Merge pull request #143 from r-devulap/maint

Simplify use of ISA specific routines through a unified API for static methods

examples/Makefile

@@ -1,30 +1,24 @@
-CXX ?= g++-12
-CFLAGS = -I../src -std=c++17 -O3 $(if $(CXXFLAGS),$(CXXFLAGS),)
-EXE = qsort32avx2 argsort kvsort qsortfp16 qsort16 qsort32 qsort64
+CXX ?= g++-13
+CFLAGS = -I../src -std=c++17 -O3
+EXE = kvsort qsortavx2 qsortavx512 qsortspr qsorticl
 
 default: all
 all : $(EXE)
 
-qsortfp16: avx512fp-16bit-qsort.cpp
-	     $(CXX) -o qsortfp16 -march=sapphirerapids $(CFLAGS) avx512fp-16bit-qsort.cpp
-
-qsort16: avx512-16bit-qsort.cpp
-	     $(CXX) -o qsort16 -march=icelake-client $(CFLAGS) avx512-16bit-qsort.cpp
-
-qsort32: avx512-32bit-qsort.cpp
-	     $(CXX) -o qsort32 -march=skylake-avx512 $(CFLAGS) avx512-32bit-qsort.cpp
+kvsort: avx512-kv.cpp
+	     $(CXX) -o kvsort -mavx512vl -mavx512dq $(CFLAGS) avx512-kv.cpp
 
-qsort32avx2: avx2-32bit-qsort.cpp
-	     $(CXX) -o qsort32avx2 -march=haswell $(CFLAGS) avx2-32bit-qsort.cpp
+qsortavx512: skx-avx2.cpp
+	     $(CXX) -o qsortavx512 -mavx512vl -mavx512dq $(CFLAGS) skx-avx2.cpp
 
-qsort64: avx512-64bit-qsort.cpp
-	     $(CXX) -o qsort64 -march=skylake-avx512 $(CFLAGS) avx512-64bit-qsort.cpp
+qsortavx2: skx-avx2.cpp
+	     $(CXX) -o qsortavx2 -mavx2 $(CFLAGS) skx-avx2.cpp
 
-argsort: avx512-argsort.cpp
-	     $(CXX) -o argsort -march=skylake-avx512 $(CFLAGS) avx512-argsort.cpp
+qsorticl: icl-16bit.cpp
+	     $(CXX) -o qsorticl -mavx512vl -mavx512bw -mavx512dq -mavx512vbmi2 $(CFLAGS) icl-16bit.cpp
 
-kvsort: avx512-kv.cpp
-	     $(CXX) -o kvsort -march=skylake-avx512 $(CFLAGS) avx512-kv.cpp
+qsortspr: spr-16bit.cpp
+	     $(CXX) -o qsortspr -mavx512vl -mavx512dq -mavx512vbmi2 -mavx512fp16  $(CFLAGS) spr-16bit.cpp
 
 clean:
 	$(RM) $(EXE)

examples/avx512-kv.cpp

@@ -1,4 +1,4 @@
-#include "avx512-64bit-keyvaluesort.hpp"
+#include "x86simdsort-static-incl.h"
 
 int main()
 {
@@ -7,17 +7,17 @@ int main()
     uint64_t arr2[size];
     double arr3[size];
     float arr4[size];
-    avx512_qsort_kv(arr1, arr1, size);
-    avx512_qsort_kv(arr1, arr2, size);
-    avx512_qsort_kv(arr1, arr3, size);
-    avx512_qsort_kv(arr2, arr1, size);
-    avx512_qsort_kv(arr2, arr2, size);
-    avx512_qsort_kv(arr2, arr3, size);
-    avx512_qsort_kv(arr3, arr1, size);
-    avx512_qsort_kv(arr3, arr2, size);
-    avx512_qsort_kv(arr1, arr4, size);
-    avx512_qsort_kv(arr2, arr4, size);
-    avx512_qsort_kv(arr3, arr4, size);
+    x86simdsortStatic::keyvalue_qsort(arr1, arr1, size);
+    x86simdsortStatic::keyvalue_qsort(arr1, arr2, size);
+    x86simdsortStatic::keyvalue_qsort(arr1, arr3, size);
+    x86simdsortStatic::keyvalue_qsort(arr2, arr1, size);
+    x86simdsortStatic::keyvalue_qsort(arr2, arr2, size);
+    x86simdsortStatic::keyvalue_qsort(arr2, arr3, size);
+    x86simdsortStatic::keyvalue_qsort(arr3, arr1, size);
+    x86simdsortStatic::keyvalue_qsort(arr3, arr2, size);
+    x86simdsortStatic::keyvalue_qsort(arr1, arr4, size);
+    x86simdsortStatic::keyvalue_qsort(arr2, arr4, size);
+    x86simdsortStatic::keyvalue_qsort(arr3, arr4, size);
     return 0;
     return 0;
 }

