add a tuning instance

example/66_gemm_bias_activation/gemm_bias_add_fp16.cpp

@@ -36,7 +36,7 @@ static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecializatio
 
 // clang-format off
 template <typename ADataType,   typename BDataType, typename DsDataType,  typename CDataType, typename CDEElementOp>
-using DeviceOpInstance_64_16_16_64 = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3<
+using DeviceOpInstance_128_32_64_64 = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3<
         ALayout,  BLayout, DsLayout, CLayout, ADataType, BDataType,
         DsDataType, CDataType, AccDataType, CShuffleDataType,
         AElementOp,  BElementOp, CDEElementOp,       GemmSpec,
@@ -52,6 +52,24 @@ using DeviceOpInstance_64_16_16_64 = ck::tensor_operation::device::DeviceGemmMul
         1,           1,
         S<1, 16, 1, 8>,      S<8, 8>,  ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v2, F16>;
 
+template <typename ADataType,   typename BDataType, typename DsDataType,  typename CDataType, typename CDEElementOp>
+using DeviceOpInstance_256_128_128_64 = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3<
+        ALayout,  BLayout, DsLayout, CLayout, ADataType, BDataType,
+        DsDataType, CDataType, AccDataType, CShuffleDataType,
+        AElementOp,  BElementOp, CDEElementOp,       GemmSpec,
+        256,  
+        128,   128,    64,  
+        8,   4,  
+        32,   32,    
+        2,    2,    
+        S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,              
+        2,              8,              8,          0,    
+        S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,            
+        1,              8,              4,          0,          
+        1,           1,                  
+        S<1, 32, 1, 8>,          S<8, 8>,  
+        ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, F16>;
+
 template <typename ADataType,   typename BDataType, typename DsDataType,  typename CDataType, typename CDEElementOp>
 using DeviceOpInstance_default = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3<
         ALayout,  BLayout, DsLayout, CLayout, ADataType, BDataType,
@@ -128,14 +146,32 @@ float run_impl(const GemmBiasAddArgs& args, const StreamConfig& config)
         return true;
     };
 
-    auto gemm =
-        DeviceOpInstance_64_16_16_64<ADataType, BDataType, DsDataType, CDataType, CDEElementOp>{};
-    if(!Run(gemm))
+    do
     {
+        if(args.M <= 512)
+        {
+            auto gemm = DeviceOpInstance_128_32_64_64<ADataType,
+                                                      BDataType,
+                                                      DsDataType,
+                                                      CDataType,
+                                                      CDEElementOp>{};
+            if(Run(gemm))
+                break;
+        }
+        else
+        {
+            auto gemm = DeviceOpInstance_256_128_128_64<ADataType,
+                                                        BDataType,
+                                                        DsDataType,
+                                                        CDataType,
+                                                        CDEElementOp>{};
+            if(Run(gemm))
+                break;
+        }
         auto gemm_def =
             DeviceOpInstance_default<ADataType, BDataType, DsDataType, CDataType, CDEElementOp>{};
         Run(gemm_def);
-    }
+    } while(0);
 
     return ave_time;
 }