[Mesa-dev] [PATCH 4/4] nir: detect more induction variables

[Eero Tamminen]

Hi,

On 28.11.2018 5.25, Timothy Arceri wrote:
> This adds allows loop analysis to detect inductions varibales that
> are incremented in both branches of an if rather than in a main
> loop block. For example:
[...]> Unfortunatly GCM could move the addition out of the if for us
> (making this patch unrequired) but we still cannot enable the GCM
> pass without regressions.

Maybe there should be a TODO comment about removing this code when GCM 
is added?


	- Eero

> This unrolls a loop in Rise of The Tomb Raider.
> 
> vkpipeline-db results (VEGA):
> 
> Totals from affected shaders:
> SGPRS: 88 -> 96 (9.09 %)
> VGPRS: 56 -> 52 (-7.14 %)
> Spilled SGPRs: 0 -> 0 (0.00 %)
> Spilled VGPRs: 0 -> 0 (0.00 %)
> Private memory VGPRs: 0 -> 0 (0.00 %)
> Scratch size: 0 -> 0 (0.00 %) dwords per thread
> Code Size: 2168 -> 4560 (110.33 %) bytes
> LDS: 0 -> 0 (0.00 %) blocks
> Max Waves: 4 -> 4 (0.00 %)
> Wait states: 0 -> 0 (0.00 %)
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=32211
> ---
>   src/compiler/nir/nir_loop_analyze.c | 36 +++++++++++++++++++++++++++++
>   1 file changed, 36 insertions(+)
> 
> diff --git a/src/compiler/nir/nir_loop_analyze.c b/src/compiler/nir/nir_loop_analyze.c
> index 8903e15105..cf97d6bf06 100644
> --- a/src/compiler/nir/nir_loop_analyze.c
> +++ b/src/compiler/nir/nir_loop_analyze.c
> @@ -245,6 +245,42 @@ compute_induction_information(loop_info_state *state)
>            if (src_var->in_if_branch || src_var->in_nested_loop)
>               break;
>   
> +         /* Detect inductions varibales that are incremented in both branches
> +          * of an unnested if rather than in a loop block.
> +          */
> +         if (is_var_phi(src_var)) {
> +            nir_phi_instr *src_phi =
> +               nir_instr_as_phi(src_var->def->parent_instr);
> +
> +            nir_op alu_op;
> +            nir_ssa_def *alu_srcs[2] = {0};
> +            nir_foreach_phi_src(src2, src_phi) {
> +               nir_loop_variable *src_var2 =
> +                  get_loop_var(src2->src.ssa, state);
> +
> +               if (!src_var2->in_if_branch || !is_var_alu(src_var2))
> +                  break;
> +
> +               nir_alu_instr *alu =
> +                  nir_instr_as_alu(src_var2->def->parent_instr);
> +               if (nir_op_infos[alu->op].num_inputs != 2)
> +                  break;
> +
> +               if (alu->src[0].src.ssa == alu_srcs[0] &&
> +                   alu->src[1].src.ssa == alu_srcs[1] &&
> +                   alu->op == alu_op) {
> +                  /* Both branches perform the same calculation so we can use
> +                   * one of them to find the induction variable.
> +                   */
> +                  src_var = src_var2;
> +               } else {
> +                  alu_srcs[0] = alu->src[0].src.ssa;
> +                  alu_srcs[1] = alu->src[1].src.ssa;
> +                  alu_op = alu->op;
> +               }
> +            }
> +         }
> +
>            if (!src_var->in_loop) {
>               biv->def_outside_loop = src_var;
>            } else if (is_var_alu(src_var)) {
>