Merge branch 'develop' into grouped_gemm_multi_abd_fixed_nk_example

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

@@ -184,7 +184,8 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                 return false;
             }
         }
-        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940")
+        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940" ||
+                ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))

include/ck/utility/inner_product.hpp

@@ -192,6 +192,8 @@ inner_product<int8x4_t, int8x4_t, int32_t>(const int8x4_t& a, const int8x4_t& b,
 #else
     c = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b), c, false);
 #endif
+#elif defined(CK_USE_AMD_V_DOT4_I32_I8_GFX11)
+    c = __builtin_amdgcn_sudot4(true, bit_cast<int32_t>(a), true, bit_cast<int32_t>(b), c, false);
 #else
     const vector_type<int8_t, 4> a_vector{a};
     const vector_type<int8_t, 4> b_vector{b};