Merge branch 'develop' into amd-develop

Dockerfile

@@ -23,11 +23,11 @@ RUN if [ "$ROCMVERSION" != "6.2" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
         sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
         sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
-    elif [ "$ROCMVERSION" = "6.2" ] && [ "$compiler_version" = "rc2" ]; then \
-        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_6.1-20.04-1_all.deb --no-check-certificate" && \
-        apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install dialog && DEBIAN_FRONTEND=noninteractive apt-get install ./amdgpu-install-internal_6.1-20.04-1_all.deb && \
-        sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 6.1 rel-48 > /etc/apt/sources.list.d/rocm-build.list' && \
-        amdgpu-repo --amdgpu-build=1736298; \
+    elif [ "$ROCMVERSION" = "6.2" ] && [ "$compiler_version" = "rc1" ]; then \
+        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_6.2-20.04-1_all.deb --no-check-certificate" && \
+        apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install dialog libpopt0 rsync && DEBIAN_FRONTEND=noninteractive apt-get install ./amdgpu-install-internal_6.2-20.04-1_all.deb && \
+        sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 6.2 rel-8 > /etc/apt/sources.list.d/rocm-build.list' && \
+        amdgpu-repo --amdgpu-build=1794148; \
     fi
 
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"

Jenkinsfile

@@ -886,10 +886,10 @@ pipeline {
                     }
                     agent{ label rocmnode("gfx90a") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS=" -O3 " """
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1100;gfx90a" -DCMAKE_CXX_FLAGS=" -O3 " """
                         execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
                                            cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
-                                           -DGPU_TARGETS="gfx908;gfx90a" \
+                                           -DGPU_TARGETS="gfx1100;gfx90a" \
                                            -DCMAKE_CXX_COMPILER="${build_compiler()}" \
                                            -DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """
                     }

client_example/25_wrapper/wrapper_basic_gemm.cpp

@@ -7,19 +7,23 @@
 #include <initializer_list>
 #include <vector>
 
+#include "ck/utility/common_header.hpp"
+// __gfx9__ defined in the above header via ck.hpp
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
+
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/utility/host_tensor.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/check_err.hpp"
-#include "ck/utility/common_header.hpp"
 #include "ck/library/utility/fill.hpp"
+#include "ck/library/utility/host_tensor.hpp"
 #include "ck/wrapper/layout.hpp"
 #include "ck/wrapper/tensor.hpp"
 #include "ck/wrapper/operations/copy.hpp"
 #include "ck/wrapper/operations/gemm.hpp"
 #include "ck/wrapper/utils/kernel_utils.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 struct SimpleDeviceMem
 {
@@ -204,6 +208,14 @@ void PerformGemm(const ck::index_t M,
 
 int main(int argc, char* argv[])
 {
+    bool is_supported = ck::is_xdl_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: xdl example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+
     using DataType = ck::half_t;
     const auto thread_layout =
         ck::wrapper::make_layout(ck::make_tuple(ck::Number<64>{}, ck::Number<4>{}),
@@ -213,3 +225,4 @@ int main(int argc, char* argv[])
         3840, 4096, 4096, tile_shape, thread_layout);
     return 0;
 }
+#endif

client_example/25_wrapper/wrapper_optimized_gemm.cpp

@@ -7,18 +7,21 @@
 #include <initializer_list>
 #include <vector>
 
-#include "ck/library/utility/host_tensor.hpp"
+#include "ck/utility/common_header.hpp"
+// __gfx9__ defined in the above header via ck.hpp
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
 
