Skip to content

Make UseRefs more memory efficient #44495

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.

Sign up for GitHub

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

Merged
Keno merged 1 commit into from
Mar 15, 2022
Merged

Make UseRefs more memory efficient #44495

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
123 changes: 63 additions & 60 deletions base/compiler/ssair/ir.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -332,70 +332,65 @@ end

const AnySSAValue = Union{SSAValue, OldSSAValue, NewSSAValue}

mutable struct UseRef
mutable struct UseRefIterator
stmt::Any
op::Int
UseRef(@nospecialize(a)) = new(a, 0)
UseRef(@nospecialize(a), op::Int) = new(a, op)
end
struct UseRefIterator
use::Tuple{UseRef, Nothing}
relevant::Bool
UseRefIterator(@nospecialize(a), relevant::Bool) = new((UseRef(a), nothing), relevant)
UseRefIterator(@nospecialize(a), relevant::Bool) = new(a, relevant)
end
getindex(it::UseRefIterator) = it.use[1].stmt
getindex(it::UseRefIterator) = it.stmt

# TODO: stack-allocation
#struct UseRef
# urs::UseRefIterator
# use::Int
#end
struct UseRef
urs::UseRefIterator
op::Int
UseRef(urs::UseRefIterator) = new(urs, 0)
UseRef(urs::UseRefIterator, op::Int) = new(urs, op)
end

struct OOBToken end; const OOB_TOKEN = OOBToken()
struct UndefToken end; const UNDEF_TOKEN = UndefToken()

function getindex(x::UseRef)
stmt = x.stmt
@noinline function _useref_getindex(@nospecialize(stmt), op::Int)
if isa(stmt, Expr) && stmt.head === :(=)
rhs = stmt.args[2]
if isa(rhs, Expr)
if is_relevant_expr(rhs)
x.op > length(rhs.args) && return OOB_TOKEN
return rhs.args[x.op]
op > length(rhs.args) && return OOB_TOKEN
return rhs.args[op]
end
end
x.op == 1 || return OOB_TOKEN
op == 1 || return OOB_TOKEN
return rhs
elseif isa(stmt, Expr) # @assert is_relevant_expr(stmt)
x.op > length(stmt.args) && return OOB_TOKEN
return stmt.args[x.op]
op > length(stmt.args) && return OOB_TOKEN
return stmt.args[op]
elseif isa(stmt, GotoIfNot)
x.op == 1 || return OOB_TOKEN
op == 1 || return OOB_TOKEN
return stmt.cond
elseif isa(stmt, ReturnNode)
isdefined(stmt, :val) || return OOB_TOKEN
x.op == 1 || return OOB_TOKEN
op == 1 || return OOB_TOKEN
return stmt.val
elseif isa(stmt, PiNode)
isdefined(stmt, :val) || return OOB_TOKEN
x.op == 1 || return OOB_TOKEN
op == 1 || return OOB_TOKEN
return stmt.val
elseif isa(stmt, UpsilonNode)
isdefined(stmt, :val) || return OOB_TOKEN
x.op == 1 || return OOB_TOKEN
op == 1 || return OOB_TOKEN
return stmt.val
elseif isa(stmt, PhiNode)
x.op > length(stmt.values) && return OOB_TOKEN
isassigned(stmt.values, x.op) || return UNDEF_TOKEN
return stmt.values[x.op]
op > length(stmt.values) && return OOB_TOKEN
isassigned(stmt.values, op) || return UNDEF_TOKEN
return stmt.values[op]
elseif isa(stmt, PhiCNode)
x.op > length(stmt.values) && return OOB_TOKEN
isassigned(stmt.values, x.op) || return UNDEF_TOKEN
return stmt.values[x.op]
op > length(stmt.values) && return OOB_TOKEN
isassigned(stmt.values, op) || return UNDEF_TOKEN
return stmt.values[op]
else
return OOB_TOKEN
end
end
@inline getindex(x::UseRef) = _useref_getindex(x.urs.stmt, x.op)
Copy link
Member

@vtjnash vtjnash Mar 7, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I had hoped our calling convention would have been enough already to make this inlining awkwardness unnecessary. What causes it to be needed?

Copy link
Member Author

@Keno Keno Mar 7, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We want UseRef to be SROA'd, so that UseRefIterator can be SROA'd. Without it UseRefIterator gets allocated.


function is_relevant_expr(e::Expr)
return e.head in (:call, :invoke, :invoke_modify,
Expand All @@ -407,45 +402,49 @@ function is_relevant_expr(e::Expr)
:new_opaque_closure)
end

function setindex!(x::UseRef, @nospecialize(v))
stmt = x.stmt
@noinline function _useref_setindex!(@nospecialize(stmt), op::Int, @nospecialize(v))
if isa(stmt, Expr) && stmt.head === :(=)
rhs = stmt.args[2]
if isa(rhs, Expr)
if is_relevant_expr(rhs)
x.op > length(rhs.args) && throw(BoundsError())
rhs.args[x.op] = v
return v
op > length(rhs.args) && throw(BoundsError())
rhs.args[op] = v
return stmt
end
end
x.op == 1 || throw(BoundsError())
op == 1 || throw(BoundsError())
stmt.args[2] = v
elseif isa(stmt, Expr) # @assert is_relevant_expr(stmt)
x.op > length(stmt.args) && throw(BoundsError())
stmt.args[x.op] = v
op > length(stmt.args) && throw(BoundsError())
stmt.args[op] = v
elseif isa(stmt, GotoIfNot)
x.op == 1 || throw(BoundsError())
x.stmt = GotoIfNot(v, stmt.dest)
op == 1 || throw(BoundsError())
stmt = GotoIfNot(v, stmt.dest)
elseif isa(stmt, ReturnNode)
x.op == 1 || throw(BoundsError())
x.stmt = typeof(stmt)(v)
op == 1 || throw(BoundsError())
stmt = typeof(stmt)(v)
elseif isa(stmt, UpsilonNode)
x.op == 1 || throw(BoundsError())
x.stmt = typeof(stmt)(v)
op == 1 || throw(BoundsError())
stmt = typeof(stmt)(v)
elseif isa(stmt, PiNode)
x.op == 1 || throw(BoundsError())
x.stmt = typeof(stmt)(v, stmt.typ)
op == 1 || throw(BoundsError())
stmt = typeof(stmt)(v, stmt.typ)
elseif isa(stmt, PhiNode)
x.op > length(stmt.values) && throw(BoundsError())
isassigned(stmt.values, x.op) || throw(BoundsError())
stmt.values[x.op] = v
op > length(stmt.values) && throw(BoundsError())
isassigned(stmt.values, op) || throw(BoundsError())
stmt.values[op] = v
elseif isa(stmt, PhiCNode)
x.op > length(stmt.values) && throw(BoundsError())
isassigned(stmt.values, x.op) || throw(BoundsError())
stmt.values[x.op] = v
op > length(stmt.values) && throw(BoundsError())
isassigned(stmt.values, op) || throw(BoundsError())
stmt.values[op] = v
else
throw(BoundsError())
end
return stmt
end

@inline function setindex!(x::UseRef, @nospecialize(v))
x.urs.stmt = _useref_setindex!(x.urs.stmt, x.op, v)
return x
end

Expand All @@ -456,18 +455,22 @@ function userefs(@nospecialize(x))
return UseRefIterator(x, relevant)
end

iterate(it::UseRefIterator) = (it.use[1].op = 0; iterate(it, nothing))
@noinline function iterate(it::UseRefIterator, ::Nothing)
it.relevant || return nothing
use = it.use[1]
@noinline function _advance(@nospecialize(stmt), op)
while true
use.op += 1
y = use[]
op += 1
y = _useref_getindex(stmt, op)
y === OOB_TOKEN && return nothing
y === UNDEF_TOKEN || return it.use
y === UNDEF_TOKEN || return op
end
end

@inline function iterate(it::UseRefIterator, op::Int=0)
it.relevant || return nothing
op = _advance(it.stmt, op)
op === nothing && return nothing
return (UseRef(it, op), op)
end

# This function is used from the show code, which may have a different
# `push!`/`used` type since it's in Base.
function scan_ssa_use!(push!, used, @nospecialize(stmt))
Expand Down