Yeah, that looks better. V2 R-B me.
--Jason
On Wed, Mar 18, 2015 at 2:56 PM, Matt Turner <matts...@gmail.com> wrote:
> total instructions in shared programs: 6263270 -> 6203091 (-0.96%)
> instructions in affected programs: 2606529 -> 2546350 (-2.31%)
> helped: 14301
> GAINED: 5
> LOST: 3
>
> Revewed-by: Jason Ekstrand <jason.ekstr...@intel.com> [v1]
> ---
> .../drivers/dri/i965/brw_fs_cmod_propagation.cpp | 25
> +++++++++++-----------
> 1 file changed, 13 insertions(+), 12 deletions(-)
>
> 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 1935f06..798fef3 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_cmod_propagation.cpp
> @@ -94,21 +94,22 @@ opt_cmod_propagation_local(bblock_t *block)
> scan_inst->dst.reg_offset != inst->src[0].reg_offset)
> break;
>
> - /* This must be done before the dst.type check because the result
> - * type of the AND will always be D, but the result of the CMP
> - * could be anything. The assumption is that the AND is just
> - * figuring out what the result of the previous comparison was
> - * instead of doing a new comparison with a different type.
> - */
> - if (inst->opcode == BRW_OPCODE_AND) {
> - if (scan_inst->opcode == BRW_OPCODE_CMP) {
> - inst->remove(block);
> - progress = true;
> - }
> -
> + /* CMP's result is the same regardless of dest type. */
> + if (inst->conditional_mod == BRW_CONDITIONAL_NZ &&
> + scan_inst->opcode == BRW_OPCODE_CMP &&
> + (inst->dst.type == BRW_REGISTER_TYPE_D ||
> + inst->dst.type == BRW_REGISTER_TYPE_UD)) {
> + inst->remove(block);
> + progress = true;
> break;
> }
>
> + /* If the AND wasn't handled by the previous case, it isn't safe
> + * to remove it.
> + */
> + if (inst->opcode == BRW_OPCODE_AND)
> + break;
> +
> /* Comparisons operate differently for ints and floats */
> if (scan_inst->dst.type != inst->dst.type)
> break;
> --
> 2.0.5
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
