From d965a1d06bf4c4d09b5a6f865e716a16aafd6454 Mon Sep 17 00:00:00 2001
From: Matteo Cicuttin <datafl4sh@toxicnet.eu>
Date: Fri, 27 Mar 2020 19:44:14 +0100
Subject: [PATCH] Updated experimental data.
---
.../AllTimes_double_update.plot | 19 ++
.../AllTimes_double_update.txt | 5 +
.../AllTimes_float_update.plot | 20 +++
.../AllTimes_float_update.txt | 6 +
.../CUDA_gflops_double_update.txt | 6 +
.../i7-3615QM_gflops_double_update.plot | 14 ++
kokkos-testing/fd_catalog/CMakeLists.txt | 24 ++-
kokkos-testing/fd_catalog/fd_main.cpp | 15 +-
kokkos-testing/fd_catalog/fd_wave_cpu.hpp | 168 ++++++++++++++----
kokkos-testing/test_mandel.cpp | 13 +-
10 files changed, 243 insertions(+), 47 deletions(-)
create mode 100644 kokkos-testing/experimental_data/AllTimes_double_update.plot
create mode 100644 kokkos-testing/experimental_data/AllTimes_double_update.txt
create mode 100644 kokkos-testing/experimental_data/AllTimes_float_update.plot
create mode 100644 kokkos-testing/experimental_data/AllTimes_float_update.txt
create mode 100644 kokkos-testing/experimental_data/CUDA_gflops_double_update.txt
create mode 100644 kokkos-testing/experimental_data/i7-3615QM_gflops_double_update.plot
diff --git a/kokkos-testing/experimental_data/AllTimes_double_update.plot b/kokkos-testing/experimental_data/AllTimes_double_update.plot
new file mode 100644
index 0000000..fe76eb7
--- /dev/null
+++ b/kokkos-testing/experimental_data/AllTimes_double_update.plot
@@ -0,0 +1,19 @@
+set term postscript enhanced color
+set output 'AllTimes_double_update.eps'
+
+set key left top
+
+set title '2D damped wave equation - double'
+
+set xlabel 'Mesh size'
+set ylabel 'Iteration time [ms]'
+
+set style data histogram
+set style fill solid border -1
+set logscale y
+set grid ytics mytics
+
+set arrow from 0,0.1 to 3,6.4 nohead lc rgb 'red'
+
+plot 'AllTimes_double_update.txt' using 2:xtic(1) ti col, '' u 3 ti col, \
+ '' u 4 ti col, '' u 5 ti col, '' u 6 ti col
diff --git a/kokkos-testing/experimental_data/AllTimes_double_update.txt b/kokkos-testing/experimental_data/AllTimes_double_update.txt
new file mode 100644
index 0000000..914a7e9
--- /dev/null
+++ b/kokkos-testing/experimental_data/AllTimes_double_update.txt
@@ -0,0 +1,5 @@
+"SIZE" "Seq" "SeqBlk" "1T" "2T" "4T"
+128 0.215397 0.0721867 0.231328 0.146418 0.150125
+256 0.884406 0.284522 0.996983 0.519418 0.420706
+512 3.49626 1.34328 3.55165 1.84903 1.3564
+1024 13.9678 5.97425 14.4531 7.56775 5.25945
diff --git a/kokkos-testing/experimental_data/AllTimes_float_update.plot b/kokkos-testing/experimental_data/AllTimes_float_update.plot
new file mode 100644
index 0000000..9a4b0dc
--- /dev/null
+++ b/kokkos-testing/experimental_data/AllTimes_float_update.plot
@@ -0,0 +1,20 @@
+set term postscript enhanced color
+set output 'AllTimes_float_update.eps'
+
+set key left top
+
+set title '2D damped wave equation - float'
+
+set xlabel 'Mesh size'
+set ylabel 'Iteration time [ms]'
+
+set style data histogram
+set style fill solid border -1
+set logscale y
+set grid ytics mytics
+
+set arrow from 0,0.1 to 3,6.4 nohead lc rgb 'red'
+
+plot 'AllTimes_float_update.txt' using 2:xtic(1) ti col, '' u 3 ti col, \
+ '' u 4 ti col, '' u 5 ti col, '' u 6 ti col
+
diff --git a/kokkos-testing/experimental_data/AllTimes_float_update.txt b/kokkos-testing/experimental_data/AllTimes_float_update.txt
new file mode 100644
index 0000000..77e1555
--- /dev/null
+++ b/kokkos-testing/experimental_data/AllTimes_float_update.txt
@@ -0,0 +1,6 @@
+"SIZE" "Seq" "SeqBlk" "1T" "2T" "4T"
+128 0.285575 0.0328261 0.266828 0.159186 0.1409
+256 1.15701 0.13435 1.07333 0.531615 0.387834
+512 4.65854 0.57531 4.21764 2.03206 1.42824
+1024 13.9774 3.15193 16.5969 8.72612 5.83051
+
diff --git a/kokkos-testing/experimental_data/CUDA_gflops_double_update.txt b/kokkos-testing/experimental_data/CUDA_gflops_double_update.txt
new file mode 100644
index 0000000..e9da9dc
--- /dev/null
+++ b/kokkos-testing/experimental_data/CUDA_gflops_double_update.txt
@@ -0,0 +1,6 @@
+"Size" "GT650M" "GTX760" "i7-3615QM"
+128 13.062 45.9258 13.1641
+256 16.8322 62.7176 13.3595
+512 17.469 67.3603 11.3188
+1024 17.0132 68.6386 10.1799
+
diff --git a/kokkos-testing/experimental_data/i7-3615QM_gflops_double_update.plot b/kokkos-testing/experimental_data/i7-3615QM_gflops_double_update.plot
new file mode 100644
index 0000000..69ac71c
--- /dev/null
+++ b/kokkos-testing/experimental_data/i7-3615QM_gflops_double_update.plot
@@ -0,0 +1,14 @@
+set term postscript enhanced color font ',20'
+set output 'i7-3615QM_gflops_double_update.eps'
+
+set title "i7-3615QM SeqBlk double"
+set logscale x
+set yrange [3:20]
+set xlabel "Mesh size"
+set ylabel "GFlops/s"
+
+set label "Intel declared: 18.4 GFlops/s" at 150,18.8
+set label "DGEMM: 3.83 GFlops/s" at 150,4.2
+set arrow 1 from 128,18.4 to 1024,18.4 nohead dt '.'
+set arrow 2 from 128,3.83 to 1024,3.83 nohead dt '.'
+plot 'CUDA_gflops_double_update.txt' using 1:4:xtic(1) w lp lw 2 ti 'SeqBlk GFlops/s'
diff --git a/kokkos-testing/fd_catalog/CMakeLists.txt b/kokkos-testing/fd_catalog/CMakeLists.txt
index 82e3739..b92eb2e 100644
--- a/kokkos-testing/fd_catalog/CMakeLists.txt
+++ b/kokkos-testing/fd_catalog/CMakeLists.txt
@@ -21,18 +21,22 @@ endif()
find_package(OpenACC)
if (OpenACC_CXX_FOUND)
+ set(OPENACC_LINK_LIBS ${LINK_LIBS})
+ if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+ set(OPENACC_LINK_LIBS ${OPENACC_LINK_LIBS} -lgomp)
+ endif()
if (ENABLE_SINGLE)
add_executable(fd_openacc_single fd_openacc.cpp)
target_compile_definitions(fd_openacc_single PUBLIC -DSINGLE_PRECISION)
target_compile_options(fd_openacc_single PUBLIC ${OpenACC_CXX_FLAGS})
- target_link_libraries(fd_openacc_single ${LINK_LIBS} -lgomp)
+ target_link_libraries(fd_openacc_single ${OPENACC_LINK_LIBS})
endif()
if (ENABLE_DOUBLE)
add_executable(fd_openacc_double fd_openacc.cpp)
target_compile_options(fd_openacc_double PUBLIC ${OpenACC_CXX_FLAGS})
- target_link_libraries(fd_openacc_double ${LINK_LIBS} -lgomp)
+ target_link_libraries(fd_openacc_double ${OPENACC_LINK_LIBS})
endif()
endif()
@@ -87,10 +91,22 @@ if (ENABLE_CUDA)
endif()
endif()
-option(ENABLE_VECTORIZER_REMARKS "Enable Clang vectorizer remarks" ON)
+option(ENABLE_VECTORIZER_REMARKS "Enable vectorizer remarks" ON)
if (ENABLE_VECTORIZER_REMARKS)
if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Rpass=loop-vectorize")
+ 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()
diff --git a/kokkos-testing/fd_catalog/fd_main.cpp b/kokkos-testing/fd_catalog/fd_main.cpp
index e0d50d5..f2d3de2 100644
--- a/kokkos-testing/fd_catalog/fd_main.cpp
+++ b/kokkos-testing/fd_catalog/fd_main.cpp
@@ -29,8 +29,9 @@ 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);
