Revert "[TRTLLM-5965] perf: Optimize MoE sort kernels for large-scale EP (NVIDIA#5435)"

This reverts commit b4dab23.

Author: syuoni

cpp/tensorrt_llm/kernels/cutlass_kernels/include/moe_kernels.h

@@ -87,6 +87,32 @@ struct LoraParams
 
 namespace cutlass_kernels
 {
+static inline size_t pad_to_multiple_of_16(size_t const& input)
+{
+    static constexpr int ALIGNMENT = 16;
+    return ALIGNMENT * ((input + ALIGNMENT - 1) / ALIGNMENT);
+}
+
+class CubKeyValueSorter
+{
+public:
+    CubKeyValueSorter();
+
+    CubKeyValueSorter(int const num_experts_per_node);
+
+    void updateNumExperts(int const num_experts_per_node);
+
+    static size_t getWorkspaceSize(size_t const num_key_value_pairs, int const num_experts_per_node);
+
+    void run(void* workspace, size_t const workspace_size, int const* keys_in, int* keys_out, int const* values_in,
+        int* values_out, size_t const num_key_value_pairs, cudaStream_t stream);
+
+private:
+    static int expertsToBits(int experts);
+    int num_experts_;
+    int num_bits_;
+};
+
 /**
  * \brief Describes what parallelism mode the MoE is using
  *
@@ -371,9 +397,9 @@ class CutlassMoeFCRunnerInterface
         ActivationType fc1_activation_type, void const* fc2_expert_weights, void const* fc2_expert_biases,
         QuantParams quant_params, int64_t const num_rows, int64_t const hidden_size, int64_t const inter_size,
         int const num_experts, int const experts_per_token, char* workspace_ptr, void* final_output,
-        int* unpermuted_row_to_permuted_row, MOEParallelismConfig parallelism_config, bool const enable_alltoall,
-        bool use_lora, LoraParams& lora_params, bool use_deepseek_fp8_block_scale, bool min_latency_mode,
-        MoeMinLatencyParams& min_latency_params, cudaStream_t stream)
+        int* expanded_source_row_to_expanded_dest_row, MOEParallelismConfig parallelism_config,
+        bool const enable_alltoall, bool use_lora, LoraParams& lora_params, bool use_deepseek_fp8_block_scale,
+        bool min_latency_mode, MoeMinLatencyParams& min_latency_params, cudaStream_t stream)
         = 0;
 
     // Aliases for profiling the gemms
@@ -387,22 +413,22 @@ class CutlassMoeFCRunnerInterface
         int const num_experts_per_node, ActivationType fc1_activation_type, float const** alpha_scale_ptr_array,
         bool bias_is_broadcast, bool use_deepseek_fp8_block_scale, cudaStream_t stream,
         cutlass_extensions::CutlassGemmConfig config, bool min_latency_mode, int* num_active_experts_per,
-        int* active_expert_global_ids)
+        int* active_expert_global_ids, int start_expert)
         = 0;
 
     virtual void gemm2(void const* const input, void* const gemm_output, void* const final_output,
         int64_t const* const expert_first_token_offset, TmaWarpSpecializedGroupedGemmInput const tma_ws_input_template,
         void const* const fc2_expert_weights, void const* const fc2_expert_biases, void const* const fc2_int_scales,
         float const* const fc2_fp8_dequant, TmaWarpSpecializedGroupedGemmInput::ElementSF const* fc2_fp4_act_flat,
         QuantParams quant_params, float const* const token_topk_unpermuted_scales,
-        float const* const token_topk_permuted_scales, int const* const unpermuted_row_to_permuted_row,
-        int const* permuted_row_to_unpermuted_row, int const* const token_selected_experts,
+        float const* const token_topk_permuted_scales, int const* const expanded_source_row_to_expanded_dest_row,
+        int const* expanded_dest_row_to_expanded_source_row, int const* const expert_for_source_row,
         int64_t const* const num_valid_tokens_ptr, int64_t const num_rows, int64_t const expanded_num_rows,
         int64_t const hidden_size, int64_t const inter_size, int const num_experts_per_node,
         int64_t const experts_per_token, float const** alpha_scale_ptr_array, bool use_lora, void* fc2_lora,
         bool use_deepseek_fp8_block_scale, cudaStream_t stream, MOEParallelismConfig parallelism_config,
         bool const enable_alltoall, cutlass_extensions::CutlassGemmConfig config, bool min_latency_mode,
-        int* num_active_experts_per, int* active_expert_global_ids)
+        int* num_active_experts_per, int* active_expert_global_ids, int start_expert)
         = 0;
 
     virtual std::pair<TmaWarpSpecializedGroupedGemmInput, TmaWarpSpecializedGroupedGemmInput>
@@ -518,9 +544,9 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
         ActivationType fc1_activation_type, void const* fc2_expert_weights, void const* fc2_expert_biases,
         QuantParams quant_params, int64_t const num_rows, int64_t const hidden_size, int64_t const inter_size,
         int const num_experts, int const experts_per_token, char* workspace_ptr, void* final_output,
-        int* unpermuted_row_to_permuted_row, MOEParallelismConfig parallelism_config, bool const enable_alltoall,
-        bool use_lora, LoraParams& lora_params, bool use_deepseek_fp8_block_scale, bool min_latency_mode,
-        MoeMinLatencyParams& min_latency_params, cudaStream_t stream) override;
+        int* expanded_source_row_to_expanded_dest_row, MOEParallelismConfig parallelism_config,
+        bool const enable_alltoall, bool use_lora, LoraParams& lora_params, bool use_deepseek_fp8_block_scale,
+        bool min_latency_mode, MoeMinLatencyParams& min_latency_params, cudaStream_t stream) override;
 
     // We make these GEMM1 & GEMM2 static because they need to be stateless for the profiler to work
     static void gemm1(MoeGemmRunner<T, WeightType, OutputType, ScaleBiasType>& gemm_runner,
@@ -539,7 +565,7 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
         int64_t const num_rows, int64_t const expanded_num_rows, int64_t const hidden_size, int64_t const inter_size,
         int const num_experts_per_node, ActivationType fc1_activation_type, float const** alpha_scale_ptr_array,
         bool bias_is_broadcast, cudaStream_t stream, cutlass_extensions::CutlassGemmConfig config,
-        bool min_latency_mode, int* num_active_experts_per, int* active_expert_global_ids);
+        bool min_latency_mode, int* num_active_experts_per, int* active_expert_global_ids, int start_expert);
 
     static void gemm2(MoeGemmRunner<T, WeightType, OutputType, ScaleBiasType>& gemm_runner,
         DeepSeekBlockScaleGemmRunner* fp8_blockscale_gemm_runner, T const* const input, void* const gemm_output,
@@ -548,14 +574,14 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
         ScaleBiasType const* const fc2_expert_biases, ScaleBiasType const* const fc2_int_scales,
         float const* const fc2_fp8_dequant, TmaWarpSpecializedGroupedGemmInput::ElementSF const* fc2_fp4_act_flat,
         QuantParams quant_params, float const* const token_topk_unpermuted_scales,
-        float const* const token_topk_permuted_scales, int const* const unpermuted_row_to_permuted_row,
-        int const* permuted_row_to_unpermuted_row, int const* const token_selected_experts,
+        float const* const token_topk_permuted_scales, int const* const expanded_source_row_to_expanded_dest_row,
+        int const* expanded_dest_row_to_expanded_source_row, int const* const expert_for_source_row,
         int64_t const* const num_valid_tokens_ptr, int64_t const num_rows, int64_t const expanded_num_rows,
         int64_t const hidden_size, int64_t const inter_size, int const num_experts_per_node,
         int64_t const experts_per_token, float const** alpha_scale_ptr_array, bool use_lora, void* fc2_lora,
         cudaStream_t stream, MOEParallelismConfig parallelism_config, bool const enable_alltoall,
         cutlass_extensions::CutlassGemmConfig config, bool min_latency_mode, int* num_active_experts_per,
-        int* active_expert_global_ids);
+        int* active_expert_global_ids, int start_expert);
 
     // Overrides to allow us to forward on to the internal functions with the pointers using the correct type
     void gemm1(void const* const input, void* const output, void* const intermediate_result,
@@ -568,7 +594,7 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
         int const num_experts_per_node, ActivationType fc1_activation_type, float const** alpha_scale_ptr_array,
         bool bias_is_broadcast, bool use_deepseek_fp8_block_scale, cudaStream_t stream,
         cutlass_extensions::CutlassGemmConfig config, bool min_latency_mode, int* num_active_experts_per,
-        int* active_expert_global_ids) override
+        int* active_expert_global_ids, int start_expert) override
     {
         auto* block_scale_gemm_runner = use_deepseek_fp8_block_scale ? getDeepSeekBlockScaleGemmRunner() : nullptr;
         return Self::gemm1(moe_gemm_runner_, block_scale_gemm_runner, static_cast<T const*>(input),
@@ -577,33 +603,33 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
             num_valid_tokens_ptr, static_cast<ScaleBiasType const*>(fc1_int_scales), fc1_fp8_dequant, fc2_fp8_quant,
             fc1_fp4_act_flat, fc2_fp4_act_flat, quant_params, num_rows, expanded_num_rows, hidden_size, inter_size,
             num_experts_per_node, fc1_activation_type, alpha_scale_ptr_array, bias_is_broadcast, stream, config,
-            min_latency_mode, num_active_experts_per, active_expert_global_ids);
+            min_latency_mode, num_active_experts_per, active_expert_global_ids, start_expert);
     }
 
     void gemm2(void const* const input, void* const gemm_output, void* const final_output,
         int64_t const* const expert_first_token_offset, TmaWarpSpecializedGroupedGemmInput const tma_ws_input_template,
         void const* const fc2_expert_weights, void const* const fc2_expert_biases, void const* const fc2_int_scales,
         float const* const fc2_fp8_dequant, TmaWarpSpecializedGroupedGemmInput::ElementSF const* fc2_fp4_act_flat,
         QuantParams quant_params, float const* const token_topk_unpermuted_scales,
-        float const* const token_topk_permuted_scales, int const* const unpermuted_row_to_permuted_row,
-        int const* permuted_row_to_unpermuted_row, int const* const token_selected_experts,
+        float const* const token_topk_permuted_scales, int const* const expanded_source_row_to_expanded_dest_row,
+        int const* expanded_dest_row_to_expanded_source_row, int const* const expert_for_source_row,
         int64_t const* const num_valid_tokens_ptr, int64_t const num_rows, int64_t const expanded_num_rows,
         int64_t const hidden_size, int64_t const inter_size, int const num_experts_per_node,
         int64_t const experts_per_token, float const** alpha_scale_ptr_array, bool use_lora, void* fc2_lora,
         bool use_deepseek_fp8_block_scale, cudaStream_t stream, MOEParallelismConfig parallelism_config,
         bool const enable_alltoall, cutlass_extensions::CutlassGemmConfig config, bool min_latency_mode,
-        int* num_active_experts_per, int* active_expert_global_ids) override
+        int* num_active_experts_per, int* active_expert_global_ids, int start_expert) override
     {
         auto* block_scale_gemm_runner = use_deepseek_fp8_block_scale ? getDeepSeekBlockScaleGemmRunner() : nullptr;
         return Self::gemm2(moe_gemm_runner_, block_scale_gemm_runner, static_cast<T const*>(input), gemm_output,
             static_cast<OutputType*>(final_output), expert_first_token_offset, tma_ws_input_template,
             static_cast<WeightType const*>(fc2_expert_weights), static_cast<ScaleBiasType const*>(fc2_expert_biases),
             static_cast<ScaleBiasType const*>(fc2_int_scales), fc2_fp8_dequant, fc2_fp4_act_flat, quant_params,
-            token_topk_unpermuted_scales, token_topk_permuted_scales, unpermuted_row_to_permuted_row,
-            permuted_row_to_unpermuted_row, token_selected_experts, num_valid_tokens_ptr, num_rows, expanded_num_rows,
-            hidden_size, inter_size, num_experts_per_node, experts_per_token, alpha_scale_ptr_array, use_lora, fc2_lora,
-            stream, parallelism_config, enable_alltoall, config, min_latency_mode, num_active_experts_per,
-            active_expert_global_ids);
+            token_topk_unpermuted_scales, token_topk_permuted_scales, expanded_source_row_to_expanded_dest_row,
+            expanded_dest_row_to_expanded_source_row, expert_for_source_row, num_valid_tokens_ptr, num_rows,
+            expanded_num_rows, hidden_size, inter_size, num_experts_per_node, experts_per_token, alpha_scale_ptr_array,
+            use_lora, fc2_lora, stream, parallelism_config, enable_alltoall, config, min_latency_mode,
+            num_active_experts_per, active_expert_global_ids, start_expert);
     }
 
     virtual size_t getGemmWorkspaceSize(int num_experts_per_node) const override
@@ -737,29 +763,30 @@ class CutlassMoeFCRunner : public CutlassMoeFCRunnerInterface
     static void BlockScaleFC2(DeepSeekBlockScaleGemmRunner& gemm_runner, T const* const input, void* const gemm_output,
         OutputType* const final_output, int64_t const* const expert_first_token_offset,
         WeightType const* const fc2_expert_weights, ScaleBiasType const* const fc2_expert_biases,
-        float const* const token_topk_unpermuted_scales, int const* const unpermuted_row_to_permuted_row,
-        int const* const permuted_row_to_unpermuted_row, int const* const token_selected_experts,
+        float const* const token_topk_unpermuted_scales, int const* const expanded_source_row_to_expanded_dest_row,
+        int const* const expanded_dest_row_to_expanded_source_row, int const* const expert_for_source_row,
         int64_t const* const num_valid_tokens_ptr, int64_t const num_rows, int64_t const expanded_num_rows,
-        int64_t const hidden_size, int64_t const inter_size, int64_t const num_experts_per_node, int64_t const k,
+        int64_t const hidden_size, int64_t const inter_size, int const num_experts_per_node, int64_t const k,
         MOEParallelismConfig parallelism_config, bool const enable_alltoall, QuantParams& quant_params,
         cudaStream_t stream);
 
     T const* applyPrequantScale(void* smoothed_act, void const* permuted_data, void const* prequant_scales,
         int64_t const* num_valid_tokens_ptr, int64_t const expanded_num_rows, int64_t const seq_len, bool const use_awq,
         cudaStream_t stream);
 
+    CubKeyValueSorter sorter_;
     MoeGemmRunner<T, WeightType, OutputType, ScaleBiasType> moe_gemm_runner_;
     std::unique_ptr<DeepSeekBlockScaleGemmRunner> blockscale_gemm_runner_;
 
     std::optional<cutlass_extensions::CutlassGemmConfig> gemm1_config_;
     std::optional<cutlass_extensions::CutlassGemmConfig> gemm2_config_;
 
     // Pointers
-    int* permuted_row_to_unpermuted_row_{};
+    int* unpermuted_token_selected_experts_{};
+    int* unpermuted_source_token_ids_{};
+    int* permuted_source_token_ids_{};
     int* permuted_token_selected_experts_{};
-    int* blocked_expert_counts_{};
-    int* blocked_expert_counts_cumsum_{};
-    int* blocked_row_to_unpermuted_row_{};
+    char* sorter_ws_{};
     T* permuted_data_{};
     float* permuted_token_final_scales_{};
 
@@ -832,6 +859,7 @@ struct GemmProfilerBackend
         mParallelismConfig = parallelism_config;
         mEnableAlltoall = enable_alltoall;
         mSM = common::getSMVersion();
+        mSorter.updateNumExperts(mNumExpertsPerNode);
 
         mScalingType = TmaWarpSpecializedGroupedGemmInput::FpXBlockScalingType::NONE;
         if (dtype == nvinfer1::DataType::kFP8
@@ -855,6 +883,7 @@ struct GemmProfilerBackend
         cudaStream_t const& stream);
 
     CutlassMoeFCRunnerInterface* mInterface;
+    CubKeyValueSorter mSorter;
 
     GemmToProfile mGemmToProfile = GemmToProfile::Undefined;
     std::vector<Config> mAllTacticsSaved;