As explained in the PR, the addition of MVE makes the parser strip 't' and 'e'
as suffixes when MVE is enabled. This leads to vadd and vsub in it blocks with
lt and le conditions to be initially parsed as vaddl and vsubl. This means the
operand parsing for these must allow for the same operands as the scalar vadd
and vsub. I had forgotten to do this and this patch remedies that oversight.
gas/ChangeLog:
2020-03-13 Andre Vieira <andre.simoesdiasvieira@arm.com>
PR 25660
* config/tc-arm.c (operand_parse_code): Add OP_RNSDMQR and OP_oRNSDMQ.
(parse_operands): Handle new operand codes.
(do_neon_dyadic_long): Make shape check accept the scalar variants.
(asm_opcode_insns): Fix operand codes for vaddl and vsubl.
* testsuite/gas/arm/mve-vaddsub-it.s: New test.
* testsuite/gas/arm/mve-vaddsub-it.d: New test.
* testsuite/gas/arm/mve-vaddsub-it-bad.s: New test.
* testsuite/gas/arm/mve-vaddsub-it-bad.l: New test.
* testsuite/gas/arm/mve-vaddsub-it-bad.d: New test.
* testsuite/gas/arm/nomve-vaddsub-it.d: New test.
Add 3 command-line options to generate lfence for load, indirect near
branch and ret to help mitigate:
https://www.intel.com/content/www/us/en/security-center/advisory/intel-sa-00334.htmlhttp://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2020-0551
1. -mlfence-after-load=[no|yes]:
-mlfence-after-load=yes generates lfence after load instructions.
2. -mlfence-before-indirect-branch=[none|all|memory|register]:
a. -mlfence-before-indirect-branch=all generates lfence before indirect
near branches via register and a warning before indirect near branches
via memory.
b. -mlfence-before-indirect-branch=memory issue a warning before
indirect near branches via memory.
c. -mlfence-before-indirect-branch=register generates lfence before
indirect near branches via register.
Note that lfence won't be generated before indirect near branches via
register with -mlfence-after-load=yes since lfence will be generated
after loading branch target register.
3. -mlfence-before-ret=[none|or|not]
a. -mlfence-before-ret=or generates or with lfence before ret.
b. -mlfence-before-ret=not generates not with lfence before ret.
A warning will be issued and lfence won't be generated before indirect
near branch and ret if the previous item is a prefix or a constant
directive, which may be used to hardcode an instruction, since there
is no clear instruction boundary.
* config/tc-i386.c (lfence_after_load): New.
(lfence_before_indirect_branch_kind): New.
(lfence_before_indirect_branch): New.
(lfence_before_ret_kind): New.
(lfence_before_ret): New.
(last_insn): New.
(load_insn_p): New.
(insert_lfence_after): New.
(insert_lfence_before): New.
(md_assemble): Call insert_lfence_before and insert_lfence_after.
Set last_insn.
(OPTION_MLFENCE_AFTER_LOAD): New.
(OPTION_MLFENCE_BEFORE_INDIRECT_BRANCH): New.
(OPTION_MLFENCE_BEFORE_RET): New.
(md_longopts): Add -mlfence-after-load=,
-mlfence-before-indirect-branch= and -mlfence-before-ret=.
(md_parse_option): Handle -mlfence-after-load=,
-mlfence-before-indirect-branch= and -mlfence-before-ret=.
(md_show_usage): Display -mlfence-after-load=,
-mlfence-before-indirect-branch= and -mlfence-before-ret=.
(i386_cons_align): New.
* config/tc-i386.h (i386_cons_align): New.
(md_cons_align): New.
* doc/c-i386.texi: Document -mlfence-after-load=,
-mlfence-before-indirect-branch= and -mlfence-before-ret=.
PR 25614
* dwarf2dbg.c (dwarf2_directive_filename): Allow a file number of
0 if the dwarf_level is 5 or more. Complain if a filename follows
a file 0.
* testsuite/gas/elf/dwarf-5-file0.s: New test.
* testsuite/gas/elf/dwarf-5-file0.d: New test driver.
