broadcom/compiler: use nir_lower_undef_to_zero

total instructions in shared programs: 13731663 -> 13721549 (-0.07%)
instructions in affected programs: 98242 -> 88128 (-10.29%)
helped: 191
HURT: 131
Instructions are helped.

total threads in shared programs: 412272 -> 412296 (<.01%)
threads in affected programs: 24 -> 48 (100.00%)
helped: 12
HURT: 0
Threads are helped.

total uniforms in shared programs: 3780693 -> 3779137 (-0.04%)
uniforms in affected programs: 10564 -> 9008 (-14.73%)
helped: 114
HURT: 7
Uniforms are helped.

total max-temps in shared programs: 2319942 -> 2319528 (-0.02%)
max-temps in affected programs: 4191 -> 3777 (-9.88%)
helped: 113
HURT: 22
Max-temps are helped.

total sfu-stalls in shared programs: 31584 -> 31616 (0.10%)
sfu-stalls in affected programs: 217 -> 249 (14.75%)
helped: 51
HURT: 54
Inconclusive result (value mean confidence interval includes 0).

total inst-and-stalls in shared programs: 13763247 -> 13753165 (-0.07%)
inst-and-stalls in affected programs: 98719 -> 88637 (-10.21%)
helped: 187
HURT: 134
Inst-and-stalls are helped.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9681>

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 1dcdbe7..cd806bf 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -1824,6 +1824,7 @@ v3d_optimize_nir(struct nir_shader *s)
                 }
 
                 NIR_PASS(progress, s, nir_opt_undef);
+                NIR_PASS(progress, s, nir_lower_undef_to_zero);
         } while (progress);
 
         nir_move_options sink_opts =
@@ -2159,18 +2160,6 @@ ntq_emit_load_const(struct v3d_compile *c, nir_load_const_instr *instr)
 }
 
 static void
-ntq_emit_ssa_undef(struct v3d_compile *c, nir_ssa_undef_instr *instr)
-{
-        struct qreg *qregs = ntq_init_ssa_def(c, &instr->def);
-
-        /* VIR needs there to be *some* value, so pick 0 (same as for
-         * ntq_setup_registers().
-         */
-        for (int i = 0; i < instr->def.num_components; i++)
-                qregs[i] = vir_uniform_ui(c, 0);
-}
-
-static void
 ntq_emit_image_size(struct v3d_compile *c, nir_intrinsic_instr *instr)
 {
         unsigned image_index = nir_src_as_uint(instr->src[0]);
@@ -3384,7 +3373,7 @@ ntq_emit_instr(struct v3d_compile *c, nir_instr *instr)
                 break;
 
         case nir_instr_type_ssa_undef:
-                ntq_emit_ssa_undef(c, nir_instr_as_ssa_undef(instr));
+                unreachable("Should've been lowered by nir_lower_undef_to_zero");
                 break;
 
         case nir_instr_type_tex: