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

GQA Memory Efficient Kernel #17920

Merged
merged 20 commits into from
Nov 2, 2023
Merged

GQA Memory Efficient Kernel #17920

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
ef82d4d
squash merge
aciddelgado Oct 12, 2023
54f2526
add unit test and fix build
aciddelgado Oct 12, 2023
e2e1157
undo work in attention_impl file
aciddelgado Oct 17, 2023
415440b
reduce tests and change default behavior for past-kv is nullptr
aciddelgado Oct 19, 2023
0e84d2d
test compatibility w/ no cuda
aciddelgado Oct 19, 2023
9ba6963
exclude from amd
aciddelgado Oct 19, 2023
6573133
fix test script
aciddelgado Oct 20, 2023
a87f211
make kernels more efficient and make present output required
aciddelgado Oct 24, 2023
16bda28
Merge branch 'main' into aciddelgado/gqa_memeff_v2
aciddelgado Oct 24, 2023
08f553d
merge main and memeff changes
aciddelgado Oct 24, 2023
eb12522
address comments
aciddelgado Oct 25, 2023
6c6aead
update ContribOperators.md
aciddelgado Oct 25, 2023
db307f3
clarify input and output formats memory efficient attention
aciddelgado Oct 27, 2023
e7a50ee
max sequence length for memory efficient attention
aciddelgado Oct 27, 2023
bc1cf0a
clang and fix test file
aciddelgado Oct 27, 2023
44ca857
undo clang on unrelated files
aciddelgado Oct 27, 2023
f08495f
Merge branch 'main' into aciddelgado/gqa_memeff_v2
aciddelgado Oct 30, 2023
660c8fd
check value and key inputs
aciddelgado Oct 31, 2023
b9c4d15
key and value dont check for nullptr since they are required
aciddelgado Oct 31, 2023
fd0ecc3
lint
aciddelgado Nov 1, 2023
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: 0 additions & 1 deletion onnxruntime/contrib_ops/cuda/bert/attention_impl.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,6 @@ size_t GetSequenceOffsetSize(int batch_size, bool has_padding) {
// There are batch_size + 1 offsets Without padding (or padding removed), and 2 * batch_size + 1 with padding.
size_t bytes = sizeof(int) * ((has_padding ? 2 * batch_size : batch_size) + 1);
return AlignSize(bytes);
;
aciddelgado marked this conversation as resolved.
Show resolved Hide resolved
}

size_t GetAttentionWorkspaceSize(
Expand Down
2 changes: 1 addition & 1 deletion onnxruntime/contrib_ops/cuda/bert/cutlass_fmha/fmha_launch_template.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@ void LaunchCutlassFmha(const MemoryEfficientAttentionParams& params) {
p.num_keys = params.kv_sequence_length;

if (params.causal) {
p.custom_mask_type = Attention::CausalFromTopLeft;
p.custom_mask_type = Attention::CausalFromBottomRight;
}

// Input format is BxSxNxH, output is BxSxNxH
Expand Down
74 changes: 60 additions & 14 deletions onnxruntime/contrib_ops/cuda/bert/group_query_attention.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
#include "contrib_ops/cuda/bert/group_query_attention_impl.h"
#include "contrib_ops/cuda/bert/group_query_attention.h"
#include "contrib_ops/cuda/bert/group_query_attention_helper.h"
#include "contrib_ops/cuda/bert/cutlass_fmha/memory_efficient_attention.h"
#include "contrib_ops/cuda/bert/flash_attention/flash_api.h"
// #include "contrib_ops/cuda/transformers/dump_cuda_tensor.h"
tianleiwu marked this conversation as resolved.
Show resolved Hide resolved
// #include "contrib_ops/cpu/utils/console_dumper.h"
Expand Down Expand Up @@ -55,6 +56,13 @@ GroupQueryAttention<T>::GroupQueryAttention(const OpKernelInfo& info)
#else
disable_flash_attention_ = true;
#endif

#if USE_MEMORY_EFFICIENT_ATTENTION
disable_memory_efficient_attention_ = sizeof(T) != 2 ||
ParseEnvironmentVariableWithDefault<bool>(attention::kDisableMemoryEfficientAttention, false);
#else
disable_memory_efficient_attention_ = true;
#endif
}

template <typename T>
Expand Down Expand Up @@ -92,18 +100,6 @@ Status GroupQueryAttention<T>::ComputeInternal(OpKernelContext* context) const {
output_shape[2] = static_cast<int64_t>(parameters.hidden_size);
Tensor* output = context->Output(0, output_shape);

std::vector<int64_t> present_dims;
if (parameters.past_kv_format == AttentionQkvFormat::Q_K_V_BSNH) {
present_dims = {
parameters.batch_size, parameters.present_sequence_length, parameters.kv_num_heads, parameters.head_size};
} else { // BNSH
present_dims = {
parameters.batch_size, parameters.kv_num_heads, parameters.present_sequence_length, parameters.head_size};
}
TensorShape present_shape(present_dims);
Tensor* present_key = context->Output(1, present_shape);
Tensor* present_value = context->Output(2, present_shape);

#if USE_FLASH_ATTENTION
bool use_flash_attention = !disable_flash_attention_ &&
onnxruntime::flash::is_supported(device_prop,
Expand Down Expand Up @@ -149,8 +145,50 @@ Status GroupQueryAttention<T>::ComputeInternal(OpKernelContext* context) const {
auto seqlens_k_buffer = GetScratchBuffer<void>(0, context->GetComputeStream()); // nullptr
#endif

// only kernel implemented for gqa right now
ORT_ENFORCE(use_flash_attention);
#if USE_MEMORY_EFFICIENT_ATTENTION
int sm = (device_prop.major * 10) + device_prop.minor;
bool use_memory_efficient_attention =
!use_flash_attention &&
!disable_memory_efficient_attention_ &&
(parameters.head_size & 7) == 0 &&
parameters.sequence_length <= parameters.past_sequence_length + parameters.kv_sequence_length &&
(sizeof(T) == 2 || parameters.sequence_length >= attention::kMinSeqLenForMemoryEfficientAttentionFp32) &&
has_memory_efficient_attention(sm, sizeof(T) == 2);
// allocate buffers
size_t kv_buffer_bytes = 0;
const bool needs_buff = (parameters.num_heads != parameters.kv_num_heads) || (parameters.past_kv_format != AttentionQkvFormat::Q_K_V_BSNH) || (past_key != nullptr && parameters.present_sequence_length != parameters.past_sequence_length + parameters.kv_sequence_length);
if (use_memory_efficient_attention && needs_buff) {
tianleiwu marked this conversation as resolved.
Show resolved Hide resolved
if (past_key == nullptr) {
kv_buffer_bytes = (sizeof(T) * parameters.batch_size * parameters.num_heads * parameters.kv_sequence_length * parameters.head_size);
tianleiwu marked this conversation as resolved.
Show resolved Hide resolved
} else {
kv_buffer_bytes = (sizeof(T) * parameters.batch_size * parameters.num_heads * (parameters.past_sequence_length + parameters.kv_sequence_length) * parameters.head_size);
}
}
size_t fmha_buffer_bytes = 0;
if (use_memory_efficient_attention && MemoryEfficientAttentionParams::need_workspace(parameters.head_size, sizeof(T) == sizeof(float))) {
fmha_buffer_bytes = (parameters.batch_size * parameters.sequence_length * parameters.num_heads * parameters.head_size * sizeof(float));
}
auto k_buffer = GetScratchBuffer<void>(kv_buffer_bytes, context->GetComputeStream());
auto v_buffer = GetScratchBuffer<void>(kv_buffer_bytes, context->GetComputeStream());
auto fmha_buffer = GetScratchBuffer<void>(fmha_buffer_bytes, context->GetComputeStream());
#else
constexpr bool use_memory_efficient_attention = false;
auto k_buffer = GetScratchBuffer<void>(0, context->GetComputeStream());
auto v_buffer = GetScratchBuffer<void>(0, context->GetComputeStream());
auto fmha_buffer = GetScratchBuffer<void>(0, context->GetComputeStream());
#endif

std::vector<int64_t> present_dims;
if (parameters.past_kv_format == AttentionQkvFormat::Q_K_V_BSNH) {
present_dims = {
parameters.batch_size, parameters.present_sequence_length, parameters.kv_num_heads, parameters.head_size};
} else { // BNSH
present_dims = {
parameters.batch_size, parameters.kv_num_heads, parameters.present_sequence_length, parameters.head_size};
}
TensorShape present_shape(present_dims);
Tensor* present_key = context->Output(1, present_shape);
Tensor* present_value = context->Output(2, present_shape);

data.query = reinterpret_cast<const CudaT*>(query->Data<T>());
data.key = reinterpret_cast<const CudaT*>(key->Data<T>());
Expand All @@ -161,6 +199,7 @@ Status GroupQueryAttention<T>::ComputeInternal(OpKernelContext* context) const {
data.present_key = (nullptr == present_key) ? nullptr : reinterpret_cast<CudaT*>(present_key->MutableData<T>());
data.present_value = (nullptr == present_value) ? nullptr : reinterpret_cast<CudaT*>(present_value->MutableData<T>());
data.use_flash_attention = use_flash_attention;
data.use_memory_efficient_attention = use_memory_efficient_attention;
if (softmax_lse_buffer != nullptr) {
data.softmax_lse = reinterpret_cast<CudaT*>(softmax_lse_buffer.get());
}
Expand All @@ -173,6 +212,13 @@ Status GroupQueryAttention<T>::ComputeInternal(OpKernelContext* context) const {
if (seqlens_k_buffer != nullptr) {
data.seqlens_k = reinterpret_cast<int*>(seqlens_k_buffer.get());
}
if (k_buffer != nullptr) {
data.k = reinterpret_cast<CudaT*>(k_buffer.get());
data.v = reinterpret_cast<CudaT*>(v_buffer.get());
}
if (fmha_buffer != nullptr) {
data.fmha_buffer = reinterpret_cast<CudaT*>(fmha_buffer.get());
}

cublasHandle_t cublas = GetCublasHandle(context);

Expand Down
1 change: 1 addition & 0 deletions onnxruntime/contrib_ops/cuda/bert/group_query_attention.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@ class GroupQueryAttention final : public CudaKernel {
bool is_past_bsnh_;
float scale_;
bool disable_flash_attention_;
bool disable_memory_efficient_attention_;
};

} // namespace cuda
Expand Down
Loading