From f1eb4ee47bb35ea25c1b69b6a91287fa4dd31ff0 Mon Sep 17 00:00:00 2001
From: Matteo Cicuttin <datafl4sh@toxicnet.eu>
Date: Sat, 28 Mar 2020 19:41:47 +0100
Subject: [PATCH] Modified MT implementation.
---
kokkos-testing/fd_catalog/CMakeLists.txt | 55 +++++++----
kokkos-testing/fd_catalog/fd_main.cpp | 14 +--
kokkos-testing/fd_catalog/fd_wave_cpu.hpp | 106 +++++++++++++++++++++-
kokkos-testing/test_thread_overhead.cpp | 9 ++
4 files changed, 152 insertions(+), 32 deletions(-)
create mode 100644 kokkos-testing/test_thread_overhead.cpp
diff --git a/kokkos-testing/fd_catalog/CMakeLists.txt b/kokkos-testing/fd_catalog/CMakeLists.txt
index b92eb2e..1927a55 100644
--- a/kokkos-testing/fd_catalog/CMakeLists.txt
+++ b/kokkos-testing/fd_catalog/CMakeLists.txt
@@ -53,11 +53,44 @@ if (ENABLE_ITERTIME_OUTPUT)
add_definitions(-DSAVE_ITERTIME)
endif()
-option(ENABLE_DAZ_FTZ "Enable Denormals Are Zero and Flush To Zero flags" ON)
-if (ENABLE_DAZ_FTZ)
+######################################################################
+## Optimization stuff
+
+option(OPT_PREFER_512bit "Prefer 512 bit vectors with AVX512 (Clang > 10 & GCC)" OFF)
+if (OPT_PREFER_512bit)
+ # https://reviews.llvm.org/D67259
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mprefer-vector-width=512")
+endif()
+
+option(OPT_AGGRESSIVE_FP "Enable DAZ, FTZ and -ffast-math" ON)
+if (OPT_AGGRESSIVE_FP)
add_definitions(-DDISALLOW_DENORMALS)
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang" OR CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math")
+ endif()
endif()
+option(ENABLE_VECTORIZER_REMARKS "Enable vectorizer remarks" ON)
+if (ENABLE_VECTORIZER_REMARKS)
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Rpass=loop-vectorize -Rpass-missed=loop-vectorize -Rpass-analysis=loop-vectorize")
+ endif()
+
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "Intel")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -qopt-report-phase=vec -qopt-report=2")
+ endif()
+
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopt-info-vec-optimized")
+ endif()
+
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "-PGI")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Minfo")
+ endif()
+endif()
+
+######################################################################
+
option(ENABLE_KOKKOS "Enable Kokkos" OFF)
if (ENABLE_KOKKOS)
FetchContent_Declare(kokkos
@@ -91,24 +124,6 @@ if (ENABLE_CUDA)
endif()
endif()
-option(ENABLE_VECTORIZER_REMARKS "Enable vectorizer remarks" ON)
-if (ENABLE_VECTORIZER_REMARKS)
- if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Rpass=loop-vectorize -Rpass-missed=loop-vectorize -Rpass-analysis=loop-vectorize")
- endif()
-
- if (CMAKE_CXX_COMPILER_ID STREQUAL "Intel")
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -qopt-report-phase=vec -qopt-report=2")
- endif()
-
- if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopt-info-vec-optimized")
- endif()
-
- if (CMAKE_CXX_COMPILER_ID STREQUAL "-Minfo")
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopt-info-vec-optimized")
- endif()
-endif()
set(CMAKE_CXX_FLAGS_DEBUG "-g")
set(CMAKE_CXX_FLAGS_RELEASE "-O3 -march=native -g -DNDEBUG")
diff --git a/kokkos-testing/fd_catalog/fd_main.cpp b/kokkos-testing/fd_catalog/fd_main.cpp
index f2d3de2..cbf749a 100644
--- a/kokkos-testing/fd_catalog/fd_main.cpp
+++ b/kokkos-testing/fd_catalog/fd_main.cpp
@@ -29,7 +29,7 @@ int main(int argc, char **argv)
_MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
std::cout << "Denormals: FTZ and DAZ" << std::endl;
#endif
-
+
_MM_SET_EXCEPTION_MASK(_MM_GET_EXCEPTION_MASK() & ~_MM_MASK_INVALID);
//_MM_SET_EXCEPTION_MASK(_MM_GET_EXCEPTION_MASK() & ~_MM_MASK_DENORM);
@@ -52,9 +52,9 @@ int main(int argc, char **argv)
wave_equation_context<T> wec(sz, sz, 1, 0.1, 0.0001, 5000);
ofs << sz << " ";
- //wec.init();
- //time = solve_sequential(wec);
- //ofs << time << " ";
+ wec.init();
+ time = solve_sequential(wec);
+ ofs << time << " ";
wec.init();
time = solve_sequential_blocked(wec);
@@ -68,9 +68,9 @@ int main(int argc, char **argv)
for (size_t threads = 1; threads < maxthreads; threads *= 2)
{
- //wec.init();
- //time = solve_multithread(wec, threads);
- //ofs << time << " ";
+ wec.init();
+ time = solve_multithread_nopool(wec, threads);
+ ofs << time << " ";
}
ofs << std::endl;
diff --git a/kokkos-testing/fd_catalog/fd_wave_cpu.hpp b/kokkos-testing/fd_catalog/fd_wave_cpu.hpp
index 73f3d75..31b620f 100644
--- a/kokkos-testing/fd_catalog/fd_wave_cpu.hpp
+++ b/kokkos-testing/fd_catalog/fd_wave_cpu.hpp
@@ -44,7 +44,7 @@ operator<<(std::ostream& os, const tile& tl)
