ir3,tu: Enable subgroup shuffles and relative shuffles

We still don't use the fast path for relative shuffles, that's left for
future work.

src/freedreno/ir3/ir3_nir.c

@@ -458,6 +458,8 @@ ir3_nir_post_finalize(struct ir3_shader *shader)
              .lower_vote_eq = true,
              .lower_subgroup_masks = true,
              .lower_read_invocation_to_cond = true,
+             .lower_shuffle = true,
+             .lower_relative_shuffle = true,
           });
    }
 

src/freedreno/vulkan/tu_device.c

@@ -830,7 +830,9 @@ tu_get_physical_device_properties_1_1(struct tu_physical_device *pdevice,
    p->subgroupSupportedStages = VK_SHADER_STAGE_COMPUTE_BIT;
    p->subgroupSupportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT |
                                     VK_SUBGROUP_FEATURE_VOTE_BIT |
-                                    VK_SUBGROUP_FEATURE_BALLOT_BIT;
+                                    VK_SUBGROUP_FEATURE_BALLOT_BIT |
+                                    VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
+                                    VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
    if (pdevice->info->a6xx.has_getfiberid) {
       p->subgroupSupportedStages |= VK_SHADER_STAGE_ALL_GRAPHICS;
       p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_QUAD_BIT;

src/freedreno/vulkan/tu_shader.c

@@ -82,6 +82,7 @@ tu_spirv_to_nir(struct tu_device *dev,
          .subgroup_ballot = true,
          .subgroup_vote = true,
          .subgroup_quad = true,
+         .subgroup_shuffle = true,
          .physical_storage_buffer_address = true,
       },
    };