-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Excessive compilation time (regression) #45395
Comments
Link to discourse thread: https://discourse.julialang.org/t/excessive-compilation-time/81114 |
All of the time here is spent in LLVM optimization and the passes that take the most time are:
|
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `export PATH=/some/recent/clang/bin:$PATH` 2. `make -C contrib/pgo-lto -j$(nproc) stage1` 3. `rm -rf profiles` 4. `./julia -O3 -e 'using Pkg; Pkg.add("Unitful"); Pkg.test("Unitful")'` 5. `make -C contrib/pgo-lto -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways.
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `export PATH=/some/recent/clang/bin:$PATH` 2. `make -C contrib/pgo-lto -j$(nproc) stage1` 3. `rm -rf profiles` 4. `./julia -O3 -e 'using Pkg; Pkg.add("Unitful"); Pkg.test("Unitful")'` 5. `make -C contrib/pgo-lto -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways.
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `cd contrib/pgo-lto` 2. `make -j$(nproc) stage1` 3. `make clean-profiles` 4. `./stage1.build/julia -O3 -e 'using Pkg; Pkg.add("LoopVectorization"); Pkg.test("LoopVectorization")'` 5. `make -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways. Turn into makefile Newline Use two out of source builds Ignore profiles + build dirs Add --icf=safe stage0 setup prebuilt clang with [cd]tors->init/fini patch
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `cd contrib/pgo-lto` 2. `make -j$(nproc) stage1` 3. `make clean-profiles` 4. `./stage1.build/julia -O3 -e 'using Pkg; Pkg.add("LoopVectorization"); Pkg.test("LoopVectorization")'` 5. `make -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways. Turn into makefile Newline Use two out of source builds Ignore profiles + build dirs Add --icf=safe stage0 setup prebuilt clang with [cd]tors->init/fini patch
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `cd contrib/pgo-lto` 2. `make -j$(nproc) stage1` 3. `make clean-profiles` 4. `./stage1.build/julia -O3 -e 'using Pkg; Pkg.add("LoopVectorization"); Pkg.test("LoopVectorization")'` 5. `make -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways. Turn into makefile Newline Use two out of source builds Ignore profiles + build dirs Add --icf=safe stage0 setup prebuilt clang with [cd]tors->init/fini patch
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `cd contrib/pgo-lto` 2. `make -j$(nproc) stage1` 3. `make clean-profiles` 4. `./stage1.build/julia -O3 -e 'using Pkg; Pkg.add("LoopVectorization"); Pkg.test("LoopVectorization")'` 5. `make -j$(nproc) stage2` This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue JuliaLang#45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 5.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions JuliaLang#2 ``` Wit PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions JuliaLang#2 ``` Or -28% time spent in LLVM. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anways. Turn into makefile Newline Use two out of source builds Ignore profiles + build dirs Add --icf=safe stage0 setup prebuilt clang with [cd]tors->init/fini patch
I don't think there is too much to say here, since compilation is expected to be super-linear in code size, other than do not embed data into programs if you don't want that. Otherwise do it. |
FWIW, v1.11 nightly with asserts timing is currently:
|
Adds a convenient way to enable PGO+LTO on Julia and LLVM together: 1. `cd contrib/pgo-lto` 2. `make -j$(nproc) stage1` 3. `make clean-profiles` 4. `./stage1.build/julia -O3 -e 'using Pkg; Pkg.add("LoopVectorization"); Pkg.test("LoopVectorization")'` 5. `make -j$(nproc) stage2` <details> <summary>* Output looks roughly like as follows</summary> ```c++ $ make -C contrib/pgo-lto top make: Entering directory '/dev/shm/julia/contrib/pgo-lto' llvm-profdata show --topn=50 /dev/shm/julia/contrib/pgo-lto/profiles/merged.prof | c++filt Instrumentation level: IR entry_first = 0 Total functions: 85943 Maximum function count: 7867557260 Maximum internal block count: 3468437590 Top 50 functions with the largest internal block counts: llvm::BitVector::operator|=(llvm::BitVector const&), max count = 7867557260 LateLowerGCFrame::ComputeLiveness(State&), max count = 3468437590 llvm::hashing::detail::hash_combine_recursive_helper::hash_combine_recursive_helper(), max count = 1742259834 llvm::SUnit::addPred(llvm::SDep const&, bool), max count = 511396575 llvm::LiveRange::overlaps(llvm::LiveRange const&, llvm::CoalescerPair const&, llvm::SlotIndexes const&) const, max count = 508061762 llvm::StringMapImpl::LookupBucketFor(llvm::StringRef), max count = 505682177 std::map<llvm::BasicBlock*, BBState, std::less<llvm::BasicBlock*>, std::allocator<std::pair<llvm::BasicBlock* const, BBState> > >::operator[](llvm::BasicBlock* const&), max count = 395628888 llvm::LiveRange::advanceTo(llvm::LiveRange::Segment const*, llvm::SlotIndex) const, max count = 384642728 llvm::LiveRange::isLiveAtIndexes(llvm::ArrayRef<llvm::SlotIndex>) const, max count = 380291040 llvm::PassRegistry::enumerateWith(llvm::PassRegistrationListener*), max count = 352313953 ijl_method_instance_add_backedge, max count = 349608221 llvm::SUnit::ComputeHeight(), max count = 336604330 llvm::LiveRange::advanceTo(llvm::LiveRange::Segment*, llvm::SlotIndex), max count = 331030109 llvm::SmallPtrSetImplBase::insert_imp(void const*), max count = 272966545 llvm::LiveIntervals::checkRegMaskInterference(llvm::LiveInterval&, llvm::BitVector&), max count = 257449540 LateLowerGCFrame::ComputeLiveSets(State&), max count = 252096274 /dev/shm/julia/src/jltypes.c:has_free_typevars, max count = 230879464 ijl_get_pgcstack, max count = 216953592 LateLowerGCFrame::RefineLiveSet(llvm::BitVector&, State&, std::vector<int, std::allocator<int> > const&), max count = 188013152 /dev/shm/julia/src/flisp/flisp.c:apply_cl, max count = 174863813 /dev/shm/julia/src/flisp/builtins.c:fl_memq, max count = 168621603 ``` </details> This results quite often in spectacular speedups for time to first X as it reduces the time spent in LLVM optimization passes by 25 or even 30%. Example 1: ```julia using LoopVectorization function f!(a, b) @turbo for i in eachindex(a) a[i] *= b[i] end return a end f!(rand(1), rand(1)) ``` ```console $ time ./julia -O3 lv.jl ``` Without PGO+LTO: 14.801s With PGO+LTO: 11.978s (-19%) Example 2: ```console $ time ./julia -e 'using Pkg; Pkg.test("Unitful");' ``` Without PGO+LTO: 1m47.688s With PGO+LTO: 1m35.704s (-11%) Example 3 (taken from issue #45395, which is almost only LLVM): ```console $ JULIA_LLVM_ARGS=-time-passes ./julia script-45395.jl ``` Without PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 101.0130 seconds (98.6253 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 53.6961 ( 54.7%) 0.1050 ( 3.8%) 53.8012 ( 53.3%) 53.8045 ( 54.6%) Unroll loops 25.5423 ( 26.0%) 0.0072 ( 0.3%) 25.5495 ( 25.3%) 25.5444 ( 25.9%) Global Value Numbering 7.1995 ( 7.3%) 0.0526 ( 1.9%) 7.2521 ( 7.2%) 7.2517 ( 7.4%) Induction Variable Simplification 6.0541 ( 5.1%) 0.0098 ( 0.3%) 5.0639 ( 5.0%) 5.0561 ( 5.1%) Combine redundant instructions #2 ``` With PGO+LTO: ``` ===-------------------------------------------------------------------------=== ... Pass execution timing report ... ===-------------------------------------------------------------------------=== Total Execution Time: 72.6507 seconds (70.1337 wall clock) ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name --- 36.0894 ( 51.7%) 0.0825 ( 2.9%) 36.1719 ( 49.8%) 36.1738 ( 51.6%) Unroll loops 16.5713 ( 23.7%) 0.0129 ( 0.5%) 16.5843 ( 22.8%) 16.5794 ( 23.6%) Global Value Numbering 5.9047 ( 8.5%) 0.0395 ( 1.4%) 5.9442 ( 8.2%) 5.9438 ( 8.5%) Induction Variable Simplification 4.7566 ( 6.8%) 0.0078 ( 0.3%) 4.7645 ( 6.6%) 4.7575 ( 6.8%) Combine redundant instructions #2 ``` Or -28% time spent in LLVM. `perf` reports show this is mostly fewer instructions and reduction in icache misses. --- Finally there's a significant reduction in binary sizes. For libLLVM.so: ``` 79M usr/lib/libLLVM-13jl.so (before) 67M usr/lib/libLLVM-13jl.so (after) ``` And it can be reduced by another 2MB with `--icf=safe` when using LLD as a linker anyways. - [x] Two out-of-source builds would be better than a single in-source build, so that it's easier to find good profile data --------- Co-authored-by: Oscar Smith <oscardssmith@gmail.com> Co-authored-by: Lilith Orion Hafner <lilithhafner@gmail.com>
This uses LLVM's BOLT to optimize libLLVM, libjulia-internal and libjulia-codegen. This improves the allinference benchmarks by about 10% largely due to the optimization of libjulia-internal. The example in issue #45395 which stresses LLVM significantly more also sees a ~10% improvement. We see a 20% improvement on ```julia @time for i in 1:100000000 string(i) end ``` When building corecompiler.ji: BOLT gives about a 16% improvement PGO+LTO gives about a 21% improvement PGO+LTO+BOLT gives about a 23% improvement This only requires a single build of LLVM and theoretically none if we change the binary builder script (i.e. we build with relocations and the `-fno-reorder-blocks-and-partition` and then we can use BOLT to get binaries with no relocations and reordered blocks and then ship both binaries?) compared to the 2 in PGO. Also, this theoretically can improve performance of a PGO+LTO build by a couple %. The only reproducible test problem I see is that the BOLT, PGO+LTO and PGO+LTO+BOLT builds all cause `readelf` to emit warnings as part of the `osutils` tests. ``` readelf: Warning: Unrecognised form: 0x22 readelf: Warning: DIE has locviews without loclist readelf: Warning: Unrecognised form: 0x23 readelf: Warning: DIE at offset 0x227399 refers to abbreviation number 14754 which does not exist readelf: Warning: Bogus end-of-siblings marker detected at offset 212aa9 in .debug_info section readelf: Warning: Bogus end-of-siblings marker detected at offset 212ab0 in .debug_info section readelf: Warning: Further warnings about bogus end-of-sibling markers suppressed ``` The unrecognised form warnings seem to be a bug in binutils, https://sourceware.org/bugzilla/show_bug.cgi?id=28981. `DIE at offset` warning I believe was fixed in binutils 2.36, https://sourceware.org/bugzilla/show_bug.cgi?id=26808, but `ld -v` says I have 2.38. I assume these are all benign. I also don't see them on CI here https://buildkite.com/julialang/julia-buildkite/builds/1507#018f00e7-0737-4a42-bcd9-d4061dc8c93e so could just be a local issue.
This uses LLVM's BOLT to optimize libLLVM, libjulia-internal and libjulia-codegen. This improves the allinference benchmarks by about 10% largely due to the optimization of libjulia-internal. The example in issue JuliaLang#45395 which stresses LLVM significantly more also sees a ~10% improvement. We see a 20% improvement on ```julia @time for i in 1:100000000 string(i) end ``` When building corecompiler.ji: BOLT gives about a 16% improvement PGO+LTO gives about a 21% improvement PGO+LTO+BOLT gives about a 23% improvement This only requires a single build of LLVM and theoretically none if we change the binary builder script (i.e. we build with relocations and the `-fno-reorder-blocks-and-partition` and then we can use BOLT to get binaries with no relocations and reordered blocks and then ship both binaries?) compared to the 2 in PGO. Also, this theoretically can improve performance of a PGO+LTO build by a couple %. The only reproducible test problem I see is that the BOLT, PGO+LTO and PGO+LTO+BOLT builds all cause `readelf` to emit warnings as part of the `osutils` tests. ``` readelf: Warning: Unrecognised form: 0x22 readelf: Warning: DIE has locviews without loclist readelf: Warning: Unrecognised form: 0x23 readelf: Warning: DIE at offset 0x227399 refers to abbreviation number 14754 which does not exist readelf: Warning: Bogus end-of-siblings marker detected at offset 212aa9 in .debug_info section readelf: Warning: Bogus end-of-siblings marker detected at offset 212ab0 in .debug_info section readelf: Warning: Further warnings about bogus end-of-sibling markers suppressed ``` The unrecognised form warnings seem to be a bug in binutils, https://sourceware.org/bugzilla/show_bug.cgi?id=28981. `DIE at offset` warning I believe was fixed in binutils 2.36, https://sourceware.org/bugzilla/show_bug.cgi?id=26808, but `ld -v` says I have 2.38. I assume these are all benign. I also don't see them on CI here https://buildkite.com/julialang/julia-buildkite/builds/1507#018f00e7-0737-4a42-bcd9-d4061dc8c93e so could just be a local issue.
I have a problem with excessive compilation time. The problem is with code automatically translated from another system (gap3) so I cannot easily change it. As an example, the function below takes 2 minutes to compile on Julia 1.7.2 (it takes 5 minutes on 1.9 nightly!); as you can see, the code is long but very simple. When you load the code, executing
@time f(4)
the first time takes 128sec (and some microseconds after that). The following should be noted:If I move the definition of
d
outside the function the compilation takes a fraction of a second. I can move d out in this example but not on some other examples where the data defines polynomials that the function evaluates at its argument (imagine that a list [1,2,3,4] in the example is replaced by something like [1+x+2y^2+4x^3] where x,y,z are arguments to the function).The time is exponential in the length of the data (if you double the length of the Dict definition the time is multiplied by 5 or 10).
Giordano’s has tested the example on Julia 1.0.5 where it compiles in 1 or 2 sec., so it is a regression.
The text was updated successfully, but these errors were encountered: