Skip to content

Implement partial sorting algorithms #13

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Apr 19, 2023
Merged

Implement partial sorting algorithms #13

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
c82b1b2
Fix build bugs
mosullivan93 Apr 7, 2023
3cd6642
Implement partial sorting algorithms
mosullivan93 Apr 7, 2023
0cc1337
Refactor test suite and add tests for partial sort
mosullivan93 Apr 7, 2023
1204561
Add benchmarks for partial sorting functions
mosullivan93 Apr 7, 2023
9029f61
Align qselect parameter meaning with nth_element
mosullivan93 Apr 10, 2023
f4bca13
Correct language on pivot and partition functions
mosullivan93 Apr 9, 2023
f8d1c22
Modify partial sort benchmarking to vary k
mosullivan93 Apr 16, 2023
a8664ed
Merge branch 'main' into main
Apr 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,4 +38,4 @@ meson:
cd builddir && ninja

clean:
$(RM) -rf $(TESTDIR)/*.o $(UTILS)/*.o testexe benchexe builddir
$(RM) -rf $(TESTDIR)/*.o $(BENCHDIR)/*.o $(UTILS)/*.o testexe benchexe builddir
11 changes: 11 additions & 0 deletions benchmarks/bench-qsort-common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,11 @@
#ifndef AVX512_BENCH_COMMON
#define AVX512_BENCH_COMMON

#include <benchmark/benchmark.h>
#include "rand_array.h"
#include "cpuinfo.h"
#include "avx512-16bit-qsort.hpp"
#include "avx512-32bit-qsort.hpp"
#include "avx512-64bit-qsort.hpp"

#endif
72 changes: 72 additions & 0 deletions benchmarks/bench_partial_qsort.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,72 @@
#include "bench-qsort-common.h"

template <typename T>
static void avx512_partial_qsort(benchmark::State& state) {
if (!cpu_has_avx512bw()) {
state.SkipWithMessage("Requires AVX512 BW ISA");
}
if ((sizeof(T) == 2) && (!cpu_has_avx512_vbmi2())) {
state.SkipWithMessage("Requires AVX512 VBMI2 ISA");
}
// Perform setup here
int64_t K = state.range(0);
size_t ARRSIZE = 10000;
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call avx512_partial_qsort */
for (auto _ : state) {
avx512_partial_qsort<T>(arr.data(), K, ARRSIZE);

state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

template <typename T>
static void stdpartialsort(benchmark::State& state) {
// Perform setup here
int64_t K = state.range(0);
size_t ARRSIZE = 10000;
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call std::partial_sort */
for (auto _ : state) {
std::partial_sort(arr.begin(), arr.begin() + K, arr.end());

state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

// Register the function as a benchmark
BENCHMARK(avx512_partial_qsort<float>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<float>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<uint32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<uint32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<int32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<int32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);

BENCHMARK(avx512_partial_qsort<double>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<double>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<uint64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<uint64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<int64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<int64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);

//BENCHMARK(avx512_partial_qsort<float16>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<uint16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<uint16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_partial_qsort<int16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdpartialsort<int16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
72 changes: 72 additions & 0 deletions benchmarks/bench_qselect.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,72 @@
#include "bench-qsort-common.h"

template <typename T>
static void avx512_qselect(benchmark::State& state) {
if (!cpu_has_avx512bw()) {
state.SkipWithMessage("Requires AVX512 BW ISA");
}
if ((sizeof(T) == 2) && (!cpu_has_avx512_vbmi2())) {
state.SkipWithMessage("Requires AVX512 VBMI2 ISA");
}
// Perform setup here
int64_t K = state.range(0);
size_t ARRSIZE = 10000;
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call avx512 quickselect */
for (auto _ : state) {
avx512_qselect<T>(arr.data(), K, ARRSIZE);

state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

template <typename T>
static void stdnthelement(benchmark::State& state) {
// Perform setup here
int64_t K = state.range(0);
size_t ARRSIZE = 10000;
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call std::nth_element */
for (auto _ : state) {
std::nth_element(arr.begin(), arr.begin() + K, arr.end());

state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

// Register the function as a benchmark
BENCHMARK(avx512_qselect<float>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<float>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<uint32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<uint32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<int32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<int32_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);