* testsuite/gas/elf/elf.exp: Run the new test.
PR 25612
* config/tc-ia64.h (DWARF2_VERISION): Fix typo.
* doc/as.texi: Fix another typo.
PR 25612
* as.c (dwarf_level): Define.
(show_usage): Add --gdwarf-3, --gdwarf-4 and --gdwarf-5.
(parse_args): Add support for the new options.
as.h (dwarf_level): Prototype.
* dwarf2dbg.c (DWARF2_VERSION): Use dwarf_level as default version
value.
* config/tc-ia64.h (DWARF2_VERISION): Update definition.
(DWARF2_LINE_VERSION): Remove definition.
* doc/as.texi: Document the new options.
Presumably as a result of various changes over the last several months,
and - for some of them - with a generalization of logic in
match_mem_size() plus mirroring of this generalization into the
broadcast handling logic of check_VecOperands(), various register-only
templates can be foled into their respective memory forms. This in
particular then also allows dropping a few more instances of IgnoreSize.
Even after commit dc2be329b9 ("i386: Only check suffix in instruction
mnemonic"), by which many of its uses have become unnecessary (some were
unnecessary even before), IgnoreSize is still used for various slightly
different purposes:
- to suppress emission of an operand size prefix,
- in Intel syntax mode to zap "derived" suffixes in certain cases and to
skip certain checks of remaining "derived" suffixes,
- to suppress ambiguous operand size / missing suffix diagnostics,
- for prefixes to suppress the "stand-alone ... prefix" warning.
Drop entirely unnecessary ones and where possible also replace instances
by the more focused (because of having just a single purpose) NoRex64.
To further restrict when IgnoreSize is needed, also generalize the logic
when to skip a template because of a present or derived L or Q suffix,
by skipping immediate operands. Additionally consider mask registers and
VecSIB there.
Note that for the time being the attribute needs to be kept in place on
MMX/SSE/etc insns (but not on VEX/EVEX encoded ones unless an operand
template of them allows for only non-SIMD-register actuals) allowing for
Dword operands - the logic when to emit a data size prefix would need
further adjustment first.
Note also that the memory forms of {,v}pinsrw get their permission for
an L or Q suffix dropped. I can only assume that it being this way was a
cut-and-paste mistake from the register forms, as the latter
specifically have NoRex64 set, and the {,v}pextrw counterparts don't
allow these suffixes either.
Convert VexW= again to their respective VexW* on lines touched anyway.
The Q and L suffix exclusion checks in match_template() ought to be
(kept) in sync as far as their FPU and SIMD aspects go. This was
already violated by only the Q one checking for active broadcast.
Convert the code such that there'll be only one instance of the logic,
the more that subsequently the logic is liable to need further
refinement / extension. (The alternative would be to drop all SIMD-ness
from the L part, but it is in principle possible to enable all sorts of
SIMD support with just a pre-386 CPU, via suitable .arch directives.)
When the template specifies any of the possible VexW settings, we can
use this instead of a separate NoRex64 to suppress the setting of REX_W.
Note that this ends up addressing an inconsistency between VEX- and
EVEX-encoded VEXTRACTPS, VPEXTR{B,W}, and VPINSR{B,W} - while the former
avoided setting VEX.W, the latter pointlessly set EVEX.W when there is a
64-bit GPR operand. Adjust the testcase to cover both cases.
Convert VexW= to their respective VexW* on lines touched anyway.
It is almost entirely redundant with Size64, and the sole case (CRC32)
where direct replacement isn't possible can easily be taken care of in
another way.
Since 16-bit addressing isn't allowed, Disp32 needs to be forced; Disp16
fails to match the templates.
The SDM leaves open whether BNDC[LNU] with a GPR operand require an
operand size override; this aspect is therefore left untouched here.
Allowing 64-bit registers is misleading here: Elsewhere these get allowed
when there's no difference between either variant, because of 32-bit
destination registers having their upper halves zeroed in 64-bit mode.
Here, however, they're source registers, and hence specifying 64-bit
registers would lead to the ambiguity of whether the upper 32 bits
actually matter.
