Use the new static methods API from x86-simd-sort

Raghuveer Devulapalli · Raghuveer Devulapalli · commit c1cf2dac9831 · 2024-04-23T12:56:28.000-07:00
diff --git a/numpy/_core/src/npysort/x86_simd_argsort.dispatch.cpp b/numpy/_core/src/npysort/x86_simd_argsort.dispatch.cpp
@@ -1,86 +1,57 @@
 #include "x86_simd_qsort.hpp"
 #ifndef __CYGWIN__
 
-#if defined(NPY_HAVE_AVX512_SKX)
-#include "x86-simd-sort/src/avx512-64bit-argsort.hpp"
-#elif defined(NPY_HAVE_AVX2)
-#include "x86-simd-sort/src/avx2-32bit-half.hpp"
-#include "x86-simd-sort/src/avx2-32bit-qsort.hpp"
-#include "x86-simd-sort/src/avx2-64bit-qsort.hpp"
-#include "x86-simd-sort/src/xss-common-argsort.h"
-#endif
-
-namespace {
-template<typename T>
-void x86_argsort(T* arr, size_t* arg, npy_intp num)
-{
-#if defined(NPY_HAVE_AVX512_SKX)
-    avx512_argsort(arr, arg, num, true);
-#elif defined(NPY_HAVE_AVX2)
-    avx2_argsort(arr, arg, num, true);
-#endif
-}
-
-template<typename T>
-void x86_argselect(T* arr, size_t* arg, npy_intp kth, npy_intp num)
-{
-#if defined(NPY_HAVE_AVX512_SKX)
-    avx512_argselect(arr, arg, kth, num, true);
-#elif defined(NPY_HAVE_AVX2)
-    avx2_argselect(arr, arg, kth, num, true);
-#endif
-}
-} // anonymous
+#include "x86-simd-sort/src/x86simdsort-static-incl.h"
 
 namespace np { namespace qsort_simd {
 
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint32_t *arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint64_t*arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(float *arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(double *arr, npy_intp* arg, npy_intp num, npy_intp kth)
 {
-    x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);
+    x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(int32_t *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(uint32_t *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(int64_t *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(uint64_t *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(float *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(double *arr, npy_intp *arg, npy_intp size)
 {
-    x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);
+    x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size, true);
 }
 
 }} // namespace np::simd
diff --git a/numpy/_core/src/npysort/x86_simd_qsort.dispatch.cpp b/numpy/_core/src/npysort/x86_simd_qsort.dispatch.cpp
@@ -1,89 +1,57 @@
 #include "x86_simd_qsort.hpp"
 #ifndef __CYGWIN__
 
-#if defined(NPY_HAVE_AVX512_SKX)
-    #include "x86-simd-sort/src/avx512-32bit-qsort.hpp"
-    #include "x86-simd-sort/src/avx512-64bit-qsort.hpp"
-    #include "x86-simd-sort/src/avx512-64bit-argsort.hpp"
-#elif defined(NPY_HAVE_AVX2)
-    #include "x86-simd-sort/src/avx2-32bit-qsort.hpp"
-    #include "x86-simd-sort/src/avx2-64bit-qsort.hpp"
-#endif
-
-namespace {
-template<typename T>
-void x86_qsort(T* arr, npy_intp num)
-{
-#if defined(NPY_HAVE_AVX512_SKX)
-    avx512_qsort(arr, num, true);
-#elif defined(NPY_HAVE_AVX2)
-    avx2_qsort(arr, num, true);
-#endif
-}
-
-template<typename T>
-void x86_qselect(T* arr, npy_intp num, npy_intp kth)
-{
-#if defined(NPY_HAVE_AVX512_SKX)
-    avx512_qselect(arr, kth, num, true);
-#elif defined(NPY_HAVE_AVX2)
-    avx2_qselect(arr, kth, num, true);
-#endif
-}
-} // anonymous
+#include "x86-simd-sort/src/x86simdsort-static-incl.h"
 
 namespace np { namespace qsort_simd {
-#if defined(NPY_HAVE_AVX512_SKX) || defined(NPY_HAVE_AVX2)
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int32_t *arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint32_t *arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int64_t*arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint64_t*arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(float *arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(double *arr, npy_intp num, npy_intp kth)
 {
-    x86_qselect(arr, num, kth);
+    x86simdsortStatic::qselect(arr, kth, num, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int32_t *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint32_t *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int64_t *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint64_t *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(float *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num, true);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(double *arr, npy_intp num)
 {
-    x86_qsort(arr, num);
+    x86simdsortStatic::qsort(arr, num, true);
 }
-#endif // NPY_HAVE_AVX512_SKX || NPY_HAVE_AVX2
-
 }} // namespace np::qsort_simd
 
