[Mesa-dev] [PATCH] glsl_to_tgsi: remove unnecessary dead code elimination pass

[Marek Olšák]

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

Marek

On Tue, May 6, 2014 at 5:40 AM, Bryan Cain <bryancain3 at gmail.com> wrote:
> With the more advanced dead code elimination pass already being run,
> eliminate_dead_code was making no difference in instruction count, and had
> an undesirable O(n^2) runtime. So remove it and rename
> eliminate_dead_code_advanced to eliminate_dead_code.
> ---
>  src/mesa/state_tracker/st_glsl_to_tgsi.cpp |   50 +++-------------------------
>  1 file changed, 5 insertions(+), 45 deletions(-)
>
> diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> index 6eb6c8a..b0e0782 100644
> --- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> +++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
> @@ -460,8 +460,7 @@ public:
>     int get_last_temp_write(int index);
>
>     void copy_propagate(void);
> -   void eliminate_dead_code(void);
> -   int eliminate_dead_code_advanced(void);
> +   int eliminate_dead_code(void);
>     void merge_registers(void);
>     void renumber_registers(void);
>
> @@ -3663,7 +3662,8 @@ glsl_to_tgsi_visitor::copy_propagate(void)
>  }
>
>  /*
> - * Tracks available PROGRAM_TEMPORARY registers for dead code elimination.
> + * On a basic block basis, tracks available PROGRAM_TEMPORARY registers for dead
> + * code elimination.
>   *
>   * The glsl_to_tgsi_visitor lazily produces code assuming that this pass
>   * will occur.  As an example, a TXP production after copy propagation but
> @@ -3676,48 +3676,9 @@ glsl_to_tgsi_visitor::copy_propagate(void)
>   * and after this pass:
>   *
>   * 0: TXP TEMP[2], INPUT[4].xyyw, texture[0], 2D;
> - *
> - * FIXME: assumes that all functions are inlined (no support for BGNSUB/ENDSUB)
> - * FIXME: doesn't eliminate all dead code inside of loops; it steps around them
> - */
> -void
> -glsl_to_tgsi_visitor::eliminate_dead_code(void)
> -{
> -   int i;
> -
> -   for (i=0; i < this->next_temp; i++) {
> -      int last_read = get_last_temp_read(i);
> -      int j = 0;
> -
> -      foreach_list_safe(node, &this->instructions) {
> -         glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *) node;
> -
> -         if (inst->dst.file == PROGRAM_TEMPORARY && inst->dst.index == i &&
> -             j > last_read)
> -         {
> -            inst->remove();
> -            delete inst;
> -         }
> -
> -         j++;
> -      }
> -   }
> -}
> -
> -/*
> - * On a basic block basis, tracks available PROGRAM_TEMPORARY registers for dead
> - * code elimination.  This is less primitive than eliminate_dead_code(), as it
> - * is per-channel and can detect consecutive writes without a read between them
> - * as dead code.  However, there is some dead code that can be eliminated by
> - * eliminate_dead_code() but not this function - for example, this function
> - * cannot eliminate an instruction writing to a register that is never read and
> - * is the only instruction writing to that register.
> - *
> - * The glsl_to_tgsi_visitor lazily produces code assuming that this pass
> - * will occur.
>   */
>  int
> -glsl_to_tgsi_visitor::eliminate_dead_code_advanced(void)
> +glsl_to_tgsi_visitor::eliminate_dead_code(void)
>  {
>     glsl_to_tgsi_instruction **writes = rzalloc_array(mem_ctx,
>                                                       glsl_to_tgsi_instruction *,
> @@ -5245,9 +5206,8 @@ get_mesa_program(struct gl_context *ctx,
>     /* Perform optimizations on the instructions in the glsl_to_tgsi_visitor. */
>     v->simplify_cmp();
>     v->copy_propagate();
> -   while (v->eliminate_dead_code_advanced());
> +   while (v->eliminate_dead_code());
>
> -   v->eliminate_dead_code();
>     v->merge_registers();
>     v->renumber_registers();
>
> --
> 1.7.9.5
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev