ##[error]Bash exited with code '1'.
2020-01-03T22:41:07.4674659Z ##[section]Starting: Checkout
2020-01-03T22:41:07.4677067Z ==============================================================================
2020-01-03T22:41:07.4677161Z Task : Get sources
2020-01-03T22:41:07.4677264Z Description : Get sources from a repository. Supports Git, TfsVC, and SVN repositories.

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

bors · 2020-01-03T22:41:32Z

💔 Test failed - checks-azure

nikic · 2020-01-03T23:21:46Z

@bors try

bors · 2020-01-03T23:21:57Z

⌛ Trying commit 26bf168 with merge 97588ae...

@ghost

[WIP] Update to LLVM 10 LLVM 10 is going to be branched soon, so it's a good time to start finding all those tasty new miscompiles and performance regressions ;) r? @ghost

bors · 2020-01-04T03:16:08Z

☀️ Try build successful - checks-azure
Build commit: 97588ae (97588aeda139309169b11654fc809e1ac5fd246c)

rust-timer · 2020-01-04T03:16:09Z

Queued 97588ae with parent c5840f9, future comparison URL.

rust-timer · 2020-01-04T05:13:01Z

Finished benchmarking try commit 97588ae, comparison URL.

nikic · 2020-01-04T11:55:02Z

Those are some pretty hefty compile-time regressions. Some local numbers for the largest diffs in LLVM timings on clap-rs (via this script):

Global Variable Optimizer #2 0.018s 0.115s 0.097s
Loop Sink 0.118s 0.002s -0.116s
Function Integration/Inlining 1.278s 1.160s -0.117s
Loop Strength Reduction ------ 0.124s 0.124s
Branch Probability Analysis 0.011s 0.139s 0.128s
Memory SSA #3 ------ 0.132s 0.132s
Memory SSA #2 ------ 0.147s 0.147s
X86 DAG->DAG Instruction Selection 1.087s 1.363s 0.277s
Global Variable Optimizer 0.087s 0.506s 0.419s
Total 14.547s 16.280s 1.733s

Haven't tested how stable these numbers are, but we clearly have a big regression in "Global Variable Optimizer" (first time I hear about this pass), some regression in X86 ISel (this tends to get slower with each release, but this seems an unusually large jump) and we now have a lot more MemorySSA runs. Previously there was just one, now there are 5 (!!). Not sure if this is an analysis preservation fail or supposed to be that way.

nikic · 2020-01-04T13:52:28Z

Looks like the five MemorySSA runs are expected, as MemorySSA is now enabled for loop passes: https://reviews.llvm.org/D58311

The GlobalOpt regression is likely https://reviews.llvm.org/D68298 and hopefully easy to address with a revert.

nikic · 2020-01-04T14:28:23Z

@bors try @rust-timer queue

rust-timer · 2020-01-04T14:28:25Z

Awaiting bors try build completion

bors · 2020-01-04T14:28:34Z

⌛ Trying commit 6011237 with merge 1b6633b46cd7705ec813cfd1cd7083e92f9da903...

bors · 2020-01-04T17:15:06Z

☀️ Try build successful - checks-azure
Build commit: 1b6633b46cd7705ec813cfd1cd7083e92f9da903 (1b6633b46cd7705ec813cfd1cd7083e92f9da903)

rust-timer · 2020-01-04T17:15:08Z

Queued 1b6633b46cd7705ec813cfd1cd7083e92f9da903 with parent 79cf5e4, future comparison URL.

rust-timer · 2020-01-04T18:36:30Z

Finished benchmarking try commit 1b6633b46cd7705ec813cfd1cd7083e92f9da903, comparison URL.

nikic · 2020-01-04T19:32:23Z

Okay, that improved things a bit, reclaiming about 5% in the best case. Still pretty bad compile-time regression overall.

nikic · 2020-01-04T23:04:56Z

@bors try @rust-timer queue

rust-timer · 2020-01-04T23:04:58Z

Awaiting bors try build completion

bors · 2020-05-21T02:53:59Z

☀️ Test successful - checks-azure
Approved by: Mark-Simulacrum
Pushing 82911b3 to master...

nnethercote · 2020-05-22T00:20:59Z

Speaking as someone who has spent a lot of time and effort on rustc's speed, I find these LLVM performance regressions extremely discouraging. Especially given that opt builds are the hardest ones to speed up from the rustc side.

@nikic: many thanks for your efforts on tracking and improving LLVM's speed. I sincerely hope that the LLVM project is taking that work seriously and that things will get better in the future.

nnethercote · 2020-05-22T00:31:39Z

More generally: an LLVM update is a large change with a lot of potential for regressions, and we are less than two weeks from the end of the dev cycle for 1.45. (No Tools Breakage Week starts on May 27.) Is there any kind of policy or guidance about the timing of landing such large changes?

cuviper · 2020-05-22T00:35:40Z

@nnethercote Your efforts are not lost, even if they appear counteracted. Imagine where we would be with this update without your changes on the Rust side!

EricRahm · 2020-05-22T00:44:24Z

For what it's worth, we ended up backing out our clang-10 upgrade on the Firefox side due to the severity of the compile time (14%) and runtime performance (11%) regressions.

nikomatsakis · 2020-05-22T01:01:07Z

Sorry, I'm not clear on this, but do we have perf results from this change? It seems like we ought to check back and discuss, perhaps at the next compiler team triage meeting. A quick glance at perf.rust-lang.org didn't show anything yet, but it may not have taken into account the effects of this PR.

cc @rust-lang/wg-prioritization -- can we add this to our next agenda to check back in?

ecstatic-morse · 2020-05-22T01:02:57Z

Personally, I would be happy to spend 10% longer compiling at -C opt-level=3 if there were an improvement in runtime performance, since you can always set a lower optimization level. However, non-codegened full builds seem to be about the same for real-world crates, or in the case of keccak and inflate even slower. This means rustc isn't really benefiting from whatever additional optimizations LLVM is doing. It would be nice to know whether this is true for the ecosystem as well.

nnethercote · 2020-05-22T01:09:09Z

Final perf results from the landing are here.

Lots of red. Interestingly, check builds (which don't run the LLVM backend) are a little improved. Which suggests that the quality of the code generated by LLVM 10 is a little better, at least for rustc itself. (Edit: as @ecstatic-morse said, the gains are mostly in the short-running benchmarks; the more realistic benchmarks see little change.)

cuviper · 2020-05-22T01:58:45Z

The net results compared to 45d050c (git merge-base beta master) don't look so dire. The wild outliers make me wonder if this is really a valid comparison though.

https://perf.rust-lang.org/compare.html?start=45d050cde277b22a755847338f2acc2c7b834141&end=82911b3bba76e73afe2881b732fe6b0edb35d5d3&stat=instructions%3Au

nnethercote · 2020-05-22T02:21:02Z

The script-servo-opt-incr-patched outlier is a known problem where slight changes in codegen boundaries have major effects on that benchmark. See rust-lang/compiler-team#281 for more details.

The net results compared to 45d050c (git merge-base beta master) don't look so dire.

Is that comparing the nightly against the most recent beta? It won't look so dire because we've made a bunch of speedups recently, particularly for debug builds. If you filter down to opt results only, things look a lot worse.

And compare it against this, which is the comparison of that beta against the revision just before the LLVM 10 upgrade. Again, filter down to opt only and we were doing great there.

Or just look at the individual opt benchmark graphs on perf.rust-lang.org, it's pretty clear there. (Ignore the summary, that's misleading because script-servo utterly dominates it because it's by far the longest-running benchmark.)

cuviper · 2020-05-22T02:28:49Z

I don't doubt at all that this PR slowed things down. I was trying to see what the overall story from 1.44 to 1.45 will look like, in spite of this.

nnethercote · 2020-06-01T23:53:49Z

Sorry, I'm not clear on this, but do we have perf results from this change? It seems like we ought to check back and discuss, perhaps at the next compiler team triage meeting. A quick glance at perf.rust-lang.org didn't show anything yet, but it may not have taken into account the effects of this PR.
cc @rust-lang/wg-prioritization -- can we add this to our next agenda to check back in?

@nikomatsakis: Any update here? The current dev cycle is almost over and the upgrade to LLVM 10 is going to cause a sizeable perf regression without any upside that I'm aware of.

cuviper · 2020-06-01T23:59:52Z

There's a maintenance upside in bugs that may be fixed in LLVM 10, without Rust having to encounter them and try to backport fixes. For instance, I just got done dealing with an LLVM 9 miscompilation in s390x, rhbz1837660, which was thankfully already fixed in LLVM 10. Of course there's a risk of LLVM regressions too, even apart from performance, but let's hope for forward progress.

tmandry · 2020-06-02T00:24:51Z

Personally I think our stance in general should be to accept slower opt builds, especially if it results in better code generation (anything else is swimming against the tide for LLVM). For major regressions and regressions to debug builds, we ideally should report to upstream and work with them to identify the cause.

For this change specifically, most regressions are for opt builds, but there are some for debug builds. But sticking to a >6mo old version of LLVM is going to cause a lot more pain than it would solve at this point, IMO. Conversely, rolling to new versions of LLVM more frequently makes it far easier to catch and identify regressions like this.

