format

example/32_batched_gemm_scale_softmax_gemm/batched_multihead_attention_backward_fp16.cpp

@@ -45,7 +45,6 @@ Kernel outputs:
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
 
-
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 

include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp

@@ -384,7 +384,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
         return transform_tensor_descriptor(
             ygrad_grid_desc_m_o,
-            make_tuple(make_unmerge_transform(make_tuple(Y_M0, Y_M1)), make_pass_through_transform(O)),
+            make_tuple(make_unmerge_transform(make_tuple(Y_M0, Y_M1)),
+                       make_pass_through_transform(O)),
             make_tuple(Sequence<0>{}, Sequence<1>{}),
             make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
     }
@@ -456,7 +457,6 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
     //
     // static auto MakeYGridDescriptor_MBlock_MPerBlock_OBlock_OPerBlock()
 
-
     //
     // dQ = alpha * dS * K
     //
@@ -700,8 +700,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
               vgrad_grid_desc_n_o_{DeviceOp::MakeVGradGridDescriptor_N_O(
                   b1_gs_gemm1ns_gemm1ks_lengths, b1_gs_gemm1ns_gemm1ks_strides)},
               /* PTrans descriptor will be constructed in kernel */
-              ygrad_grid_desc_m0_o_m1_{
-                  DeviceOp::MakeYGradGridDescriptor_M0_O_M1(y_grid_desc_m_o_)},
+              ygrad_grid_desc_m0_o_m1_{DeviceOp::MakeYGradGridDescriptor_M0_O_M1(y_grid_desc_m_o_)},
               // batch offsets
               a_grid_desc_g_m_k_{
                   Transform::MakeAGridDescriptor_G_M_K(a_gs_ms_ks_lengths, a_gs_ms_ks_strides)},

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp

@@ -214,43 +214,22 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
     // PGrad Gemm
     struct PGradGemmTile_M_N_O_
     {
-
     };
 
     template <index_t BlockSize_, index_t BlockSliceLength_M_, index_t BlockSliceLength_O_>
     struct YDotYGrad_M_O_
     {
         static constexpr index_t SrcScalarPerVetor = 16 / sizeof(DataType);
-        static constexpr auto ThreadClusterLength_O = Number<BlockSliceLength_O_ / SrcScalarPerVetor>{};
+        static constexpr auto ThreadClusterLength_O =
+            Number<BlockSliceLength_O_ / SrcScalarPerVetor>{};
         static constexpr auto ThreadClusterLength_M = Number<BlockSize_ / ThreadClusterLength_O>{};
         static constexpr auto ThreadSliceLength_O   = Number<SrcScalarPerVetor>{};
         static constexpr auto ThreadSliceLength_M =
             Number<BlockSliceLength_M_ * ThreadClusterLength_O / BlockSize_>{};
 
-        // static_assert(BlockSliceLength_O_ % SrcScalarPerVetor == 0, "");
-        // static_assert(BlockSize_ % ThreadClusterLength_O == 0, "");
         static_assert(ThreadClusterLength_O * ThreadSliceLength_O == BlockSliceLength_O_, "");
         static_assert(ThreadClusterLength_M * ThreadSliceLength_M == BlockSliceLength_M_, "");
 
-        using BlockwiseSumReduce = PartitionedBlockwiseReduction<
-            FloatGemmAcc,
-            BlockSize_,
-            Sequence<ThreadClusterLength_M, ThreadClusterLength_O>,
-            Sequence<0, 1>,
-            reduce::Add,
-            false>; // propagateNaN
-
-        // using ThreadReduceSrcDesc_M_O = decltype(make_naive_tensor_descriptor_packed(
-        //     make_tuple(ThreadSliceLength_M, ThreadSliceLength_O)));
-        // using ThreadReduceDstDesc_M =
-        //     decltype(make_naive_tensor_descriptor_packed(make_tuple(ThreadSliceLength_M)));
-        // using ThreadwiseSumReduce =
-        //     ThreadwiseReduction<FloatGemmAcc,
-        //                         ThreadReduceSrcDesc_M_O,
-        //                         ThreadReduceDstDesc_M,
-        //                         reduce::Add,
-        //                         false>; // propagateNaN
-
         using SrcBufType = StaticBuffer<AddressSpaceEnum::Vgpr,
                                         DataType,
                                         ThreadSliceLength_M * ThreadSliceLength_O,
@@ -1271,11 +1250,8 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
         //
         // dP
         //
-        constexpr auto y_thread_desc_m0_m1_o0_o1 =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           YDotYGrad_M_O::ThreadSliceLength_M,
-                                                           I1,
-                                                           YDotYGrad_M_O::ThreadSliceLength_O));
+        constexpr auto y_thread_desc_m0_m1_o0_o1 = make_naive_tensor_descriptor_packed(make_tuple(
+            I1, YDotYGrad_M_O::ThreadSliceLength_M, I1, YDotYGrad_M_O::ThreadSliceLength_O));
         constexpr auto y_thread_cluster_desc =
             make_cluster_descriptor(Sequence<I1,
                                              YDotYGrad_M_O::ThreadClusterLength_M,
@@ -1316,10 +1292,12 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
         constexpr auto y_dot_ygrad_block_desc_mblock_mrepeat_mwave_mperxdl =
             make_naive_tensor_descriptor(make_tuple(I1, P_M0, P_M1, P_M2),
                                          make_tuple(P_M0 * P_M1 * P_M2, P_M1 * P_M2, P_M2, I1));
-        // y_dot_ygrad thread buffer for calculating sgrad; reuse LSE thread descriptor because
-        // per-thread LSE data and y_dot_ygrad is tiled the same way
         constexpr auto y_dot_ygrad_thread_desc_mblock_mrepeat_mwave_mperxdl =
-            lse_thread_desc_mblock_mrepeat_mwave_mperxdl;
+            lse_thread_desc_mblock_mrepeat_mwave_mperxdl; // reuse LSE thread descriptor because
+                                                          // per-thread LSE data and y_dot_ygrad is
+                                                          // tiled the same way
+
+        // TODO ANT: dP Gemm can reuse first blockwise gemm and pipeline
 
         auto y_dot_ygrad_thread_copy_lds_to_vgpr = ThreadwiseTensorSliceTransfer_v2<
             FloatGemmAcc,
@@ -1404,7 +1382,8 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
         block_sync_lds();
         static_for<0, YDotYGrad_M_O::ThreadSliceLength_M, 1>{}([&](auto iM) {
             const auto idx_on_block = y_thread_data_on_block_idx[I1] + iM;
-            y_dot_ygrad_block_accum_buf.AtomicAdd(idx_on_block, true, y_dot_ygrad_thread_accum_buf[iM]);
+            y_dot_ygrad_block_accum_buf.AtomicAdd(
+                idx_on_block, true, y_dot_ygrad_thread_accum_buf[iM]);
         });
         block_sync_lds();
 
@@ -1416,7 +1395,7 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
         }
 #endif
 
-        // distribute to threads
+        // distribute y_dot_ygrad to threads; LDS accum buffer can be safely accessed after barrier
         y_dot_ygrad_thread_copy_lds_to_vgpr.Run(
             y_dot_ygrad_block_desc_mblock_mrepeat_mwave_mperxdl,
             y_dot_ygrad_block_accum_buf,
@@ -1667,9 +1646,9 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
 
         } while(++gemm1_k_block_outer_index < num_gemm1_k_block_outer_loop); // end j loop
 
-        // TODO ANT:
-        // shuffle dQ and write
-        #if 0
+// TODO ANT:
+// shuffle dQ and write
+#if 0
         {
             static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
                               Gemm1NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
@@ -1865,7 +1844,7 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
                 }
             });
         }
-        #endif
+#endif
     }
 };