BENCHMARK(avx512_qselect<double>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<double>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<uint64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<uint64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<int64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<int64_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);

//BENCHMARK(avx512_qselect<float16>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<uint16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<uint16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(avx512_qselect<int16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
BENCHMARK(stdnthelement<int16_t>)->Arg(10)->Arg(100)->Arg(1000)->Arg(5000);
76 changes: 3 additions & 73 deletions benchmarks/bench_qsort.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,73 +1,3 @@
#include <benchmark/benchmark.h>
#include "rand_array.h"
#include "cpuinfo.h"
#include "avx512-16bit-qsort.hpp"
#include "avx512-32bit-qsort.hpp"
#include "avx512-64bit-qsort.hpp"

template <typename T>
static void avx512_qsort(benchmark::State& state) {
if (!cpu_has_avx512bw()) {
state.SkipWithMessage("Requires AVX512 BW ISA");
}
if ((sizeof(T) == 2) && (!cpu_has_avx512_vbmi2())) {
state.SkipWithMessage("Requires AVX512 VBMI2 ISA");
}
// Perform setup here
size_t ARRSIZE = state.range(0);
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements is reverse order */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call avx512 quicksort */
for (auto _ : state) {
avx512_qsort<T>(arr.data(), ARRSIZE);
state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

template <typename T>
static void stdsort(benchmark::State& state) {
// Perform setup here
size_t ARRSIZE = state.range(0);
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements is reverse order */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call avx512 quicksort */
for (auto _ : state) {
std::sort(arr.begin(), arr.end());
state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

// Register the function as a benchmark
BENCHMARK(avx512_qsort<float>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<float>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int32_t>)->Arg(10000)->Arg(1000000);

BENCHMARK(avx512_qsort<double>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<double>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int64_t>)->Arg(10000)->Arg(10000000);

//BENCHMARK(avx512_qsort<float16>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int16_t>)->Arg(10000)->Arg(10000000);
#include "bench_qsort.hpp"
#include "bench_qselect.hpp"
#include "bench_partial_qsort.hpp"
68 changes: 68 additions & 0 deletions benchmarks/bench_qsort.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,68 @@
#include "bench-qsort-common.h"

template <typename T>
static void avx512_qsort(benchmark::State& state) {
if (!cpu_has_avx512bw()) {
state.SkipWithMessage("Requires AVX512 BW ISA");
}
if ((sizeof(T) == 2) && (!cpu_has_avx512_vbmi2())) {
state.SkipWithMessage("Requires AVX512 VBMI2 ISA");
}
// Perform setup here
size_t ARRSIZE = state.range(0);
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call avx512 quicksort */
for (auto _ : state) {
avx512_qsort<T>(arr.data(), ARRSIZE);
state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

template <typename T>
static void stdsort(benchmark::State& state) {
// Perform setup here
size_t ARRSIZE = state.range(0);
std::vector<T> arr;
std::vector<T> arr_bkp;

/* Initialize elements */
arr = get_uniform_rand_array<T>(ARRSIZE);
arr_bkp = arr;

/* call std::sort */
for (auto _ : state) {
std::sort(arr.begin(), arr.end());
state.PauseTiming();
arr = arr_bkp;
state.ResumeTiming();
}
}

// Register the function as a benchmark
BENCHMARK(avx512_qsort<float>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<float>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int32_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int32_t>)->Arg(10000)->Arg(1000000);

BENCHMARK(avx512_qsort<double>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<double>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int64_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int64_t>)->Arg(10000)->Arg(1000000);

//BENCHMARK(avx512_qsort<float16>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<uint16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<uint16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(avx512_qsort<int16_t>)->Arg(10000)->Arg(1000000);
BENCHMARK(stdsort<int16_t>)->Arg(10000)->Arg(1000000);
Loading