See: https://reviews.llvm.org/D70779 https://reviews.llvm.org/D70779#C1703593NL568 LLVM 10 merged into master at: rust-lang#67759

bors added the S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. label Jan 3, 2020

bors added a commit that referenced this pull request Jan 3, 2020
Auto merge of#67759- nikic:llvm-10, r=<try>
97588ae
[WIP] Update to LLVM 10 LLVM 10 is going to be branched soon, so it's a good time to start finding all those tasty new miscompiles and performance regressions ;) r? @ghost

tmandry mentioned this pull request May 21, 2020
Update the Fuchsia linker defaults #72116
Merged

bors added the merged-by-bors This PR was explicitly merged by bors. label May 21, 2020

bors merged commit 82911b3 into rust-lang:masterMay 21, 2020

This was referenced May 25, 2020
libunwind changes needed to run code in sgx environment via rust-sgx. rust-lang/llvm-project#57
Merged
x86 LVI hardening rust-lang/llvm-project#58
Merged

josephlr mentioned this pull request Jun 1, 2020
Missing codegen option corresponding to clang's prefer-vector-width #53312
Closed

trevyn added a commit to trevyn/rust that referenced this pull request Jun 7, 2020
Rename "cyclone" to "apple-a7" per changes in upstream LLVM
6f6620b
See: https://reviews.llvm.org/D70779 https://reviews.llvm.org/D70779#C1703593NL568 LLVM 10 merged into master at: rust-lang#67759

trevyn mentioned this pull request Jun 7, 2020
Rename "cyclone" to "apple-a7" per changes in upstream LLVM #73086
Merged

cuviper mentioned this pull request Jun 10, 2020
llvm-objcopy produces 160MB binary. It used to be 74KB #73201
Closed

aquarhead mentioned this pull request Jul 23, 2020
Upgrade to LLVM 10 foniod/redbpf#66
Merged

dtolnay mentioned this pull request Jul 24, 2020
Tracking issue for new LLVM pass manager (-Znew-llvm-pass-manager) #74705
Closed
4 tasks

This was referenced Aug 1, 2020
ELF output contains lots of zero data: linker script issue? rust-embedded/riscv-rt#60
Closed
Rust binary (after objcopy) too large icebreaker-fpga/icebreaker-litex-examples#6
Open

viktorchvatal mentioned this pull request Sep 3, 2020
Large performance drop in compiled binary in stable rust 1.45.2 vs 1.44.0 on x86_64 linux #76247
Open

eggyal mentioned this pull request Apr 6, 2021
Moving all elements out of a Vec generates worse assembly since 1.45 #78918
Open

cuviper mentioned this pull request Mar 29, 2024
Rust 1.45+ generates suboptimal control flow for ARMv7 atomic compare exchange #123113
Open

Uh oh!

Uh oh!

Conversation

nikic commented Dec 31, 2019• edited by Mark-Simulacrum Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rust-highfive commented Dec 31, 2019

Uh oh!

nikic commented Dec 31, 2019

Uh oh!

nikic commented Jan 1, 2020

Uh oh!

nikic commented Jan 2, 2020

Uh oh!

nikic commented Jan 2, 2020

Uh oh!

nikic commented Jan 3, 2020

Uh oh!

rust-timer commented Jan 3, 2020

Uh oh!

bors commented Jan 3, 2020

Uh oh!

rust-highfive commented Jan 3, 2020

Uh oh!

bors commented Jan 3, 2020

Uh oh!

nikic commented Jan 3, 2020

Uh oh!

bors commented Jan 3, 2020

Uh oh!

bors commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

nikic commented Jan 4, 2020

Uh oh!

nikic commented Jan 4, 2020

Uh oh!

nikic commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

bors commented Jan 4, 2020

Uh oh!

bors commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

nikic commented Jan 4, 2020

Uh oh!

nikic commented Jan 4, 2020

Uh oh!

rust-timer commented Jan 4, 2020

Uh oh!

bors commented May 21, 2020

Uh oh!

nnethercote commented May 22, 2020

Uh oh!

nnethercote commented May 22, 2020

Uh oh!

cuviper commented May 22, 2020

Uh oh!

EricRahm commented May 22, 2020

Uh oh!

nikomatsakis commented May 22, 2020

Uh oh!

ecstatic-morse commented May 22, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nnethercote commented May 22, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cuviper commented May 22, 2020

Uh oh!

nnethercote commented May 22, 2020

Uh oh!

nikic commented Dec 31, 2019•
edited by Mark-Simulacrum
Loading

ecstatic-morse commented May 22, 2020•
edited
Loading

nnethercote commented May 22, 2020•
edited
Loading

cuviper commented Jun 1, 2020•
edited
Loading