[Mesa-dev] [PATCH 15/24] i965: Use LZD to implement nir_op_ufind_msb

[Matt Turner]

On Thu, Jun 30, 2016 at 3:33 PM, Francisco Jerez <currojerez at riseup.net> wrote:
> Matt Turner <mattst88 at gmail.com> writes:
>
>> On Wed, Jun 29, 2016 at 2:04 PM, Ian Romanick <idr at freedesktop.org> wrote:
>>> From: Ian Romanick <ian.d.romanick at intel.com>
>>>
>>> This uses one less instruction.
>>
>> Add FBH to the list of stupid instructions.
>>
>>> Signed-off-by: Ian Romanick <ian.d.romanick at intel.com>
>>> ---
>>>  src/mesa/drivers/dri/i965/brw_fs.h               |  4 ++++
>>>  src/mesa/drivers/dri/i965/brw_fs_generator.cpp   |  3 +++
>>>  src/mesa/drivers/dri/i965/brw_fs_nir.cpp         | 26 +++++++++++++++++++++++-
>>>  src/mesa/drivers/dri/i965/brw_vec4.h             |  4 ++++
>>>  src/mesa/drivers/dri/i965/brw_vec4_generator.cpp |  3 +++
>>>  src/mesa/drivers/dri/i965/brw_vec4_nir.cpp       | 22 ++++++++++++++++++++
>>>  6 files changed, 61 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h
>>> index 4237197..22ce092 100644
>>> --- a/src/mesa/drivers/dri/i965/brw_fs.h
>>> +++ b/src/mesa/drivers/dri/i965/brw_fs.h
>>> @@ -237,6 +237,10 @@ public:
>>>                           nir_tex_instr *instr);
>>>     void nir_emit_jump(const brw::fs_builder &bld,
>>>                        nir_jump_instr *instr);
>>> +   void nir_emit_find_msb_using_lzd(const brw::fs_builder &bld,
>>> +                                    const fs_reg &result,
>>> +                                    const fs_reg &src,
>>> +                                    bool is_signed);
>>>     fs_reg get_nir_src(const nir_src &src);
>>>     fs_reg get_nir_src_imm(const nir_src &src);
>>>     fs_reg get_nir_dest(const nir_dest &dest);
>>> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
>>> index d25d26a..bda4a26 100644
>>> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
>>> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
>>> @@ -1761,6 +1761,9 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
>>>           /* FBL only supports UD type for dst. */
>>>           brw_FBL(p, retype(dst, BRW_REGISTER_TYPE_UD), src[0]);
>>>           break;
>>> +      case BRW_OPCODE_LZD:
>>> +         brw_LZD(p, dst, src[0]);
>>> +         break;
>>>        case BRW_OPCODE_CBIT:
>>>           assert(devinfo->gen >= 7);
>>>           /* CBIT only supports UD type for dst. */
>>> diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>>> index b3f5dfd..f15bf3e 100644
>>> --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>>> +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
>>> @@ -617,6 +617,25 @@ fs_visitor::optimize_frontfacing_ternary(nir_alu_instr *instr,
>>>  }
>>>
>>>  void
>>> +fs_visitor::nir_emit_find_msb_using_lzd(const fs_builder &bld,
>>> +                                        const fs_reg &result,
>>> +                                        const fs_reg &src,
>>> +                                        bool is_signed)
>>> +{
>>> +   fs_inst *inst;
>>> +
>>> +   bld.LZD(retype(result, BRW_REGISTER_TYPE_UD), src);
>>> +
>>> +   /* LZD counts from the MSB side, while GLSL's findMSB() wants the count
>>> +    * from the LSB side. Subtract the result from 31 to convert the MSB
>>> +    * count into an LSB count.  If no bits are set, LZD will return 32.
>>> +    * 31-32 = -1, which is exactly what findMSB() is supposed to return.
>>> +    */
>>> +   inst = bld.ADD(result, retype(result, BRW_REGISTER_TYPE_D), brw_imm_d(31));
>>> +   inst->src[0].negate = true;
>>> +}
>>
>> I'd personally be inclined to just inline these functions. I know they
>> grow somewhat in the next patches... whatever your preference is.
>
> It seems to grow quite a bit in PATCH 16, and it's used in multiple
> places, right?  How about we keep it factored out but make it a
> stand-alone function instead of a visitor method?  It doesn't look like
> it uses *any* internal or external data structures of fs_visitor, it
> doesn't even dereference the 'this' pointer at all AFAICT, so you could
> likely improve encapsulation somewhat by making it a static function
> local to the brw_*_nir.cpp source files.

That sounds like a fine plan to me.