gallo/atrip
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Add the slicing into the GPU

Browse Source
This commit is contained in:
Alejandro Gallo 2022-12-05 21:16:30 +01:00
parent 658397ebd7
commit ad542fe856
2 changed files with 71 additions and 55 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
.dir-locals.el
View File

@ -13,7 +13,9 @@
(format "%s/include/" root) (format "%s/include/" root)
(format "%s/" root) (format "%s/" root)
(format "%s/bench/" root) (format "%s/bench/" root)
(format "%s/build/main/" root))))) (format "%s/build/main/" root)))
(setq-local flycheck-clang-include-path
flycheck-gcc-include-path)))
(eval . (flycheck-mode)) (eval . (flycheck-mode))
(eval . (outline-minor-mode)) (eval . (outline-minor-mode))
(indent-tabs-mode . nil) (indent-tabs-mode . nil)

122
include/atrip/Unions.hpp
View File

@ -19,8 +19,14 @@
namespace atrip { namespace atrip {
template <typename F=double> template <typename F=double>
static
void sliceIntoVector void sliceIntoVector
( std::vector<F> &v #if defined(ATRIP_SOURCES_IN_GPU)
( DataPtr<F> &source
#else
( std::vector<F> &source
#endif
, size_t sliceSize
, CTF::Tensor<F> &toSlice , CTF::Tensor<F> &toSlice
, std::vector<int64_t> const low , std::vector<int64_t> const low
, std::vector<int64_t> const up , std::vector<int64_t> const up
@ -44,18 +50,30 @@ namespace atrip {
<< "\n"; << "\n";
#ifndef ATRIP_DONT_SLICE #ifndef ATRIP_DONT_SLICE
toSlice.slice( toSlice_.low.data() toSlice.slice(toSlice_.low.data(),
, toSlice_.up.data() toSlice_.up.data(),
, 0.0 0.0,
, origin origin,
, origin_.low.data() origin_.low.data(),
, origin_.up.data() origin_.up.data(),
, 1.0); 1.0);
memcpy(v.data(), toSlice.data, sizeof(F) * v.size());
#else #else
# pragma message("WARNING: COMPILING WITHOUT SLICING THE TENSORS") # pragma message("WARNING: COMPILING WITHOUT SLICING THE TENSORS")
#endif #endif
#if defined(ATRIP_SOURCES_IN_GPU)
WITH_CHRONO("cuda:sources",
_CHECK_CUDA_SUCCESS("copying sources data to device",
cuMemcpyHtoD(source,
toSlice.data,
sliceSize));
)
#else
memcpy(source.data(),
toSlice.data,
sizeof(F) * sliceSize);
#endif
} }
@ -80,16 +98,15 @@ namespace atrip {
void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override
{ {
const int Nv = this->sliceLength[0]
, No = this->sliceLength[1]
, a = this->rankMap.find({static_cast<size_t>(Atrip::rank), it});
;
const int
Nv = this->sliceLength[0],
No = this->sliceLength[1],
a = this->rankMap.find({static_cast<size_t>(Atrip::rank), it});
sliceIntoVector<F>( this->sources[it] sliceIntoVector<F>(this->sources[it], this->sliceSize,
, to, {0, 0, 0}, {Nv, No, No} to, {0, 0, 0}, {Nv, No, No},
, from, {a, 0, 0, 0}, {a+1, Nv, No, No} from, {a, 0, 0, 0}, {a+1, Nv, No, No});
);
} }
@ -118,14 +135,13 @@ namespace atrip {
void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override
{ {
const int No = this->sliceLength[0] const int
, a = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}) No = this->sliceLength[0],
; a = this->rankMap.find({static_cast<size_t>(Atrip::rank), it});
sliceIntoVector<F>( this->sources[it] sliceIntoVector<F>(this->sources[it], this->sliceSize,
, to, {0, 0, 0}, {No, No, No} to, {0, 0, 0}, {No, No, No},
, from, {0, 0, 0, a}, {No, No, No, a+1} from, {0, 0, 0, a}, {No, No, No, a+1});
);
} }
}; };
@ -153,18 +169,17 @@ namespace atrip {
void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override { void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override {
const int Nv = this->sliceLength[0] const int
, No = this->sliceLength[1] Nv = this->sliceLength[0],
, el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}) No = this->sliceLength[1],
, a = el % Nv el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}),
, b = el / Nv a = el % Nv,
; b = el / Nv;
sliceIntoVector<F>( this->sources[it] sliceIntoVector<F>(this->sources[it], this->sliceSize,
, to, {0, 0}, {Nv, No} to, {0, 0}, {Nv, No},
, from, {a, b, 0, 0}, {a+1, b+1, Nv, No} from, {a, b, 0, 0}, {a+1, b+1, Nv, No});
);
} }
@ -191,17 +206,17 @@ namespace atrip {
void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override { void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override {
const int Nv = from.lens[0] const int
, No = this->sliceLength[1] Nv = from.lens[0],
, el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}) No = this->sliceLength[1],
, a = el % Nv el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}),
, b = el / Nv a = el % Nv,
; b = el / Nv;
sliceIntoVector<F>( this->sources[it]
, to, {0, 0}, {No, No} sliceIntoVector<F>(this->sources[it], this->sliceSize,
, from, {a, b, 0, 0}, {a+1, b+1, No, No} to, {0, 0}, {No, No},
); from, {a, b, 0, 0}, {a+1, b+1, No, No});
} }
@ -231,17 +246,16 @@ namespace atrip {
void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override { void sliceIntoBuffer(size_t it, CTF::Tensor<F> &to, CTF::Tensor<F> const& from) override {
// TODO: maybe generalize this with ABHH // TODO: maybe generalize this with ABHH
const int Nv = from.lens[0] const int
, No = this->sliceLength[1] Nv = from.lens[0],
, el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}) No = this->sliceLength[1],
, a = el % Nv el = this->rankMap.find({static_cast<size_t>(Atrip::rank), it}),
, b = el / Nv a = el % Nv,
; b = el / Nv;
sliceIntoVector<F>( this->sources[it] sliceIntoVector<F>(this->sources[it], this->sliceSize,
, to, {0, 0}, {No, No} to, {0, 0}, {No, No},
, from, {a, b, 0, 0}, {a+1, b+1, No, No} from, {a, b, 0, 0}, {a+1, b+1, No, No});
);
} }