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/absint.jl b/src/absint.jl
index 06a06d3d..3484f35e 100644
--- a/src/absint.jl
+++ b/src/absint.jl
@@ -6,11 +6,11 @@
const JL_MAX_TAGS = 64 # see `enum jl_small_typeof_tags` in julia.h
function unbind(@nospecialize(val))
- if val isa Core.Binding
- return val.value
- else
- return val
- end
+ if val isa Core.Binding
+ return val.value
+ else
+ return val
+ end
end
function absint(@nospecialize(arg::LLVM.Value), partial::Bool = false, istracked::Bool=false, typetag::Bool=false)::Tuple{Bool, Any}
@@ -28,7 +28,7 @@ function absint(@nospecialize(arg::LLVM.Value), partial::Bool = false, istracked
return (true, v)
end
end
- @assert !startswith(gname, "ejl_inserted") "Could not find ejl_inserted variable in map $gname"
+ @assert !startswith(gname, "ejl_inserted") "Could not find ejl_inserted variable in map $gname"
end
if isa(ce, LLVM.LoadInst)
gv = operands(ce)[1]
@@ -373,7 +373,7 @@ function abs_typeof(
end
for (k, v) in JuliaEnzymeNameMap
if gname == "ejl_" * k
- return (true, Core.Typeof(unbind(v)), GPUCompiler.BITS_REF)
+ return (true, Core.Typeof(unbind(v)), GPUCompiler.BITS_REF)
end
end
end
@@ -464,14 +464,14 @@ function abs_typeof(
nm == "jl_gc_alloc_typed" ||
nm == "ijl_gc_alloc_typed"
vals = absint(operands(arg)[3], partial, false, #=typetag=#true)
- @assert !(vals[2] isa Core.Binding)
+ @assert !(vals[2] isa Core.Binding)
return (vals[1], vals[2], vals[1] ? GPUCompiler.BITS_REF : nothing)
end
# Type tag is arg 3
if nm == "jl_alloc_genericmemory_unchecked" ||
nm == "ijl_alloc_genericmemory_unchecked"
vals = absint(operands(arg)[3], partial, true, #=typetag=#true)
- @assert !(vals[2] isa Core.Binding)
+ @assert !(vals[2] isa Core.Binding)
return (vals[1], vals[2], vals[1] ? GPUCompiler.MUT_REF : nothing)
end
# Type tag is arg 1
@@ -486,13 +486,13 @@ function abs_typeof(
nm == "jl_alloc_genericmemory" ||
nm == "ijl_alloc_genericmemory"
vals = absint(operands(arg)[1], partial, false, #=typetag=#true)
- @assert !(vals[2] isa Core.Binding)
+ @assert !(vals[2] isa Core.Binding)
return (vals[1], vals[2], vals[1] ? GPUCompiler.MUT_REF : nothing)
end
if nm == "jl_new_structt" || nm == "ijl_new_structt"
vals = absint(operands(arg)[1], partial, false, #=typetag=#true)
- @assert !(vals[2] isa Core.Binding)
+ @assert !(vals[2] isa Core.Binding)
return (vals[1], vals[2], vals[1] ? GPUCompiler.MUT_REF : nothing)
end
@@ -511,7 +511,7 @@ function abs_typeof(
if nm == "jl_new_structv" || nm == "ijl_new_structv"
@assert index == 2
vals = absint(operands(arg)[index], partial, false, #=typetag=#true)
- @assert !(vals[2] isa Core.Binding)
+ @assert !(vals[2] isa Core.Binding)
return (vals[1], vals[2], vals[1] ? GPUCompiler.MUT_REF : nothing)
end
@@ -545,11 +545,11 @@ function abs_typeof(
if nm == "jl_f__apply_iterate" || nm == "ijl_f__apply_iterate"
index += 1
legal, iterfn = absint(operands(arg)[index])
- iterfn = unbind(iterfn)
+ iterfn = unbind(iterfn)
index += 1
if legal && iterfn == Base.iterate
legal0, combfn = absint(operands(arg)[index])
- combfn = unbind(combfn)
