Skip to content

Fixing failures in debug run: #247

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

Merged
merged 2 commits into from
Jul 30, 2020
Merged

Fixing failures in debug run: #247

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a722ac0
fixing un-initialized data; illegal usage of c10::optional
jjsjann123 Jul 30, 2020
d8e958b
clang-format
jjsjann123 Jul 30, 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
31 changes: 19 additions & 12 deletions torch/csrc/jit/codegen/cuda/manager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,6 +147,7 @@ class CudaFusionManager {
}

void debugPrint(const TensorTypePtr& type) {
printf("\nsizes:");
if (auto sizes = type->symbolic_sizes().sizes()) {
// for (const auto& shape_symbol : sizes.value()) {
int rank = static_cast<int>(sizes->size());
Expand All @@ -165,24 +166,27 @@ class CudaFusionManager {
int rank = static_cast<int>(stride_properties->size());
printf("\nstride: ");
for (int i = 0; i < rank; i++) {
if (auto val = (*stride_properties)[i]->stride_) {
printf("%ld, ", val.value());
if ((*stride_properties)[i].has_value() &&
(*stride_properties)[i]->stride_.has_value()) {
printf("%ld, ", (*stride_properties)[i]->stride_.value());
} else {
printf("?, ");
}
}
printf("\nstride index: ");
for (int i = 0; i < rank; i++) {
if (auto val = (*stride_properties)[i]->stride_index_) {
printf("%ld, ", val.value());
if ((*stride_properties)[i].has_value() &&
(*stride_properties)[i]->stride_index_.has_value()) {
printf("%ld, ", (*stride_properties)[i]->stride_index_.value());
} else {
printf("?, ");
}
}
printf("\ncontiguous: ");
for (int i = 0; i < rank; i++) {
if (auto val = (*stride_properties)[i]->contiguous_) {
printf("%d, ", val.value());
if ((*stride_properties)[i].has_value() &&
(*stride_properties)[i]->contiguous_.has_value()) {
printf("%d, ", (*stride_properties)[i]->contiguous_.value());
} else {
printf("?, ");
}
Expand All @@ -196,7 +200,7 @@ class CudaFusionManager {
at::DimVector restorePermutation(at::DimVector permuted) {
int rank = static_cast<int>(permuted.size());
at::DimVector permutation(rank, -1);
for (int i; i < rank; i++) {
for (int i = 0; i < rank; i++) {
permutation[permuted[i]] = i;
}
return permutation;
Expand All @@ -220,17 +224,19 @@ class CudaFusionManager {

// TODO: this does not support broadcast yet;
for (int i = 0; i < rank; i++) {
if (auto index = (*stride_properties)[i]->stride_index_) {
ordered_axes.insert(*index);
if ((*stride_properties)[i].has_value() &&
(*stride_properties)[i]->stride_index_.has_value()) {
ordered_axes.insert((*stride_properties)[i]->stride_index_.value());
}
}

int unallocated_axis = 0;
// we push from slowest to fastest
for (int i = rank - 1; i >= 0; i--) {
if (auto index = (*stride_properties)[i]->stride_index_) {
// pushing axis index to current entry in permute_seq;
permute_seq.emplace_back(*index);
if ((*stride_properties)[i].has_value() &&
(*stride_properties)[i]->stride_index_.has_value()) {
permute_seq.emplace_back(
(*stride_properties)[i]->stride_index_.value());
} else {
// no designated axis for this slot, so we push an axis w/o designated
// order;
Expand Down Expand Up @@ -282,6 +288,7 @@ class CudaFusionManager {

auto strategy = getSortStrideScheme(acc_type);
// TODO: early return if permutation is no-op;

auto restore_strategy = restorePermutation(strategy);

std::vector<at::Tensor> permuted_outputs;
Expand Down