broadcom/compiler: implement more subgroup intrinsics

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11620>

src/broadcom/compiler/nir_to_vir.c

@@ -3241,6 +3241,57 @@ ntq_emit_intrinsic(struct v3d_compile *c, nir_intrinsic_instr *instr)
                 break;
         }
 
+        case nir_intrinsic_load_subgroup_size:
+                ntq_store_dest(c, &instr->dest, 0,
+                               vir_uniform_ui(c, V3D_CHANNELS));
+                break;
+
+        case nir_intrinsic_load_subgroup_invocation:
+                ntq_store_dest(c, &instr->dest, 0, vir_EIDX(c));
+                break;
+
+        case nir_intrinsic_elect: {
+                /* Flafirst reads A flags so we need to make sure it reflects
+                 * currently active lanes.
+                 *
+                 * MSF returns 0 for disabled lanes in compute shaders so
+                 * PUSHZ will set A=1 for disabled lanes. We want the inverse
+                 * of this but we don't have any means to negate the A flags
+                 * directly, but we can do it by repeating the same operation
+                 * with NORZ (A = ~A & ~Z).
+                 */
+                assert(c->s->info.stage == MESA_SHADER_COMPUTE);
+                vir_set_pf(c, vir_MSF_dest(c, vir_nop_reg()), V3D_QPU_PF_PUSHZ);
+                vir_set_uf(c, vir_MSF_dest(c, vir_nop_reg()), V3D_QPU_UF_NORZ);
+
+                /* If we are under non-uniform control flow we also need to
+                 * AND the A flags with the current execute mask.
+                 */
+                if (vir_in_nonuniform_control_flow(c)) {
+                        const uint32_t bidx = c->cur_block->index;
+                        vir_set_uf(c, vir_XOR_dest(c, vir_nop_reg(),
+                                                    c->execute,
+                                                    vir_uniform_ui(c, bidx)),
+                                   V3D_QPU_UF_ANDZ);
+                }
+
+                struct qreg first = vir_FLAFIRST(c);
+
+                /* Produce a boolean result from Flafirst */
+                vir_set_pf(c, vir_XOR_dest(c, vir_nop_reg(),
+                                           first, vir_uniform_ui(c, 1)),
+                                           V3D_QPU_PF_PUSHZ);
+                struct qreg result =
+                        vir_MOV(c, vir_SEL(c, V3D_QPU_COND_IFA,
+                                           vir_uniform_ui(c, ~0),
+                                           vir_uniform_ui(c, 0)));
+                c->flags_temp = result.index;
+                c->flags_cond = V3D_QPU_COND_IFA;
+
+                ntq_store_dest(c, &instr->dest, 0, result);
+                break;
+        }
+
         case nir_intrinsic_load_num_subgroups:
                 unreachable("Should have been lowered");
                 break;