[Mesa-dev] [PATCH] clover: Return the minimum required value for CL_DEVICE_SINGLE_FP_CONFIG

[Matt Arsenault]

> On Mar 6, 2015, at 8:56 AM, Francisco Jerez <currojerez at riseup.net> wrote:
> 
> Tom Stellard <tom at stellard.net <mailto:tom at stellard.net>> writes:
> 
>> On Thu, Mar 05, 2015 at 08:42:25PM +0200, Francisco Jerez wrote:
>>> Tom Stellard <thomas.stellard at amd.com> writes:
>>> 
>>>> This means dropping CL_FP_DENORM from the current return value.
>>>> ---
>>>> src/gallium/state_trackers/clover/api/device.cpp | 4 +++-
>>>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>>> 
>>>> diff --git a/src/gallium/state_trackers/clover/api/device.cpp b/src/gallium/state_trackers/clover/api/device.cpp
>>>> index b1f556f..db3b931 100644
>>>> --- a/src/gallium/state_trackers/clover/api/device.cpp
>>>> +++ b/src/gallium/state_trackers/clover/api/device.cpp
>>>> @@ -201,8 +201,10 @@ clGetDeviceInfo(cl_device_id d_dev, cl_device_info param,
>>>>       break;
>>>> 
>>>>    case CL_DEVICE_SINGLE_FP_CONFIG:
>>>> +      // This is the "mandated minimum single precision floating-point
>>>> +      // capability"
>>> 
>>> Could you add that this is according to the OpenCL 1.1 specification?
>>> OpenCL 1.2 is even weaker (CL_FP_INF_NAN is not required, only one of
>>> CL_FP_ROUND_TO_ZERO or CL_FP_ROUND_TO_NEAREST is required, and no FP
>>> capabilities at all are required for custom devices as Jan pointed out).
>>> 
>>>>       buf.as_scalar<cl_device_fp_config>() =
>>>> -         CL_FP_DENORM | CL_FP_INF_NAN | CL_FP_ROUND_TO_NEAREST;
>>>> +         CL_FP_INF_NAN | CL_FP_ROUND_TO_NEAREST;
>>> 
>>> I'm okay with this change, but I'm curious, is this motivated by your
>>> architecture not supporting denorms?
>>> 
>> 
>> It can, but supporting them hurts performance.
>> 
> Sounds like you want to advertise denorm support and rely on the
> -cl-denorms-are-zero compiler option to decide whether to flush them to
> zero or not?

This is true for newer devices which have more instructions as fast with denormal support. For the currently supported devices, the performance difference is quite extreme and the denormal support is not that useful.





> 
>> -Tom
>> 
>>>>       break;
>>>> 
>>>>    case CL_DEVICE_DOUBLE_FP_CONFIG:
>>>> -- 
>>>> 2.0.4
>> 
>> 
>> 
>> 
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20150306/32480ab1/attachment-0001.html>