merge master

driver/src/conv_driver.cpp

@@ -64,8 +64,8 @@ int main(int argc, char* argv[])
     using ConvStrides   = Sequence<1, 1>;
     using ConvDilations = Sequence<1, 1>;
 
-    using LeftPads                   = Sequence<0, 0>;
-    using RightPads                  = Sequence<0, 0>;
+    using LeftPads  = Sequence<0, 0>;
+    using RightPads = Sequence<0, 0>;
 #elif 0
     constexpr index_t N  = 1;
     constexpr index_t C  = 16;
@@ -108,7 +108,7 @@ int main(int argc, char* argv[])
 
     using LeftPads  = Sequence<1, 1>;
     using RightPads = Sequence<1, 1>;
-#elif 1
+#elif 0
     constexpr index_t N  = 1;
     constexpr index_t C  = 16;
     constexpr index_t HI = 540;
@@ -772,7 +772,7 @@ int main(int argc, char* argv[])
         LeftPads{},
         RightPads{},
         nrepeat);
-#elif 0
+#elif 1
     device_dynamic_convolution_forward_implicit_gemm_v4r4_nhwc_kyxc_nhwk<in_data_t,
                                                                          in_vector_size,
                                                                          acc_data_t,

script/cmake-rocm3.7.sh → script/cmake-rocm.sh

@@ -16,7 +16,8 @@ cmake
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                                                  \
 ${MY_PROJECT_SOURCE}
 
-#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -save-temps=$CWD"                                                                   \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0"                   \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0 -mllvm -print-before=amdgpu-codegenprepare -mllvm -print-module-scope" \
 #-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -gline-tables-only -save-temps=$CWD"                                                \
 #-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0"                                               \
 #-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0 -save-temps=$CWD"                              \