[Mesa-dev] [PATCH] radeonsi: compute the absolute value before RSQ

[Samuel Pitoiset]

On 01/05/2017 06:49 PM, Roland Scheidegger wrote:
> Meh, I'm not really a big fan of such hacks. GPUs have support for NaNs
> since ages, and while glsl is lenient the point stands that returning a
> NaN is a more correct result, so doing extra work to get a wrong result
> doesn't look all that great to me.
> FWIW dx10 requires NaNs as results (for both sqrt and rsq). Maybe app
> specific quirks (if said apps can't be fixed) would be better...
> But well, it's your driver, so whatever floats your boat.

I think the main idea here is to reduce the number of bug reports when 
applications use D3D-style for RSQ/SQRT instead of what GLSL spec says.

>
> Roland
>
>
> Am 05.01.2017 um 18:37 schrieb Marek Olšák:
>> Shouldn't we also use abs for SQRT? For example, this adds abs for
>> both RSQ and SQRT:
>>
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__cgit.freedesktop.org_-7Emareko_mesa_commit_-3Fid-3D5e0fb661a8e6ac5f7b2245dd31595155128e0664&d=DgIGaQ&c=uilaK90D4TOVoH58JNXRgQ&r=_QIjpv-UJ77xEQY8fIYoQtr5qv8wKrPJc7v7_-CYAb0&m=wUYi2DhIZXHwD19wGIoEgceIF0cE0chTe7swjZ4MLVs&s=ZqrwtvXGdvXw-YJQtVnWo7yBeH5lhM3jtnYaa6vFZJY&e=
>>
>> Marek
>>
>> On Thu, Jan 5, 2017 at 5:47 PM, Samuel Pitoiset
>> <samuel.pitoiset at gmail.com> wrote:
>>> As explained by Nicolai, it seems like D3D always compute the
>>> absolute value while GLSL says that the result of inversesqrt()
>>> is undefined if x <= 0. Using the absolute value looks like safer
>>> especially when the game has been ported from D3D to GL.
>>>
>>> This gets rid of the NaN values in the "Spec Ops: The Line" game
>>> as well as the black squares.
>>>
>>> Bugzilla: https://urldefense.proofpoint.com/v2/url?u=https-3A__bugs.freedesktop.org_show-5Fbug.cgi-3Fid-3D97338&d=DgIGaQ&c=uilaK90D4TOVoH58JNXRgQ&r=_QIjpv-UJ77xEQY8fIYoQtr5qv8wKrPJc7v7_-CYAb0&m=wUYi2DhIZXHwD19wGIoEgceIF0cE0chTe7swjZ4MLVs&s=gjxDxmhEtQDYO34rucXJ6nIDojabOhpTAx1rNUP6X-g&e=
>>>
>>> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
>>> ---
>>>
>>> Nouveau also computes the absolute value before emitting RSQ.
>>>
>>>  src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c | 9 ++++++---
>>>  1 file changed, 6 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
>>> index 1966752cc0..ec6d6b0534 100644
>>> --- a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
>>> +++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
>>> @@ -754,9 +754,12 @@ static void emit_rsq(const struct lp_build_tgsi_action *action,
>>>                      struct lp_build_tgsi_context *bld_base,
>>>                      struct lp_build_emit_data *emit_data)
>>>  {
>>> -       LLVMValueRef sqrt =
>>> -               lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_SQRT,
>>> -                                        emit_data->args[0]);
>>> +       LLVMValueRef abs, sqrt;
>>> +
>>> +       abs = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_ABS,
>>> +                                      emit_data->args[0]);
>>> +
>>> +       sqrt = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_SQRT, abs);
>>>
>>>         emit_data->output[emit_data->chan] =
>>>                 lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_DIV,
>>> --
>>> 2.11.0
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_mesa-2Ddev&d=DgIGaQ&c=uilaK90D4TOVoH58JNXRgQ&r=_QIjpv-UJ77xEQY8fIYoQtr5qv8wKrPJc7v7_-CYAb0&m=wUYi2DhIZXHwD19wGIoEgceIF0cE0chTe7swjZ4MLVs&s=boy8sEj2W3FP0LnjHsSnUxWrR98aM22Qq-uxvFIqMlQ&e=
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.freedesktop.org_mailman_listinfo_mesa-2Ddev&d=DgIGaQ&c=uilaK90D4TOVoH58JNXRgQ&r=_QIjpv-UJ77xEQY8fIYoQtr5qv8wKrPJc7v7_-CYAb0&m=wUYi2DhIZXHwD19wGIoEgceIF0cE0chTe7swjZ4MLVs&s=boy8sEj2W3FP0LnjHsSnUxWrR98aM22Qq-uxvFIqMlQ&e=
>>
>