add to device

b60595f9 · letaoqin · 83d13bff · b60595f9 · b60595f9
Commit b60595f9 authored Aug 11, 2023 by letaoqin
2 changed files
--- a/example/52_flash_atten_bias/batched_multihead_attention_bias_backward_v2.cpp
+++ b/example/52_flash_atten_bias/batched_multihead_attention_bias_backward_v2.cpp
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_mha_bwd_xdl_cshuffle_qloop_v2.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_mha_bwd_xdl_cshuffle_qloop_v2.hpp
@@ -278,6 +278,7 @@ template <index_t NumDimG,
          index_t BBlockTransferSrcScalarPerVector,
          index_t BBlockTransferDstScalarPerVector_BK1,
          bool BBlockLdsExtraN,
+          index_t D0BlockTransferSrcScalarPerVector,
          typename B1BlockTransferThreadClusterLengths_BK0_N_BK1,
          typename B1BlockTransferThreadClusterArrangeOrder,
          typename B1BlockTransferSrcAccessOrder,
@@ -302,7 +303,7 @@ struct DeviceBatchedMultiheadAttentionBackward_Qloop_Xdl_CShuffle_V2
    static constexpr index_t NumAcc1Bias = Acc1BiasDataType::Size();

    // TODO: implement bias combination
-    static_assert(NumAcc0Bias == 0 && NumAcc0Bias == 0, "Bias addition is unimplemented");
+    static_assert(NumAcc1Bias == 0, "Bias addition is unimplemented");

    using DeviceOp = DeviceBatchedMultiheadAttentionBackward_Qloop_Xdl_CShuffle_V2;