Skip to content

OMP threads #300

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
raver119 wants to merge 18 commits into
base: master
Choose a base branch
from
Open

OMP threads #300

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
f48ddfa
omp threads backported
raver119 Mar 6, 2020
8df8bc8
omp scalar reduce
raver119 Mar 6, 2020
d71c6d5
timing
raver119 Mar 6, 2020
5a2d4f4
timing
raver119 Mar 6, 2020
c78067c
minor tweaks
raver119 Mar 7, 2020
bcc8401
minor tweaks
raver119 Mar 7, 2020
aabe9d5
namespace change
raver119 Mar 7, 2020
53d8688
num_threads
raver119 Mar 8, 2020
c1a6a4b
one minor fix
raver119 Mar 8, 2020
c9b057a
Merge branch 'master' into r119_omp_threads
raver119 Mar 9, 2020
e617a6d
Merge branch 'master' into r119_omp_threads
raver119 Mar 9, 2020
2ce5899
Merge remote-tracking branch 'origin/r119_omp_threads' into r119_omp_…
raver119 Mar 9, 2020
9b95cf6
Merge branch 'master' into r119_omp_threads
raver119 Mar 10, 2020
2ad10d1
reduce float tweaks
raver119 Mar 10, 2020
42aa7e0
minor tweak
raver119 Mar 10, 2020
19b735f
minor bugfix
raver119 Mar 16, 2020
5ee1efe
Merge branch 'master' into r119_omp_threads
raver119 Mar 16, 2020
9a5e92e
avgpool3d/maxpool3d + bp parallelism fixed
raver119 Mar 16, 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
4 changes: 2 additions & 2 deletions libnd4j/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,9 @@ if (SD_CUDA)
enable_language(CUDA)
set(CMAKE_CUDA_STANDARD 11)

set(DEFAULT_ENGINE "samediff::ENGINE_CUDA")
set(DEFAULT_ENGINE "sd::ENGINE_CUDA")
else()
set(DEFAULT_ENGINE "samediff::ENGINE_CPU")
set(DEFAULT_ENGINE "sd::ENGINE_CPU")
endif()

# MSVC runtime lib can be either "MultiThreaded" or "MultiThreadedDLL", /MT and /MD respectively
Expand Down
8 changes: 4 additions & 4 deletions libnd4j/include/array/DataTypeConversions.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,7 +56,7 @@ namespace sd {
buffer[e] = canKeep ? static_cast<T>(tmp[e]) : BitwiseUtils::swap_bytes<T>(static_cast<T>(tmp[e]));
};

samediff::Threads::parallel_for(func, 0, length);
sd::Threads::parallel_for(func, 0, length);
#endif

delete[] tmp;
Expand Down Expand Up @@ -114,7 +114,7 @@ namespace sd {
buffer[e] = canKeep ? static_cast<T>(tmp[e]) : BitwiseUtils::swap_bytes<T>(static_cast<T>(tmp[e]));
};

samediff::Threads::parallel_for(func, 0, length);
sd::Threads::parallel_for(func, 0, length);
#endif

delete[] tmp;
Expand Down Expand Up @@ -142,7 +142,7 @@ namespace sd {
buffer[e] = canKeep ? static_cast<T>(tmp[e]) : BitwiseUtils::swap_bytes<T>(static_cast<T>(tmp[e]));
};

samediff::Threads::parallel_for(func, 0, length);
sd::Threads::parallel_for(func, 0, length);
#endif
delete[] tmp;
}
Expand All @@ -168,7 +168,7 @@ namespace sd {
buffer[e] = canKeep ? static_cast<T>(tmp[e]) : BitwiseUtils::swap_bytes<T>(static_cast<T>(tmp[e]));
};

samediff::Threads::parallel_for(func, 0, length);
sd::Threads::parallel_for(func, 0, length);
#endif
delete[] tmp;
}
Expand Down
20 changes: 10 additions & 10 deletions libnd4j/include/array/NDArray.hXX
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -514,7 +514,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<const cha
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

tickWriteHost();
syncToDevice();
Expand Down Expand Up @@ -581,7 +581,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<std::stri
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);


