unnecessary scratch_load/store when using built in

include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp

@@ -414,7 +414,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
         constexpr auto KPack = math::integer_least_multiple(K1, WmmaK);
 
         auto blockwise_gemm =
-            BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle_FIFO<BlockSize,
+            BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle<BlockSize,
                                                          FloatA,
                                                          FloatB,
                                                          FloatAcc,

include/ck/utility/amd_wmma.hpp

@@ -23,11 +23,11 @@ struct intrin_wmma_f32_16x16x16_f16_w32<16, 16>
     {
         // * Inline assembly need to elimate the duplicated data load, compiler won't help you
         // delete them.
-        amd_assembly_wmma_f32_16x16x16_f16_w32(
-            reg_a, reg_b, reg_c.template AsType<float8_t>()(Number<0>{}));
-        // reg_c.template AsType<float8_t>()(Number<0>{}) =
-        // __builtin_amdgcn_wmma_f32_16x16x16_f16_w32( reg_a, reg_b, reg_c.template
-        // AsType<float8_t>()[Number<0>{}]);
+        // amd_assembly_wmma_f32_16x16x16_f16_w32(
+            // reg_a, reg_b, reg_c.template AsType<float8_t>()(Number<0>{}));
+        reg_c.template AsType<float8_t>()(Number<0>{}) =
+        __builtin_amdgcn_wmma_f32_16x16x16_f16_w32( reg_a, reg_b, reg_c.template
+        AsType<float8_t>()[Number<0>{}]);
     }
 };