/* Given a gpu_prog "prog" and the corresponding transformed AST * "tree", print the entire CUDA code to "p". * "types" collects the types for which a definition has already * been printed. */ static __isl_give isl_printer *print_cuda(__isl_take isl_printer *p, struct gpu_prog *prog, __isl_keep isl_ast_node *tree, struct gpu_types *types, void *user) { struct cuda_info *cuda = user; isl_printer *kernel; kernel = isl_printer_to_file(isl_printer_get_ctx(p), cuda->kernel_c); kernel = isl_printer_set_output_format(kernel, ISL_FORMAT_C); kernel = gpu_print_types(kernel, types, prog); isl_printer_free(kernel); if (!kernel) return isl_printer_free(p); texture_optimization(prog,tree); p = print_host_code(p, prog, tree, cuda); isl_printer *tp = isl_printer_to_file(isl_printer_get_ctx(p), cuda->kernel_h); tp = isl_printer_set_output_format(tp, ISL_FORMAT_C); print_texture_or_surface_decl(tp, prog); isl_printer_free(tp); printf("\n Clearing done"); fflush(stdout); // Clearing is_texture or is_surface required clear_texture_flags(prog); return p; }
/* Given a gpu_prog "prog" and the corresponding transformed AST * "tree", print the entire OpenCL code to "p". */ static __isl_give isl_printer *print_opencl(__isl_take isl_printer *p, struct gpu_prog *prog, __isl_keep isl_ast_node *tree, struct gpu_types *types, void *user) { struct opencl_info *opencl = user; opencl->kprinter = isl_printer_set_output_format(opencl->kprinter, ISL_FORMAT_C); if (any_double_elements(prog)) opencl->kprinter = opencl_enable_double_support( opencl->kprinter); if (opencl->options->opencl_print_kernel_types) opencl->kprinter = gpu_print_types(opencl->kprinter, types, prog); if (!opencl->kprinter) return isl_printer_free(p); p = ppcg_start_block(p); p = opencl_print_host_macros(p); p = gpu_print_local_declarations(p, prog); p = opencl_declare_device_arrays(p, prog); p = opencl_setup(p, opencl->input, opencl); p = opencl_allocate_device_arrays(p, prog); p = opencl_print_host_code(p, prog, tree, opencl); p = opencl_release_device_arrays(p, prog); p = opencl_release_cl_objects(p, opencl); p = ppcg_end_block(p); return p; }