intel/compiler: Use nir_opt_conditional_discard

anv vkpipeline-db results for SKL:

total instructions in shared programs: 3622461 -> 3611281 (-0.31%)
instructions in affected programs: 396452 -> 385272 (-2.82%)
helped: 2062
HURT: 1

total cycles in shared programs: 1458144669 -> 1458105320 (<.01%)
cycles in affected programs: 4171830 -> 4132481 (-0.94%)
helped: 1874
HURT: 180

total loops in shared programs: 2437 -> 2437 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total spills in shared programs: 8745 -> 8748 (0.03%)
spills in affected programs: 8 -> 11 (37.50%)
helped: 1
HURT: 1

total fills in shared programs: 23392 -> 23395 (0.01%)
fills in affected programs: 8 -> 11 (37.50%)
helped: 1
HURT: 1

LOST:   0
GAINED: 1

No changes to shader-db on i965 or iris.  The glsl compiler already
does a similar optimization.

Improvement suggested by Daniel Schürmann.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>

diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
index 675fe026695dea431eef8f68e55006ae3a5edc19..ae81c65b94a34a48a4eb692d1f4313db7693e1b5 100644 (file)
--- a/src/intel/compiler/brw_nir.c
+++ b/src/intel/compiler/brw_nir.c
@@ -580,6 +580,7 @@ brw_nir_optimize(nir_shader *nir, const struct brw_compiler *compiler,
          OPT(nir_opt_dce);
       }
       OPT(nir_opt_if, false);
+      OPT(nir_opt_conditional_discard);
       if (nir->options->max_unroll_iterations != 0) {
          OPT(nir_opt_loop_unroll, indirect_mask);
       }