Conversation
Contributor
|
Your PR requires formatting changes to meet the project's style guidelines. Click here to view the suggested changes.diff --git a/src/compiler/optimize.jl b/src/compiler/optimize.jl
index 5651bc55..61bf69f4 100644
--- a/src/compiler/optimize.jl
+++ b/src/compiler/optimize.jl
@@ -79,7 +79,7 @@ function optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine)
function middle_optimize!(second_stage=false)
@dispose pb = NewPMPassBuilder() begin
registerEnzymeAndPassPipeline!(pb)
- register!(pb, RestoreAllocaType())
+ register!(pb, RestoreAllocaType())
add!(pb, NewPMAAManager()) do aam
add!(aam, ScopedNoAliasAA())
add!(aam, TypeBasedAA())
@@ -102,7 +102,7 @@ function optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine)
add!(fpm, ReassociatePass())
add!(fpm, EarlyCSEPass())
add!(fpm, AllocOptPass())
- add!(fpm, RestoreAllocaType())
+ add!(fpm, RestoreAllocaType())
add!(fpm, NewPMLoopPassManager(use_memory_ssa=true)) do lpm
add!(lpm, LoopIdiomRecognizePass())
@@ -122,7 +122,7 @@ function optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine)
# todo peeling=false?
add!(fpm, LoopUnrollPass(opt_level=2, partial=false)) # what opt level?
add!(fpm, AllocOptPass())
- add!(fpm, RestoreAllocaType())
+ add!(fpm, RestoreAllocaType())
add!(fpm, SROAPass())
add!(fpm, GVNPass())
@@ -137,7 +137,7 @@ function optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine)
add!(fpm, JumpThreadingPass())
add!(fpm, DSEPass())
add!(fpm, AllocOptPass())
- add!(fpm, RestoreAllocaType())
+ add!(fpm, RestoreAllocaType())
add!(fpm, SimplifyCFGPass())
@@ -229,7 +229,7 @@ function addOptimizationPasses!(mpm::LLVM.NewPMPassManager)
# merging the `alloca` for the unboxed data and the `alloca` created by the `alloc_opt`
# pass.
- add!(fpm, AllocOptPass())
+ add!(fpm, AllocOptPass())
add!(fpm, RestoreAllocaType())
# consider AggressiveInstCombinePass at optlevel > 2
@@ -292,7 +292,7 @@ function addOptimizationPasses!(mpm::LLVM.NewPMPassManager)
# More dead allocation (store) deletion before loop optimization
# consider removing this:
- add!(fpm, AllocOptPass())
+ add!(fpm, AllocOptPass())
add!(fpm, RestoreAllocaType())
# see if all of the constant folding has exposed more loops
@@ -446,13 +446,13 @@ function post_optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine, machine::Bool
registerEnzymeAndPassPipeline!(pb)
register!(pb, ReinsertGCMarkerPass())
register!(pb, SafeAtomicToRegularStorePass())
- register!(pb, RestoreAllocaType())
+ register!(pb, RestoreAllocaType())
add!(pb, NewPMAAManager()) do aam
add!(aam, ScopedNoAliasAA())
add!(aam, TypeBasedAA())
add!(aam, BasicAA())
end
- add!(pb, NewPMModulePassManager()) do mpm
+ add!(pb, NewPMModulePassManager()) do mpm
addOptimizationPasses!(mpm)
if machine
# TODO enable validate_return_roots
diff --git a/src/llvm/attributes.jl b/src/llvm/attributes.jl
index 1b5128ea..3c5e1381 100644
--- a/src/llvm/attributes.jl
+++ b/src/llvm/attributes.jl
@@ -1,5 +1,5 @@
const nofreefns = Set{String}((
- "jl_genericmemory_copyto",
+ "jl_genericmemory_copyto",
"utf8proc_toupper",
"ClientGetAddressableDevices",
"ClientNumAddressableDevices",
diff --git a/src/llvm/transforms.jl b/src/llvm/transforms.jl
index ec4f1401..a449155a 100644
--- a/src/llvm/transforms.jl
+++ b/src/llvm/transforms.jl
@@ -1,5 +1,5 @@
function restore_alloca_type!(f::LLVM.Function)
- replaceAndErase = Tuple{LLVM.AllocaInst,Type, LLVMType}[]
+ replaceAndErase = Tuple{LLVM.AllocaInst, Type, LLVMType}[]
dl = datalayout(LLVM.parent(f))
for bb in blocks(f), inst in instructions(bb)
@@ -39,12 +39,12 @@ function restore_alloca_type!(f::LLVM.Function)
cst = al2
if value_type(cst) != value_type(al)
cst = bitcast!(b, cst, value_type(al))
- end
+ end
LLVM.replace_uses!(al, cst)
LLVM.API.LLVMInstructionEraseFromParent(al)
metadata(inst)["enzymejl_allocart"] = MDNode(LLVM.Metadata[MDString(string(convert(UInt, unsafe_to_pointer(RT))))])
end
- return length(replaceAndErase) != 0
+ return length(replaceAndErase) != 0
end
# Rewrite calls with "jl_roots" to only have the jl_value_t attached and not { { {} addrspace(10)*, [1 x [2 x i64]], i64, i64 }, [2 x i64] } %unbox110183_replacementA
@@ -2691,7 +2691,7 @@ function removeDeadArgs!(mod::LLVM.Module, tm::LLVM.TargetMachine, post_gc_fixup
propagate_returned!(mod)
LLVM.@dispose pb = NewPMPassBuilder() begin
registerEnzymeAndPassPipeline!(pb)
- register!(pb, RestoreAllocaType())
+ register!(pb, RestoreAllocaType())
add!(pb, NewPMModulePassManager()) do mpm
add!(mpm, NewPMFunctionPassManager()) do fpm
add!(fpm, InstCombinePass())
@@ -2720,7 +2720,7 @@ function removeDeadArgs!(mod::LLVM.Module, tm::LLVM.TargetMachine, post_gc_fixup
LLVM.@dispose pb = NewPMPassBuilder() begin
registerEnzymeAndPassPipeline!(pb)
register!(pb, EnzymeAttributorPass())
- register!(pb, RestoreAllocaType())
+ register!(pb, RestoreAllocaType())
add!(pb, NewPMModulePassManager()) do mpm
add!(mpm, NewPMFunctionPassManager()) do fpm
add!(fpm, InstCombinePass())
diff --git a/src/typeutils/lltypes.jl b/src/typeutils/lltypes.jl
index 0723dcd7..5b325943 100644
--- a/src/typeutils/lltypes.jl
+++ b/src/typeutils/lltypes.jl
@@ -98,7 +98,7 @@ function strip_tracked_pointers(@nospecialize(T::LLVM.LLVMType))
for (i, t) in enumerate(LLVM.elements(ty))
push!(subtypes, strip_tracked_pointers(t))
end
- return LLVM.StructType(subtypes; packed=LLVM.ispacked(T))
+ return LLVM.StructType(subtypes; packed = LLVM.ispacked(T))
end
throw(AssertionError("Unknown composite type")) |
wsmoses
commented
Jan 16, 2026
|
|
||
| EnzymeAttributorPass() = NewPMModulePass("enzyme_attributor", enzyme_attributor_pass!) | ||
| ReinsertGCMarkerPass() = NewPMFunctionPass("reinsert_gcmarker", reinsert_gcmarker_pass!) | ||
| RestoreAllocaType() = NewPMFunctionPass("restore_alloca_type", restore_alloca_type!) |
Member
Author
There was a problem hiding this comment.
@vchuravy can you check what obvious thing I'm doing wrong here?
Member
There was a problem hiding this comment.
register!(pb, RestoreAllocaType())
Collaborator
|
With this on 1.12 backports does SciML work? |
Member
Author
|
Well we need to fix the "I'm silly and didn't register the pass correctly here", then we can test |
Collaborator
|
Given JuliaLang/julia#60699 is this still necessary? They will probably be both in the same release regardless |
Member
Author
|
Yes it's separately necessary to determine the type of the alloca |
Return a boolean indicating if replacements and erasures occurred.
Closed
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2902 +/- ##
==========================================
- Coverage 66.98% 66.89% -0.10%
==========================================
Files 58 58
Lines 21216 21308 +92
==========================================
+ Hits 14212 14254 +42
- Misses 7004 7054 +50 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Contributor
Benchmark Results
Benchmark PlotsA plot of the benchmark results has been uploaded as an artifact at https://github.com/EnzymeAD/Enzyme.jl/actions/runs/21123951548/artifacts/5172079042. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Requires JuliaLang/julia#60695 to be effective.
Fixes numerous 1.12 problems.
Essentially suppose we have a gc alloc of a complex64. LLVMAlloc opt will change this into an [16 x i8]. Then sroa will turn that into 16 individual byte-sized extract/insert -- which we cannot actually handle properly