[Mesa-dev] [PATCH 1/4] i965/fs: Generate better code for ir_triop_csel.

[Ian Romanick]

Yeah... I recall observing this when I was doing other bool related
optimization work, but I never got around to fixing it.

Reviewed-by: Ian Romanick <ian.d.romanick at intel.com>

On 10/15/2014 08:51 PM, Kenneth Graunke wrote:
> Previously, we generated an extra CMP instruction:
> 
>    cmp.ge.f0(8)   g4<1>D          g2<0,1,0>F      0F
>    cmp.nz.f0(8)   null            g4<8,8,1>D      0D
>    (+f0) sel(8)   g120<1>F        g2.4<0,1,0>F    g3<0,1,0>F
> 
> The first operand is always a boolean, and we want to predicate the SEL
> on that.  Rather than producing a boolean value and comparing it against
> zero, we can just produce a condition code in the flag register.
> 
> Now we generate:
> 
>    cmp.ge.f0(8)    null            g2<0,1,0>F      0F
>    (+f0) sel(8)    g124<1>F        g2.4<0,1,0>F    g3<0,1,0>F
> 
> total instructions in shared programs: 5473459 -> 5473253 (-0.00%)
> instructions in affected programs:     6219 -> 6013 (-3.31%)
> 
> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
> ---
>  src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 18 +++++++++++++-----
>  1 file changed, 13 insertions(+), 5 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> index 8a11b04..3ba5492 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> @@ -487,6 +487,19 @@ fs_visitor::visit(ir_expression *ir)
>           return;
>        break;
>  
> +   case ir_triop_csel:
> +      ir->operands[1]->accept(this);
> +      op[1] = this->result;
> +      ir->operands[2]->accept(this);
> +      op[2] = this->result;
> +
> +      emit_bool_to_cond_code(ir->operands[0]);
> +
> +      this->result = fs_reg(this, ir->type);
> +      inst = emit(SEL(this->result, op[1], op[2]));
> +      inst->predicate = BRW_PREDICATE_NORMAL;
> +      return;
> +
>     case ir_unop_interpolate_at_centroid:
>     case ir_binop_interpolate_at_offset:
>     case ir_binop_interpolate_at_sample:
> @@ -1023,11 +1036,6 @@ fs_visitor::visit(ir_expression *ir)
>        break;
>  
>     case ir_triop_csel:
> -      emit(CMP(reg_null_d, op[0], fs_reg(0), BRW_CONDITIONAL_NZ));
> -      inst = emit(BRW_OPCODE_SEL, this->result, op[1], op[2]);
> -      inst->predicate = BRW_PREDICATE_NORMAL;
> -      break;
> -
>     case ir_unop_interpolate_at_centroid:
>     case ir_binop_interpolate_at_offset:
>     case ir_binop_interpolate_at_sample:
>