Merge pull request #105 from r-devulap/key-value

Add key-value sort to runtime dispatch

benchmarks/bench-all.cpp

@@ -44,3 +44,4 @@
 #include "bench-partial-qsort.hpp"
 #include "bench-qselect.hpp"
 #include "bench-qsort.hpp"
+#include "bench-keyvalue.hpp"

benchmarks/bench-keyvalue.hpp

@@ -0,0 +1,48 @@
+#include "x86simdsort-scalar.h"
+
+template <typename T, class... Args>
+static void scalarkvsort(benchmark::State &state, Args &&...args)
+{
+    // Get args
+    auto args_tuple = std::make_tuple(std::move(args)...);
+    size_t arrsize = std::get<0>(args_tuple);
+    std::string arrtype = std::get<1>(args_tuple);
+    // set up array
+    std::vector<T> key = get_array<T>(arrtype, arrsize);
+    std::vector<T> val = get_array<T>("random", arrsize);
+    std::vector<T> key_bkp = key;
+    // benchmark
+    for (auto _ : state) {
+        xss::scalar::keyvalue_qsort(key.data(), val.data(), arrsize, false);
+        state.PauseTiming();
+        key = key_bkp;
+        state.ResumeTiming();
+    }
+}
+
+template <typename T, class... Args>
+static void simdkvsort(benchmark::State &state, Args &&...args)
+{
+    auto args_tuple = std::make_tuple(std::move(args)...);
+    size_t arrsize = std::get<0>(args_tuple);
+    std::string arrtype = std::get<1>(args_tuple);
+    // set up array
+    std::vector<T> key = get_array<T>(arrtype, arrsize);
+    std::vector<T> val = get_array<T>("random", arrsize);
+    std::vector<T> key_bkp = key;
+    // benchmark
+    for (auto _ : state) {
+        x86simdsort::keyvalue_qsort(key.data(), val.data(), arrsize);
+        state.PauseTiming();
+        key = key_bkp;
+        state.ResumeTiming();
+    }
+}
+
+#define BENCH_BOTH_KVSORT(type) \
+    BENCH_SORT(simdkvsort, type) \
+    BENCH_SORT(scalarkvsort, type)
+
+BENCH_BOTH_KVSORT(uint64_t)
+BENCH_BOTH_KVSORT(int64_t)
+BENCH_BOTH_KVSORT(double)

lib/x86simdsort-internal.h

@@ -9,6 +9,9 @@ namespace avx512 {
     // quicksort
     template <typename T>
     XSS_HIDE_SYMBOL void qsort(T *arr, size_t arrsize, bool hasnan = false);
+    // key-value quicksort
+    template <typename T1, typename T2>
+    XSS_EXPORT_SYMBOL void keyvalue_qsort(T1 *key, T2* val, size_t arrsize, bool hasnan = false);
     // quickselect
     template <typename T>
     XSS_HIDE_SYMBOL void
@@ -30,6 +33,9 @@ namespace avx2 {
     // quicksort
     template <typename T>
     XSS_HIDE_SYMBOL void qsort(T *arr, size_t arrsize, bool hasnan = false);
+    // key-value quicksort
+    template <typename T1, typename T2>
+    XSS_EXPORT_SYMBOL void keyvalue_qsort(T1 *key, T2* val, size_t arrsize, bool hasnan = false);
     // quickselect
     template <typename T>
     XSS_HIDE_SYMBOL void
@@ -51,6 +57,9 @@ namespace scalar {
     // quicksort
     template <typename T>
     XSS_HIDE_SYMBOL void qsort(T *arr, size_t arrsize, bool hasnan = false);
+    // key-value quicksort
+    template <typename T1, typename T2>
+    XSS_EXPORT_SYMBOL void keyvalue_qsort(T1 *key, T2* val, size_t arrsize, bool hasnan = false);
     // quickselect
     template <typename T>
     XSS_HIDE_SYMBOL void

