Merge commit 'b38bb492a1a55b5abb0c345962143c0f9c482cfb' into develop

CMakeLists.txt

@@ -122,7 +122,7 @@ add_compile_options(
 # Recent change in compiler makes this warning ON by default, which led to compile errors.
 add_compile_options(-Wno-nrvo)
 
-if(NOT DISABLE_DL_KERNELS)
+if(NOT DISABLE_DL_KERNELS AND GPU_TARGETS MATCHES "gfx103|gfx10-3-generic")
     add_definitions(-DDL_KERNELS)
     set(DL_KERNELS "ON")
     set(CK_ENABLE_DL_KERNELS "ON")

include/ck/utility/amd_inline_asm.hpp

@@ -86,7 +86,7 @@ inline __device__ f8x8_t amd_assembly_i4_to_fp8x8(int a)
 
     return bit_cast<f8x8_t>(((static_cast<uint64_t>(fp8x4_1) << 32) | fp8x4_0));
 }
-
+#ifdef DL_KERNELS
 // c0 += inner_product(a, b0)
 // c1 += inner_product(a, b1)
 __device__ void amd_assembly_outer_product_1x2(float a, float b0, float b1, float& c0, float& c1)
@@ -430,6 +430,6 @@ __device__ void amd_assembly_outer_product_1x4(int8x16_t a,
                                    c2,
                                    c3);
 }
-
+#endif
 } // namespace ck
 #endif