[PATCH]: RFC: Add power7 support to the rs6000 (part 11 of 12)
Michael Meissner
meissner@linux.vnet.ibm.com
Thu Jun 4 21:53:00 GMT 2009
Note, I needed to split rs6000.c into two patches, the first part was submitted
in part 10. The offsets in the second patch will offset by 1017 lines due to
the lines added in the first patch.
2009-06-04 Michael Meissner <meissner@linux.vnet.ibm.com>
Pat Haugen <pthaugen@us.ibm.com>
Revital1 Eres <ERES@il.ibm.com>
* config/rs6000/rs6000.c (def_builtin): Replace an abort with a
call to fatal_error.
(bdesc_3arg): Add VSX builtins.
(bdesc_2arg): Ditto.
(bdesc_altivec_preds): Ditto.
(bdesc_abs): Ditto.
(bdesc_1arg): Ditto.
(altivec_expand_predicate_builtin): Rewrite how vector predicates
are generated. Add VSX support.
(rs6000_expand_ternop_builtin): Add VSX support. Repalce if tests
with a switch.
(altivec_expand_ld_builtin): Change to use vector memory expanders
instead of altivec insns.
(altivec_expand_st_builtin): Ditto.
(altivec_expand_builtin): Add VSX support. Change how vector
predicates are defined.
(rs6000_expand_builtin): Delete some trailing whitespace. Add VSX
builtin support.
(rs6000_init_builtins): Add VSX support. Switch to use
builtin_function_type to create builtin function types.
(altivec_init_builtins): Add VSX support.
(rs6000_common_init_builtins): Ditto.
(rs6000_secondary_memory_needed_rtx): Add -mdebug=addr support.
(rs6000_reload_register_type): Simplify the register types for
secondary reload.
(rs6000_secondary_reload): New function to deal with Altivec/VSX
types only providing reg+reg addressing in the vector unit, and
needing reg+offset addressing if they get moved to the GPRs.
(rs6000_secondary_reload_inner): New function for secondary reload
support.
(rs6000_preferred_reload_class): Move PREFERRED_RELOAD_CLASS code
here from the macro. Add VSX support.
(rs6000_secondary_memory_needed): Move SECONDARY_MEMORY_NEEDED
code here from the macro. Add VSX support.
(rs6000_secondary_reload_class): Make static. Add VSX support.
(print_operand): Fix error message to have correct string. Add %x
to print a register as a VSX register instead of a float or
Altivec register. Add VSX support.
(rs6000_emit_vector_compare_inner): Rewrite vector compare and
select support. Add VSX.
(rs6000_emit_vector_compare): Ditto.
(get_vsel_insn): Delete.
(rs6000_emit_vector_select): Ditto.
(rs6000_emit_vector_cond_expr): Ditto.
(rs6000_emit_int_cmove): Add support for -misel on 64-bit systems.
(rs6000_emit_minmax): On Altivec/VSX, directly emit the min/max
operation.
(create_TOC_reference): Add -mdebug=addr support.
(emit_frame_save): Add VSX support.
(output_toc): Add an assert to make sure we have a CONST_INT.
(insn_must_be_first_in_group): Add power7 support.
(insn_must_be_last_in_group): Ditto.
(rs6000_handle_altivec_attribute): Add VSX support.
(rs6000_mangle_type): Ditto.
(rs6000_register_move_cost): Add -mdebug=cost support. Add VSX
support.
(rs6000_memory_move_cost): Ditto.
(rs6000_emit_popcount): Add support for -mpopcntd.
(rs6000_vector_mode_supported_p): Add VSX support.
(rs6000_hash_indirect_jump_p): Return true if an indrect jump was
used in the program to turn off using the CTR register for count
down loops.
(rs6000_set_indirect_jump): Note that we have an indirect jump in
the program.
Index: gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/config/rs6000/rs6000.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/config/rs6000/rs6000.c) (revision 148152)
+++ gcc/config/rs6000/rs6000.c (.../gcc/config/rs6000/rs6000.c) (working copy)
@@ -7180,7 +8197,8 @@ def_builtin (int mask, const char *name,
if ((mask & target_flags) || TARGET_PAIRED_FLOAT)
{
if (rs6000_builtin_decls[code])
- abort ();
+ fatal_error ("internal error: builtin function to %s already processed.",
+ name);
rs6000_builtin_decls[code] =
add_builtin_function (name, type, code, BUILT_IN_MD,
@@ -7203,14 +8221,26 @@ static const struct builtin_description
{ MASK_ALTIVEC, CODE_FOR_altivec_vmsumuhs, "__builtin_altivec_vmsumuhs", ALTIVEC_BUILTIN_VMSUMUHS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmsumshs, "__builtin_altivec_vmsumshs", ALTIVEC_BUILTIN_VMSUMSHS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vnmsubfp, "__builtin_altivec_vnmsubfp", ALTIVEC_BUILTIN_VNMSUBFP },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v2df, "__builtin_altivec_vperm_2df", ALTIVEC_BUILTIN_VPERM_2DF },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v2di, "__builtin_altivec_vperm_2di", ALTIVEC_BUILTIN_VPERM_2DI },
{ MASK_ALTIVEC, CODE_FOR_altivec_vperm_v4sf, "__builtin_altivec_vperm_4sf", ALTIVEC_BUILTIN_VPERM_4SF },
{ MASK_ALTIVEC, CODE_FOR_altivec_vperm_v4si, "__builtin_altivec_vperm_4si", ALTIVEC_BUILTIN_VPERM_4SI },
{ MASK_ALTIVEC, CODE_FOR_altivec_vperm_v8hi, "__builtin_altivec_vperm_8hi", ALTIVEC_BUILTIN_VPERM_8HI },
- { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v16qi, "__builtin_altivec_vperm_16qi", ALTIVEC_BUILTIN_VPERM_16QI },
- { MASK_ALTIVEC, CODE_FOR_altivec_vsel_v4sf, "__builtin_altivec_vsel_4sf", ALTIVEC_BUILTIN_VSEL_4SF },
- { MASK_ALTIVEC, CODE_FOR_altivec_vsel_v4si, "__builtin_altivec_vsel_4si", ALTIVEC_BUILTIN_VSEL_4SI },
- { MASK_ALTIVEC, CODE_FOR_altivec_vsel_v8hi, "__builtin_altivec_vsel_8hi", ALTIVEC_BUILTIN_VSEL_8HI },
- { MASK_ALTIVEC, CODE_FOR_altivec_vsel_v16qi, "__builtin_altivec_vsel_16qi", ALTIVEC_BUILTIN_VSEL_16QI },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v16qi_uns, "__builtin_altivec_vperm_16qi", ALTIVEC_BUILTIN_VPERM_16QI },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v2di_uns, "__builtin_altivec_vperm_2di_uns", ALTIVEC_BUILTIN_VPERM_2DI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v4si_uns, "__builtin_altivec_vperm_4si_uns", ALTIVEC_BUILTIN_VPERM_4SI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v8hi_uns, "__builtin_altivec_vperm_8hi_uns", ALTIVEC_BUILTIN_VPERM_8HI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vperm_v16qi_uns, "__builtin_altivec_vperm_16qi_uns", ALTIVEC_BUILTIN_VPERM_16QI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v4sf, "__builtin_altivec_vsel_4sf", ALTIVEC_BUILTIN_VSEL_4SF },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v4si, "__builtin_altivec_vsel_4si", ALTIVEC_BUILTIN_VSEL_4SI },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v8hi, "__builtin_altivec_vsel_8hi", ALTIVEC_BUILTIN_VSEL_8HI },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v16qi, "__builtin_altivec_vsel_16qi", ALTIVEC_BUILTIN_VSEL_16QI },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v2df, "__builtin_altivec_vsel_2df", ALTIVEC_BUILTIN_VSEL_2DF },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v2di, "__builtin_altivec_vsel_2di", ALTIVEC_BUILTIN_VSEL_2DI },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v4si_uns, "__builtin_altivec_vsel_4si_uns", ALTIVEC_BUILTIN_VSEL_4SI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v8hi_uns, "__builtin_altivec_vsel_8hi_uns", ALTIVEC_BUILTIN_VSEL_8HI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v16qi_uns, "__builtin_altivec_vsel_16qi_uns", ALTIVEC_BUILTIN_VSEL_16QI_UNS },
+ { MASK_ALTIVEC, CODE_FOR_vector_select_v2di_uns, "__builtin_altivec_vsel_2di_uns", ALTIVEC_BUILTIN_VSEL_2DI_UNS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vsldoi_v16qi, "__builtin_altivec_vsldoi_16qi", ALTIVEC_BUILTIN_VSLDOI_16QI },
{ MASK_ALTIVEC, CODE_FOR_altivec_vsldoi_v8hi, "__builtin_altivec_vsldoi_8hi", ALTIVEC_BUILTIN_VSLDOI_8HI },
{ MASK_ALTIVEC, CODE_FOR_altivec_vsldoi_v4si, "__builtin_altivec_vsldoi_4si", ALTIVEC_BUILTIN_VSLDOI_4SI },
@@ -7232,6 +8262,56 @@ static const struct builtin_description
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_perm", ALTIVEC_BUILTIN_VEC_PERM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sel", ALTIVEC_BUILTIN_VEC_SEL },
+ { MASK_VSX, CODE_FOR_vsx_fmaddv2df4, "__builtin_vsx_xvmadddp", VSX_BUILTIN_XVMADDDP },
+ { MASK_VSX, CODE_FOR_vsx_fmsubv2df4, "__builtin_vsx_xvmsubdp", VSX_BUILTIN_XVMSUBDP },
+ { MASK_VSX, CODE_FOR_vsx_fnmaddv2df4, "__builtin_vsx_xvnmadddp", VSX_BUILTIN_XVNMADDDP },
+ { MASK_VSX, CODE_FOR_vsx_fnmsubv2df4, "__builtin_vsx_xvnmsubdp", VSX_BUILTIN_XVNMSUBDP },
+
+ { MASK_VSX, CODE_FOR_vsx_fmaddv4sf4, "__builtin_vsx_xvmaddsp", VSX_BUILTIN_XVMADDSP },
+ { MASK_VSX, CODE_FOR_vsx_fmsubv4sf4, "__builtin_vsx_xvmsubsp", VSX_BUILTIN_XVMSUBSP },
+ { MASK_VSX, CODE_FOR_vsx_fnmaddv4sf4, "__builtin_vsx_xvnmaddsp", VSX_BUILTIN_XVNMADDSP },
+ { MASK_VSX, CODE_FOR_vsx_fnmsubv4sf4, "__builtin_vsx_xvnmsubsp", VSX_BUILTIN_XVNMSUBSP },
+
+ { MASK_VSX, CODE_FOR_vector_select_v2di, "__builtin_vsx_xxsel_2di", VSX_BUILTIN_XXSEL_2DI },
+ { MASK_VSX, CODE_FOR_vector_select_v2df, "__builtin_vsx_xxsel_2df", VSX_BUILTIN_XXSEL_2DF },
+ { MASK_VSX, CODE_FOR_vector_select_v4sf, "__builtin_vsx_xxsel_4sf", VSX_BUILTIN_XXSEL_4SF },
+ { MASK_VSX, CODE_FOR_vector_select_v4si, "__builtin_vsx_xxsel_4si", VSX_BUILTIN_XXSEL_4SI },
+ { MASK_VSX, CODE_FOR_vector_select_v8hi, "__builtin_vsx_xxsel_8hi", VSX_BUILTIN_XXSEL_8HI },
+ { MASK_VSX, CODE_FOR_vector_select_v16qi, "__builtin_vsx_xxsel_16qi", VSX_BUILTIN_XXSEL_16QI },
+ { MASK_VSX, CODE_FOR_vector_select_v2di_uns, "__builtin_vsx_xxsel_2di_uns", VSX_BUILTIN_XXSEL_2DI_UNS },
+ { MASK_VSX, CODE_FOR_vector_select_v4si_uns, "__builtin_vsx_xxsel_4si_uns", VSX_BUILTIN_XXSEL_4SI_UNS },
+ { MASK_VSX, CODE_FOR_vector_select_v8hi_uns, "__builtin_vsx_xxsel_8hi_uns", VSX_BUILTIN_XXSEL_8HI_UNS },
+ { MASK_VSX, CODE_FOR_vector_select_v16qi_uns, "__builtin_vsx_xxsel_16qi_uns", VSX_BUILTIN_XXSEL_16QI_UNS },
+
+ { MASK_VSX, CODE_FOR_altivec_vperm_v2di, "__builtin_vsx_vperm_2di", VSX_BUILTIN_VPERM_2DI },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v2df, "__builtin_vsx_vperm_2df", VSX_BUILTIN_VPERM_2DF },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v4sf, "__builtin_vsx_vperm_4sf", VSX_BUILTIN_VPERM_4SF },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v4si, "__builtin_vsx_vperm_4si", VSX_BUILTIN_VPERM_4SI },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v8hi, "__builtin_vsx_vperm_8hi", VSX_BUILTIN_VPERM_8HI },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v16qi, "__builtin_vsx_vperm_16qi", VSX_BUILTIN_VPERM_16QI },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v2di_uns, "__builtin_vsx_vperm_2di_uns", VSX_BUILTIN_VPERM_2DI_UNS },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v4si_uns, "__builtin_vsx_vperm_4si_uns", VSX_BUILTIN_VPERM_4SI_UNS },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v8hi_uns, "__builtin_vsx_vperm_8hi_uns", VSX_BUILTIN_VPERM_8HI_UNS },
+ { MASK_VSX, CODE_FOR_altivec_vperm_v16qi_uns, "__builtin_vsx_vperm_16qi_uns", VSX_BUILTIN_VPERM_16QI_UNS },
+
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v2df, "__builtin_vsx_xxpermdi_2df", VSX_BUILTIN_XXPERMDI_2DF },
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v2di, "__builtin_vsx_xxpermdi_2di", VSX_BUILTIN_XXPERMDI_2DI },
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v4sf, "__builtin_vsx_xxpermdi_4sf", VSX_BUILTIN_XXPERMDI_4SF },
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v4si, "__builtin_vsx_xxpermdi_4si", VSX_BUILTIN_XXPERMDI_4SI },
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v8hi, "__builtin_vsx_xxpermdi_8hi", VSX_BUILTIN_XXPERMDI_8HI },
+ { MASK_VSX, CODE_FOR_vsx_xxpermdi_v16qi, "__builtin_vsx_xxpermdi_16qi", VSX_BUILTIN_XXPERMDI_16QI },
+ { MASK_VSX, CODE_FOR_nothing, "__builtin_vsx_xxpermdi", VSX_BUILTIN_VEC_XXPERMDI },
+ { MASK_VSX, CODE_FOR_vsx_set_v2df, "__builtin_vsx_set_2df", VSX_BUILTIN_SET_2DF },
+ { MASK_VSX, CODE_FOR_vsx_set_v2di, "__builtin_vsx_set_2di", VSX_BUILTIN_SET_2DI },
+
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v2di, "__builtin_vsx_xxsldwi_2di", VSX_BUILTIN_XXSLDWI_2DI },
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v2df, "__builtin_vsx_xxsldwi_2df", VSX_BUILTIN_XXSLDWI_2DF },
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v4sf, "__builtin_vsx_xxsldwi_4sf", VSX_BUILTIN_XXSLDWI_4SF },
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v4si, "__builtin_vsx_xxsldwi_4si", VSX_BUILTIN_XXSLDWI_4SI },
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v8hi, "__builtin_vsx_xxsldwi_8hi", VSX_BUILTIN_XXSLDWI_8HI },
+ { MASK_VSX, CODE_FOR_vsx_xxsldwi_v16qi, "__builtin_vsx_xxsldwi_16qi", VSX_BUILTIN_XXSLDWI_16QI },
+ { MASK_VSX, CODE_FOR_nothing, "__builtin_vsx_xxsldwi", VSX_BUILTIN_VEC_XXSLDWI },
+
{ 0, CODE_FOR_paired_msub, "__builtin_paired_msub", PAIRED_BUILTIN_MSUB },
{ 0, CODE_FOR_paired_madd, "__builtin_paired_madd", PAIRED_BUILTIN_MADD },
{ 0, CODE_FOR_paired_madds0, "__builtin_paired_madds0", PAIRED_BUILTIN_MADDS0 },
@@ -7284,18 +8364,18 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_altivec_vcfux, "__builtin_altivec_vcfux", ALTIVEC_BUILTIN_VCFUX },
{ MASK_ALTIVEC, CODE_FOR_altivec_vcfsx, "__builtin_altivec_vcfsx", ALTIVEC_BUILTIN_VCFSX },
{ MASK_ALTIVEC, CODE_FOR_altivec_vcmpbfp, "__builtin_altivec_vcmpbfp", ALTIVEC_BUILTIN_VCMPBFP },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpequb, "__builtin_altivec_vcmpequb", ALTIVEC_BUILTIN_VCMPEQUB },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpequh, "__builtin_altivec_vcmpequh", ALTIVEC_BUILTIN_VCMPEQUH },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpequw, "__builtin_altivec_vcmpequw", ALTIVEC_BUILTIN_VCMPEQUW },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpeqfp, "__builtin_altivec_vcmpeqfp", ALTIVEC_BUILTIN_VCMPEQFP },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgefp, "__builtin_altivec_vcmpgefp", ALTIVEC_BUILTIN_VCMPGEFP },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtub, "__builtin_altivec_vcmpgtub", ALTIVEC_BUILTIN_VCMPGTUB },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtsb, "__builtin_altivec_vcmpgtsb", ALTIVEC_BUILTIN_VCMPGTSB },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtuh, "__builtin_altivec_vcmpgtuh", ALTIVEC_BUILTIN_VCMPGTUH },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtsh, "__builtin_altivec_vcmpgtsh", ALTIVEC_BUILTIN_VCMPGTSH },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtuw, "__builtin_altivec_vcmpgtuw", ALTIVEC_BUILTIN_VCMPGTUW },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtsw, "__builtin_altivec_vcmpgtsw", ALTIVEC_BUILTIN_VCMPGTSW },
- { MASK_ALTIVEC, CODE_FOR_altivec_vcmpgtfp, "__builtin_altivec_vcmpgtfp", ALTIVEC_BUILTIN_VCMPGTFP },
+ { MASK_ALTIVEC, CODE_FOR_vector_eqv16qi, "__builtin_altivec_vcmpequb", ALTIVEC_BUILTIN_VCMPEQUB },
+ { MASK_ALTIVEC, CODE_FOR_vector_eqv8hi, "__builtin_altivec_vcmpequh", ALTIVEC_BUILTIN_VCMPEQUH },
+ { MASK_ALTIVEC, CODE_FOR_vector_eqv4si, "__builtin_altivec_vcmpequw", ALTIVEC_BUILTIN_VCMPEQUW },
+ { MASK_ALTIVEC, CODE_FOR_vector_eqv4sf, "__builtin_altivec_vcmpeqfp", ALTIVEC_BUILTIN_VCMPEQFP },
+ { MASK_ALTIVEC, CODE_FOR_vector_gev4sf, "__builtin_altivec_vcmpgefp", ALTIVEC_BUILTIN_VCMPGEFP },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtuv16qi, "__builtin_altivec_vcmpgtub", ALTIVEC_BUILTIN_VCMPGTUB },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtuv8hi, "__builtin_altivec_vcmpgtsb", ALTIVEC_BUILTIN_VCMPGTSB },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtuv4si, "__builtin_altivec_vcmpgtuh", ALTIVEC_BUILTIN_VCMPGTUH },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtv16qi, "__builtin_altivec_vcmpgtsh", ALTIVEC_BUILTIN_VCMPGTSH },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtv8hi, "__builtin_altivec_vcmpgtuw", ALTIVEC_BUILTIN_VCMPGTUW },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtv4si, "__builtin_altivec_vcmpgtsw", ALTIVEC_BUILTIN_VCMPGTSW },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtv4sf, "__builtin_altivec_vcmpgtfp", ALTIVEC_BUILTIN_VCMPGTFP },
{ MASK_ALTIVEC, CODE_FOR_altivec_vctsxs, "__builtin_altivec_vctsxs", ALTIVEC_BUILTIN_VCTSXS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vctuxs, "__builtin_altivec_vctuxs", ALTIVEC_BUILTIN_VCTUXS },
{ MASK_ALTIVEC, CODE_FOR_umaxv16qi3, "__builtin_altivec_vmaxub", ALTIVEC_BUILTIN_VMAXUB },
@@ -7319,14 +8399,18 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_sminv4si3, "__builtin_altivec_vminsw", ALTIVEC_BUILTIN_VMINSW },
{ MASK_ALTIVEC, CODE_FOR_sminv4sf3, "__builtin_altivec_vminfp", ALTIVEC_BUILTIN_VMINFP },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmuleub, "__builtin_altivec_vmuleub", ALTIVEC_BUILTIN_VMULEUB },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vmuleub, "__builtin_altivec_vmuleub_uns", ALTIVEC_BUILTIN_VMULEUB_UNS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmulesb, "__builtin_altivec_vmulesb", ALTIVEC_BUILTIN_VMULESB },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmuleuh, "__builtin_altivec_vmuleuh", ALTIVEC_BUILTIN_VMULEUH },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vmuleuh, "__builtin_altivec_vmuleuh_uns", ALTIVEC_BUILTIN_VMULEUH_UNS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmulesh, "__builtin_altivec_vmulesh", ALTIVEC_BUILTIN_VMULESH },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmuloub, "__builtin_altivec_vmuloub", ALTIVEC_BUILTIN_VMULOUB },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vmuloub, "__builtin_altivec_vmuloub_uns", ALTIVEC_BUILTIN_VMULOUB_UNS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmulosb, "__builtin_altivec_vmulosb", ALTIVEC_BUILTIN_VMULOSB },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmulouh, "__builtin_altivec_vmulouh", ALTIVEC_BUILTIN_VMULOUH },