examples/icl-16bit.cpp

@@ -0,0 +1,11 @@
+#include "x86simdsort-static-incl.h"
+
+int main()
+{
+    const int size = 1000;
+    short arr[size];
+    x86simdsortStatic::qsort(arr, size);
+    x86simdsortStatic::qselect(arr, 10, size);
+    x86simdsortStatic::partial_qsort(arr, 10, size);
+    return 0;
+}

examples/skx-avx2.cpp

@@ -0,0 +1,19 @@
+#include "x86simdsort-static-incl.h"
+
+int main()
+{
+    const int size = 1000;
+    double arrd[size];
+    float arrf[size];
+    x86simdsortStatic::qsort(arrf, size);
+    x86simdsortStatic::qsort(arrd, size);
+    x86simdsortStatic::qselect(arrf, 10, size);
+    x86simdsortStatic::qselect(arrd, 10, size);
+    x86simdsortStatic::partial_qsort(arrf, 10, size);
+    x86simdsortStatic::partial_qsort(arrd, 10, size);
+    auto arg1 = x86simdsortStatic::argsort(arrf, size);
+    auto arg2 = x86simdsortStatic::argselect(arrf, 10, size);
+    auto arg3 = x86simdsortStatic::argsort(arrd, size);
+    auto arg4 = x86simdsortStatic::argselect(arrd, 10, size);
+    return 0;
+}

examples/spr-16bit.cpp

@@ -0,0 +1,11 @@
+#include "x86simdsort-static-incl.h"
+
+int main()
+{
+    const int size = 1000;
+    _Float16 arr[size];
+    x86simdsortStatic::qsort(arr, size);
+    x86simdsortStatic::qselect(arr, 10, size);
+    x86simdsortStatic::partial_qsort(arr, 10, size);
+    return 0;
+}

lib/x86simdsort-avx2.cpp

@@ -1,39 +1,36 @@
 // AVX2 specific routines:
-#include "avx2-32bit-qsort.hpp"
-#include "avx2-64bit-qsort.hpp"
-#include "avx2-32bit-half.hpp"
-#include "xss-common-argsort.h"
+#include "x86simdsort-static-incl.h"
 #include "x86simdsort-internal.h"
 
 #define DEFINE_ALL_METHODS(type) \
     template <> \
     void qsort(type *arr, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx2_qsort(arr, arrsize, hasnan, descending); \
+        x86simdsortStatic::qsort(arr, arrsize, hasnan, descending); \
     } \
     template <> \
     void qselect( \
             type *arr, size_t k, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx2_qselect(arr, k, arrsize, hasnan, descending); \
+        x86simdsortStatic::qselect(arr, k, arrsize, hasnan, descending); \
     } \
     template <> \
     void partial_qsort( \
             type *arr, size_t k, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx2_partial_qsort(arr, k, arrsize, hasnan, descending); \
+        x86simdsortStatic::partial_qsort(arr, k, arrsize, hasnan, descending); \
     } \
     template <> \
     std::vector<size_t> argsort( \
             type *arr, size_t arrsize, bool hasnan, bool descending) \
     { \
-        return avx2_argsort(arr, arrsize, hasnan, descending); \
+        return x86simdsortStatic::argsort(arr, arrsize, hasnan, descending); \
     } \
     template <> \
     std::vector<size_t> argselect( \
             type *arr, size_t k, size_t arrsize, bool hasnan) \
     { \
-        return avx2_argselect(arr, k, arrsize, hasnan); \
+        return x86simdsortStatic::argselect(arr, k, arrsize, hasnan); \
     }
 
 namespace xss {

lib/x86simdsort-icl.cpp

@@ -1,13 +1,13 @@
 // ICL specific routines:
-#include "avx512-16bit-qsort.hpp"
+#include "x86simdsort-static-incl.h"
 #include "x86simdsort-internal.h"
 
