LWPCK-2429: Device grouped GEMM uses Async Memcpy

Resolving merge conflicts

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

 #pragma once
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -603,11 +603,11 @@ struct DeviceGroupedGemmMultipleD_Dl : public DeviceGroupedGemm<ALayout,
             }
 
             hipGetErrorString(
-                hipMemcpyWithStream(arg.p_workspace_,
-                                    arg.gemm_desc_kernel_arg_.data(),
-                                    arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
-                                    hipMemcpyHostToDevice,
-                                    stream_config.stream_id_));
+                hipMemcpyAsync(arg.p_workspace_,
+                               arg.gemm_desc_kernel_arg_.data(),
+                               arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
+                               hipMemcpyHostToDevice,
+                               stream_config.stream_id_));
 
             auto launch_kernel = [&](auto has_main_k_block_loop,
                                      auto has_double_tail_k_block_loop) {

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

@@ -761,11 +761,11 @@ struct DeviceGroupedGemmMultipleDSplitKXdlCShuffleTwoStage
             float time{0.f};
 
             hip_check_error(
-                hipMemcpyWithStream(dev_gemm_kargs,
-                                    arg.gemm_kernel_args_.data(),
-                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
-                                    hipMemcpyHostToDevice,
-                                    stream_config.stream_id_));
+                hipMemcpyAsync(dev_gemm_kargs,
+                               arg.gemm_kernel_args_.data(),
+                               arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                               hipMemcpyHostToDevice,
+                               stream_config.stream_id_));
 
             auto preprocess = [&]() {
                 hip_check_error(hipMemsetAsync(

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

@@ -940,10 +940,10 @@ struct DeviceGroupedGemmMultipleDXdlCShuffleTileLoop
                              const void* p_host_kernel_args) const
     {
         arg.p_dev_gemm_args_ = p_dev_kernel_args;
-        hip_check_error(hipMemcpy(p_dev_kernel_args,
-                                  p_host_kernel_args,
-                                  GetDeviceKernelArgSize(&arg),
-                                  hipMemcpyHostToDevice));
+        hip_check_error(hipMemcpyAsync(p_dev_kernel_args,
+                                       p_host_kernel_args,
+                                       GetDeviceKernelArgSize(&arg),
+                                       hipMemcpyHostToDevice));
     }
 
     virtual void SetDeviceKernelArgs(BaseArgument* p_arg,

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

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

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

@@ -557,12 +557,12 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
                 }
             }
 
-            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
-                                                  arg.gemm_desc_kernel_arg_.data(),
-                                                  arg.gemm_desc_kernel_arg_.size() *
-                                                      sizeof(GemmBiasTransKernelArg),
-                                                  hipMemcpyHostToDevice,
-                                                  stream_config.stream_id_));
+            hipGetErrorString(
+                hipMemcpyAsync(arg.p_workspace_,
+                               arg.gemm_desc_kernel_arg_.data(),
+                               arg.gemm_desc_kernel_arg_.size() * sizeof(GemmBiasTransKernelArg),
+                               hipMemcpyHostToDevice,
+                               stream_config.stream_id_));
 
             float ave_time = 0;
 

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

@@ -421,11 +421,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
             }
 
             hip_check_error(
-                hipMemcpyWithStream(arg.p_workspace_,
-                                    arg.gemm_kernel_args_.data(),
-                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
-                                    hipMemcpyHostToDevice,
-                                    stream_config.stream_id_));
+                hipMemcpyAsync(arg.p_workspace_,
+                               arg.gemm_kernel_args_.data(),
+                               arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                               hipMemcpyHostToDevice,
+                               stream_config.stream_id_));
 
             float ave_time = 0;
 

profiler/include/profiler/profile_grouped_gemm_impl.hpp

@@ -302,6 +302,13 @@ bool profile_grouped_gemm_impl(int do_verification,
                                                                   rtol,
                                                                   atol);
 
+                        instance_pass =
+                            instance_pass && ck::utils::check_err(c_m_n_device_results[i],
+                                                                  c_m_n_host_results[i],
+                                                                  "Error: Incorrect results!",
+                                                                  rtol,
+                                                                  atol);
+
                         if(do_log)
                         {
                             LogRangeAsType<float>(std::cout << "a : ", a_m_k[i].mData, ",")