+ combfn = unbind(combfn)
index += 1
if legal0 && combfn == Core.apply_type && partial
return (true, Type, GPUCompiler.BITS_REF)
@@ -887,7 +887,7 @@ function abs_typeof(
legal, val = absint(arg, partial)
if legal
- val = unbind(val)
+ val = unbind(val)
return (true, Core.Typeof(val), GPUCompiler.BITS_REF)
end
return (false, nothing, nothing)
diff --git a/src/api.jl b/src/api.jl
index 75d7b0a8..93ccfeac 100644
--- a/src/api.jl
+++ b/src/api.jl
@@ -1502,11 +1502,11 @@ function EnzymeDumpModuleRef(mod)
end
function EnzymeDumpValueRef(mod)
- ccall((:EnzymeDumpValueRef, libEnzyme), Cvoid, (LLVM.API.LLVMValueRef,), mod)
+ return ccall((:EnzymeDumpValueRef, libEnzyme), Cvoid, (LLVM.API.LLVMValueRef,), mod)
end
function EnzymeDumpTypeRef(mod)
- ccall((:EnzymeDumpTypeRef, libEnzyme), Cvoid, (LLVM.API.LLVMValueRef,), mod)
+ return ccall((:EnzymeDumpTypeRef, libEnzyme), Cvoid, (LLVM.API.LLVMValueRef,), mod)
end
EnzymeComputeByteOffsetOfGEP(B, V, T) = LLVM.Value(
diff --git a/src/compiler.jl b/src/compiler.jl
index 78b160d0..ee1b2cd7 100644
--- a/src/compiler.jl
+++ b/src/compiler.jl
@@ -1694,10 +1694,10 @@ function shadow_alloc_rewrite(V::LLVM.API.LLVMValueRef, gutils::API.EnzymeGradie
index += 1
found = Any[]
legal, Ty = absint(operands(arg)[index], partial)
- Ty = unbind(Ty)
+ Ty = unbind(Ty)
if legal && Ty == NTuple
legal, Ty = absint(operands(arg)[index+2])
- Ty = unbind(Ty)
+ Ty = unbind(Ty)
if legal
# count should represent {the total size in bytes, the aligned size of each element}
B = LLVM.IRBuilder()
@@ -5402,8 +5402,8 @@ function GPUCompiler.compile_unhooked(output::Symbol, job::CompilerJob{<:EnzymeT
@static if VERSION < v"1.11-"
else
legal2, obj = absint(inst)
- obj = unbind(obj)
- if legal2 && is_memory_instance(obj)
+ obj = unbind(obj)
+ if legal2 && is_memory_instance(obj)
metadata(inst)["nonnull"] = MDNode(LLVM.Metadata[])
end
end
@@ -5632,7 +5632,7 @@ end
string(cur)
slegal, foundv = absint(cur)
if slegal
- foundv = unbind(foundv)
+ foundv = unbind(foundv)
resstr *= "of type " * string(foundv)
end
emit_error(builder, user, resstr, EnzymeMutabilityException)
@@ -6468,7 +6468,7 @@ const DumpLLVMCall = Ref(false)
end
reinsert_gcmarker!(llvm_f)
- Enzyme.Compiler.JIT.prepare!(mod)
+ Enzyme.Compiler.JIT.prepare!(mod)
if DumpLLVMCall[]
API.EnzymeDumpModuleRef(mod.ref)
end
@@ -6594,7 +6594,7 @@ function _thunk(job, postopt::Bool = true)::Tuple{LLVM.Module, Vector{Any}, Stri
end
else
propagate_returned!(mod)
- Compiler.JIT.prepare!(mod)
+ Compiler.JIT.prepare!(mod)
end
mstr
else
diff --git a/src/compiler/optimize.jl b/src/compiler/optimize.jl
index 0ccb9dfe..c77e7788 100644
--- a/src/compiler/optimize.jl
+++ b/src/compiler/optimize.jl
@@ -74,13 +74,13 @@ function optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine)
function middle_optimize!(second_stage=false)
@dispose pb = NewPMPassBuilder() begin
- registerEnzymeAndPassPipeline!(pb)
+ registerEnzymeAndPassPipeline!(pb)
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
add!(mpm, CPUFeaturesPass()) # why is this duplicated?
