From cb46ed1121130e39d06f9794131bff5b51ee17e6 Mon Sep 17 00:00:00 2001
From: Berenger Bramas <Berenger.Bramas@inria.fr>
Date: Fri, 24 Jan 2020 11:09:14 +0100
Subject: [PATCH] Add counters
---
README.md | 14 ++
sort512.hpp | 353 ++++++++++++++++++++++++++++++++++++++++++++++++
sort512kv.hpp | 200 +++++++++++++++++++++++++++
sort512test.cpp | 83 ++++++++++++
4 files changed, 650 insertions(+)
diff --git a/README.md b/README.md
index 95dd32e..3631a9a 100644
--- a/README.md
+++ b/README.md
@@ -48,3 +48,17 @@ Anyone can test the code without having a KNL by using the Intel SDE
## Knowing more about the number of instructions
Please checkout the branch "feature/counters" to have access to counters.
+
+You are actually in, if you want to get performance then checkout the master branch instead.
+
+You must use also use `-std=c++17` (for inline static variables)
+
+You can access the following functions to print and reset the counters:
+```
+Sort512::PrintCounters();
+Sort512::ResetCounters();
+// and
+
+Sort512kv::PrintCounters();
+Sort512kv::ResetCounters();
+```
diff --git a/sort512.hpp b/sort512.hpp
index fd2f62e..e8e79e6 100644
--- a/sort512.hpp
+++ b/sort512.hpp
@@ -22,6 +22,7 @@
/// Intel : -xCOMMON-AVX512 -xCORE-AVX512 -qopenmp
///
/// Or use "-march=native -mtune=native" if you are already on the right platform ("native can be replaced by "knl" or "skylake")
+/// You are in the branch with counters! You must use also use -std=c++17 (for inline static variables)
//////////////////////////////////////////////////////////
#ifndef SORT512_HPP
#define SORT512_HPP
@@ -37,7 +38,40 @@
#include "parallelInplace.hpp"
#endif
+#include <iostream>
namespace Sort512 {
+ inline static long int globalCptMin = 0;
+ inline static long int globalCptMax = 0;
+ inline static long int globalCptMove = 0;
+ inline static long int globalCptPermute = 0;
+ inline static long int globalCptSet = 0;
+ inline static long int globalCptLoad = 0;
+ inline static long int globalCptStore = 0;
+ inline static long int globalCptCmp = 0;
+ inline static long int globalCptCompress = 0;
+inline void PrintCounters(){
+ std::cout << "Counter min: " << globalCptMin << "\n";
+ std::cout << "Counter max: " << globalCptMax << "\n";
+ std::cout << "Counter mov: " << globalCptMove << "\n";
+ std::cout << "Counter perm: " << globalCptPermute << "\n";
+ std::cout << "Counter set: " << globalCptSet << "\n";
+ std::cout << "Counter load: " << globalCptLoad << "\n";
+ std::cout << "Counter store: " << globalCptStore << "\n";
+ std::cout << "Counter cmp: " << globalCptCmp << "\n";
+ std::cout << "Counter compress: " << globalCptCompress << "\n";
+ std::cout << " Total : " << globalCptMin + globalCptMax + globalCptMove + globalCptPermute + globalCptSet + globalCptLoad + globalCptStore + globalCptCmp + globalCptCompress << "\n";
+}
+inline void ResetCounters(){
+ globalCptMin = 0;
+ globalCptMax = 0;
+ globalCptMove = 0;
+ globalCptPermute = 0;
+ globalCptSet = 0;
+ globalCptLoad = 0;
+ globalCptStore = 0;
+ globalCptCmp = 0;
+ globalCptCompress = 0;
+}
///////////////////////////////////////////////////////////
/// AVX Sort functions
@@ -46,6 +80,11 @@ namespace Sort512 {
/// Double
inline __m512d CoreSmallSort(__m512d input){
+ globalCptMin += 6;
+ globalCptMax += 6;
+ globalCptMove += 6;
+ globalCptPermute += 6;
+ globalCptSet += 6;
{
__m512i idxNoNeigh = _mm512_set_epi64(6, 7, 4, 5, 2, 3, 0, 1);
__m512d permNeigh = _mm512_permutexvar_pd(idxNoNeigh, input);
@@ -93,11 +132,17 @@ inline __m512d CoreSmallSort(__m512d input){
}
inline void CoreSmallSort(double* __restrict__ ptr1){
+ globalCptLoad += 1;
_mm512_storeu_pd(ptr1, CoreSmallSort(_mm512_loadu_pd(ptr1)));
}
inline void CoreExchangeSort2V(__m512d& input, __m512d& input2){
+ globalCptMin += 7;
+ globalCptMax += 7;
+ globalCptMove += 6;
+ globalCptPermute += 7;
+ globalCptSet += 4;
{
__m512i idxNoNeigh = _mm512_set_epi64(0, 1, 2, 3, 4, 5, 6, 7);
__m512d permNeigh = _mm512_permutexvar_pd(idxNoNeigh, input);
@@ -140,6 +185,11 @@ inline void CoreExchangeSort2V(__m512d& input, __m512d& input2){
}
inline void CoreSmallSort2(__m512d& input, __m512d& input2){
+ globalCptMin += 12;
+ globalCptMax += 12;
+ globalCptMove += 12;
+ globalCptPermute += 12;
+ globalCptSet += 6;
{
__m512i idxNoNeigh = _mm512_set_epi64(6, 7, 4, 5, 2, 3, 0, 1);
__m512d permNeigh = _mm512_permutexvar_pd(idxNoNeigh, input);
@@ -210,6 +260,8 @@ inline void CoreSmallSort2(__m512d& input, __m512d& input2){
}
inline void CoreSmallSort2(double* __restrict__ ptr1, double* __restrict__ ptr2 ){
+ globalCptLoad += 2;
+ globalCptStore += 2;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
CoreSmallSort2(input1, input2);
@@ -219,6 +271,11 @@ inline void CoreSmallSort2(double* __restrict__ ptr1, double* __restrict__ ptr2
inline void CoreSmallSort3(__m512d& input, __m512d& input2, __m512d& input3 ){
+ globalCptMin += 11;
+ globalCptMax += 11;
+ globalCptMove += 9;
+ globalCptPermute += 10;
+ globalCptSet += 4;
CoreSmallSort2(input, input2);
input3 = CoreSmallSort(input3);
{
@@ -280,6 +337,8 @@ inline void CoreSmallSort3(__m512d& input, __m512d& input2, __m512d& input3 ){
}
inline void CoreSmallSort3(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3 ){
+ globalCptLoad += 3;
+ globalCptStore += 3;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -291,6 +350,11 @@ inline void CoreSmallSort3(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort4(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4 ){
+ globalCptMin += 16;
+ globalCptMax += 16;
+ globalCptMove += 12;
+ globalCptPermute += 14;
+ globalCptSet += 4;
CoreSmallSort2(input, input2);
CoreSmallSort2(input3, input4);
{
@@ -375,6 +439,8 @@ inline void CoreSmallSort4(__m512d& input, __m512d& input2, __m512d& input3, __m
inline void CoreSmallSort4(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3, double* __restrict__ ptr4 ){
+ globalCptLoad += 4;
+ globalCptStore += 4;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -388,6 +454,11 @@ inline void CoreSmallSort4(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort5(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4, __m512d& input5 ){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 15;
+ globalCptPermute += 16;
+ globalCptSet += 4;
CoreSmallSort4(input, input2, input3, input4);
input5 = CoreSmallSort(input5);
{
@@ -491,6 +562,8 @@ inline void CoreSmallSort5(__m512d& input, __m512d& input2, __m512d& input3, __m
inline void CoreSmallSort5(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3,
double* __restrict__ ptr4, double* __restrict__ ptr5 ){
+ globalCptLoad += 5;
+ globalCptStore += 5;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -506,6 +579,11 @@ inline void CoreSmallSort5(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort6(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4, __m512d& input5, __m512d& input6 ){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 18;
