Skip to content

Add complex dtype support to op_sum #10559

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
merged 1 commit into from
May 1, 2025
Merged

Add complex dtype support to op_sum #10559

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
47 changes: 39 additions & 8 deletions kernels/portable/cpu/op_sum.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,25 +52,56 @@ Tensor& sum_dim_out(
}
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "sum.IntList_out";
ET_SWITCH_REALHBBF16_TYPES(in.scalar_type(), ctx, op_name, CTYPE_IN, [&] {
ET_SWITCH_REALHBBF16_TYPES(out.scalar_type(), ctx, op_name, CTYPE_OUT, [&] {
CTYPE_OUT* out_data = out.mutable_data_ptr<CTYPE_OUT>();

if (executorch::runtime::isComplexType(in.scalar_type())) {
ET_KERNEL_CHECK(
ctx, in.scalar_type() == out.scalar_type(), InvalidArgument, out);

ET_SWITCH_COMPLEXH_TYPES(in.scalar_type(), ctx, op_name, CTYPE, [&] {
CTYPE* out_data = out.mutable_data_ptr<CTYPE>();
const bool success = parallel_for_each_reduce_over_dim_list_output_index(
in, dim_list, out, [&](const auto begin, const auto end) {
for (const auto out_ix : c10::irange(begin, end)) {
CTYPE_OUT sum = 0;
CTYPE sum(0, 0);
if (plan.has_value()) {
sum = plan->execute<CTYPE_IN, CTYPE_OUT>(
[](CTYPE_IN v) { return static_cast<CTYPE_OUT>(v); },
[](CTYPE_OUT outv, CTYPE_OUT acc) { return acc + outv; },
sum = plan->execute<CTYPE, CTYPE>(
[](CTYPE v) { return v; },
[](CTYPE outv, CTYPE acc) { return acc + outv; },
out_ix);
}
out_data[out_ix] = sum;
}
});
ET_KERNEL_CHECK_MSG(ctx, success, Internal, , "parallel_for failed");
});
});
} else {
ET_SWITCH_REALHBBF16_TYPES(in.scalar_type(), ctx, op_name, CTYPE_IN, [&] {
ET_SWITCH_REALHBBF16_TYPES(
out.scalar_type(), ctx, op_name, CTYPE_OUT, [&] {
CTYPE_OUT* out_data = out.mutable_data_ptr<CTYPE_OUT>();
const bool success =
parallel_for_each_reduce_over_dim_list_output_index(
in, dim_list, out, [&](const auto begin, const auto end) {
for (const auto out_ix : c10::irange(begin, end)) {
CTYPE_OUT sum = 0;
if (plan.has_value()) {
sum = plan->execute<CTYPE_IN, CTYPE_OUT>(
[](CTYPE_IN v) {
return static_cast<CTYPE_OUT>(v);
},
[](CTYPE_OUT outv, CTYPE_OUT acc) {
return acc + outv;
},
out_ix);
}
out_data[out_ix] = sum;
}
});
ET_KERNEL_CHECK_MSG(
ctx, success, Internal, , "parallel_for failed");
});
});
}

return out;
}
Expand Down
89 changes: 89 additions & 0 deletions kernels/test/op_sum_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,85 @@ class OpSumOutTest : public OperatorTest {
self, optional_dim_list, /*keepdim=*/false, dtype, out));
}

template <typename CTYPE, ScalarType DTYPE>
void test_complex_dtype() {
TensorFactory<DTYPE> tf;

Tensor self = tf.make(
{2, 3, 2},
{CTYPE(1, 1),
CTYPE(2, 2),
CTYPE(3, 3),
CTYPE(4, 4),
CTYPE(5, 5),
CTYPE(6, 6),

CTYPE(7, 7),
CTYPE(8, 8),
CTYPE(9, 9),
CTYPE(10, 10),
CTYPE(11, 11),
CTYPE(12, 12)});

Tensor out1 = tf.make(
{2, 3, 1},
{
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
});
int64_t dims_1[1] = {2};
optional<ArrayRef<int64_t>> dim_list1{ArrayRef<int64_t>{dims_1, 1}};
optional<ScalarType> dtype = DTYPE;

op_sum_intlist_out(self, dim_list1, true, dtype, out1);

Tensor expected1 = tf.make(
{2, 3, 1},
{CTYPE(3, 3),
CTYPE(7, 7),
CTYPE(11, 11),

CTYPE(15, 15),
CTYPE(19, 19),
CTYPE(23, 23)});

EXPECT_TENSOR_CLOSE(out1, expected1);

Tensor out2 = tf.make(
{2, 1, 2},
{
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
CTYPE(0, 0),
});
int64_t dims_2[1] = {1};
optional<ArrayRef<int64_t>> dim_list2{ArrayRef<int64_t>{dims_2, 1}};

op_sum_intlist_out(self, dim_list2, true, dtype, out2);

Tensor expected2 = tf.make(
{2, 1, 2}, {CTYPE(9, 9), CTYPE(12, 12), CTYPE(27, 27), CTYPE(30, 30)});
EXPECT_TENSOR_CLOSE(out2, expected2);

Tensor out3 = tf.make(
{1, 1, 1},
{
CTYPE(0, 0),
});
optional<ArrayRef<int64_t>> null_dim_list;

op_sum_intlist_out(self, null_dim_list, true, dtype, out3);

Tensor expected3 = tf.make({1, 1, 1}, {CTYPE(78, 78)});

EXPECT_TENSOR_CLOSE(out3, expected3);
}

template <ScalarType IN_DTYPE, ScalarType OUT_DTYPE>
void test_sum_dim_out_dtype() {
TensorFactory<IN_DTYPE> tf_in;
Expand Down Expand Up @@ -366,6 +445,16 @@ TEST_F(OpSumOutTest, TypeConversionTest) {
// clang-format on
}

TEST_F(OpSumOutTest, AllComplexDtypesSupported) {
#define TEST_ENTRY(ctype, dtype) test_complex_dtype<ctype, ScalarType::dtype>();
if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
ET_FORALL_COMPLEX_TYPES(TEST_ENTRY);
} else {
ET_FORALL_COMPLEXH_TYPES(TEST_ENTRY);
}
#undef TEST_ENTRY
}

TEST_F(OpSumOutTest, InfinityAndNANTest) {
TensorFactory<ScalarType::Float> tf_float;
// clang-format off
Expand Down
Loading