add!(mpm, NewPMFunctionPassManager()) do fpm
diff --git a/src/compiler/orcv2.jl b/src/compiler/orcv2.jl
index da9ec7c7..df747497 100644
--- a/src/compiler/orcv2.jl
+++ b/src/compiler/orcv2.jl
@@ -187,22 +187,22 @@ function prepare!(mod)
end
for g in collect(globals(mod))
if !startswith(LLVM.name(g), "ejl_inserted\$")
- continue
+ continue
end
_, ogname, load1, initaddr = split(LLVM.name(g), "\$")
load1 = load1 == "true"
- initaddr = parse(UInt, initaddr)
+ initaddr = parse(UInt, initaddr)
ptr = Base.reinterpret(Ptr{Ptr{Cvoid}}, initaddr)
if load1
- ptr = Base.unsafe_load(ptr, :unordered)
+ ptr = Base.unsafe_load(ptr, :unordered)
end
-
+
obj = Base.unsafe_pointer_to_objref(ptr)
-
+
# Let's try a de-bind for 1.10 lux
if isa(obj, Core.Binding)
- ptr = Compiler.unsafe_to_ptr(obj.value)
+ ptr = Compiler.unsafe_to_ptr(obj.value)
end
ptr = reinterpret(UInt, ptr)
@@ -212,6 +212,7 @@ function prepare!(mod)
replace_uses!(g, ptr)
Compiler.eraseInst(mod, g)
end
+ return
end
function get_trampoline(job)
@@ -258,7 +259,7 @@ function get_trampoline(job)
Compiler.eraseInst(mod, other_func)
end
- prepare!(mod)
+ prepare!(mod)
tsm = move_to_threadsafe(mod)
il = LLVM.IRCompileLayer(lljit)
diff --git a/src/compiler/utils.jl b/src/compiler/utils.jl
index 701e8e4d..96fb7ac5 100644
--- a/src/compiler/utils.jl
+++ b/src/compiler/utils.jl
@@ -317,7 +317,7 @@ function declare_pgcstack!(mod::LLVM.Module)
)
end
-function emit_pgcstack(B::LLVM.IRBuilder, name::String="")
+function emit_pgcstack(B::LLVM.IRBuilder, name::String = "")
curent_bb = position(B)
fn = LLVM.parent(curent_bb)
mod = LLVM.parent(fn)
@@ -353,20 +353,20 @@ function reinsert_gcmarker!(func::LLVM.Function, @nospecialize(PB::Union{Nothing
context(LLVM.parent(func))
B = IRBuilder()
entry_bb = first(blocks(func))
- if PB !== nothing && LLVM.name(Base.position(PB)) == "allocsForInversion"
- B = PB
- elseif !isempty(instructions(entry_bb))
- if PB === nothing || Base.position(PB) != entry_bb
- position!(B, first(instructions(entry_bb)))
- else
- B = PB
- end
+ if PB !== nothing && LLVM.name(Base.position(PB)) == "allocsForInversion"
+ B = PB
+ elseif !isempty(instructions(entry_bb))
+ if PB === nothing || Base.position(PB) != entry_bb
+ position!(B, first(instructions(entry_bb)))
+ else
+ B = PB
+ end
else
- if PB === nothing || Base.position(PB) != entry_bb
- position!(B, entry_bb)
- else
- B = PB
- end
+ if PB === nothing || Base.position(PB) != entry_bb
+ position!(B, entry_bb)
+ else
+ B = PB
+ end
end
emit_pgcstack(B, "newly_emitted_pgc_stack")
else
diff --git a/src/compiler/validation.jl b/src/compiler/validation.jl
index 1138d8b1..ed7da86f 100644
--- a/src/compiler/validation.jl
+++ b/src/compiler/validation.jl
@@ -239,77 +239,77 @@ function check_ir!(interp, @nospecialize(job::CompilerJob), errors::Vector{IRErr
inst = LLVM.Instruction(iter)
iter = LLVM.API.LLVMGetNextInstruction(iter)
- if isa(value_type(inst), LLVM.PointerType) && addrspace(value_type(inst)) == Tracked
- inst0, _ = get_base_and_offset(inst; offsetAllowed=false, inttoptr=true)
- if isa(inst0, LLVM.LoadInst) && addrspace(value_type(operands(inst0)[1])) == 0
- addr = operands(inst0)[1]
- addr, off = get_base_and_offset(addr; offsetAllowed=true, inttoptr=true)
- gname = nothing
- load1 = false
- if isa(addr, LLVM.GlobalVariable) && haskey(metadata(addr), "julia.constgv")
- paddr = addr
- addr = LLVM.initializer(paddr)
- gname = LLVM.name(paddr)*"\$false"
- addr, _ = get_base_and_offset(addr; offsetAllowed=false, inttoptr=true)
- elseif isa(addr, LLVM.LoadInst)
- paddr = operands(addr)[1]
- if isa(paddr, LLVM.GlobalVariable) && haskey(metadata(paddr), "julia.constgv")
- addr = LLVM.initializer(paddr)
- gname = LLVM.name(paddr)*"\$true"
- addr, _ = get_base_and_offset(addr; offsetAllowed=false, inttoptr=true)
- load1 = true
- end
- elseif isa(addr, LLVM.ConstantInt)
- gname = string(convert(UInt, addr))*"\$true"
- load1 = true
- end
+ if isa(value_type(inst), LLVM.PointerType) && addrspace(value_type(inst)) == Tracked
+ inst0, _ = get_base_and_offset(inst; offsetAllowed = false, inttoptr = true)
+ if isa(inst0, LLVM.LoadInst) && addrspace(value_type(operands(inst0)[1])) == 0
+ addr = operands(inst0)[1]
+ addr, off = get_base_and_offset(addr; offsetAllowed = true, inttoptr = true)
+ gname = nothing
+ load1 = false
+ if isa(addr, LLVM.GlobalVariable) && haskey(metadata(addr), "julia.constgv")
+ paddr = addr
+ addr = LLVM.initializer(paddr)
+ gname = LLVM.name(paddr) * "\$false"
+ addr, _ = get_base_and_offset(addr; offsetAllowed = false, inttoptr = true)
+ elseif isa(addr, LLVM.LoadInst)
+ paddr = operands(addr)[1]
+ if isa(paddr, LLVM.GlobalVariable) && haskey(metadata(paddr), "julia.constgv")
+ addr = LLVM.initializer(paddr)
+ gname = LLVM.name(paddr) * "\$true"
+ addr, _ = get_base_and_offset(addr; offsetAllowed = false, inttoptr = true)
+ load1 = true
+ end
+ elseif isa(addr, LLVM.ConstantInt)
+ gname = string(convert(UInt, addr)) * "\$true"
+ load1 = true
+ end
- if isa(addr, LLVM.ConstantInt)
-
- initaddr = convert(UInt, addr) + off
- if gname isa String
- gname = gname *"\$$initaddr"
- end
- ptr = Base.reinterpret(Ptr{Ptr{Cvoid}}, initaddr)
- if load1
- ptr = Base.unsafe_load(ptr, :unordered)
- if ptr == C_NULL
- continue
- end
- end
- obj = Base.unsafe_pointer_to_objref(ptr)
- if obj === nothing
- continue
- end
- obj0 = obj
-
- # TODO we can use this to make it properly relocatable
- if isa(obj, Core.Binding)
- obj = obj.value
- if gname === nothing
- obj0 = obj
- end
- end
+ if isa(addr, LLVM.ConstantInt)
- # We really don't want to mess with the atomic baked in loads here
- #if obj isa Base.ReentrantLock
- # continue
- #end
-
- b = IRBuilder()
- position!(b, inst)
- newf = unsafe_to_llvm(b, obj0; insert_name_if_not_exists=gname)
- replace_uses!(inst, newf)
- LLVM.API.LLVMInstructionEraseFromParent(inst)
- continue
- end
- end
- end
+ initaddr = convert(UInt, addr) + off
+ if gname isa String
+ gname = gname * "\$$initaddr"
+ end
+ ptr = Base.reinterpret(Ptr{Ptr{Cvoid}}, initaddr)
+ if load1
+ ptr = Base.unsafe_load(ptr, :unordered)
+ if ptr == C_NULL
+ continue
+ end
+ end
+ obj = Base.unsafe_pointer_to_objref(ptr)
+ if obj === nothing
+ continue
+ end
+ obj0 = obj
+
+ # TODO we can use this to make it properly relocatable
+ if isa(obj, Core.Binding)
+ obj = obj.value
+ if gname === nothing
+ obj0 = obj
+ end
+ end
+
+ # We really don't want to mess with the atomic baked in loads here
+ #if obj isa Base.ReentrantLock
+ # continue
+ #end
+
+ b = IRBuilder()
+ position!(b, inst)
+ newf = unsafe_to_llvm(b, obj0; insert_name_if_not_exists = gname)
+ replace_uses!(inst, newf)
+ LLVM.API.LLVMInstructionEraseFromParent(inst)
+ continue
+ end
+ end
+ end
if isa(inst, LLVM.CallInst)
push!(calls, inst)
# remove illegal invariant.load and jtbaa_const invariants
elseif isa(inst, LLVM.LoadInst)
- fn_got, _ = get_base_and_offset(operands(inst)[1]; offsetAllowed=false, inttoptr=false)
+ fn_got, _ = get_base_and_offset(operands(inst)[1]; offsetAllowed = false, inttoptr = false)
fname = String(name(fn_got))
match_ = match(r"^jlplt_(.*)_\d+_got$", fname)
@@ -852,11 +852,11 @@ function check_ir!(interp, @nospecialize(job::CompilerJob), errors::Vector{IRErr
flib = ops[1]
fname = ops[2]
- if isa(flib, LLVM.ConstantExpr) || isa(flib, LLVM.GlobalVariable)
- legal, flib2 = absint(flib)
- if legal
- flib = unbind(flib2)
- end
+ if isa(flib, LLVM.ConstantExpr) || isa(flib, LLVM.GlobalVariable)
+ legal, flib2 = absint(flib)
+ if legal
+ flib = unbind(flib2)
+ end
end
if isa(flib, GlobalRef) && isdefined(flib.mod, flib.name)
flib = getfield(flib.mod, flib.name)
@@ -1006,7 +1006,7 @@ function check_ir!(interp, @nospecialize(job::CompilerJob), errors::Vector{IRErr
iteroff = 2
legal, iterlib = absint(operands(inst)[iteroff+1])
- iterlib = unbind(iterlib)
+ iterlib = unbind(iterlib)
if legal && iterlib == Base.iterate
legal, GT, byref = abs_typeof(operands(inst)[4+1], true)
funcoff = 3
@@ -1096,7 +1096,7 @@ function check_ir!(interp, @nospecialize(job::CompilerJob), errors::Vector{IRErr
push!(tys, typ)
end
legal, flib = absint(operands(inst)[offset+1])
- flib = unbind(flib)
+ flib = unbind(flib)
if legal && isa(flib, Core.MethodInstance)
if !Base.isvarargtype(flib.specTypes.parameters[end])
@assert length(tys) == length(flib.specTypes.parameters)
@@ -1251,7 +1251,7 @@ function check_ir!(interp, @nospecialize(job::CompilerJob), errors::Vector{IRErr
push!(tys, typ)
end
legal, flib = absint(operands(inst)[offset+1])
- flib = unbind(flib)
+ flib = unbind(flib)
if legal && isa(flib, Core.MethodInstance)
if !Base.isvarargtype(flib.specTypes.parameters[end])
if length(tys) != length(flib.specTypes.parameters)
diff --git a/src/errors.jl b/src/errors.jl
index e89da601..027bcfd0 100644
--- a/src/errors.jl
+++ b/src/errors.jl
@@ -1068,18 +1068,18 @@ function julia_error(
print(io, "Current scope: \n")
print(io, ir)
end
- legal, obj = absint(val)
- if legal
- obj0 = obj
- obj = unbind(obj)
- println(io, "\nValue of type: ", Core.Typeof(obj))
- println(io , " of value : ", obj)
- if obj0 isa Core.Binding
- println(io , " binding : ", obj0)
- end
- println(io)
- end
- if !isa(val, LLVM.Argument) && !isa(val, LLVM.GlobalVariable)
+ legal, obj = absint(val)
+ if legal
+ obj0 = obj
+ obj = unbind(obj)
+ println(io, "\nValue of type: ", Core.Typeof(obj))
+ println(io, " of value : ", obj)
+ if obj0 isa Core.Binding
+ println(io, " binding : ", obj0)
+ end
+ println(io)
+ end
+ if !isa(val, LLVM.Argument) && !isa(val, LLVM.GlobalVariable)
print(io, "\n Inverted pointers: \n")
ip = API.EnzymeGradientUtilsInvertedPointersToString(gutils)
sval = Base.unsafe_string(ip)
@@ -1302,15 +1302,15 @@ function julia_error(
end
legal2, obj = absint(cur)
- obj0 = obj
+ obj0 = obj
# Only do so for the immediate operand/etc to a phi, since otherwise we will make multiple
if legal2
- obj = unbind(obj)
- if is_memory_instance(obj) || (obj isa Core.SimpleVector && length(obj) == 0)
- return make_batched(ncur, prevbb)
- end
+ obj = unbind(obj)
+ if is_memory_instance(obj) || (obj isa Core.SimpleVector && length(obj) == 0)
+ return make_batched(ncur, prevbb)
+ end
if active_reg(TT, world) == ActiveState &&
- ( isa(cur, LLVM.ConstantExpr) || isa(cur, LLVM.GlobalVariable)) &&
+ (isa(cur, LLVM.ConstantExpr) || isa(cur, LLVM.GlobalVariable)) &&
cur == data2
if width == 1
if mode == API.DEM_ForwardMode
@@ -1351,8 +1351,8 @@ else
larg, off = get_base_and_offset(operands(cur)[1])
if isa(larg, LLVM.LoadInst)
legal2, obj = absint(larg)
- obj = unbind(obj)
- if legal2 && is_memory_instance(obj)
+ obj = unbind(obj)
+ if legal2 && is_memory_instance(obj)
return make_batched(ncur, prevbb)
end
end
@@ -1361,10 +1361,10 @@ end
badval = if legal2
sv = string(obj) * " of type" * " " * string(TT)
- if obj0 isa Core.Binding
- sv = sv *" binded at "*string(obj0)
- end
- sv
+ if obj0 isa Core.Binding
+ sv = sv * " binded at " * string(obj0)
+ end
+ sv
else
"Unknown object of type" * " " * string(TT)
end
@@ -1525,7 +1525,7 @@ end
end
end
- if isa(cur, LLVM.LoadInst) || isa(cur, LLVM.BitCastInst) || isa(cur, LLVM.AddrSpaceCastInst) || (isa(cur, LLVM.GetElementPtrInst) && all(Base.Fix2(isa, LLVM.ConstantInt), operands(cur)[2:end])) || (isa(cur,LLVM.ConstantExpr) && opcode(cur) in (LLVM.API.LLVMBitCast, LLVM.API.LLVMAddrSpaceCast, LLVM.API.LLVMGetElementPtr))
+ if isa(cur, LLVM.LoadInst) || isa(cur, LLVM.BitCastInst) || isa(cur, LLVM.AddrSpaceCastInst) || (isa(cur, LLVM.GetElementPtrInst) && all(Base.Fix2(isa, LLVM.ConstantInt), operands(cur)[2:end])) || (isa(cur, LLVM.ConstantExpr) && opcode(cur) in (LLVM.API.LLVMBitCast, LLVM.API.LLVMAddrSpaceCast, LLVM.API.LLVMGetElementPtr))
lhs = make_replacement(operands(cur)[1], prevbb)
if illegal
return ncur
diff --git a/src/jlrt.jl b/src/jlrt.jl
index ad18cc90..00a4e324 100644
--- a/src/jlrt.jl
+++ b/src/jlrt.jl
@@ -20,12 +20,12 @@ function emit_allocobj!(
T_pint8 = LLVM.PointerType(T_int8)
pgcstack = reinsert_gcmarker!(fn, B)
- bc = bitcast!(B, pgcstack, T_ppjlvalue, LLVM.name(pgcstack)*"_bc")
-
+ bc = bitcast!(B, pgcstack, T_ppjlvalue, LLVM.name(pgcstack) * "_bc")
+
ct = inbounds_gep!(
B,
T_pjlvalue,
- bc,
+ bc,
[LLVM.ConstantInt(current_task_offset())],
)
@@ -457,7 +457,7 @@ function emit_apply_type!(B::LLVM.IRBuilder, @nospecialize(Ty::Type), args::Vect
for arg in args
slegal, foundv = absint(arg)
if slegal
- push!(found, unbind(foundv))
+ push!(found, unbind(foundv))
else
legal = false
break
@@ -512,7 +512,7 @@ function emit_tuple!(B::LLVM.IRBuilder, args::Vector{LLVM.Value})::LLVM.Value
for arg in args
slegal, foundv = absint(arg)
if slegal
- push!(found, unbind(foundv))
+ push!(found, unbind(foundv))
else
legal = false
break
@@ -874,7 +874,7 @@ function emit_layout_of_type!(B::LLVM.IRBuilder, @nospecialize(ty::LLVM.Value))
ls = get_layout_struct()
lptr = LLVM.PointerType(ls, 10)
if legal
- JTy = unbind(JTy)
+ JTy = unbind(JTy)
return LLVM.const_inttoptr(LLVM.ConstantInt(Base.reinterpret(UInt, JTy.layout)), lptr)
end
@assert !isa(ty, LLVM.ConstantExpr)
@@ -893,7 +893,7 @@ end
function emit_type_layout_elsz!(B::LLVM.IRBuilder, @nospecialize(ty::LLVM.Value))
legal, JTy = absint(ty)
if legal
- JTy = unbind(JTy)
+ JTy = unbind(JTy)
@assert JTy isa Type
res = Compiler.datatype_layoutsize(JTy)
return LLVM.ConstantInt(res)
diff --git a/src/llvm/transforms.jl b/src/llvm/transforms.jl
index 5cecd431..7725027a 100644
--- a/src/llvm/transforms.jl
+++ b/src/llvm/transforms.jl
@@ -442,7 +442,7 @@ function memcpy_alloca_to_loadstore(mod::LLVM.Module)
@static if VERSION < v"1.11-"
else
legal2, obj = absint(src)
- if legal2 && is_memory_instance(unbind(obj))
+ if legal2 && is_memory_instance(unbind(obj))
metadata(src)["nonnull"] = MDNode(LLVM.Metadata[])
end
end
diff --git a/src/sugar.jl b/src/sugar.jl
index f244d1be..5ede8791 100644
--- a/src/sugar.jl
+++ b/src/sugar.jl
@@ -107,9 +107,9 @@ end
LLVM.position!(builder, exit)
LLVM.ret!(builder, obj)
-
+
Compiler.reinsert_gcmarker!(llvm_f)
- Compiler.JIT.prepare!(mod)
+ Compiler.JIT.prepare!(mod)
string(mod)
end
diff --git a/src/typeutils/jltypes.jl b/src/typeutils/jltypes.jl
index c2877660..7beaf3e4 100644
--- a/src/typeutils/jltypes.jl
+++ b/src/typeutils/jltypes.jl
@@ -412,13 +412,13 @@ end
@inline remove_innerty(::Type{<:BatchMixedDuplicated}) = MixedDuplicated
@inline function is_memory_instance(@nospecialize(obj))
- @static if VERSION < v"1.11"
- return false
- else
- if obj isa Memory
- return obj == typeof(obj).instance
+ @static if VERSION < v"1.11"
+ return false
+ else
+ if obj isa Memory
+ return obj == typeof(obj).instance
end
- return false
- end
+ return false
+ end
end
diff --git a/src/utils.jl b/src/utils.jl
index c0ca94de..1a05e158 100644
--- a/src/utils.jl
+++ b/src/utils.jl
@@ -92,7 +92,7 @@ end
