update block sparse attention kernel.

Signed-off-by: Fanrong Li <[email protected]>

fix params issue

Signed-off-by: yuhangh <[email protected]>

Author: lfr-0531

cpp/tensorrt_llm/common/attentionOp.h

@@ -494,11 +494,11 @@ class AttentionOp
             mPosShiftEnabled, mPagedContextFMHA, mFP8ContextFMHA, mFP8AttenOutput, mFP8ContextMLA, mFP8GenerationMLA,
             mChunkPrefillBufferBatchSize, mDenseContextFMHA, mHasFullAttentionMask, mIsSpecDecodingEnabled,
             mUseSpecDecoding, mIsSpecDecTree, mSpecDecodingIsGenerationLengthVariable, mSpecDecodingMaxGenerationLength,
-            mIsMLAEnabled, mIsGenerationMLA, mUseGenFlashMLA, mUseSparseAttention, mMLAParams.data(), mCpSize, mCpRank,
-            mCpGroup, mNumAttnHeads, mNumAttnKVHeads, mNumKVHeadsOrigin, mAttnTpSize, mAttnTpRank, mAttnCpSize,
-            mAttnCpRank, mUlyssesMQABroadcast, mEnableContextFMHA, mFMHAForceFP32Acc, mMultiBlockMode, mEnableXQA,
-            mUseKVCache, mSkipAttn, mFuseFp4Quant, mRuntimeSparseAttentionParams.data(), mNbMultiBlockSemaphores,
-            mAttentionChunkSize.value_or(-1));
+            mIsMLAEnabled, mIsGenerationMLA, mUseGenFlashMLA, mUseSparseAttention, mUseTllmGenSparseAttention,
+            mMLAParams.data(), mCpSize, mCpRank, mCpGroup, mNumAttnHeads, mNumAttnKVHeads, mNumKVHeadsOrigin,
+            mAttnTpSize, mAttnTpRank, mAttnCpSize, mAttnCpRank, mUlyssesMQABroadcast, mEnableContextFMHA,
+            mFMHAForceFP32Acc, mMultiBlockMode, mEnableXQA, mUseKVCache, mSkipAttn, mFuseFp4Quant,
+            mRuntimeSparseAttentionParams.data(), mNbMultiBlockSemaphores, mAttentionChunkSize.value_or(-1));
     };
 
 private: