[Mesa-dev] [PATCH] gallium/radeon: add HUD queries for GPU temperature and clocks

[Nicolai Hähnle]

On 14.02.2017 18:20, Marek Olšák wrote:
> On Tue, Feb 14, 2017 at 9:23 AM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
>> On 13.02.2017 23:08, Samuel Pitoiset wrote:
>>>
>>> Only the Radeon kernel driver exposed the GPU temperature and
>>> the shader/memory clocks, this implements the same functionality
>>> for the AMDGPU kernel driver.
>>>
>>> These queries will return 0 if the DRM version is less than 3.10,
>>> I don't explicitely check the version here because the query
>>> codepath is already a bit messy.
>>>
>>> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
>>> ---
>>>  src/gallium/drivers/radeon/r600_query.c       | 12 ++++++------
>>>  src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c | 19 ++++++++++++++++++-
>>>  2 files changed, 24 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/src/gallium/drivers/radeon/r600_query.c
>>> b/src/gallium/drivers/radeon/r600_query.c
>>> index 05741d33c9..4959621e11 100644
>>> --- a/src/gallium/drivers/radeon/r600_query.c
>>> +++ b/src/gallium/drivers/radeon/r600_query.c
>>> @@ -1768,6 +1768,10 @@ static struct pipe_driver_query_info
>>> r600_driver_query_list[] = {
>>>         XG(GPIN, "GPIN_003",            GPIN_NUM_SPI,           UINT,
>>> AVERAGE),
>>>         XG(GPIN, "GPIN_004",            GPIN_NUM_SE,            UINT,
>>> AVERAGE),
>>>
>>> +       X("temperature",                GPU_TEMPERATURE,        UINT64,
>>> AVERAGE),
>>> +       X("shader-clock",               CURRENT_GPU_SCLK,       HZ,
>>> AVERAGE),
>>> +       X("memory-clock",               CURRENT_GPU_MCLK,       HZ,
>>> AVERAGE),
>>> +
>>>         /* The following queries must be at the end of the list because
>>> their
>>>          * availability is adjusted dynamically based on the DRM version.
>>> */
>>>         X("GPU-load",                   GPU_LOAD,               UINT64,
>>> AVERAGE),
>>> @@ -1792,10 +1796,6 @@ static struct pipe_driver_query_info
>>> r600_driver_query_list[] = {
>>>         X("GPU-dma-busy",               GPU_DMA_BUSY,           UINT64,
>>> AVERAGE),
>>>         X("GPU-scratch-ram-busy",       GPU_SCRATCH_RAM_BUSY,   UINT64,
>>> AVERAGE),
>>>         X("GPU-ce-busy",                GPU_CE_BUSY,            UINT64,
>>> AVERAGE),
>>> -
>>> -       X("temperature",                GPU_TEMPERATURE,        UINT64,
>>> AVERAGE),
>>> -       X("shader-clock",               CURRENT_GPU_SCLK,       HZ,
>>> AVERAGE),
>>> -       X("memory-clock",               CURRENT_GPU_MCLK,       HZ,
>>> AVERAGE),
>>>  };
>>>
>>>  #undef X
>>> @@ -1808,9 +1808,9 @@ static unsigned r600_get_num_queries(struct
>>> r600_common_screen *rscreen)
>>>                 return ARRAY_SIZE(r600_driver_query_list);
>>>         else if (rscreen->info.drm_major == 3) {
>>>                 if (rscreen->chip_class >= VI)
>>> -                       return ARRAY_SIZE(r600_driver_query_list) - 3;
>>> +                       return ARRAY_SIZE(r600_driver_query_list);
>>>                 else
>>> -                       return ARRAY_SIZE(r600_driver_query_list) - 10;
>>> +                       return ARRAY_SIZE(r600_driver_query_list) - 7;
>>>         }
>>>         else
>>>                 return ARRAY_SIZE(r600_driver_query_list) - 25;
>>> diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
>>> b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
>>> index c3dfda53f0..d99df9e941 100644
>>> --- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
>>> +++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
>>> @@ -63,6 +63,18 @@
>>>  #define AMDGPU_INFO_NUM_EVICTIONS              0x18
>>>  #endif
>>>
>>> +#ifndef AMDGPU_INFO_CURRENT_GPU_SCLK
>>> +#define AMDGPU_INFO_CURRENT_GPU_SCLK   0x1c
>>> +#endif
>>> +
>>> +#ifndef AMDGPU_INFO_CURRENT_GPU_MCLK
>>> +#define AMDGPU_INFO_CURRENT_GPU_MCLK   0x1d
>>> +#endif
>>> +
>>> +#ifndef AMDGPU_INFO_CURRENT_GPU_TEMP
>>> +#define AMDGPU_INFO_CURRENT_GPU_TEMP   0x1e
>>> +#endif
>>> +
>>>  static struct util_hash_table *dev_tab = NULL;
>>>  pipe_static_mutex(dev_tab_mutex);
>>>
>>> @@ -459,9 +471,14 @@ static uint64_t amdgpu_query_value(struct
>>> radeon_winsys *rws,
>>>        amdgpu_query_heap_info(ws->dev, AMDGPU_GEM_DOMAIN_GTT, 0, &heap);
>>>        return heap.heap_usage;
>>>     case RADEON_GPU_TEMPERATURE:
>>> +      amdgpu_query_info(ws->dev, AMDGPU_INFO_CURRENT_GPU_TEMP, 4,
>>> &retval);
>>> +      return retval;
>>>     case RADEON_CURRENT_SCLK:
>>> +      amdgpu_query_info(ws->dev, AMDGPU_INFO_CURRENT_GPU_SCLK, 4,
>>> &retval);
>>> +      return retval;
>>>     case RADEON_CURRENT_MCLK:
>>> -      return 0;
>>> +      amdgpu_query_info(ws->dev, AMDGPU_INFO_CURRENT_GPU_MCLK, 4,
>>> &retval);
>>> +      return retval;
>>
>>
>> I think these should be guarded with a DRM (kernel) version check. Also,
>> what Emil said.
>
> Let's ignore the DRM version if the query doesn't send any messages to
> dmesg. Reporting 0 in the HUD is already self-explanatory. (at least
> for us) We don't even check the DRM version for older queries that are
> not universally supported.

Fair enough, as long as it doesn't result in dmesg spam it's no big deal.

Cheers,
Nicolai

>
> Marek
>