[Mesa-dev] [PATCH] i965/fs: Add support for removing NOT.NZ and NOT.Z instructions.

[Ian Romanick]

This was not fully baked.  I'll send out a fixed version later.

On 05/08/2015 07:05 PM, Ian Romanick wrote:
> From: Ian Romanick <ian.d.romanick at intel.com>
> 
> Shader-db results:
> 
> GM45 and Iron Lake:
> total instructions in shared programs: 7888585 -> 7888585 (0.00%)
> instructions in affected programs:     0 -> 0
> 
> Sandy Bridge, Ivy Bridge, Haswell, and Broadwell:
> total instructions in shared programs: 9598608 -> 9598572 (-0.00%)
> instructions in affected programs:     6506 -> 6470 (-0.55%)
> helped:                                36
> 
> Signed-off-by: Ian Romanick <ian.d.romanick at intel.com>
> ---
>  src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp b/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp
> index 469f2ea..d72a83a 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp
> @@ -59,7 +59,8 @@ opt_cmod_propagation_local(bblock_t *block)
>  
>        if ((inst->opcode != BRW_OPCODE_AND &&
>             inst->opcode != BRW_OPCODE_CMP &&
> -           inst->opcode != BRW_OPCODE_MOV) ||
> +           inst->opcode != BRW_OPCODE_MOV &&
> +           inst->opcode != BRW_OPCODE_NOT) ||
>            inst->predicate != BRW_PREDICATE_NONE ||
>            !inst->dst.is_null() ||
>            inst->src[0].file != GRF ||
> @@ -86,6 +87,11 @@ opt_cmod_propagation_local(bblock_t *block)
>            inst->conditional_mod != BRW_CONDITIONAL_NZ)
>           continue;
>  
> +      if (inst->opcode == BRW_OPCODE_NOT &&
> +          inst->conditional_mod != BRW_CONDITIONAL_Z &&
> +          inst->conditional_mod != BRW_CONDITIONAL_NZ)
> +         continue;
> +
>        bool read_flag = false;
>        foreach_inst_in_block_reverse_starting_from(fs_inst, scan_inst, inst,
>                                                    block) {
>