#define BLK_ROWS 56
#define BLK_COLS 56
-//#define KERNEL_BLK_USE_MEMCPY
+#define KERNEL_BLK_USE_MEMCPY
#define KERNEL_BLK_USE_TEMP
template<typename T>
@@ -101,7 +101,7 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
assert(i < BLK_ROWS+2*WAVE_8_HALO_SIZE);
memcpy(s_curr[i], g_curr.data(ofs_y, ofs_x), (BLK_COLS+2*WAVE_8_HALO_SIZE)*sizeof(T));
#else
- #pragma ivdep
+ //#pragma ivdep
#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < BLK_COLS+2*WAVE_8_HALO_SIZE; j++)
{
@@ -114,7 +114,7 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
#endif /* KERNEL_BLK_USE_TEMP */
for (size_t i = 0; i < BLK_ROWS; i++)
{
- #pragma ivdep
+ //#pragma ivdep
//#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < BLK_COLS; j++)
{
@@ -337,6 +337,7 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
g_next(maxrow-1, j) = 0;
}
+ #pragma loop distribute(enable)
for (size_t i = 0; i < maxrow; i++)
{
g_next(i, 0) = 0;
@@ -349,7 +350,7 @@ void
wave_2D_kernel(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
fd_grid<T>& g_next,
const wave_2D_params<T>& params,
- size_t thread_id = 0, size_t num_threads = 1)
+ size_t from = 0, size_t to = 0)
{
int maxrow = params.maxrow;
int maxcol = params.maxcol;
@@ -357,6 +358,9 @@ wave_2D_kernel(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
T c = params.velocity;
T a = params.damping;
+ if (to == 0)
+ to = maxrow;
+
assert(maxcol > 1);
assert(maxrow > 1);
@@ -373,8 +377,9 @@ wave_2D_kernel(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
T one_minus_adt = (1.0 - a*dt);
T two_minus_adt = (2.0 - a*dt);
- for (size_t i = thread_id; i < maxrow; i += num_threads)
+ for (size_t i = from; i < to; i++)
{
+ #pragma clang loop vectorize(enable)
for (size_t j = 0; j < maxcol; j++)
{
T lapl = 0.0;
@@ -614,3 +619,94 @@ double solve_multithread(wave_equation_context<T>& wec, size_t nths)
return time/wec.maxiter;
}
+template<typename T>
+double solve_multithread_nopool(wave_equation_context<T>& wec, size_t nths)
+{
+ /* Simulation parameters */
+ wave_2D_params<T> params;
+ params.maxcol = wec.g_curr.domain_cols();
+ params.maxrow = wec.g_curr.domain_rows();
+ params.dt = wec.dt;
+ params.velocity = wec.velocity;
+ params.damping = wec.damping;
+
+
+ /* Multithreading stuff */
+ std::mutex cv_mtx;
+ std::condition_variable prod_cv;
+ std::condition_variable cons_cv;
+ std::vector<bool> thread_done(nths);
+ bool iteration_finished = false;
+
+ auto thread_lambda = [&](size_t thread_id, size_t num_threads) {
+#ifdef DISALLOW_DENORMALS
+ _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
+ _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
+#endif
+
+ size_t count = params.maxrow / num_threads;
+ size_t remainder = params.maxrow % num_threads;
+
+ size_t from = 0;
+ size_t to = 0;
+
+ if (thread_id < remainder)
+ {
+ from = thread_id * (count + 1);
+ to = from + count;
+ }
+ else
+ {
+ from = thread_id * count + remainder;
+ to = from + (count - 1);
+ }
+
+ wave_2D_kernel(wec.g_prev, wec.g_curr, wec.g_next, params, from, to);
+ };
+
+
+ double time = 0.0;
+
+ for (size_t iter = 0; iter < wec.maxiter; iter++)
+ {
+ auto start = std::chrono::high_resolution_clock::now();
+
+ std::vector<std::thread> threads(nths);
+ for (size_t i = 0; i < nths; i++)
+ threads[i] = std::thread(thread_lambda, i, nths);
+
+ for (auto& th : threads)
+ th.join();
+
+ auto stop = std::chrono::high_resolution_clock::now();
+ std::chrono::duration<double, std::milli> ms = stop - start;
+ time += ms.count();
+
+ std::swap(wec.g_prev, wec.g_curr);
+ std::swap(wec.g_curr, wec.g_next);
+
+#ifdef HAVE_SILO
+#ifdef SAVE_TIMESTEPS
+ if ( (iter%100) == 0 )
+ {
+ std::stringstream ss;
+ ss << "wave_mt_" << iter << ".silo";
+ visit_dump(wec.g_curr, ss.str());
+ }
+#endif /* SAVE_TIMESTEPS */
+#endif /* HAVE_SILO */
+ }
+
+ std::cout << "[Wave][MT] Iteration Time (" << nths << " threads): ";
+ std::cout << time/wec.maxiter << "ms" << std::endl;
+ std::cout << "[Wave][MT] Wall Time (" << nths << " threads): ";
+ std::cout << time << "ms" << std::endl;
+
+#ifdef HAVE_SILO
+ visit_dump(wec.g_curr, "wave_mt_lastiter.silo");
+#endif /* HAVE_SILO */
+
+ return time/wec.maxiter;
+}
+
+
diff --git a/kokkos-testing/test_thread_overhead.cpp b/kokkos-testing/test_thread_overhead.cpp
new file mode 100644
index 0000000..291ed57
--- /dev/null
+++ b/kokkos-testing/test_thread_overhead.cpp
@@ -0,0 +1,9 @@
+#include <thread>
+
+int main(int argc, char** argv)
+{
+ for (volatile int i = 0; i < 500000; i++)
+ std::thread([](){}).detach();
+ return 0;
+}
+
--
GitLab