Skip to content

[SYCL] Implement the rest of geometric built-ins #8718

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 4 commits into from
Apr 11, 2023
Merged

[SYCL] Implement the rest of geometric built-ins #8718

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ac9d01b
[SYCL] Implement the rest of geometric built-ins
KornevNikita Mar 20, 2023
72751b3
Merge branch 'intel:sycl' into marray-geometric-funcs
KornevNikita Apr 5, 2023
eb1fcd3
Refactoring + adding tests from llvm-test-suite
dm-vodopyanov Apr 10, 2023
07d4b2f
Apply CR comments
dm-vodopyanov Apr 11, 2023
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
76 changes: 76 additions & 0 deletions sycl/include/sycl/builtins.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,18 @@ namespace detail {
template <class T, size_t N> vec<T, 2> to_vec2(marray<T, N> x, size_t start) {
return {x[start], x[start + 1]};
}
template <class T, size_t N> vec<T, N> to_vec(marray<T, N> x) {
vec<T, N> vec;
for (size_t i = 0; i < N; i++)
vec[i] = x[i];
return vec;
}
template <class T, int N> marray<T, N> to_marray(vec<T, N> x) {
marray<T, N> marray;
for (size_t i = 0; i < N; i++)
marray[i] = x[i];
return marray;
}
} // namespace detail

#ifdef __SYCL_DEVICE_ONLY__
Expand Down Expand Up @@ -1715,6 +1727,70 @@ fast_normalize(T p) __NOEXC {
return __sycl_std::__invoke_fast_normalize<T>(p);
}

// marray geometric functions

#define __SYCL_MARRAY_GEOMETRIC_FUNCTION_OVERLOAD_IMPL(NAME, ...) \
vec<detail::marray_element_t<T>, T::size()> result_v; \
result_v = NAME(__VA_ARGS__); \
return detail::to_marray(result_v);

template <typename T>
std::enable_if_t<detail::is_gencrossmarray<T>::value, T> cross(T p0,
T p1) __NOEXC {
__SYCL_MARRAY_GEOMETRIC_FUNCTION_OVERLOAD_IMPL(cross, detail::to_vec(p0),
detail::to_vec(p1))
}

template <typename T>
std::enable_if_t<detail::is_gengeomarray<T>::value, T> normalize(T p) __NOEXC {
__SYCL_MARRAY_GEOMETRIC_FUNCTION_OVERLOAD_IMPL(normalize, detail::to_vec(p))
}

template <typename T>
std::enable_if_t<detail::is_gengeomarrayfloat<T>::value, T>
fast_normalize(T p) __NOEXC {
__SYCL_MARRAY_GEOMETRIC_FUNCTION_OVERLOAD_IMPL(fast_normalize,
detail::to_vec(p))
}

#undef __SYCL_MARRAY_GEOMETRIC_FUNCTION_OVERLOAD_IMPL

#define __SYCL_MARRAY_GEOMETRIC_FUNCTION_IS_GENGEOMARRAY_BINOP_OVERLOAD(NAME) \
template <typename T> \
std::enable_if_t<detail::is_gengeomarray<T>::value, \
detail::marray_element_t<T>> \
NAME(T p0, T p1) __NOEXC { \
return NAME(detail::to_vec(p0), detail::to_vec(p1)); \
}

// clang-format off
__SYCL_MARRAY_GEOMETRIC_FUNCTION_IS_GENGEOMARRAY_BINOP_OVERLOAD(dot)
__SYCL_MARRAY_GEOMETRIC_FUNCTION_IS_GENGEOMARRAY_BINOP_OVERLOAD(distance)
// clang-format on

#undef __SYCL_MARRAY_GEOMETRIC_FUNCTION_IS_GENGEOMARRAY_BINOP_OVERLOAD

template <typename T>
std::enable_if_t<detail::is_gengeomarray<T>::value, detail::marray_element_t<T>>
length(T p) __NOEXC {
return __sycl_std::__invoke_length<detail::marray_element_t<T>>(
detail::to_vec(p));
}

template <typename T>
std::enable_if_t<detail::is_gengeomarrayfloat<T>::value,
detail::marray_element_t<T>>
fast_distance(T p0, T p1) __NOEXC {
return fast_distance(detail::to_vec(p0), detail::to_vec(p1));
}

template <typename T>
std::enable_if_t<detail::is_gengeomarrayfloat<T>::value,
detail::marray_element_t<T>>
fast_length(T p) __NOEXC {
return fast_length(detail::to_vec(p));
}

/* SYCL 1.2.1 ---- 4.13.7 Relational functions. -----------------------------*/
/* SYCL 2020 ---- 4.17.9 Relational functions. -----------------------------*/

