EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

[MISC] fix format check error (#654)

Browse Source 
This pr makes format.sh works as expect.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-04-29 11:14:19 +08:00
committed by GitHub
parent 1fce70a2fb
commit 0dae55a9a3
17 changed files with 45 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
csrc/kernels/pos_encoding_kernels.cpp
View File

@@ -29,7 +29,7 @@ using vllm_ascend::AccType;
using vllm_ascend::local_mem_copy;
template <typename scalar_t, bool isNeox> class RotaryEmbedding {
// NOTE(ganyi): we use 512B as load stride for pipe, need to find another way to
// retrive this size from runtime for more Soc support
// retrieve this size from runtime for more Soc support
static int constexpr loadSize = 512;
using dst_t = scalar_t;
using acc_t = typename AccType<scalar_t>::type;
@@ -66,7 +66,7 @@ public:
pipe_->InitBuffer(inQue_, 1 /* buffer_num */, loadSize /* buffer_size */);
pipe_->InitBuffer(inQueSinCos_, 1 /* buffer_num */, rotDim_ * sizeof(scalar_t) /* buffer_size */);
pipe_->InitBuffer(outQue_, 1 /* buffer_num */, loadSize /* buffer_size */);
// 2 temperary calculation buffer
// 2 temporary calculation buffer
calcTmpBufferOffset_ = 0;
// 1 upcast buffer for bf16 (headSize)
upcastInputBufferOffset_ = calcTmpBufferOffset_ + sizeof(acc_t) * embedDim_ * 2;
@@ -75,10 +75,10 @@ public:
// 2 sin cos upcast buffer for bf16
cosSinUpcastBufferOffset_ = upcastTempBufferOffset_ + sizeof(acc_t) * 2 * embedDim_;
// 2. bf16 path: needs 2 cos sin upcast buffer size
// 3. fp16 path: needs 2 temperary calculation buffer size
// 3. fp16 path: needs 2 temporary calculation buffer size
tempBufferSize_ = cosSinUpcastBufferOffset_ + 2 * embedDim_ * sizeof(acc_t);
// need to consider upcast the bf16 to fp32, so we might need 4 buffer just in case
// 2 temperary buffer, 2 input buffer, 1 cos buffer, 1 sin buffer, 2 scale buffer (headSize), 2 zp
// 2 temporary buffer, 2 input buffer, 1 cos buffer, 1 sin buffer, 2 scale buffer (headSize), 2 zp
// buffer(headSize int8), 1 dst_temp buffer(headSize, int32)
pipe_->InitBuffer(calcBuf_, tempBufferSize_ /* buffer_size */);
if constexpr (!std::is_same_v<scalar_t, acc_t>) {