+ globalCptPermute += 20;
+ globalCptSet += 4;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort2(input5, input6);
{
@@ -630,6 +708,8 @@ inline void CoreSmallSort6(__m512d& input, __m512d& input2, __m512d& input3, __m
inline void CoreSmallSort6(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3,
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6 ){
+ globalCptLoad += 6;
+ globalCptStore += 6;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -648,6 +728,11 @@ inline void CoreSmallSort6(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort7(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7 ){
+ globalCptMin += 30;
+ globalCptMax += 30;
+ globalCptMove += 21;
+ globalCptPermute += 24;
+ globalCptSet += 4;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort3(input5, input6, input7);
{
@@ -793,6 +878,8 @@ inline void CoreSmallSort7(__m512d& input, __m512d& input2, __m512d& input3, __m
inline void CoreSmallSort7(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3,
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6,
double* __restrict__ ptr7){
+ globalCptLoad += 7;
+ globalCptStore += 7;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -813,6 +900,11 @@ inline void CoreSmallSort7(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort8(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8 ){
+ globalCptMin += 36;
+ globalCptMax += 36;
+ globalCptMove += 24;
+ globalCptPermute += 28;
+ globalCptSet += 4;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort4(input5, input6, input7, input8);
{
@@ -984,6 +1076,8 @@ inline void CoreSmallSort8(__m512d& input, __m512d& input2, __m512d& input3, __m
inline void CoreSmallSort8(double* __restrict__ ptr1, double* __restrict__ ptr2, double* __restrict__ ptr3,
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6,
double* __restrict__ ptr7, double* __restrict__ ptr8 ){
+ globalCptLoad += 8;
+ globalCptStore += 8;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -1005,6 +1099,11 @@ inline void CoreSmallSort8(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallEnd1(__m512d& input){
+ globalCptMin += 3;
+ globalCptMax += 3;
+ globalCptMove += 3;
+ globalCptPermute += 3;
+ globalCptSet += 3;
{
__m512i idxNoNeigh = _mm512_set_epi64(3, 2, 1, 0, 7, 6, 5, 4);
__m512d permNeigh = _mm512_permutexvar_pd(idxNoNeigh, input);
@@ -1029,6 +1128,11 @@ inline void CoreSmallEnd1(__m512d& input){
}
inline void CoreSmallEnd2(__m512d& input, __m512d& input2){
+ globalCptMin += 7;
+ globalCptMax += 7;
+ globalCptMove += 6;
+ globalCptPermute += 6;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input2, inputCopy);
@@ -1070,6 +1174,11 @@ inline void CoreSmallEnd2(__m512d& input, __m512d& input2){
}
inline void CoreSmallEnd3(__m512d& input, __m512d& input2, __m512d& input3){
+ globalCptMin += 11;
+ globalCptMax += 11;
+ globalCptMove += 9;
+ globalCptPermute += 9;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input3, inputCopy);
@@ -1128,6 +1237,11 @@ inline void CoreSmallEnd3(__m512d& input, __m512d& input2, __m512d& input3){
}
inline void CoreSmallEnd4(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4){
+ globalCptMin += 16;
+ globalCptMax += 16;
+ globalCptMove += 12;
+ globalCptPermute += 12;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input3, inputCopy);
@@ -1209,6 +1323,11 @@ inline void CoreSmallEnd4(__m512d& input, __m512d& input2, __m512d& input3, __m5
inline void CoreSmallEnd5(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 15;
+ globalCptPermute += 15;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input5, inputCopy);
@@ -1307,6 +1426,11 @@ inline void CoreSmallEnd5(__m512d& input, __m512d& input2, __m512d& input3, __m5
inline void CoreSmallEnd6(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 18;
+ globalCptPermute += 18;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input5, inputCopy);
@@ -1427,6 +1551,11 @@ inline void CoreSmallEnd6(__m512d& input, __m512d& input2, __m512d& input3, __m5
inline void CoreSmallEnd7(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7){
+ globalCptMin += 30;
+ globalCptMax += 30;
+ globalCptMove += 21;
+ globalCptPermute += 21;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input5, inputCopy);
@@ -1570,6 +1699,11 @@ inline void CoreSmallEnd7(__m512d& input, __m512d& input2, __m512d& input3, __m5
inline void CoreSmallEnd8(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8 ){
+ globalCptMin += 36;
+ globalCptMax += 36;
+ globalCptMove += 24;
+ globalCptPermute += 24;
+ globalCptSet += 3;
{
__m512d inputCopy = input;
input = _mm512_min_pd(input5, inputCopy);
@@ -1740,6 +1874,10 @@ inline void CoreSmallEnd8(__m512d& input, __m512d& input2, __m512d& input3, __m5
inline void CoreSmallSort9(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9){
+ globalCptMin += 1;
+ globalCptMax += 1;
+ globalCptPermute += 1;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
input9 = CoreSmallSort(input9);
{
@@ -1760,6 +1898,8 @@ inline void CoreSmallSort9(double* __restrict__ ptr1, double* __restrict__ ptr2,
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6,
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9){
+ globalCptLoad += 9;
+ globalCptStore += 9;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -1786,6 +1926,10 @@ inline void CoreSmallSort9(double* __restrict__ ptr1, double* __restrict__ ptr2,
inline void CoreSmallSort10(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10){
+ globalCptMin += 2;
+ globalCptMax += 2;
+ globalCptPermute += 2;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort2(input9, input10);
{
@@ -1809,6 +1953,8 @@ inline void CoreSmallSort10(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6,
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9, double* __restrict__ ptr10){
+ globalCptLoad += 10;
+ globalCptStore += 10;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -1836,6 +1982,10 @@ inline void CoreSmallSort10(double* __restrict__ ptr1, double* __restrict__ ptr2
inline void CoreSmallSort11(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11){
+ globalCptMin += 3;
+ globalCptMax += 3;
+ globalCptPermute += 3;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort3(input9, input10, input11);
{
@@ -1862,6 +2012,8 @@ inline void CoreSmallSort11(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr4, double* __restrict__ ptr5, double* __restrict__ ptr6,
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11){
+ globalCptLoad += 11;
+ globalCptStore += 11;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -1891,6 +2043,10 @@ inline void CoreSmallSort11(double* __restrict__ ptr1, double* __restrict__ ptr2
inline void CoreSmallSort12(__m512d& input, __m512d& input2, __m512d& input3, __m512d& input4,