lib/x86simdsort-scalar.h

@@ -3,6 +3,27 @@
 #include <numeric>
 
 namespace xss {
+namespace utils {
+/* O(1) permute array in place: stolen from
+ * http://www.davidespataro.it/apply-a-permutation-to-a-vector */
+template<typename T>
+void apply_permutation_in_place(T* arr, std::vector<size_t> arg)
+{
+    for(size_t i = 0 ; i < arg.size() ; i++) {
+        size_t curr = i;
+        size_t next = arg[curr];
+        while(next != i)
+        {
+            std::swap(arr[curr], arr[next]);
+            arg[curr] = curr;
+            curr = next;
+            next = arg[next];
+        }
+        arg[curr] = curr;
+    }
+}
+} // utils
+
 namespace scalar {
     template <typename T>
     void qsort(T *arr, size_t arrsize, bool hasnan)
@@ -57,6 +78,13 @@ namespace scalar {
                          compare_arg<T, std::less<T>>(arr));
         return arg;
     }
+    template <typename T1, typename T2>
+    void keyvalue_qsort(T1 *key, T2* val, size_t arrsize, bool hasnan)
+    {
+        std::vector<size_t> arg = argsort(key, arrsize, hasnan);
+        utils::apply_permutation_in_place(key, arg);
+        utils::apply_permutation_in_place(val, arg);
+    }
 
 } // namespace scalar
 } // namespace xss

lib/x86simdsort-skx.cpp

@@ -1,5 +1,6 @@
 // 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-internal.h"
@@ -32,6 +33,14 @@
         return avx512_argselect(arr, k, arrsize, hasnan); \
     }
 
+#define DEFINE_KEYVALUE_METHODS(type1, type2) \
+    template <> \
+    void keyvalue_qsort(type1 *key, type2* val, size_t arrsize, bool hasnan) \
+    { \
+        avx512_qsort_kv(key, val, arrsize, hasnan); \
+    } \
+
+
 namespace xss {
 namespace avx512 {
     DEFINE_ALL_METHODS(uint32_t)
@@ -40,5 +49,14 @@ namespace avx512 {
     DEFINE_ALL_METHODS(uint64_t)
     DEFINE_ALL_METHODS(int64_t)
     DEFINE_ALL_METHODS(double)
+    DEFINE_KEYVALUE_METHODS(double, uint64_t)
+    DEFINE_KEYVALUE_METHODS(double, int64_t)
+    DEFINE_KEYVALUE_METHODS(double, double)
+    DEFINE_KEYVALUE_METHODS(uint64_t, uint64_t)
+    DEFINE_KEYVALUE_METHODS(uint64_t, int64_t)
+    DEFINE_KEYVALUE_METHODS(uint64_t, double)
+    DEFINE_KEYVALUE_METHODS(int64_t, uint64_t)
+    DEFINE_KEYVALUE_METHODS(int64_t, int64_t)
+    DEFINE_KEYVALUE_METHODS(int64_t, double)
 } // namespace avx512
 } // namespace xss

lib/x86simdsort.cpp

@@ -51,6 +51,8 @@ dispatch_requested(std::string_view cpurequested,
     return false;
 }
 
+namespace x86simdsort {
+
 #define CAT_(a, b) a##b
 #define CAT(a, b) CAT_(a, b)
 
@@ -120,6 +122,33 @@ dispatch_requested(std::string_view cpurequested,
                 return; \
             } \
         } \
+    } \
+
+#define DISPATCH_KEYVALUE_SORT(TYPE1, TYPE2, ISA) \
+    static void (CAT(CAT(*internal_kv_qsort_, TYPE1), TYPE2))(TYPE1*, TYPE2*, size_t, bool) = NULL; \
+    template <> \
+    void keyvalue_qsort(TYPE1 *key, TYPE2* val, size_t arrsize, bool hasnan) \
+    { \
+        (CAT(CAT(*internal_kv_qsort_, TYPE1), TYPE2))(key, val, arrsize, hasnan); \
+    } \
+    static __attribute__((constructor)) void \
+    CAT(CAT(resolve_keyvalue_qsort_, TYPE1), TYPE2)(void) \
+    { \
+        CAT(CAT(internal_kv_qsort_, TYPE1), TYPE2) = &xss::scalar::keyvalue_qsort<TYPE1, TYPE2>; \
+        __builtin_cpu_init(); \
+        std::string_view preferred_cpu = find_preferred_cpu(ISA); \
+        if constexpr (dispatch_requested("avx512", ISA)) { \
+            if (preferred_cpu.find("avx512") != std::string_view::npos) { \
+                CAT(CAT(internal_kv_qsort_, TYPE1), TYPE2) = &xss::avx512::keyvalue_qsort<TYPE1, TYPE2>; \
+                return; \
+            } \
+        } \
+        if constexpr (dispatch_requested("avx2", ISA)) { \
+            if (preferred_cpu.find("avx2") != std::string_view::npos) { \
+                CAT(CAT(internal_kv_qsort_, TYPE1), TYPE2) = &xss::avx2::keyvalue_qsort<TYPE1, TYPE2>; \
+                return; \
+            } \
+        } \
     }
 
