[Mesa-dev] [PATCH 1/2] nir: Replace fcsel with fmov instead of imov in nir_opt_undef

[Boyan Ding]

2015-08-22 5:36 GMT+08:00 Matt Turner <mattst88 at gmail.com>:
> On Fri, Aug 21, 2015 at 6:42 AM, Boyan Ding <boyan.j.ding at gmail.com> wrote:
>> Signed-off-by: Boyan Ding <boyan.j.ding at gmail.com>
>> ---
>>  src/glsl/nir/nir_opt_undef.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/src/glsl/nir/nir_opt_undef.c b/src/glsl/nir/nir_opt_undef.c
>> index 4ab27a8..fb9c6bb 100644
>> --- a/src/glsl/nir/nir_opt_undef.c
>> +++ b/src/glsl/nir/nir_opt_undef.c
>> @@ -63,7 +63,7 @@ opt_undef_alu(nir_alu_instr *instr)
>>        memset(&empty_src, 0, sizeof(empty_src));
>>        nir_instr_rewrite_src(&instr->instr, &instr->src[1].src, empty_src);
>>        nir_instr_rewrite_src(&instr->instr, &instr->src[2].src, empty_src);
>> -      instr->op = nir_op_imov;
>> +      instr->op = instr->op == nir_op_bcsel ? nir_op_imov : nir_op_fmov;
>>
>>        return true;
>>     }
>> --
>
> I don't think this is right (or was a problem). The b in bcsel simply
> means that the condition in a boolean, whereas fcsel means that the
> condition is a float (i.e., it performs the operation (src0 != 0.0f) ?
> src1 : src2). So you cannot determine the types of src1 or src2 from
> this.

I understand now. Sorry for the noise.

Thanks,
Boyan Ding