/* Make header */
ofs << "\"SIZE\" \"Seq\" \"SeqBlk\" ";
@@ -51,9 +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);
@@ -67,9 +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(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 06f452c..73f3d75 100644
--- a/kokkos-testing/fd_catalog/fd_wave_cpu.hpp
+++ b/kokkos-testing/fd_catalog/fd_wave_cpu.hpp
@@ -44,6 +44,8 @@ operator<<(std::ostream& os, const tile& tl)
#define BLK_ROWS 56
#define BLK_COLS 56
+//#define KERNEL_BLK_USE_MEMCPY
+#define KERNEL_BLK_USE_TEMP
template<typename T>
void
@@ -51,9 +53,6 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
fd_grid<T>& g_next,
const wave_2D_params<T>& params)
{
- /* 1) memcpy is faster than for
- * 2) if you put inner block processing, you kill perf
- */
int maxrow = params.maxrow;
int maxcol = params.maxcol;
T dt = params.dt;
@@ -71,10 +70,12 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
static const T w4 = -1.0/560.0;
static const T w[9] = { w4, w3, w2, w1, w0, w1, w2, w3, w4 };
+#ifdef KERNEL_BLK_USE_TEMP
T s_curr[BLK_ROWS+2*WAVE_8_HALO_SIZE][BLK_COLS+2*WAVE_8_HALO_SIZE];
+#endif
- T kx2 = c*c * dt*dt * (maxcol-1)*(maxcol-1);
- T ky2 = c*c * dt*dt * (maxrow-1)*(maxrow-1);
+ T kx2 = (c*dt*(maxcol-1)) * (c*dt*(maxcol-1));
+ T ky2 = (c*dt*(maxrow-1)) * (c*dt*(maxrow-1));
T one_minus_adt = (1.0 - a*dt);
T two_minus_adt = (2.0 - a*dt);
@@ -91,19 +92,34 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
{
/* Here all the dimensions are known at compile time, let the
* compiler do its job. */
+#ifdef KERNEL_BLK_USE_TEMP
for (size_t i = 0; i < BLK_ROWS+2*WAVE_8_HALO_SIZE; i++)
{
+#ifdef KERNEL_BLK_USE_MEMCPY
int ofs_x = bj - WAVE_8_HALO_SIZE;
int ofs_y = bi + i - WAVE_8_HALO_SIZE;
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 clang loop vectorize(assume_safety)
+ for (size_t j = 0; j < BLK_COLS+2*WAVE_8_HALO_SIZE; j++)
+ {
+ int ofs_x = bj + j - WAVE_8_HALO_SIZE;
+ int ofs_y = bi + i - WAVE_8_HALO_SIZE;
+ s_curr[i][j] = g_curr(ofs_y, ofs_x);
+ }
+#endif /* KERNEL_BLK_USE_MEMCPY */
}
-
+#endif /* KERNEL_BLK_USE_TEMP */
for (size_t i = 0; i < BLK_ROWS; i++)
{
+ #pragma ivdep
+ //#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < BLK_COLS; j++)
{
T lapl = 0.0;
+#ifdef KERNEL_BLK_USE_TEMP
for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
@@ -111,8 +127,19 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
T val = lapl -
- one_minus_adt * g_prev(bi+i, bj+j) +
- two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+#else
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i, bj+j+k);
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i+k, bj+j);
+
+ T val = lapl -
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * g_curr(bi+i, bj+j);
+#endif /* KERNEL_BLK_USE_TEMP */
assert(j < BLK_COLS);
g_next(bi+i, bj+j) = val;
@@ -124,22 +151,37 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
/* Do upper blocks */
{
size_t bi = blocks_y*BLK_ROWS;
+
for (size_t bj = 0; bj < blocks_x*BLK_COLS; bj += BLK_COLS)
{
- for (size_t i = 0; i < rem_y+WAVE_8_HALO_SIZE; i++)
+#ifdef KERNEL_BLK_USE_TEMP
+ for (size_t i = 0; i < rem_y+2*WAVE_8_HALO_SIZE; i++)
{
-
+#ifdef KERNEL_BLK_USE_MEMCPY
int ofs_x = bj - WAVE_8_HALO_SIZE;
int ofs_y = bi + i - WAVE_8_HALO_SIZE;
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 clang loop vectorize(assume_safety)
+ for (size_t j = 0; j < BLK_COLS+2*WAVE_8_HALO_SIZE; j++)
+ {
+ int ofs_x = bj + j - WAVE_8_HALO_SIZE;
+ int ofs_y = bi + i - WAVE_8_HALO_SIZE;
+ s_curr[i][j] = g_curr(ofs_y, ofs_x);
+ }
+#endif
}
+#endif /* KERNEL_BLK_USE_TEMP */
for (size_t i = 0; i < rem_y; i++)
{
+ #pragma ivdep
+ //#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < BLK_COLS; j++)
{
T lapl = 0.0;
+#ifdef KERNEL_BLK_USE_TEMP
for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
@@ -147,8 +189,19 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
T val = lapl -
- one_minus_adt * g_prev(bi+i, bj+j) +
- two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+#else
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i, bj+j+k);
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i+k, bj+j);
+
+ T val = lapl -
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * g_curr(bi+i, bj+j);
+#endif /* KERNEL_BLK_USE_TEMP */
assert(j < BLK_COLS);
g_next(bi+i, bj+j) = val;
@@ -162,29 +215,54 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
size_t bi = blocks_y*BLK_ROWS;
size_t bj = blocks_x*BLK_COLS;
- for (size_t i = 0; i < rem_y+WAVE_8_HALO_SIZE; i++)
+#ifdef KERNEL_BLK_USE_TEMP
+ for (size_t i = 0; i < rem_y+2*WAVE_8_HALO_SIZE; i++)
{
-
+#ifdef KERNEL_BLK_USE_MEMCPY
int ofs_x = bj - WAVE_8_HALO_SIZE;
int ofs_y = bi + i - WAVE_8_HALO_SIZE;
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));
+ memcpy(s_curr[i], g_curr.data(ofs_y, ofs_x), (rem_x+2*WAVE_8_HALO_SIZE)*sizeof(T));
+#else
+ #pragma clang loop vectorize(assume_safety)
+ for (size_t j = 0; j < rem_x+2*WAVE_8_HALO_SIZE; j++)
+ {
+ int ofs_x = bj + j - WAVE_8_HALO_SIZE;
+ int ofs_y = bi + i - WAVE_8_HALO_SIZE;
+ s_curr[i][j] = g_curr(ofs_y, ofs_x);
+ }
+#endif
}
+#endif /* KERNEL_BLK_USE_TEMP */
for (size_t i = 0; i < rem_y; i++)
{
+ #pragma ivdep
+ //#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < rem_x; j++)
{
T lapl = 0.0;
- for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
- lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
-
- for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
- lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
+#ifdef KERNEL_BLK_USE_TEMP
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
- T val = lapl -
+ T val = lapl -
one_minus_adt * g_prev(bi+i, bj+j) +
two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+#else
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i, bj+j+k);
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i+k, bj+j);
+
+ T val = lapl -
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * g_curr(bi+i, bj+j);
+#endif /* KERNEL_BLK_USE_TEMP */
assert(j < BLK_COLS);
g_next(bi+i, bj+j) = val;
@@ -197,29 +275,54 @@ wave_2D_kernel_blk(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
{
size_t bj = blocks_x*BLK_COLS;
+#ifdef KERNEL_BLK_USE_TEMP
for (size_t i = 0; i < BLK_ROWS+2*WAVE_8_HALO_SIZE; i++)
{
-
+#ifdef KERNEL_BLK_USE_MEMCPY
int ofs_x = bj - WAVE_8_HALO_SIZE;
int ofs_y = bi + i - WAVE_8_HALO_SIZE;
assert(i < BLK_ROWS+2*WAVE_8_HALO_SIZE);
- memcpy(s_curr[i], g_curr.data(ofs_y, ofs_x), (rem_x+WAVE_8_HALO_SIZE)*sizeof(T));
+ memcpy(s_curr[i], g_curr.data(ofs_y, ofs_x), (rem_x+2*WAVE_8_HALO_SIZE)*sizeof(T));
+#else
+ #pragma clang loop vectorize(assume_safety)
+ for (size_t j = 0; j < rem_x+2*WAVE_8_HALO_SIZE; j++)
+ {
+ int ofs_x = bj + j - WAVE_8_HALO_SIZE;
+ int ofs_y = bi + i - WAVE_8_HALO_SIZE;
+ s_curr[i][j] = g_curr(ofs_y, ofs_x);
+ }
+#endif
}
+#endif /* KERNEL_BLK_USE_TEMP */
for (size_t i = 0; i < BLK_ROWS; i++)
{
+ #pragma ivdep
+ //#pragma clang loop vectorize(assume_safety)
for (size_t j = 0; j < rem_x; j++)
{
T lapl = 0.0;
- for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
- lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
-
- for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
- lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
+#ifdef KERNEL_BLK_USE_TEMP
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE+k];
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * s_curr[i+WAVE_8_HALO_SIZE+k][j+WAVE_8_HALO_SIZE];
- T val = lapl -
+ T val = lapl -
one_minus_adt * g_prev(bi+i, bj+j) +
two_minus_adt * s_curr[i+WAVE_8_HALO_SIZE][j+WAVE_8_HALO_SIZE];
+#else
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += kx2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i, bj+j+k);
+
+ for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
+ lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * g_curr(bi+i+k, bj+j);
+
+ T val = lapl -
+ one_minus_adt * g_prev(bi+i, bj+j) +
+ two_minus_adt * g_curr(bi+i, bj+j);
+#endif /* KERNEL_BLK_USE_TEMP */
assert(j < BLK_COLS);
g_next(bi+i, bj+j) = val;
@@ -265,8 +368,8 @@ wave_2D_kernel(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
static const T w4 = -1.0/560.0;
static const T w[9] = { w4, w3, w2, w1, w0, w1, w2, w3, w4 };
- T kx2 = c*c * dt*dt * (maxcol-1)*(maxcol-1);
- T ky2 = c*c * dt*dt * (maxrow-1)*(maxrow-1);
+ T kx2 = (c*dt*(maxcol-1)) * (c*dt*(maxcol-1));
+ T ky2 = (c*dt*(maxrow-1)) * (c*dt*(maxrow-1));
T one_minus_adt = (1.0 - a*dt);
T two_minus_adt = (2.0 - a*dt);
@@ -281,7 +384,8 @@ wave_2D_kernel(const fd_grid<T>& g_prev, const fd_grid<T>& g_curr,
for (int k = -WAVE_8_HALO_SIZE; k <= WAVE_8_HALO_SIZE; k++)
lapl += ky2 * w[k+WAVE_8_HALO_SIZE] * g_curr(i+k,j);
- T val = lapl - one_minus_adt * g_prev(i, j) + two_minus_adt * g_curr(i, j);
+ T val = lapl - one_minus_adt * g_prev(i, j);
+ val += two_minus_adt * g_curr(i, j);
if ( (i == 0) or (j == 0) or (i == maxrow-1) or (j == maxcol-1) )
val = 0;
diff --git a/kokkos-testing/test_mandel.cpp b/kokkos-testing/test_mandel.cpp
index 6365842..832b340 100644
--- a/kokkos-testing/test_mandel.cpp
+++ b/kokkos-testing/test_mandel.cpp
@@ -3,6 +3,9 @@
#include <vector>
#include <chrono>
+#include <pmmintrin.h>
+#include <xmmintrin.h>
+
#define YMIN (-1.5)
#define YMAX (1.5)
#define XMIN (-2.5)
@@ -56,8 +59,6 @@ void gpu_mandel(T *mandel, int maxiter)
mandel[i*SIZE_X + j] = iter;
}
-#endif /* __NVCC__ */
-
template<typename T>
void compute_gpu(std::vector<T>& mandel, int maxiter)
{
@@ -65,10 +66,12 @@ void compute_gpu(std::vector<T>& mandel, int maxiter)
dim3 threads_per_block(WAVE_8_KER_COLS, WAVE_8_KER_ROWS);
}
+#endif /* __NVCC__ */
+
int main(void)
{
size_t maxiter = 1000;
- using T = double;
+ using T = float;
std::vector<T> mandel(SIZE_X*SIZE_Y);
@@ -76,7 +79,9 @@ int main(void)
size_t flops = 0;
- #pragma omp parallel for
+ //_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
+ //_MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
+
for (size_t i = 0; i < SIZE_Y; i++)
{
for (size_t j = 0; j < SIZE_X; j++)
--
GitLab