[Mesa-dev] [PATCH 2/4] i965: Use greater-equal cmod to implement maximum.

[Ian Romanick]

This patch is

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

On 02/10/2015 11:09 AM, Matt Turner wrote:
> The docs specifically call out SEL with .l and .ge as the
> implementations of MIN and MAX respectively. Among other things, SEL
> with these conditional mods are commutative.
> ---
>  src/mesa/drivers/dri/i965/brw_fs_visitor.cpp      | 5 ++++-
>  src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp    | 6 +++---
>  src/mesa/drivers/dri/i965/brw_vec4_vp.cpp         | 4 ++--
>  src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp | 2 +-
>  4 files changed, 10 insertions(+), 7 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> index 6cddcf5..bca2139 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> @@ -322,6 +322,9 @@ void
>  fs_visitor::emit_minmax(enum brw_conditional_mod conditionalmod, const fs_reg &dst,
>                          const fs_reg &src0, const fs_reg &src1)
>  {
> +   assert(conditionalmod == BRW_CONDITIONAL_GE ||
> +          conditionalmod == BRW_CONDITIONAL_L);
> +
>     fs_inst *inst;
>  
>     if (brw->gen >= 6) {
> @@ -1948,7 +1951,7 @@ fs_visitor::rescale_texcoord(fs_reg coordinate, int coord_components,
>  	    chan = offset(chan, i);
>  
>  	    inst = emit(BRW_OPCODE_SEL, chan, chan, fs_reg(0.0f));
> -	    inst->conditional_mod = BRW_CONDITIONAL_G;
> +	    inst->conditional_mod = BRW_CONDITIONAL_GE;
>  
>  	    /* Our parameter comes in as 1.0/width or 1.0/height,
>  	     * because that's what people normally want for doing
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
> index e6a7ed0..d13c716 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
> @@ -513,7 +513,7 @@ vec4_visitor::emit_unpack_snorm_4x8(const dst_reg &dst, src_reg src0)
>     emit(MUL(scaled, src_reg(f), src_reg(1.0f / 127.0f)));
>  
>     dst_reg max(this, glsl_type::vec4_type);
> -   emit_minmax(BRW_CONDITIONAL_G, max, src_reg(scaled), src_reg(-1.0f));
> +   emit_minmax(BRW_CONDITIONAL_GE, max, src_reg(scaled), src_reg(-1.0f));
>     emit_minmax(BRW_CONDITIONAL_L, dst, src_reg(max), src_reg(1.0f));
>  }
>  
> @@ -541,7 +541,7 @@ void
>  vec4_visitor::emit_pack_snorm_4x8(const dst_reg &dst, const src_reg &src0)
>  {
>     dst_reg max(this, glsl_type::vec4_type);
> -   emit_minmax(BRW_CONDITIONAL_G, max, src0, src_reg(-1.0f));
> +   emit_minmax(BRW_CONDITIONAL_GE, max, src0, src_reg(-1.0f));
>  
>     dst_reg min(this, glsl_type::vec4_type);
>     emit_minmax(BRW_CONDITIONAL_L, min, src_reg(max), src_reg(1.0f));
> @@ -1673,7 +1673,7 @@ vec4_visitor::visit(ir_expression *ir)
>        emit_minmax(BRW_CONDITIONAL_L, result_dst, op[0], op[1]);
>        break;
>     case ir_binop_max:
> -      emit_minmax(BRW_CONDITIONAL_G, result_dst, op[0], op[1]);
> +      emit_minmax(BRW_CONDITIONAL_GE, result_dst, op[0], op[1]);
>        break;
>  
>     case ir_binop_pow:
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp b/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
> index 434f032..e2d4b7c 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_vp.cpp
> @@ -227,7 +227,7 @@ vec4_vs_visitor::emit_program_code()
>                 /* if (tmp.y < 0) tmp.y = 0; */
>                 src_reg tmp_y = swizzle(src[0], BRW_SWIZZLE_YYYY);
>                 result.writemask = WRITEMASK_Z;
> -               emit_minmax(BRW_CONDITIONAL_G, result, tmp_y, src_reg(0.0f));
> +               emit_minmax(BRW_CONDITIONAL_GE, result, tmp_y, src_reg(0.0f));
>  
>                 src_reg clamped_y(result);
>                 clamped_y.swizzle = BRW_SWIZZLE_ZZZZ;
> @@ -314,7 +314,7 @@ vec4_vs_visitor::emit_program_code()
>        }
>  
>        case OPCODE_MAX:
> -         emit_minmax(BRW_CONDITIONAL_G, dst, src[0], src[1]);
> +         emit_minmax(BRW_CONDITIONAL_GE, dst, src[0], src[1]);
>           break;
>  
>        case OPCODE_MIN:
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp
> index 72b6ef0..a48b730 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_vs_visitor.cpp
> @@ -97,7 +97,7 @@ vec4_vs_visitor::emit_prolog()
>                 dst.type = brw_type_for_base_type(glsl_type::vec4_type);
>                 emit(MOV(dst, src_reg(reg_d)));
>                 emit(MUL(dst, src_reg(dst), src_reg(es3_normalize_factor)));
> -               emit_minmax(BRW_CONDITIONAL_G, dst, src_reg(dst), src_reg(-1.0f));
> +               emit_minmax(BRW_CONDITIONAL_GE, dst, src_reg(dst), src_reg(-1.0f));
>              } else {
>                 /* The following equations are from the OpenGL 3.2 specification:
>                  *
>