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11, __m512d& input12){
+ globalCptMin += 4;
+ globalCptMax += 4;
+ globalCptPermute += 4;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort4(input9, input10, input11, input12);
{
@@ -1921,6 +2077,8 @@ inline void CoreSmallSort12(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11,
double* __restrict__ ptr12){
+ globalCptLoad += 12;
+ globalCptStore += 12;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -1953,6 +2111,10 @@ inline void CoreSmallSort13(__m512d& input, __m512d& input2, __m512d& input3, __
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11, __m512d& input12,
__m512d& input13){
+ globalCptMin += 5;
+ globalCptMax += 5;
+ globalCptPermute += 5;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort5(input9, input10, input11, input12, input13);
{
@@ -1986,6 +2148,8 @@ inline void CoreSmallSort13(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11,
double* __restrict__ ptr12, double* __restrict__ ptr13){
+ globalCptLoad += 13;
+ globalCptStore += 13;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -2020,6 +2184,10 @@ inline void CoreSmallSort14(__m512d& input, __m512d& input2, __m512d& input3, __
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11, __m512d& input12,
__m512d& input13, __m512d& input14){
+ globalCptMin += 6;
+ globalCptMax += 6;
+ globalCptPermute += 6;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort6(input9, input10, input11, input12, input13, input14);
{
@@ -2056,6 +2224,8 @@ inline void CoreSmallSort14(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr7, double* __restrict__ ptr8,
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11,
double* __restrict__ ptr12, double* __restrict__ ptr13, double* __restrict__ ptr14){
+ globalCptLoad += 14;
+ globalCptStore += 14;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -2092,6 +2262,10 @@ inline void CoreSmallSort15(__m512d& input, __m512d& input2, __m512d& input3, __
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11, __m512d& input12,
__m512d& input13, __m512d& input14, __m512d& input15){
+ globalCptMin += 7;
+ globalCptMax += 7;
+ globalCptPermute += 7;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort7(input9, input10, input11, input12, input13, input14, input15);
{
@@ -2132,6 +2306,8 @@ inline void CoreSmallSort15(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11,
double* __restrict__ ptr12, double* __restrict__ ptr13, double* __restrict__ ptr14,
double* __restrict__ ptr15){
+ globalCptLoad += 15;
+ globalCptStore += 15;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -2171,6 +2347,10 @@ inline void CoreSmallSort16(__m512d& input, __m512d& input2, __m512d& input3, __
__m512d& input5, __m512d& input6, __m512d& input7, __m512d& input8,
__m512d& input9, __m512d& input10, __m512d& input11, __m512d& input12,
__m512d& input13, __m512d& input14, __m512d& input15, __m512d& input16){
+ globalCptMin += 8;
+ globalCptMax += 8;
+ globalCptPermute += 8;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort8(input9, input10, input11, input12, input13, input14, input15, input16);
{
@@ -2214,6 +2394,8 @@ inline void CoreSmallSort16(double* __restrict__ ptr1, double* __restrict__ ptr2
double* __restrict__ ptr9, double* __restrict__ ptr10, double* __restrict__ ptr11,
double* __restrict__ ptr12, double* __restrict__ ptr13, double* __restrict__ ptr14,
double* __restrict__ ptr15, double* __restrict__ ptr16){
+ globalCptLoad += 16;
+ globalCptStore += 16;
__m512d input1 = _mm512_loadu_pd(ptr1);
__m512d input2 = _mm512_loadu_pd(ptr2);
__m512d input3 = _mm512_loadu_pd(ptr3);
@@ -2253,6 +2435,10 @@ inline void CoreSmallSort16(double* __restrict__ ptr1, double* __restrict__ ptr2
inline void SmallSort16V(double* __restrict__ ptr, const size_t length){
+ globalCptMax += 17;
+ globalCptLoad += 136;
+ globalCptStore += 120;
+ globalCptCompress += 16;
// length is limited to 4 times size of a vec
const int nbValuesInVec = 8;
const int nbVecs = (length+nbValuesInVec-1)/nbValuesInVec;
@@ -2635,6 +2821,11 @@ inline void SmallSort16V(double* __restrict__ ptr, const size_t length){
/// Int
inline __m512i CoreSmallSort(__m512i input){
+ globalCptMin += 10;
+ globalCptMax += 10;
+ globalCptMove += 10;
+ globalCptPermute += 10;
+ globalCptSet += 10;
{
__m512i idxNoNeigh = _mm512_set_epi32(14, 15, 12, 13, 10, 11, 8, 9,
6, 7, 4, 5, 2, 3, 0, 1);
@@ -2720,11 +2911,17 @@ inline __m512i CoreSmallSort(__m512i input){
}
inline void CoreSmallSort(int* __restrict__ ptr1){
+ globalCptLoad += 1;
_mm512_storeu_si512(ptr1, CoreSmallSort(_mm512_loadu_si512(ptr1)));
}
inline void CoreExchangeSort2V(__m512i& input, __m512i& input2 ){
+ globalCptMin += 9;
+ globalCptMax += 9;
+ globalCptMove += 8;
+ globalCptPermute += 9;
+ globalCptSet += 5;
{
__m512i idxNoNeigh = _mm512_set_epi32(0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15);
@@ -2783,6 +2980,11 @@ inline void CoreExchangeSort2V(__m512i& input, __m512i& input2 ){
}
inline void CoreSmallSort2(__m512i& input, __m512i& input2 ){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 20;
+ globalCptPermute += 20;
+ globalCptSet += 10;
{
__m512i idxNoNeigh = _mm512_set_epi32(14, 15, 12, 13, 10, 11, 8, 9,
6, 7, 4, 5, 2, 3, 0, 1);
@@ -2907,6 +3109,8 @@ inline void CoreSmallSort2(__m512i& input, __m512i& input2 ){
}
inline void CoreSmallSort2(int* __restrict__ ptr1, int* __restrict__ ptr2 ){
+ globalCptLoad += 2;
+ globalCptStore += 2;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
CoreSmallSort2(input1, input2);
@@ -2916,6 +3120,11 @@ inline void CoreSmallSort2(int* __restrict__ ptr1, int* __restrict__ ptr2 ){
inline void CoreSmallSort3(__m512i& input, __m512i& input2, __m512i& input3 ){
+ globalCptMin += 14;
+ globalCptMax += 14;
+ globalCptMove += 12;
+ globalCptPermute += 13;
+ globalCptSet += 5;
CoreSmallSort2(input, input2);
input3 = CoreSmallSort(input3);
{
@@ -2997,6 +3206,8 @@ inline void CoreSmallSort3(__m512i& input, __m512i& input2, __m512i& input3 ){
}
inline void CoreSmallSort3(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3 ){
+ globalCptLoad += 3;
+ globalCptStore += 3;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3007,6 +3218,11 @@ inline void CoreSmallSort3(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
}
inline void CoreSmallSort4(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4 ){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 16;
+ globalCptPermute += 18;
+ globalCptSet += 5;
CoreSmallSort2(input, input2);
CoreSmallSort2(input3, input4);
{
@@ -3114,6 +3330,8 @@ inline void CoreSmallSort4(__m512i& input, __m512i& input2, __m512i& input3, __m
}
inline void CoreSmallSort4(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4 ){
+ globalCptLoad += 4;
