Skip to content
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

fix #29885, make grisu digit buffer task-local #29907

Merged
merged 1 commit into from
Nov 28, 2018
Merged

fix #29885, make grisu digit buffer task-local #29907

Show file tree
Hide file tree
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
23 changes: 17 additions & 6 deletions base/grisu/grisu.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,17 +19,26 @@ include("grisu/bignum.jl")
const DIGITS = Vector{UInt8}(undef, 309+17)
const BIGNUMS = [Bignums.Bignum(),Bignums.Bignum(),Bignums.Bignum(),Bignums.Bignum()]

# thread-safe code should use a per-thread DIGITS buffer DIGITSs[Threads.threadid()]
# NOTE: DIGITS[s] is deprecated; you should use getbuf() instead.
const DIGITSs = [DIGITS]
const BIGNUMSs = [BIGNUMS]
function __init__()
Threads.resize_nthreads!(DIGITSs)
Threads.resize_nthreads!(BIGNUMSs)
end

function getbuf()
tls = task_local_storage()
d = get(tls, :DIGITS, nothing)
if d === nothing
d = Vector{UInt8}(undef, 309+17)
tls[:DIGITS] = d
end
return d::Vector{UInt8}
end

"""
(len, point, neg) = Grisu.grisu(v::AbstractFloat, mode, requested_digits,
buffer=DIGITSs[Threads.threadid()], bignums=BIGNUMSs[Threads.threadid()])
(len, point, neg) = Grisu.grisu(v::AbstractFloat, mode, requested_digits, [buffer], [bignums])

Convert the number `v` to decimal using the Grisu algorithm.

Expand All @@ -38,7 +47,7 @@ Convert the number `v` to decimal using the Grisu algorithm.
- `Grisu.FIXED`: round to `requested_digits` digits.
- `Grisu.PRECISION`: round to `requested_digits` significant digits.

The characters are written as bytes to `buffer`, with a terminating NUL byte, and `bignums` are used internally as part of the correction step.
The characters are written as bytes to `buffer`, with a terminating NUL byte, and `bignums` are used internally as part of the correction step. You can call `Grisu.getbuf()` to obtain a suitable task-local buffer.

The returned tuple contains:

Expand Down Expand Up @@ -94,7 +103,8 @@ function _show(io::IO, x::AbstractFloat, mode, n::Int, typed, compact)
return
end
typed && isa(x,Float16) && print(io, "Float16(")
(len,pt,neg),buffer = grisu(x,mode,n),DIGITSs[Threads.threadid()]
buffer = getbuf()
len, pt, neg = grisu(x,mode,n,buffer)
pdigits = pointer(buffer)
if mode == PRECISION
while len > 1 && buffer[len] == 0x30
Expand Down Expand Up @@ -182,7 +192,8 @@ function _print_shortest(io::IO, x::AbstractFloat, dot::Bool, mode, n::Int)
isnan(x) && return print(io, "NaN")
x < 0 && print(io,'-')
isinf(x) && return print(io, "Inf")
(len,pt,neg),buffer = grisu(x,mode,n),DIGITSs[Threads.threadid()]
buffer = getbuf()
len, pt, neg = grisu(x,mode,n,buffer)
pdigits = pointer(buffer)
e = pt-len
k = -9<=e<=9 ? 1 : 2
Expand Down
52 changes: 26 additions & 26 deletions base/printf.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -213,7 +213,7 @@ end

# note: if print_fixed is changed, print_fixed_width should be changed accordingly
function print_fixed(out, precision, pt, ndigits, trailingzeros=true)
pdigits = pointer(DIGITSs[Threads.threadid()])
pdigits = pointer(Grisu.getbuf())
if pt <= 0
# 0.0dddd0
print(out, '0')
Expand Down Expand Up @@ -346,7 +346,7 @@ function gen_d(flags::String, width::Int, precision::Int, c::Char)
push!(blk.args, pad(width-1, zeros, '0'))
end
# print integer
push!(blk.args, :(unsafe_write(out, pointer(DIGITSs[Threads.threadid()]), pt)))
push!(blk.args, :(unsafe_write(out, pointer($Grisu.getbuf()), pt)))
# print padding
if padding !== nothing && '-' in flags
push!(blk.args, pad(width-precision, padding, ' '))
Expand Down Expand Up @@ -405,7 +405,7 @@ function gen_f(flags::String, width::Int, precision::Int, c::Char)
if precision > 0
push!(blk.args, :(print_fixed(out,$precision,pt,len)))
else
push!(blk.args, :(unsafe_write(out, pointer(DIGITSs[Threads.threadid()]), len)))
push!(blk.args, :(unsafe_write(out, pointer($Grisu.getbuf()), len)))
push!(blk.args, :(while pt >= (len+=1) print(out,'0') end))
'#' in flags && push!(blk.args, :(print(out, '.')))
end
Expand Down Expand Up @@ -438,9 +438,9 @@ function gen_e(flags::String, width::Int, precision::Int, c::Char, inside_g::Boo
end
# interpret the number
if precision < 0; precision = 6; end
ndigits = min(precision+1,length(DIGITSs[Threads.threadid()])-1)
ndigits = min(precision+1,length(Grisu.getbuf())-1)
push!(blk.args, :((do_out, args) = ini_dec(out,$x,$ndigits, $flags, $width, $precision, $c)))
push!(blk.args, :(digits = DIGITSs[Threads.threadid()]))
push!(blk.args, :(digits = $Grisu.getbuf()))
ifblk = Expr(:if, :do_out, Expr(:block))
push!(blk.args, ifblk)
blk = ifblk.args[2]
Expand Down Expand Up @@ -554,10 +554,10 @@ function gen_a(flags::String, width::Int, precision::Int, c::Char)
if precision < 0
push!(blk.args, :((do_out, args) = $fn(out,$x, $flags, $width, $precision, $c)))
else
ndigits = min(precision+1,length(DIGITSs[Threads.threadid()])-1)
ndigits = min(precision+1,length(Grisu.getbuf())-1)
push!(blk.args, :((do_out, args) = $fn(out,$x,$ndigits, $flags, $width, $precision, $c)))
end
push!(blk.args, :(digits = DIGITSs[Threads.threadid()]))
push!(blk.args, :(digits = $Grisu.getbuf()))
ifblk = Expr(:if, :do_out, Expr(:block))
push!(blk.args, ifblk)
blk = ifblk.args[2]
Expand Down Expand Up @@ -750,7 +750,7 @@ function gen_g(flags::String, width::Int, precision::Int, c::Char)
#
x, ex, blk = special_handler(flags,width)
if precision < 0; precision = 6; end
ndigits = min(precision+1,length(DIGITSs[Threads.threadid()])-1)
ndigits = min(precision+1,length(Grisu.getbuf())-1)
# See if anyone else wants to handle it
push!(blk.args, :((do_out, args) = ini_dec(out,$x,$ndigits, $flags, $width, $precision, $c)))
ifblk = Expr(:if, :do_out, Expr(:block))
Expand Down Expand Up @@ -859,7 +859,7 @@ end

function decode_oct(d::Integer)
neg, x = handlenegative(d)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
@handle_zero x digits
pt = i = div((sizeof(x)<<3)-leading_zeros(x)+2,3)
while i > 0
Expand All @@ -874,7 +874,7 @@ function decode_0ct(d::Integer)
neg, x = handlenegative(d)
# doesn't need special handling for zero
pt = i = div((sizeof(x)<<3)-leading_zeros(x)+5,3)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
while i > 0
digits[i] = 48+(x&0x7)
x >>= 3
Expand All @@ -885,7 +885,7 @@ end

function decode_dec(d::Integer)
neg, x = handlenegative(d)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
@handle_zero x digits
pt = i = Base.ndigits0z(x)
while i > 0
Expand All @@ -898,7 +898,7 @@ end

function decode_hex(d::Integer, symbols::AbstractArray{UInt8,1})
neg, x = handlenegative(d)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
@handle_zero x digits
pt = i = (sizeof(x)<<1)-(leading_zeros(x)>>2)
while i > 0
Expand All @@ -918,7 +918,7 @@ decode_HEX(x::Integer) = decode_hex(x,HEX_symbols)
function decode(b::Int, x::BigInt)
neg = x.size < 0
pt = Base.ndigits(x, base=abs(b))
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
length(digits) < pt+1 && resize!(digits, pt+1)
neg && (x.size = -x.size)
GMP.MPZ.get_str!(digits, b, x)
Expand All @@ -932,7 +932,7 @@ decode_HEX(x::BigInt) = decode(-16, x)

function decode_0ct(x::BigInt)
neg = x.size < 0
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
digits[1] = '0'
if x.size == 0
return Int32(1), Int32(1), neg
Expand Down Expand Up @@ -961,12 +961,12 @@ end
#

