implicit gemm v3 added lds double buffer

driver/device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw.hpp

@@ -3,6 +3,7 @@
 #include "device.hpp"
 #include "gridwise_convolution_wrapper.hip.hpp"
 #include "gridwise_convolution_implicit_gemm_v3_nchw_cyxk_nkhw.hip.hpp"
+#include "gridwise_convolution_implicit_gemm_v3_lds_double_buffer_nchw_cyxk_nkhw.hip.hpp"
 
 template <class T, class InDesc, class WeiDesc, class OutDesc>
 void device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw(InDesc,
@@ -98,8 +99,10 @@ void device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw(InDesc,
     for(index_t i = 0; i < nrepeat; ++i)
     {
         constexpr auto gridwise_conv =
-#if 1
+#if 0
             GridwiseConvolutionImplicitGemm_v3_nchw_cyxk_nkhw
+#else
+            GridwiseConvolutionImplicitGemm_v3_lds_double_buffer_nchw_cyxk_nkhw
 #endif
             <GridSize,
              BlockSize,

driver/driver.hip.cpp

@@ -422,7 +422,7 @@ int main(int argc, char* argv[])
 
     constexpr index_t HPad = 0;
     constexpr index_t WPad = 0;
-#elif 1
+#elif 0
     // 3x3, 34x34
     constexpr index_t N  = 64;
     constexpr index_t C  = 256;
@@ -446,7 +446,7 @@ int main(int argc, char* argv[])
 
     constexpr index_t HPad = 0;
     constexpr index_t WPad = 0;
-#elif 0
+#elif 1
     // 3x3 filter, 28x28 image
     constexpr index_t N  = 128;
     constexpr index_t C  = 256;