 #endif // __CYGWIN__
diff --git a/numpy/_core/src/npysort/x86_simd_qsort_16bit.dispatch.cpp b/numpy/_core/src/npysort/x86_simd_qsort_16bit.dispatch.cpp
@@ -1,36 +1,30 @@
 #include "x86_simd_qsort.hpp"
 #ifndef __CYGWIN__
 
-#if defined(NPY_HAVE_AVX512_SPR)
-    #include "x86-simd-sort/src/avx512fp16-16bit-qsort.hpp"
-    #include "x86-simd-sort/src/avx512-16bit-qsort.hpp"
-#elif defined(NPY_HAVE_AVX512_ICL)
-    #include "x86-simd-sort/src/avx512-16bit-qsort.hpp"
-#endif
+#include "x86-simd-sort/src/x86simdsort-static-incl.h"
 
 namespace np { namespace qsort_simd {
 
 /*
  * QSelect dispatch functions:
  */
-#if defined(NPY_HAVE_AVX512_ICL) || defined(NPY_HAVE_AVX512_SPR)
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(Half *arr, npy_intp num, npy_intp kth)
 {
 #if defined(NPY_HAVE_AVX512_SPR)
-    avx512_qselect(reinterpret_cast<_Float16*>(arr), kth, num, true);
+    x86simdsortStatic::qselect(reinterpret_cast<_Float16*>(arr), kth, num, true);
 #else
     avx512_qselect_fp16(reinterpret_cast<uint16_t*>(arr), kth, num, true);
 #endif
 }
 
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint16_t *arr, npy_intp num, npy_intp kth)
 {
-    avx512_qselect(arr, kth, num);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 
 template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int16_t *arr, npy_intp num, npy_intp kth)
 {
-    avx512_qselect(arr, kth, num);
+    x86simdsortStatic::qselect(arr, kth, num);
 }
 
 /*
@@ -39,20 +33,19 @@ template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int16_t *arr, npy_intp num, npy_
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(Half *arr, npy_intp size)
 {
 #if defined(NPY_HAVE_AVX512_SPR)
-    avx512_qsort(reinterpret_cast<_Float16*>(arr), size, true);
+    x86simdsortStatic::qsort(reinterpret_cast<_Float16*>(arr), size, true);
 #else
     avx512_qsort_fp16(reinterpret_cast<uint16_t*>(arr), size, true);
 #endif
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint16_t *arr, npy_intp size)
 {
-    avx512_qsort(arr, size);
+    x86simdsortStatic::qsort(arr, size);
 }
 template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int16_t *arr, npy_intp size)
 {
-    avx512_qsort(arr, size);
+    x86simdsortStatic::qsort(arr, size);
 }
-#endif // NPY_HAVE_AVX512_ICL || SPR
 
 }} // namespace np::qsort_simd
 

Original file line number	Diff line number	Diff line change
`@@ -1,86 +1,57 @@`
`1`	`1`	`#include "x86_simd_qsort.hpp"`
`2`	`2`	`#ifndef __CYGWIN__`
`3`	`3`
`4`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`5`		`-#include "x86-simd-sort/src/avx512-64bit-argsort.hpp"`
`6`		`-#elif defined(NPY_HAVE_AVX2)`
`7`		`-#include "x86-simd-sort/src/avx2-32bit-half.hpp"`
`8`		`-#include "x86-simd-sort/src/avx2-32bit-qsort.hpp"`
`9`		`-#include "x86-simd-sort/src/avx2-64bit-qsort.hpp"`
`10`		`-#include "x86-simd-sort/src/xss-common-argsort.h"`
`11`		`-#endif`
`12`		`-`
`13`		`-namespace {`
`14`		`-template<typename T>`
`15`		`-void x86_argsort(T* arr, size_t* arg, npy_intp num)`
`16`		`-{`
`17`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`18`		`- avx512_argsort(arr, arg, num, true);`
`19`		`-#elif defined(NPY_HAVE_AVX2)`
`20`		`- avx2_argsort(arr, arg, num, true);`
`21`		`-#endif`
`22`		`-}`
`23`		`-`
`24`		`-template<typename T>`
`25`		`-void x86_argselect(T* arr, size_t* arg, npy_intp kth, npy_intp num)`
`26`		`-{`
`27`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`28`		`- avx512_argselect(arr, arg, kth, num, true);`
`29`		`-#elif defined(NPY_HAVE_AVX2)`
`30`		`- avx2_argselect(arr, arg, kth, num, true);`
`31`		`-#endif`
`32`		`-}`
`33`		`-} // anonymous`
	`4`	`+#include "x86-simd-sort/src/x86simdsort-static-incl.h"`
`34`	`5`
`35`	`6`	`namespace np { namespace qsort_simd {`
`36`	`7`
`37`	`8`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int32_t arr, npy_intp arg, npy_intp num, npy_intp kth)`
`38`	`9`	`{`
`39`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`10`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
`40`	`11`	`}`
`41`	`12`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint32_t arr, npy_intp arg, npy_intp num, npy_intp kth)`
`42`	`13`	`{`
`43`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`14`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
`44`	`15`	`}`
`45`	`16`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(int64_tarr, npy_intp arg, npy_intp num, npy_intp kth)`
`46`	`17`	`{`
`47`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`18`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
`48`	`19`	`}`
`49`	`20`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(uint64_tarr, npy_intp arg, npy_intp num, npy_intp kth)`
`50`	`21`	`{`
`51`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`22`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
`52`	`23`	`}`
`53`	`24`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(float arr, npy_intp arg, npy_intp num, npy_intp kth)`
`54`	`25`	`{`
`55`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`26`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num, true);`
`56`	`27`	`}`
`57`	`28`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSelect)(double arr, npy_intp arg, npy_intp num, npy_intp kth)`
`58`	`29`	`{`
`59`		`- x86_argselect(arr, reinterpret_cast<size_t*>(arg), kth, num);`
	`30`	`+ x86simdsortStatic::argselect(arr, reinterpret_cast<size_t*>(arg), kth, num, true);`
`60`	`31`	`}`
`61`	`32`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(int32_t arr, npy_intp arg, npy_intp size)`
`62`	`33`	`{`
`63`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`34`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);`
`64`	`35`	`}`
`65`	`36`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(uint32_t arr, npy_intp arg, npy_intp size)`
`66`	`37`	`{`
`67`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`38`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);`
`68`	`39`	`}`
`69`	`40`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(int64_t arr, npy_intp arg, npy_intp size)`
`70`	`41`	`{`
`71`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`42`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);`
`72`	`43`	`}`
