Skip to content
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

Jump to bottom

Move computeAt logic to separate file/class. #128

Closed
wants to merge 5 commits into from
Closed

Move computeAt logic to separate file/class. #128

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
129e04e
Move computeAt logic to separate file/class.
csarofeen Jun 26, 2020
e11c8c3
Clang tidy.
csarofeen Jun 26, 2020
2556f0d
getDependencyChainsTo changed to getAllUseChains
csarofeen Jun 26, 2020
5e2fe4a
Continue refactoring.
csarofeen Jun 26, 2020
eee5d21
Fix computeAt.
csarofeen Jun 26, 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 caffe2/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -424,6 +424,7 @@ if(NOT INTERN_BUILD_MOBILE OR NOT BUILD_CAFFE2_MOBILE)
${TORCH_SRC_DIR}/csrc/autograd/functions/comm.cpp
${TORCH_SRC_DIR}/csrc/cuda/comm.cpp
${TORCH_SRC_DIR}/csrc/jit/codegen/cuda/arith.cpp
${TORCH_SRC_DIR}/csrc/jit/codegen/cuda/compute_at.cpp
${TORCH_SRC_DIR}/csrc/jit/codegen/cuda/dispatch.cpp
${TORCH_SRC_DIR}/csrc/jit/codegen/cuda/expr_evaluator.cpp
${TORCH_SRC_DIR}/csrc/jit/codegen/cuda/fusion.cpp
Expand Down
28 changes: 25 additions & 3 deletions test/cpp/jit/test_gpu.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1125,7 +1125,7 @@ void testGPU_FusionAdvancedComputeAt() {
TORCH_CHECK(tv0->getComputeAtView() == tv3);
TORCH_CHECK(tv1->getComputeAtView() == tv4);
TORCH_CHECK(tv2->getComputeAtView() == tv4);
TORCH_CHECK(tv3->getComputeAtView() == tv5);
TORCH_CHECK(tv3->getComputeAtView() == tv6);
TORCH_CHECK(tv4->getComputeAtView() == tv5);
TORCH_CHECK(tv5->getComputeAtView() == tv6);
TORCH_CHECK(!tv6->hasComputeAt());
Expand Down Expand Up @@ -2131,7 +2131,8 @@ void testGPU_FusionRFactorReplay() {

// Replay casp, replay new_domain2 as new_domain
// reordered_new_domain[I0oi{16}, I0oo*I0i{32}, ir1oi{4}rf, R(R1oo*R1i{8})rf]
TensorDomain* casp = TransformReplay::replayCasP(new_domain2, new_domain, 2);
auto replay_casp = TransformReplay::replayCasP(new_domain2, new_domain, 2);
TensorDomain* casp = replay_casp.first;
// new_domain[I0oi{16}, I0oo*I0i{32}, ir1oi{4}rf, R(R1oo*R1i{8})rf]
// casp[I0oi{16}, I0oo*I0i{32}, R1oi{4}]

Expand All @@ -2140,7 +2141,8 @@ void testGPU_FusionRFactorReplay() {
// new_domain[I0oi{16}, I0oo*I0i{32} , ir1oi{4}rf,
// R(R1oo*R1i{8})rf]

TensorDomain* pasc = TransformReplay::replayPasC(new_domain, casp, 2);
auto replay_pasc = TransformReplay::replayPasC(new_domain, casp, 2);
TensorDomain* pasc = replay_pasc.first;
// pasc [I0oi{16}, (I0oo*I0i{32})o, I(Ioo*I0i)i{2}, ir1oi{4}rf,
// R(R1oo*R1i{8})rf]

Expand Down Expand Up @@ -3448,6 +3450,26 @@ void testGPU_FusionBCastReduce() {
!tv2->axis(2)->isBroadcast() && !tv2->axis(2)->isReduction());
}

// Multiple consumer reduction with computeAt
// https://github.com/csarofeen/pytorch/issues/110
void testGPU_FusionReductionMultiConsumer() {
torch::jit::fuser::cuda::CudaKernel prog;
Fusion& fusion = *prog.fusion_;
FusionGuard fg(&fusion);
TensorView* tv0 = makeDummyTensor(2);
fusion.addInput(tv0);
auto tv1 = unaryOp(UnaryOpType::Exp, tv0);
auto tv2 = reductionOp(BinaryOpType::Max, {-1}, new Float(0), tv1);
auto tv3 = reductionOp(BinaryOpType::Min, {-1}, new Float(0), tv1);
auto tv4 = add(tv2, tv3);
fusion.addOutput(tv4);
tv1->computeAt(tv2, -1);

TORCH_CHECK(
(tv1->getComputeAtView() == tv2 || tv1->getComputeAtView() == tv3) &&
tv1->getThisComputeAtAxis() == 2 && tv1->getRelativeComputeAtAxis() == 2);
}

} // namespace jit
} // namespace torch
#endif // #if defined(USE_CUDA)
3 changes: 2 additions & 1 deletion test/cpp/jit/tests.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,7 +156,8 @@ namespace jit {
_(GPU_FusionNonRedAxisBind) \
_(GPU_FusionBCastInnerDim) \
_(GPU_FusionBCastReduce) \
_(GPU_FusionSplitBCast)
_(GPU_FusionSplitBCast) \
_(GPU_FusionReductionMultiConsumer)
#else
#define TH_FORALL_TESTS_CUDA(_) \
_(ArgumentSpec) \
Expand Down
1 change: 1 addition & 0 deletions tools/build_variables.bzl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -305,6 +305,7 @@ libtorch_cuda_sources = [
"torch/csrc/autograd/profiler_cuda.cpp",
"torch/csrc/autograd/functions/comm.cpp",
"torch/csrc/jit/codegen/cuda/arith.cpp",
"torch/csrc/jit/codegen/cuda/compute_at.cpp",
"torch/csrc/jit/codegen/cuda/dispatch.cpp",
"torch/csrc/jit/codegen/cuda/expr_evaluator.cpp",
"torch/csrc/jit/codegen/cuda/fusion.cpp",
Expand Down
Loading