Skip to content

Feature/raja vec #83

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

Open
vsrana01 wants to merge 15 commits into
base: develop
Choose a base branch
from
Open

Feature/raja vec #83

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
290fb27
Initial vec feature added in add
May 15, 2020
ef92cd4
temp remove RAJA submodule
May 15, 2020
ab57da5
Attemtp to add RAJAvec lib
May 15, 2020
e38062a
Added DAXPY kernel w/ vec
May 15, 2020
3a18563
Adding RAJA back as submodule
May 21, 2020
1d6bbea
Test push
Nov 16, 2020
c3192c4
Initial kernels add to test vectorization
Nov 16, 2020
7179925
Merged in develop for feature/variant-output
Nov 16, 2020
cdccd30
Added Vectorized Kernels
Dec 7, 2020
793fc36
merge feature/rajaVec2
Dec 7, 2020
0cd73e6
Mod'd ADD-Seq file
Dec 7, 2020
e37904a
Mod'd 2mm
Dec 7, 2020
1a909d0
Polybench kernel
Dec 10, 2020
1ae4806
Merge branch 'feature/rajaVec2' into feature/rajaVec
Dec 10, 2020
b53e74d
Added polybench 2mm
Dec 10, 2020
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
1 change: 1 addition & 0 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@ if (PERFSUITE_ENABLE_WARNINGS)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra -Werror")
endif()

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx2")
Copy link

@ajkunen ajkunen May 21, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This flag should go in the scripts/lc-builds/XXX files. Also I think there is a architecture agnostic flag for (at least gnu and clang) like -march=native or something... in case the machine has SSE, AVX, AVX2 or AVX512, it will pick the best one.

Copy link
Author

@vsrana01 vsrana01 May 21, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I saw the lc-builds file and this is where I have it now. Not sure how this slipped in..but thanks for the feedback!

set(ENABLE_TESTS Off CACHE BOOL "Enable BLT and RAJA tests")
set(ENABLE_EXAMPLES Off CACHE BOOL "Enable RAJA examples")
set(ENABLE_EXERCISES Off CACHE BOOL "Enable RAJA exercises")
Expand Down
1 change: 1 addition & 0 deletions scripts/lc-builds/toss3_icpc19.1.0.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@ module load cmake/3.14.5
cmake \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_CXX_COMPILER=/usr/tce/packages/intel/intel-19.1.0/bin/icpc \
-DCMAKE_CXX_FLAGS="-xCORE-AVX2" \
-C ${RAJA_HOSTCONFIG} \
-DENABLE_OPENMP=On \
-DCMAKE_INSTALL_PREFIX=../install_${BUILD_SUFFIX} \
Expand Down
19 changes: 19 additions & 0 deletions src/basic/DAXPY-Seq.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,25 @@ void DAXPY::runSeqVariant(VariantID vid)

break;
}

case RAJA_Vec : {

startTimer();

for (RepIndex_type irep = 0; irep < run_reps; ++irep) {

RAJA::forall<RAJA::vector_exec<vector_t>>(RAJA::TypedRangeSegment<I>(ibegin, iend),
[=](VecI i)
{
Y[i] += a * X[i];
});

}
stopTimer();

break;
}

#endif

default : {
Expand Down
8 changes: 7 additions & 1 deletion src/basic/DAXPY.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,13 @@
#define DAXPY_DATA_SETUP \
Real_ptr x = m_x; \
Real_ptr y = m_y; \
Real_type a = m_a;
Real_type a = m_a; \
RAJA_INDEX_VALUE_T(I, int, "I");\
using vector_t = RAJA::StreamVector<Real_type, 2>;\
using VecI = RAJA::VectorIndex<I, vector_t>;\
RAJA::TypedView<Real_type, RAJA::Layout<1>, I> X(x, getRunSize()); \
RAJA::TypedView<Real_type, RAJA::Layout<1>, I> Y(y, getRunSize()); \


#define DAXPY_BODY \
y[i] += a * x[i] ;
Expand Down
1 change: 1 addition & 0 deletions src/common/KernelBase.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,6 +88,7 @@ void KernelBase::runKernel(VariantID vid)
#if defined(RUN_RAJA_SEQ)
case Lambda_Seq :
case RAJA_Seq :
case RAJA_Vec:
#endif
{
runSeqVariant(vid);
Expand Down
1 change: 1 addition & 0 deletions src/common/RAJAPerfSuite.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,6 +214,7 @@ static const std::string VariantNames [] =
#if defined(RUN_RAJA_SEQ)
std::string("Lambda_Seq"),
std::string("RAJA_Seq"),
std::string("RAJA_Vec"),
#endif

#if defined(RAJA_ENABLE_OPENMP) && defined(RUN_OPENMP)
Expand Down
1 change: 1 addition & 0 deletions src/common/RAJAPerfSuite.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,6 +185,7 @@ enum VariantID {
#if defined(RUN_RAJA_SEQ)
Lambda_Seq,
RAJA_Seq,
RAJA_Vec,
#endif

#if defined(RAJA_ENABLE_OPENMP) && defined(RUN_OPENMP)
Expand Down
37 changes: 18 additions & 19 deletions src/polybench/POLYBENCH_2MM-OMP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,26 +53,25 @@ void POLYBENCH_2MM::runOpenMPVariant(VariantID vid)

POLYBENCH_2MM_VIEWS_RAJA;

auto poly_2mm_lam1 = [=](Index_type /*i*/, Index_type /*j*/, Index_type /*k*/, Real_type &dot) {
auto poly_2mm_lam1 = [=](Real_type &dot) {
Copy link
Member

@rhornung67 rhornung67 May 21, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@vsrana01 have you checked whether these changes do not adversely affect performance across a range of compilers?

I don't think we want to make changes like this in this PR, since they are orthogonal. We can think about adding additional variants like this later to further stress compilers.

Copy link

@ajkunen ajkunen May 21, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rhornung67 These changes may be needed with the latest Lambda changes: if not all segments are in active loops, the Lambda form will static_assert out

POLYBENCH_2MM_BODY1_RAJA;
};
auto poly_2mm_lam2 = [=](Index_type i, Index_type j, Index_type k,
Real_type &dot) {
POLYBENCH_2MM_BODY2_RAJA;
};
auto poly_2mm_lam3 = [=](Index_type i, Index_type j, Index_type /*k*/,
auto poly_2mm_lam3 = [=](Index_type i, Index_type j,
Real_type &dot) {
POLYBENCH_2MM_BODY3_RAJA;
};
auto poly_2mm_lam4 = [=](Index_type /*i*/, Index_type /*l*/, Index_type /*j*/,
Real_type &dot) {
auto poly_2mm_lam4 = [=](Real_type &dot) {
POLYBENCH_2MM_BODY4_RAJA;
};
auto poly_2mm_lam5 = [=](Index_type i, Index_type l, Index_type j,
Real_type &dot) {
POLYBENCH_2MM_BODY5_RAJA;
};
auto poly_2mm_lam6 = [=](Index_type i, Index_type l, Index_type /*j*/,
auto poly_2mm_lam6 = [=](Index_type i, Index_type l,
Real_type &dot) {
POLYBENCH_2MM_BODY6_RAJA;
};
Expand Down Expand Up @@ -168,23 +167,23 @@ void POLYBENCH_2MM::runOpenMPVariant(VariantID vid)
RAJA::KernelPolicy<
RAJA::statement::Collapse<RAJA::omp_parallel_collapse_exec,
RAJA::ArgList<0, 1>,
RAJA::statement::Lambda<0>,
RAJA::statement::Lambda<0, RAJA::Params<0>>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1>
RAJA::statement::Lambda<1, RAJA::Segs<0,1,2>, RAJA::Params<0>>
>,
RAJA::statement::Lambda<2>
RAJA::statement::Lambda<2, RAJA::Segs<0,1>, RAJA::Params<0>>
>
>;
#else // without collapse...
using EXEC_POL =
RAJA::KernelPolicy<
RAJA::statement::For<0, RAJA::omp_parallel_for_exec,
RAJA::statement::For<1, RAJA::loop_exec,
RAJA::statement::Lambda<0>,
RAJA::statement::Lambda<0, RAJA::Params<0>>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1>
RAJA::statement::Lambda<1, RAJA::Segs<0,1,2>, RAJA::Params<0>>
>,
RAJA::statement::Lambda<2>
RAJA::statement::Lambda<2, RAJA::Segs<0,1>, RAJA::Params<0>>
>
>
>;
Expand All @@ -194,21 +193,21 @@ void POLYBENCH_2MM::runOpenMPVariant(VariantID vid)
for (RepIndex_type irep = 0; irep < run_reps; ++irep) {

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nj},
RAJA::RangeSegment{0, nk}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nj),
RAJA::RangeSegment(0, nk)),
RAJA::tuple<Real_type> {0.0},

poly_2mm_lam1,
poly_2mm_lam2,
poly_2mm_lam3
);