tickWriteHost();
Expand Down Expand Up @@ -647,7 +647,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<std::u16s
}
}
};
samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

tickWriteHost();
syncToDevice();
Expand Down Expand Up @@ -713,7 +713,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<const cha
}
}
};
samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

tickWriteHost();
syncToDevice();
Expand Down Expand Up @@ -779,7 +779,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<std::u32s
}
}
};
samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

tickWriteHost();
syncToDevice();
Expand Down Expand Up @@ -845,7 +845,7 @@ NDArray::NDArray(const std::vector<Nd4jLong>& shape, const std::vector<const cha
}
}
};
samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

tickWriteHost();
syncToDevice();
Expand Down Expand Up @@ -2392,7 +2392,7 @@ NDArray NDArray::asS() const {
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

registerPrimaryUse({ &res }, { this });

Expand Down Expand Up @@ -3490,7 +3490,7 @@ NDArray NDArray::dup(const char newOrder) const {
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

return NDArray(getShapeAsVector(), strings, dataType(), getContext());
}
Expand All @@ -3503,7 +3503,7 @@ NDArray NDArray::dup(const char newOrder) const {
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

return NDArray(getShapeAsVector(), strings, dataType(), getContext());
}
Expand All @@ -3515,7 +3515,7 @@ NDArray NDArray::dup(const char newOrder) const {
}
};

samediff::Threads::parallel_for(func, 0, lengthOf(), 1);
sd::Threads::parallel_for(func, 0, lengthOf(), 1);

return NDArray(getShapeAsVector(), strings, dataType(), getContext());
}
Expand Down
10 changes: 5 additions & 5 deletions libnd4j/include/array/cpu/NDArray.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ void NDArray::fillAsTriangular(const float val, int lower, int upper, NDArray& t
}
};

samediff::Threads::parallel_for(func, 0, zLen);
sd::Threads::parallel_for(func, 0, zLen);
}
BUILD_SINGLE_TEMPLATE(template void NDArray::fillAsTriangular, (const float val, int lower, int upper, NDArray& target, const char direction), LIBND4J_TYPES);

Expand Down Expand Up @@ -166,7 +166,7 @@ static void templatedSwap(void *xBuffer, void *yBuffer, Nd4jLong length) {
}
};

samediff::Threads::parallel_for(func, 0, length);
sd::Threads::parallel_for(func, 0, length);
}
BUILD_SINGLE_TEMPLATE(template void templatedSwap, (void *xBuffer, void *yBuffer, Nd4jLong length), LIBND4J_TYPES);

Expand Down Expand Up @@ -279,7 +279,7 @@ NDArray NDArray::tile(const std::vector<Nd4jLong>& reps) const {
}
};

samediff::Threads::parallel_for(func, 0, resultLen);
sd::Threads::parallel_for(func, 0, resultLen);
}
else {

Expand All @@ -291,7 +291,7 @@ NDArray NDArray::tile(const std::vector<Nd4jLong>& reps) const {
}
};

samediff::Threads::parallel_for(func, 0, resultLen);
sd::Threads::parallel_for(func, 0, resultLen);
}
result.tickWriteHost();
return result;
Expand Down Expand Up @@ -410,7 +410,7 @@ static void repeat_(const NDArray& input, NDArray& output, const std::vector<int
}
};

samediff::Threads::parallel_for(func, 0, zLen);
sd::Threads::parallel_for(func, 0, zLen);
}

//////////////////////////////////////////////////////////////////////////
Expand Down
30 changes: 15 additions & 15 deletions libnd4j/include/array/cpu/NDArrayLambda.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ void NDArray::applyTriplewiseLambda(NDArray& second, NDArray& third, const std::
z[e] = func(f[e], s[e], t[e]);
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {
if (f == z) {