+ { MASK_ALTIVEC, CODE_FOR_altivec_vmulouh, "__builtin_altivec_vmulouh_uns", ALTIVEC_BUILTIN_VMULOUH_UNS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vmulosh, "__builtin_altivec_vmulosh", ALTIVEC_BUILTIN_VMULOSH },
- { MASK_ALTIVEC, CODE_FOR_altivec_norv4si3, "__builtin_altivec_vnor", ALTIVEC_BUILTIN_VNOR },
+ { MASK_ALTIVEC, CODE_FOR_norv4si3, "__builtin_altivec_vnor", ALTIVEC_BUILTIN_VNOR },
{ MASK_ALTIVEC, CODE_FOR_iorv4si3, "__builtin_altivec_vor", ALTIVEC_BUILTIN_VOR },
{ MASK_ALTIVEC, CODE_FOR_altivec_vpkuhum, "__builtin_altivec_vpkuhum", ALTIVEC_BUILTIN_VPKUHUM },
{ MASK_ALTIVEC, CODE_FOR_altivec_vpkuwum, "__builtin_altivec_vpkuwum", ALTIVEC_BUILTIN_VPKUWUM },
@@ -7337,9 +8421,9 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_altivec_vpkshus, "__builtin_altivec_vpkshus", ALTIVEC_BUILTIN_VPKSHUS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vpkuwus, "__builtin_altivec_vpkuwus", ALTIVEC_BUILTIN_VPKUWUS },
{ MASK_ALTIVEC, CODE_FOR_altivec_vpkswus, "__builtin_altivec_vpkswus", ALTIVEC_BUILTIN_VPKSWUS },
- { MASK_ALTIVEC, CODE_FOR_altivec_vrlb, "__builtin_altivec_vrlb", ALTIVEC_BUILTIN_VRLB },
- { MASK_ALTIVEC, CODE_FOR_altivec_vrlh, "__builtin_altivec_vrlh", ALTIVEC_BUILTIN_VRLH },
- { MASK_ALTIVEC, CODE_FOR_altivec_vrlw, "__builtin_altivec_vrlw", ALTIVEC_BUILTIN_VRLW },
+ { MASK_ALTIVEC, CODE_FOR_vrotlv16qi3, "__builtin_altivec_vrlb", ALTIVEC_BUILTIN_VRLB },
+ { MASK_ALTIVEC, CODE_FOR_vrotlv8hi3, "__builtin_altivec_vrlh", ALTIVEC_BUILTIN_VRLH },
+ { MASK_ALTIVEC, CODE_FOR_vrotlv4si3, "__builtin_altivec_vrlw", ALTIVEC_BUILTIN_VRLW },
{ MASK_ALTIVEC, CODE_FOR_vashlv16qi3, "__builtin_altivec_vslb", ALTIVEC_BUILTIN_VSLB },
{ MASK_ALTIVEC, CODE_FOR_vashlv8hi3, "__builtin_altivec_vslh", ALTIVEC_BUILTIN_VSLH },
{ MASK_ALTIVEC, CODE_FOR_vashlv4si3, "__builtin_altivec_vslw", ALTIVEC_BUILTIN_VSLW },
@@ -7374,8 +8458,42 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_altivec_vsumsws, "__builtin_altivec_vsumsws", ALTIVEC_BUILTIN_VSUMSWS },
{ MASK_ALTIVEC, CODE_FOR_xorv4si3, "__builtin_altivec_vxor", ALTIVEC_BUILTIN_VXOR },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_add", ALTIVEC_BUILTIN_VEC_ADD },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vaddfp", ALTIVEC_BUILTIN_VEC_VADDFP },
+ { MASK_VSX, CODE_FOR_addv2df3, "__builtin_vsx_xvadddp", VSX_BUILTIN_XVADDDP },
+ { MASK_VSX, CODE_FOR_subv2df3, "__builtin_vsx_xvsubdp", VSX_BUILTIN_XVSUBDP },
+ { MASK_VSX, CODE_FOR_mulv2df3, "__builtin_vsx_xvmuldp", VSX_BUILTIN_XVMULDP },
+ { MASK_VSX, CODE_FOR_divv2df3, "__builtin_vsx_xvdivdp", VSX_BUILTIN_XVDIVDP },
+ { MASK_VSX, CODE_FOR_sminv2df3, "__builtin_vsx_xvmindp", VSX_BUILTIN_XVMINDP },
+ { MASK_VSX, CODE_FOR_smaxv2df3, "__builtin_vsx_xvmaxdp", VSX_BUILTIN_XVMAXDP },
+ { MASK_VSX, CODE_FOR_vsx_tdivv2df3, "__builtin_vsx_xvtdivdp", VSX_BUILTIN_XVTDIVDP },
+ { MASK_VSX, CODE_FOR_vector_eqv2df, "__builtin_vsx_xvcmpeqdp", VSX_BUILTIN_XVCMPEQDP },
+ { MASK_VSX, CODE_FOR_vector_gtv2df, "__builtin_vsx_xvcmpgtdp", VSX_BUILTIN_XVCMPGTDP },
+ { MASK_VSX, CODE_FOR_vector_gev2df, "__builtin_vsx_xvcmpgedp", VSX_BUILTIN_XVCMPGEDP },
+
+ { MASK_VSX, CODE_FOR_addv4sf3, "__builtin_vsx_xvaddsp", VSX_BUILTIN_XVADDSP },
+ { MASK_VSX, CODE_FOR_subv4sf3, "__builtin_vsx_xvsubsp", VSX_BUILTIN_XVSUBSP },
+ { MASK_VSX, CODE_FOR_mulv4sf3, "__builtin_vsx_xvmulsp", VSX_BUILTIN_XVMULSP },
+ { MASK_VSX, CODE_FOR_divv4sf3, "__builtin_vsx_xvdivsp", VSX_BUILTIN_XVDIVSP },
+ { MASK_VSX, CODE_FOR_sminv4sf3, "__builtin_vsx_xvminsp", VSX_BUILTIN_XVMINSP },
+ { MASK_VSX, CODE_FOR_smaxv4sf3, "__builtin_vsx_xvmaxsp", VSX_BUILTIN_XVMAXSP },
+ { MASK_VSX, CODE_FOR_vsx_tdivv4sf3, "__builtin_vsx_xvtdivsp", VSX_BUILTIN_XVTDIVSP },
+ { MASK_VSX, CODE_FOR_vector_eqv4sf, "__builtin_vsx_xvcmpeqsp", VSX_BUILTIN_XVCMPEQSP },
+ { MASK_VSX, CODE_FOR_vector_gtv4sf, "__builtin_vsx_xvcmpgtsp", VSX_BUILTIN_XVCMPGTSP },
+ { MASK_VSX, CODE_FOR_vector_gev4sf, "__builtin_vsx_xvcmpgesp", VSX_BUILTIN_XVCMPGESP },
+
+ { MASK_VSX, CODE_FOR_smindf3, "__builtin_vsx_xsmindp", VSX_BUILTIN_XSMINDP },
+ { MASK_VSX, CODE_FOR_smaxdf3, "__builtin_vsx_xsmaxdp", VSX_BUILTIN_XSMAXDP },
+
+ { MASK_VSX, CODE_FOR_vsx_concat_v2df, "__builtin_vsx_concat_2df", VSX_BUILTIN_CONCAT_2DF },
+ { MASK_VSX, CODE_FOR_vsx_concat_v2di, "__builtin_vsx_concat_2di", VSX_BUILTIN_CONCAT_2DI },
+ { MASK_VSX, CODE_FOR_vsx_splat_v2df, "__builtin_vsx_splat_2df", VSX_BUILTIN_SPLAT_2DF },
+ { MASK_VSX, CODE_FOR_vsx_splat_v2di, "__builtin_vsx_splat_2di", VSX_BUILTIN_SPLAT_2DI },
+ { MASK_VSX, CODE_FOR_vsx_xxmrghw_v4sf, "__builtin_vsx_xxmrghw", VSX_BUILTIN_XXMRGHW_4SF },
+ { MASK_VSX, CODE_FOR_vsx_xxmrghw_v4si, "__builtin_vsx_xxmrghw_4si", VSX_BUILTIN_XXMRGHW_4SI },
+ { MASK_VSX, CODE_FOR_vsx_xxmrglw_v4sf, "__builtin_vsx_xxmrglw", VSX_BUILTIN_XXMRGLW_4SF },
+ { MASK_VSX, CODE_FOR_vsx_xxmrglw_v4si, "__builtin_vsx_xxmrglw_4si", VSX_BUILTIN_XXMRGLW_4SI },
+
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_add", ALTIVEC_BUILTIN_VEC_ADD },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vaddfp", ALTIVEC_BUILTIN_VEC_VADDFP },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vadduwm", ALTIVEC_BUILTIN_VEC_VADDUWM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vadduhm", ALTIVEC_BUILTIN_VEC_VADDUHM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vaddubm", ALTIVEC_BUILTIN_VEC_VADDUBM },
@@ -7387,8 +8505,8 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vadduhs", ALTIVEC_BUILTIN_VEC_VADDUHS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vaddsbs", ALTIVEC_BUILTIN_VEC_VADDSBS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vaddubs", ALTIVEC_BUILTIN_VEC_VADDUBS },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_and", ALTIVEC_BUILTIN_VEC_AND },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_andc", ALTIVEC_BUILTIN_VEC_ANDC },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_and", ALTIVEC_BUILTIN_VEC_AND },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_andc", ALTIVEC_BUILTIN_VEC_ANDC },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_avg", ALTIVEC_BUILTIN_VEC_AVG },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vavgsw", ALTIVEC_BUILTIN_VEC_VAVGSW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vavguw", ALTIVEC_BUILTIN_VEC_VAVGUW },
@@ -7413,8 +8531,8 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vcmpgtub", ALTIVEC_BUILTIN_VEC_VCMPGTUB },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_cmple", ALTIVEC_BUILTIN_VEC_CMPLE },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_cmplt", ALTIVEC_BUILTIN_VEC_CMPLT },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_max", ALTIVEC_BUILTIN_VEC_MAX },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmaxfp", ALTIVEC_BUILTIN_VEC_VMAXFP },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_max", ALTIVEC_BUILTIN_VEC_MAX },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vmaxfp", ALTIVEC_BUILTIN_VEC_VMAXFP },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmaxsw", ALTIVEC_BUILTIN_VEC_VMAXSW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmaxuw", ALTIVEC_BUILTIN_VEC_VMAXUW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmaxsh", ALTIVEC_BUILTIN_VEC_VMAXSH },
@@ -7429,8 +8547,8 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmrglw", ALTIVEC_BUILTIN_VEC_VMRGLW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmrglh", ALTIVEC_BUILTIN_VEC_VMRGLH },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmrglb", ALTIVEC_BUILTIN_VEC_VMRGLB },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_min", ALTIVEC_BUILTIN_VEC_MIN },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vminfp", ALTIVEC_BUILTIN_VEC_VMINFP },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_min", ALTIVEC_BUILTIN_VEC_MIN },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vminfp", ALTIVEC_BUILTIN_VEC_VMINFP },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vminsw", ALTIVEC_BUILTIN_VEC_VMINSW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vminuw", ALTIVEC_BUILTIN_VEC_VMINUW },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vminsh", ALTIVEC_BUILTIN_VEC_VMINSH },
@@ -7447,8 +8565,8 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmulouh", ALTIVEC_BUILTIN_VEC_VMULOUH },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmulosb", ALTIVEC_BUILTIN_VEC_VMULOSB },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vmuloub", ALTIVEC_BUILTIN_VEC_VMULOUB },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_nor", ALTIVEC_BUILTIN_VEC_NOR },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_or", ALTIVEC_BUILTIN_VEC_OR },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_nor", ALTIVEC_BUILTIN_VEC_NOR },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_or", ALTIVEC_BUILTIN_VEC_OR },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_pack", ALTIVEC_BUILTIN_VEC_PACK },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vpkuwum", ALTIVEC_BUILTIN_VEC_VPKUWUM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vpkuhum", ALTIVEC_BUILTIN_VEC_VPKUHUM },
@@ -7481,8 +8599,8 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsrab", ALTIVEC_BUILTIN_VEC_VSRAB },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_srl", ALTIVEC_BUILTIN_VEC_SRL },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sro", ALTIVEC_BUILTIN_VEC_SRO },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sub", ALTIVEC_BUILTIN_VEC_SUB },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsubfp", ALTIVEC_BUILTIN_VEC_VSUBFP },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_sub", ALTIVEC_BUILTIN_VEC_SUB },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vsubfp", ALTIVEC_BUILTIN_VEC_VSUBFP },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsubuwm", ALTIVEC_BUILTIN_VEC_VSUBUWM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsubuhm", ALTIVEC_BUILTIN_VEC_VSUBUHM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsububm", ALTIVEC_BUILTIN_VEC_VSUBUBM },
@@ -7500,7 +8618,10 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vsum4ubs", ALTIVEC_BUILTIN_VEC_VSUM4UBS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sum2s", ALTIVEC_BUILTIN_VEC_SUM2S },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sums", ALTIVEC_BUILTIN_VEC_SUMS },
- { MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_xor", ALTIVEC_BUILTIN_VEC_XOR },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_xor", ALTIVEC_BUILTIN_VEC_XOR },
+
+ { MASK_VSX, CODE_FOR_nothing, "__builtin_vec_mul", VSX_BUILTIN_VEC_MUL },
+ { MASK_VSX, CODE_FOR_nothing, "__builtin_vec_div", VSX_BUILTIN_VEC_DIV },
{ 0, CODE_FOR_divv2sf3, "__builtin_paired_divv2sf3", PAIRED_BUILTIN_DIVV2SF3 },
{ 0, CODE_FOR_addv2sf3, "__builtin_paired_addv2sf3", PAIRED_BUILTIN_ADDV2SF3 },
@@ -7664,30 +8785,58 @@ struct builtin_description_predicates
{
const unsigned int mask;
const enum insn_code icode;
- const char *opcode;
const char *const name;
const enum rs6000_builtins code;
};
static const struct builtin_description_predicates bdesc_altivec_preds[] =
{
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4sf, "*vcmpbfp.", "__builtin_altivec_vcmpbfp_p", ALTIVEC_BUILTIN_VCMPBFP_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4sf, "*vcmpeqfp.", "__builtin_altivec_vcmpeqfp_p", ALTIVEC_BUILTIN_VCMPEQFP_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4sf, "*vcmpgefp.", "__builtin_altivec_vcmpgefp_p", ALTIVEC_BUILTIN_VCMPGEFP_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4sf, "*vcmpgtfp.", "__builtin_altivec_vcmpgtfp_p", ALTIVEC_BUILTIN_VCMPGTFP_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4si, "*vcmpequw.", "__builtin_altivec_vcmpequw_p", ALTIVEC_BUILTIN_VCMPEQUW_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4si, "*vcmpgtsw.", "__builtin_altivec_vcmpgtsw_p", ALTIVEC_BUILTIN_VCMPGTSW_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v4si, "*vcmpgtuw.", "__builtin_altivec_vcmpgtuw_p", ALTIVEC_BUILTIN_VCMPGTUW_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v8hi, "*vcmpgtuh.", "__builtin_altivec_vcmpgtuh_p", ALTIVEC_BUILTIN_VCMPGTUH_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v8hi, "*vcmpgtsh.", "__builtin_altivec_vcmpgtsh_p", ALTIVEC_BUILTIN_VCMPGTSH_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v8hi, "*vcmpequh.", "__builtin_altivec_vcmpequh_p", ALTIVEC_BUILTIN_VCMPEQUH_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v16qi, "*vcmpequb.", "__builtin_altivec_vcmpequb_p", ALTIVEC_BUILTIN_VCMPEQUB_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v16qi, "*vcmpgtsb.", "__builtin_altivec_vcmpgtsb_p", ALTIVEC_BUILTIN_VCMPGTSB_P },
- { MASK_ALTIVEC, CODE_FOR_altivec_predicate_v16qi, "*vcmpgtub.", "__builtin_altivec_vcmpgtub_p", ALTIVEC_BUILTIN_VCMPGTUB_P },
-
- { MASK_ALTIVEC, CODE_FOR_nothing, NULL, "__builtin_vec_vcmpeq_p", ALTIVEC_BUILTIN_VCMPEQ_P },
- { MASK_ALTIVEC, CODE_FOR_nothing, NULL, "__builtin_vec_vcmpgt_p", ALTIVEC_BUILTIN_VCMPGT_P },
- { MASK_ALTIVEC, CODE_FOR_nothing, NULL, "__builtin_vec_vcmpge_p", ALTIVEC_BUILTIN_VCMPGE_P }
+ { MASK_ALTIVEC, CODE_FOR_altivec_vcmpbfp_p, "__builtin_altivec_vcmpbfp_p",
+ ALTIVEC_BUILTIN_VCMPBFP_P },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_vector_eq_v4sf_p,
+ "__builtin_altivec_vcmpeqfp_p", ALTIVEC_BUILTIN_VCMPEQFP_P },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_vector_ge_v4sf_p,
+ "__builtin_altivec_vcmpgefp_p", ALTIVEC_BUILTIN_VCMPGEFP_P },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_vector_gt_v4sf_p,
+ "__builtin_altivec_vcmpgtfp_p", ALTIVEC_BUILTIN_VCMPGTFP_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_eq_v4si_p, "__builtin_altivec_vcmpequw_p",
+ ALTIVEC_BUILTIN_VCMPEQUW_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gt_v4si_p, "__builtin_altivec_vcmpgtsw_p",
+ ALTIVEC_BUILTIN_VCMPGTSW_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtu_v4si_p, "__builtin_altivec_vcmpgtuw_p",
+ ALTIVEC_BUILTIN_VCMPGTUW_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_eq_v8hi_p, "__builtin_altivec_vcmpequh_p",
+ ALTIVEC_BUILTIN_VCMPEQUH_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gt_v8hi_p, "__builtin_altivec_vcmpgtsh_p",
+ ALTIVEC_BUILTIN_VCMPGTSH_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtu_v8hi_p, "__builtin_altivec_vcmpgtuh_p",
+ ALTIVEC_BUILTIN_VCMPGTUH_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_eq_v16qi_p, "__builtin_altivec_vcmpequb_p",
+ ALTIVEC_BUILTIN_VCMPEQUB_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gt_v16qi_p, "__builtin_altivec_vcmpgtsb_p",
+ ALTIVEC_BUILTIN_VCMPGTSB_P },
+ { MASK_ALTIVEC, CODE_FOR_vector_gtu_v16qi_p, "__builtin_altivec_vcmpgtub_p",
+ ALTIVEC_BUILTIN_VCMPGTUB_P },
+
+ { MASK_VSX, CODE_FOR_vector_eq_v4sf_p, "__builtin_vsx_xvcmpeqsp_p",
+ VSX_BUILTIN_XVCMPEQSP_P },
+ { MASK_VSX, CODE_FOR_vector_ge_v4sf_p, "__builtin_vsx_xvcmpgesp_p",
+ VSX_BUILTIN_XVCMPGESP_P },
+ { MASK_VSX, CODE_FOR_vector_gt_v4sf_p, "__builtin_vsx_xvcmpgtsp_p",
+ VSX_BUILTIN_XVCMPGTSP_P },
+ { MASK_VSX, CODE_FOR_vector_eq_v2df_p, "__builtin_vsx_xvcmpeqdp_p",
+ VSX_BUILTIN_XVCMPEQDP_P },
+ { MASK_VSX, CODE_FOR_vector_ge_v2df_p, "__builtin_vsx_xvcmpgedp_p",
+ VSX_BUILTIN_XVCMPGEDP_P },
+ { MASK_VSX, CODE_FOR_vector_gt_v2df_p, "__builtin_vsx_xvcmpgtdp_p",
+ VSX_BUILTIN_XVCMPGTDP_P },
+
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vcmpeq_p",
+ ALTIVEC_BUILTIN_VCMPEQ_P },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vcmpgt_p",
+ ALTIVEC_BUILTIN_VCMPGT_P },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_nothing, "__builtin_vec_vcmpge_p",
+ ALTIVEC_BUILTIN_VCMPGE_P }
};
/* SPE predicates. */
@@ -7745,7 +8894,11 @@ static const struct builtin_description
{ MASK_ALTIVEC, CODE_FOR_absv16qi2, "__builtin_altivec_abs_v16qi", ALTIVEC_BUILTIN_ABS_V16QI },
{ MASK_ALTIVEC, CODE_FOR_altivec_abss_v4si, "__builtin_altivec_abss_v4si", ALTIVEC_BUILTIN_ABSS_V4SI },
{ MASK_ALTIVEC, CODE_FOR_altivec_abss_v8hi, "__builtin_altivec_abss_v8hi", ALTIVEC_BUILTIN_ABSS_V8HI },
- { MASK_ALTIVEC, CODE_FOR_altivec_abss_v16qi, "__builtin_altivec_abss_v16qi", ALTIVEC_BUILTIN_ABSS_V16QI }
+ { MASK_ALTIVEC, CODE_FOR_altivec_abss_v16qi, "__builtin_altivec_abss_v16qi", ALTIVEC_BUILTIN_ABSS_V16QI },
+ { MASK_VSX, CODE_FOR_absv2df2, "__builtin_vsx_xvabsdp", VSX_BUILTIN_XVABSDP },
+ { MASK_VSX, CODE_FOR_vsx_nabsv2df2, "__builtin_vsx_xvnabsdp", VSX_BUILTIN_XVNABSDP },
+ { MASK_VSX, CODE_FOR_absv4sf2, "__builtin_vsx_xvabssp", VSX_BUILTIN_XVABSSP },
+ { MASK_VSX, CODE_FOR_vsx_nabsv4sf2, "__builtin_vsx_xvnabssp", VSX_BUILTIN_XVNABSSP },
};
/* Simple unary operations: VECb = foo (unsigned literal) or VECb =
@@ -7771,6 +8924,61 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_altivec_vupklpx, "__builtin_altivec_vupklpx", ALTIVEC_BUILTIN_VUPKLPX },
{ MASK_ALTIVEC, CODE_FOR_altivec_vupklsh, "__builtin_altivec_vupklsh", ALTIVEC_BUILTIN_VUPKLSH },
+ { MASK_VSX, CODE_FOR_negv2df2, "__builtin_vsx_xvnegdp", VSX_BUILTIN_XVNEGDP },
+ { MASK_VSX, CODE_FOR_sqrtv2df2, "__builtin_vsx_xvsqrtdp", VSX_BUILTIN_XVSQRTDP },
+ { MASK_VSX, CODE_FOR_vsx_rsqrtev2df2, "__builtin_vsx_xvrsqrtedp", VSX_BUILTIN_XVRSQRTEDP },
+ { MASK_VSX, CODE_FOR_vsx_tsqrtv2df2, "__builtin_vsx_xvtsqrtdp", VSX_BUILTIN_XVTSQRTDP },
+ { MASK_VSX, CODE_FOR_vsx_frev2df2, "__builtin_vsx_xvredp", VSX_BUILTIN_XVREDP },
+
+ { MASK_VSX, CODE_FOR_negv4sf2, "__builtin_vsx_xvnegsp", VSX_BUILTIN_XVNEGSP },
+ { MASK_VSX, CODE_FOR_sqrtv4sf2, "__builtin_vsx_xvsqrtsp", VSX_BUILTIN_XVSQRTSP },
+ { MASK_VSX, CODE_FOR_vsx_rsqrtev4sf2, "__builtin_vsx_xvrsqrtesp", VSX_BUILTIN_XVRSQRTESP },
+ { MASK_VSX, CODE_FOR_vsx_tsqrtv4sf2, "__builtin_vsx_xvtsqrtsp", VSX_BUILTIN_XVTSQRTSP },
+ { MASK_VSX, CODE_FOR_vsx_frev4sf2, "__builtin_vsx_xvresp", VSX_BUILTIN_XVRESP },
+
+ { MASK_VSX, CODE_FOR_vsx_xscvdpsp, "__builtin_vsx_xscvdpsp", VSX_BUILTIN_XSCVDPSP },
+ { MASK_VSX, CODE_FOR_vsx_xscvdpsp, "__builtin_vsx_xscvspdp", VSX_BUILTIN_XSCVSPDP },
+ { MASK_VSX, CODE_FOR_vsx_xvcvdpsp, "__builtin_vsx_xvcvdpsp", VSX_BUILTIN_XVCVDPSP },
+ { MASK_VSX, CODE_FOR_vsx_xvcvspdp, "__builtin_vsx_xvcvspdp", VSX_BUILTIN_XVCVSPDP },
+
+ { MASK_VSX, CODE_FOR_vsx_fix_truncv2dfv2di2, "__builtin_vsx_xvcvdpsxds", VSX_BUILTIN_XVCVDPSXDS },
+ { MASK_VSX, CODE_FOR_vsx_fixuns_truncv2dfv2di2, "__builtin_vsx_xvcvdpuxds", VSX_BUILTIN_XVCVDPUXDS },
+ { MASK_VSX, CODE_FOR_vsx_fixuns_truncv2dfv2di2, "__builtin_vsx_xvcvdpuxds_uns", VSX_BUILTIN_XVCVDPUXDS_UNS },
+ { MASK_VSX, CODE_FOR_vsx_floatv2div2df2, "__builtin_vsx_xvcvsxddp", VSX_BUILTIN_XVCVSXDDP },
+ { MASK_VSX, CODE_FOR_vsx_floatunsv2div2df2, "__builtin_vsx_xvcvuxddp", VSX_BUILTIN_XVCVUXDDP },
+ { MASK_VSX, CODE_FOR_vsx_floatunsv2div2df2, "__builtin_vsx_xvcvuxddp_uns", VSX_BUILTIN_XVCVUXDDP_UNS },
+
+ { MASK_VSX, CODE_FOR_vsx_fix_truncv4sfv4si2, "__builtin_vsx_xvcvspsxws", VSX_BUILTIN_XVCVSPSXWS },
+ { MASK_VSX, CODE_FOR_vsx_fixuns_truncv4sfv4si2, "__builtin_vsx_xvcvspuxws", VSX_BUILTIN_XVCVSPUXWS },
+ { MASK_VSX, CODE_FOR_vsx_floatv4siv4sf2, "__builtin_vsx_xvcvsxwsp", VSX_BUILTIN_XVCVSXWSP },
+ { MASK_VSX, CODE_FOR_vsx_floatunsv4siv4sf2, "__builtin_vsx_xvcvuxwsp", VSX_BUILTIN_XVCVUXWSP },
+
+ { MASK_VSX, CODE_FOR_vsx_xvcvdpsxws, "__builtin_vsx_xvcvdpsxws", VSX_BUILTIN_XVCVDPSXWS },
+ { MASK_VSX, CODE_FOR_vsx_xvcvdpuxws, "__builtin_vsx_xvcvdpuxws", VSX_BUILTIN_XVCVDPUXWS },
+ { MASK_VSX, CODE_FOR_vsx_xvcvsxwdp, "__builtin_vsx_xvcvsxwdp", VSX_BUILTIN_XVCVSXWDP },
+ { MASK_VSX, CODE_FOR_vsx_xvcvuxwdp, "__builtin_vsx_xvcvuxwdp", VSX_BUILTIN_XVCVUXWDP },
+ { MASK_VSX, CODE_FOR_vsx_xvrdpi, "__builtin_vsx_xvrdpi", VSX_BUILTIN_XVRDPI },
+ { MASK_VSX, CODE_FOR_vsx_xvrdpic, "__builtin_vsx_xvrdpic", VSX_BUILTIN_XVRDPIC },
+ { MASK_VSX, CODE_FOR_vsx_floorv2df2, "__builtin_vsx_xvrdpim", VSX_BUILTIN_XVRDPIM },
+ { MASK_VSX, CODE_FOR_vsx_ceilv2df2, "__builtin_vsx_xvrdpip", VSX_BUILTIN_XVRDPIP },
+ { MASK_VSX, CODE_FOR_vsx_btruncv2df2, "__builtin_vsx_xvrdpiz", VSX_BUILTIN_XVRDPIZ },
+
+ { MASK_VSX, CODE_FOR_vsx_xvcvspsxds, "__builtin_vsx_xvcvspsxds", VSX_BUILTIN_XVCVSPSXDS },
+ { MASK_VSX, CODE_FOR_vsx_xvcvspuxds, "__builtin_vsx_xvcvspuxds", VSX_BUILTIN_XVCVSPUXDS },
+ { MASK_VSX, CODE_FOR_vsx_xvcvsxdsp, "__builtin_vsx_xvcvsxdsp", VSX_BUILTIN_XVCVSXDSP },
+ { MASK_VSX, CODE_FOR_vsx_xvcvuxdsp, "__builtin_vsx_xvcvuxdsp", VSX_BUILTIN_XVCVUXDSP },
+ { MASK_VSX, CODE_FOR_vsx_xvrspi, "__builtin_vsx_xvrspi", VSX_BUILTIN_XVRSPI },
+ { MASK_VSX, CODE_FOR_vsx_xvrspic, "__builtin_vsx_xvrspic", VSX_BUILTIN_XVRSPIC },
+ { MASK_VSX, CODE_FOR_vsx_floorv4sf2, "__builtin_vsx_xvrspim", VSX_BUILTIN_XVRSPIM },
+ { MASK_VSX, CODE_FOR_vsx_ceilv4sf2, "__builtin_vsx_xvrspip", VSX_BUILTIN_XVRSPIP },
+ { MASK_VSX, CODE_FOR_vsx_btruncv4sf2, "__builtin_vsx_xvrspiz", VSX_BUILTIN_XVRSPIZ },
+
+ { MASK_VSX, CODE_FOR_vsx_xsrdpi, "__builtin_vsx_xsrdpi", VSX_BUILTIN_XSRDPI },
+ { MASK_VSX, CODE_FOR_vsx_xsrdpic, "__builtin_vsx_xsrdpic", VSX_BUILTIN_XSRDPIC },
+ { MASK_VSX, CODE_FOR_vsx_floordf2, "__builtin_vsx_xsrdpim", VSX_BUILTIN_XSRDPIM },
+ { MASK_VSX, CODE_FOR_vsx_ceildf2, "__builtin_vsx_xsrdpip", VSX_BUILTIN_XSRDPIP },
+ { MASK_VSX, CODE_FOR_vsx_btruncdf2, "__builtin_vsx_xsrdpiz", VSX_BUILTIN_XSRDPIZ },
+
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_abs", ALTIVEC_BUILTIN_VEC_ABS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_abss", ALTIVEC_BUILTIN_VEC_ABSS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_ceil", ALTIVEC_BUILTIN_VEC_CEIL },
@@ -7791,6 +8999,11 @@ static struct builtin_description bdesc_
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vupklsh", ALTIVEC_BUILTIN_VEC_VUPKLSH },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_vupklsb", ALTIVEC_BUILTIN_VEC_VUPKLSB },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_floatv4siv4sf2, "__builtin_vec_float_sisf", VECTOR_BUILTIN_FLOAT_V4SI_V4SF },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_unsigned_floatv4siv4sf2, "__builtin_vec_uns_float_sisf", VECTOR_BUILTIN_UNSFLOAT_V4SI_V4SF },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_fix_truncv4sfv4si2, "__builtin_vec_fix_sfsi", VECTOR_BUILTIN_FIX_V4SF_V4SI },
+ { MASK_ALTIVEC|MASK_VSX, CODE_FOR_fixuns_truncv4sfv4si2, "__builtin_vec_fixuns_sfsi", VECTOR_BUILTIN_FIXUNS_V4SF_V4SI },
+
/* The SPE unary builtins must start with SPE_BUILTIN_EVABS and
end with SPE_BUILTIN_EVSUBFUSIAAW. */
{ 0, CODE_FOR_spe_evabs, "__builtin_spe_evabs", SPE_BUILTIN_EVABS },
@@ -7987,8 +9200,7 @@ rs6000_expand_binop_builtin (enum insn_c
}
static rtx
-altivec_expand_predicate_builtin (enum insn_code icode, const char *opcode,
- tree exp, rtx target)
+altivec_expand_predicate_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, scratch;
tree cr6_form = CALL_EXPR_ARG (exp, 0);
@@ -8027,8 +9239,7 @@ altivec_expand_predicate_builtin (enum i
scratch = gen_reg_rtx (mode0);
- pat = GEN_FCN (icode) (scratch, op0, op1,
- gen_rtx_SYMBOL_REF (Pmode, opcode));
+ pat = GEN_FCN (icode) (scratch, op0, op1);
if (! pat)
return 0;
emit_insn (pat);
@@ -8295,11 +9506,12 @@ rs6000_expand_ternop_builtin (enum insn_
|| arg2 == error_mark_node)
return const0_rtx;
- if (icode == CODE_FOR_altivec_vsldoi_v4sf
- || icode == CODE_FOR_altivec_vsldoi_v4si
- || icode == CODE_FOR_altivec_vsldoi_v8hi
- || icode == CODE_FOR_altivec_vsldoi_v16qi)
+ switch (icode)
{
+ case CODE_FOR_altivec_vsldoi_v4sf:
+ case CODE_FOR_altivec_vsldoi_v4si:
+ case CODE_FOR_altivec_vsldoi_v8hi:
+ case CODE_FOR_altivec_vsldoi_v16qi:
/* Only allow 4-bit unsigned literals. */
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
@@ -8308,6 +9520,40 @@ rs6000_expand_ternop_builtin (enum insn_
error ("argument 3 must be a 4-bit unsigned literal");
return const0_rtx;
}
+ break;
+
+ case CODE_FOR_vsx_xxpermdi_v2df:
+ case CODE_FOR_vsx_xxpermdi_v2di:
+ case CODE_FOR_vsx_xxsldwi_v16qi:
+ case CODE_FOR_vsx_xxsldwi_v8hi:
+ case CODE_FOR_vsx_xxsldwi_v4si:
+ case CODE_FOR_vsx_xxsldwi_v4sf:
+ case CODE_FOR_vsx_xxsldwi_v2di:
+ case CODE_FOR_vsx_xxsldwi_v2df:
+ /* Only allow 2-bit unsigned literals. */
+ STRIP_NOPS (arg2);
+ if (TREE_CODE (arg2) != INTEGER_CST
+ || TREE_INT_CST_LOW (arg2) & ~0x3)
+ {
+ error ("argument 3 must be a 2-bit unsigned literal");
+ return const0_rtx;
+ }
+ break;
+
+ case CODE_FOR_vsx_set_v2df:
+ case CODE_FOR_vsx_set_v2di:
+ /* Only allow 1-bit unsigned literals. */
+ STRIP_NOPS (arg2);
+ if (TREE_CODE (arg2) != INTEGER_CST
+ || TREE_INT_CST_LOW (arg2) & ~0x1)
+ {
+ error ("argument 3 must be a 1-bit unsigned literal");
+ return const0_rtx;
+ }
+ break;
+
+ default:
+ break;
}
if (target == 0
@@ -8347,16 +9593,16 @@ altivec_expand_ld_builtin (tree exp, rtx
switch (fcode)
{
case ALTIVEC_BUILTIN_LD_INTERNAL_16qi:
- icode = CODE_FOR_altivec_lvx_v16qi;
+ icode = CODE_FOR_vector_load_v16qi;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_8hi:
- icode = CODE_FOR_altivec_lvx_v8hi;
+ icode = CODE_FOR_vector_load_v8hi;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_4si:
- icode = CODE_FOR_altivec_lvx_v4si;
+ icode = CODE_FOR_vector_load_v4si;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_4sf:
- icode = CODE_FOR_altivec_lvx_v4sf;
+ icode = CODE_FOR_vector_load_v4sf;
break;
default:
*expandedp = false;
@@ -8400,16 +9646,16 @@ altivec_expand_st_builtin (tree exp, rtx
switch (fcode)
{
case ALTIVEC_BUILTIN_ST_INTERNAL_16qi:
- icode = CODE_FOR_altivec_stvx_v16qi;
+ icode = CODE_FOR_vector_store_v16qi;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_8hi:
- icode = CODE_FOR_altivec_stvx_v8hi;
+ icode = CODE_FOR_vector_store_v8hi;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_4si:
- icode = CODE_FOR_altivec_stvx_v4si;
+ icode = CODE_FOR_vector_store_v4si;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_4sf:
- icode = CODE_FOR_altivec_stvx_v4sf;
+ icode = CODE_FOR_vector_store_v4sf;
break;
default:
*expandedp = false;
@@ -8615,8 +9861,10 @@ altivec_expand_builtin (tree exp, rtx ta
enum machine_mode tmode, mode0;
unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
- if (fcode >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
- && fcode <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ if ((fcode >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
+ && fcode <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ || (fcode >= VSX_BUILTIN_OVERLOADED_FIRST
+ && fcode <= VSX_BUILTIN_OVERLOADED_LAST))
{
*expandedp = true;
error ("unresolved overload for Altivec builtin %qF", fndecl);
@@ -8724,18 +9972,24 @@ altivec_expand_builtin (tree exp, rtx ta
case ALTIVEC_BUILTIN_VEC_INIT_V8HI:
case ALTIVEC_BUILTIN_VEC_INIT_V16QI:
case ALTIVEC_BUILTIN_VEC_INIT_V4SF:
+ case VSX_BUILTIN_VEC_INIT_V2DF:
+ case VSX_BUILTIN_VEC_INIT_V2DI:
return altivec_expand_vec_init_builtin (TREE_TYPE (exp), exp, target);
case ALTIVEC_BUILTIN_VEC_SET_V4SI:
case ALTIVEC_BUILTIN_VEC_SET_V8HI:
case ALTIVEC_BUILTIN_VEC_SET_V16QI:
case ALTIVEC_BUILTIN_VEC_SET_V4SF:
+ case VSX_BUILTIN_VEC_SET_V2DF:
+ case VSX_BUILTIN_VEC_SET_V2DI:
return altivec_expand_vec_set_builtin (exp);
case ALTIVEC_BUILTIN_VEC_EXT_V4SI:
case ALTIVEC_BUILTIN_VEC_EXT_V8HI:
case ALTIVEC_BUILTIN_VEC_EXT_V16QI:
case ALTIVEC_BUILTIN_VEC_EXT_V4SF:
+ case VSX_BUILTIN_VEC_EXT_V2DF:
+ case VSX_BUILTIN_VEC_EXT_V2DI:
return altivec_expand_vec_ext_builtin (exp, target);
default:
@@ -8753,8 +10007,7 @@ altivec_expand_builtin (tree exp, rtx ta
dp = bdesc_altivec_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_altivec_preds); i++, dp++)
if (dp->code == fcode)
- return altivec_expand_predicate_builtin (dp->icode, dp->opcode,
- exp, target);
+ return altivec_expand_predicate_builtin (dp->icode, exp, target);
/* LV* are funky. We initialized them differently. */
switch (fcode)
@@ -8804,6 +10057,26 @@ altivec_expand_builtin (tree exp, rtx ta
/* Expand the builtin in EXP and store the result in TARGET. Store
true in *EXPANDEDP if we found a builtin to expand. */
static rtx
+vsx_expand_builtin (tree exp, rtx target ATTRIBUTE_UNUSED, bool *expandedp)
+{
+ tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
+ unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
+
+ if (fcode >= VSX_BUILTIN_OVERLOADED_FIRST
+ && fcode <= VSX_BUILTIN_OVERLOADED_LAST)
+ {
+ *expandedp = true;
+ error ("unresolved overload for vsx builtin %qF", fndecl);
+ return const0_rtx;
+ }
+
+ *expandedp = false;
+ return NULL_RTX;
+}
+
+/* Expand the builtin in EXP and store the result in TARGET. Store
+ true in *EXPANDEDP if we found a builtin to expand. */
+static rtx
paired_expand_builtin (tree exp, rtx target, bool * expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
@@ -9248,13 +10521,18 @@ rs6000_expand_builtin (tree exp, rtx tar
bool success;
if (fcode == RS6000_BUILTIN_RECIP)
- return rs6000_expand_binop_builtin (CODE_FOR_recipdf3, exp, target);
+ return rs6000_expand_binop_builtin (CODE_FOR_recipdf3, exp, target);
if (fcode == RS6000_BUILTIN_RECIPF)
- return rs6000_expand_binop_builtin (CODE_FOR_recipsf3, exp, target);
+ return rs6000_expand_binop_builtin (CODE_FOR_recipsf3, exp, target);
if (fcode == RS6000_BUILTIN_RSQRTF)
- return rs6000_expand_unop_builtin (CODE_FOR_rsqrtsf2, exp, target);
+ return rs6000_expand_unop_builtin (CODE_FOR_rsqrtsf2, exp, target);
+
+ if (fcode == POWER7_BUILTIN_BPERMD)
+ return rs6000_expand_binop_builtin (((TARGET_64BIT)
+ ? CODE_FOR_bpermd_di
+ : CODE_FOR_bpermd_si), exp, target);
if (fcode == ALTIVEC_BUILTIN_MASK_FOR_LOAD
|| fcode == ALTIVEC_BUILTIN_MASK_FOR_STORE)
@@ -9315,6 +10593,13 @@ rs6000_expand_builtin (tree exp, rtx tar
if (success)
return ret;
}
+ if (TARGET_VSX)
+ {
+ ret = vsx_expand_builtin (exp, target, &success);
+
+ if (success)
+ return ret;
+ }
if (TARGET_SPE)
{
ret = spe_expand_builtin (exp, target, &success);
@@ -9330,7 +10615,7 @@ rs6000_expand_builtin (tree exp, rtx tar
return ret;
}
- gcc_assert (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT);
+ gcc_assert (TARGET_ALTIVEC || TARGET_VSX || TARGET_SPE || TARGET_PAIRED_FLOAT);
/* Handle simple unary operations. */
d = (struct builtin_description *) bdesc_1arg;
@@ -9360,6 +10645,8 @@ rs6000_init_builtins (void)
V2SI_type_node = build_vector_type (intSI_type_node, 2);
V2SF_type_node = build_vector_type (float_type_node, 2);
+ V2DI_type_node = build_vector_type (intDI_type_node, 2);
+ V2DF_type_node = build_vector_type (double_type_node, 2);
V4HI_type_node = build_vector_type (intHI_type_node, 4);
V4SI_type_node = build_vector_type (intSI_type_node, 4);
V4SF_type_node = build_vector_type (float_type_node, 4);
@@ -9369,6 +10656,7 @@ rs6000_init_builtins (void)
unsigned_V16QI_type_node = build_vector_type (unsigned_intQI_type_node, 16);
unsigned_V8HI_type_node = build_vector_type (unsigned_intHI_type_node, 8);
unsigned_V4SI_type_node = build_vector_type (unsigned_intSI_type_node, 4);
+ unsigned_V2DI_type_node = build_vector_type (unsigned_intDI_type_node, 2);
opaque_V2SF_type_node = build_opaque_vector_type (float_type_node, 2);
opaque_V2SI_type_node = build_opaque_vector_type (intSI_type_node, 2);
@@ -9382,6 +10670,7 @@ rs6000_init_builtins (void)
bool_char_type_node = build_distinct_type_copy (unsigned_intQI_type_node);
bool_short_type_node = build_distinct_type_copy (unsigned_intHI_type_node);
bool_int_type_node = build_distinct_type_copy (unsigned_intSI_type_node);
+ bool_long_type_node = build_distinct_type_copy (unsigned_intDI_type_node);
pixel_type_node = build_distinct_type_copy (unsigned_intHI_type_node);
long_integer_type_internal_node = long_integer_type_node;
@@ -9392,9 +10681,36 @@ rs6000_init_builtins (void)
uintHI_type_internal_node = unsigned_intHI_type_node;
intSI_type_internal_node = intSI_type_node;
uintSI_type_internal_node = unsigned_intSI_type_node;
+ intDI_type_internal_node = intDI_type_node;
+ uintDI_type_internal_node = unsigned_intDI_type_node;
float_type_internal_node = float_type_node;
+ double_type_internal_node = float_type_node;
void_type_internal_node = void_type_node;
+ /* Initialize the modes for builtin_function_type, mapping a machine mode to
+ tree type node. */
+ builtin_mode_to_type[QImode][0] = integer_type_node;
+ builtin_mode_to_type[HImode][0] = integer_type_node;
+ builtin_mode_to_type[SImode][0] = intSI_type_node;
+ builtin_mode_to_type[SImode][1] = unsigned_intSI_type_node;
+ builtin_mode_to_type[DImode][0] = intDI_type_node;
+ builtin_mode_to_type[DImode][1] = unsigned_intDI_type_node;
+ builtin_mode_to_type[SFmode][0] = float_type_node;
+ builtin_mode_to_type[DFmode][0] = double_type_node;
+ builtin_mode_to_type[V2SImode][0] = V2SI_type_node;
+ builtin_mode_to_type[V2SFmode][0] = V2SF_type_node;
+ builtin_mode_to_type[V2DImode][0] = V2DI_type_node;
+ builtin_mode_to_type[V2DImode][1] = unsigned_V2DI_type_node;
+ builtin_mode_to_type[V2DFmode][0] = V2DF_type_node;
+ builtin_mode_to_type[V4HImode][0] = V4HI_type_node;
+ builtin_mode_to_type[V4SImode][0] = V4SI_type_node;
+ builtin_mode_to_type[V4SImode][1] = unsigned_V4SI_type_node;
+ builtin_mode_to_type[V4SFmode][0] = V4SF_type_node;
+ builtin_mode_to_type[V8HImode][0] = V8HI_type_node;
+ builtin_mode_to_type[V8HImode][1] = unsigned_V8HI_type_node;
+ builtin_mode_to_type[V16QImode][0] = V16QI_type_node;
+ builtin_mode_to_type[V16QImode][1] = unsigned_V16QI_type_node;
+
tdecl = build_decl (TYPE_DECL, get_identifier ("__bool char"),
bool_char_type_node);
TYPE_NAME (bool_char_type_node) = tdecl;
@@ -9415,6 +10731,7 @@ rs6000_init_builtins (void)
bool_V16QI_type_node = build_vector_type (bool_char_type_node, 16);
bool_V8HI_type_node = build_vector_type (bool_short_type_node, 8);
bool_V4SI_type_node = build_vector_type (bool_int_type_node, 4);
+ bool_V2DI_type_node = build_vector_type (bool_long_type_node, 2);
pixel_V8HI_type_node = build_vector_type (pixel_type_node, 8);
tdecl = build_decl (TYPE_DECL, get_identifier ("__vector unsigned char"),
@@ -9465,39 +10782,69 @@ rs6000_init_builtins (void)
TYPE_NAME (pixel_V8HI_type_node) = tdecl;
(*lang_hooks.decls.pushdecl) (tdecl);
+ if (TARGET_VSX)
+ {
+ tdecl = build_decl (TYPE_DECL, get_identifier ("__vector double"),
+ unsigned_V2DI_type_node);
+ TYPE_NAME (V2DF_type_node) = tdecl;
+ (*lang_hooks.decls.pushdecl) (tdecl);
+
+ tdecl = build_decl (TYPE_DECL, get_identifier ("__vector long"),
+ unsigned_V2DI_type_node);
+ TYPE_NAME (V2DI_type_node) = tdecl;
+ (*lang_hooks.decls.pushdecl) (tdecl);
+
+ tdecl = build_decl (TYPE_DECL, get_identifier ("__vector unsigned long"),
+ unsigned_V2DI_type_node);
+ TYPE_NAME (unsigned_V2DI_type_node) = tdecl;
+ (*lang_hooks.decls.pushdecl) (tdecl);
+
+ tdecl = build_decl (TYPE_DECL, get_identifier ("__vector __bool long"),
+ bool_V2DI_type_node);
+ TYPE_NAME (bool_V2DI_type_node) = tdecl;
+ (*lang_hooks.decls.pushdecl) (tdecl);
+ }
+
if (TARGET_PAIRED_FLOAT)
paired_init_builtins ();
if (TARGET_SPE)
spe_init_builtins ();
if (TARGET_ALTIVEC)
altivec_init_builtins ();
- if (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT)
+ if (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT || TARGET_VSX)
rs6000_common_init_builtins ();
if (TARGET_PPC_GFXOPT)
{
- tree ftype = build_function_type_list (float_type_node,
- float_type_node,
- float_type_node,
- NULL_TREE);
+ tree ftype = builtin_function_type (SFmode, SFmode, SFmode, VOIDmode,
+ RS6000_BUILTIN_RECIPF,
+ "__builtin_recipdivf");
def_builtin (MASK_PPC_GFXOPT, "__builtin_recipdivf", ftype,
RS6000_BUILTIN_RECIPF);
- ftype = build_function_type_list (float_type_node,
- float_type_node,
- NULL_TREE);
+ ftype = builtin_function_type (SFmode, SFmode, VOIDmode, VOIDmode,
+ RS6000_BUILTIN_RSQRTF,
+ "__builtin_rsqrtf");
def_builtin (MASK_PPC_GFXOPT, "__builtin_rsqrtf", ftype,
RS6000_BUILTIN_RSQRTF);
}
if (TARGET_POPCNTB)
{
- tree ftype = build_function_type_list (double_type_node,
- double_type_node,
- double_type_node,
- NULL_TREE);
+ tree ftype = builtin_function_type (DFmode, DFmode, DFmode, VOIDmode,
+ RS6000_BUILTIN_RECIP,
+ "__builtin_recipdiv");
def_builtin (MASK_POPCNTB, "__builtin_recipdiv", ftype,
RS6000_BUILTIN_RECIP);
}
+ if (TARGET_POPCNTD)
+ {
+ enum machine_mode mode = (TARGET_64BIT) ? DImode : SImode;
+ tree ftype = builtin_function_type (mode, mode, mode, VOIDmode,
+ POWER7_BUILTIN_BPERMD,
+ "__builtin_bpermd");
+ def_builtin (MASK_POPCNTD, "__builtin_bpermd", ftype,
+ POWER7_BUILTIN_BPERMD);
+ }
#if TARGET_XCOFF
/* AIX libm provides clog as __clog. */
@@ -9926,6 +11273,10 @@ altivec_init_builtins (void)
= build_function_type_list (integer_type_node,
integer_type_node, V4SF_type_node,
V4SF_type_node, NULL_TREE);
+ tree int_ftype_int_v2df_v2df
+ = build_function_type_list (integer_type_node,
+ integer_type_node, V2DF_type_node,
+ V2DF_type_node, NULL_TREE);
tree v4si_ftype_v4si
= build_function_type_list (V4SI_type_node, V4SI_type_node, NULL_TREE);
tree v8hi_ftype_v8hi
@@ -9934,6 +11285,8 @@ altivec_init_builtins (void)
= build_function_type_list (V16QI_type_node, V16QI_type_node, NULL_TREE);
tree v4sf_ftype_v4sf
= build_function_type_list (V4SF_type_node, V4SF_type_node, NULL_TREE);
+ tree v2df_ftype_v2df
+ = build_function_type_list (V2DF_type_node, V2DF_type_node, NULL_TREE);
tree void_ftype_pcvoid_int_int
= build_function_type_list (void_type_node,
pcvoid_type_node, integer_type_node,
@@ -10036,8 +11389,10 @@ altivec_init_builtins (void)
{
enum machine_mode mode1;
tree type;
- bool is_overloaded = dp->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
- && dp->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST;
+ bool is_overloaded = ((dp->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
+ && dp->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ || (dp->code >= VSX_BUILTIN_OVERLOADED_FIRST
+ && dp->code <= VSX_BUILTIN_OVERLOADED_LAST));
if (is_overloaded)
mode1 = VOIDmode;
@@ -10061,6 +11416,9 @@ altivec_init_builtins (void)
case V4SFmode:
type = int_ftype_int_v4sf_v4sf;
break;
+ case V2DFmode:
+ type = int_ftype_int_v2df_v2df;
+ break;
default:
gcc_unreachable ();
}
@@ -10091,6 +11449,9 @@ altivec_init_builtins (void)
case V4SFmode:
type = v4sf_ftype_v4sf;
break;
+ case V2DFmode:
+ type = v2df_ftype_v2df;
+ break;
default:
gcc_unreachable ();
}
@@ -10150,6 +11511,19 @@ altivec_init_builtins (void)
def_builtin (MASK_ALTIVEC, "__builtin_vec_init_v4sf", ftype,
ALTIVEC_BUILTIN_VEC_INIT_V4SF);
+ if (TARGET_VSX)
+ {
+ ftype = build_function_type_list (V2DF_type_node, double_type_node,
+ double_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_init_v2df", ftype,
+ VSX_BUILTIN_VEC_INIT_V2DF);
+
+ ftype = build_function_type_list (V2DI_type_node, intDI_type_node,
+ intDI_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_init_v2di", ftype,
+ VSX_BUILTIN_VEC_INIT_V2DI);
+ }
+
/* Access to the vec_set patterns. */
ftype = build_function_type_list (V4SI_type_node, V4SI_type_node,
intSI_type_node,
@@ -10163,7 +11537,7 @@ altivec_init_builtins (void)
def_builtin (MASK_ALTIVEC, "__builtin_vec_set_v8hi", ftype,
ALTIVEC_BUILTIN_VEC_SET_V8HI);
- ftype = build_function_type_list (V8HI_type_node, V16QI_type_node,
+ ftype = build_function_type_list (V16QI_type_node, V16QI_type_node,
intQI_type_node,
integer_type_node, NULL_TREE);
def_builtin (MASK_ALTIVEC, "__builtin_vec_set_v16qi", ftype,
@@ -10172,9 +11546,24 @@ altivec_init_builtins (void)
ftype = build_function_type_list (V4SF_type_node, V4SF_type_node,
float_type_node,
integer_type_node, NULL_TREE);
- def_builtin (MASK_ALTIVEC, "__builtin_vec_set_v4sf", ftype,
+ def_builtin (MASK_ALTIVEC|MASK_VSX, "__builtin_vec_set_v4sf", ftype,
ALTIVEC_BUILTIN_VEC_SET_V4SF);
+ if (TARGET_VSX)
+ {
+ ftype = build_function_type_list (V2DF_type_node, V2DF_type_node,
+ double_type_node,
+ integer_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_set_v2df", ftype,
+ VSX_BUILTIN_VEC_SET_V2DF);
+
+ ftype = build_function_type_list (V2DI_type_node, V2DI_type_node,
+ intDI_type_node,
+ integer_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_set_v2di", ftype,
+ VSX_BUILTIN_VEC_SET_V2DI);
+ }
+
/* Access to the vec_extract patterns. */
ftype = build_function_type_list (intSI_type_node, V4SI_type_node,
integer_type_node, NULL_TREE);
@@ -10193,539 +11582,376 @@ altivec_init_builtins (void)
ftype = build_function_type_list (float_type_node, V4SF_type_node,
integer_type_node, NULL_TREE);
- def_builtin (MASK_ALTIVEC, "__builtin_vec_ext_v4sf", ftype,
+ def_builtin (MASK_ALTIVEC|MASK_VSX, "__builtin_vec_ext_v4sf", ftype,
ALTIVEC_BUILTIN_VEC_EXT_V4SF);
+
+ if (TARGET_VSX)
+ {
+ ftype = build_function_type_list (double_type_node, V2DF_type_node,
+ integer_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_ext_v2df", ftype,
+ VSX_BUILTIN_VEC_EXT_V2DF);
+
+ ftype = build_function_type_list (intDI_type_node, V2DI_type_node,
+ integer_type_node, NULL_TREE);
+ def_builtin (MASK_VSX, "__builtin_vec_ext_v2di", ftype,
+ VSX_BUILTIN_VEC_EXT_V2DI);
+ }
}
-static void
-rs6000_common_init_builtins (void)
+/* Hash function for builtin functions with up to 3 arguments and a return
+ type. */
+static unsigned
+builtin_hash_function (const void *hash_entry)
{
- const struct builtin_description *d;
- size_t i;
+ unsigned ret = 0;
+ int i;
+ const struct builtin_hash_struct *bh =
+ (const struct builtin_hash_struct *) hash_entry;
- tree v2sf_ftype_v2sf_v2sf_v2sf
- = build_function_type_list (V2SF_type_node,
- V2SF_type_node, V2SF_type_node,
- V2SF_type_node, NULL_TREE);
-
- tree v4sf_ftype_v4sf_v4sf_v16qi
- = build_function_type_list (V4SF_type_node,
- V4SF_type_node, V4SF_type_node,
- V16QI_type_node, NULL_TREE);
- tree v4si_ftype_v4si_v4si_v16qi
- = build_function_type_list (V4SI_type_node,
- V4SI_type_node, V4SI_type_node,
- V16QI_type_node, NULL_TREE);
- tree v8hi_ftype_v8hi_v8hi_v16qi
- = build_function_type_list (V8HI_type_node,
- V8HI_type_node, V8HI_type_node,
- V16QI_type_node, NULL_TREE);
- tree v16qi_ftype_v16qi_v16qi_v16qi
- = build_function_type_list (V16QI_type_node,
- V16QI_type_node, V16QI_type_node,
- V16QI_type_node, NULL_TREE);
- tree v4si_ftype_int
- = build_function_type_list (V4SI_type_node, integer_type_node, NULL_TREE);
- tree v8hi_ftype_int
- = build_function_type_list (V8HI_type_node, integer_type_node, NULL_TREE);
- tree v16qi_ftype_int
- = build_function_type_list (V16QI_type_node, integer_type_node, NULL_TREE);
- tree v8hi_ftype_v16qi
- = build_function_type_list (V8HI_type_node, V16QI_type_node, NULL_TREE);
- tree v4sf_ftype_v4sf
- = build_function_type_list (V4SF_type_node, V4SF_type_node, NULL_TREE);
+ for (i = 0; i < 4; i++)
+ {
+ ret = (ret * (unsigned)MAX_MACHINE_MODE) + ((unsigned)bh->mode[i]);
+ ret = (ret * 2) + bh->uns_p[i];
+ }
- tree v2si_ftype_v2si_v2si
- = build_function_type_list (opaque_V2SI_type_node,
- opaque_V2SI_type_node,
- opaque_V2SI_type_node, NULL_TREE);
-
- tree v2sf_ftype_v2sf_v2sf_spe
- = build_function_type_list (opaque_V2SF_type_node,
- opaque_V2SF_type_node,
- opaque_V2SF_type_node, NULL_TREE);
-
- tree v2sf_ftype_v2sf_v2sf
- = build_function_type_list (V2SF_type_node,
- V2SF_type_node,
- V2SF_type_node, NULL_TREE);
-
-
- tree v2si_ftype_int_int
- = build_function_type_list (opaque_V2SI_type_node,
- integer_type_node, integer_type_node,
- NULL_TREE);
+ return ret;
+}
- tree opaque_ftype_opaque
- = build_function_type_list (opaque_V4SI_type_node,
- opaque_V4SI_type_node, NULL_TREE);
+/* Compare builtin hash entries H1 and H2 for equivalence. */
+static int
+builtin_hash_eq (const void *h1, const void *h2)
+{
+ const struct builtin_hash_struct *p1 = (const struct builtin_hash_struct *) h1;
+ const struct builtin_hash_struct *p2 = (const struct builtin_hash_struct *) h2;
- tree v2si_ftype_v2si
- = build_function_type_list (opaque_V2SI_type_node,
- opaque_V2SI_type_node, NULL_TREE);
-
- tree v2sf_ftype_v2sf_spe
- = build_function_type_list (opaque_V2SF_type_node,
- opaque_V2SF_type_node, NULL_TREE);
-
- tree v2sf_ftype_v2sf
- = build_function_type_list (V2SF_type_node,
- V2SF_type_node, NULL_TREE);
-
- tree v2sf_ftype_v2si
- = build_function_type_list (opaque_V2SF_type_node,
- opaque_V2SI_type_node, NULL_TREE);
-
- tree v2si_ftype_v2sf
- = build_function_type_list (opaque_V2SI_type_node,
- opaque_V2SF_type_node, NULL_TREE);
-
- tree v2si_ftype_v2si_char
- = build_function_type_list (opaque_V2SI_type_node,
- opaque_V2SI_type_node,
- char_type_node, NULL_TREE);
-
- tree v2si_ftype_int_char
- = build_function_type_list (opaque_V2SI_type_node,
- integer_type_node, char_type_node, NULL_TREE);
-
- tree v2si_ftype_char
- = build_function_type_list (opaque_V2SI_type_node,
- char_type_node, NULL_TREE);
+ return ((p1->mode[0] == p2->mode[0])
+ && (p1->mode[1] == p2->mode[1])
+ && (p1->mode[2] == p2->mode[2])
+ && (p1->mode[3] == p2->mode[3])
+ && (p1->uns_p[0] == p2->uns_p[0])
+ && (p1->uns_p[1] == p2->uns_p[1])
+ && (p1->uns_p[2] == p2->uns_p[2])
+ && (p1->uns_p[3] == p2->uns_p[3]));
+}
+
+/* Map types for builtin functions with an explicit return type and up to 3
+ arguments. Functions with fewer than 3 arguments use VOIDmode as the type
+ of the argument. */
+static tree
+builtin_function_type (enum machine_mode mode_ret, enum machine_mode mode_arg0,
+ enum machine_mode mode_arg1, enum machine_mode mode_arg2,
+ enum rs6000_builtins builtin, const char *name)
+{
+ struct builtin_hash_struct h;
+ struct builtin_hash_struct *h2;
+ void **found;
+ int num_args = 3;
+ int i;
+ tree ret_type = NULL_TREE;
+ tree arg_type[3] = { NULL_TREE, NULL_TREE, NULL_TREE };
+ tree args;
+
+ /* Create builtin_hash_table. */
+ if (builtin_hash_table == NULL)
+ builtin_hash_table = htab_create_ggc (1500, builtin_hash_function,
+ builtin_hash_eq, NULL);
+
+ h.type = NULL_TREE;
+ h.mode[0] = mode_ret;
+ h.mode[1] = mode_arg0;
+ h.mode[2] = mode_arg1;
+ h.mode[3] = mode_arg2;
+ h.uns_p[0] = 0;
+ h.uns_p[1] = 0;
+ h.uns_p[2] = 0;
+ h.uns_p[3] = 0;
+
+ /* If the builtin is a type that produces unsigned results or takes unsigned
+ arguments, and it is returned as a decl for the vectorizer (such as
+ widening multiplies, permute), make sure the arguments and return value
+ are type correct. */
+ switch (builtin)
+ {
+ /* unsigned 2 argument functions. */
+ case ALTIVEC_BUILTIN_VMULEUB_UNS:
+ case ALTIVEC_BUILTIN_VMULEUH_UNS:
+ case ALTIVEC_BUILTIN_VMULOUB_UNS:
+ case ALTIVEC_BUILTIN_VMULOUH_UNS:
+ h.uns_p[0] = 1;
+ h.uns_p[1] = 1;
+ h.uns_p[2] = 1;
+ break;
+
+ /* unsigned 3 argument functions. */
+ case ALTIVEC_BUILTIN_VPERM_16QI_UNS:
+ case ALTIVEC_BUILTIN_VPERM_8HI_UNS:
+ case ALTIVEC_BUILTIN_VPERM_4SI_UNS:
+ case ALTIVEC_BUILTIN_VPERM_2DI_UNS:
+ case ALTIVEC_BUILTIN_VSEL_16QI_UNS:
+ case ALTIVEC_BUILTIN_VSEL_8HI_UNS:
+ case ALTIVEC_BUILTIN_VSEL_4SI_UNS:
+ case ALTIVEC_BUILTIN_VSEL_2DI_UNS:
+ case VSX_BUILTIN_VPERM_16QI_UNS:
+ case VSX_BUILTIN_VPERM_8HI_UNS:
+ case VSX_BUILTIN_VPERM_4SI_UNS:
+ case VSX_BUILTIN_VPERM_2DI_UNS:
+ case VSX_BUILTIN_XXSEL_16QI_UNS:
+ case VSX_BUILTIN_XXSEL_8HI_UNS:
+ case VSX_BUILTIN_XXSEL_4SI_UNS:
+ case VSX_BUILTIN_XXSEL_2DI_UNS:
+ h.uns_p[0] = 1;
+ h.uns_p[1] = 1;
+ h.uns_p[2] = 1;
+ h.uns_p[3] = 1;
+ break;
+
+ /* signed permute functions with unsigned char mask. */
+ case ALTIVEC_BUILTIN_VPERM_16QI:
+ case ALTIVEC_BUILTIN_VPERM_8HI:
+ case ALTIVEC_BUILTIN_VPERM_4SI:
+ case ALTIVEC_BUILTIN_VPERM_4SF:
+ case ALTIVEC_BUILTIN_VPERM_2DI:
+ case ALTIVEC_BUILTIN_VPERM_2DF:
+ case VSX_BUILTIN_VPERM_16QI:
+ case VSX_BUILTIN_VPERM_8HI:
+ case VSX_BUILTIN_VPERM_4SI:
+ case VSX_BUILTIN_VPERM_4SF:
+ case VSX_BUILTIN_VPERM_2DI:
+ case VSX_BUILTIN_VPERM_2DF:
+ h.uns_p[3] = 1;
+ break;
+
+ /* unsigned args, signed return. */
+ case VSX_BUILTIN_XVCVUXDDP_UNS:
+ case VECTOR_BUILTIN_UNSFLOAT_V4SI_V4SF:
+ h.uns_p[1] = 1;
+ break;
+
+ /* signed args, unsigned return. */
+ case VSX_BUILTIN_XVCVDPUXDS_UNS:
+ case VECTOR_BUILTIN_FIXUNS_V4SF_V4SI:
+ h.uns_p[0] = 1;
+ break;
- tree int_ftype_int_int
- = build_function_type_list (integer_type_node,
- integer_type_node, integer_type_node,
- NULL_TREE);
+ default:
+ break;
+ }
- tree opaque_ftype_opaque_opaque
- = build_function_type_list (opaque_V4SI_type_node,
- opaque_V4SI_type_node, opaque_V4SI_type_node, NULL_TREE);
- tree v4si_ftype_v4si_v4si
- = build_function_type_list (V4SI_type_node,
- V4SI_type_node, V4SI_type_node, NULL_TREE);
- tree v4sf_ftype_v4si_int
- = build_function_type_list (V4SF_type_node,
- V4SI_type_node, integer_type_node, NULL_TREE);
- tree v4si_ftype_v4sf_int
- = build_function_type_list (V4SI_type_node,
- V4SF_type_node, integer_type_node, NULL_TREE);
- tree v4si_ftype_v4si_int
- = build_function_type_list (V4SI_type_node,
- V4SI_type_node, integer_type_node, NULL_TREE);
- tree v8hi_ftype_v8hi_int
- = build_function_type_list (V8HI_type_node,
- V8HI_type_node, integer_type_node, NULL_TREE);
- tree v16qi_ftype_v16qi_int
- = build_function_type_list (V16QI_type_node,
- V16QI_type_node, integer_type_node, NULL_TREE);
- tree v16qi_ftype_v16qi_v16qi_int
- = build_function_type_list (V16QI_type_node,
- V16QI_type_node, V16QI_type_node,
- integer_type_node, NULL_TREE);
- tree v8hi_ftype_v8hi_v8hi_int
- = build_function_type_list (V8HI_type_node,
- V8HI_type_node, V8HI_type_node,
- integer_type_node, NULL_TREE);
- tree v4si_ftype_v4si_v4si_int
- = build_function_type_list (V4SI_type_node,
- V4SI_type_node, V4SI_type_node,
- integer_type_node, NULL_TREE);
- tree v4sf_ftype_v4sf_v4sf_int
- = build_function_type_list (V4SF_type_node,
- V4SF_type_node, V4SF_type_node,
- integer_type_node, NULL_TREE);
- tree v4sf_ftype_v4sf_v4sf
- = build_function_type_list (V4SF_type_node,
- V4SF_type_node, V4SF_type_node, NULL_TREE);
- tree opaque_ftype_opaque_opaque_opaque
- = build_function_type_list (opaque_V4SI_type_node,
- opaque_V4SI_type_node, opaque_V4SI_type_node,
- opaque_V4SI_type_node, NULL_TREE);
- tree v4sf_ftype_v4sf_v4sf_v4si
- = build_function_type_list (V4SF_type_node,
- V4SF_type_node, V4SF_type_node,
- V4SI_type_node, NULL_TREE);
- tree v4sf_ftype_v4sf_v4sf_v4sf
- = build_function_type_list (V4SF_type_node,
- V4SF_type_node, V4SF_type_node,
- V4SF_type_node, NULL_TREE);
- tree v4si_ftype_v4si_v4si_v4si
- = build_function_type_list (V4SI_type_node,
- V4SI_type_node, V4SI_type_node,
- V4SI_type_node, NULL_TREE);
- tree v8hi_ftype_v8hi_v8hi
- = build_function_type_list (V8HI_type_node,
- V8HI_type_node, V8HI_type_node, NULL_TREE);
- tree v8hi_ftype_v8hi_v8hi_v8hi
- = build_function_type_list (V8HI_type_node,
- V8HI_type_node, V8HI_type_node,
- V8HI_type_node, NULL_TREE);
- tree v4si_ftype_v8hi_v8hi_v4si
- = build_function_type_list (V4SI_type_node,
- V8HI_type_node, V8HI_type_node,
- V4SI_type_node, NULL_TREE);
- tree v4si_ftype_v16qi_v16qi_v4si
- = build_function_type_list (V4SI_type_node,
- V16QI_type_node, V16QI_type_node,
- V4SI_type_node, NULL_TREE);
- tree v16qi_ftype_v16qi_v16qi
- = build_function_type_list (V16QI_type_node,
- V16QI_type_node, V16QI_type_node, NULL_TREE);
- tree v4si_ftype_v4sf_v4sf
- = build_function_type_list (V4SI_type_node,
- V4SF_type_node, V4SF_type_node, NULL_TREE);
- tree v8hi_ftype_v16qi_v16qi
- = build_function_type_list (V8HI_type_node,
- V16QI_type_node, V16QI_type_node, NULL_TREE);
- tree v4si_ftype_v8hi_v8hi
- = build_function_type_list (V4SI_type_node,
- V8HI_type_node, V8HI_type_node, NULL_TREE);
- tree v8hi_ftype_v4si_v4si
- = build_function_type_list (V8HI_type_node,
- V4SI_type_node, V4SI_type_node, NULL_TREE);
- tree v16qi_ftype_v8hi_v8hi
- = build_function_type_list (V16QI_type_node,
- V8HI_type_node, V8HI_type_node, NULL_TREE);
- tree v4si_ftype_v16qi_v4si
- = build_function_type_list (V4SI_type_node,
- V16QI_type_node, V4SI_type_node, NULL_TREE);
- tree v4si_ftype_v16qi_v16qi
- = build_function_type_list (V4SI_type_node,
- V16QI_type_node, V16QI_type_node, NULL_TREE);
- tree v4si_ftype_v8hi_v4si
- = build_function_type_list (V4SI_type_node,
- V8HI_type_node, V4SI_type_node, NULL_TREE);
- tree v4si_ftype_v8hi
- = build_function_type_list (V4SI_type_node, V8HI_type_node, NULL_TREE);
- tree int_ftype_v4si_v4si
- = build_function_type_list (integer_type_node,
- V4SI_type_node, V4SI_type_node, NULL_TREE);
- tree int_ftype_v4sf_v4sf
- = build_function_type_list (integer_type_node,
- V4SF_type_node, V4SF_type_node, NULL_TREE);
- tree int_ftype_v16qi_v16qi
- = build_function_type_list (integer_type_node,
- V16QI_type_node, V16QI_type_node, NULL_TREE);
- tree int_ftype_v8hi_v8hi
- = build_function_type_list (integer_type_node,
- V8HI_type_node, V8HI_type_node, NULL_TREE);
+ /* Figure out how many args are present. */
+ while (num_args > 0 && h.mode[num_args] == VOIDmode)
+ num_args--;
+
+ if (num_args == 0)
+ fatal_error ("internal error: builtin function %s had no type", name);
+
+ ret_type = builtin_mode_to_type[h.mode[0]][h.uns_p[0]];
+ if (!ret_type && h.uns_p[0])
+ ret_type = builtin_mode_to_type[h.mode[0]][0];
+
+ if (!ret_type)
+ fatal_error ("internal error: builtin function %s had an unexpected "
+ "return type %s", name, GET_MODE_NAME (h.mode[0]));
+
+ for (i = 0; i < num_args; i++)
+ {
+ int m = (int) h.mode[i+1];
+ int uns_p = h.uns_p[i+1];
- /* Add the simple ternary operators. */
+ arg_type[i] = builtin_mode_to_type[m][uns_p];
+ if (!arg_type[i] && uns_p)
+ arg_type[i] = builtin_mode_to_type[m][0];
+
+ if (!arg_type[i])
+ fatal_error ("internal error: builtin function %s, argument %d "
+ "had unexpected argument type %s", name, i,
+ GET_MODE_NAME (m));
+ }
+
+ found = htab_find_slot (builtin_hash_table, &h, INSERT);
+ if (*found == NULL)
+ {
+ h2 = GGC_NEW (struct builtin_hash_struct);
+ *h2 = h;
+ *found = (void *)h2;
+ args = void_list_node;
+
+ for (i = num_args - 1; i >= 0; i--)
+ args = tree_cons (NULL_TREE, arg_type[i], args);
+
+ h2->type = build_function_type (ret_type, args);
+ }
+
+ return ((struct builtin_hash_struct *)(*found))->type;
+}
+
+static void
+rs6000_common_init_builtins (void)
+{
+ const struct builtin_description *d;
+ size_t i;
+
+ tree opaque_ftype_opaque = NULL_TREE;
+ tree opaque_ftype_opaque_opaque = NULL_TREE;
+ tree opaque_ftype_opaque_opaque_opaque = NULL_TREE;
+ tree v2si_ftype_qi = NULL_TREE;
+ tree v2si_ftype_v2si_qi = NULL_TREE;
+ tree v2si_ftype_int_qi = NULL_TREE;
+
+ if (!TARGET_PAIRED_FLOAT)
+ {
+ builtin_mode_to_type[V2SImode][0] = opaque_V2SI_type_node;
+ builtin_mode_to_type[V2SFmode][0] = opaque_V2SF_type_node;
+ }
+
+ /* Add the ternary operators. */
d = bdesc_3arg;
for (i = 0; i < ARRAY_SIZE (bdesc_3arg); i++, d++)
{
- enum machine_mode mode0, mode1, mode2, mode3;
tree type;
- bool is_overloaded = d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
- && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST;
+ int mask = d->mask;
- if (is_overloaded)
- {
- mode0 = VOIDmode;
- mode1 = VOIDmode;
- mode2 = VOIDmode;
- mode3 = VOIDmode;
+ if ((mask != 0 && (mask & target_flags) == 0)
+ || (mask == 0 && !TARGET_PAIRED_FLOAT))
+ continue;
+
+ if ((d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
+ && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ || (d->code >= VSX_BUILTIN_OVERLOADED_FIRST
+ && d->code <= VSX_BUILTIN_OVERLOADED_LAST))
+ {
+ if (! (type = opaque_ftype_opaque_opaque_opaque))
+ type = opaque_ftype_opaque_opaque_opaque
+ = build_function_type_list (opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ NULL_TREE);
}
else
{
- if (d->name == 0 || d->icode == CODE_FOR_nothing)
+ enum insn_code icode = d->icode;
+ if (d->name == 0 || icode == CODE_FOR_nothing)
continue;
- mode0 = insn_data[d->icode].operand[0].mode;
- mode1 = insn_data[d->icode].operand[1].mode;
- mode2 = insn_data[d->icode].operand[2].mode;
- mode3 = insn_data[d->icode].operand[3].mode;
+ type = builtin_function_type (insn_data[icode].operand[0].mode,
+ insn_data[icode].operand[1].mode,
+ insn_data[icode].operand[2].mode,
+ insn_data[icode].operand[3].mode,
+ d->code, d->name);
}
- /* When all four are of the same mode. */
- if (mode0 == mode1 && mode1 == mode2 && mode2 == mode3)
- {
- switch (mode0)
- {
- case VOIDmode:
- type = opaque_ftype_opaque_opaque_opaque;
- break;
- case V4SImode:
- type = v4si_ftype_v4si_v4si_v4si;
- break;
- case V4SFmode:
- type = v4sf_ftype_v4sf_v4sf_v4sf;
- break;
- case V8HImode:
- type = v8hi_ftype_v8hi_v8hi_v8hi;
- break;
- case V16QImode:
- type = v16qi_ftype_v16qi_v16qi_v16qi;
- break;
- case V2SFmode:
- type = v2sf_ftype_v2sf_v2sf_v2sf;
- break;
- default:
- gcc_unreachable ();
- }
- }
- else if (mode0 == mode1 && mode1 == mode2 && mode3 == V16QImode)
- {
- switch (mode0)
- {
- case V4SImode:
- type = v4si_ftype_v4si_v4si_v16qi;
- break;
- case V4SFmode:
- type = v4sf_ftype_v4sf_v4sf_v16qi;
- break;
- case V8HImode:
- type = v8hi_ftype_v8hi_v8hi_v16qi;
- break;
- case V16QImode:
- type = v16qi_ftype_v16qi_v16qi_v16qi;
- break;
- default:
- gcc_unreachable ();
- }
- }
- else if (mode0 == V4SImode && mode1 == V16QImode && mode2 == V16QImode
- && mode3 == V4SImode)
- type = v4si_ftype_v16qi_v16qi_v4si;
- else if (mode0 == V4SImode && mode1 == V8HImode && mode2 == V8HImode
- && mode3 == V4SImode)
- type = v4si_ftype_v8hi_v8hi_v4si;
- else if (mode0 == V4SFmode && mode1 == V4SFmode && mode2 == V4SFmode
- && mode3 == V4SImode)
- type = v4sf_ftype_v4sf_v4sf_v4si;
-
- /* vchar, vchar, vchar, 4-bit literal. */
- else if (mode0 == V16QImode && mode1 == mode0 && mode2 == mode0
- && mode3 == QImode)
- type = v16qi_ftype_v16qi_v16qi_int;
-
- /* vshort, vshort, vshort, 4-bit literal. */
- else if (mode0 == V8HImode && mode1 == mode0 && mode2 == mode0
- && mode3 == QImode)
- type = v8hi_ftype_v8hi_v8hi_int;
-
- /* vint, vint, vint, 4-bit literal. */
- else if (mode0 == V4SImode && mode1 == mode0 && mode2 == mode0
- && mode3 == QImode)
- type = v4si_ftype_v4si_v4si_int;
-
- /* vfloat, vfloat, vfloat, 4-bit literal. */
- else if (mode0 == V4SFmode && mode1 == mode0 && mode2 == mode0
- && mode3 == QImode)
- type = v4sf_ftype_v4sf_v4sf_int;
-
- else
- gcc_unreachable ();
-
def_builtin (d->mask, d->name, type, d->code);
}
- /* Add the simple binary operators. */
+ /* Add the binary operators. */
d = (struct builtin_description *) bdesc_2arg;
for (i = 0; i < ARRAY_SIZE (bdesc_2arg); i++, d++)
{
enum machine_mode mode0, mode1, mode2;
tree type;
- bool is_overloaded = d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
- && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST;
+ int mask = d->mask;
- if (is_overloaded)
- {
- mode0 = VOIDmode;
- mode1 = VOIDmode;
- mode2 = VOIDmode;
+ if ((mask != 0 && (mask & target_flags) == 0)
+ || (mask == 0 && !TARGET_PAIRED_FLOAT))
+ continue;
+
+ if ((d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
+ && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ || (d->code >= VSX_BUILTIN_OVERLOADED_FIRST
+ && d->code <= VSX_BUILTIN_OVERLOADED_LAST))
+ {
+ if (! (type = opaque_ftype_opaque_opaque))
+ type = opaque_ftype_opaque_opaque
+ = build_function_type_list (opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ NULL_TREE);
}
else
{
- if (d->name == 0 || d->icode == CODE_FOR_nothing)
+ enum insn_code icode = d->icode;
+ if (d->name == 0 || icode == CODE_FOR_nothing)
continue;
- mode0 = insn_data[d->icode].operand[0].mode;
- mode1 = insn_data[d->icode].operand[1].mode;
- mode2 = insn_data[d->icode].operand[2].mode;
- }
+ mode0 = insn_data[icode].operand[0].mode;
+ mode1 = insn_data[icode].operand[1].mode;
+ mode2 = insn_data[icode].operand[2].mode;
- /* When all three operands are of the same mode. */
- if (mode0 == mode1 && mode1 == mode2)
- {
- switch (mode0)
+ if (mode0 == V2SImode && mode1 == V2SImode && mode2 == QImode)
{
- case VOIDmode:
- type = opaque_ftype_opaque_opaque;
- break;
- case V4SFmode:
- type = v4sf_ftype_v4sf_v4sf;
- break;
- case V4SImode:
- type = v4si_ftype_v4si_v4si;
- break;
- case V16QImode:
- type = v16qi_ftype_v16qi_v16qi;
- break;
- case V8HImode:
- type = v8hi_ftype_v8hi_v8hi;
- break;
- case V2SImode:
- type = v2si_ftype_v2si_v2si;
- break;
- case V2SFmode:
- if (TARGET_PAIRED_FLOAT)
- type = v2sf_ftype_v2sf_v2sf;
- else
- type = v2sf_ftype_v2sf_v2sf_spe;
- break;
- case SImode:
- type = int_ftype_int_int;
- break;
- default:
- gcc_unreachable ();
+ if (! (type = v2si_ftype_v2si_qi))
+ type = v2si_ftype_v2si_qi
+ = build_function_type_list (opaque_V2SI_type_node,
+ opaque_V2SI_type_node,
+ char_type_node,
+ NULL_TREE);
}
- }
-
- /* A few other combos we really don't want to do manually. */
-
- /* vint, vfloat, vfloat. */
- else if (mode0 == V4SImode && mode1 == V4SFmode && mode2 == V4SFmode)
- type = v4si_ftype_v4sf_v4sf;
-
- /* vshort, vchar, vchar. */
- else if (mode0 == V8HImode && mode1 == V16QImode && mode2 == V16QImode)
- type = v8hi_ftype_v16qi_v16qi;
-
- /* vint, vshort, vshort. */
- else if (mode0 == V4SImode && mode1 == V8HImode && mode2 == V8HImode)
- type = v4si_ftype_v8hi_v8hi;
-
- /* vshort, vint, vint. */
- else if (mode0 == V8HImode && mode1 == V4SImode && mode2 == V4SImode)
- type = v8hi_ftype_v4si_v4si;
-
- /* vchar, vshort, vshort. */
- else if (mode0 == V16QImode && mode1 == V8HImode && mode2 == V8HImode)
- type = v16qi_ftype_v8hi_v8hi;
-
- /* vint, vchar, vint. */
- else if (mode0 == V4SImode && mode1 == V16QImode && mode2 == V4SImode)
- type = v4si_ftype_v16qi_v4si;
- /* vint, vchar, vchar. */
- else if (mode0 == V4SImode && mode1 == V16QImode && mode2 == V16QImode)
- type = v4si_ftype_v16qi_v16qi;
-
- /* vint, vshort, vint. */
- else if (mode0 == V4SImode && mode1 == V8HImode && mode2 == V4SImode)
- type = v4si_ftype_v8hi_v4si;
-
- /* vint, vint, 5-bit literal. */
- else if (mode0 == V4SImode && mode1 == V4SImode && mode2 == QImode)
- type = v4si_ftype_v4si_int;
-
- /* vshort, vshort, 5-bit literal. */
- else if (mode0 == V8HImode && mode1 == V8HImode && mode2 == QImode)
- type = v8hi_ftype_v8hi_int;
-
- /* vchar, vchar, 5-bit literal. */
- else if (mode0 == V16QImode && mode1 == V16QImode && mode2 == QImode)
- type = v16qi_ftype_v16qi_int;
-
- /* vfloat, vint, 5-bit literal. */
- else if (mode0 == V4SFmode && mode1 == V4SImode && mode2 == QImode)
- type = v4sf_ftype_v4si_int;
-
- /* vint, vfloat, 5-bit literal. */
- else if (mode0 == V4SImode && mode1 == V4SFmode && mode2 == QImode)
- type = v4si_ftype_v4sf_int;
-
- else if (mode0 == V2SImode && mode1 == SImode && mode2 == SImode)
- type = v2si_ftype_int_int;
-
- else if (mode0 == V2SImode && mode1 == V2SImode && mode2 == QImode)
- type = v2si_ftype_v2si_char;
-
- else if (mode0 == V2SImode && mode1 == SImode && mode2 == QImode)
- type = v2si_ftype_int_char;
-
- else
- {
- /* int, x, x. */
- gcc_assert (mode0 == SImode);
- switch (mode1)
+ else if (mode0 == V2SImode && GET_MODE_CLASS (mode1) == MODE_INT
+ && mode2 == QImode)
{
- case V4SImode:
- type = int_ftype_v4si_v4si;
- break;
- case V4SFmode:
- type = int_ftype_v4sf_v4sf;
- break;
- case V16QImode:
- type = int_ftype_v16qi_v16qi;
- break;
- case V8HImode:
- type = int_ftype_v8hi_v8hi;
- break;
- default:
- gcc_unreachable ();
+ if (! (type = v2si_ftype_int_qi))
+ type = v2si_ftype_int_qi
+ = build_function_type_list (opaque_V2SI_type_node,
+ integer_type_node,
+ char_type_node,
+ NULL_TREE);
}
+
+ else
+ type = builtin_function_type (mode0, mode1, mode2, VOIDmode,
+ d->code, d->name);
}
def_builtin (d->mask, d->name, type, d->code);
}
- /* Add the simple unary operators. */
+ /* Add the unary operators. */
d = (struct builtin_description *) bdesc_1arg;
for (i = 0; i < ARRAY_SIZE (bdesc_1arg); i++, d++)
{
enum machine_mode mode0, mode1;
tree type;
- bool is_overloaded = d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
- && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST;
+ int mask = d->mask;
- if (is_overloaded)
- {
- mode0 = VOIDmode;
- mode1 = VOIDmode;
- }
+ if ((mask != 0 && (mask & target_flags) == 0)
+ || (mask == 0 && !TARGET_PAIRED_FLOAT))
+ continue;
+
+ if ((d->code >= ALTIVEC_BUILTIN_OVERLOADED_FIRST
+ && d->code <= ALTIVEC_BUILTIN_OVERLOADED_LAST)
+ || (d->code >= VSX_BUILTIN_OVERLOADED_FIRST
+ && d->code <= VSX_BUILTIN_OVERLOADED_LAST))
+ {
+ if (! (type = opaque_ftype_opaque))
+ type = opaque_ftype_opaque
+ = build_function_type_list (opaque_V4SI_type_node,
+ opaque_V4SI_type_node,
+ NULL_TREE);
+ }
else
{
- if (d->name == 0 || d->icode == CODE_FOR_nothing)
+ enum insn_code icode = d->icode;
+ if (d->name == 0 || icode == CODE_FOR_nothing)
continue;
- mode0 = insn_data[d->icode].operand[0].mode;
- mode1 = insn_data[d->icode].operand[1].mode;
- }
+ mode0 = insn_data[icode].operand[0].mode;
+ mode1 = insn_data[icode].operand[1].mode;
- if (mode0 == V4SImode && mode1 == QImode)
- type = v4si_ftype_int;
- else if (mode0 == V8HImode && mode1 == QImode)
- type = v8hi_ftype_int;
- else if (mode0 == V16QImode && mode1 == QImode)
- type = v16qi_ftype_int;
- else if (mode0 == VOIDmode && mode1 == VOIDmode)
- type = opaque_ftype_opaque;
- else if (mode0 == V4SFmode && mode1 == V4SFmode)
- type = v4sf_ftype_v4sf;
- else if (mode0 == V8HImode && mode1 == V16QImode)
- type = v8hi_ftype_v16qi;
- else if (mode0 == V4SImode && mode1 == V8HImode)
- type = v4si_ftype_v8hi;
- else if (mode0 == V2SImode && mode1 == V2SImode)
- type = v2si_ftype_v2si;
- else if (mode0 == V2SFmode && mode1 == V2SFmode)
- {
- if (TARGET_PAIRED_FLOAT)
- type = v2sf_ftype_v2sf;
- else
- type = v2sf_ftype_v2sf_spe;
- }
- else if (mode0 == V2SFmode && mode1 == V2SImode)
- type = v2sf_ftype_v2si;
- else if (mode0 == V2SImode && mode1 == V2SFmode)
- type = v2si_ftype_v2sf;
- else if (mode0 == V2SImode && mode1 == QImode)
- type = v2si_ftype_char;
- else
- gcc_unreachable ();
+ if (mode0 == V2SImode && mode1 == QImode)
+ {
+ if (! (type = v2si_ftype_qi))
+ type = v2si_ftype_qi
+ = build_function_type_list (opaque_V2SI_type_node,
+ char_type_node,
+ NULL_TREE);
+ }
+
+ else
+ type = builtin_function_type (mode0, mode1, VOIDmode, VOIDmode,
+ d->code, d->name);
+ }
def_builtin (d->mask, d->name, type, d->code);
}
@@ -11506,8 +12732,10 @@ rtx
rs6000_secondary_memory_needed_rtx (enum machine_mode mode)
{
static bool eliminated = false;
+ rtx ret;
+
if (mode != SDmode)
- return assign_stack_local (mode, GET_MODE_SIZE (mode), 0);
+ ret = assign_stack_local (mode, GET_MODE_SIZE (mode), 0);
else
{
rtx mem = cfun->machine->sdmode_stack_slot;
@@ -11519,8 +12747,20 @@ rs6000_secondary_memory_needed_rtx (enum
cfun->machine->sdmode_stack_slot = mem;
eliminated = true;
}
- return mem;
+ ret = mem;
}
+
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nrs6000_secondary_memory_needed_rtx, mode %s, rtx:\n",
+ GET_MODE_NAME (mode));
+ if (!ret)
+ fprintf (stderr, "\tNULL_RTX\n");
+ else
+ debug_rtx (ret);
+ }
+
+ return ret;
}
static tree
@@ -11555,6 +12795,445 @@ rs6000_check_sdmode (tree *tp, int *walk
return NULL_TREE;
}
+enum reload_reg_type {
+ GPR_REGISTER_TYPE,
+ VECTOR_REGISTER_TYPE,
+ OTHER_REGISTER_TYPE
+};
+
+static enum reload_reg_type
+rs6000_reload_register_type (enum reg_class rclass)
+{
+ switch (rclass)
+ {
+ case GENERAL_REGS:
+ case BASE_REGS:
+ return GPR_REGISTER_TYPE;
+
+ case FLOAT_REGS:
+ case ALTIVEC_REGS:
+ case VSX_REGS:
+ return VECTOR_REGISTER_TYPE;
+
+ default:
+ return OTHER_REGISTER_TYPE;
+ }
+}
+
+/* Inform reload about cases where moving X with a mode MODE to a register in
+ RCLASS requires an extra scratch or immediate register. Return the class
+ needed for the immediate register.