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nl},
RAJA::RangeSegment{0, nj}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nl),
RAJA::RangeSegment(0, nj)),
RAJA::tuple<Real_type> {0.0},

poly_2mm_lam4,
poly_2mm_lam5,
Expand Down
47 changes: 22 additions & 25 deletions src/polybench/POLYBENCH_2MM-Seq.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,27 +43,24 @@ void POLYBENCH_2MM::runSeqVariant(VariantID vid)

POLYBENCH_2MM_VIEWS_RAJA;

auto poly_2mm_lam1 = [=](Index_type /*i*/, Index_type /*j*/, Index_type /*k*/, Real_type &dot) {
auto poly_2mm_lam1 = [=](Real_type &dot) {
Copy link
Member

@rhornung67 rhornung67 May 21, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same comment as previous one.

POLYBENCH_2MM_BODY1_RAJA;
};
auto poly_2mm_lam2 = [=](Index_type i, Index_type j, Index_type k,
Real_type &dot) {
POLYBENCH_2MM_BODY2_RAJA;
};
auto poly_2mm_lam3 = [=](Index_type i, Index_type j, Index_type /*k*/,
Real_type &dot) {
auto poly_2mm_lam3 = [=](Index_type i, Index_type j, Real_type &dot) {
POLYBENCH_2MM_BODY3_RAJA;
};
auto poly_2mm_lam4 = [=](Index_type /*i*/, Index_type /*l*/, Index_type /*j*/,
Real_type &dot) {
auto poly_2mm_lam4 = [=](Real_type &dot) {
POLYBENCH_2MM_BODY4_RAJA;
};
auto poly_2mm_lam5 = [=](Index_type i, Index_type l, Index_type j,
Real_type &dot) {
POLYBENCH_2MM_BODY5_RAJA;
};
auto poly_2mm_lam6 = [=](Index_type i, Index_type l, Index_type /*j*/,
Real_type &dot) {
auto poly_2mm_lam6 = [=](Index_type i, Index_type l, Real_type &dot) {
POLYBENCH_2MM_BODY6_RAJA;
};

Expand Down Expand Up @@ -137,36 +134,36 @@ void POLYBENCH_2MM::runSeqVariant(VariantID vid)

using EXEC_POL =
RAJA::KernelPolicy<
RAJA::statement::For<0, RAJA::loop_exec,
RAJA::statement::For<1, RAJA::loop_exec,
RAJA::statement::Lambda<0>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1>
>,
RAJA::statement::Lambda<2>
RAJA::statement::For<0, RAJA::loop_exec,
RAJA::statement::For<1, RAJA::loop_exec,
RAJA::statement::Lambda<0, RAJA::Params<0>>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1, RAJA::Segs<0,1,2>, RAJA::Params<0>>,
RAJA::statement::Lambda<2, RAJA::Segs<0,1>, RAJA::Params<0>>
>
>
>
>;

startTimer();
for (RepIndex_type irep = 0; irep < run_reps; ++irep) {

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nj},
RAJA::RangeSegment{0, nk}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nj),
RAJA::RangeSegment(0, nk)),
RAJA::tuple<Real_type> {0.0},

poly_2mm_lam1,
poly_2mm_lam2,
poly_2mm_lam1,
poly_2mm_lam2,
poly_2mm_lam3
);
);

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nl},
RAJA::RangeSegment{0, nj}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nl),
RAJA::RangeSegment(0, nj)),
RAJA::tuple<Real_type> {0.0},

poly_2mm_lam4,
poly_2mm_lam5,
Expand Down
52 changes: 24 additions & 28 deletions src/polybench/POLYBENCH_3MM-OMP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,41 +62,37 @@ void POLYBENCH_3MM::runOpenMPVariant(VariantID vid)

POLYBENCH_3MM_VIEWS_RAJA;

auto poly_3mm_lam1 = [=] (Index_type /*i*/, Index_type /*j*/, Index_type /*k*/,
Real_type &dot) {
auto poly_3mm_lam1 = [=] ( Real_type &dot) {
POLYBENCH_3MM_BODY1_RAJA;
};
auto poly_3mm_lam2 = [=] (Index_type i, Index_type j, Index_type k,
Real_type &dot) {
POLYBENCH_3MM_BODY2_RAJA;
};
auto poly_3mm_lam3 = [=] (Index_type i, Index_type j, Index_type /*k*/,
auto poly_3mm_lam3 = [=] (Index_type i, Index_type j,
Real_type &dot) {
POLYBENCH_3MM_BODY3_RAJA;
};

auto poly_3mm_lam4 = [=] (Index_type /*j*/, Index_type /*l*/, Index_type /*m*/,
Real_type &dot) {
auto poly_3mm_lam4 = [=] (Real_type &dot) {
POLYBENCH_3MM_BODY4_RAJA;
};
auto poly_3mm_lam5 = [=] (Index_type j, Index_type l, Index_type m,
Real_type &dot) {
POLYBENCH_3MM_BODY5_RAJA;
};
auto poly_3mm_lam6 = [=] (Index_type j, Index_type l, Index_type /*m*/,
auto poly_3mm_lam6 = [=] (Index_type j, Index_type l,
Real_type &dot) {
POLYBENCH_3MM_BODY6_RAJA;
};
auto poly_3mm_lam7 = [=] (Index_type /*i*/, Index_type /*l*/, Index_type /*j*/,
Real_type &dot) {
auto poly_3mm_lam7 = [=] (Real_type &dot) {
POLYBENCH_3MM_BODY7_RAJA;
};
auto poly_3mm_lam8 = [=] (Index_type i, Index_type l, Index_type j,
Real_type &dot) {
POLYBENCH_3MM_BODY8_RAJA;
};
auto poly_3mm_lam9 = [=] (Index_type i, Index_type l, Index_type /*j*/,
Real_type &dot) {
auto poly_3mm_lam9 = [=] (Index_type i, Index_type l, Real_type &dot) {
POLYBENCH_3MM_BODY9_RAJA;
};

Expand Down Expand Up @@ -221,23 +217,23 @@ void POLYBENCH_3MM::runOpenMPVariant(VariantID vid)
RAJA::KernelPolicy<
RAJA::statement::Collapse<RAJA::omp_parallel_collapse_exec,
RAJA::ArgList<0, 1>,
RAJA::statement::Lambda<0>,
RAJA::statement::Lambda<0, RAJA::Params<0>>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1>
RAJA::statement::Lambda<1, RAJA::Segs<0,1,2>, RAJA::Params<0>>
>,
RAJA::statement::Lambda<2>
RAJA::statement::Lambda<2, RAJA::Segs<0,1>, RAJA::Params<0>>
>
>;
#else
using EXEC_POL =
RAJA::KernelPolicy<
RAJA::statement::For<0, RAJA::omp_parallel_for_exec,
RAJA::statement::For<1, RAJA::loop_exec,
RAJA::statement::Lambda<0>,
RAJA::statement::Lambda<0, RAJA::Params<0>>,
RAJA::statement::For<2, RAJA::loop_exec,
RAJA::statement::Lambda<1>
RAJA::statement::Lambda<1, RAJA::Segs<0,1,2>, RAJA::Params<0>>
>,
RAJA::statement::Lambda<2>
RAJA::statement::Lambda<2, RAJA::Segs<0,1>, RAJA::Params<0>>
>
>
>;
Expand All @@ -247,10 +243,10 @@ void POLYBENCH_3MM::runOpenMPVariant(VariantID vid)
for (RepIndex_type irep = 0; irep < run_reps; ++irep) {

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nj},
RAJA::RangeSegment{0, nk}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nj),
RAJA::RangeSegment(0, nk)),
RAJA::tuple<Real_type> {0.0},

poly_3mm_lam1,
poly_3mm_lam2,
Expand All @@ -259,10 +255,10 @@ void POLYBENCH_3MM::runOpenMPVariant(VariantID vid)
);

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, nj},
RAJA::RangeSegment{0, nl},
RAJA::RangeSegment{0, nm}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, nj),
RAJA::RangeSegment(0, nl),
RAJA::RangeSegment(0, nm)),
RAJA::tuple<Real_type> {0.0},

poly_3mm_lam4,
poly_3mm_lam5,
Expand All @@ -271,10 +267,10 @@ void POLYBENCH_3MM::runOpenMPVariant(VariantID vid)
);

RAJA::kernel_param<EXEC_POL>(
RAJA::make_tuple(RAJA::RangeSegment{0, ni},
RAJA::RangeSegment{0, nl},
RAJA::RangeSegment{0, nj}),
RAJA::make_tuple(static_cast<Real_type>(0.0)),
RAJA::make_tuple(RAJA::RangeSegment(0, ni),
RAJA::RangeSegment(0, nl),
RAJA::RangeSegment(0, nj)),
RAJA::tuple<Real_type>{0.0},

poly_3mm_lam7,
poly_3mm_lam8,
Expand Down
Loading