+ globalCptStore += 4;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3127,6 +3345,11 @@ inline void CoreSmallSort4(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort5(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4, __m512i& input5 ){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 20;
+ globalCptPermute += 21;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4);
input5 = CoreSmallSort(input5);
{
@@ -3256,6 +3479,8 @@ inline void CoreSmallSort5(__m512i& input, __m512i& input2, __m512i& input3, __m
}
inline void CoreSmallSort5(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4, int* __restrict__ ptr5 ){
+ globalCptLoad += 5;
+ globalCptStore += 5;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3272,6 +3497,11 @@ inline void CoreSmallSort5(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort6(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6){
+ globalCptMin += 31;
+ globalCptMax += 31;
+ globalCptMove += 24;
+ globalCptPermute += 26;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort2(input5, input6);
{
@@ -3427,6 +3657,8 @@ inline void CoreSmallSort6(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort6(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6 ){
+ globalCptLoad += 6;
+ globalCptStore += 6;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3445,6 +3677,11 @@ inline void CoreSmallSort6(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort7(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7){
+ globalCptMin += 37;
+ globalCptMax += 37;
+ globalCptMove += 28;
+ globalCptPermute += 31;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort3(input5, input6, input7);
{
@@ -3624,6 +3861,8 @@ inline void CoreSmallSort7(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort7(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7 ){
+ globalCptLoad += 7;
+ globalCptStore += 7;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3643,6 +3882,11 @@ inline void CoreSmallSort7(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort8(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8 ){
+ globalCptMin += 44;
+ globalCptMax += 44;
+ globalCptMove += 32;
+ globalCptPermute += 36;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4);
CoreSmallSort4(input5, input6, input7, input8);
{
@@ -3851,6 +4095,8 @@ inline void CoreSmallSort8(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort8(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8 ){
+ globalCptLoad += 8;
+ globalCptStore += 8;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -3871,6 +4117,11 @@ inline void CoreSmallSort8(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
}
inline void CoreSmallEnd1(__m512i& input){
+ globalCptMin += 4;
+ globalCptMax += 4;
+ globalCptMove += 4;
+ globalCptPermute += 4;
+ globalCptSet += 4;
{
__m512i idxNoNeigh = _mm512_set_epi32(7, 6, 5, 4, 3, 2, 1, 0,
15, 14, 13, 12, 11, 10, 9, 8);
@@ -3906,6 +4157,11 @@ inline void CoreSmallEnd1(__m512i& input){
}
inline void CoreSmallEnd2(__m512i& input, __m512i& input2){
+ globalCptMin += 9;
+ globalCptMax += 9;
+ globalCptMove += 8;
+ globalCptPermute += 8;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input2, inputCopy);
@@ -3962,6 +4218,11 @@ inline void CoreSmallEnd2(__m512i& input, __m512i& input2){
}
inline void CoreSmallEnd3(__m512i& input, __m512i& input2, __m512i& input3){
+ globalCptMin += 14;
+ globalCptMax += 14;
+ globalCptMove += 12;
+ globalCptPermute += 12;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input3, inputCopy);
@@ -4039,6 +4300,11 @@ inline void CoreSmallEnd3(__m512i& input, __m512i& input2, __m512i& input3){
}
inline void CoreSmallEnd4(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 16;
+ globalCptPermute += 16;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input3, inputCopy);
@@ -4143,6 +4409,11 @@ inline void CoreSmallEnd4(__m512i& input, __m512i& input2, __m512i& input3, __m5
inline void CoreSmallEnd5(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 20;
+ globalCptPermute += 20;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input5, inputCopy);
@@ -4268,6 +4539,11 @@ inline void CoreSmallEnd5(__m512i& input, __m512i& input2, __m512i& input3, __m5
inline void CoreSmallEnd6(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6){
+ globalCptMin += 31;
+ globalCptMax += 31;
+ globalCptMove += 24;
+ globalCptPermute += 24;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input5, inputCopy);
@@ -4419,6 +4695,11 @@ inline void CoreSmallEnd6(__m512i& input, __m512i& input2, __m512i& input3, __m5
inline void CoreSmallEnd7(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7){
+ globalCptMin += 38;
+ globalCptMax += 38;
+ globalCptMove += 28;
+ globalCptPermute += 28;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input5, inputCopy);
@@ -4601,6 +4882,11 @@ inline void CoreSmallEnd7(__m512i& input, __m512i& input2, __m512i& input3, __m5
inline void CoreSmallEnd8(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8 ){
+ globalCptMin += 45;
+ globalCptMax += 45;
+ globalCptMove += 32;
+ globalCptPermute += 32;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
input = _mm512_min_epi32(input5, inputCopy);
@@ -4815,6 +5101,10 @@ inline void CoreSmallEnd8(__m512i& input, __m512i& input2, __m512i& input3, __m5
inline void CoreSmallSort9(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9){
+ globalCptMin += 1;
+ globalCptMax += 1;
+ globalCptPermute += 1;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
input9 = CoreSmallSort(input9);
{
@@ -4833,6 +5123,8 @@ inline void CoreSmallSort9(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort9(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9){
+ globalCptLoad += 9;
+ globalCptStore += 9;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -4858,6 +5150,10 @@ inline void CoreSmallSort9(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort10(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10){
+ globalCptMin += 2;
+ globalCptMax += 2;
+ globalCptPermute += 2;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort2(input9, input10);
{
@@ -4879,6 +5175,8 @@ inline void CoreSmallSort10(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort10(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10){
+ globalCptLoad += 10;
+ globalCptStore += 10;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -4906,6 +5204,10 @@ inline void CoreSmallSort10(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort11(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11){
+ globalCptMin += 3;
+ globalCptMax += 3;
+ globalCptPermute += 3;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort3(input9, input10, input11);
{
@@ -4930,6 +5232,8 @@ inline void CoreSmallSort11(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort11(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11){
+ globalCptLoad += 11;
+ globalCptStore += 11;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -4959,6 +5263,10 @@ inline void CoreSmallSort11(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void CoreSmallSort12(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11, __m512i& input12 ){
+ globalCptMin += 4;
+ globalCptMax += 4;
+ globalCptPermute += 4;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort4(input9, input10, input11, input12);
{
@@ -4986,6 +5294,8 @@ inline void CoreSmallSort12(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort12(int* __restrict__ ptr1, int* __restrict__ ptr2, int* __restrict__ ptr3, int* __restrict__ ptr4,
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11, int* __restrict__ ptr12){
+ globalCptLoad += 12;
+ globalCptStore += 12;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -5018,6 +5328,10 @@ inline void CoreSmallSort13(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11, __m512i& input12,
__m512i& input13 ){
+ globalCptMin += 5;
+ globalCptMax += 5;
+ globalCptPermute += 5;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort5(input9, input10, input11, input12, input13);
{
@@ -5049,6 +5363,8 @@ inline void CoreSmallSort13(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11, int* __restrict__ ptr12,
int* __restrict__ ptr13){
+ globalCptLoad += 13;
+ globalCptStore += 13;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -5084,6 +5400,10 @@ inline void CoreSmallSort14(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11, __m512i& input12,
__m512i& input13, __m512i& input14 ){
+ globalCptMin += 6;
+ globalCptMax += 6;
+ globalCptPermute += 6;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort6(input9, input10, input11, input12, input13, input14);
{
@@ -5118,6 +5438,8 @@ inline void CoreSmallSort14(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11, int* __restrict__ ptr12,
int* __restrict__ ptr13, int* __restrict__ ptr14){
+ globalCptLoad += 14;
+ globalCptStore += 14;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -5155,6 +5477,10 @@ inline void CoreSmallSort15(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11, __m512i& input12,
__m512i& input13, __m512i& input14, __m512i& input15 ){
+ globalCptMin += 7;
+ globalCptMax += 7;
+ globalCptPermute += 7;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort7(input9, input10, input11, input12, input13, input14, input15);
{
@@ -5192,6 +5518,8 @@ inline void CoreSmallSort15(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11, int* __restrict__ ptr12,
int* __restrict__ ptr13, int* __restrict__ ptr14, int* __restrict__ ptr15){
+ globalCptLoad += 15;
+ globalCptStore += 15;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -5231,6 +5559,10 @@ inline void CoreSmallSort16(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input9, __m512i& input10, __m512i& input11, __m512i& input12,
__m512i& input13, __m512i& input14, __m512i& input15, __m512i& input16 ){
+ globalCptMin += 8;
+ globalCptMax += 8;
+ globalCptPermute += 8;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8);
CoreSmallSort8(input9, input10, input11, input12, input13, input14, input15, input16);
{
@@ -5271,6 +5603,8 @@ inline void CoreSmallSort16(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
int* __restrict__ ptr5, int* __restrict__ ptr6, int* __restrict__ ptr7, int* __restrict__ ptr8,
int* __restrict__ ptr9, int* __restrict__ ptr10, int* __restrict__ ptr11, int* __restrict__ ptr12,
int* __restrict__ ptr13, int* __restrict__ ptr14, int* __restrict__ ptr15, int* __restrict__ ptr16){
+ globalCptLoad += 16;
+ globalCptStore += 16;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr2);
__m512i input3 = _mm512_loadu_si512(ptr3);
@@ -5310,6 +5644,10 @@ inline void CoreSmallSort16(int* __restrict__ ptr1, int* __restrict__ ptr2, int*
inline void SmallSort16V(int* __restrict__ ptr, const size_t length){
+ globalCptSet += 16;
+ globalCptLoad += 136;
+ globalCptStore += 120;
+ globalCptCompress += 16;
// length is limited to 4 times size of a vec
const int nbValuesInVec = 16;
const int nbVecs = (length+nbValuesInVec-1)/nbValuesInVec;
@@ -5729,6 +6067,10 @@ inline int popcount(__mmask16 mask){
template <class IndexType>
static inline IndexType Partition512(int array[], IndexType left, IndexType right,
const int pivot){
+ globalCptSet += 1;
+ globalCptLoad += 5;
+ globalCptCmp += 4;
+ globalCptCompress += 8;
const IndexType S = 16;//(512/8)/sizeof(int);
if(right-left+1 < 2*S){
@@ -5821,6 +6163,10 @@ static inline IndexType Partition512(int array[], IndexType left, IndexType righ
template <class IndexType>
static inline IndexType Partition512(double array[], IndexType left, IndexType right,
const double pivot){
+ globalCptSet += 1;
+ globalCptLoad += 5;
+ globalCptCmp += 4;
+ globalCptCompress += 8;
const IndexType S = 8;//(512/8)/sizeof(double);
if(right-left+1 < 2*S){
@@ -5991,6 +6337,10 @@ static inline void CoreSortTaskPartition(SortType array[], const IndexType left,
template <class SortType, class IndexType = size_t>
static inline void SortOmpPartition(SortType array[], const IndexType size){
+ globalCptMax += 1;
+ // const int nbTasksRequiere = (omp_get_max_threads() * 5);
+ // int deep = 0;
+ // while( (1 << deep) < nbTasksRequiere ) deep += 1;
int deep = 0;
while( (IndexType(1) << deep) < size ) deep += 1;
@@ -6005,6 +6355,7 @@ static inline void SortOmpPartition(SortType array[], const IndexType size){
template <class SortType, class IndexType = size_t>
static inline void SortOmpMerge(SortType array[], const IndexType size){
+ globalCptMax += 1;
const long int MAX_THREADS = 128;
const long int LOG2_MAX_THREADS = 7;
int done[LOG2_MAX_THREADS][MAX_THREADS] = {0};
@@ -6059,6 +6410,7 @@ static inline void SortOmpMerge(SortType array[], const IndexType size){
template <class SortType, class IndexType = size_t>
static inline void SortOmpMergeDeps(SortType array[], const IndexType size){
+ globalCptMax += 1;
int nbParts = 1;
while(nbParts < omp_get_max_threads()){
nbParts <<= 1;
@@ -6103,6 +6455,7 @@ static inline void SortOmpMergeDeps(SortType array[], const IndexType size){
template <class SortType, class IndexType = size_t>
static inline void SortOmpParMerge(SortType array[], const IndexType size){
+ globalCptMax += 1;
if(size < omp_get_max_threads()){
CoreSort<SortType,IndexType>(array,0,size-1);
return;
diff --git a/sort512kv.hpp b/sort512kv.hpp
index 3434190..0915429 100644
--- a/sort512kv.hpp
+++ b/sort512kv.hpp
@@ -22,6 +22,7 @@
/// Intel : -xCOMMON-AVX512 -xCORE-AVX512 -qopenmp
///
/// Or use "-march=native -mtune=native" if you are already on the right platform ("native can be replaced by "knl" or "skylake")
+/// You are in the branch with counters! You must use also use -std=c++17 (for inline static variables)
//////////////////////////////////////////////////////////
#ifndef SORT512KV_HPP
#define SORT512LV_HPP
@@ -35,7 +36,40 @@
#include <omp.h>
#endif
+#include <iostream>
namespace Sort512kv {
+ inline static long int globalCptMin = 0;
+ inline static long int globalCptMax = 0;
+ inline static long int globalCptMove = 0;
+ inline static long int globalCptPermute = 0;
+ inline static long int globalCptSet = 0;
+ inline static long int globalCptLoad = 0;
+ inline static long int globalCptStore = 0;
+ inline static long int globalCptCmp = 0;
+ inline static long int globalCptCompress = 0;
+inline void PrintCounters(){
+ std::cout << "Counter min: " << globalCptMin << "\n";
+ std::cout << "Counter max: " << globalCptMax << "\n";
+ std::cout << "Counter mov: " << globalCptMove << "\n";
+ std::cout << "Counter perm: " << globalCptPermute << "\n";
+ std::cout << "Counter set: " << globalCptSet << "\n";
+ std::cout << "Counter load: " << globalCptLoad << "\n";
+ std::cout << "Counter store: " << globalCptStore << "\n";
+ std::cout << "Counter cmp: " << globalCptCmp << "\n";
+ std::cout << "Counter compress: " << globalCptCompress << "\n";
+ std::cout << " Total : " << globalCptMin + globalCptMax + globalCptMove + globalCptPermute + globalCptSet + globalCptLoad + globalCptStore + globalCptCmp + globalCptCompress << "\n";
+}
+inline void ResetCounters(){
+ globalCptMin = 0;
+ globalCptMax = 0;
+ globalCptMove = 0;
+ globalCptPermute = 0;
+ globalCptSet = 0;
+ globalCptLoad = 0;
+ globalCptStore = 0;
+ globalCptCmp = 0;
+ globalCptCompress = 0;
+}
///////////////////////////////////////////////////////////
/// AVX Sort functions
@@ -44,6 +78,11 @@ namespace Sort512kv {
/// Int
inline void CoreSmallSort(__m512i& input, __m512i& values){
+ globalCptMin += 10;
+ globalCptMax += 10;
+ globalCptMove += 10;
+ globalCptPermute += 20;
+ globalCptSet += 10;
{
__m512i idxNoNeigh = _mm512_set_epi32(14, 15, 12, 13, 10, 11, 8, 9,
6, 7, 4, 5, 2, 3, 0, 1);
@@ -177,6 +216,8 @@ inline void CoreSmallSort(__m512i& input, __m512i& values){
}
inline void CoreSmallSort(int* __restrict__ ptr1, int* __restrict__ ptrVal){
+ globalCptLoad += 2;
+ globalCptStore += 2;
__m512i v = _mm512_loadu_si512(ptr1);
__m512i v_val = _mm512_loadu_si512(ptrVal);
CoreSmallSort(v, v_val);
@@ -188,6 +229,11 @@ inline void CoreSmallSort(int* __restrict__ ptr1, int* __restrict__ ptrVal){
inline void CoreExchangeSort2V(__m512i& input, __m512i& input2,
__m512i& input_val, __m512i& input2_val){
+ globalCptMin += 9;
+ globalCptMax += 9;
+ globalCptMove += 10;
+ globalCptPermute += 18;
+ globalCptSet += 5;
{
__m512i idxNoNeigh = _mm512_set_epi32(0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15);
@@ -290,6 +336,11 @@ inline void CoreExchangeSort2V(__m512i& input, __m512i& input2,
inline void CoreSmallSort2(__m512i& input, __m512i& input2,
__m512i& input_val, __m512i& input2_val){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 20;
+ globalCptPermute += 40;
+ globalCptSet += 10;
{
__m512i idxNoNeigh = _mm512_set_epi32(14, 15, 12, 13, 10, 11, 8, 9,
6, 7, 4, 5, 2, 3, 0, 1);
@@ -495,6 +546,8 @@ inline void CoreSmallSort2(__m512i& input, __m512i& input2,
inline void CoreSmallSort2(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 4;
+ globalCptStore += 4;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr1+16);
__m512i input1_val = _mm512_loadu_si512(values);
@@ -509,6 +562,11 @@ inline void CoreSmallSort2(int* __restrict__ ptr1, int* __restrict__ values ){
inline void CoreSmallSort3(__m512i& input, __m512i& input2, __m512i& input3,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val){
+ globalCptMin += 14;
+ globalCptMax += 14;
+ globalCptMove += 16;
+ globalCptPermute += 26;
+ globalCptSet += 5;
CoreSmallSort2(input, input2, input_val, input2_val);
CoreSmallSort(input3, input3_val);
{
@@ -653,6 +711,8 @@ inline void CoreSmallSort3(__m512i& input, __m512i& input2, __m512i& input3,
inline void CoreSmallSort3(int* __restrict__ ptr1, int* __restrict__ values){
+ globalCptLoad += 6;
+ globalCptStore += 6;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr1+16);
__m512i input3 = _mm512_loadu_si512(ptr1+32);
@@ -673,6 +733,11 @@ inline void CoreSmallSort3(int* __restrict__ ptr1, int* __restrict__ values){
inline void CoreSmallSort4(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 24;
+ globalCptPermute += 36;
+ globalCptSet += 5;
CoreSmallSort2(input, input2, input_val, input2_val);
CoreSmallSort2(input3, input4, input3_val, input4_val);
{
@@ -873,6 +938,8 @@ inline void CoreSmallSort4(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort4(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 8;
+ globalCptStore += 8;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr1+16);
__m512i input3 = _mm512_loadu_si512(ptr1+32);
@@ -896,6 +963,11 @@ inline void CoreSmallSort4(int* __restrict__ ptr1, int* __restrict__ values ){
inline void CoreSmallSort5(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4, __m512i& input5,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val, __m512i& input5_val){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 30;
+ globalCptPermute += 42;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4,
input_val, input2_val, input3_val, input4_val);
CoreSmallSort(input5, input5_val);
@@ -1140,6 +1212,8 @@ inline void CoreSmallSort5(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort5(int* __restrict__ ptr1, int* __restrict__ values){
+ globalCptLoad += 10;
+ globalCptStore += 10;
__m512i input1 = _mm512_loadu_si512(ptr1);
__m512i input2 = _mm512_loadu_si512(ptr1+1*16);
__m512i input3 = _mm512_loadu_si512(ptr1+2*16);
@@ -1170,6 +1244,11 @@ inline void CoreSmallSort6(__m512i& input, __m512i& input2, __m512i& input3, __m
__m512i& input5, __m512i& input6,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val){
+ globalCptMin += 31;
+ globalCptMax += 31;
+ globalCptMove += 38;
+ globalCptPermute += 52;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4,
input_val, input2_val, input3_val, input4_val);
CoreSmallSort2(input5, input6, input5_val, input6_val);
@@ -1465,6 +1544,8 @@ inline void CoreSmallSort6(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort6(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 12;
+ globalCptStore += 12;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -1499,6 +1580,11 @@ inline void CoreSmallSort7(__m512i& input, __m512i& input2, __m512i& input3, __m
__m512i& input5, __m512i& input6, __m512i& input7,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val){
+ globalCptMin += 37;
+ globalCptMax += 37;
+ globalCptMove += 46;
+ globalCptPermute += 62;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4,
input_val, input2_val, input3_val, input4_val);
CoreSmallSort3(input5, input6, input7,
@@ -1850,6 +1936,8 @@ inline void CoreSmallSort7(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort7(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 14;
+ globalCptStore += 14;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -1888,6 +1976,11 @@ inline void CoreSmallSort8(__m512i& input, __m512i& input2, __m512i& input3, __m
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val){
+ globalCptMin += 44;
+ globalCptMax += 44;
+ globalCptMove += 56;
+ globalCptPermute += 72;
+ globalCptSet += 5;
CoreSmallSort4(input, input2, input3, input4,
input_val, input2_val, input3_val, input4_val);
CoreSmallSort4(input5, input6, input7, input8,
@@ -2305,6 +2398,8 @@ inline void CoreSmallSort8(__m512i& input, __m512i& input2, __m512i& input3, __m
}
inline void CoreSmallSort8(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 16;
+ globalCptStore += 16;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -2344,6 +2439,11 @@ inline void CoreSmallSort8(int* __restrict__ ptr1, int* __restrict__ values ){
inline void CoreSmallEnd1(__m512i& input, __m512i& values){
+ globalCptMin += 4;
+ globalCptMax += 4;
+ globalCptMove += 4;
+ globalCptPermute += 8;
+ globalCptSet += 4;
{
__m512i idxNoNeigh = _mm512_set_epi32(7, 6, 5, 4, 3, 2, 1, 0,
15, 14, 13, 12, 11, 10, 9, 8);
@@ -2400,6 +2500,11 @@ inline void CoreSmallEnd1(__m512i& input, __m512i& values){
inline void CoreSmallEnd2(__m512i& input, __m512i& input2,
__m512i& input_val, __m512i& input2_val){
+ globalCptMin += 9;
+ globalCptMax += 9;
+ globalCptMove += 10;
+ globalCptPermute += 16;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input2, inputCopy);
@@ -2497,6 +2602,11 @@ inline void CoreSmallEnd2(__m512i& input, __m512i& input2,
inline void CoreSmallEnd3(__m512i& input, __m512i& input2, __m512i& input3,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val){
+ globalCptMin += 14;
+ globalCptMax += 14;
+ globalCptMove += 16;
+ globalCptPermute += 24;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input3, inputCopy);
@@ -2635,6 +2745,11 @@ inline void CoreSmallEnd3(__m512i& input, __m512i& input2, __m512i& input3,
inline void CoreSmallEnd4(__m512i& input, __m512i& input2, __m512i& input3, __m512i& input4,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val){
+ globalCptMin += 20;
+ globalCptMax += 20;
+ globalCptMove += 24;
+ globalCptPermute += 32;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input3, inputCopy);
@@ -2829,6 +2944,11 @@ inline void CoreSmallEnd5(__m512i& input, __m512i& input2, __m512i& input3, __m5
__m512i& input5,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val){
+ globalCptMin += 25;
+ globalCptMax += 25;
+ globalCptMove += 30;
+ globalCptPermute += 40;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input5, inputCopy);
@@ -3064,6 +3184,11 @@ inline void CoreSmallEnd6(__m512i& input, __m512i& input2, __m512i& input3, __m5
__m512i& input5, __m512i& input6,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val){
+ globalCptMin += 31;
+ globalCptMax += 31;
+ globalCptMove += 38;
+ globalCptPermute += 48;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input5, inputCopy);
@@ -3355,6 +3480,11 @@ inline void CoreSmallEnd7(__m512i& input, __m512i& input2, __m512i& input3, __m5
__m512i& input5, __m512i& input6, __m512i& input7,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val){
+ globalCptMin += 38;
+ globalCptMax += 38;
+ globalCptMove += 48;
+ globalCptPermute += 56;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input5, inputCopy);
@@ -3713,6 +3843,11 @@ inline void CoreSmallEnd8(__m512i& input, __m512i& input2, __m512i& input3, __m5
__m512i& input5, __m512i& input6, __m512i& input7, __m512i& input8,
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val){
+ globalCptMin += 45;
+ globalCptMax += 45;
+ globalCptMove += 57;
+ globalCptPermute += 64;
+ globalCptSet += 4;
{
__m512i inputCopy = input;
__m512i tmp_input = _mm512_min_epi32(input5, inputCopy);
@@ -4138,6 +4273,11 @@ inline void CoreSmallSort9(__m512i& input, __m512i& input2, __m512i& input3, __m
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val){
+ globalCptMin += 1;
+ globalCptMax += 1;
+ globalCptMove += 2;
+ globalCptPermute += 2;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort(input9, input9_val);
@@ -4166,6 +4306,8 @@ inline void CoreSmallSort9(__m512i& input, __m512i& input2, __m512i& input3, __m
inline void CoreSmallSort9(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 18;
+ globalCptStore += 18;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4213,6 +4355,11 @@ inline void CoreSmallSort10(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val, __m512i& input10_val){
+ globalCptMin += 2;
+ globalCptMax += 2;
+ globalCptMove += 4;
+ globalCptPermute += 4;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort2(input9, input10, input9_val, input10_val);
@@ -4255,6 +4402,8 @@ inline void CoreSmallSort10(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort10(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 20;
+ globalCptStore += 20;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4307,6 +4456,11 @@ inline void CoreSmallSort11(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input_val, __m512i& input2_val, __m512i& input3_val, __m512i& input4_val,
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val){
+ globalCptMin += 3;
+ globalCptMax += 3;
+ globalCptMove += 6;
+ globalCptPermute += 6;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort3(input9, input10, input11,
@@ -4363,6 +4517,8 @@ inline void CoreSmallSort11(__m512i& input, __m512i& input2, __m512i& input3, __
}
inline void CoreSmallSort11(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 22;
+ globalCptStore += 22;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4418,6 +4574,11 @@ inline void CoreSmallSort12(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val ,
__m512i& input12_val){
+ globalCptMin += 4;
+ globalCptMax += 4;
+ globalCptMove += 8;
+ globalCptPermute += 8;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort4(input9, input10, input11, input12,
@@ -4486,6 +4647,8 @@ inline void CoreSmallSort12(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort12(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 24;
+ globalCptStore += 24;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4548,6 +4711,11 @@ inline void CoreSmallSort13(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val ,
__m512i& input12_val, __m512i& input13_val){
+ globalCptMin += 5;
+ globalCptMax += 5;
+ globalCptMove += 10;
+ globalCptPermute += 10;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort5(input9, input10, input11, input12, input13,
@@ -4629,6 +4797,8 @@ inline void CoreSmallSort13(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort13(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 26;
+ globalCptStore += 26;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4695,6 +4865,11 @@ inline void CoreSmallSort14(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input5_val, __m512i& input6_val, __m512i& input7_val, __m512i& input8_val,