+
+ For VSX and Altivec, we may need a register to convert sp+offset into
+ reg+sp. */
+
+static enum reg_class
+rs6000_secondary_reload (bool in_p,
+ rtx x,
+ enum reg_class rclass,
+ enum machine_mode mode,
+ secondary_reload_info *sri)
+{
+ enum reg_class ret = ALL_REGS;
+ enum insn_code icode;
+ bool default_p = false;
+
+ sri->icode = CODE_FOR_nothing;
+
+ /* Convert vector loads and stores into gprs to use an additional base
+ register. */
+ icode = rs6000_vector_reload[mode][in_p != false];
+ if (icode != CODE_FOR_nothing)
+ {
+ ret = NO_REGS;
+ sri->icode = CODE_FOR_nothing;
+ sri->extra_cost = 0;
+
+ if (GET_CODE (x) == MEM)
+ {
+ rtx addr = XEXP (x, 0);
+
+ /* Loads to and stores from gprs can do reg+offset, and wouldn't need
+ an extra register in that case, but it would need an extra
+ register if the addressing is reg+reg or (reg+reg)&(-16). */
+ if (rclass == GENERAL_REGS || rclass == BASE_REGS)
+ {
+ if (!legitimate_indirect_address_p (addr, false)
+ && !rs6000_legitimate_offset_address_p (TImode, addr, false))
+ {
+ sri->icode = icode;
+ /* account for splitting the loads, and converting the
+ address from reg+reg to reg. */
+ sri->extra_cost = (((TARGET_64BIT) ? 3 : 5)
+ + ((GET_CODE (addr) == AND) ? 1 : 0));
+ }
+ }
+ /* Loads to and stores from vector registers can only do reg+reg
+ addressing. Altivec registers can also do (reg+reg)&(-16). */
+ else if (rclass == VSX_REGS || rclass == ALTIVEC_REGS
+ || rclass == FLOAT_REGS || rclass == NO_REGS)
+ {
+ if (!VECTOR_MEM_ALTIVEC_P (mode)
+ && GET_CODE (addr) == AND
+ && GET_CODE (XEXP (addr, 1)) == CONST_INT
+ && INTVAL (XEXP (addr, 1)) == -16
+ && (legitimate_indirect_address_p (XEXP (addr, 0), false)
+ || legitimate_indexed_address_p (XEXP (addr, 0), false)))
+ {
+ sri->icode = icode;
+ sri->extra_cost = ((GET_CODE (XEXP (addr, 0)) == PLUS)
+ ? 2 : 1);
+ }
+ else if (!legitimate_indirect_address_p (addr, false)
+ && (rclass == NO_REGS
+ || !legitimate_indexed_address_p (addr, false)))
+ {
+ sri->icode = icode;
+ sri->extra_cost = 1;
+ }
+ else
+ icode = CODE_FOR_nothing;
+ }
+ /* Any other loads, including to pseudo registers which haven't been
+ assigned to a register yet, default to require a scratch
+ register. */
+ else
+ {
+ sri->icode = icode;
+ sri->extra_cost = 2;
+ }
+ }
+ else if (REG_P (x))
+ {
+ int regno = true_regnum (x);
+
+ icode = CODE_FOR_nothing;
+ if (regno < 0 || regno >= FIRST_PSEUDO_REGISTER)
+ default_p = true;
+ else
+ {
+ enum reg_class xclass = REGNO_REG_CLASS (regno);
+ enum reload_reg_type rtype1 = rs6000_reload_register_type (rclass);
+ enum reload_reg_type rtype2 = rs6000_reload_register_type (xclass);
+
+ /* If memory is needed, use default_secondary_reload to create the
+ stack slot. */
+ if (rtype1 != rtype2 || rtype1 == OTHER_REGISTER_TYPE)
+ default_p = true;
+ else
+ ret = NO_REGS;
+ }
+ }
+ else
+ default_p = true;
+ }
+ else
+ default_p = true;
+
+ if (default_p)
+ ret = default_secondary_reload (in_p, x, rclass, mode, sri);
+
+ gcc_assert (ret != ALL_REGS);
+
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr,
+ "\nrs6000_secondary_reload, return %s, in_p = %s, rclass = %s, "
+ "mode = %s",
+ reg_class_names[ret],
+ in_p ? "true" : "false",
+ reg_class_names[rclass],
+ GET_MODE_NAME (mode));
+
+ if (default_p)
+ fprintf (stderr, ", default secondary reload");
+
+ if (sri->icode != CODE_FOR_nothing)
+ fprintf (stderr, ", reload func = %s, extra cost = %d\n",
+ insn_data[sri->icode].name, sri->extra_cost);
+ else
+ fprintf (stderr, "\n");
+
+ debug_rtx (x);
+ }
+
+ return ret;
+}
+
+/* Fixup reload addresses for Altivec or VSX loads/stores to change SP+offset
+ to SP+reg addressing. */
+
+void
+rs6000_secondary_reload_inner (rtx reg, rtx mem, rtx scratch, bool store_p)
+{
+ int regno = true_regnum (reg);
+ enum machine_mode mode = GET_MODE (reg);
+ enum reg_class rclass;
+ rtx addr;
+ rtx and_op2 = NULL_RTX;
+ rtx addr_op1;
+ rtx addr_op2;
+ rtx scratch_or_premodify = scratch;
+ rtx and_rtx;
+ rtx cc_clobber;
+
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nrs6000_secondary_reload_inner, type = %s\n",
+ store_p ? "store" : "load");
+ fprintf (stderr, "reg:\n");
+ debug_rtx (reg);
+ fprintf (stderr, "mem:\n");
+ debug_rtx (mem);
+ fprintf (stderr, "scratch:\n");
+ debug_rtx (scratch);
+ }
+
+ gcc_assert (regno >= 0 && regno < FIRST_PSEUDO_REGISTER);
+ gcc_assert (GET_CODE (mem) == MEM);
+ rclass = REGNO_REG_CLASS (regno);
+ addr = XEXP (mem, 0);
+
+ switch (rclass)
+ {
+ /* GPRs can handle reg + small constant, all other addresses need to use
+ the scratch register. */
+ case GENERAL_REGS:
+ case BASE_REGS:
+ if (GET_CODE (addr) == AND)
+ {
+ and_op2 = XEXP (addr, 1);
+ addr = XEXP (addr, 0);
+ }
+
+ if (GET_CODE (addr) == PRE_MODIFY)
+ {
+ scratch_or_premodify = XEXP (addr, 0);
+ gcc_assert (REG_P (scratch_or_premodify));
+ gcc_assert (GET_CODE (XEXP (addr, 1)) == PLUS);
+ addr = XEXP (addr, 1);
+ }
+
+ if (GET_CODE (addr) == PLUS
+ && (!rs6000_legitimate_offset_address_p (TImode, addr, false)
+ || and_op2 != NULL_RTX))
+ {
+ addr_op1 = XEXP (addr, 0);
+ addr_op2 = XEXP (addr, 1);
+ gcc_assert (legitimate_indirect_address_p (addr_op1, false));
+
+ if (!REG_P (addr_op2)
+ && (GET_CODE (addr_op2) != CONST_INT
+ || !satisfies_constraint_I (addr_op2)))
+ {
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr,
+ "\nMove plus addr to register %s, mode = %s: ",
+ rs6000_reg_names[REGNO (scratch)],
+ GET_MODE_NAME (mode));
+ debug_rtx (addr_op2);
+ }
+ rs6000_emit_move (scratch, addr_op2, Pmode);
+ addr_op2 = scratch;
+ }
+
+ emit_insn (gen_rtx_SET (VOIDmode,
+ scratch_or_premodify,
+ gen_rtx_PLUS (Pmode,
+ addr_op1,
+ addr_op2)));
+
+ addr = scratch_or_premodify;
+ scratch_or_premodify = scratch;
+ }
+ else if (!legitimate_indirect_address_p (addr, false)
+ && !rs6000_legitimate_offset_address_p (TImode, addr, false))
+ {
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nMove addr to register %s, mode = %s: ",
+ rs6000_reg_names[REGNO (scratch_or_premodify)],
+ GET_MODE_NAME (mode));
+ debug_rtx (addr);
+ }
+ rs6000_emit_move (scratch_or_premodify, addr, Pmode);
+ addr = scratch_or_premodify;
+ scratch_or_premodify = scratch;
+ }
+ break;
+
+ /* Float/Altivec registers can only handle reg+reg addressing. Move
+ other addresses into a scratch register. */
+ case FLOAT_REGS:
+ case VSX_REGS:
+ case ALTIVEC_REGS:
+
+ /* With float regs, we need to handle the AND ourselves, since we can't
+ use the Altivec instruction with an implicit AND -16. Allow scalar
+ loads to float registers to use reg+offset even if VSX. */
+ if (GET_CODE (addr) == AND
+ && (rclass != ALTIVEC_REGS || GET_MODE_SIZE (mode) != 16
+ || GET_CODE (XEXP (addr, 1)) != CONST_INT
+ || INTVAL (XEXP (addr, 1)) != -16
+ || !VECTOR_MEM_ALTIVEC_P (mode)))
+ {
+ and_op2 = XEXP (addr, 1);
+ addr = XEXP (addr, 0);
+ }
+
+ /* If we aren't using a VSX load, save the PRE_MODIFY register and use it
+ as the address later. */
+ if (GET_CODE (addr) == PRE_MODIFY
+ && (!VECTOR_MEM_VSX_P (mode)
+ || and_op2 != NULL_RTX
+ || !legitimate_indexed_address_p (XEXP (addr, 1), false)))
+ {
+ scratch_or_premodify = XEXP (addr, 0);
+ gcc_assert (legitimate_indirect_address_p (scratch_or_premodify,
+ false));
+ gcc_assert (GET_CODE (XEXP (addr, 1)) == PLUS);
+ addr = XEXP (addr, 1);
+ }
+
+ if (legitimate_indirect_address_p (addr, false) /* reg */
+ || legitimate_indexed_address_p (addr, false) /* reg+reg */
+ || GET_CODE (addr) == PRE_MODIFY /* VSX pre-modify */
+ || (GET_CODE (addr) == AND /* Altivec memory */
+ && GET_CODE (XEXP (addr, 1)) == CONST_INT
+ && INTVAL (XEXP (addr, 1)) == -16
+ && VECTOR_MEM_ALTIVEC_P (mode))
+ || (rclass == FLOAT_REGS /* legacy float mem */
+ && GET_MODE_SIZE (mode) == 8
+ && and_op2 == NULL_RTX
+ && scratch_or_premodify == scratch
+ && rs6000_legitimate_offset_address_p (mode, addr, false)))
+ ;
+
+ else if (GET_CODE (addr) == PLUS)
+ {
+ addr_op1 = XEXP (addr, 0);
+ addr_op2 = XEXP (addr, 1);
+ gcc_assert (REG_P (addr_op1));
+
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nMove plus addr to register %s, mode = %s: ",
+ rs6000_reg_names[REGNO (scratch)], GET_MODE_NAME (mode));
+ debug_rtx (addr_op2);
+ }
+ rs6000_emit_move (scratch, addr_op2, Pmode);
+ emit_insn (gen_rtx_SET (VOIDmode,
+ scratch_or_premodify,
+ gen_rtx_PLUS (Pmode,
+ addr_op1,
+ scratch)));
+ addr = scratch_or_premodify;
+ scratch_or_premodify = scratch;
+ }
+
+ else if (GET_CODE (addr) == SYMBOL_REF || GET_CODE (addr) == CONST
+ || GET_CODE (addr) == CONST_INT || REG_P (addr))
+ {
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nMove addr to register %s, mode = %s: ",
+ rs6000_reg_names[REGNO (scratch_or_premodify)],
+ GET_MODE_NAME (mode));
+ debug_rtx (addr);
+ }
+
+ rs6000_emit_move (scratch_or_premodify, addr, Pmode);
+ addr = scratch_or_premodify;
+ scratch_or_premodify = scratch;
+ }
+
+ else
+ gcc_unreachable ();
+
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+
+ /* If the original address involved a pre-modify that we couldn't use the VSX
+ memory instruction with update, and we haven't taken care of already,
+ store the address in the pre-modify register and use that as the
+ address. */
+ if (scratch_or_premodify != scratch && scratch_or_premodify != addr)
+ {
+ emit_insn (gen_rtx_SET (VOIDmode, scratch_or_premodify, addr));
+ addr = scratch_or_premodify;
+ }
+
+ /* If the original address involved an AND -16 and we couldn't use an ALTIVEC
+ memory instruction, recreate the AND now, including the clobber which is
+ generated by the general ANDSI3/ANDDI3 patterns for the
+ andi. instruction. */
+ if (and_op2 != NULL_RTX)
+ {
+ if (! legitimate_indirect_address_p (addr, false))
+ {
+ emit_insn (gen_rtx_SET (VOIDmode, scratch, addr));
+ addr = scratch;
+ }
+
+ if (TARGET_DEBUG_ADDR)
+ {
+ fprintf (stderr, "\nAnd addr to register %s, mode = %s: ",
+ rs6000_reg_names[REGNO (scratch)], GET_MODE_NAME (mode));
+ debug_rtx (and_op2);
+ }
+
+ and_rtx = gen_rtx_SET (VOIDmode,
+ scratch,
+ gen_rtx_AND (Pmode,
+ addr,
+ and_op2));
+
+ cc_clobber = gen_rtx_CLOBBER (CCmode, gen_rtx_SCRATCH (CCmode));
+ emit_insn (gen_rtx_PARALLEL (VOIDmode,
+ gen_rtvec (2, and_rtx, cc_clobber)));
+ addr = scratch;
+ }
+
+ /* Adjust the address if it changed. */
+ if (addr != XEXP (mem, 0))
+ {
+ mem = change_address (mem, mode, addr);
+ if (TARGET_DEBUG_ADDR)
+ fprintf (stderr, "\nrs6000_secondary_reload_inner, mem adjusted.\n");
+ }
+
+ /* Now create the move. */
+ if (store_p)
+ emit_insn (gen_rtx_SET (VOIDmode, mem, reg));
+ else
+ emit_insn (gen_rtx_SET (VOIDmode, reg, mem));
+
+ return;
+}
+
+/* Target hook to return the cover classes for Integrated Register Allocator.
+ Cover classes is a set of non-intersected register classes covering all hard
+ registers used for register allocation purpose. Any move between two
+ registers of a cover class should be cheaper than load or store of the
+ registers. The value is array of register classes with LIM_REG_CLASSES used
+ as the end marker.
+
+ We need two IRA_COVER_CLASSES, one for pre-VSX, and the other for VSX to
+ account for the Altivec and Floating registers being subsets of the VSX
+ register set under VSX, but distinct register sets on pre-VSX machines. */
+
+static const enum reg_class *
+rs6000_ira_cover_classes (void)
+{
+ static const enum reg_class cover_pre_vsx[] = IRA_COVER_CLASSES_PRE_VSX;
+ static const enum reg_class cover_vsx[] = IRA_COVER_CLASSES_VSX;
+
+ return (TARGET_VSX) ? cover_vsx : cover_pre_vsx;
+}
/* Allocate a 64-bit stack slot to be used for copying SDmode
values through if this function has any SDmode references. */
@@ -11605,13 +13284,163 @@ rs6000_instantiate_decls (void)
instantiate_decl_rtl (cfun->machine->sdmode_stack_slot);
}
+/* Given an rtx X being reloaded into a reg required to be
+ in class CLASS, return the class of reg to actually use.
+ In general this is just CLASS; but on some machines
+ in some cases it is preferable to use a more restrictive class.
+
+ On the RS/6000, we have to return NO_REGS when we want to reload a
+ floating-point CONST_DOUBLE to force it to be copied to memory.
+
+ We also don't want to reload integer values into floating-point
+ registers if we can at all help it. In fact, this can
+ cause reload to die, if it tries to generate a reload of CTR
+ into a FP register and discovers it doesn't have the memory location
+ required.
+
+ ??? Would it be a good idea to have reload do the converse, that is
+ try to reload floating modes into FP registers if possible?
+ */
+
+static enum reg_class
+rs6000_preferred_reload_class (rtx x, enum reg_class rclass)
+{
+ enum machine_mode mode = GET_MODE (x);
+
+ if (VECTOR_UNIT_VSX_P (mode)
+ && x == CONST0_RTX (mode) && VSX_REG_CLASS_P (rclass))
+ return rclass;
+
+ if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (mode)
+ && (rclass == ALTIVEC_REGS || rclass == VSX_REGS)
+ && easy_vector_constant (x, mode))
+ return ALTIVEC_REGS;
+
+ if (CONSTANT_P (x) && reg_classes_intersect_p (rclass, FLOAT_REGS))
+ return NO_REGS;
+
+ if (GET_MODE_CLASS (mode) == MODE_INT && rclass == NON_SPECIAL_REGS)
+ return GENERAL_REGS;
+
+ /* For VSX, prefer the traditional registers for DF if the address is of the
+ form reg+offset because we can use the non-VSX loads. Prefer the Altivec
+ registers if Altivec is handling the vector operations (i.e. V16QI, V8HI,
+ and V4SI). */
+ if (rclass == VSX_REGS && VECTOR_MEM_VSX_P (mode))
+ {
+ if (mode == DFmode && GET_CODE (x) == MEM)
+ {
+ rtx addr = XEXP (x, 0);
+
+ if (legitimate_indirect_address_p (addr, false)) /* reg */
+ return VSX_REGS;
+
+ if (legitimate_indexed_address_p (addr, false)) /* reg+reg */
+ return VSX_REGS;
+
+ if (GET_CODE (addr) == PRE_MODIFY
+ && legitimate_indexed_address_p (XEXP (addr, 0), false))
+ return VSX_REGS;
+
+ return FLOAT_REGS;
+ }
+
+ if (VECTOR_UNIT_ALTIVEC_P (mode))
+ return ALTIVEC_REGS;
+
+ return rclass;
+ }
+
+ return rclass;
+}
+
+/* Debug version of rs6000_preferred_reload_class. */
+static enum reg_class
+rs6000_debug_preferred_reload_class (rtx x, enum reg_class rclass)
+{
+ enum reg_class ret = rs6000_preferred_reload_class (x, rclass);
+
+ fprintf (stderr,
+ "\nrs6000_preferred_reload_class, return %s, rclass = %s, "
+ "mode = %s, x:\n",
+ reg_class_names[ret], reg_class_names[rclass],
+ GET_MODE_NAME (GET_MODE (x)));
+ debug_rtx (x);
+
+ return ret;
+}
+
+/* If we are copying between FP or AltiVec registers and anything else, we need
+ a memory location. The exception is when we are targeting ppc64 and the
+ move to/from fpr to gpr instructions are available. Also, under VSX, you
+ can copy vector registers from the FP register set to the Altivec register
+ set and vice versa. */
+
+static bool
+rs6000_secondary_memory_needed (enum reg_class class1,
+ enum reg_class class2,
+ enum machine_mode mode)
+{
+ if (class1 == class2)
+ return false;
+
+ /* Under VSX, there are 3 register classes that values could be in (VSX_REGS,
+ ALTIVEC_REGS, and FLOAT_REGS). We don't need to use memory to copy
+ between these classes. But we need memory for other things that can go in
+ FLOAT_REGS like SFmode. */
+ if (TARGET_VSX
+ && (VECTOR_MEM_VSX_P (mode) || VECTOR_UNIT_VSX_P (mode))
+ && (class1 == VSX_REGS || class1 == ALTIVEC_REGS
+ || class1 == FLOAT_REGS))
+ return (class2 != VSX_REGS && class2 != ALTIVEC_REGS
+ && class2 != FLOAT_REGS);
+
+ if (class1 == VSX_REGS || class2 == VSX_REGS)
+ return true;
+
+ if (class1 == FLOAT_REGS
+ && (!TARGET_MFPGPR || !TARGET_POWERPC64
+ || ((mode != DFmode)
+ && (mode != DDmode)
+ && (mode != DImode))))
+ return true;
+
+ if (class2 == FLOAT_REGS
+ && (!TARGET_MFPGPR || !TARGET_POWERPC64
+ || ((mode != DFmode)
+ && (mode != DDmode)
+ && (mode != DImode))))
+ return true;
+
+ if (class1 == ALTIVEC_REGS || class2 == ALTIVEC_REGS)
+ return true;
+
+ return false;
+}
+
+/* Debug version of rs6000_secondary_memory_needed. */
+static bool
+rs6000_debug_secondary_memory_needed (enum reg_class class1,
+ enum reg_class class2,
+ enum machine_mode mode)
+{
+ bool ret = rs6000_secondary_memory_needed (class1, class2, mode);
+
+ fprintf (stderr,
+ "rs6000_secondary_memory_needed, return: %s, class1 = %s, "
+ "class2 = %s, mode = %s\n",
+ ret ? "true" : "false", reg_class_names[class1],
+ reg_class_names[class2], GET_MODE_NAME (mode));
+
+ return ret;
+}
+
/* Return the register class of a scratch register needed to copy IN into
or out of a register in RCLASS in MODE. If it can be done directly,
NO_REGS is returned. */
-enum reg_class
-rs6000_secondary_reload_class (enum reg_class rclass,
- enum machine_mode mode ATTRIBUTE_UNUSED,
+static enum reg_class
+rs6000_secondary_reload_class (enum reg_class rclass, enum machine_mode mode,
rtx in)
{
int regno;
@@ -11667,6 +13496,13 @@ rs6000_secondary_reload_class (enum reg_
&& (rclass == FLOAT_REGS || rclass == NON_SPECIAL_REGS))
return (mode != SDmode) ? NO_REGS : GENERAL_REGS;
+ /* Memory, and FP/altivec registers can go into fp/altivec registers under
+ VSX. */
+ if (TARGET_VSX
+ && (regno == -1 || VSX_REGNO_P (regno))
+ && VSX_REG_CLASS_P (rclass))
+ return NO_REGS;
+
/* Memory, and AltiVec registers can go into AltiVec registers. */
if ((regno == -1 || ALTIVEC_REGNO_P (regno))
&& rclass == ALTIVEC_REGS)
@@ -11680,6 +13516,83 @@ rs6000_secondary_reload_class (enum reg_
/* Otherwise, we need GENERAL_REGS. */
return GENERAL_REGS;
}
+
+/* Debug version of rs6000_secondary_reload_class. */
+static enum reg_class
+rs6000_debug_secondary_reload_class (enum reg_class rclass,
+ enum machine_mode mode, rtx in)
+{
+ enum reg_class ret = rs6000_secondary_reload_class (rclass, mode, in);
+ fprintf (stderr,
+ "\nrs6000_secondary_reload_class, return %s, rclass = %s, "
+ "mode = %s, input rtx:\n",
+ reg_class_names[ret], reg_class_names[rclass],
+ GET_MODE_NAME (mode));
+ debug_rtx (in);
+
+ return ret;
+}
+
+/* Return nonzero if for CLASS a mode change from FROM to TO is invalid. */
+
+static bool
+rs6000_cannot_change_mode_class (enum machine_mode from,
+ enum machine_mode to,
+ enum reg_class rclass)
+{
+ unsigned from_size = GET_MODE_SIZE (from);
+ unsigned to_size = GET_MODE_SIZE (to);
+
+ if (from_size != to_size)
+ {
+ enum reg_class xclass = (TARGET_VSX) ? VSX_REGS : FLOAT_REGS;
+ return ((from_size < 8 || to_size < 8 || TARGET_IEEEQUAD)
+ && reg_classes_intersect_p (xclass, rclass));
+ }
+
+ if (TARGET_E500_DOUBLE
+ && ((((to) == DFmode) + ((from) == DFmode)) == 1
+ || (((to) == TFmode) + ((from) == TFmode)) == 1
+ || (((to) == DDmode) + ((from) == DDmode)) == 1
+ || (((to) == TDmode) + ((from) == TDmode)) == 1
+ || (((to) == DImode) + ((from) == DImode)) == 1))
+ return true;
+
+ /* Since the VSX register set includes traditional floating point registers
+ and altivec registers, just check for the size being different instead of
+ trying to check whether the modes are vector modes. Otherwise it won't
+ allow say DF and DI to change classes. */
+ if (TARGET_VSX && VSX_REG_CLASS_P (rclass))
+ return (from_size != 8 && from_size != 16);
+
+ if (TARGET_ALTIVEC && rclass == ALTIVEC_REGS
+ && (ALTIVEC_VECTOR_MODE (from) + ALTIVEC_VECTOR_MODE (to)) == 1)
+ return true;
+
+ if (TARGET_SPE && (SPE_VECTOR_MODE (from) + SPE_VECTOR_MODE (to)) == 1
+ && reg_classes_intersect_p (GENERAL_REGS, rclass))
+ return true;
+
+ return false;
+}
+
+/* Debug version of rs6000_cannot_change_mode_class. */
+static bool
+rs6000_debug_cannot_change_mode_class (enum machine_mode from,
+ enum machine_mode to,
+ enum reg_class rclass)
+{
+ bool ret = rs6000_cannot_change_mode_class (from, to, rclass);
+
+ fprintf (stderr,
+ "rs6000_cannot_change_mode_class, return %s, from = %s, "
+ "to = %s, rclass = %s\n",
+ ret ? "true" : "false",
+ GET_MODE_NAME (from), GET_MODE_NAME (to),
+ reg_class_names[rclass]);
+
+ return ret;
+}
/* Given a comparison operation, return the bit number in CCR to test. We
know this is a valid comparison.
@@ -11973,7 +13886,7 @@ print_operand (FILE *file, rtx x, int co
case 'c':
/* X is a CR register. Print the number of the GT bit of the CR. */
if (GET_CODE (x) != REG || ! CR_REGNO_P (REGNO (x)))
- output_operand_lossage ("invalid %%E value");
+ output_operand_lossage ("invalid %%c value");
else
fprintf (file, "%d", 4 * (REGNO (x) - CR0_REGNO) + 1);
return;
@@ -12410,6 +14323,26 @@ print_operand (FILE *file, rtx x, int co
fprintf (file, "%d", i + 1);
return;
+ case 'x':
+ /* X is a FPR or Altivec register used in a VSX context. */
+ if (GET_CODE (x) != REG || !VSX_REGNO_P (REGNO (x)))
+ output_operand_lossage ("invalid %%x value");
+ else
+ {
+ int reg = REGNO (x);
+ int vsx_reg = (FP_REGNO_P (reg)
+ ? reg - 32
+ : reg - FIRST_ALTIVEC_REGNO + 32);
+
+#ifdef TARGET_REGNAMES
+ if (TARGET_REGNAMES)
+ fprintf (file, "%%vs%d", vsx_reg);
+ else
+#endif
+ fprintf (file, "%d", vsx_reg);
+ }
+ return;
+
case 'X':
if (GET_CODE (x) == MEM
&& (legitimate_indexed_address_p (XEXP (x, 0), 0)
@@ -12522,13 +14455,16 @@ print_operand (FILE *file, rtx x, int co
/* Fall through. Must be [reg+reg]. */
}
- if (TARGET_ALTIVEC
+ if (VECTOR_MEM_ALTIVEC_P (GET_MODE (x))
&& GET_CODE (tmp) == AND
&& GET_CODE (XEXP (tmp, 1)) == CONST_INT
&& INTVAL (XEXP (tmp, 1)) == -16)
tmp = XEXP (tmp, 0);
+ else if (VECTOR_MEM_VSX_P (GET_MODE (x))
+ && GET_CODE (tmp) == PRE_MODIFY)
+ tmp = XEXP (tmp, 1);
if (GET_CODE (tmp) == REG)
- fprintf (file, "0,%s", reg_names[REGNO (tmp)]);
+ fprintf (file, "%s,%s", reg_names[0], reg_names[REGNO (tmp)]);
else
{
if (!GET_CODE (tmp) == PLUS
@@ -13274,55 +15210,34 @@ output_e500_flip_gt_bit (rtx dst, rtx sr
return string;
}
-/* Return insn index for the vector compare instruction for given CODE,
- and DEST_MODE, OP_MODE. Return INSN_NOT_AVAILABLE if valid insn is
- not available. */
+/* Return insn for VSX or Altivec comparisons. */
-static int
-get_vec_cmp_insn (enum rtx_code code,
- enum machine_mode dest_mode,
- enum machine_mode op_mode)
+static rtx
+rs6000_emit_vector_compare_inner (enum rtx_code code, rtx op0, rtx op1)
{
- if (!TARGET_ALTIVEC)
- return INSN_NOT_AVAILABLE;
+ rtx mask;
+ enum machine_mode mode = GET_MODE (op0);
switch (code)
{
- case EQ:
- if (dest_mode == V16QImode && op_mode == V16QImode)
- return UNSPEC_VCMPEQUB;
- if (dest_mode == V8HImode && op_mode == V8HImode)
- return UNSPEC_VCMPEQUH;
- if (dest_mode == V4SImode && op_mode == V4SImode)
- return UNSPEC_VCMPEQUW;
- if (dest_mode == V4SImode && op_mode == V4SFmode)
- return UNSPEC_VCMPEQFP;
+ default:
break;
+
case GE:
- if (dest_mode == V4SImode && op_mode == V4SFmode)
- return UNSPEC_VCMPGEFP;
+ if (GET_MODE_CLASS (mode) == MODE_VECTOR_INT)
+ return NULL_RTX;
+
+ case EQ:
case GT:
- if (dest_mode == V16QImode && op_mode == V16QImode)
- return UNSPEC_VCMPGTSB;
- if (dest_mode == V8HImode && op_mode == V8HImode)
- return UNSPEC_VCMPGTSH;
- if (dest_mode == V4SImode && op_mode == V4SImode)
- return UNSPEC_VCMPGTSW;
- if (dest_mode == V4SImode && op_mode == V4SFmode)
- return UNSPEC_VCMPGTFP;
- break;
case GTU:
- if (dest_mode == V16QImode && op_mode == V16QImode)
- return UNSPEC_VCMPGTUB;
- if (dest_mode == V8HImode && op_mode == V8HImode)
- return UNSPEC_VCMPGTUH;
- if (dest_mode == V4SImode && op_mode == V4SImode)
- return UNSPEC_VCMPGTUW;
- break;
- default:
- break;
+ mask = gen_reg_rtx (mode);
+ emit_insn (gen_rtx_SET (VOIDmode,
+ mask,
+ gen_rtx_fmt_ee (code, mode, op0, op1)));
+ return mask;
}
- return INSN_NOT_AVAILABLE;
+
+ return NULL_RTX;
}
/* Emit vector compare for operands OP0 and OP1 using code RCODE.
@@ -13333,129 +15248,114 @@ rs6000_emit_vector_compare (enum rtx_cod
rtx op0, rtx op1,
enum machine_mode dmode)
{
- int vec_cmp_insn;
rtx mask;
- enum machine_mode dest_mode;
- enum machine_mode op_mode = GET_MODE (op1);
+ bool swap_operands = false;
+ bool try_again = false;
- gcc_assert (TARGET_ALTIVEC);
+ gcc_assert (VECTOR_UNIT_ALTIVEC_OR_VSX_P (dmode));
gcc_assert (GET_MODE (op0) == GET_MODE (op1));
- /* Floating point vector compare instructions uses destination V4SImode.
- Move destination to appropriate mode later. */
- if (dmode == V4SFmode)
- dest_mode = V4SImode;
- else
- dest_mode = dmode;
-
- mask = gen_reg_rtx (dest_mode);
- vec_cmp_insn = get_vec_cmp_insn (rcode, dest_mode, op_mode);
+ /* See if the comparison works as is. */
+ mask = rs6000_emit_vector_compare_inner (rcode, op0, op1);
+ if (mask)
+ return mask;
- if (vec_cmp_insn == INSN_NOT_AVAILABLE)
+ switch (rcode)
{
- bool swap_operands = false;
- bool try_again = false;
- switch (rcode)
- {
- case LT:
- rcode = GT;
- swap_operands = true;
- try_again = true;
- break;
- case LTU:
- rcode = GTU;
- swap_operands = true;
- try_again = true;
- break;
- case NE:
- case UNLE:
- case UNLT:
- case UNGE:
- case UNGT:
- /* Invert condition and try again.
- e.g., A != B becomes ~(A==B). */
- {
- enum rtx_code rev_code;
- enum insn_code nor_code;
- rtx eq_rtx;
-
- rev_code = reverse_condition_maybe_unordered (rcode);
- eq_rtx = rs6000_emit_vector_compare (rev_code, op0, op1,
- dest_mode);
-
- nor_code = optab_handler (one_cmpl_optab, (int)dest_mode)->insn_code;
- gcc_assert (nor_code != CODE_FOR_nothing);
- emit_insn (GEN_FCN (nor_code) (mask, eq_rtx));
+ case LT:
+ rcode = GT;
+ swap_operands = true;
+ try_again = true;
+ break;
+ case LTU:
+ rcode = GTU;
+ swap_operands = true;
+ try_again = true;
+ break;
+ case NE:
+ case UNLE:
+ case UNLT:
+ case UNGE:
+ case UNGT:
+ /* Invert condition and try again.
+ e.g., A != B becomes ~(A==B). */
+ {
+ enum rtx_code rev_code;
+ enum insn_code nor_code;
+ rtx mask2;
+
+ rev_code = reverse_condition_maybe_unordered (rcode);
+ if (rev_code == UNKNOWN)
+ return NULL_RTX;
+
+ nor_code = optab_handler (one_cmpl_optab, (int)dmode)->insn_code;
+ if (nor_code == CODE_FOR_nothing)
+ return NULL_RTX;
+
+ mask2 = rs6000_emit_vector_compare (rev_code, op0, op1, dmode);
+ if (!mask2)
+ return NULL_RTX;
+
+ mask = gen_reg_rtx (dmode);
+ emit_insn (GEN_FCN (nor_code) (mask, mask2));
+ return mask;
+ }
+ break;
+ case GE:
+ case GEU:
+ case LE:
+ case LEU:
+ /* Try GT/GTU/LT/LTU OR EQ */
+ {
+ rtx c_rtx, eq_rtx;
+ enum insn_code ior_code;
+ enum rtx_code new_code;
- if (dmode != dest_mode)
- {
- rtx temp = gen_reg_rtx (dest_mode);
- convert_move (temp, mask, 0);
- return temp;
- }
- return mask;
- }
- break;
- case GE:
- case GEU:
- case LE:
- case LEU:
- /* Try GT/GTU/LT/LTU OR EQ */
+ switch (rcode)
{
- rtx c_rtx, eq_rtx;
- enum insn_code ior_code;
- enum rtx_code new_code;
-
- switch (rcode)
- {
- case GE:
- new_code = GT;
- break;
-
- case GEU:
- new_code = GTU;
- break;
+ case GE:
+ new_code = GT;
+ break;
- case LE:
- new_code = LT;
- break;
+ case GEU:
+ new_code = GTU;
+ break;
- case LEU:
- new_code = LTU;
- break;
+ case LE:
+ new_code = LT;
+ break;
- default:
- gcc_unreachable ();
- }
+ case LEU:
+ new_code = LTU;
+ break;
- c_rtx = rs6000_emit_vector_compare (new_code,
- op0, op1, dest_mode);
- eq_rtx = rs6000_emit_vector_compare (EQ, op0, op1,
- dest_mode);
-
- ior_code = optab_handler (ior_optab, (int)dest_mode)->insn_code;
- gcc_assert (ior_code != CODE_FOR_nothing);
- emit_insn (GEN_FCN (ior_code) (mask, c_rtx, eq_rtx));
- if (dmode != dest_mode)
- {
- rtx temp = gen_reg_rtx (dest_mode);
- convert_move (temp, mask, 0);
- return temp;
- }
- return mask;
+ default:
+ gcc_unreachable ();
}
- break;
- default:
- gcc_unreachable ();
- }
- if (try_again)
- {
- vec_cmp_insn = get_vec_cmp_insn (rcode, dest_mode, op_mode);
- /* You only get two chances. */
- gcc_assert (vec_cmp_insn != INSN_NOT_AVAILABLE);
- }
+ ior_code = optab_handler (ior_optab, (int)dmode)->insn_code;
+ if (ior_code == CODE_FOR_nothing)
+ return NULL_RTX;
+
+ c_rtx = rs6000_emit_vector_compare (new_code, op0, op1, dmode);
+ if (!c_rtx)
+ return NULL_RTX;
+
+ eq_rtx = rs6000_emit_vector_compare (EQ, op0, op1, dmode);
+ if (!eq_rtx)
+ return NULL_RTX;
+
+ mask = gen_reg_rtx (dmode);
+ emit_insn (GEN_FCN (ior_code) (mask, c_rtx, eq_rtx));
+ return mask;
+ }
+ break;
+ default:
+ return NULL_RTX;
+ }
+ if (try_again)
+ {
if (swap_operands)
{
rtx tmp;
@@ -13463,91 +15363,84 @@ rs6000_emit_vector_compare (enum rtx_cod
op0 = op1;
op1 = tmp;
}
- }
-
- emit_insn (gen_rtx_SET (VOIDmode, mask,
- gen_rtx_UNSPEC (dest_mode,
- gen_rtvec (2, op0, op1),
- vec_cmp_insn)));
- if (dmode != dest_mode)
- {
- rtx temp = gen_reg_rtx (dest_mode);
- convert_move (temp, mask, 0);
- return temp;
- }
- return mask;
-}
-
-/* Return vector select instruction for MODE. Return INSN_NOT_AVAILABLE, if
- valid insn doesn exist for given mode. */
-static int
-get_vsel_insn (enum machine_mode mode)
-{
- switch (mode)
- {
- case V4SImode:
- return UNSPEC_VSEL4SI;
- break;
- case V4SFmode:
- return UNSPEC_VSEL4SF;
- break;
- case V8HImode:
- return UNSPEC_VSEL8HI;
- break;
- case V16QImode:
- return UNSPEC_VSEL16QI;
- break;
- default:
- return INSN_NOT_AVAILABLE;
- break;
+ mask = rs6000_emit_vector_compare_inner (rcode, op0, op1);
+ if (mask)
+ return mask;
}
- return INSN_NOT_AVAILABLE;
-}
-
-/* Emit vector select insn where DEST is destination using
- operands OP1, OP2 and MASK. */
-
-static void
-rs6000_emit_vector_select (rtx dest, rtx op1, rtx op2, rtx mask)
-{
- rtx t, temp;
- enum machine_mode dest_mode = GET_MODE (dest);
- int vsel_insn_index = get_vsel_insn (GET_MODE (dest));
- temp = gen_reg_rtx (dest_mode);
-
- /* For each vector element, select op1 when mask is 1 otherwise
- select op2. */
- t = gen_rtx_SET (VOIDmode, temp,
- gen_rtx_UNSPEC (dest_mode,
- gen_rtvec (3, op2, op1, mask),
- vsel_insn_index));
- emit_insn (t);
- emit_move_insn (dest, temp);
- return;
+ /* You only get two chances. */
+ return NULL_RTX;
}
-/* Emit vector conditional expression.
- DEST is destination. OP1 and OP2 are two VEC_COND_EXPR operands.
- CC_OP0 and CC_OP1 are the two operands for the relation operation COND. */
+/* Emit vector conditional expression. DEST is destination. OP_TRUE and
+ OP_FALSE are two VEC_COND_EXPR operands. CC_OP0 and CC_OP1 are the two
+ operands for the relation operation COND. */
int
-rs6000_emit_vector_cond_expr (rtx dest, rtx op1, rtx op2,
+rs6000_emit_vector_cond_expr (rtx dest, rtx op_true, rtx op_false,
rtx cond, rtx cc_op0, rtx cc_op1)
{
enum machine_mode dest_mode = GET_MODE (dest);
enum rtx_code rcode = GET_CODE (cond);
+ enum machine_mode cc_mode = CCmode;
rtx mask;
+ rtx cond2;
+ rtx tmp;
+ bool invert_move = false;
- if (!TARGET_ALTIVEC)
+ if (VECTOR_UNIT_NONE_P (dest_mode))
return 0;
+ switch (rcode)
+ {
+ /* Swap operands if we can, and fall back to doing the operation as
+ specified, and doing a NOR to invert the test. */
+ case NE:
+ case UNLE:
+ case UNLT:
+ case UNGE:
+ case UNGT:
+ /* Invert condition and try again.
+ e.g., A = (B != C) ? D : E becomes A = (B == C) ? E : D. */
+ invert_move = true;
+ rcode = reverse_condition_maybe_unordered (rcode);
+ if (rcode == UNKNOWN)
+ return 0;
+ break;
+
+ /* Mark unsigned tests with CCUNSmode. */
+ case GTU:
+ case GEU:
+ case LTU:
+ case LEU:
+ cc_mode = CCUNSmode;
+ break;
+
+ default:
+ break;
+ }
+
/* Get the vector mask for the given relational operations. */
mask = rs6000_emit_vector_compare (rcode, cc_op0, cc_op1, dest_mode);
- rs6000_emit_vector_select (dest, op1, op2, mask);
+ if (!mask)
+ return 0;
+
+ if (invert_move)
+ {
+ tmp = op_true;
+ op_true = op_false;
+ op_false = tmp;
+ }
+ cond2 = gen_rtx_fmt_ee (NE, cc_mode, mask, const0_rtx);
+ emit_insn (gen_rtx_SET (VOIDmode,
+ dest,
+ gen_rtx_IF_THEN_ELSE (dest_mode,
+ cond2,
+ op_true,
+ op_false)));
return 1;
}
@@ -13743,9 +15636,9 @@ static int
rs6000_emit_int_cmove (rtx dest, rtx op, rtx true_cond, rtx false_cond)
{
rtx condition_rtx, cr;
+ enum machine_mode mode = GET_MODE (XEXP (op, 0));
- /* All isel implementations thus far are 32-bits. */
- if (GET_MODE (XEXP (op, 0)) != SImode)
+ if (mode != SImode && (!TARGET_POWERPC64 || mode != DImode))
return 0;
/* We still have to do the compare, because isel doesn't do a
@@ -13754,12 +15647,24 @@ rs6000_emit_int_cmove (rtx dest, rtx op,
condition_rtx = rs6000_generate_compare (op, SImode);
cr = XEXP (condition_rtx, 0);
- if (GET_MODE (cr) == CCmode)
- emit_insn (gen_isel_signed (dest, condition_rtx,
- true_cond, false_cond, cr));
+ if (mode == SImode)
+ {
+ if (GET_MODE (cr) == CCmode)
+ emit_insn (gen_isel_signed_si (dest, condition_rtx,
+ true_cond, false_cond, cr));
+ else
+ emit_insn (gen_isel_unsigned_si (dest, condition_rtx,
+ true_cond, false_cond, cr));
+ }
else
- emit_insn (gen_isel_unsigned (dest, condition_rtx,
- true_cond, false_cond, cr));
+ {
+ if (GET_MODE (cr) == CCmode)
+ emit_insn (gen_isel_signed_di (dest, condition_rtx,
+ true_cond, false_cond, cr));
+ else
+ emit_insn (gen_isel_unsigned_di (dest, condition_rtx,
+ true_cond, false_cond, cr));
+ }
return 1;
}
@@ -13786,6 +15691,15 @@ rs6000_emit_minmax (rtx dest, enum rtx_c
enum rtx_code c;
rtx target;
+ /* VSX/altivec have direct min/max insns. */
+ if ((code == SMAX || code == SMIN) && VECTOR_UNIT_ALTIVEC_OR_VSX_P (mode))
+ {
+ emit_insn (gen_rtx_SET (VOIDmode,
+ dest,
+ gen_rtx_fmt_ee (code, mode, op0, op1)));
+ return;
+ }
+
if (code == SMAX || code == SMIN)
c = GE;
else
@@ -15451,6 +17365,19 @@ uses_TOC (void)
rtx
create_TOC_reference (rtx symbol)
{
+ if (TARGET_DEBUG_ADDR)
+ {
+ if (GET_CODE (symbol) == SYMBOL_REF)
+ fprintf (stderr, "\ncreate_TOC_reference, (symbol_ref %s)\n",
+ XSTR (symbol, 0));
+ else
+ {
+ fprintf (stderr, "\ncreate_TOC_reference, code %s:\n",
+ GET_RTX_NAME (GET_CODE (symbol)));
+ debug_rtx (symbol);
+ }
+ }
+
if (!can_create_pseudo_p ())
df_set_regs_ever_live (TOC_REGISTER, true);
return gen_rtx_PLUS (Pmode,
@@ -15759,6 +17686,7 @@ emit_frame_save (rtx frame_reg, rtx fram
/* Some cases that need register indexed addressing. */
if ((TARGET_ALTIVEC_ABI && ALTIVEC_VECTOR_MODE (mode))
+ || (TARGET_VSX && VSX_VECTOR_MODE (mode))
|| (TARGET_E500_DOUBLE && mode == DFmode)
|| (TARGET_SPE_ABI
&& SPE_VECTOR_MODE (mode)
@@ -18414,7 +20342,8 @@ output_toc (FILE *file, rtx x, int label
if (GET_CODE (x) == CONST)
{
- gcc_assert (GET_CODE (XEXP (x, 0)) == PLUS);
+ gcc_assert (GET_CODE (XEXP (x, 0)) == PLUS
+ && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT);
base = XEXP (XEXP (x, 0), 0);
offset = INTVAL (XEXP (XEXP (x, 0), 1));
@@ -18866,6 +20795,7 @@ rs6000_adjust_cost (rtx insn, rtx link,
|| rs6000_cpu_attr == CPU_PPC7450
|| rs6000_cpu_attr == CPU_POWER4
|| rs6000_cpu_attr == CPU_POWER5
+ || (rs6000_cpu_attr == CPU_POWER7 && TARGET_POWER7_ADJUST_COST)
|| rs6000_cpu_attr == CPU_CELL)
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0))
@@ -19105,6 +21035,35 @@ rs6000_adjust_cost (rtx insn, rtx link,
return cost;
}
+/* Debug version of rs6000_adjust_cost. */
+
+static int
+rs6000_debug_adjust_cost (rtx insn, rtx link, rtx dep_insn, int cost)
+{
+ int ret = rs6000_adjust_cost (insn, link, dep_insn, cost);
+
+ if (ret != cost)
+ {
+ const char *dep;
+
+ switch (REG_NOTE_KIND (link))
+ {
+ default: dep = "unknown depencency"; break;
+ case REG_DEP_TRUE: dep = "data dependency"; break;
+ case REG_DEP_OUTPUT: dep = "output dependency"; break;
+ case REG_DEP_ANTI: dep = "anti depencency"; break;
+ }
+
+ fprintf (stderr,
+ "\nrs6000_adjust_cost, final cost = %d, orig cost = %d, "
+ "%s, insn:\n", ret, cost, dep);
+
+ debug_rtx (insn);
+ }
+
+ return ret;
+}
+
/* The function returns a true if INSN is microcoded.
Return false otherwise. */
@@ -19386,6 +21345,7 @@ rs6000_issue_rate (void)
case CPU_POWER4:
case CPU_POWER5:
case CPU_POWER6:
+ case CPU_POWER7:
return 5;
default:
return 1;
@@ -19987,6 +21947,41 @@ insn_must_be_first_in_group (rtx insn)
break;
}
break;
+ case PROCESSOR_POWER7:
+ type = get_attr_type (insn);
+
+ switch (type)
+ {
+ case TYPE_CR_LOGICAL:
+ case TYPE_MFCR:
+ case TYPE_MFCRF:
+ case TYPE_MTCR:
+ case TYPE_IDIV:
+ case TYPE_LDIV:
+ case TYPE_COMPARE:
+ case TYPE_DELAYED_COMPARE:
+ case TYPE_VAR_DELAYED_COMPARE:
+ case TYPE_ISYNC:
+ case TYPE_LOAD_L:
+ case TYPE_STORE_C:
+ case TYPE_LOAD_U:
+ case TYPE_LOAD_UX:
+ case TYPE_LOAD_EXT:
+ case TYPE_LOAD_EXT_U:
+ case TYPE_LOAD_EXT_UX:
+ case TYPE_STORE_U:
+ case TYPE_STORE_UX:
+ case TYPE_FPLOAD_U:
+ case TYPE_FPLOAD_UX:
+ case TYPE_FPSTORE_U:
+ case TYPE_FPSTORE_UX:
+ case TYPE_MFJMPR:
+ case TYPE_MTJMPR:
+ return true;
+ default:
+ break;
+ }
+ break;
default:
break;
}
@@ -20048,6 +22043,23 @@ insn_must_be_last_in_group (rtx insn)
break;
}
break;
+ case PROCESSOR_POWER7:
+ type = get_attr_type (insn);
+
+ switch (type)
+ {
+ case TYPE_ISYNC:
+ case TYPE_SYNC:
+ case TYPE_LOAD_L:
+ case TYPE_STORE_C:
+ case TYPE_LOAD_EXT_U:
+ case TYPE_LOAD_EXT_UX:
+ case TYPE_STORE_UX:
+ return true;
+ default:
+ break;
+ }
+ break;
default:
break;
}
@@ -20610,7 +22622,8 @@ rs6000_handle_altivec_attribute (tree *n
mode = TYPE_MODE (type);
/* Check for invalid AltiVec type qualifiers. */
- if (type == long_unsigned_type_node || type == long_integer_type_node)
+ if ((type == long_unsigned_type_node || type == long_integer_type_node)
+ && !TARGET_VSX)
{
if (TARGET_64BIT)
error ("use of %<long%> in AltiVec types is invalid for 64-bit code");
@@ -20620,8 +22633,8 @@ rs6000_handle_altivec_attribute (tree *n
else if (type == long_long_unsigned_type_node
|| type == long_long_integer_type_node)
error ("use of %<long long%> in AltiVec types is invalid");
- else if (type == double_type_node)
- error ("use of %<double%> in AltiVec types is invalid");
+ else if (type == double_type_node && !TARGET_VSX)
+ error ("use of %<double%> in AltiVec types is invalid without -mvsx");
else if (type == long_double_type_node)
error ("use of %<long double%> in AltiVec types is invalid");
else if (type == boolean_type_node)
@@ -20647,6 +22660,8 @@ rs6000_handle_altivec_attribute (tree *n
result = (unsigned_p ? unsigned_V16QI_type_node : V16QI_type_node);
break;
case SFmode: result = V4SF_type_node; break;
+ case DFmode: result = V2DF_type_node; break;
+ case DImode: result = V2DI_type_node; break;
/* If the user says 'vector int bool', we may be handed the 'bool'
attribute _before_ the 'vector' attribute, and so select the
proper type in the 'b' case below. */
@@ -20658,6 +22673,7 @@ rs6000_handle_altivec_attribute (tree *n
case 'b':
switch (mode)
{
+ case DImode: case V2DImode: result = bool_V2DI_type_node; break;
case SImode: case V4SImode: result = bool_V4SI_type_node; break;
case HImode: case V8HImode: result = bool_V8HI_type_node; break;
case QImode: case V16QImode: result = bool_V16QI_type_node;
@@ -20702,6 +22718,7 @@ rs6000_mangle_type (const_tree type)
if (type == bool_short_type_node) return "U6__bools";
if (type == pixel_type_node) return "u7__pixel";
if (type == bool_int_type_node) return "U6__booli";
+ if (type == bool_long_type_node) return "U6__booll";
/* Mangle IBM extended float long double as `g' (__float128) on
powerpc*-linux where long-double-64 previously was the default. */
@@ -22168,6 +24185,43 @@ rs6000_rtx_costs (rtx x, int code, int o
return false;
}
+/* Debug form of r6000_rtx_costs that is selected if -mdebug=cost. */
+
+static bool
+rs6000_debug_rtx_costs (rtx x, int code, int outer_code, int *total,
+ bool speed)
+{
+ bool ret = rs6000_rtx_costs (x, code, outer_code, total, speed);
+
+ fprintf (stderr,
+ "\nrs6000_rtx_costs, return = %s, code = %s, outer_code = %s, "
+ "total = %d, speed = %s, x:\n",
+ ret ? "complete" : "scan inner",
+ GET_RTX_NAME (code),
+ GET_RTX_NAME (outer_code),
+ *total,
+ speed ? "true" : "false");
+
+ debug_rtx (x);
+
+ return ret;
+}
+
+/* Debug form of ADDRESS_COST that is selected if -mdebug=cost. */
+
+static int
+rs6000_debug_address_cost (rtx x, bool speed)
+{
+ int ret = TARGET_ADDRESS_COST (x, speed);
+
+ fprintf (stderr, "\nrs6000_address_cost, return = %d, speed = %s, x:\n",
+ ret, speed ? "true" : "false");
+ debug_rtx (x);
+
+ return ret;
+}
+
+
/* A C expression returning the cost of moving data from a register of class
CLASS1 to one of CLASS2. */
@@ -22175,6 +24229,8 @@ int
rs6000_register_move_cost (enum machine_mode mode,
enum reg_class from, enum reg_class to)
{
+ int ret;
+
/* Moves from/to GENERAL_REGS. */
if (reg_classes_intersect_p (to, GENERAL_REGS)
|| reg_classes_intersect_p (from, GENERAL_REGS))
@@ -22182,34 +24238,48 @@ rs6000_register_move_cost (enum machine_
if (! reg_classes_intersect_p (to, GENERAL_REGS))
from = to;
- if (from == FLOAT_REGS || from == ALTIVEC_REGS)
- return (rs6000_memory_move_cost (mode, from, 0)
- + rs6000_memory_move_cost (mode, GENERAL_REGS, 0));
+ if (from == FLOAT_REGS || from == ALTIVEC_REGS || from == VSX_REGS)
+ ret = (rs6000_memory_move_cost (mode, from, 0)
+ + rs6000_memory_move_cost (mode, GENERAL_REGS, 0));
/* It's more expensive to move CR_REGS than CR0_REGS because of the
shift. */
else if (from == CR_REGS)
- return 4;
+ ret = 4;
/* Power6 has slower LR/CTR moves so make them more expensive than
memory in order to bias spills to memory .*/
else if (rs6000_cpu == PROCESSOR_POWER6
&& reg_classes_intersect_p (from, LINK_OR_CTR_REGS))
- return 6 * hard_regno_nregs[0][mode];
+ ret = 6 * hard_regno_nregs[0][mode];
else
/* A move will cost one instruction per GPR moved. */
- return 2 * hard_regno_nregs[0][mode];
+ ret = 2 * hard_regno_nregs[0][mode];
}
+ /* If we have VSX, we can easily move between FPR or Altivec registers. */
+ else if (VECTOR_UNIT_VSX_P (mode)
+ && reg_classes_intersect_p (to, VSX_REGS)
+ && reg_classes_intersect_p (from, VSX_REGS))
+ ret = 2 * hard_regno_nregs[32][mode];
+
/* Moving between two similar registers is just one instruction. */
else if (reg_classes_intersect_p (to, from))
- return (mode == TFmode || mode == TDmode) ? 4 : 2;
+ ret = (mode == TFmode || mode == TDmode) ? 4 : 2;
/* Everything else has to go through GENERAL_REGS. */
else
- return (rs6000_register_move_cost (mode, GENERAL_REGS, to)
- + rs6000_register_move_cost (mode, from, GENERAL_REGS));
+ ret = (rs6000_register_move_cost (mode, GENERAL_REGS, to)
+ + rs6000_register_move_cost (mode, from, GENERAL_REGS));
+
+ if (TARGET_DEBUG_COST)
+ fprintf (stderr,
+ "rs6000_register_move_cost:, ret=%d, mode=%s, from=%s, to=%s\n",
+ ret, GET_MODE_NAME (mode), reg_class_names[from],
+ reg_class_names[to]);
+
+ return ret;
}
/* A C expressions returning the cost of moving data of MODE from a register to
@@ -22219,14 +24289,23 @@ int
rs6000_memory_move_cost (enum machine_mode mode, enum reg_class rclass,
int in ATTRIBUTE_UNUSED)
{
+ int ret;
+
if (reg_classes_intersect_p (rclass, GENERAL_REGS))
- return 4 * hard_regno_nregs[0][mode];
+ ret = 4 * hard_regno_nregs[0][mode];
else if (reg_classes_intersect_p (rclass, FLOAT_REGS))
- return 4 * hard_regno_nregs[32][mode];
+ ret = 4 * hard_regno_nregs[32][mode];
else if (reg_classes_intersect_p (rclass, ALTIVEC_REGS))
- return 4 * hard_regno_nregs[FIRST_ALTIVEC_REGNO][mode];
+ ret = 4 * hard_regno_nregs[FIRST_ALTIVEC_REGNO][mode];
else
- return 4 + rs6000_register_move_cost (mode, rclass, GENERAL_REGS);
+ ret = 4 + rs6000_register_move_cost (mode, rclass, GENERAL_REGS);
+
+ if (TARGET_DEBUG_COST)
+ fprintf (stderr,
+ "rs6000_memory_move_cost: ret=%d, mode=%s, rclass=%s, in=%d\n",
+ ret, GET_MODE_NAME (mode), reg_class_names[rclass], in);
+
+ return ret;
}
/* Returns a code for a target-specific builtin that implements
@@ -22442,8 +24521,8 @@ rs6000_emit_swrsqrtsf (rtx dst, rtx src)
emit_label (XEXP (label, 0));
}
-/* Emit popcount intrinsic on TARGET_POPCNTB targets. DST is the
- target, and SRC is the argument operand. */
+/* Emit popcount intrinsic on TARGET_POPCNTB (Power5) and TARGET_POPCNTD
+ (Power7) targets. DST is the target, and SRC is the argument operand. */
void
rs6000_emit_popcount (rtx dst, rtx src)
@@ -22451,6 +24530,16 @@ rs6000_emit_popcount (rtx dst, rtx src)
enum machine_mode mode = GET_MODE (dst);
rtx tmp1, tmp2;
+ /* Use the PPC ISA 2.06 popcnt{w,d} instruction if we can. */
+ if (TARGET_POPCNTD)
+ {
+ if (mode == SImode)
+ emit_insn (gen_popcntwsi2 (dst, src));
+ else
+ emit_insn (gen_popcntddi2 (dst, src));
+ return;
+ }
+
tmp1 = gen_reg_rtx (mode);
if (mode == SImode)
@@ -22867,7 +24956,7 @@ rs6000_vector_mode_supported_p (enum mac
if (TARGET_SPE && SPE_VECTOR_MODE (mode))
return true;
- else if (TARGET_ALTIVEC && ALTIVEC_VECTOR_MODE (mode))
+ else if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (mode))
return true;
else
@@ -22928,4 +25017,24 @@ rs6000_final_prescan_insn (rtx insn, rtx
}
}
+/* Return true if the function has an indirect jump or a table jump. The compiler
+ prefers the ctr register for such jumps, which interferes with using the decrement
+ ctr register and branch. */
+
+bool
+rs6000_has_indirect_jump_p (void)
+{
+ gcc_assert (cfun && cfun->machine);
+ return cfun->machine->indirect_jump_p;
+}
+
+/* Remember when we've generated an indirect jump. */
+
+void
+rs6000_set_indirect_jump (void)
+{
+ gcc_assert (cfun && cfun->machine);
+ cfun->machine->indirect_jump_p = true;
+}
+
#include "gt-rs6000.h"
--
Michael Meissner, IBM
4 Technology Place Drive, MS 2203A, Westford, MA, 01886, USA
meissner@linux.vnet.ibm.com
More information about the Gcc-patches
mailing list