 #define ISA_LIST(...) \
@@ -128,7 +157,6 @@ dispatch_requested(std::string_view cpurequested,
         __VA_ARGS__ \
     }
 
-namespace x86simdsort {
 #ifdef __FLT16_MAX__
 DISPATCH(qsort, _Float16, ISA_LIST("avx512_spr"))
 DISPATCH(qselect, _Float16, ISA_LIST("avx512_spr"))
@@ -168,4 +196,14 @@ DISPATCH_ALL(argselect,
              (ISA_LIST("avx512_skx")),
              (ISA_LIST("avx512_skx")))
 
+DISPATCH_KEYVALUE_SORT(uint64_t, int64_t, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(uint64_t, uint64_t, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(uint64_t, double, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(int64_t, int64_t, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(int64_t, uint64_t, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(int64_t, double, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(double, int64_t, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(double, double, (ISA_LIST("avx512_skx")))
+DISPATCH_KEYVALUE_SORT(double, uint64_t, (ISA_LIST("avx512_skx")))
+
 } // namespace x86simdsort

lib/x86simdsort.h

@@ -34,5 +34,10 @@ template <typename T>
 XSS_EXPORT_SYMBOL std::vector<size_t>
 argselect(T *arr, size_t k, size_t arrsize, bool hasnan = false);
 
+// argselect
+template <typename T1, typename T2>
+XSS_EXPORT_SYMBOL void
+keyvalue_qsort(T1 *key, T2* val, size_t arrsize, bool hasnan = false);
+
 } // namespace x86simdsort
 #endif

run-bench.py

@@ -31,6 +31,9 @@
     elif "argsort" in args.benchcompare:
         baseline = "scalarargsort.*" + filterb
         contender = "simdargsort.*" + filterb
+    elif "keyvalue" in args.benchcompare:
+        baseline = "scalarkvsort.*" + filterb
+        contender = "simdkvsort.*" + filterb
     else:
         parser.print_help(sys.stderr)
         parser.error("ERROR: Unknown argument '%s'" % args.benchcompare)

src/avx512-64bit-keyvaluesort.hpp

@@ -542,8 +542,8 @@ heapify(type1_t *keys, type2_t *indexes, arrsize_t idx, arrsize_t size)
     arrsize_t i = idx;
     while (true) {
         arrsize_t j = 2 * i + 1;
-        if (j >= size || j < 0) { break; }
-        int k = j + 1;
+        if (j >= size) { break; }
+        arrsize_t k = j + 1;
         if (k < size && keys[j] < keys[k]) { j = k; }
         if (keys[j] < keys[i]) { break; }
         std::swap(keys[i], keys[j]);
@@ -558,8 +558,9 @@ template <typename vtype1,
 X86_SIMD_SORT_INLINE void
 heap_sort(type1_t *keys, type2_t *indexes, arrsize_t size)
 {
-    for (arrsize_t i = size / 2 - 1; i >= 0; i--) {
+    for (arrsize_t i = size / 2 - 1; ; i--) {
         heapify<vtype1, vtype2>(keys, indexes, i, size);
+        if (i == 0) { break; }
     }
     for (arrsize_t i = size - 1; i > 0; i--) {
         std::swap(keys[0], keys[i]);
@@ -614,8 +615,9 @@ X86_SIMD_SORT_INLINE void qsort_64bit_(type1_t *keys,
 
 template <typename T1, typename T2>
 X86_SIMD_SORT_INLINE void
-avx512_qsort_kv(T1 *keys, T2 *indexes, arrsize_t arrsize)
+avx512_qsort_kv(T1 *keys, T2 *indexes, arrsize_t arrsize, bool hasnan = false)
 {
+    UNUSED(hasnan);
     if (arrsize > 1) {
         if constexpr (std::is_floating_point_v<T1>) {
             arrsize_t nan_count

tests/meson.build

@@ -6,6 +6,12 @@ libtests += static_library('tests_qsort',
   include_directories : [lib, utils],
   )
 
+libtests += static_library('tests_kvsort',
+  files('test-keyvalue.cpp', ),
+  dependencies: gtest_dep,
+  include_directories : [lib, utils],
+  )
+
 #if cancompilefp16
 #  libtests += static_library('tests_qsortfp16',
 #    files('test-qsortfp16.cpp', ),

tests/test-keyvalue.cpp

@@ -0,0 +1,67 @@
+/*******************************************
+ * * Copyright (C) 2022-2023 Intel Corporation
+ * * SPDX-License-Identifier: BSD-3-Clause
+ * *******************************************/
+
+#include "rand_array.h"
+#include "x86simdsort.h"
+#include "x86simdsort-scalar.h"
+#include <gtest/gtest.h>
+
+template <typename T>
+class simdkvsort : public ::testing::Test {
+public:
+    simdkvsort()
+    {
+        std::iota(arrsize.begin(), arrsize.end(), 1);
+        arrtype = {"random",
+                   "constant",
+                   "sorted",
+                   "reverse",
+                   "smallrange",
+                   "max_at_the_end",
+                   "rand_max"};
+    }
+    std::vector<std::string> arrtype;
+    std::vector<size_t> arrsize = std::vector<size_t>(1024);
+};
+
+TYPED_TEST_SUITE_P(simdkvsort);
+
+TYPED_TEST_P(simdkvsort, test_kvsort)
+{
+    using T1 = typename std::tuple_element<0, decltype(TypeParam())>::type;
+    using T2 = typename std::tuple_element<1, decltype(TypeParam())>::type;
+    for (auto type : this->arrtype) {
+        bool hasnan = (type == "rand_with_nan") ? true : false;
+        for (auto size : this->arrsize) {
+            std::vector<T1> key = get_array<T1>(type, size);
+            std::vector<T2> val = get_array<T2>(type, size);
+            std::vector<T1> key_bckp = key;
+            std::vector<T2> val_bckp = val;
+            x86simdsort::keyvalue_qsort(key.data(), val.data(), size, hasnan);
+            xss::scalar::keyvalue_qsort(key_bckp.data(), val_bckp.data(), size, hasnan);
+            ASSERT_EQ(key, key_bckp);
+            const bool hasDuplicates = std::adjacent_find(key.begin(), key.end()) != key.end();
+            if (!hasDuplicates) {
+                ASSERT_EQ(val, val_bckp);
+            }
+            key.clear(); val.clear();
+            key_bckp.clear(); val_bckp.clear();
+        }
+    }
+}
+
+REGISTER_TYPED_TEST_SUITE_P(simdkvsort, test_kvsort);
+
+using QKVSortTestTypes = testing::Types<std::tuple<double, double>,
+                                      std::tuple<double, uint64_t>,
+                                      std::tuple<double, int64_t>,
+                                      std::tuple<uint64_t, double>,
+                                      std::tuple<uint64_t, uint64_t>,
+                                      std::tuple<uint64_t, int64_t>,
+                                      std::tuple<int64_t, double>,
+                                      std::tuple<int64_t, uint64_t>,
+                                      std::tuple<int64_t, int64_t>>;
+
+INSTANTIATE_TYPED_TEST_SUITE_P(xss, simdkvsort, QKVSortTestTypes);