[Mesa-dev] [PATCH 3/4] gallium: add PIPE_CAP_TGSI_DDIV capability

Nicolai Hähnle nicolai.haehnle at amd.com
Mon Jan 16 17:46:40 UTC 2017


On 16.01.2017 18:14, Ilia Mirkin wrote:
> FWIW nouveau has an internal OP_DIV (which we lower to mul + rcp, but
> probably shouldn't). It'd be trivial to hook up DDIV to come out as
> OP_DIV in nv50_ir_from_tgsi.cpp. I suspect hacking something into
> r600_asm should be moderately easy as well.

That's a fair point, the only question is one of logistics. The missing 
drivers are nvc0 and r600.

I could go ahead and push patch #2 ("add DDIV instruction") immediately, 
then it's a question of how fast the nvc0 and r600 gaps can be filled in 
- in part due to the 17.0 branch point (which IIRC was actually supposed 
to have been last Friday? Cc Emil). Though I'd say this should count as 
a bug fix anyway :)

Do I get a R-b for patch #2 from you guys already?

Nicolai

> On Mon, Jan 16, 2017 at 12:08 PM, Roland Scheidegger <sroland at vmware.com> wrote:
>> Is a cap bit actually needed here or can we require drivers to
>> unconditionally support this (as long as they support doubles, of
>> course...)?
>> d3d11 has a requirement (if doubles are supported) that ddiv has an
>> accuracy of 0.5 ULP - emulating with mul + rcp clearly isn't allowed
>> (rcp, both the float and double variant, actually only has 1.0 ULP
>> requirement, but it doesn't even matter).
>> (Note that div, in contrast to ddiv, has explicitly allowed a*(1/b)
>> implementation.)
>>
>> My guess is you'd actually never want this lowered. Noone uses doubles
>> for ordinary graphics for a reason - I suspect when someone does use
>> doubles he'd rather wants full accuracy...
>>
>> Roland
>>
>>
>> Am 16.01.2017 um 17:20 schrieb Nicolai Hähnle:
>>> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>>>
>>> For drivers to indicate that they don't want double-precision divides
>>> to be lowered into rcp+mul.
>>> ---
>>>  src/gallium/docs/source/screen.rst               | 2 ++
>>>  src/gallium/drivers/freedreno/freedreno_screen.c | 3 ++-
>>>  src/gallium/drivers/i915/i915_screen.c           | 1 +
>>>  src/gallium/drivers/ilo/ilo_screen.c             | 1 +
>>>  src/gallium/drivers/llvmpipe/lp_screen.c         | 1 +
>>>  src/gallium/drivers/nouveau/nv30/nv30_screen.c   | 1 +
>>>  src/gallium/drivers/nouveau/nv50/nv50_screen.c   | 1 +
>>>  src/gallium/drivers/nouveau/nvc0/nvc0_screen.c   | 1 +
>>>  src/gallium/drivers/r300/r300_screen.c           | 1 +
>>>  src/gallium/drivers/r600/r600_pipe.c             | 3 ++-
>>>  src/gallium/drivers/radeonsi/si_pipe.c           | 3 ++-
>>>  src/gallium/drivers/softpipe/sp_screen.c         | 1 +
>>>  src/gallium/drivers/svga/svga_screen.c           | 1 +
>>>  src/gallium/drivers/swr/swr_screen.cpp           | 1 +
>>>  src/gallium/drivers/vc4/vc4_screen.c             | 1 +
>>>  src/gallium/drivers/virgl/virgl_screen.c         | 1 +
>>>  src/gallium/include/pipe/p_defines.h             | 1 +
>>>  17 files changed, 21 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/gallium/docs/source/screen.rst b/src/gallium/docs/source/screen.rst
>>> index 000551a..4d79104 100644
>>> --- a/src/gallium/docs/source/screen.rst
>>> +++ b/src/gallium/docs/source/screen.rst
>>> @@ -369,6 +369,8 @@ The integer capabilities:
>>>  * ``PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY``: Tell the GLSL compiler to use
>>>    the minimum amount of optimizations just to be able to do all the linking
>>>    and lowering.
>>> +* ``PIPE_CAP_TGSI_DDIV``: Indicates that the TGSI DDIV instruction for double
>>> +  division is supported and preferred over a DRCP+DMUL sequence.
>>>
>>>
>>>  .. _pipe_capf:
>>> diff --git a/src/gallium/drivers/freedreno/freedreno_screen.c b/src/gallium/drivers/freedreno/freedreno_screen.c
>>> index 2ff89eb..d6fdb77 100644
>>> --- a/src/gallium/drivers/freedreno/freedreno_screen.c
>>> +++ b/src/gallium/drivers/freedreno/freedreno_screen.c
>>> @@ -295,7 +295,8 @@ fd_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
>>>       case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
>>>       case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
>>>       case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>> -        case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_TGSI_DDIV:
>>>               return 0;
>>>
>>>       case PIPE_CAP_MAX_VIEWPORTS:
>>> diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
>>> index 18578c0..8791105 100644
>>> --- a/src/gallium/drivers/i915/i915_screen.c
>>> +++ b/src/gallium/drivers/i915/i915_screen.c
>>> @@ -297,6 +297,7 @@ i915_get_param(struct pipe_screen *screen, enum pipe_cap cap)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_MAX_VIEWPORTS:
>>> diff --git a/src/gallium/drivers/ilo/ilo_screen.c b/src/gallium/drivers/ilo/ilo_screen.c
>>> index 20a0e8d..916c1c9 100644
>>> --- a/src/gallium/drivers/ilo/ilo_screen.c
>>> +++ b/src/gallium/drivers/ilo/ilo_screen.c
>>> @@ -520,6 +520,7 @@ ilo_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_VENDOR_ID:
>>> diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
>>> index 4501df4..3248c4d 100644
>>> --- a/src/gallium/drivers/llvmpipe/lp_screen.c
>>> +++ b/src/gallium/drivers/llvmpipe/lp_screen.c
>>> @@ -341,6 +341,7 @@ llvmpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>     }
>>>     /* should only get here on unhandled cases */
>>> diff --git a/src/gallium/drivers/nouveau/nv30/nv30_screen.c b/src/gallium/drivers/nouveau/nv30/nv30_screen.c
>>> index 19df068..586a85c 100644
>>> --- a/src/gallium/drivers/nouveau/nv30/nv30_screen.c
>>> +++ b/src/gallium/drivers/nouveau/nv30/nv30_screen.c
>>> @@ -206,6 +206,7 @@ nv30_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_VENDOR_ID:
>>> diff --git a/src/gallium/drivers/nouveau/nv50/nv50_screen.c b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
>>> index 5637001..3cf5724 100644
>>> --- a/src/gallium/drivers/nouveau/nv50/nv50_screen.c
>>> +++ b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
>>> @@ -258,6 +258,7 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_VENDOR_ID:
>>> diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
>>> index b6e4c6c..de00565 100644
>>> --- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
>>> +++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
>>> @@ -275,6 +275,7 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_VENDOR_ID:
>>> diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
>>> index 50a42ef..bfce22e 100644
>>> --- a/src/gallium/drivers/r300/r300_screen.c
>>> +++ b/src/gallium/drivers/r300/r300_screen.c
>>> @@ -228,6 +228,7 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>>>          case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>          case PIPE_CAP_NATIVE_FENCE_FD:
>>>          case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +        case PIPE_CAP_TGSI_DDIV:
>>>              return 0;
>>>
>>>          /* SWTCL-only features. */
>>> diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
>>> index a0da24f..263d09c 100644
>>> --- a/src/gallium/drivers/r600/r600_pipe.c
>>> +++ b/src/gallium/drivers/r600/r600_pipe.c
>>> @@ -376,7 +376,8 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>>>       case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
>>>       case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>       case PIPE_CAP_NATIVE_FENCE_FD:
>>> -        case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_TGSI_DDIV:
