v3d: Enable the late algebraic optimizations to get real subs.

This worked better than my original v3d-local pass for just subs, and is a
huge win over not producing subs.

total instructions in shared programs: 6408469 -> 6167932 (-3.75%)
total threads in shared programs: 153784 -> 154104 (0.21%)
total uniforms in shared programs: 2157078 -> 1905823 (-11.65%)
total max-temps in shared programs: 904546 -> 895796 (-0.97%)
total spills in shared programs: 4959 -> 4993 (0.69%)
total fills in shared programs: 6558 -> 6670 (1.71%)
total sfu-stalls in shared programs: 25845 -> 25175 (-2.59%)
total inst-and-stalls in shared programs: 6434314 -> 6193107 (-3.75%)

Reviewed-by: Daniel Schürmann <[email protected]>
Reviewed-by: Connor Abbott <[email protected]>

1 files changed, 16 insertions, 0 deletions

diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 917ccfeaef1..802448b1e03 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -939,6 +939,22 @@ uint64_t *v3d_compile(const struct v3d_compiler *compiler,
         NIR_PASS_V(c->s, nir_lower_idiv);
 
         v3d_optimize_nir(c->s);
+
+        /* Do late algebraic optimization to turn add(a, neg(b)) back into
+         * subs, then the mandatory cleanup after algebraic.  Note that it may
+         * produce fnegs, and if so then we need to keep running to squash
+         * fneg(fneg(a)).
+         */
+        bool more_late_algebraic = true;
+        while (more_late_algebraic) {
+                more_late_algebraic = false;
+                NIR_PASS(more_late_algebraic, c->s, nir_opt_algebraic_late);
+                NIR_PASS_V(c->s, nir_opt_constant_folding);
+                NIR_PASS_V(c->s, nir_copy_prop);
+                NIR_PASS_V(c->s, nir_opt_dce);
+                NIR_PASS_V(c->s, nir_opt_cse);
+        }
+
         NIR_PASS_V(c->s, nir_lower_bool_to_int32);
         NIR_PASS_V(c->s, nir_convert_from_ssa, true);