[Mesa-dev] [PATCH 11/13] nir: Add lrp lowering for doubles in opt_algebraic

[Samuel Iglesias Gonsálvez]

On 27/04/16 21:18, Jason Ekstrand wrote:
> On Tue, Apr 12, 2016 at 1:05 AM, Samuel Iglesias Gonsálvez <
> siglesias at igalia.com> wrote:
> 
>> Some hardware (i965 on Broadwell generation, for example) does not support
>> natively the execution of lrp instruction with double arguments.
>>
>> Add 'lower_lrp_double' flag to lower this instruction in that case.
>>
>> Signed-off-by: Samuel Iglesias Gonsálvez <siglesias at igalia.com>
>> ---
>>  src/compiler/nir/nir.h                | 2 ++
>>  src/compiler/nir/nir_opt_algebraic.py | 7 +++++--
>>  2 files changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
>> index 0d17ce0..0bfc849 100644
>> --- a/src/compiler/nir/nir.h
>> +++ b/src/compiler/nir/nir.h
>> @@ -1582,6 +1582,8 @@ typedef struct nir_shader_compiler_options {
>>     bool lower_fdiv;
>>     bool lower_ffma;
>>     bool lower_flrp;
>> +   /** Lowers flrp when it does not supports doubles */
>>
> 
> "does not support"
> 
> 
>> +   bool lower_flrp_double;
>>
> 
> Maybe rename the first one to lower_flrp32 and call this lower_flrp64?
> 
> 
>>     bool lower_fpow;
>>     bool lower_fsat;
>>     bool lower_fsqrt;
>> diff --git a/src/compiler/nir/nir_opt_algebraic.py
>> b/src/compiler/nir/nir_opt_algebraic.py
>> index ec8929a..f971585 100644
>> --- a/src/compiler/nir/nir_opt_algebraic.py
>> +++ b/src/compiler/nir/nir_opt_algebraic.py
>> @@ -92,11 +92,14 @@ optimizations = [
>>     (('~flrp', 0.0, a, b), ('fmul', a, b)),
>>     (('~flrp', a, b, ('b2f', c)), ('bcsel', c, b, a),
>> 'options->lower_flrp'),
>>     (('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a),
>> 'options->lower_flrp'),
>> +   (('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a),
>> '!options->lower_flrp && options->lower_flrp_double', 64),
>>
> 
> With the new mechanism, the first one because 'flrp at 32' and this becomes
> 'flrp at 64'
> 
> Also, I think we can just treat lower_flrp and lower_flrp64 as orthogonal
> and require drivers to specify both if they want it.
> 

OK, I will send today a v2 with those changes.

Thanks!

Sam

> 
>>     (('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'),
>>     (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('fmul',
>> b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp'),
>> -   (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg',         c ))), ('fmul',
>> b,         c )), ('flrp', a, b, c), '!options->lower_flrp'),
>> +   (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg',         c ))), ('fmul',
>> b,         c )), ('flrp', a, b, c), '!options->lower_flrp', 32),
>> +   (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg',         c ))), ('fmul',
>> b,         c )), ('flrp', a, b, c), '!options->lower_flrp &&
>> !options->lower_flrp_double', 64),
>>     (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))),
>> ('bcsel', c, b, a), 'options->lower_flrp'),
>> -   (('~fadd', a, ('fmul',         c , ('fadd', b, ('fneg', a)))),
>> ('flrp', a, b, c), '!options->lower_flrp'),
>> +   (('~fadd', a, ('fmul',         c , ('fadd', b, ('fneg', a)))),
>> ('flrp', a, b, c), '!options->lower_flrp', 32),
>> +   (('~fadd', a, ('fmul',         c , ('fadd', b, ('fneg', a)))),
>> ('flrp', a, b, c), '!options->lower_flrp && !options->lower_flrp_double',
>> 64),
>>     (('ffma', a, b, c), ('fadd', ('fmul', a, b), c),
>> 'options->lower_ffma'),
>>     (('~fadd', ('fmul', a, b), c), ('ffma', a, b, c),
>> '!options->lower_ffma'),
>>     # Comparison simplifications
>> --
>> 2.5.0
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>
>