error: use of undeclared identifier '__builtin_amdgcn_sched_barrier'; did you mean '__builtin_amdgcn_s_getreg'? __builtin_amdgcn_sched_barrier(0);

[jeffhammond]

I am using 0345963eef4f92e9c5eab608bb8557b5463a1dcb on an MI-210 (gfx90a) system.

~/composable_kernel/build$ git clean -dfx ; cmake -DCMAKE_PREFIX_PATH=/opt/rocm -DCMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -DCMAKE_C_COMPILER=/opt/rocm/bin/hipcc -DCMAKE_CXX_FLAGS="-O3" -DCMAKE_BUILD_TYPE=Release -DGPU_TARGETS="gfx90a" ..

I see many of the following errors. What is wrong?

In file included from ~/composable_kernel/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp:9:
In file included from ~/composable_kernel/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp:16:
In file included from ~/composable_kernel/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp:11:
In file included from ~/composable_kernel/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp:6:
In file included from ~/composable_kernel/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp:7:
~/composable_kernel/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp:543:13: error: use of undeclared identifier '__builtin_amdgcn_sched_barrier'; did you mean '__builtin_amdgcn_s_getreg'?
            __builtin_amdgcn_sched_barrier(0);
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            __builtin_amdgcn_s_getreg
/opt/rocm-5.2.0/include/hip/amd_detail/amd_device_functions.h:980:22: note: '__builtin_amdgcn_s_getreg' declared here
    unsigned cu_id = __builtin_amdgcn_s_getreg(
                     ^

[rosenrodt]

ROCm 5.3 should be fine. What's the hipcc version?

[ppanchad-amd]

@jeffhammond Please check if your issue still exist with the latest ROCm 6.2? If resolved, please close the ticket. Thanks

[jeffhammond]

it is no longer relevant to me.