not supported by the current arch, only change the name if
its contents are the same as prev_name.
(get_specific): If the the architecture doesn't match, fail.
* elf32-mips.c (mips_elf_next_relocation): Rename from
mips_elf_next_lo16_relocation, and generalize to look
for any relocation type.
(elf_mips_howto_table): Make R_MIPS_PC16 pcrel_offset.
(elf_mips_gnu_rel_hi16): Howto for R_MIPS_GNU_REL_HI16.
(elf_mips_gnu_rel_lo16): Howto for R_MIPS_GNU_REL_LO16.
(elf_mips_gnu_rel16_s2): Howto for R_MIPS_GNU_REL16_S2.
(elf_mips_gnu_pcrel64): Howto for R_MIPS_PC64.
(elf_mips_gnu_pcrel32): Howto for R_MIPS_PC32.
(bfd_elf32_bfd_reloc_type_lookup): Add new relocs.
(mips_rtype_to_howto): Likewise.
(mips_elf_calculate_relocation): Handle new relocs.
(_bfd_mips_elf_relocate_section): REL_HI16/REL_LO16 relocs
are paired. The addend for R_MIPS_GNU_REL16_S2
is shifted right two bits.
In gas/:
* config/tc-mips.c (mips_ip): Don't put stuff in .rodata
when embedded-pic.
* config/tc-mips.c (SWITCH_TABLE): The ELF embedded-pic
implementation doesn't have special handling for switch
statements.
(macro_build): Allow for code in sections other than .text.
(macro): Likewise.
(mips_ip): Likewise.
(md_apply_fix): Do pc-relative relocation madness for MIPS ELF.
Don't perform relocs if we will be outputting them.
(tc_gen_reloc): For ELF, just use fx_addnumber for pc-relative
relocations. Allow BFD_RELOC_16_PCREL_S2 relocs when
embedded-pic.
In gas/testsuite/:
* gas/mips/empic.d: New file.
* gas/mips/empic.s: New file.
* gas/mips/mips16-e.d: New file.
* gas/mips/mips16-e.s: New file.
* gas/mips/mips16-f.d: New file.
* gas/mips/mips16-f.s: New file.
* gas/mips/mips.exp: Add empic, mips16-e. Add mips16-f as an
expected failure.
In include/elf:
* mips.h: Add R_MIPS_GNU_REL_HI16, R_MIPS_GNU_REL_LO16,
R_MIPS_GNU_REL16_S2, R_MIPS_PC64 and R_MIPS_PC32 relocation
numbers.
* doc/internals.texi (CPU backend): Mention that
line_separator_chars do not break up comments. Fix typos for
LEX_AT and LEX_NAME descriptions. Document operands for
TC_EQUAL_IN_INSN, md_operand and md_section_align. Correct
description of md_create_short_jump usage. Document argument for
md_undefined_symbol.
the errors exposed by this addition. These were intel mode
"fi... word ptr", "fi... dword ptr", "jmp Imm seg, Imm offset", "out dx,al".
The failure with intel "out dx,al" was also present in att "out al,dx".
Extend testsuite to catch this case too.
* config/tc-mips.c (mips_gp32): New variable.
(macro_build) Use mips_gp32.
(mips_ip): Ditto.
(md_longopts): Add "-mgp32" and "-mgp64".
(md_parse_option): Add OPTION_GP32 and OPTION_GP64.
Reinstate bits of sh4 support that got accidentally deleted.
Add sh-dsp support.
bfd:
* archures.c (bfd_mach_sh2, bfd_mach_sh_dsp): New macros.
(bfd_mach_sh3_dsp): Likewise.
(bfd_mach_sh4): Reinstate.
(bfd_default_scan): Recognize 7410, 7708, 7729 and 7750.
* bfd-in2.h: Regenerate.
* coff-sh.c (struct sh_opcode): flags is no longer short.
(USESAS, USESAS_REG, USESR8, SETSAS, SETSAS_REG): New macros.
(sh_opcode41, sh_opcode42): Integrate as sh_opcode41.
(sh_opcode01, sh_opcode02, sh_opcode40): Add sh-dsp opcodes.
(sh_opcode41, sh_opcode4, sh_opcode80): Likewise.
(sh_opcodes): No longer const.
(sh_dsp_opcodef0, sh_dsp_opcodef): New arrays.
(sh_insn_uses_reg): Check for USESAS and USESR8.
(sh_insn_sets_reg, sh_insns_conflict): Check for SETSAS.
(_bfd_sh_align_load_span): Return early for SH4.
Modify sh_opcodes lookup table for sh-dsp / sh3-dsp.
Take into account that field b of a parallel processing insn
could be mistaken for a separate insn.
* cpu-sh.c (arch_info_struct): New array elements for
sh2, sh-dsp and sh3-dsp.
Reinstate element for sh4.
(SH2_NEXT, SH_DSP_NEXT, SH3_DSP_NEXT): New macros.
(SH4_NEXT): Reinstate.
(SH3_NEXT, SH3E_NEXT): Adjust.
* elf-bfd.h (_sh_elf_set_mach_from_flags): Declare.
* elf32-sh.c (sh_elf_set_private_flags): New function.
(sh_elf_copy_private_data, sh_elf_set_mach_from_flags): Likewise.
(sh_elf_merge_private_data): New function.
(elf_backend_object_p, bfd_elf32_bfd_set_private_bfd_flags): Define.
(bfd_elf32_bfd_copy_private_bfd_data): Define.
(bfd_elf32_bfd_merge_private_bfd_data): Change to
sh_elf_merge_private_data.
gas:
* config/tc-sh.c ("elf/sh.h"): Include.
(sh_dsp, valid_arch, reg_x, reg_y, reg_efg): New static variables.
(md.begin): Initialize target_arch.
Only include opcodes in has table that match selected architecture.
(parse_reg): Recognize register names for sh-dsp.
(parse_at): Recognize post-modify addressing.
(get_operands): The leading space is now optional.
(get_specific): Remove FDREG_N support. Add support for sh-dsp
arguments. Update valid_arch.
(build_Mytes): Add support for SDT_REG_N.
(find_cooked_opcode): New function, broken out of md_assemble.
(assemble_ppi, sh_elf_final_processing): New functions.
(md_assemble): Use find_cooked_opcode and assemble_ppi.
(md_longopts, md_parse_option): New option: -dsp.
* config/tc-sh.h (elf_tc_final_processing): Define.
(sh_elf_final_processing): Declare.
include/elf:
* sh.h: (EF_SH_MACH_MASK, EF_SH_UNKNOWN, EF_SH1, EF_SH2): New macros.
(EF_SH3, EF_SH_HAS_DSP, EF_SH_DSP, EF_SH3_DSP): Likewise.
(EF_SH_HAS_FP, EF_SH3E, EF_SH4, EF_SH_MERGE_MACH): Likewise.
opcodes:
* sh-dis.c (print_movxy, print_insn_ddt, print_dsp_reg): New functions.
(print_insn_ppi): Likewise.
(print_insn_shx): Use info->mach to select appropriate insn set.
Add support for sh-dsp. Remove FD_REG_N support.
* sh-opc.h (sh_nibble_type): Add new values for sh-dsp support.
(sh_arg_type): Likewise. Remove FD_REG_N.
(sh_dsp_reg_nums): New enum.
(arch_sh1, arch_sh2, arch_sh3, arch_sh3e, arch_sh4): New macros.
(arch_sh_dsp, arch_sh3_dsp, arch_sh1_up, arch_sh2_up): Likewise.
(arch_sh3_up, arch_sh3e_up, arch_sh4_up, arch_sh_dsp_up): Likewise.
(arch_sh3_dsp_up): Likewise.
(sh_opcode_info): New field: arch.
(sh_table): Split up insn with FD_REG_N into ones with F_REG_N and
D_REG_N. Fill in arch field. Add sh-dsp insns.
already defined.
* config/tc-ppc.h [OBJ_XCOFF] (OBJ_COPY_SYMBOL_ATTRIBUTES):
New macro.
* config/tc-ppc.c (ppc_fix_adjustable): Don't look at the frag
of a symbol when we really care about its value.
assembler. ie. You will be able to do "as --em=i386aout" on an x86
linux-elf assembler to generate aout format object files, rather than
using a separate assembler. The aout emulation is enabled by giving
"--enable-targets=i386-linuxaout" to configure.
Oh yeah, there's a couple of fixes too. Error messages shouldn't be
passed to printf in the format arg just in case someone puts a `%' in
the message.
* gas/i386/general.{s,l}: Move 16 bit jmp and call tests from here
* gas/i386/jump16.{s,d}: To here.
* gas/i386/jump.{s,d}: New files. Duplicate 32 bit jmp and call
tests in general.s so that objdump is exercised.
* gas/i386/i386.exp: Call new tests.
Alexandre Oliva