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

Jump to bottom

fixed Float16 from Float64 and BigFloat #42837

Merged
merged 7 commits into from
Nov 4, 2021
Merged

fixed Float16 from Float64 and BigFloat #42837

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
02f4424
fixed Float16 from Float64 and BigFloat
oscardssmith Oct 28, 2021
847b086
fix impl, add tests
oscardssmith Oct 29, 2021
055bf60
fix impl, add tests
oscardssmith Oct 29, 2021
abfb719
comment subnormal mask
oscardssmith Oct 30, 2021
b821022
Julia isn't C
oscardssmith Nov 1, 2021
352aef7
remove stdio
oscardssmith Nov 1, 2021
41533ef
fix nextfloat 0.0
oscardssmith Nov 1, 2021
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
Prev Previous commit
fix nextfloat 0.0
  • Loading branch information
@oscardssmith
oscardssmith committed Nov 1, 2021
commit 41533ef17cde7e618327399446d5cb90d2b3acad
5 changes: 4 additions & 1 deletion base/mpfr.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -344,7 +344,10 @@ function Float16(x::BigFloat) :: Float16
if (resi&0x7fffffff) < 0x38800000 # if Float16(res) is subnormal
#shift so that the mantissa lines up where it would for normal Float16
shift = 113-((resi & 0x7f800000)>>23)
shift<23 && (resi >>= shift)
if shift<23
resi |= 0x0080_0000 # set implicit bit
resi >>= shift
end
end
if (resi & 0x1fff == 0x1000) # if we are halfway between 2 Float16 values
vchuravy marked this conversation as resolved.
Show resolved Hide resolved
# adjust the value by 1 ULP in the direction that will make Float16(res) give the right answer
Expand Down
4 changes: 3 additions & 1 deletion src/runtime_intrinsics.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,8 +211,10 @@ JL_DLLEXPORT uint16_t __truncdfhf2(double param)
if ((resi&0x7fffffffu) < 0x38800000u){ // if Float16(res) is subnormal
// shift so that the mantissa lines up where it would for normal Float16
uint32_t shift = 113u-((resi & 0x7f800000u)>>23u);
if (shift<23u)
if (shift<23u) {
resi |= 0x00800000; // set implicit bit
resi >>= shift;
}
}
if ((resi & 0x1fffu) == 0x1000u) { // if we are halfway between 2 Float16 values
memcpy(&resi, &res, sizeof(res));
Expand Down
5 changes: 2 additions & 3 deletions test/float16.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -209,16 +209,15 @@ const minsubf16_32 = Float32(minsubf16)
for i in 1:2^16
f = reinterpret(Float16, UInt16(i-1))
isfinite(f) || continue
abs(f)<=eps(f) && continue
if f < 0
epsdown = T(eps(f))/2
epsup = issubnormal(f) ? epsdown : T(eps(nextfloat(f)))/2
else
epsup = T(eps(f))/2
epsdown = issubnormal(f) ? epsup : T(eps(prevfloat(f)))/2
end
@test isequal(f, Float16(nextfloat(T(f) - epsdown)))
@test isequal(f, Float16(prevfloat(T(f) + epsup)))
@test isequal(f*(-1)^(f === Float16(0)), Float16(nextfloat(T(f) - epsdown)))
@test isequal(f*(-1)^(f === -Float16(0)), Float16(prevfloat(T(f) + epsup)))
@test isequal(prevfloat(f), Float16(prevfloat(T(f) - epsdown)))
@test isequal(nextfloat(f), Float16(nextfloat(T(f) + epsup)))
end
Expand Down