[Performance][MLA][ROCm] Remove redundant D2D copy in deepseek (#27457)

Signed-off-by: ganyi <ygan@amd.com>
This commit is contained in:
Pleaplusone
2025-11-26 12:45:28 +08:00
committed by GitHub
parent 53d7f1f601
commit d9d342d214
5 changed files with 49 additions and 41 deletions
+12 -15
View File
@@ -16,7 +16,8 @@ __global__ void merge_attn_states_kernel(
scalar_t* output, float* output_lse, const scalar_t* prefix_output,
const float* prefix_lse, const scalar_t* suffix_output,
const float* suffix_lse, const uint num_tokens, const uint num_heads,
const uint head_size) {
const uint head_size, const uint prefix_head_stride,
const uint output_head_stride) {
using pack_128b_t = uint4;
const uint pack_size = 16 / sizeof(scalar_t);
const uint threads_per_head = head_size / pack_size;
@@ -34,11 +35,13 @@ __global__ void merge_attn_states_kernel(
const uint head_idx = token_head_idx % num_heads;
const uint pack_offset = pack_idx * pack_size; // (0~15)*8, etc.
const uint head_offset =
token_idx * num_heads * head_size + head_idx * head_size;
const scalar_t* prefix_head_ptr = prefix_output + head_offset;
const scalar_t* suffix_head_ptr = suffix_output + head_offset;
scalar_t* output_head_ptr = output + head_offset;
const uint src_head_offset = token_idx * num_heads * prefix_head_stride +
head_idx * prefix_head_stride;
const uint dst_head_offset = token_idx * num_heads * output_head_stride +
head_idx * output_head_stride;
const scalar_t* prefix_head_ptr = prefix_output + src_head_offset;
const scalar_t* suffix_head_ptr = suffix_output + src_head_offset;
scalar_t* output_head_ptr = output + dst_head_offset;
float p_lse = prefix_lse[head_idx * num_tokens + token_idx];
float s_lse = suffix_lse[head_idx * num_tokens + token_idx];
@@ -140,7 +143,7 @@ __global__ void merge_attn_states_kernel(
reinterpret_cast<float*>(prefix_lse.data_ptr()), \
reinterpret_cast<scalar_t*>(suffix_output.data_ptr()), \
reinterpret_cast<float*>(suffix_lse.data_ptr()), num_tokens, \
num_heads, head_size); \
num_heads, head_size, prefix_head_stride, output_head_stride); \
}
/*@brief Merges the attention states from prefix and suffix
@@ -166,17 +169,11 @@ void merge_attn_states_launcher(torch::Tensor& output,
const uint num_tokens = output.size(0);
const uint num_heads = output.size(1);
const uint head_size = output.size(2);
const uint prefix_head_stride = prefix_output.stride(1);
const uint output_head_stride = output.stride(1);
const uint pack_size = 16 / sizeof(scalar_t);
TORCH_CHECK(head_size % pack_size == 0,
"headsize must be multiple of pack_size:", pack_size);
TORCH_CHECK(output.stride(-2) == head_size && output.stride(-1) == 1,
"output heads must be contiguous in memory");
TORCH_CHECK(
prefix_output.stride(-2) == head_size && prefix_output.stride(-1) == 1,
"prefix_output heads must be contiguous in memory");
TORCH_CHECK(
suffix_output.stride(-2) == head_size && suffix_output.stride(-1) == 1,
"suffix_output heads must be contiguous in memory");
float* output_lse_ptr = nullptr;
if (output_lse.has_value()) {
output_lse_ptr = output_lse.value().data_ptr<float>();
+1 -2
View File
@@ -52,14 +52,13 @@ void paged_attention_v2(
const int64_t blocksparse_vert_stride, const int64_t blocksparse_block_size,
const int64_t blocksparse_head_sliding_step);
#ifndef USE_ROCM
void merge_attn_states(torch::Tensor& output,
std::optional<torch::Tensor> output_lse,
const torch::Tensor& prefix_output,
const torch::Tensor& prefix_lse,
const torch::Tensor& suffix_output,
const torch::Tensor& suffix_lse);
#ifndef USE_ROCM
void convert_vertical_slash_indexes(
torch::Tensor& block_count, // [BATCH, N_HEADS, NUM_ROWS]
torch::Tensor& block_offset, // [BATCH, N_HEADS, NUM_ROWS, NNZ_S]
+1 -2
View File
@@ -63,7 +63,6 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
" int blocksparse_head_sliding_step) -> ()");
ops.impl("paged_attention_v2", torch::kCUDA, &paged_attention_v2);
#ifndef USE_ROCM
// Merge attn states
// Implements section 2.2 of https://www.arxiv.org/pdf/2501.01005
// can be used to combine partial attention results (in the split-KV case)
@@ -76,7 +75,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
" Tensor suffix_output,"
" Tensor suffix_lse) -> ()");
ops.impl("merge_attn_states", torch::kCUDA, &merge_attn_states);
#ifndef USE_ROCM
ops.def(
"convert_vertical_slash_indexes("
" Tensor! block_count, Tensor! block_offset, "