`73`	`44`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(uint64_t arr, npy_intp arg, npy_intp size)`
`74`	`45`	`{`
`75`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`46`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size);`
`76`	`47`	`}`
`77`	`48`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(float arr, npy_intp arg, npy_intp size)`
`78`	`49`	`{`
`79`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`50`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size, true);`
`80`	`51`	`}`
`81`	`52`	`template<> void NPY_CPU_DISPATCH_CURFX(ArgQSort)(double arr, npy_intp arg, npy_intp size)`
`82`	`53`	`{`
`83`		`- x86_argsort(arr, reinterpret_cast<size_t*>(arg), size);`
	`54`	`+ x86simdsortStatic::argsort(arr, reinterpret_cast<size_t*>(arg), size, true);`
`84`	`55`	`}`
`85`	`56`
`86`	`57`	`}} // namespace np::simd`
Original file line number	Diff line number	Diff line change
`@@ -1,89 +1,57 @@`
`1`	`1`	`#include "x86_simd_qsort.hpp"`
`2`	`2`	`#ifndef __CYGWIN__`
`3`	`3`
`4`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`5`		`- #include "x86-simd-sort/src/avx512-32bit-qsort.hpp"`
`6`		`- #include "x86-simd-sort/src/avx512-64bit-qsort.hpp"`
`7`		`- #include "x86-simd-sort/src/avx512-64bit-argsort.hpp"`
`8`		`-#elif defined(NPY_HAVE_AVX2)`
`9`		`- #include "x86-simd-sort/src/avx2-32bit-qsort.hpp"`
`10`		`- #include "x86-simd-sort/src/avx2-64bit-qsort.hpp"`
`11`		`-#endif`
`12`		`-`
`13`		`-namespace {`
`14`		`-template<typename T>`
`15`		`-void x86_qsort(T* arr, npy_intp num)`
`16`		`-{`
`17`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`18`		`- avx512_qsort(arr, num, true);`
`19`		`-#elif defined(NPY_HAVE_AVX2)`
`20`		`- avx2_qsort(arr, num, true);`
`21`		`-#endif`
`22`		`-}`
`23`		`-`
`24`		`-template<typename T>`
`25`		`-void x86_qselect(T* arr, npy_intp num, npy_intp kth)`
`26`		`-{`
`27`		`-#if defined(NPY_HAVE_AVX512_SKX)`
`28`		`- avx512_qselect(arr, kth, num, true);`
`29`		`-#elif defined(NPY_HAVE_AVX2)`
`30`		`- avx2_qselect(arr, kth, num, true);`
`31`		`-#endif`
`32`		`-}`
`33`		`-} // anonymous`
	`4`	`+#include "x86-simd-sort/src/x86simdsort-static-incl.h"`
`34`	`5`
`35`	`6`	`namespace np { namespace qsort_simd {`
`36`		`-#if defined(NPY_HAVE_AVX512_SKX) \|\| defined(NPY_HAVE_AVX2)`
`37`	`7`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int32_t *arr, npy_intp num, npy_intp kth)`
`38`	`8`	`{`
`39`		`- x86_qselect(arr, num, kth);`
	`9`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`40`	`10`	`}`
`41`	`11`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint32_t *arr, npy_intp num, npy_intp kth)`
`42`	`12`	`{`
`43`		`- x86_qselect(arr, num, kth);`
	`13`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`44`	`14`	`}`
`45`	`15`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int64_t*arr, npy_intp num, npy_intp kth)`
`46`	`16`	`{`
`47`		`- x86_qselect(arr, num, kth);`
	`17`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`48`	`18`	`}`
`49`	`19`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint64_t*arr, npy_intp num, npy_intp kth)`
`50`	`20`	`{`
`51`		`- x86_qselect(arr, num, kth);`
	`21`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`52`	`22`	`}`
`53`	`23`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(float *arr, npy_intp num, npy_intp kth)`
`54`	`24`	`{`
`55`		`- x86_qselect(arr, num, kth);`
	`25`	`+ x86simdsortStatic::qselect(arr, kth, num, true);`
`56`	`26`	`}`
`57`	`27`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(double *arr, npy_intp num, npy_intp kth)`
`58`	`28`	`{`
`59`		`- x86_qselect(arr, num, kth);`
	`29`	`+ x86simdsortStatic::qselect(arr, kth, num, true);`
`60`	`30`	`}`
`61`	`31`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int32_t *arr, npy_intp num)`
`62`	`32`	`{`
`63`		`- x86_qsort(arr, num);`
	`33`	`+ x86simdsortStatic::qsort(arr, num);`
`64`	`34`	`}`
