edited README, added device_gemm_dl op with extra gfx908 to support the GEMM example

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

@@ -485,7 +485,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030")
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" || ck::get_device_name() == "gfx908")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);

python/README.md

@@ -4,17 +4,21 @@ There are 2 directories: shared and normal. The normal directory contains one in
 generates multiple instances and compiles them into a shared library.
 
 ## Normal
-To generate the cpp file and executable: 
+To generate the cpp file and executable:  
+
 `python3 gemm_ex.py`
 
-Run the executable (same layout as CK examples: verification, initialization, run kernel # of times)
+Run the executable (same layout as CK examples: verification, initialization, run kernel # of times):  
+
 `./ex 0 1 5`
 
 ## Shared
-Generate all instances, make shared library and executable
+Generate all instances, make shared library and executable:  
+
 `python3 driver.py`
 
-Run the executable (same layout as CK examples)
+Run the executable (same layout as CK examples):  
+
 `./example 0 1 5`
 
 * There's a main.cpp file needed for the executable included, so be careful when deleting the generated cpp files for the instances