On Mon, Nov 7, 2011 at 5:32 PM, Michael Meissner <meiss...@linux.vnet.ibm.com> wrote:
> This patch rewrites the way builtins are handled in the rs6000 port so that > like the x86, when you do #pragma GCC target or attribute((target(...))) and > it > enables or disables the builtins based on the current target attributes. It > also defines or undefines the various hardware macros (__ARCH_<xxx>, __VSX__, > and __ALTIVEC__), and enables/disables the vector keyword. > > I've done bootstrap builds with make check, and there is one minor regression, > which I will fix as a bug (gcc.target/powerpc/recip-5.c fails due to not > vectorizing the recip builtin) after this patch goes in. I build a > linuxpaired > compiler and verified that the paired builtins are being handled correctly. > > I have some more ideas of cleanup (moving the large table in rs6000-c.c into > rs6000-builtins.def, and eliminating the various bdesc arrays, but for now, it > works, and hopefully is a bit cleaner than before). > > Is this ok to check in? > > [gcc] > 2011-11-07 Michael Meissner <meiss...@tlinux.vnet.ibm.com> > > * config/rs6000/rs6000-builtins.def: Completely rewrite builtin > handling to centralize more of the builtins in this file. Change > some builtin enumerations to be more consistant. Use a new mask > to hold the current builtins, including SPE and PAIRED builtins > which no longer are set via target_flags masks. Add > -mdebug=builtin debug support. For power machines, define all > Altivec and VSX buitins when the compiler starts, but don't allow > the use of a builtin unless the appropriate switch is used, or > #pragma GCC target is used to change the options. If the user > uses #pragma GCC target, update the appropriate hardware macros. > * common/config/rs6000/rs6000-common.c (rs6000_handle_option): > Ditto. > * config/rs6000/rs6000.opt (rs6000_builtin_mask): Ditto. > * config/rs6000/rs6000-c.c (rs6000_macro_to_expand): Ditto. > (rs6000_define_or_undefine_macro): Ditto. > (rs6000_target_modify_macros): Ditto. > (rs6000_cpu_cpp_builtins): Ditto. > (altivec_overloaded_builtins): Ditto. > (altivec_build_resolved_builtin): Ditto. > * config/rs6000/rs6000.c (rs6000_target_modify_macros_ptr): > Ditto. > (rs6000_builtin_info): Ditto. > (def_builtin): Ditto. > (enable_mask_for_builtins): Ditto. > (DEBUG_FMT_X): Ditto. > (rs6000_debug_reg_global): Ditto. > (rs6000_builtin_mask_calculate): Ditto. > (rs6000_option_override_internal): Ditto. > (rs6000_builtin_conversion): Ditto. > (rs6000_builtin_vectorized_function): Ditto. > (bdesc_3arg): Ditto. > (bdesc_dst): Ditto. > (bdesc_2arg): Ditto. > (builtin_description_predicates): Ditto. > (bdesc_altivec_preds): Ditto. > (bdesc_spe_predicates): Ditto. > (bdesc_spe_evsel): Ditto. > (bdesc_paired_preds): Ditto. > (bdesc_abs): Ditto. > (bdesc_1arg): Ditto. > (rs6000_overloaded_builtin_p): Ditto. > (rs6000_expand_unop_builtin): Ditto. > (bdesc_2arg_spe): Ditto. > (spe_expand_builtin): Ditto. > (rs6000_invalid_builtin): Ditto. > (rs6000_expand_builtin): Ditto. > (rs6000_init_builtins): Ditto. > (spe_init_builtins): Ditto. > (paired_init_builtins): Ditto. > (altivec_init_builtins): Ditto. > (builtin_function_type): Ditto. > (rs6000_common_init_builtins): Ditto. > (rs6000_builtin_reciprocal): Ditto. > (rs6000_builtin_mask_names): Ditto. > (rs6000_pragma_target_parse): Ditto. > (rs6000_function_specific_print): Ditto. > * config/rs6000/rs6000.h (MASK_DEBUG_BUILTIN): Ditto. > (MASK_DEBUG_ALL): Ditto. > (TARGET_DEBUG_BUILTIN): Ditto. > (TARGET_EXTRA_BUILTINS): Ditto. > (REGISTER_TARGET_PRAGMAS): Ditto. > (enum rs6000_btc): Ditto. > (RS6000_BTC_*): Ditto. > (RS6000_BTM_*): Ditto. > (enum rs6000_builtins): Ditto. > * config/rs6000/rs6000-protos.h (rs6000_overloaded_builtin_p): > Ditto. > (rs6000_target_modify_macros): Ditto. > (rs6000_target_modify_macros_ptr): Ditto. > > [gcc/testsuite] > 2011-11-07 Michael Meissner <meiss...@linux.vnet.ibm.com> > > * testsuite/gcc.target/powerpc/ppc-target-4.c: New test for > builtin functions being defined and undefined based on the #pragma > GCC target settings. This is okay. Thanks, David
