4x4

driver/driver.hip.cpp

@@ -580,7 +580,7 @@ int main(int argc, char* argv[])
 
     constexpr index_t HPad = 0;
     constexpr index_t WPad = 0;
-#elif 1
+#elif 0
     // 1x1 filter, 14x14 image, C = 2048
     constexpr index_t N  = 128;
     constexpr index_t C  = 2048;

src/include/gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn.hip.hpp

@@ -323,7 +323,7 @@ class gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn
                     (p_wei_block + wei_cyxk_block_desc.Get1dIndex(0, y, x, 0),
                      p_in_block + y * Wi + x,
                      p_out_thread,
-                     f_accum);
+                     f_accum, Number<in_block_element_space>());
                 }
             }
         }