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

Revert "[Kernel] add custom moe ops for prefill" (#4806)

Browse Source 
Reverts vllm-project/vllm-ascend#4194 as it broke CI in
https://github.com/vllm-project/vllm-ascend/actions/runs/20030369087/job/57437687382?pr=4791

Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
Mengqing Cao
2025-12-08 23:20:32 +08:00
committed by GitHub
parent 432b861cae
commit 7e70da9fb7
39 changed files with 2 additions and 5562 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
csrc/dispatch_layout/op_host/aclnn_dispatch_layout.h
View File

@@ -1,50 +0,0 @@
#ifndef ACLNN_DISPATCH_LAYOUT_H_
#define ACLNN_DISPATCH_LAYOUT_H_
#include "aclnn/acl_meta.h"
#ifdef __cplusplus
extern "C" {
#endif
/* funtion: aclnnDispatchLayoutGetWorkspaceSize
* topkIdx : required
* numTokens : required
* numRanks : required
* numExperts : required
* numTopk : required
* numTokensPerRank : required
* numTokensPerExpert : required
* isTokenInRank : required
* workspaceSize : size of workspace(output).
* executor : executor context(output).
*/
__attribute__((visibility("default"))) aclnnStatus aclnnDispatchLayoutGetWorkspaceSize(
const aclTensor *topkIdx,
int64_t numTokens,
int64_t numRanks,
int64_t numExperts,
int64_t numTopk,
const aclTensor *numTokensPerRank,
const aclTensor *numTokensPerExpert,
const aclTensor *isTokenInRank,
uint64_t *workspaceSize,
aclOpExecutor **executor);
/* funtion: aclnnDispatchLayout
* workspace : workspace memory addr(input).
* workspaceSize : size of workspace(input).
* executor : executor context(input).
* stream : acl stream.
*/
__attribute__((visibility("default"))) aclnnStatus aclnnDispatchLayout(
void *workspace,
uint64_t workspaceSize,
aclOpExecutor *executor,
aclrtStream stream);
#ifdef __cplusplus
}
#endif
#endif