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val ,
__m512i& input12_val, __m512i& input13_val, __m512i& input14_val){
+ globalCptMin += 6;
+ globalCptMax += 6;
+ globalCptMove += 12;
+ globalCptPermute += 12;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort6(input9, input10, input11, input12, input13, input14,
@@ -4789,6 +4964,8 @@ inline void CoreSmallSort14(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort14(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 28;
+ globalCptStore += 28;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -4859,6 +5036,11 @@ inline void CoreSmallSort15(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val ,
__m512i& input12_val, __m512i& input13_val, __m512i& input14_val,
__m512i& input15_val){
+ globalCptMin += 7;
+ globalCptMax += 7;
+ globalCptMove += 14;
+ globalCptPermute += 14;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort7(input9, input10, input11, input12, input13, input14, input15,
@@ -4966,6 +5148,8 @@ inline void CoreSmallSort15(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort15(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 30;
+ globalCptStore += 30;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -5041,6 +5225,11 @@ inline void CoreSmallSort16(__m512i& input, __m512i& input2, __m512i& input3, __
__m512i& input9_val, __m512i& input10_val, __m512i& input11_val ,
__m512i& input12_val, __m512i& input13_val, __m512i& input14_val,
__m512i& input15_val,__m512i& input16_val){
+ globalCptMin += 8;
+ globalCptMax += 8;
+ globalCptMove += 16;
+ globalCptPermute += 16;
+ globalCptSet += 1;
CoreSmallSort8(input, input2, input3, input4, input5, input6, input7, input8,
input_val, input2_val, input3_val, input4_val, input5_val, input6_val, input7_val, input8_val);
CoreSmallSort8(input9, input10, input11, input12, input13, input14, input15, input16,
@@ -5162,6 +5351,8 @@ inline void CoreSmallSort16(__m512i& input, __m512i& input2, __m512i& input3, __
inline void CoreSmallSort16(int* __restrict__ ptr1, int* __restrict__ values ){
+ globalCptLoad += 32;
+ globalCptStore += 32;
__m512i input0 = _mm512_loadu_si512(ptr1+0*16);
__m512i input1 = _mm512_loadu_si512(ptr1+1*16);
__m512i input2 = _mm512_loadu_si512(ptr1+2*16);
@@ -5233,6 +5424,10 @@ inline void CoreSmallSort16(int* __restrict__ ptr1, int* __restrict__ values ){
inline void SmallSort16V(int* __restrict__ ptr, int* __restrict__ values, const size_t length){
+ globalCptSet += 32;
+ globalCptLoad += 272;
+ globalCptStore += 240;
+ globalCptCompress += 32;
// length is limited to 4 times size of a vec
const int nbValuesInVec = 16;
const int nbVecs = (length+nbValuesInVec-1)/nbValuesInVec;
@@ -5959,6 +6154,10 @@ inline int popcount(__mmask16 mask){
template <class IndexType>
static inline IndexType Partition512(int array[], int values[], IndexType left, IndexType right,
const int pivot){
+ globalCptSet += 1;
+ globalCptLoad += 10;
+ globalCptCmp += 4;
+ globalCptCompress += 16;
const IndexType S = 16;//(512/8)/sizeof(int);
if(right-left+1 < 2*S){
@@ -6163,6 +6362,7 @@ static inline void CoreSortTaskPartition(SortType array[], SortType values[], co
template <class SortType, class IndexType = size_t>
static inline void SortOmpPartition(SortType array[], SortType values[], const IndexType size){
+ globalCptMax += 1;
// const int nbTasksRequiere = (omp_get_max_threads() * 5);
// int deep = 0;
// while( (1 << deep) < nbTasksRequiere ) deep += 1;
diff --git a/sort512test.cpp b/sort512test.cpp
index 43bb1ed..0a417de 100644
--- a/sort512test.cpp
+++ b/sort512test.cpp
@@ -14,6 +14,7 @@
/// - SKL
/// Gcc : g++ -DNDEBUG -O3 -funroll-loops -faggressive-loop-optimizations -std=c++11 -mavx512f -mavx512cd -mavx512vl -mavx512bw -mavx512dq -fopenmp sort512test.cpp -o sort512test.gcc.exe
/// Intel : icpc -DNDEBUG -O3 -std=c++11 -xCOMMON-AVX512 -xCORE-AVX512 -qopenmp sort512test.cpp -o sort512test.intel.exe
+/// You are in the branch with counters! You must use also use -std=c++17 (for inline static variables)
//////////////////////////////////////////////////////////
#include "sort512.hpp"
@@ -1603,50 +1604,132 @@ int main(){
testPopcount();
testSortVec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSortVec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort2Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort2Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort3Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort3Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort4Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort4Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort5Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort5Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort6Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort6Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort7Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort7Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort8Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort8Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort9Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort9Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort10Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort10Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort11Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort11Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort12Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort12Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort13Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort13Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort14Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort14Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort15Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort15Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort16Vec();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSort16Vec_pair();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSmallVecSort<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSmallVecSort<double>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testSmallVecSort_pair<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testQs512<double>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testQs512<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testQs512_pair<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testPartition<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testPartition<double>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
testPartition_pair<int>();
+ std::cout <<"Counters \n"; Sort512::PrintCounters(); Sort512::ResetCounters();
+ std::cout <<"Counters KV \n"; Sort512kv::PrintCounters(); Sort512kv::ResetCounters();
if(test_res != 0){
std::cout << "Test failed!" << std::endl;
--
GitLab