Skip to content

Add f16 support for modulus and norm #3258

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
christophe-murphy merged 10 commits into from
Mar 7, 2025
Merged

Add f16 support for modulus and norm #3258

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
385a283
Add support for f16 for modulus operations
umar456 Jun 7, 2022
d05c2c6
Update float math functions to use f<op>f functions to maintain types
umar456 Jun 7, 2022
7d9ed1d
Add additional binary tests for integer types
umar456 Jun 7, 2022
68b5505
Add tests for norm
umar456 Jun 10, 2022
c388d51
Add support for half for norm
umar456 Jun 10, 2022
821df19
Added tests for norm and modulus. Made consistent modulus for ints in…
edwinsolisf Feb 4, 2025
186b306
Added more mod tests
edwinsolisf Feb 18, 2025
faf943d
Updated documentation to reflect status quo of the mod and rem expect…
edwinsolisf Feb 28, 2025
37e2982
Merge branch 'master' into half_mod
christophe-murphy Mar 7, 2025
dfe2ab9
Update copyright
christophe-murphy Mar 7, 2025
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
Next Next commit
Added more mod tests
  • Loading branch information
@edwinsolisf
edwinsolisf committed Feb 18, 2025
commit 186b30636e4b4a00a73984dee6451ea3250e6e07
6 changes: 1 addition & 5 deletions src/backend/cpu/binary.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,11 +88,7 @@ LOGIC_CPLX_FN(double, af_or_t, ||)

template<typename T>
static T __mod(T lhs, T rhs) {
T res = lhs % rhs; // Same as other backends

// Does this break compatibility?
// return (res < 0) ? abs(rhs - res) : res;
return res;
return lhs % rhs; // Same as other backends
}

template<typename T>
Expand Down
39 changes: 26 additions & 13 deletions test/math.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,31 +149,44 @@ TEST(Math, Not) {

TEST(Math, Modulus) {
af::dim4 shape(2, 2);
std::vector<long long> aData{1, 1, 1, 1};
std::vector<long long> aData{3, 3, 3, 3};
std::vector<long long> bData{2, 2, 2, 2};

auto a = af::array(shape, aData.data(), afHost);
auto b = af::array(shape, bData.data(), afHost);
auto rem = a % b;
auto diff = rem - a;

auto neg_rem = -a % b;
auto neg_diff = neg_rem + a;

ASSERT_ARRAYS_EQ(af::constant(1, shape, s64), rem);
ASSERT_ARRAYS_EQ(af::constant(0, shape, s64), diff);

ASSERT_ARRAYS_EQ(af::constant(-1, shape, s64), neg_rem);
ASSERT_ARRAYS_EQ(af::constant(0, shape, s64), neg_diff);
}

Comment thread
christophe-murphy marked this conversation as resolved.
TEST(Math, ModulusHalf) {
TEST(Math, ModulusFloat) {
SUPPORTED_TYPE_CHECK(half_float::half);
auto a = af::constant(3, {2, 2}, af::dtype::f16);
auto b = af::constant(2, {2, 2}, af::dtype::f16);
auto a32 = af::constant(3, {2, 2}, af::dtype::f32);
auto b32 = af::constant(2, {2, 2}, af::dtype::f32);
auto rem32 = a32 % b32;
af::dim4 shape(2, 2);

auto a = af::constant(3, shape, af::dtype::f16);
auto b = af::constant(2, shape, af::dtype::f16);
auto a32 = af::constant(3, shape, af::dtype::f32);
auto b32 = af::constant(2, shape, af::dtype::f32);
auto a64 = af::constant(3, shape, af::dtype::f64);
auto b64 = af::constant(2, shape, af::dtype::f64);

auto rem = a % b;
auto rem32 = a32 % b32;
auto rem64 = a64 % b64;

auto neg_rem = -a % b;
auto neg_rem32 = -a32 % b32;
auto neg_rem64 = -a64 % b64;

ASSERT_ARRAYS_EQ(af::constant(1, shape, af::dtype::f16), rem);
ASSERT_ARRAYS_EQ(af::constant(1, shape, af::dtype::f32), rem32);
ASSERT_ARRAYS_EQ(af::constant(1, shape, af::dtype::f64), rem64);

ASSERT_ARRAYS_EQ(af::constant(-1, shape, af::dtype::f16), neg_rem);
ASSERT_ARRAYS_EQ(af::constant(-1, shape, af::dtype::f32), neg_rem32);
ASSERT_ARRAYS_EQ(af::constant(-1, shape, af::dtype::f64), neg_rem64);

ASSERT_ARRAYS_EQ(rem32.as(f16), rem);
}