Expand Down
12 changes: 12 additions & 0 deletions sycl/include/sycl/detail/generic_type_lists.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,12 @@ using vector_geo_float_list =
using vector_geo_double_list =
type_list<vec<double, 1>, vec<double, 2>, vec<double, 3>, vec<double, 4>>;

using marray_geo_float_list =
type_list<marray<float, 2>, marray<float, 3>, marray<float, 4>>;

using marray_geo_double_list =
type_list<marray<double, 2>, marray<double, 3>, marray<double, 4>>;

using geo_half_list = type_list<scalar_geo_half_list, vector_geo_half_list>;

using geo_float_list = type_list<scalar_geo_float_list, vector_geo_float_list>;
Expand All @@ -121,6 +127,9 @@ using scalar_geo_list = type_list<scalar_geo_half_list, scalar_geo_float_list,
using vector_geo_list = type_list<vector_geo_half_list, vector_geo_float_list,
vector_geo_double_list>;

using marray_geo_list =
type_list<marray_geo_float_list, marray_geo_double_list>;

using geo_list = type_list<scalar_geo_list, vector_geo_list>;

// cross floating point types
Expand All @@ -133,6 +142,9 @@ using cross_double_list = type_list<vec<double, 3>, vec<double, 4>>;
using cross_floating_list =
type_list<cross_float_list, cross_double_list, cross_half_list>;

using cross_marray_list = type_list<marray<float, 3>, marray<float, 4>,
marray<double, 3>, marray<double, 4>>;

using scalar_default_char_list = type_list<char>;

using vector_default_char_list =
Expand Down
9 changes: 9 additions & 0 deletions sycl/include/sycl/detail/generic_type_traits.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,6 +70,12 @@ using is_gengeofloat = is_contained<T, gtl::geo_float_list>;
template <typename T>
using is_gengeodouble = is_contained<T, gtl::geo_double_list>;

template <typename T>
using is_gengeomarrayfloat = is_contained<T, gtl::marray_geo_float_list>;

template <typename T>
using is_gengeomarray = is_contained<T, gtl::marray_geo_list>;

template <typename T> using is_gengeohalf = is_contained<T, gtl::geo_half_list>;

template <typename T>
Expand Down Expand Up @@ -97,6 +103,9 @@ using is_gencrosshalf = is_contained<T, gtl::cross_half_list>;
template <typename T>
using is_gencross = is_contained<T, gtl::cross_floating_list>;

template <typename T>
using is_gencrossmarray = is_contained<T, gtl::cross_marray_list>;

template <typename T>
using is_charn = is_contained<T, gtl::vector_default_char_list>;

Expand Down
137 changes: 137 additions & 0 deletions sycl/test-e2e/Basic/built-ins/marray_geometric.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,137 @@
// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
// RUN: %CPU_RUN_PLACEHOLDER %t.out
// RUN: %GPU_RUN_PLACEHOLDER %t.out
// RUN: %ACC_RUN_PLACEHOLDER %t.out

#include <CL/sycl.hpp>

#define TEST(FUNC, MARRAY_ELEM_TYPE, DIM, EXPECTED, DELTA, ...) \
{ \
{ \
MARRAY_ELEM_TYPE result[DIM]; \
{ \
sycl::buffer<MARRAY_ELEM_TYPE> b(result, sycl::range{DIM}); \
Queue.submit([&](sycl::handler &cgh) { \
sycl::accessor res_access{b, cgh}; \
cgh.single_task([=]() { \
sycl::marray<MARRAY_ELEM_TYPE, DIM> res = FUNC(__VA_ARGS__); \
for (int i = 0; i < DIM; i++) \
res_access[i] = res[i]; \
}); \
}); \
} \
for (int i = 0; i < DIM; i++) { \
assert(abs(result[i] - EXPECTED[i]) <= DELTA); \
} \
} \
}

#define TEST2(FUNC, TYPE, EXPECTED, DELTA, ...) \
{ \
{ \
TYPE result; \
{ \
sycl::buffer<TYPE> b(&result, 1); \
Queue.submit([&](sycl::handler &cgh) { \
sycl::accessor res_access{b, cgh}; \
cgh.single_task([=]() { res_access[0] = FUNC(__VA_ARGS__); }); \
}); \
} \
assert(abs(result - EXPECTED) <= DELTA); \
} \
}

#define EXPECTED(TYPE, ...) ((TYPE[]){__VA_ARGS__})

int main() {
sycl::device Dev;
sycl::queue Queue(Dev);
// clang-format off
sycl::marray<float, 2> MFloatD2 = {1.f, 2.f};
sycl::marray<float, 2> MFloatD2_2 = {3.f, 5.f};
sycl::marray<float, 3> MFloatD3 = {1.f, 2.f, 3.f};
sycl::marray<float, 3> MFloatD3_2 = {1.f, 5.f, 7.f};
sycl::marray<float, 4> MFloatD4 = {1.f, 2.f, 3.f, 4.f};
sycl::marray<float, 4> MFloatD4_2 = {1.f, 5.f, 7.f, 4.f};

sycl::marray<double, 2> MDoubleD2 = {1.0, 2.0};
sycl::marray<double, 2> MDoubleD2_2 = {3.0, 5.0};
sycl::marray<double, 3> MDoubleD3 = {1.0, 2.0, 3.0};
sycl::marray<double, 3> MDoubleD3_2 = {1.0, 5.0, 7.0};
sycl::marray<double, 4> MDoubleD4 = {1.0, 2.0, 3.0, 4.0};
sycl::marray<double, 4> MDoubleD4_2 = {1.0, 5.0, 7.0, 4.0};
// clang-format on

