Disable ActiveWorkgroupsPerCU for different arch in wmma kernels (#3566)

Author: bartekxk

include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_wmma_cshuffle_v3.hpp

@@ -314,6 +314,10 @@ struct DeviceBatchedGemmMultiD_Wmma_CShuffleV3
     {
         ActiveWorkgroupsPerCU()
         {
+            if(!ck::is_gfx11_supported() && !ck::is_gfx12_supported())
+            {
+                return;
+            }
             constexpr int dynamic_smem_size = 0;
             int max_occupancy               = 0;
 

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_two_stage_wmma_cshuffle_v3.hpp

@@ -466,6 +466,10 @@ struct DeviceGroupedConvBwdWeightTwoStage_Wmma_CShuffleV3
     {
         ActiveWorkgroupsPerCU()
         {
+            if(!ck::is_gfx11_supported() && !ck::is_gfx12_supported())
+            {
+                return;
+            }
             constexpr int dynamic_smem_size = 0;
             constexpr index_t minimum_occupancy =
                 BlkGemmPipeSched == BlockGemmPipelineScheduler::Intrawave ? 1 : 2;

include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle_v3.hpp

@@ -415,6 +415,10 @@ struct DeviceGroupedConvBwdWeight_Wmma_CShuffleV3
     {
         ActiveWorkgroupsPerCU()
         {
+            if(!ck::is_gfx11_supported() && !ck::is_gfx12_supported())
+            {
+                return;
+            }
             constexpr int dynamic_smem_size = 0;
             constexpr index_t minimum_occupancy =
                 BlkGemmPipeSched == BlockGemmPipelineScheduler::Intrawave ? 1 : 2;