radv: do not load the local invocation index when it's unused

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/amd/common/ac_nir_to_llvm.c

@@ -751,7 +751,8 @@ static void create_function(struct nir_to_llvm_context *ctx,
 					       &ctx->num_work_groups);
 		}
 		add_sgpr_argument(&args, ctx->ac.v3i32, &ctx->workgroup_ids);
-		add_sgpr_argument(&args, ctx->ac.i32, &ctx->tg_size);
+		if (ctx->shader_info->info.cs.uses_local_invocation_idx)
+			add_sgpr_argument(&args, ctx->ac.i32, &ctx->tg_size);
 		add_vgpr_argument(&args, ctx->ac.v3i32, &ctx->local_invocation_ids);
 		break;
 	case MESA_SHADER_VERTEX:

src/amd/common/ac_shader_info.c

@@ -58,6 +58,9 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, struct ac_shader_info *info)
 		}
 		break;
 	}
+	case nir_intrinsic_load_local_invocation_index:
+		info->cs.uses_local_invocation_idx = true;
+		break;
 	case nir_intrinsic_load_sample_id:
 		info->ps.force_persample = true;
 		break;

src/amd/common/ac_shader_info.h

@@ -45,6 +45,7 @@ struct ac_shader_info {
 		bool uses_grid_size;
 		bool uses_block_id[3];
 		bool uses_thread_id[3];
+		bool uses_local_invocation_idx;
 	} cs;
 };
 

src/amd/vulkan/radv_shader.c

@@ -400,7 +400,7 @@ radv_fill_shader_variant(struct radv_device *device,
 			S_00B84C_TGID_Z_EN(info->cs.uses_block_id[2]) |
 			S_00B84C_TIDIG_COMP_CNT(info->cs.uses_thread_id[2] ? 2 :
 						info->cs.uses_thread_id[1] ? 1 : 0) |
-			S_00B84C_TG_SIZE_EN(1) |
+			S_00B84C_TG_SIZE_EN(info->cs.uses_local_invocation_idx) |
 			S_00B84C_LDS_SIZE(variant->config.lds_size);
 		break;
 	}