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/check_err.hpp"
-#include "ck/utility/common_header.hpp"
 #include "ck/library/utility/fill.hpp"
+#include "ck/library/utility/host_tensor.hpp"
 #include "ck/wrapper/layout.hpp"
 #include "ck/wrapper/tensor.hpp"
 #include "ck/wrapper/operations/copy.hpp"
 #include "ck/wrapper/operations/gemm.hpp"
 #include "ck/wrapper/utils/kernel_utils.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 struct SimpleDeviceMem
 {
@@ -296,6 +299,14 @@ void PerformGemm(const ck::index_t M,
 
 int main(int argc, char* argv[])
 {
+    bool is_supported = ck::is_xdl_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: xdl example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+
     using DataType = ck::half_t;
     const auto thread_layout =
         ck::wrapper::make_layout(ck::make_tuple(ck::Number<4>{}, ck::Number<64>{}, ck::Number<1>{}),
@@ -305,3 +316,4 @@ int main(int argc, char* argv[])
         3840, 4096, 4096, tile_shape, thread_layout);
     return 0;
 }
+#endif

docs/sphinx/requirements.in

-rocm-docs-core==1.4.0
+rocm-docs-core==1.4.1
 sphinxcontrib-bibtex==2.6.2

docs/sphinx/requirements.txt

@@ -103,7 +103,7 @@ requests==2.31.0
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==1.4.0
+rocm-docs-core==1.4.1
     # via -r requirements.in
 six==1.16.0
     # via

example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp

@@ -17,6 +17,7 @@
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 #include "ck/library/utility/check_err.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 struct AlphaBetaAdd
 {
@@ -175,6 +176,14 @@ int main(int argc, char* argv[])
         exit(0);
     }
 
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
             using namespace ck::literals;

example/02_gemm_bilinear/gemm_bilinear_wmma_int8.cpp

@@ -17,6 +17,7 @@
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 #include "ck/library/utility/check_err.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 struct AlphaBetaAdd
 {
@@ -175,6 +176,14 @@ int main(int argc, char* argv[])
         exit(0);
     }
 
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
             using namespace ck::literals;

example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp

@@ -2,6 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common_wmma.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 // kernel data types
 using InKernelDataType       = FP16;
@@ -23,4 +24,14 @@ using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
 
 #include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv);
+}

example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp

@@ -2,6 +2,7 @@
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common_wmma.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 // kernel data types
 using InKernelDataType       = I8;
@@ -23,4 +24,14 @@ using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
 
 #include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16.cpp

@@ -27,6 +27,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_n = Softmax(A_g_m_k * B0_g
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -163,4 +164,14 @@ using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm<
 
 #include "run_batched_gemm_scale_softmax_gemm_permute_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp

@@ -27,6 +27,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_n = Softmax(A_g_m_k * B0_g
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -285,4 +286,14 @@ using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm<
 
 #include "run_batched_gemm_scale_softmax_gemm_permute_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/cross_attention_forward_wmma_fp16.cpp

@@ -27,6 +27,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_n = Softmax(A_g_m_k * B0_g
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -351,4 +352,14 @@ using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm<
 
 #include "run_cross_attention_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/grouped_query_attention_forward_wmma_fp16.cpp

@@ -28,6 +28,7 @@ Example is GQA-4
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -299,4 +300,14 @@ using ReferenceGemm1Instance =
 
 #include "run_grouped_query_attention_forward_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/multi_query_attention_forward_wmma_fp16.cpp

@@ -26,6 +26,7 @@ Shazeer, Noam. “Fast Transformer Decoding: One Write-Head Is All You Need.”
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -284,4 +285,14 @@ using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm_
 
 #include "run_multi_query_attention_forward_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/32_batched_gemm_scale_softmax_gemm/self_attention_forward_wmma_fp16.cpp

@@ -27,6 +27,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_n = Softmax(A_g_m_k * B0_g
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -329,4 +330,14 @@ using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm<
 
 #include "run_self_attention_wmma.inc"
 
-int main(int argc, char* argv[]) { return run(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run(argc, argv);
+}

example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp"
 #include "common.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 using OutDataType      = FP16;
 using WeiDataType      = FP16;
@@ -31,4 +32,14 @@ using DeviceConvInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDat
 
 #include "run_grouped_conv_bwd_data_example.inc"
 
-int main(int argc, char* argv[]) { return run_grouped_conv_bwd_data_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    bool is_supported = ck::is_gfx11_supported();
+    if(!is_supported)
+    {
+        std::cout << "WARNING: wmma example not supported on the platform " << ck::get_device_name()
+                  << std::endl;
+        return 0;
+    }
+    return run_grouped_conv_bwd_data_example(argc, argv);
+}

example/CMakeLists.txt

@@ -181,7 +181,7 @@ endfunction(add_example_executable_no_testing EXAMPLE_NAME)
 # add all example subdir
 file(GLOB dir_list LIST_DIRECTORIES true *)
 FOREACH(subdir ${dir_list})
-    IF(IS_DIRECTORY "${subdir}")
+    if(IS_DIRECTORY "${subdir}" AND EXISTS "${subdir}/CMakeLists.txt")
         add_subdirectory(${subdir})
     ENDIF()
 ENDFOREACH()

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

@@ -47,12 +47,12 @@ __global__ void
 #endif
         kernel_grouped_conv_bwd_weight_xdl_cshuffle_v3(
             typename GridwiseGemm::Argument karg,
-            const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
-            const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
-            const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            [[maybe_unused]] const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
+            [[maybe_unused]] const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
+            [[maybe_unused]] const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
                 c_grid_desc_mblock_mperblock_nblock_nperblock,
-            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
-            const index_t num_k_per_block)
+            [[maybe_unused]] const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
+            [[maybe_unused]] const index_t num_k_per_block)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx94__))
@@ -103,12 +103,12 @@ __global__ void
 #endif
         kernel_grouped_conv_bwd_weight_xdl_cshuffle_v3_2lds(
             typename GridwiseGemm::Argument karg,
-            const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
-            const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
-            const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            [[maybe_unused]] const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
+            [[maybe_unused]] const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
+            [[maybe_unused]] const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
                 c_grid_desc_mblock_mperblock_nblock_nperblock,
-            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
-            const index_t num_k_per_block)
+            [[maybe_unused]] const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
+            [[maybe_unused]] const index_t num_k_per_block)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_abd_xdl_cshuffle_v3.hpp

@@ -69,14 +69,15 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
 #endif
-        kernel_grouped_conv_fwd_xdl_cshuffle_v3(typename GridwiseGemm::Argument karg,
-                                                const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
-                                                const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
-                                                const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
-                                                    c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                const ComputePtrOffset compute_ptr_offset_of_groups,
-                                                const ComputePtrOffset compute_ptr_offset_of_n,
-                                                const index_t groups_count)
+        kernel_grouped_conv_fwd_xdl_cshuffle_v3(
+            typename GridwiseGemm::Argument karg,
+            [[maybe_unused]] const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
+            [[maybe_unused]] const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
+            [[maybe_unused]] const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+                c_grid_desc_mblock_mperblock_nblock_nperblock,
+            [[maybe_unused]] const ComputePtrOffset compute_ptr_offset_of_groups,
+            [[maybe_unused]] const ComputePtrOffset compute_ptr_offset_of_n,
+            [[maybe_unused]] const index_t groups_count)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
     // offset base pointer for each work-group
@@ -132,13 +133,13 @@ __global__ void
 #endif
         kernel_grouped_conv_fwd_xdl_cshuffle_v3_2lds(
             typename GridwiseGemm::Argument karg,
-            const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
-            const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
-            const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            [[maybe_unused]] const AGridDesc_AK0_M_K1 a_grid_desc_ak0_m_ak1,
+            [[maybe_unused]] const BGridDesc_BK0_N_K1 b_grid_desc_bk0_n_bk1,
+            [[maybe_unused]] const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
                 c_grid_desc_mblock_mperblock_nblock_nperblock,
-            const ComputePtrOffset compute_ptr_offset_of_groups,
-            const ComputePtrOffset compute_ptr_offset_of_n,
-            const index_t groups_count)
+            [[maybe_unused]] const ComputePtrOffset compute_ptr_offset_of_groups,
+            [[maybe_unused]] const ComputePtrOffset compute_ptr_offset_of_n,
+            [[maybe_unused]] const index_t groups_count)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx9__))
     // offset base pointer for each work-group