broadcom/compiler: use skip_helpers with textures, UBOs and SSBOs

Set the per-pixel mask based on the value of skip_helpers.

This slightly increase the performance on several traces.

fps_avg  helped:  gl_gfxbench_trex.trace:                           22.30 -> 22.79 (2.20%)

total fps_avg in all runs: 55.18 -> 55.71 (0.97%)
total fps_avg in affected (through threshold) runs: 22.30 -> 22.79 (2.20%)
helped: 1
HURT: 0

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38759>

src/broadcom/compiler/nir_to_vir.c

@@ -651,18 +651,14 @@ ntq_emit_tmu_general(struct v3d_compile *c, nir_intrinsic_instr *instr,
                                 v3d_tmu_get_type_from_op(tmu_op, !is_load) ==
                                 V3D_TMU_OP_TYPE_ATOMIC;
 
-                        /* Only load per-quad if we can be certain that all
-                         * lines in the quad are active. Notice that demoted
-                         * invocations, unlike terminated ones, are still
-                         * active: we want to skip memory writes for them but
-                         * loads should still work.
+                        /* Only load per-quad if we can't skip helper
+                         * invocations.
                          */
                         uint32_t perquad =
-                                is_load && !vir_in_nonuniform_control_flow(c) &&
-                                ((c->s->info.stage == MESA_SHADER_FRAGMENT &&
-                                  c->s->info.fs.needs_coarse_quad_helper_invocations &&
-                                  !c->emitted_discard) ||
-                                 c->s->info.uses_wide_subgroup_intrinsics) ?
+                                is_load &&
+                                c->s->info.stage == MESA_SHADER_FRAGMENT &&
+                                nir_intrinsic_has_access(instr) &&
+                                !(nir_intrinsic_access(instr) & ACCESS_SKIP_HELPERS) ?
                                 GENERAL_TMU_LOOKUP_PER_QUAD :
                                 GENERAL_TMU_LOOKUP_PER_PIXEL;
                         config = 0xffffff00 | tmu_op << 3 | perquad;

src/broadcom/compiler/v3d_tex.c

@@ -307,12 +307,15 @@ v3d_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
 
         /* To handle the cases were we can't just use p1_unpacked_default */
         bool non_default_p1_config = nir_tex_instr_need_sampler(instr) ||
-                output_type_32_bit;
+                output_type_32_bit ||
+                (c->s->info.stage == MESA_SHADER_FRAGMENT &&
+                 !instr->skip_helpers);
 
         if (non_default_p1_config) {
                 struct V3D42_TMU_CONFIG_PARAMETER_1 p1_unpacked = {
                         .output_type_32_bit = output_type_32_bit,
-
+                        .per_pixel_mask_enable = (c->s->info.stage != MESA_SHADER_FRAGMENT ||
+                                                  instr->skip_helpers),
                         .unnormalized_coordinates = (instr->sampler_dim ==
                                                      GLSL_SAMPLER_DIM_RECT),
                 };

src/broadcom/compiler/vir.c

@@ -1768,6 +1768,13 @@ intrinsic_try_skip_helpers(nir_intrinsic_instr *intr, UNUSED void *data)
 {
         switch(intr->intrinsic) {
                 case nir_intrinsic_image_load:
+                case nir_intrinsic_load_uniform:
+                case nir_intrinsic_load_ubo:
+                case nir_intrinsic_load_ssbo:
+                case nir_intrinsic_load_scratch:
+                case nir_intrinsic_load_shared:
+                case nir_intrinsic_load_global:
+                case nir_intrinsic_load_global_constant:
                         return true;
                 default:
                         return false;