Only launch splitkv combine kernel when necessary

example/ck_tile/01_fmha/fmha_fwd.cpp

@@ -11,6 +11,7 @@
 #include <array>
 #include <cstring>
 #include <functional>
+#include <map>
 #include <numeric>
 #include <ostream>
 #include <string>
@@ -176,61 +177,14 @@ auto get_elimit<FmhaFwdFp8>(std::string init_method)
     }
 }
 
-int num_splits_heuristic(int batch_nhead_mblocks, int num_SMs, int num_n_blocks, int max_splits)
-{
-    // If we have enough to almost fill the SMs, then just use 1 split
-    if(batch_nhead_mblocks >= 0.8f * num_SMs)
-    {
-        return 1;
-    }
-    max_splits           = std::min({max_splits, num_SMs, num_n_blocks});
-    float max_efficiency = 0.f;
-    std::vector<float> efficiency;
-    efficiency.reserve(max_splits);
-    auto ceildiv = [](int a, int b) { return (a + b - 1) / b; };
-    // Some splits are not eligible. For example, if we have 64 blocks and choose 11 splits,
-    // we'll have 6 * 10 + 4 blocks. If we choose 12 splits, we'll have 6 * 11 + (-2) blocks
-    // (i.e. it's 11 splits anyway).
-    // So we check if the number of blocks per split is the same as the previous num_splits.
-    auto is_split_eligible = [&ceildiv, &num_n_blocks](int num_splits) {
-        return num_splits == 1 ||
-               ceildiv(num_n_blocks, num_splits) != ceildiv(num_n_blocks, num_splits - 1);
-    };
-    for(int num_splits = 1; num_splits <= max_splits; num_splits++)
-    {
-        if(!is_split_eligible(num_splits))
-        {
-            efficiency.push_back(0.f);
-        }
-        else
-        {
-            float n_waves = float(batch_nhead_mblocks * num_splits) / num_SMs;
-            float eff     = n_waves / ceil(n_waves);
-            // printf("num_splits = %d, eff = %f\n", num_splits, eff);
-            if(eff > max_efficiency)
-            {
-                max_efficiency = eff;
-            }
-            efficiency.push_back(eff);
-        }
-    }
-    for(int num_splits = 1; num_splits <= max_splits; num_splits++)
-    {
-        if(!is_split_eligible(num_splits))
-        {
-            continue;
-        }
-        if(efficiency[num_splits - 1] >= 0.85 * max_efficiency)
-        {
-            // printf("num_splits chosen = %d\n", num_splits);
-            return num_splits;
-        }
-    }
-    return 1;
-}
-
-int override_num_splits_if_necessary(
-    int batch, int nhead, int max_seqlen_q, int hdim_v, float p_drop, int num_splits)
+int override_num_splits_if_necessary(int batch,
+                                     int nhead,
+                                     int max_seqlen_q,
+                                     int hdim_q,
+                                     int hdim_v,
+                                     float p_drop,
+                                     bool is_prefill,
+                                     int num_splits)
 {
     int device;
     auto status = hipGetDevice(&device);
@@ -246,17 +200,41 @@ int override_num_splits_if_necessary(
         return num_splits;
     }
 
-    // tile size should match the generate.py
-    const int kM0 = 64;
-    const int kN1 = hdim_v;
+    const int kM0 = [&] {
+        // get kM0 for prefill phase
+        if(is_prefill)
+        {
+            return 128;
+        }
+
+        // get kM0 for decode phase
+        /// TODO: take dtype=fp8/bf8 into consideration
+        const std::map<int, int> hdim_to_m0 = {
+            {32, 32},
+            {64, 64},
+            // {96,  64},
+            {128, 64},
+            {256, 64},
+        };
+
+        for(auto [hdim, m0] : hdim_to_m0)
+        {
+            if(hdim_q <= hdim && hdim_v <= hdim)
+            {
+                return m0;
+            }
+        }
+
+        return 64; // meet unsupported hdim_q/hdim_v
+    }();
+    // const int kN1 = hdim_v;
 
     const int num_m_blocks = ck_tile::integer_divide_ceil(max_seqlen_q, kM0);
-    const int num_n_blocks = ck_tile::integer_divide_ceil(hdim_v, kN1);
+    // const int num_n_blocks = ck_tile::integer_divide_ceil(hdim_v, kN1); // always 1
 
     if(num_splits < 1 && p_drop == 0.0f)
     {
-        return num_splits_heuristic(
-            batch * nhead * num_m_blocks, props.multiProcessorCount * 2, num_n_blocks, 128);
+        return num_splits_heuristic(batch * nhead * num_m_blocks, props.multiProcessorCount * 2, 8);
     }
 
     return num_splits;
@@ -556,8 +534,15 @@ bool run(const ck_tile::ArgParser& arg_parser)
     // legalize num_splits according to other options
     if(num_splits < 1)
     {
-        num_splits = override_num_splits_if_necessary(
-            batch, nhead, max_seqlen_q, hdim_v, p_drop, num_splits);
+        num_splits = override_num_splits_if_necessary(batch,
+                                                      nhead,
+                                                      max_seqlen_q,
+                                                      hdim_q,
+                                                      hdim_v,
+                                                      p_drop,
+                                                      /*is_prefill=*/mode == mode_enum::group &&
+                                                          0 < page_block_size,
+                                                      num_splits);
     }
     if(128 < num_splits)
     {
@@ -632,12 +617,13 @@ bool run(const ck_tile::ArgParser& arg_parser)
     auto [rotary_cos_host, rotary_sin_host] = generate_rotary_cos_sin<KDataType>(
         std::max(shape_seqlen_q, shape_seqlen_k), rotary_dim, seed);
 
+    // lse_acc_host & o_acc_host are only used when 1 < num_spilts
     ck_tile::HostTensor<LSEDataType> lse_acc_host(
-        1 < num_splits || use_kvcache
+        1 < num_splits
             ? std::array<ck_tile::index_t, 4>{shape_batch, nhead, num_splits, shape_seqlen_q}
             : std::array<ck_tile::index_t, 4>{1, 1, 1, 1});
     ck_tile::HostTensor<OaccDataType> o_acc_host(
-        1 < num_splits || use_kvcache ? std::array<ck_tile::index_t, 5>{shape_batch,
+        1 < num_splits ? std::array<ck_tile::index_t, 5>{shape_batch,
                                                          nhead,
                                                          num_splits,
                                                          shape_seqlen_q,
@@ -1043,9 +1029,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
             }
             else if constexpr(std::is_same_v<fmha_fwd_splitkv_args, std::decay_t<decltype(args)>>)
             {
-                args.lse_acc_ptr = lse_acc_buf.GetDeviceBuffer();
-                args.o_acc_ptr   = o_acc_buf.GetDeviceBuffer();
-
+                // lse_acc_buf & o_acc_buf are only used when 1 < num_spilts
                 args.block_table_ptr =
                     (0 < page_block_size ? block_table_buf.GetDeviceBuffer() : nullptr);
                 args.batch_stride_block_table = batch_stride_block_table;
@@ -1057,6 +1041,11 @@ bool run(const ck_tile::ArgParser& arg_parser)
 
                 args.num_splits = num_splits;
 
+                if(1 < num_splits)
+                {
+                    args.lse_acc_ptr = lse_acc_buf.GetDeviceBuffer();
+                    args.o_acc_ptr   = o_acc_buf.GetDeviceBuffer();
+
                     args.stride_o_acc         = stride_o_acc;
                     args.nhead_stride_lse_acc = nhead_stride_lse_acc;
                     args.nhead_stride_o_acc   = nhead_stride_o_acc;
@@ -1065,6 +1054,21 @@ bool run(const ck_tile::ArgParser& arg_parser)
                     args.split_stride_lse_acc = split_stride_lse_acc;
                     args.split_stride_o_acc   = split_stride_o_acc;
                 }
+                else
+                {
+                    // following attribues are ignored by fmha_fwd_splitkv()
+                    args.lse_acc_ptr = nullptr;
+                    args.o_acc_ptr   = nullptr;
+
+                    args.stride_o_acc         = 0;
+                    args.nhead_stride_lse_acc = 0;
+                    args.nhead_stride_o_acc   = 0;
+                    args.batch_stride_lse_acc = 0;
+                    args.batch_stride_o_acc   = 0;
+                    args.split_stride_lse_acc = 0;
+                    args.split_stride_o_acc   = 0;
+                }
+            }
         }
     };
 

example/ck_tile/01_fmha/fmha_fwd.hpp

@@ -12,6 +12,7 @@
 #include "mask.hpp"
 #include "rotary.hpp"
 
+#include <array>
 #include <type_traits>
 #include <utility>
 #include <variant>
@@ -422,12 +423,13 @@ auto fmha_fwd_splitkv_create_kargs_and_grids(fmha_fwd_splitkv_args args)
         // create group mode kernel arguments
         if constexpr(Kernel::kIsGroupMode)
         {
-            return Kernel::MakeKargs(args.q_ptr,
+            return Kernel::MakeKargs(
+                args.q_ptr,
                 args.k_ptr,
                 args.v_ptr,
                 args.bias_ptr,
-                                     args.lse_acc_ptr,
-                                     args.o_acc_ptr,
+                (1 < args.num_splits ? args.lse_acc_ptr : args.lse_ptr),
+                (1 < args.num_splits ? args.o_acc_ptr : args.o_ptr),
                 args.batch,
                 args.seqstart_q_ptr,
                 args.seqstart_k_ptr,
@@ -447,29 +449,30 @@ auto fmha_fwd_splitkv_create_kargs_and_grids(fmha_fwd_splitkv_args args)
                 args.stride_k,
                 args.stride_v,
                 args.stride_bias,
-                                     args.stride_o_acc,
+                (1 < args.num_splits ? args.stride_o_acc : args.stride_o),
                 args.nhead_stride_q,
                 args.nhead_stride_k,
                 args.nhead_stride_v,
                 args.nhead_stride_bias,
-                                     args.nhead_stride_lse_acc,
-                                     args.nhead_stride_o_acc,
+                (1 < args.num_splits ? args.nhead_stride_lse_acc : args.nhead_stride_lse),
+                (1 < args.num_splits ? args.nhead_stride_o_acc : args.nhead_stride_o),
                 args.batch_stride_k, // only used for paged-kvcache
                 args.batch_stride_v, // only used for paged-kvcache
-                                     args.split_stride_lse_acc,
-                                     args.split_stride_o_acc,
+                (1 < args.num_splits ? args.split_stride_lse_acc : 0),
+                (1 < args.num_splits ? args.split_stride_o_acc : 0),
                 args.window_size_left,
                 args.window_size_right,
                 args.mask_type);
         }
         else
         { // create batch mode kernel arguments
-            return Kernel::MakeKargs(args.q_ptr,
+            return Kernel::MakeKargs(
+                args.q_ptr,
                 args.k_ptr,
                 args.v_ptr,
                 args.bias_ptr,
-                                     args.lse_acc_ptr,
-                                     args.o_acc_ptr,
+                (1 < args.num_splits ? args.lse_acc_ptr : args.lse_ptr),
+                (1 < args.num_splits ? args.o_acc_ptr : args.o_ptr),
                 args.batch,
                 args.seqlen_q,
                 args.seqlen_k,
@@ -489,21 +492,21 @@ auto fmha_fwd_splitkv_create_kargs_and_grids(fmha_fwd_splitkv_args args)
                 args.stride_k,
                 args.stride_v,
                 args.stride_bias,
-                                     args.stride_o_acc,
+                (1 < args.num_splits ? args.stride_o_acc : args.stride_o),
                 args.nhead_stride_q,
                 args.nhead_stride_k,
                 args.nhead_stride_v,
                 args.nhead_stride_bias,
-                                     args.nhead_stride_lse_acc,
-                                     args.nhead_stride_o_acc,
+                (1 < args.num_splits ? args.nhead_stride_lse_acc : args.nhead_stride_lse),
+                (1 < args.num_splits ? args.nhead_stride_o_acc : args.nhead_stride_o),
                 args.batch_stride_q,
                 args.batch_stride_k,
                 args.batch_stride_v,
                 args.batch_stride_bias,
-                                     args.batch_stride_lse_acc,
-                                     args.batch_stride_o_acc,
-                                     args.split_stride_lse_acc,
-                                     args.split_stride_o_acc,
+                (1 < args.num_splits ? args.batch_stride_lse_acc : args.batch_stride_lse),
+                (1 < args.num_splits ? args.batch_stride_o_acc : args.batch_stride_o),
+                (1 < args.num_splits ? args.split_stride_lse_acc : 0),
+                (1 < args.num_splits ? args.split_stride_o_acc : 0),
                 args.window_size_left,
                 args.window_size_right,
                 args.mask_type);
@@ -821,3 +824,40 @@ struct fmha_fwd_appendkv_traits
 float fmha_fwd_appendkv(fmha_fwd_appendkv_traits,
                         fmha_fwd_appendkv_args,
                         const ck_tile::stream_config&);
+
+template <typename Int = int>
+Int num_splits_heuristic(Int batch_nhead_mblocks, Int num_SMs, Int max_splits)
+{
+    // If we have enough to almost fill the SMs, then just use 1 split
+    if(batch_nhead_mblocks >= 0.8f * num_SMs)
+    {
+        return 1;
+    }
+
+    max_splits = std::min({max_splits, num_SMs});
+
+    constexpr std::array<Int, 5> num_splits_array = {1, 2, 4, 8, 16};
+
+    float max_efficiency = 0.f;
+    std::array<float, num_splits_array.size()> efficiency;
+
+    for(size_t idx = 0; idx < num_splits_array.size() && num_splits_array[idx] <= max_splits; ++idx)
+    {
+        float n_blocks = float(batch_nhead_mblocks * num_splits_array[idx]) / num_SMs;
+        float eff      = n_blocks / std::ceil(n_blocks);
+
+        if(eff > max_efficiency)
+        {
+            max_efficiency = eff;
+        }
+        efficiency[idx] = eff;
+    }
+    for(size_t idx = 0; idx < num_splits_array.size() && num_splits_array[idx] <= max_splits; ++idx)
+    {
+        if(efficiency[idx] >= 0.85 * max_efficiency)
+        {
+            return num_splits_array[idx];
+        }
+    }
+    return 1;
+}