Expand All @@ -40,7 +40,7 @@ void NDArray::applyTriplewiseLambda(NDArray& second, NDArray& third, const std::
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {

auto loop = PRAGMA_THREADS_FOR {
Expand All @@ -54,7 +54,7 @@ void NDArray::applyTriplewiseLambda(NDArray& second, NDArray& third, const std::
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
}
}
}
Expand Down Expand Up @@ -97,7 +97,7 @@ void NDArray::applyPairwiseLambda(const NDArray& other, const std::function<T(T,
z[e] = func(f[e], s[e]);
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {
if (f == z) {

Expand All @@ -110,7 +110,7 @@ void NDArray::applyPairwiseLambda(const NDArray& other, const std::function<T(T,
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {

auto loop = PRAGMA_THREADS_FOR {
Expand All @@ -123,7 +123,7 @@ void NDArray::applyPairwiseLambda(const NDArray& other, const std::function<T(T,
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
}
}
}
Expand Down Expand Up @@ -160,7 +160,7 @@ void NDArray::applyLambda(const std::function<T(T)>& func, NDArray& target) {
z[e] = func(f[e]);
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {
if (f == z) {

Expand All @@ -172,7 +172,7 @@ void NDArray::applyLambda(const std::function<T(T)>& func, NDArray& target) {
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {

auto loop = PRAGMA_THREADS_FOR {
Expand All @@ -184,7 +184,7 @@ void NDArray::applyLambda(const std::function<T(T)>& func, NDArray& target) {
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
}
}
}
Expand Down Expand Up @@ -221,7 +221,7 @@ void NDArray::applyIndexedLambda(const std::function<T(Nd4jLong, T)>& func, NDAr
z[e] = func(e, f[e]);
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {
if (f == z) {

Expand All @@ -233,7 +233,7 @@ void NDArray::applyIndexedLambda(const std::function<T(Nd4jLong, T)>& func, NDAr
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {

auto loop = PRAGMA_THREADS_FOR {
Expand All @@ -245,7 +245,7 @@ void NDArray::applyIndexedLambda(const std::function<T(Nd4jLong, T)>& func, NDAr
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
}
}
}
Expand Down Expand Up @@ -287,7 +287,7 @@ void NDArray::applyIndexedPairwiseLambda(NDArray& other, const std::function<T(N
z[e] = func((Nd4jLong) e, f[e], s[e]);
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {
if (f == z) {

Expand All @@ -300,7 +300,7 @@ void NDArray::applyIndexedPairwiseLambda(NDArray& other, const std::function<T(N
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
} else {

auto loop = PRAGMA_THREADS_FOR {
Expand All @@ -313,7 +313,7 @@ void NDArray::applyIndexedPairwiseLambda(NDArray& other, const std::function<T(N
}
};

samediff::Threads::parallel_for(loop, 0, _length);
sd::Threads::parallel_for(loop, 0, _length);
}
}
}
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/BlockingQueue.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
#include <atomic>
#include <condition_variable>

namespace samediff {
namespace sd {
template <typename T>
class BlockingQueue {
private:
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/CallableInterface.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@
#include <mutex>
#include <condition_variable>

namespace samediff {
namespace sd {
/**
* This class is suited for passing functions to execution threads without queues
*/
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/CallableWithArguments.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
#include <condition_variable>
#include <system/op_boilerplate.h>

namespace samediff {
namespace sd {
class CallableWithArguments {
FUNC_DO _function_do;
FUNC_1D _function_1d;
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/Engine.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@
#ifndef SD_ENGINE_H
#define SD_ENGINE_H

namespace samediff {
namespace sd {
enum Engine {
ENGINE_CPU = 0,
ENGINE_CUDA = 1,
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/ExecutionMode.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@
#ifndef SD_EXECUTIONMODE_H
#define SD_EXECUTIONMODE_H

namespace samediff {
namespace sd {
enum ExecutionMode {
MODE_UNDEFINED = 0,
MODE_TRAINING = 1,
Expand Down
2 changes: 1 addition & 1 deletion libnd4j/include/execution/ThreadPool.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@
#include <execution/Ticket.h>
#include <queue>

namespace samediff {
namespace sd {
class ND4J_EXPORT ThreadPool {
private:
static ThreadPool* _INSTANCE;
Expand Down
Loading