Status of ROCm runtime upstream kernels

Status of ROCm runtime upstream kernels

[Tom Stellard]

Hi Felix,

In a mailing list post[1], you mentioned that you had the ROCm
runtime working on an upstream kernel, which kernel version was this and
was it only Kaveri and Carrizo?

Thanks,
Tom


[1] https://lists.freedesktop.org/archives/amd-gfx/2017-September/012951.html
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: Status of ROCm runtime upstream kernels

[Felix Kuehling]

Yes, it's only on CZ and KV for now.

You also need a custom Thunk from here because the ioctl ABI is getting
changed in the upstreaming process:
https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/fxkamd/drm-next-wip

KV support in the user mode package is currently lacking. I was able to
hack the OpenCL driver to enable KV support with a simple patch:

commit 2761ff8bb444cf93de4d3d182b38d1da140df986
Author: Felix Kuehling <Felix.Kuehling@amd.com>
Date:   Fri Sep 8 16:14:23 2017 -0400

    Enable Kaveri
    
    Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>

diff --git a/runtime/device/rocm/rocdevice.cpp b/runtime/device/rocm/rocdevice.cpp
index 16db78a..edd3cbd 100644
--- a/runtime/device/rocm/rocdevice.cpp
+++ b/runtime/device/rocm/rocdevice.cpp
@@ -76,6 +76,8 @@ static HsaDeviceId getHsaDeviceId(hsa_agent_t device, uint32_t& pci_id) {
   }
 
   switch (gfxipVersion) {
+    case 700:
+      return HSA_SPECTRE_ID;
     case 701:
       return HSA_HAWAII_ID;
     case 801:


Regards,
  Felix


On 2017-09-25 03:58 PM, Tom Stellard wrote:
> Hi Felix,
>
> In a mailing list post[1], you mentioned that you had the ROCm
> runtime working on an upstream kernel, which kernel version was this and
> was it only Kaveri and Carrizo?
>
> Thanks,
> Tom
>
>
> [1] https://lists.freedesktop.org/archives/amd-gfx/2017-September/012951.html

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: Status of ROCm runtime upstream kernels

[Tom Stellard]

On 09/25/2017 01:11 PM, Felix Kuehling wrote:
> Yes, it's only on CZ and KV for now.
> 

And which kernel version do I need?

-Tom

> You also need a custom Thunk from here because the ioctl ABI is getting
> changed in the upstreaming process:
> https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/fxkamd/drm-next-wip
> 
> KV support in the user mode package is currently lacking. I was able to
> hack the OpenCL driver to enable KV support with a simple patch:
> 
> commit 2761ff8bb444cf93de4d3d182b38d1da140df986
> Author: Felix Kuehling <Felix.Kuehling@amd.com>
> Date:   Fri Sep 8 16:14:23 2017 -0400
> 
>     Enable Kaveri
>     
>     Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
> 
> diff --git a/runtime/device/rocm/rocdevice.cpp b/runtime/device/rocm/rocdevice.cpp
> index 16db78a..edd3cbd 100644
> --- a/runtime/device/rocm/rocdevice.cpp
> +++ b/runtime/device/rocm/rocdevice.cpp
> @@ -76,6 +76,8 @@ static HsaDeviceId getHsaDeviceId(hsa_agent_t device, uint32_t& pci_id) {
>    }
>  
>    switch (gfxipVersion) {
> +    case 700:
> +      return HSA_SPECTRE_ID;
>      case 701:
>        return HSA_HAWAII_ID;
>      case 801:
> 
> 
> Regards,
>   Felix
> 
> 
> On 2017-09-25 03:58 PM, Tom Stellard wrote:
>> Hi Felix,
>>
>> In a mailing list post[1], you mentioned that you had the ROCm
>> runtime working on an upstream kernel, which kernel version was this and
>> was it only Kaveri and Carrizo?
>>
>> Thanks,
>> Tom
>>
>>
>> [1] https://lists.freedesktop.org/archives/amd-gfx/2017-September/012951.html
> 

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: Status of ROCm runtime upstream kernels

[Felix Kuehling]

On 2017-09-25 04:26 PM, Tom Stellard wrote:
> On 09/25/2017 01:11 PM, Felix Kuehling wrote:
>> Yes, it's only on CZ and KV for now.
>>
> And which kernel version do I need?

Current amd-staging-drm-next should work. The necessary patches are
making their way into Linux 4.14.

Regards,
  Felix

>
> -Tom
>
>> You also need a custom Thunk from here because the ioctl ABI is getting
>> changed in the upstreaming process:
>> https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/fxkamd/drm-next-wip
>>
>> KV support in the user mode package is currently lacking. I was able to
>> hack the OpenCL driver to enable KV support with a simple patch:
>>
>> commit 2761ff8bb444cf93de4d3d182b38d1da140df986
>> Author: Felix Kuehling <Felix.Kuehling@amd.com>
>> Date:   Fri Sep 8 16:14:23 2017 -0400
>>
>>     Enable Kaveri
>>     
>>     Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
>>
>> diff --git a/runtime/device/rocm/rocdevice.cpp b/runtime/device/rocm/rocdevice.cpp
>> index 16db78a..edd3cbd 100644
>> --- a/runtime/device/rocm/rocdevice.cpp
>> +++ b/runtime/device/rocm/rocdevice.cpp
>> @@ -76,6 +76,8 @@ static HsaDeviceId getHsaDeviceId(hsa_agent_t device, uint32_t& pci_id) {
>>    }
>>  
>>    switch (gfxipVersion) {
>> +    case 700:
>> +      return HSA_SPECTRE_ID;
>>      case 701:
>>        return HSA_HAWAII_ID;
>>      case 801:
>>
>>
>> Regards,
>>   Felix
>>
>>
>> On 2017-09-25 03:58 PM, Tom Stellard wrote:
>>> Hi Felix,
>>>
>>> In a mailing list post[1], you mentioned that you had the ROCm
>>> runtime working on an upstream kernel, which kernel version was this and
>>> was it only Kaveri and Carrizo?
>>>
>>> Thanks,
>>> Tom
>>>
>>>
>>> [1] https://lists.freedesktop.org/archives/amd-gfx/2017-September/012951.html

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx