[Mesa-dev] [PATCH 06/10] i965: Relax restriction on scheduling last instruction.

[Francisco Jerez]

Matt Turner <mattst88 at gmail.com> writes:

> I think when this code was written, basic blocks were always ended by a
> control flow instruction or an end-of-thread message. That's no longer
> the case, and removing this restriction actually helps things:
>
>    instructions in affected programs: 7267 -> 7244 (-0.32%)
>    helped: 4
>
>    total cycles in shared programs: 66559580 -> 66431900 (-0.19%)
>    cycles in affected programs: 28310152 -> 28182472 (-0.45%)
>    helped: 9577
>    HURT: 879
>
>    GAINED: 2
>
> The addition of the is_control_flow() checks is not a functional change,
> since the add_insts_from_block() does not put them in the list of
> instructions to schedule. I plan to change this in a later patch.

Reviewed-by: Francisco Jerez <currojerez at riseup.net>

> ---
>  .../drivers/dri/i965/brw_schedule_instructions.cpp | 23 +++-------------------
>  1 file changed, 3 insertions(+), 20 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> index 66eb07e..46b45a5 100644
> --- a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> @@ -923,15 +923,6 @@ fs_instruction_scheduler::calculate_deps()
>      */
>     schedule_node *last_fixed_grf_write = NULL;
>  
> -   /* The last instruction always needs to still be the last
> -    * instruction.  Either it's flow control (IF, ELSE, ENDIF, DO,
> -    * WHILE) and scheduling other things after it would disturb the
> -    * basic block, or it's FB_WRITE and we should do a better job at
> -    * dead code elimination anyway.
> -    */
> -   schedule_node *last = (schedule_node *)instructions.get_tail();
> -   add_barrier_deps(last);
> -
>     memset(last_grf_write, 0, sizeof(last_grf_write));
>     memset(last_mrf_write, 0, sizeof(last_mrf_write));
>  
> @@ -940,7 +931,8 @@ fs_instruction_scheduler::calculate_deps()
>        fs_inst *inst = (fs_inst *)n->inst;
>  
>        if (inst->opcode == FS_OPCODE_PLACEHOLDER_HALT ||
> -         inst->has_side_effects())
> +          inst->is_control_flow() ||
> +          inst->has_side_effects())
>           add_barrier_deps(n);
>  
>        /* read-after-write deps. */
> @@ -1166,15 +1158,6 @@ vec4_instruction_scheduler::calculate_deps()
>      */
>     schedule_node *last_fixed_grf_write = NULL;
>  
> -   /* The last instruction always needs to still be the last instruction.
> -    * Either it's flow control (IF, ELSE, ENDIF, DO, WHILE) and scheduling
> -    * other things after it would disturb the basic block, or it's the EOT
> -    * URB_WRITE and we should do a better job at dead code eliminating
> -    * anything that could have been scheduled after it.
> -    */
> -   schedule_node *last = (schedule_node *)instructions.get_tail();
> -   add_barrier_deps(last);
> -
>     memset(last_grf_write, 0, sizeof(last_grf_write));
>     memset(last_mrf_write, 0, sizeof(last_mrf_write));
>  
> @@ -1182,7 +1165,7 @@ vec4_instruction_scheduler::calculate_deps()
>     foreach_in_list(schedule_node, n, &instructions) {
>        vec4_instruction *inst = (vec4_instruction *)n->inst;
>  
> -      if (inst->has_side_effects())
> +      if (inst->is_control_flow() || inst->has_side_effects())
>           add_barrier_deps(n);
>  
>        /* read-after-write deps. */
> -- 
> 2.4.10
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160314/12a9ceb3/attachment.sig>