Use macro to toggle pipeline v2 optimization

include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp

@@ -100,7 +100,7 @@ struct GridwiseGemmPipeline_v2
                 // global read i + 2
                 b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
 
-#if 1
+#if defined(ENABLE_PIPELINE_V2_OPT)
                 __builtin_amdgcn_sched_group_barrier(0x020, 2, 0); // VMEM read
                 __builtin_amdgcn_sched_group_barrier(0x008, 1, 0); // MFMA
 

test/gemm/CMakeLists.txt

@@ -20,7 +20,9 @@ add_library(gemm_standalone_xdl_fp16_instances STATIC
     # instance/gemm_f16_tn_instance.cpp
     # instance/gemm_f16_tt_instance.cpp
 )
-set_source_files_properties(instance/gemm_f16_nn_instance.cpp PROPERTIES COMPILE_FLAGS "-save-temps -Wno-gnu-line-marker -amdgpu-enable-max-ilp-scheduling-strategy")
+set_source_files_properties(instance/gemm_f16_nn_instance.cpp PROPERTIES
+    COMPILE_FLAGS "-save-temps -Wno-gnu-line-marker -amdgpu-enable-max-ilp-scheduling-strategy"
+    COMPILE_DEFINITIONS ENABLE_PIPELINE_V2_OPT)
 
 add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
 target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)