function decode_dec(x::SmallFloatingPoint)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
if x == 0.0
digits[1] = '0'
return (Int32(1), Int32(1), false)
end
len,pt,neg = grisu(x,Grisu.FIXED,0)
len,pt,neg = grisu(x,Grisu.FIXED,0,digits)
if len == 0
digits[1] = '0'
return (Int32(1), Int32(1), false)
Expand All @@ -991,9 +991,9 @@ fix_dec(x::Real, n::Int) = fix_dec(float(x),n)
fix_dec(x::Integer, n::Int) = decode_dec(x)

function fix_dec(x::SmallFloatingPoint, n::Int)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
if n > length(digits)-1; n = length(digits)-1; end
len,pt,neg = grisu(x,Grisu.FIXED,n)
len,pt,neg = grisu(x,Grisu.FIXED,n,digits)
if len == 0
digits[1] = '0'
return (Int32(1), Int32(1), neg)
Expand All @@ -1013,7 +1013,7 @@ ini_dec(x::Real, n::Int) = ini_dec(float(x),n)
function ini_dec(d::Integer, n::Int)
neg, x = handlenegative(d)
k = ndigits(x)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
if k <= n
pt = k
for i = k:-1:1
Expand Down Expand Up @@ -1045,24 +1045,24 @@ end

function ini_dec(x::SmallFloatingPoint, n::Int)
if x == 0.0
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), DIGITSs[Threads.threadid()], '0', n)
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), Grisu.getbuf(), '0', n)
return Int32(1), Int32(1), signbit(x)
else
len,pt,neg = grisu(x,Grisu.PRECISION,n)
len,pt,neg = grisu(x,Grisu.PRECISION,n,Grisu.getbuf())
end
return Int32(len), Int32(pt), neg
end

function ini_dec(x::BigInt, n::Int)
if x.size == 0
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), DIGITSs[Threads.threadid()], '0', n)
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), Grisu.getbuf(), '0', n)
return Int32(1), Int32(1), false
end
d = Base.ndigits0z(x)
if d <= n
info = decode_dec(x)
d == n && return info
p = convert(Ptr{Cvoid}, DIGITSs[Threads.threadid()]) + info[2]
p = convert(Ptr{Cvoid}, Grisu.getbuf()) + info[2]
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), p, '0', n - info[2])
return info
end
Expand All @@ -1081,7 +1081,7 @@ ini_hex(x::Real, symbols::AbstractArray{UInt8,1}) = ini_hex(float(x), symbols)

function ini_hex(x::SmallFloatingPoint, n::Int, symbols::AbstractArray{UInt8,1})
x = Float64(x)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
if x == 0.0
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), digits, '0', n)
return Int32(1), Int32(0), signbit(x)
Expand All @@ -1107,7 +1107,7 @@ end

function ini_hex(x::SmallFloatingPoint, symbols::AbstractArray{UInt8,1})
x = Float64(x)
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
if x == 0.0
ccall(:memset, Ptr{Cvoid}, (Ptr{Cvoid}, Cint, Csize_t), digits, '0', 1)
return Int32(1), Int32(0), signbit(x)
Expand Down Expand Up @@ -1174,7 +1174,7 @@ function bigfloat_printf(out, d::BigFloat, flags::String, width::Int, precision:
write(fmt, UInt8(0))
printf_fmt = take!(fmt)
@assert length(printf_fmt) == fmt_len
digits = DIGITSs[Threads.threadid()]
digits = Grisu.getbuf()
bufsiz = length(digits)
lng = ccall((:mpfr_snprintf,:libmpfr), Int32,
(Ptr{UInt8}, Culong, Ptr{UInt8}, Ref{BigFloat}...),
Expand Down
11 changes: 11 additions & 0 deletions test/grisu.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1751,3 +1751,14 @@ len,point,neg = Grisu.grisu(1.0, Grisu.FIXED, 0, buffer)
@test 1 >= len-1
@test "1" == unsafe_string(pointer(buffer))
@test !neg

# issue #29885
@sync let p = Pipe(), q = Pipe()
Base.link_pipe!(p, reader_supports_async=true, writer_supports_async=true)
Base.link_pipe!(q, reader_supports_async=true, writer_supports_async=true)
@async write(p, zeros(UInt8, 2^18))
@async (print(p, 12.345); close(p.in))
@async print(q, 9.8)
read(p, 2^18)
@test read(p, String) == "12.345"
end
Copy link
Member

Choose a reason for hiding this comment

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

wrap in @sync (so we'll collect any errors)