5 Commits

Author SHA1 Message Date
Wiebe van Breukelen
38664f6acb Fix compiler warnings 2025-10-22 15:20:50 +02:00
mmancini-skao
76998a137a Fix error in Taylor expansion (#36)
Replaced term3 with term4 in calculations for t4.
2025-10-20 17:09:35 +02:00
Bram Veenboer
500d35070e Fix formatting (#35)
* Run pre-commit

* Skip line-length check in cmake-lint
2025-10-10 09:19:18 +02:00
Dantali0n
bfe752433f Fixes #30, Add CMake steps to install python bindings (#31) 2025-09-17 20:03:28 +02:00
Bram Veenboer
8fe8314905 Update GPU backend (#29)
* Update GPU memory management
* Add allocate_memory and free_memory
2025-09-03 09:16:28 +02:00
6 changed files with 44 additions and 13 deletions

View File

@@ -7,4 +7,5 @@ repos:
rev: v0.6.13
hooks:
- id: cmake-format
- id: cmake-lint
- id: cmake-lint
args: [--disabled-codes=C0301]

View File

@@ -12,6 +12,11 @@ option(TRIGDX_BUILD_TESTS "Build tests" ON)
option(TRIGDX_BUILD_BENCHMARKS "Build tests" ON)
option(TRIGDX_BUILD_PYTHON "Build Python interface" ON)
# Add compiler flags
set(CMAKE_CXX_FLAGS
"${CMAKE_CXX_FLAGS} -Wall -Wnon-virtual-dtor -Wduplicated-branches -Wvla -Wpointer-arith -Wextra -Wno-unused-parameter"
)
list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
configure_file(
${CMAKE_CURRENT_SOURCE_DIR}/cmake/trigdx_config.hpp.in

View File

@@ -8,5 +8,16 @@ if(NOT pybind11_FOUND)
FetchContent_MakeAvailable(pybind11)
endif()
# Needed to set ${Python_VERSION_MAJOR} and ${Python_VERSION_MINOR}
find_package(Python REQUIRED)
pybind11_add_module(pytrigdx bindings.cpp)
target_link_libraries(pytrigdx PRIVATE trigdx)
set_target_properties(pytrigdx PROPERTIES OUTPUT_NAME "trigdx")
set(PYTHON_SITE_PACKAGES
"${CMAKE_INSTALL_LIBDIR}/python${Python_VERSION_MAJOR}.${Python_VERSION_MINOR}/site-packages/trigdx"
)
install(TARGETS pytrigdx DESTINATION ${PYTHON_SITE_PACKAGES})
install(FILES __init__.py DESTINATION ${PYTHON_SITE_PACKAGES})

16
python/__init__.py Normal file
View File

@@ -0,0 +1,16 @@
from .trigdx import Reference, Lookup16K, Lookup32K, LookupAVX16K, LookupAVX32K
try:
from .trigdx import MKL
except ImportError:
pass
try:
from .trigdx import GPU
except ImportError:
pass
try:
from .trigdx import LookupXSIMD16K, LookupXSIMD32K
except ImportError:
pass

View File

@@ -72,7 +72,9 @@ void bind_backend(py::module &m, const char *name) {
.def("compute_sincosf", &compute_sincos<float>);
}
PYBIND11_MODULE(pytrigdx, m) {
PYBIND11_MODULE(trigdx, m) {
m.doc() = "TrigDx python bindings";
py::class_<Backend, std::shared_ptr<Backend>>(m, "Backend")
.def("init", &Backend::init);
@@ -91,4 +93,4 @@ PYBIND11_MODULE(pytrigdx, m) {
bind_backend<LookupXSIMDBackend<16384>>(m, "LookupXSIMD16K");
bind_backend<LookupXSIMDBackend<32768>>(m, "LookupXSIMD32K");
#endif
}
}

View File

@@ -20,8 +20,8 @@ template <std::size_t NR_SAMPLES> struct lookup_table {
cos_values[i] = cosf(i * PI_FRAC);
}
}
std::array<float, NR_SAMPLES> cos_values;
std::array<float, NR_SAMPLES> sin_values;
std::array<float, NR_SAMPLES> cos_values;
};
template <std::size_t NR_SAMPLES> struct cosf_dispatcher {
@@ -33,7 +33,6 @@ template <std::size_t NR_SAMPLES> struct cosf_dispatcher {
constexpr uint_fast32_t VL = b_type::size;
const uint_fast32_t VS = n - n % VL;
const uint_fast32_t Q_PI = NR_SAMPLES / 4U;
const b_type scale = b_type::broadcast(lookup_table_.SCALE);
const b_type pi_frac = b_type::broadcast(lookup_table_.PI_FRAC);
const m_type mask = m_type::broadcast(lookup_table_.MASK);
@@ -42,7 +41,7 @@ template <std::size_t NR_SAMPLES> struct cosf_dispatcher {
const b_type term2 = b_type::broadcast(lookup_table_.TERM2); // 1/2!
const b_type term3 = b_type::broadcast(lookup_table_.TERM3); // 1/3!
const b_type term4 = b_type::broadcast(lookup_table_.TERM4); // 1/4!
const m_type quarter_pi = m_type::broadcast(Q_PI);
uint_fast32_t i;
for (i = 0; i < VS; i += VL) {
const b_type vx = b_type::load(a + i, Tag());
@@ -60,7 +59,7 @@ template <std::size_t NR_SAMPLES> struct cosf_dispatcher {
const b_type dx4 = xsimd::mul(dx2, dx);
const b_type t2 = xsimd::mul(dx2, term2);
const b_type t3 = xsimd::mul(dx3, term3);
const b_type t4 = xsimd::mul(dx4, term3);
const b_type t4 = xsimd::mul(dx4, term4);
const b_type cosdx = xsimd::add(xsimd::sub(term1, t2), t4);
@@ -98,7 +97,6 @@ template <std::size_t NR_SAMPLES> struct sinf_dispatcher {
constexpr uint_fast32_t VL = b_type::size;
const uint_fast32_t VS = n - n % VL;
const uint_fast32_t Q_PI = NR_SAMPLES / 4U;
const b_type scale = b_type::broadcast(lookup_table_.SCALE);
const b_type pi_frac = b_type::broadcast(lookup_table_.PI_FRAC);
const m_type mask = m_type::broadcast(lookup_table_.MASK);
@@ -107,7 +105,7 @@ template <std::size_t NR_SAMPLES> struct sinf_dispatcher {
const b_type term2 = b_type::broadcast(lookup_table_.TERM2); // 1/2!
const b_type term3 = b_type::broadcast(lookup_table_.TERM3); // 1/3!
const b_type term4 = b_type::broadcast(lookup_table_.TERM4); // 1/4!
const m_type quarter_pi = m_type::broadcast(Q_PI);
uint_fast32_t i;
for (i = 0; i < VS; i += VL) {
const b_type vx = b_type::load(a + i, Tag());
@@ -120,7 +118,7 @@ template <std::size_t NR_SAMPLES> struct sinf_dispatcher {
const b_type dx4 = xsimd::mul(dx2, dx);
const b_type t2 = xsimd::mul(dx2, term2);
const b_type t3 = xsimd::mul(dx3, term3);
const b_type t4 = xsimd::mul(dx4, term3);
const b_type t4 = xsimd::mul(dx4, term4);
const b_type cosdx = xsimd::add(xsimd::sub(term1, t2), t4);
const b_type sindx = xsimd::sub(dx, t3);
@@ -160,7 +158,6 @@ template <std::size_t NR_SAMPLES> struct sin_cosf_dispatcher {
constexpr uint_fast32_t VL = b_type::size;
const uint_fast32_t VS = n - n % VL;
const uint_fast32_t Q_PI = NR_SAMPLES / 4U;
const b_type scale = b_type::broadcast(lookup_table_.SCALE);
const m_type mask = m_type::broadcast(lookup_table_.MASK);
const b_type pi_frac = b_type::broadcast(lookup_table_.PI_FRAC);
@@ -170,7 +167,6 @@ template <std::size_t NR_SAMPLES> struct sin_cosf_dispatcher {
const b_type term3 = b_type::broadcast(lookup_table_.TERM3); // 1/3!
const b_type term4 = b_type::broadcast(lookup_table_.TERM4); // 1/4!
const m_type quarter_pi = m_type::broadcast(Q_PI);
uint_fast32_t i;
for (i = 0; i < VS; i += VL) {
const b_type vx = b_type::load(a + i, Tag());
@@ -183,7 +179,7 @@ template <std::size_t NR_SAMPLES> struct sin_cosf_dispatcher {
const b_type dx4 = xsimd::mul(dx2, dx);
const b_type t2 = xsimd::mul(dx2, term2);
const b_type t3 = xsimd::mul(dx3, term3);
const b_type t4 = xsimd::mul(dx4, term3);
const b_type t4 = xsimd::mul(dx4, term4);
idx = xsimd::bitwise_and(idx, mask);
b_type sinv = b_type::gather(lookup_table_.sin_values.data(), idx);