 namespace xss {
 namespace avx512 {
     template <>
     void qsort(uint16_t *arr, size_t size, bool hasnan, bool descending)
     {
-        avx512_qsort(arr, size, hasnan, descending);
+        x86simdsortStatic::qsort(arr, size, hasnan, descending);
     }
     template <>
     void qselect(uint16_t *arr,
@@ -16,7 +16,7 @@ namespace avx512 {
                  bool hasnan,
                  bool descending)
     {
-        avx512_qselect(arr, k, arrsize, hasnan, descending);
+        x86simdsortStatic::qselect(arr, k, arrsize, hasnan, descending);
     }
     template <>
     void partial_qsort(uint16_t *arr,
@@ -25,12 +25,12 @@ namespace avx512 {
                        bool hasnan,
                        bool descending)
     {
-        avx512_partial_qsort(arr, k, arrsize, hasnan, descending);
+        x86simdsortStatic::partial_qsort(arr, k, arrsize, hasnan, descending);
     }
     template <>
     void qsort(int16_t *arr, size_t size, bool hasnan, bool descending)
     {
-        avx512_qsort(arr, size, hasnan, descending);
+        x86simdsortStatic::qsort(arr, size, hasnan, descending);
     }
     template <>
     void qselect(int16_t *arr,
@@ -39,7 +39,7 @@ namespace avx512 {
                  bool hasnan,
                  bool descending)
     {
-        avx512_qselect(arr, k, arrsize, hasnan, descending);
+        x86simdsortStatic::qselect(arr, k, arrsize, hasnan, descending);
     }
     template <>
     void partial_qsort(int16_t *arr,
@@ -48,7 +48,7 @@ namespace avx512 {
                        bool hasnan,
                        bool descending)
     {
-        avx512_partial_qsort(arr, k, arrsize, hasnan, descending);
+        x86simdsortStatic::partial_qsort(arr, k, arrsize, hasnan, descending);
     }
 } // namespace avx512
 } // namespace xss

lib/x86simdsort-skx.cpp

@@ -1,71 +1,68 @@
 // SKX specific routines:
-#include "avx512-32bit-qsort.hpp"
-#include "avx512-64bit-keyvaluesort.hpp"
-#include "avx512-64bit-argsort.hpp"
-#include "avx512-64bit-qsort.hpp"
+#include "x86simdsort-static-incl.h"
 #include "x86simdsort-internal.h"
 
 #define DEFINE_ALL_METHODS(type) \
     template <> \
     void qsort(type *arr, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx512_qsort(arr, arrsize, hasnan, descending); \
+        x86simdsortStatic::qsort(arr, arrsize, hasnan, descending); \
     } \
     template <> \
     void qselect( \
             type *arr, size_t k, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx512_qselect(arr, k, arrsize, hasnan, descending); \
+        x86simdsortStatic::qselect(arr, k, arrsize, hasnan, descending); \
     } \
     template <> \
     void partial_qsort( \
             type *arr, size_t k, size_t arrsize, bool hasnan, bool descending) \
     { \
-        avx512_partial_qsort(arr, k, arrsize, hasnan, descending); \
+        x86simdsortStatic::partial_qsort(arr, k, arrsize, hasnan, descending); \
     } \
     template <> \
     std::vector<size_t> argsort( \
             type *arr, size_t arrsize, bool hasnan, bool descending) \
     { \
-        return avx512_argsort(arr, arrsize, hasnan, descending); \
+        return x86simdsortStatic::argsort(arr, arrsize, hasnan, descending); \
     } \
     template <> \
     std::vector<size_t> argselect( \
             type *arr, size_t k, size_t arrsize, bool hasnan) \
     { \
-        return avx512_argselect(arr, k, arrsize, hasnan); \
+        return x86simdsortStatic::argselect(arr, k, arrsize, hasnan); \
     }
 
 #define DEFINE_KEYVALUE_METHODS(type) \
     template <> \
     void keyvalue_qsort(type *key, uint64_t *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     } \
     template <> \
     void keyvalue_qsort(type *key, int64_t *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     } \
     template <> \
     void keyvalue_qsort(type *key, double *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     } \
     template <> \
     void keyvalue_qsort(type *key, uint32_t *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     } \
     template <> \
     void keyvalue_qsort(type *key, int32_t *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     } \
     template <> \
     void keyvalue_qsort(type *key, float *val, size_t arrsize, bool hasnan) \
     { \
-        avx512_qsort_kv(key, val, arrsize, hasnan); \
+        x86simdsortStatic::keyvalue_qsort(key, val, arrsize, hasnan); \
     }
 
 namespace xss {