enable fwd conv on navi4x

example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt

 list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942 gfx950)
-list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
+list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102 gfx1200)
 
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)

example/30_grouped_conv_fwd_multiple_d/common.hpp

@@ -93,7 +93,7 @@ struct ExecutionConfig final
 #define DefaultConvParam                                                     \
     ck::utils::conv::ConvParam                                               \
     {                                                                        \
-        2, 32, 2, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, { 1, 1 } \
+        2, 32, 2, 32, 32, {3, 3}, {14, 14}, {2, 2}, {1, 1}, {1, 1}, { 1, 1 } \
     }
 
 inline void print_help_msg()

example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp

@@ -93,7 +93,7 @@ struct ExecutionConfig final
 #define DefaultConvParam                                                     \
     ck::utils::conv::ConvParam                                               \
     {                                                                        \
-        2, 32, 2, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, { 1, 1 } \
+        2, 32, 2, 32, 32, {3, 3}, {14, 14}, {2, 2}, {1, 1}, {1, 1}, { 1, 1 } \
     }
 
 inline void print_help_msg()

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

@@ -581,7 +581,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
         namespace ctc = tensor_layout::convolution;
 
         // check device
-        if(ck::is_navi3_supported())
+        if(ck::is_navi3_supported() || ck::is_navi4_supported())
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
             {

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

@@ -340,7 +340,8 @@ struct GridwiseGemmMultipleD_Wmma
 
     static constexpr auto MWaves = MPerBlock / (MRepeat * MPerWmma);
     static constexpr auto NWaves = NPerBlock / (NRepeat * NPerWmma);
-    static constexpr auto WmmaK  = K1 == 16 ? 32 : 16;
+    // static constexpr auto WmmaK  = K1 == 16 ? 32 : 16;
+    static constexpr auto WmmaK = 16;
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 

library/include/ck/library/utility/check_err.hpp

@@ -156,7 +156,7 @@ check_err(const Range& out,
         {
             max_err = err > max_err ? err : max_err;
             err_count++;
-            // if(err_count < 5)
+            if(err_count < 5)
             {
                 std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
                           << "] != ref[" << i << "]: " << o << " != " << r << std::endl;