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

JSON tokenizer memory optimizations #16978

Merged
merged 9 commits into from
Oct 23, 2024
Merged

JSON tokenizer memory optimizations #16978

Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
faf8b91
add nvtx resource adapter and to json benchmark
karthikeyann Oct 1, 2024
a0c8818
add CUDF_FUNC_RANGE
karthikeyann Oct 1, 2024
14a7aa2
add benchmark and options
karthikeyann Oct 1, 2024
e30b79d
easy opts
shrshi Oct 2, 2024
e9172e0
cleanup
shrshi Oct 5, 2024
d2d0711
Merge branch 'branch-24.12' into json-pda-memory
shrshi Oct 5, 2024
b9c3971
more cleanup
shrshi Oct 5, 2024
24d39cb
adding comments
shrshi Oct 5, 2024
3a9d88c
Merge branch 'branch-24.12' into json-pda-memory
shrshi Oct 22, 2024
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
43 changes: 31 additions & 12 deletions cpp/src/io/json/nested_json_gpu.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1448,10 +1448,6 @@ void get_stack_context(device_span<SymbolT const> json_in,
// Number of stack operations in the input (i.e., number of '{', '}', '[', ']' outside of quotes)
cudf::detail::device_scalar<SymbolOffsetT> d_num_stack_ops(stream);

// Sequence of stack symbols and their position in the original input (sparse representation)
rmm::device_uvector<StackSymbolT> stack_ops{json_in.size(), stream};
rmm::device_uvector<SymbolOffsetT> stack_op_indices{json_in.size(), stream};

// Prepare finite-state transducer that only selects '{', '}', '[', ']' outside of quotes
constexpr auto max_translation_table_size =
to_stack_op::NUM_SYMBOL_GROUPS * to_stack_op::TT_NUM_STATES;
Expand All @@ -1468,11 +1464,26 @@ void get_stack_context(device_span<SymbolT const> json_in,

// "Search" for relevant occurrence of brackets and braces that indicate the beginning/end
// of structs/lists
// Run FST to estimate the sizes of translated buffers
json_to_stack_ops_fst.Transduce(json_in.begin(),
static_cast<SymbolOffsetT>(json_in.size()),
thrust::make_discard_iterator(),
thrust::make_discard_iterator(),
d_num_stack_ops.data(),
to_stack_op::start_state,
stream);

auto stack_ops_bufsize = d_num_stack_ops.value(stream);
// Sequence of stack symbols and their position in the original input (sparse representation)
rmm::device_uvector<StackSymbolT> stack_ops{stack_ops_bufsize, stream};
rmm::device_uvector<SymbolOffsetT> stack_op_indices{stack_ops_bufsize, stream};

// Run bracket-brace FST to retrieve starting positions of structs and lists
json_to_stack_ops_fst.Transduce(json_in.begin(),
static_cast<SymbolOffsetT>(json_in.size()),
stack_ops.data(),
stack_op_indices.data(),
d_num_stack_ops.data(),
thrust::make_discard_iterator(),
to_stack_op::start_state,
stream);

Expand Down Expand Up @@ -1508,6 +1519,7 @@ std::pair<rmm::device_uvector<PdaTokenT>, rmm::device_uvector<SymbolOffsetT>> pr
device_span<SymbolOffsetT const> token_indices,
rmm::cuda_stream_view stream)
{
CUDF_FUNC_RANGE();
// Instantiate FST for post-processing the token stream to remove all tokens that belong to an
// invalid JSON line
token_filter::UnwrapTokenFromSymbolOp sgid_op{};
Expand Down Expand Up @@ -1643,21 +1655,28 @@ std::pair<rmm::device_uvector<PdaTokenT>, rmm::device_uvector<SymbolOffsetT>> ge
// see a JSON-line delimiter as the very first item
SymbolOffsetT const delimiter_offset =
(format == tokenizer_pda::json_format_cfg_t::JSON_LINES_RECOVER ? 1 : 0);
rmm::device_uvector<PdaTokenT> tokens{max_token_out_count + delimiter_offset, stream, mr};
rmm::device_uvector<SymbolOffsetT> tokens_indices{
max_token_out_count + delimiter_offset, stream, mr};

// Run FST to estimate the size of output buffers
json_to_tokens_fst.Transduce(zip_in,
static_cast<SymbolOffsetT>(json_in.size()),
tokens.data() + delimiter_offset,
tokens_indices.data() + delimiter_offset,
thrust::make_discard_iterator(),
thrust::make_discard_iterator(),
num_written_tokens.data(),
tokenizer_pda::start_state,
stream);

auto const num_total_tokens = num_written_tokens.value(stream) + delimiter_offset;
tokens.resize(num_total_tokens, stream);
tokens_indices.resize(num_total_tokens, stream);
rmm::device_uvector<PdaTokenT> tokens{num_total_tokens, stream, mr};
rmm::device_uvector<SymbolOffsetT> tokens_indices{num_total_tokens, stream, mr};

// Run FST to translate the input JSON string into tokens and indices at which they occur
json_to_tokens_fst.Transduce(zip_in,
static_cast<SymbolOffsetT>(json_in.size()),
tokens.data() + delimiter_offset,
tokens_indices.data() + delimiter_offset,
thrust::make_discard_iterator(),
tokenizer_pda::start_state,
stream);

if (delimiter_offset == 1) {
tokens.set_element(0, token_t::LineEnd, stream);
Expand Down
Loading