8000 ENH: Use Highway's VQSort on AArch64 by Mousius · Pull Request #24018 · numpy/numpy · GitHub
[go: up one dir, main page]
Skip to content

ENH: Use Highway's VQSort on AArch64 #24018

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 8 commits into from
Nov 24, 2023
Merged

ENH: Use Highway's VQSort on AArch64 #24018

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
a8082fe
ENH: Use Highway's VQSort on AArch64
Mousius Jun 20, 2023
edac5aa
Better gating of Highway
Mousius Jun 22, 2023
11ab024
Use VQSortStatic
Mousius Jun 26, 2023
003c3a4
Add argsort to meson
Mousius Sep 19, 2023
87f6d19
Fix up Highway gate in Meson
Mousius Sep 19, 2023
4d139b3
Re-align submodules with main
Mousius Oct 2, 2023
59443e8
Switch off simd_select for AArch64
Mousius Oct 2, 2023
278e1de
Minor text fixes
Mousius Oct 2, 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
Prev Previous commit
Next Next commit
Switch off simd_select for AArch64
  • Loading branch information
@Mousius
Mousius committed Nov 20, 2023
commit 59443e8a5ef68133bcb791ffca042d3a54eeca71
5 changes: 4 additions & 1 deletion numpy/_core/src/npysort/selection.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@
#include "simd_qsort.hpp"

#define NOT_USED NPY_UNUSED(unused)
#define DISABLE_HIGHWAY_OPTIMIZATION (defined(__arm__) || defined(__aarch64__))
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
#define DISABLE_HIGHWAY_OPTIMIZATION (defined(__arm__) || defined(__aarch64__))

Based on the presented suggestion, there would be no purpose in keeping this #definition.


template<typename T>
inline bool quickselect_dispatch(T* v, npy_intp num, npy_intp kth)
Expand Down Expand Up @@ -55,12 +56,14 @@ inline bool quickselect_dispatch(T* v, npy_intp num, npy_intp kth)
#endif
NPY_CPU_DISPATCH_CALL_XB(dispfunc = np::qsort_simd::template QSelect, <TF>);
}
#if !DISABLE_HIGHWAY_OPTIMIZATION
else if constexpr (sizeof(T) == sizeof(uint32_t) || sizeof(T) == sizeof(uint64_t)) {
#ifndef NPY_DISABLE_OPTIMIZATION
#include "simd_qsort.dispatch.h"
#endif
NPY_CPU_DISPATCH_CALL_XB(dispfunc = np::qsort_simd::template QSelect, <TF>);
}
#endif
if (dispfunc) {
(*dispfunc)(reinterpret_cast<TF*>(v), num, kth);
return true;
Expand All @@ -85,7 +88,7 @@ inline bool argquickselect_dispatch(T* v, npy_intp* arg, npy_intp num, npy_intp
sizeof(npy_intp) == sizeof(int64_t)) {
using TF = typename np::meta::FixedWidth<T>::Type;
#ifndef NPY_DISABLE_OPTIMIZATION
#include "simd_qsort.dispatch.h"
#include "simd_argsort.dispatch.h"
#endif
void (*dispfunc)(TF*, npy_intp*, npy_intp, npy_intp) = nullptr;
NPY_CPU_DISPATCH_CALL_XB(dispfunc = np::qsort_simd::template ArgQSelect, <TF>);
Expand Down
31 changes: 29 additions & 2 deletions numpy/_core/src/npysort/simd_argsort.dispatch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,14 +7,39 @@
// 'baseline' option isn't specified within targets.

#include "simd_qsort.hpp"
#ifndef __CYGWIN__

#if defined(NPY_HAVE_AVX512_SKX) && !defined(_MSC_VER)
#if defined(NPY_HAVE_AVX512_SKX)
#include "x86-simd-sort/src/avx512-64bit-argsort.hpp"
#endif

namespace np { namespace qsort_simd {

#if defined(NPY_HAVE_AVX512_SKX) && !defined(_MSC_VER)
#if defined(NPY_HAVE_AVX512_SKX)
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(float *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(double *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(int32_t *arr, npy_intp *arg, npy_intp size)
{
avx512_argsort(arr, reinterpret_cast<int64_t*>(arg), size);
Expand Down Expand Up @@ -42,3 +67,5 @@ template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(double *arr, npy_intp *arg, npy
#endif

}} // namespace np::simd

#endif // __CYGWIN__
24 changes: 0 additions & 24 deletions numpy/_core/src/npysort/simd_qsort.dispatch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,30 +23,6 @@
namespace np { namespace qsort_simd {

#if defined(NPY_HAVE_AVX512_SKX)
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(float *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(double *arr, npy_intp* arg, npy_intp num, npy_intp kth)
{
avx512_argselect(arr, reinterpret_cast<int64_t*>(arg), kth, num);
}
template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int32_t *arr, npy_intp num, npy_intp kth)
{
avx512_qselect(arr, kth, num, true);
Expand Down
0