Skip to content

clean up unused variables in int_nbit_split_embedding_codegen_forward_cpu #927

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

Closed
wants to merge 1 commit into from
Closed

clean up unused variables in int_nbit_split_embedding_codegen_forward_cpu #927

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
64 changes: 26 additions & 38 deletions fbgemm_gpu/codegen/embedding_forward_quantized_cpu_template.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -169,10 +169,6 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
{% if weighted %}
const float* indice_weights_acc = indice_weights.data_ptr<float>();
{% endif %}
// Empty array filled with zeros (thus accumulating to zero).
// max-D = 1024, max-sizeof(T) = sizeof(float) = 4.
alignas(32) static constexpr std::array<uint8_t, 1024 * 4> zero_row = {0};
std::vector<__m256> acc; //, {{ kMaxVecsPerThread }} > acc;

AT_DISPATCH_INDEX_TYPES(indices.scalar_type(), "int_nbit_split_embedding_codegen_forward_", [&] () {
const auto* indices_acc = indices.data_ptr<index_t>();
Expand Down Expand Up @@ -207,11 +203,19 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
float>::type;

bool success = true;
bool has_weight = {{ "true" if weighted else "false" }};
bool normalize_by_lengths = static_cast<PoolingMode>(pooling_mode) == PoolingMode::MEAN;

index_t index_size = offsets_acc[(t + 1) * B] - *offsets_begin_ptr;
const float* indice_weights_ptr = nullptr;
{% if weighted %}
indice_weights_ptr = indice_weights_acc + *offsets_begin_ptr;
{% endif %}
if (weight_ty == SparseType::FP32) {
auto kernel = fbgemm::GenerateEmbeddingSpMDMWithStrides<float, index_t, index_t, fbgemm_out_t, /*THREAD_LOCAL=*/true>(
D,
{{ "true" if weighted else "false" }},
static_cast<PoolingMode>(pooling_mode) == PoolingMode::MEAN,
has_weight,
normalize_by_lengths,
/*prefetch=*/16,
/*is_weight_positional=*/false,
/*use_offsets=*/true,
Expand All @@ -220,22 +224,18 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
/*scale_bias_last=*/false);
success = kernel(
B,
offsets_acc[(t + 1) * B] - *offsets_begin_ptr,
index_size,
num_rows,
reinterpret_cast<const float*>(weights),
indices_acc + *offsets_begin_ptr,
offsets_begin_ptr,
{% if weighted %}
indice_weights_acc + *offsets_begin_ptr,
{% else %}
nullptr,
{% endif %}
indice_weights_ptr,
reinterpret_cast<fbgemm_out_t*>(output_acc + D_start));
} else if (weight_ty == SparseType::FP16) {
auto kernel = fbgemm::GenerateEmbeddingSpMDMWithStrides<float16, index_t, index_t, fbgemm_out_t, /*THREAD_LOCAL=*/true>(
D,
{{ "true" if weighted else "false" }},
static_cast<PoolingMode>(pooling_mode) == PoolingMode::MEAN,
has_weight,
normalize_by_lengths,
/*prefetch=*/16,
/*is_weight_positional=*/false,
/*use_offsets=*/true,
Expand All @@ -244,22 +244,18 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
/*scale_bias_last=*/false);
success = kernel(
B,
offsets_acc[(t + 1) * B] - *offsets_begin_ptr,
index_size,
num_rows,
reinterpret_cast<const float16*>(weights),
indices_acc + *offsets_begin_ptr,
offsets_begin_ptr,
{% if weighted %}
indice_weights_acc + *offsets_begin_ptr,
{% else %}
nullptr,
{% endif %}
indice_weights_ptr,
reinterpret_cast<fbgemm_out_t*>(output_acc + D_start));
} else if (weight_ty == SparseType::INT8) {
auto kernel = fbgemm::GenerateEmbeddingSpMDMWithStrides<uint8_t, index_t, index_t, fbgemm_out_t, /*THREAD_LOCAL=*/true>(
D,
{{ "true" if weighted else "false" }},
static_cast<PoolingMode>(pooling_mode) == PoolingMode::MEAN,
has_weight,
normalize_by_lengths,
/*prefetch=*/16,
/*is_weight_positional=*/false,
/*use_offsets=*/true,
Expand All @@ -268,23 +264,19 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
/*scale_bias_last=*/false);
success = kernel(
B,
offsets_acc[(t + 1) * B] - *offsets_begin_ptr,
index_size,
num_rows,
reinterpret_cast<const uint8_t*>(weights),
weights,
indices_acc + *offsets_begin_ptr,
offsets_begin_ptr,
{% if weighted %}
indice_weights_acc + *offsets_begin_ptr,
{% else %}
nullptr,
{% endif %}
indice_weights_ptr,
reinterpret_cast<fbgemm_out_t*>(output_acc + D_start));
} else if (weight_ty == SparseType::INT4) {
auto kernel = fbgemm::GenerateEmbeddingSpMDMNBitWithStrides<index_t, index_t, fbgemm_out_t, /*THREAD_LOCAL=*/true>(
/*bit_rate=*/4,
D,
{{ "true" if weighted else "false" }},
static_cast<PoolingMode>(pooling_mode) == PoolingMode::MEAN,
has_weight,
normalize_by_lengths,
/*prefetch=*/16,
/*is_weight_positional=*/false,
/*use_offsets=*/true,
Expand All @@ -293,16 +285,12 @@ Tensor int_nbit_split_embedding_codegen_forward_{{ wdesc }}_cpu(
/*scale_bias_last=*/false);
success = kernel(
B,
offsets_acc[(t + 1) * B] - *offsets_begin_ptr,
index_size,
num_rows,
reinterpret_cast<const uint8_t*>(weights),
weights,
indices_acc + *offsets_begin_ptr,
offsets_begin_ptr,
{% if weighted %}
indice_weights_acc + *offsets_begin_ptr,
{% else %}
nullptr,
{% endif %}
indice_weights_ptr,
reinterpret_cast<fbgemm_out_t*>(output_acc + D_start));
} else {
throw std::logic_error("Unsupported SparseType: " + std::to_string(static_cast<int>(weight_ty)));
Expand Down