`65`	`35`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint32_t *arr, npy_intp num)`
`66`	`36`	`{`
`67`		`- x86_qsort(arr, num);`
	`37`	`+ x86simdsortStatic::qsort(arr, num);`
`68`	`38`	`}`
`69`	`39`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int64_t *arr, npy_intp num)`
`70`	`40`	`{`
`71`		`- x86_qsort(arr, num);`
	`41`	`+ x86simdsortStatic::qsort(arr, num);`
`72`	`42`	`}`
`73`	`43`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint64_t *arr, npy_intp num)`
`74`	`44`	`{`
`75`		`- x86_qsort(arr, num);`
	`45`	`+ x86simdsortStatic::qsort(arr, num);`
`76`	`46`	`}`
`77`	`47`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(float *arr, npy_intp num)`
`78`	`48`	`{`
`79`		`- x86_qsort(arr, num);`
	`49`	`+ x86simdsortStatic::qsort(arr, num, true);`
`80`	`50`	`}`
`81`	`51`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(double *arr, npy_intp num)`
`82`	`52`	`{`
`83`		`- x86_qsort(arr, num);`
	`53`	`+ x86simdsortStatic::qsort(arr, num, true);`
`84`	`54`	`}`
`85`		`-#endif // NPY_HAVE_AVX512_SKX \|\| NPY_HAVE_AVX2`
`86`		`-`
`87`	`55`	`}} // namespace np::qsort_simd`
`88`	`56`
`89`	`57`	`#endif // __CYGWIN__`
Original file line number	Diff line number	Diff line change
`@@ -1,36 +1,30 @@`
`1`	`1`	`#include "x86_simd_qsort.hpp"`
`2`	`2`	`#ifndef __CYGWIN__`
`3`	`3`
`4`		`-#if defined(NPY_HAVE_AVX512_SPR)`
`5`		`- #include "x86-simd-sort/src/avx512fp16-16bit-qsort.hpp"`
`6`		`- #include "x86-simd-sort/src/avx512-16bit-qsort.hpp"`
`7`		`-#elif defined(NPY_HAVE_AVX512_ICL)`
`8`		`- #include "x86-simd-sort/src/avx512-16bit-qsort.hpp"`
`9`		`-#endif`
	`4`	`+#include "x86-simd-sort/src/x86simdsort-static-incl.h"`
`10`	`5`
`11`	`6`	`namespace np { namespace qsort_simd {`
`12`	`7`
`13`	`8`	`/*`
`14`	`9`	`* QSelect dispatch functions:`
`15`	`10`	`*/`
`16`		`-#if defined(NPY_HAVE_AVX512_ICL) \|\| defined(NPY_HAVE_AVX512_SPR)`
`17`	`11`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(Half *arr, npy_intp num, npy_intp kth)`
`18`	`12`	`{`
`19`	`13`	`#if defined(NPY_HAVE_AVX512_SPR)`
`20`		`- avx512_qselect(reinterpret_cast<_Float16*>(arr), kth, num, true);`
	`14`	`+ x86simdsortStatic::qselect(reinterpret_cast<_Float16*>(arr), kth, num, true);`
`21`	`15`	`#else`
`22`	`16`	`avx512_qselect_fp16(reinterpret_cast<uint16_t*>(arr), kth, num, true);`
`23`	`17`	`#endif`
`24`	`18`	`}`
`25`	`19`
`26`	`20`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(uint16_t *arr, npy_intp num, npy_intp kth)`
`27`	`21`	`{`
`28`		`- avx512_qselect(arr, kth, num);`
	`22`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`29`	`23`	`}`
`30`	`24`
`31`	`25`	`template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int16_t *arr, npy_intp num, npy_intp kth)`
`32`	`26`	`{`
`33`		`- avx512_qselect(arr, kth, num);`
	`27`	`+ x86simdsortStatic::qselect(arr, kth, num);`
`34`	`28`	`}`
`35`	`29`
`36`	`30`	`/*`
`@@ -39,20 +33,19 @@ template<> void NPY_CPU_DISPATCH_CURFX(QSelect)(int16_t *arr, npy_intp num, npy_`
`39`	`33`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(Half *arr, npy_intp size)`
`40`	`34`	`{`
`41`	`35`	`#if defined(NPY_HAVE_AVX512_SPR)`
`42`		`- avx512_qsort(reinterpret_cast<_Float16*>(arr), size, true);`
	`36`	`+ x86simdsortStatic::qsort(reinterpret_cast<_Float16*>(arr), size, true);`
`43`	`37`	`#else`
`44`	`38`	`avx512_qsort_fp16(reinterpret_cast<uint16_t*>(arr), size, true);`
`45`	`39`	`#endif`
`46`	`40`	`}`
`47`	`41`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(uint16_t *arr, npy_intp size)`
`48`	`42`	`{`
`49`		`- avx512_qsort(arr, size);`
	`43`	`+ x86simdsortStatic::qsort(arr, size);`
`50`	`44`	`}`
`51`	`45`	`template<> void NPY_CPU_DISPATCH_CURFX(QSort)(int16_t *arr, npy_intp size)`
`52`	`46`	`{`
`53`		`- avx512_qsort(arr, size);`
	`47`	`+ x86simdsortStatic::qsort(arr, size);`
`54`	`48`	`}`
`55`		`-#endif // NPY_HAVE_AVX512_ICL \|\| SPR`
`56`	`49`
`57`	`50`	`}} // namespace np::qsort_simd`
`58`	`51`