>>>               return 0;
>>>
>>>       case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
>>> diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
>>> index 842eb59..fdf9bee 100644
>>> --- a/src/gallium/drivers/radeonsi/si_pipe.c
>>> +++ b/src/gallium/drivers/radeonsi/si_pipe.c
>>> @@ -413,7 +413,8 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>>>       case PIPE_CAP_CULL_DISTANCE:
>>>       case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
>>>       case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>> -        case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +     case PIPE_CAP_TGSI_DDIV:
>>>               return 1;
>>>
>>>       case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
>>> diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
>>> index 93c7503..ab1a877 100644
>>> --- a/src/gallium/drivers/softpipe/sp_screen.c
>>> +++ b/src/gallium/drivers/softpipe/sp_screen.c
>>> @@ -291,6 +291,7 @@ softpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>     case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
>>>        return 4;
>>> diff --git a/src/gallium/drivers/svga/svga_screen.c b/src/gallium/drivers/svga/svga_screen.c
>>> index 3e27474..2ae4ee7 100644
>>> --- a/src/gallium/drivers/svga/svga_screen.c
>>> +++ b/src/gallium/drivers/svga/svga_screen.c
>>> @@ -422,6 +422,7 @@ svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>     }
>>>
>>> diff --git a/src/gallium/drivers/swr/swr_screen.cpp b/src/gallium/drivers/swr/swr_screen.cpp
>>> index 5012388..bca6047 100644
>>> --- a/src/gallium/drivers/swr/swr_screen.cpp
>>> +++ b/src/gallium/drivers/swr/swr_screen.cpp
>>> @@ -315,6 +315,7 @@ swr_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
>>>     case PIPE_CAP_NATIVE_FENCE_FD:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>
>>>     case PIPE_CAP_VENDOR_ID:
>>> diff --git a/src/gallium/drivers/vc4/vc4_screen.c b/src/gallium/drivers/vc4/vc4_screen.c
>>> index 3da6202..7318425 100644
>>> --- a/src/gallium/drivers/vc4/vc4_screen.c
>>> +++ b/src/gallium/drivers/vc4/vc4_screen.c
>>> @@ -242,6 +242,7 @@ vc4_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
>>>          case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>          case PIPE_CAP_NATIVE_FENCE_FD:
>>>          case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +        case PIPE_CAP_TGSI_DDIV:
>>>                  return 0;
>>>
>>>                  /* Stream output. */
>>> diff --git a/src/gallium/drivers/virgl/virgl_screen.c b/src/gallium/drivers/virgl/virgl_screen.c
>>> index 0a18aa0..05bdd37 100644
>>> --- a/src/gallium/drivers/virgl/virgl_screen.c
>>> +++ b/src/gallium/drivers/virgl/virgl_screen.c
>>> @@ -252,6 +252,7 @@ virgl_get_param(struct pipe_screen *screen, enum pipe_cap param)
>>>     case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
>>>     case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
>>>     case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
>>> +   case PIPE_CAP_TGSI_DDIV:
>>>        return 0;
>>>     case PIPE_CAP_VENDOR_ID:
>>>        return 0x1af4;
>>> diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
>>> index 7dfd7dd..28e134c 100644
>>> --- a/src/gallium/include/pipe/p_defines.h
>>> +++ b/src/gallium/include/pipe/p_defines.h
>>> @@ -744,6 +744,7 @@ enum pipe_cap
>>>     PIPE_CAP_TGSI_CAN_READ_OUTPUTS,
>>>     PIPE_CAP_NATIVE_FENCE_FD,
>>>     PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY,
>>> +   PIPE_CAP_TGSI_DDIV,
>>>  };
>>>
>>>  #define PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50 (1 << 0)
>>>
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev



More information about the mesa-dev mailing list