export unsafe_to_ptr
# This mimicks literal_pointer_val / literal_pointer_val_slot
-function unsafe_to_llvm(B::LLVM.IRBuilder, @nospecialize(val); insert_name_if_not_exists::Union{String, Nothing}=nothing)::LLVM.Value
+function unsafe_to_llvm(B::LLVM.IRBuilder, @nospecialize(val); insert_name_if_not_exists::Union{String, Nothing} = nothing)::LLVM.Value
T_jlvalue = LLVM.StructType(LLVM.LLVMType[])
T_prjlvalue = LLVM.PointerType(T_jlvalue, Tracked)
T_prjlvalue_UT = LLVM.PointerType(T_jlvalue)
@@ -106,65 +106,65 @@ function unsafe_to_llvm(B::LLVM.IRBuilder, @nospecialize(val); insert_name_if_no
end
end
end
-
+
function setup_global(k, v)
- k0 = k
+ k0 = k
mod = LLVM.parent(LLVM.parent(LLVM.position(B)))
globs = LLVM.globals(mod)
if Base.haskey(globs, "ejl_" * k)
return globs["ejl_"*k]
end
-
- force_inactive = false
- if insert_name_if_not_exists isa String
- k = "inserted\$"*insert_name_if_not_exists
- if !haskey(Compiler.JuliaEnzymeNameMap, k)
- Compiler.JuliaEnzymeNameMap[k] = val
- end
- # Since the legacy behavior was to force inactive for global constants, we retain that here (for now)
- force_inactive = true
- end
- if Base.haskey(globs, "ejl_" * k)
- return globs["ejl_"*k]
+ force_inactive = false
+ if insert_name_if_not_exists isa String
+ k = "inserted\$" * insert_name_if_not_exists
+ if !haskey(Compiler.JuliaEnzymeNameMap, k)
+ Compiler.JuliaEnzymeNameMap[k] = val
end
+ # Since the legacy behavior was to force inactive for global constants, we retain that here (for now)
+ force_inactive = true
+ end
+
+ if Base.haskey(globs, "ejl_" * k)
+ return globs["ejl_" * k]
+ end
gv = LLVM.GlobalVariable(mod, T_jlvalue, "ejl_" * k, Tracked)
API.SetMD(gv, "enzyme_ta_norecur", LLVM.MDNode(LLVM.Metadata[]))
- inactive = force_inactive || Enzyme.Compiler.is_memory_instance(v)
- if !inactive && v isa Core.SimpleVector && length(v) == 0
- inactive = true
- end
- if !inactive && world isa UInt
+ inactive = force_inactive || Enzyme.Compiler.is_memory_instance(v)
+ if !inactive && v isa Core.SimpleVector && length(v) == 0
+ inactive = true
+ end
+ if !inactive && world isa UInt
legal, jTy, byref = Compiler.abs_typeof(gv, true)
if legal
curent_bb = position(B)
fn = LLVM.parent(curent_bb)
- state = Enzyme.Compiler.active_reg(jTy, world)
- inactive = state == Enzyme.Compiler.AnyState ||state == Enzyme.Compiler.ActiveState
+ state = Enzyme.Compiler.active_reg(jTy, world)
+ inactive = state == Enzyme.Compiler.AnyState ||state == Enzyme.Compiler.ActiveState
end
end
- if inactive
- API.SetMD(gv, "enzyme_inactive", LLVM.MDNode(LLVM.Metadata[]))
- end
+ if inactive
+ API.SetMD(gv, "enzyme_inactive", LLVM.MDNode(LLVM.Metadata[]))
+ end
return gv
end
for (k, v) in Compiler.JuliaGlobalNameMap
if v === val
- return setup_global(k, v)
+ return setup_global(k, v)
end
end
for (k, v) in Compiler.JuliaEnzymeNameMap
if v === val
- return setup_global(k, v)
+ return setup_global(k, v)
end
end
if insert_name_if_not_exists !== nothing
- return setup_global(insert_name_if_not_exists, val)
+ return setup_global(insert_name_if_not_exists, val)
end
# XXX: This prevents code from being runtime relocatable
diff --git a/test/locks.jl b/test/locks.jl
index 9e5daf67..8106a1f8 100644
--- a/test/locks.jl
+++ b/test/locks.jl
@@ -4,9 +4,9 @@ using Test
const my_cache_lock = ReentrantLock()
function my_lock()
- lock(my_cache_lock);
- unlock(my_cache_lock);
- return nothing
+ lock(my_cache_lock)
+ unlock(my_cache_lock)
+ return nothing
end
@testset "Lock forward" begin |
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/19791982609/artifacts/4714998493. |
vchuravy
reviewed
Nov 25, 2025
src/compiler/orcv2.jl
Outdated
|
|
||
| get_tm() = tm[] | ||
