update the target criteria

include/ck/tensor_operation/gpu/device/device_batched_gemm_xdl.hpp

@@ -46,7 +46,7 @@ __global__ void
             const ComputeBasePrtOfBatch compute_base_ptr_of_batch_,
             const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp

@@ -38,7 +38,7 @@ __global__ void
                                         c_grid_desc_mblock_mperblock_nblock_nperblock,
                                     const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp

@@ -39,7 +39,7 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (!__HIP_DEVICE_COMPILE__  || defined(__gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
@@ -76,7 +76,7 @@ __global__ void
             const BElementwiseOperation b_element_op,
             const CElementwiseOperation c_element_op)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (!__HIP_DEVICE_COMPILE__  || defined(__gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp

@@ -37,7 +37,7 @@ __global__ void
                                 const CElementwiseOperation c_element_op,
                                 const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp

@@ -39,7 +39,7 @@ __global__ void
                                   const CElementwiseOperation c_element_op,
                                   const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined!__HIP_DEVICE_COMPILE__  || (__gfx908__) || defined(__gfx90a__))
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp

@@ -42,7 +42,7 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp

@@ -45,7 +45,7 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp

@@ -49,7 +49,7 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if (defined(__gfx908__) || defined(__gfx90a__))
+#if (defined(!__HIP_DEVICE_COMPILE__  || __gfx908__) || defined(__gfx90a__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(