fix call indexing adaptor issue

include/ck_tile/core/algorithm/indexing_adaptor.hpp

@@ -65,8 +65,8 @@ struct indexing_adaptor
     CK_TILE_HOST_DEVICE constexpr indexing_adaptor() = default;
     CK_TILE_HOST_DEVICE constexpr indexing_adaptor(const IndexingType* idx) : cached_idx_(idx) {}
     const IndexingType* cached_idx_;
-    mutable index_t  preUpIndex = 0;
-    mutable index_t  preLowIndex = 0;
+    mutable index_t pre_up_index_  = 0;
+    mutable index_t pre_low_index_ = 0;
 
     template <typename LowIdx, typename UpIdx>
     CK_TILE_HOST_DEVICE constexpr void calculate_lower_index(LowIdx& idx_low,
@@ -77,12 +77,14 @@ struct indexing_adaptor
 
         idx_low(number<0>{}) = *(cached_idx_ + idx_up[number<0>{}]);
 
-        preUpIndex = idx_up[number<0>{}];
-        preLowIndex = idx_low(number<0>{});
-        if(threadIdx.x == 0 && blockIdx.x == 0 && blockIdx.y == 0 && blockIdx.z == 0)
+        pre_up_index_  = idx_up[number<0>{}];
+        pre_low_index_ = idx_low(number<0>{});
+#if 0
+        if(threadIdx.x == 65 && blockIdx.x == 0 && blockIdx.y == 1 && blockIdx.z == 0)
         {
             printf("\n first index from  %d to  %d  \n", idx_up[number<0>{}], idx_low(number<0>{}));
         }
+#endif
     }
 
     template <typename LowIdxDiff, typename UpIdxDiff, typename LowIdx, typename UpIdx>
@@ -96,14 +98,14 @@ struct indexing_adaptor
                           UpIdx::size() == 1,
                       "wrong! inconsistent # of dimension");
 
-        int up_index  = idx_diff_up[number<0>{}] + preUpIndex;
+        int up_index              = idx_diff_up[number<0>{}] + pre_up_index_;
         int low_index             = *(cached_idx_ + up_index);
-        idx_diff_low(number<0>{}) = low_index - preLowIndex;
+        idx_diff_low(number<0>{}) = low_index - pre_low_index_;
 
-        preUpIndex = up_index;
-        preLowIndex = low_index;
-
-        if(threadIdx.x == 0 && blockIdx.x == 0 && blockIdx.y == 0 && blockIdx.z == 0)
+        pre_up_index_  = up_index;
+        pre_low_index_ = low_index;
+#if 0
+        if(threadIdx.x == 65 && blockIdx.x == 0 && blockIdx.y == 1 && blockIdx.z == 0)
         {
             printf("\n index form %d to %d, diff  from  %d to  %d  \n",
                    up_index,
@@ -111,6 +113,7 @@ struct indexing_adaptor
                    idx_diff_up[number<0>{}],
                    idx_diff_low(number<0>{}));
         }
+#endif
 
         // pass the diff to lower, but not changing the actually index
     }

include/ck_tile/ops/fused_moe/kernel/fused_moegemm_general_kernel.hpp

@@ -268,8 +268,8 @@ struct FusedMoeGemmGlKernel
         auto topk_weight =
             reinterpret_cast<const TopkWeightDataType*>(kargs.sorted_weight_ptr)[sorted_token_id];
 
-        const index_t* sorted_token_ids_ptr = reinterpret_cast<const index_t*>(
-            &(reinterpret_cast<const index_t*>(kargs.sorted_token_ids_ptr)[sorted_token_id]));
+        const index_t* sorted_token_ids_ptr =
+            reinterpret_cast<const index_t*>(kargs.sorted_token_ids_ptr);
 
         const auto a_window = [&]() {
             // A is already pre-padded in previous kernel

include/ck_tile/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_gl.hpp

@@ -104,19 +104,20 @@ struct FusedMoeGemmPipeline_FlatmmGl
         ignore = hidden_size;
         ignore = intermediate_size;
 
-        auto a_copy_dram_window =
-            make_tile_window(a_window_.get_bottom_tensor_view(),
+        auto a_copy_dram_window = make_tile_window(
+            a_window_.get_bottom_tensor_view(),
             make_tuple(number<BlockShape::Block_M0>{}, number<BlockShape::Block_K0>{}),
             a_window_.get_window_origin(),
             Policy::template MakeGlobalTileDistribution_A<Problem>());
         auto a_dram = load_tile(a_copy_dram_window);
+#if 0
         //check a matrix gather right or not
         constexpr auto a_spans = decltype(a_dram)::get_distributed_spans();
         int counter = 0;
         sweep_tile_span(a_spans[number<0>{}], [&](auto idxm) {
                 sweep_tile_span(a_spans[number<1>{}], [&](auto idxk){
                     constexpr auto i_j_idx = make_tuple(idxm, idxk);
-                    if(threadIdx.x == 0 && blockIdx.x == 0 && blockIdx.y == 0 && blockIdx.z == 0){
+                    if(threadIdx.x == 65 && blockIdx.x == 0 && blockIdx.y == 1 && blockIdx.z == 0){
                        counter = counter + 1;
                        index_t idm_0 = idxm.impl_.at(0);
                        index_t idn_0 = idxk.impl_.at(0);
@@ -124,8 +125,8 @@ struct FusedMoeGemmPipeline_FlatmmGl
                     }
                 });
             });
-
-        ignore = a_spans;
+#endif
+        ignore = a_dram;
     }
 };