| get_jit() = jit[].jit | ||
| get_dylib() = dylib[] |
Member
There was a problem hiding this comment.
I would just define it as:
get_dylib() = JITDylib(lljit)
Member
Author
There was a problem hiding this comment.
it was used [I tried to do the llvm.define when we added to our list, but that led to issues], so now just doing a pass over ir doing the rewrite.
removed since
Member
Author
|
@vchuravy interestingly [on a debug build] I can't repro the ci failures. any luck on your side? |
wsmoses
commented
Nov 26, 2025
| continue | ||
| end | ||
| _, ogname, load1, initaddr = split(LLVM.name(g), "\$") | ||
|
|
Member
Author
There was a problem hiding this comment.
@vchuravy here we have the original name of the global variable that julia emitted it as [as well as the initial address/etc]. If there's any way we can lookup a portable version of this from the name, we have our relocatable code. E.g. if I could cglobal(:jl_global_7353) [where just the name was consistent but the runtime pointer differed, we'd be done]
Member
There was a problem hiding this comment.
No, we can't rely on names. Names are currently session local and in the future compilation local
On the Julia side this rougly looks like:
We are building a list of pointers (GOT like) that will be initialized during the loading process, we create a side table of Julia Object => Index in GOT, which will be serialized during the saving of a package image, upon loading we deserialize that table (basically an IDDict) and populate the GOT.
Now this might be something GPUCompiler doesn't yet fully expose.
Member
Author
There was a problem hiding this comment.
Yeah if we can somehow get access to that dict directly we could replace this with the correct object per session.
We actually store all the objects right now in our own dict, but it then begs the question of how we keep that dict relocatable to different sessions
Member
Author
|
@vchuravy @gbaraldi @oscardssmith are there any jl_global's passed as a ptr{ptr{jlvaluet}} or ptr{jlaluet} that are different to load dependeing on what thread you're on? |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2810 +/- ##
==========================================
+ Coverage 67.79% 67.83% +0.03%
==========================================
Files 58 58
Lines 20723 20801 +78
==========================================
+ Hits 14050 14111 +61
- Misses 6673 6690 +17 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Member
I don't understand the question? The only object that is different to load is the ptls |
Member
Author
|
Resolved ish, but a question I have is some jlvaluet are passed a s a pointer to the value others as a pointer to the pointer. Sometimes these aren't initialized and the inner ptr is null and or the final value is nothing. I've reverse engineered that behavior and we don't load those such values but is good to confirm |
Member
|
So GPU Compiler currently does: https://github.com/JuliaGPU/GPUCompiler.jl/blob/531dccd906cb14b9ff7fc7951ab13243bda24433/src/jlgen.jl#L751-L774 But I think we would need https://github.com/JuliaLang/julia/blob/ee6bb20bc86ec6fc9905e2769293ce1da4c0c4cf/src/aotcompile.cpp#L112-L143 to get a mapping from GlobalVariable to Julia value. |
Member
Could those be jl_binding_t? https://github.com/JuliaLang/julia/blob/ee6bb20bc86ec6fc9905e2769293ce1da4c0c4cf/src/aotcompile.cpp#L128 |
Member
Author
|
No they are not and we explicitly check for those (which appear to only happen on 1.10) |
Member
Author
|
okay figured it out, we needed to add an atomic to the load |
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
2 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.
No description provided.