On 10/17/2024 11:53 AM, Kenneth Feng wrote:
> disable deep sleep during the compute workload for the
> potential performance loss on smu v14.0.2/3
> 
> V2: consider the case when the compute workload is set twice
> 
> Signed-off-by: Kenneth Feng <[email protected]>
> ---
>  drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0_2_ppt.c | 10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0_2_ppt.c 
> b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0_2_ppt.c
> index da448523ec60..618a8e17c457 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0_2_ppt.c
> +++ b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0_2_ppt.c
> @@ -1787,7 +1787,7 @@ static int smu_v14_0_2_set_power_profile_mode(struct 
> smu_context *smu,
>       DpmActivityMonitorCoeffInt_t *activity_monitor =
>               &(activity_monitor_external.DpmActivityMonitorCoeffInt);
>       int workload_type, ret = 0;
> -
> +     uint32_t prev_profile_mode = smu->power_profile_mode;
>       smu->power_profile_mode = input[size];
>  
>       if (smu->power_profile_mode >= PP_SMC_POWER_PROFILE_COUNT) {
> @@ -1845,6 +1845,14 @@ static int smu_v14_0_2_set_power_profile_mode(struct 
> smu_context *smu,
>               }
>       }
>  
> +     if (smu->power_profile_mode == PP_SMC_POWER_PROFILE_COMPUTE &&
> +             prev_profile_mode != PP_SMC_POWER_PROFILE_COMPUTE)

This check shouldn't be required. If the current profile is the same as
the target profile, the code shouldn't reach here - that check should be
at an earlier stage than this.

Thanks,
Lijo

> +             smu_v14_0_deep_sleep_control(smu, false);
> +
> +     if (smu->power_profile_mode != PP_SMC_POWER_PROFILE_COMPUTE &&
> +             prev_profile_mode == PP_SMC_POWER_PROFILE_COMPUTE)
> +             smu_v14_0_deep_sleep_control(smu, true);
> +
>       /* conv PP_SMC_POWER_PROFILE* to WORKLOAD_PPLIB_*_BIT */
>       workload_type = smu_cmn_to_asic_specific_index(smu,
>                                                      
> CMN2ASIC_MAPPING_WORKLOAD,

Reply via email to