TEST(sycl::cross, float, 3, EXPECTED(float, -1.f, -4.f, 3.f), 0, MFloatD3,
MFloatD3_2);
TEST(sycl::cross, float, 4, EXPECTED(float, -1.f, -4.f, 3.f, 0.f), 0,
MFloatD4, MFloatD4_2);
if (Dev.has(sycl::aspect::fp64)) {
TEST(sycl::cross, double, 3, EXPECTED(double, -1.f, -4.f, 3.f), 0,
MDoubleD3, MDoubleD3_2);
TEST(sycl::cross, double, 4, EXPECTED(double, -1.f, -4.f, 3.f, 0.f), 0,
MDoubleD4, MDoubleD4_2);
}

TEST2(sycl::dot, float, 13.f, 0, MFloatD2, MFloatD2_2);
TEST2(sycl::dot, float, 32.f, 0, MFloatD3, MFloatD3_2);
TEST2(sycl::dot, float, 48.f, 0, MFloatD4, MFloatD4_2);
if (Dev.has(sycl::aspect::fp64)) {
TEST2(sycl::dot, double, 13, 0, MDoubleD2, MDoubleD2_2);
TEST2(sycl::dot, double, 32, 0, MDoubleD3, MDoubleD3_2);
TEST2(sycl::dot, double, 48, 0, MDoubleD4, MDoubleD4_2);
}

TEST2(sycl::length, float, 2.236068f, 1e-6, MFloatD2);
TEST2(sycl::length, float, 3.741657f, 1e-6, MFloatD3);
TEST2(sycl::length, float, 5.477225f, 1e-6, MFloatD4);
if (Dev.has(sycl::aspect::fp64)) {
TEST2(sycl::length, double, 2.236068, 1e-6, MDoubleD2);
TEST2(sycl::length, double, 3.741657, 1e-6, MDoubleD3);
TEST2(sycl::length, double, 5.477225, 1e-6, MDoubleD4);
}

TEST2(sycl::distance, float, 3.605551f, 1e-6, MFloatD2, MFloatD2_2);
TEST2(sycl::distance, float, 5.f, 0, MFloatD3, MFloatD3_2);
TEST2(sycl::distance, float, 5.f, 0, MFloatD4, MFloatD4_2);
if (Dev.has(sycl::aspect::fp64)) {
TEST2(sycl::distance, double, 3.605551, 1e-6, MDoubleD2, MDoubleD2_2);
TEST2(sycl::distance, double, 5.0, 0, MDoubleD3, MDoubleD3_2);
TEST2(sycl::distance, double, 5.0, 0, MDoubleD4, MDoubleD4_2);
}

TEST(sycl::normalize, float, 2, EXPECTED(float, 0.447213f, 0.894427f), 1e-6,
MFloatD2);
TEST(sycl::normalize, float, 3,
EXPECTED(float, 0.267261f, 0.534522f, 0.801784f), 1e-6, MFloatD3);
TEST(sycl::normalize, float, 4,
EXPECTED(float, 0.182574f, 0.365148f, 0.547723f, 0.730297f), 1e-6,
MFloatD4);
if (Dev.has(sycl::aspect::fp64)) {
TEST(sycl::normalize, double, 2, EXPECTED(double, 0.447213, 0.894427), 1e-6,
MDoubleD2);
TEST(sycl::normalize, double, 3,
EXPECTED(double, 0.267261, 0.534522, 0.801784), 1e-6, MDoubleD3);
TEST(sycl::normalize, double, 4,
EXPECTED(double, 0.182574, 0.365148, 0.547723, 0.730297), 1e-6,
MDoubleD4);
}

TEST2(sycl::fast_distance, float, 3.605551f, 1e-6, MFloatD2, MFloatD2_2);
TEST2(sycl::fast_distance, float, 5.f, 0, MFloatD3, MFloatD3_2);
TEST2(sycl::fast_distance, float, 5.f, 0, MFloatD4, MFloatD4_2);

TEST2(sycl::fast_length, float, 2.236068f, 1e-6, MFloatD2);
TEST2(sycl::fast_length, float, 3.741657f, 1e-6, MFloatD3);
TEST2(sycl::fast_length, float, 5.477225f, 1e-6, MFloatD4);

TEST(sycl::fast_normalize, float, 2, EXPECTED(float, 0.447213f, 0.894427f),
1e-3, MFloatD2);
TEST(sycl::fast_normalize, float, 3,
EXPECTED(float, 0.267261f, 0.534522f, 0.801784f), 1e-3, MFloatD3);
TEST(sycl::fast_normalize, float, 4,
EXPECTED(float, 0.182574f, 0.365148f, 0.547723f, 0.730297f), 1e-3,
MFloatD4);

return 0;
}