Skip to content

Commit

Permalink
Fix process_bulk case in constexpr
Browse files Browse the repository at this point in the history 
Need else in if constexpr, otherwise the code below still requires
the implementation to have process_one, even if never called.
(GCC 12.3.0)
  • Loading branch information
@frankosterfeld
frankosterfeld committed Jun 21, 2023
1 parent 01c9174 commit 90a9bb4
Showing 1 changed file with 38 additions and 38 deletions.
76 changes: 38 additions & 38 deletions include/node.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -631,50 +631,50 @@ class node : protected std::tuple<Arguments...> {
const bool success = consume_readers(self(), samples_to_process);
forward_tags();
return success ? ret : work_return_t::ERROR;
}

using input_simd_types = meta::simdize<typename input_types::template apply<std::tuple>>;
using output_simd_types = meta::simdize<typename output_types::template apply<std::tuple>>;

std::integral_constant<std::size_t, (meta::simdize_size_v<input_simd_types> == 0 ? std::size_t(stdx::simd_abi::max_fixed_size<double>)
: std::min(std::size_t(stdx::simd_abi::max_fixed_size<double>), meta::simdize_size_v<input_simd_types> * 4))>
width{};

if constexpr ((is_sink_node or meta::simdize_size_v<output_simd_types> != 0) and ((is_source_node and requires(Derived &d) {
{ d.process_one_simd(width) };
}) or (meta::simdize_size_v<input_simd_types> != 0 and traits::node::can_process_simd<Derived>))) {
// SIMD loop
std::size_t i = 0;
for (; i + width <= samples_to_process; i += width) {
const auto &results = simdize_tuple_load_and_apply(width, input_spans, i, [&](const auto &...input_simds) { return invoke_process_one_simd(width, input_simds...); });
meta::tuple_for_each([i](auto &output_range, const auto &result) { result.copy_to(output_range.data() + i, stdx::element_aligned); }, writers_tuple, results);
}
simd_epilogue(width, [&](auto w) {
if (i + w <= samples_to_process) {
const auto results = simdize_tuple_load_and_apply(w, input_spans, i, [&](auto &&...input_simds) { return invoke_process_one_simd(w, input_simds...); });
meta::tuple_for_each([i](auto &output_range, auto &result) { result.copy_to(output_range.data() + i, stdx::element_aligned); }, writers_tuple, results);
i += w;
}
});
} else {
// Non-SIMD loop
for (std::size_t i = 0; i < samples_to_process; ++i) {
const auto results = std::apply([this, i](auto &...inputs) { return this->invoke_process_one(inputs[i]...); }, input_spans);
meta::tuple_for_each([i](auto &output_range, auto &result) { output_range[i] = std::move(result); }, writers_tuple, results);
using input_simd_types = meta::simdize<typename input_types::template apply<std::tuple>>;
using output_simd_types = meta::simdize<typename output_types::template apply<std::tuple>>;

std::integral_constant<std::size_t, (meta::simdize_size_v<input_simd_types> == 0 ? std::size_t(stdx::simd_abi::max_fixed_size<double>)
: std::min(std::size_t(stdx::simd_abi::max_fixed_size<double>), meta::simdize_size_v<input_simd_types> * 4))>
width{};

if constexpr ((is_sink_node or meta::simdize_size_v<output_simd_types> != 0) and ((is_source_node and requires(Derived &d) {
{ d.process_one_simd(width) };
}) or (meta::simdize_size_v<input_simd_types> != 0 and traits::node::can_process_simd<Derived>))) {
// SIMD loop
std::size_t i = 0;
for (; i + width <= samples_to_process; i += width) {
const auto &results = simdize_tuple_load_and_apply(width, input_spans, i, [&](const auto &...input_simds) { return invoke_process_one_simd(width, input_simds...); });
meta::tuple_for_each([i](auto &output_range, const auto &result) { result.copy_to(output_range.data() + i, stdx::element_aligned); }, writers_tuple, results);
}
simd_epilogue(width, [&](auto w) {
if (i + w <= samples_to_process) {
const auto results = simdize_tuple_load_and_apply(w, input_spans, i, [&](auto &&...input_simds) { return invoke_process_one_simd(w, input_simds...); });
meta::tuple_for_each([i](auto &output_range, auto &result) { result.copy_to(output_range.data() + i, stdx::element_aligned); }, writers_tuple, results);
i += w;
}
});
} else {
// Non-SIMD loop
for (std::size_t i = 0; i < samples_to_process; ++i) {
const auto results = std::apply([this, i](auto &...inputs) { return this->invoke_process_one(inputs[i]...); }, input_spans);
meta::tuple_for_each([i](auto &output_range, auto &result) { output_range[i] = std::move(result); }, writers_tuple, results);
}
}
}

write_to_outputs(samples_to_process, writers_tuple);
write_to_outputs(samples_to_process, writers_tuple);

const bool success = consume_readers(self(), samples_to_process);
const bool success = consume_readers(self(), samples_to_process);

#ifdef _DEBUG
if (!success) {
fmt::print("Node {} failed to consume {} values from inputs\n", self().name(), samples_to_process);
#ifdef _DEBUG
if (!success) {
fmt::print("Node {} failed to consume {} values from inputs\n", self().name(), samples_to_process);
}
#endif
forward_tags();
return success ? work_return_t::OK : work_return_t::ERROR;
}
#endif
forward_tags();
return success ? work_return_t::OK : work_return_t::ERROR;
} // end: work_return_t work() noexcept { ..}
};

Expand Down

0 comments on commit 90a9bb4

Please sign in to comment.