Additionally, for proper code generation in 16-bit mode, IgnoreSize is
needed on both.
And finally, just like for e.g. MONITOR/MWAIT, add variants with all
input registers explicitly specified.
gas/
* config/tc-riscv.c: Support the modifier %got_pcrel_hi.
* doc/c-riscv.texi: Add documentation.
* testsuite/gas/riscv/no-relax-reloc.d: Add test case for the new
modifier %got_pcrel_hi.
* testsuite/gas/riscv/no-relax-reloc.s: Likewise.
* testsuite/gas/riscv/relax-reloc.d: Likewise.
* testsuite/gas/riscv/relax-reloc.s: Likewise.
* config/tc-arm.c (md_apply_fix): Warn if a PC-relative load is
detected in a section which does not have at least 4 byte
alignment.
* testsuite/gas/arm/armv8-ar-it-bad.s: Add alignment directive.
* testsuite/gas/arm/ldr-t.s: Likewise.
* testsuite/gas/arm/sp-pc-usage-t.s: Likewise.
* testsuite/gas/arm/sp-pc-usage-t.d: Finish test at end of
disassembly, ignoring any NOPs that may have been inserted because
of section alignment.
* testsuite/gas/arm/ldr-t.d: Likewise.
According to intel SDM manual, not all compare flag-modifying instructions
are marcro-fusible with subsequent jcc instructions. For those non-fusible
instructions, -malign-branch doesn't need to align them, only jcc itself
needs to be aligned.
Here are 2 restrictions which separate macro-fusible instruction from not
Restriction 1:
If TEST/AND/CMP/ADD/SUB/INC/DEC is one of the following format:
cmp m, imm
add m, imm
sub m, imm
test m, imm
and m, imm
inc m
dec m
it is unfusible with any jcc instruction.
Restriction 2:
/* Table 3-2. Macro-Fusible Instructions in Haswell Microarchitecture
Note it also works for Skylake and Cascadelake.
---------------------------------------------------------------------
| JCC | ADD/SUB/CMP | INC/DEC | TEST/AND |
| ------ | ----------- | ------- | -------- |
| Jo | N | N | Y |
| Jno | N | N | Y |
| Jc/Jb | Y | N | Y |
| Jae/Jnb | Y | N | Y |
| Je/Jz | Y | Y | Y |
| Jne/Jnz | Y | Y | Y |
| Jna/Jbe | Y | N | Y |
| Ja/Jnbe | Y | N | Y |
| Js | N | N | Y |
| Jns | N | N | Y |
| Jp/Jpe | N | N | Y |
| Jnp/Jpo | N | N | Y |
| Jl/Jnge | Y | Y | Y |
| Jge/Jnl | Y | Y | Y |
| Jle/Jng | Y | Y | Y |
| Jg/Jnle | Y | Y | Y |
Update maybe_fused_with_jcc_p to check if operands of CMP like instructions
can be fused with condition jump.
* gas/config/tc-i386.h (i386_tc_frag_data): Add member mf_type.
(TC_FRAG_INIT): Init mf_type.
* gas/config/tc-i386.c (enum mf_jcc_kind): New enum.
(enum mf_cmp_kind): Ditto.
(maybe_fused_with_jcc_p): Add argument mf_cmp_p to get
mf_type of corresponding instructons, exclude unfusible
instructions.
(add_fused_jcc_padding_frag_p): Likewise.
(add_branch_padding_frag_p): Likewise.
(output_insn): Record mf_type for corresponding instructions.
(i386_macro_fusible_p): New function.
(i386_next_fusible_jcc_frag): Rename from i386_next_jcc_frag,
add argument cmp_fragP to return next fusible jcc frag only.
(i386_classify_machine_dependant_frag): Seperate macro-fusible
instructions from condition jump.
* gas/testsuite/gas/i386/align-branch-9.s: New file.
* gas/testsuite/gas/i386/align-branch-9.d: Ditto.
* gas/testsuite/gas/i386/x86-64-align-branch-9.s: Ditto.
* gas/testsuite/gas/i386/x86-64-align-branch-9.d: Ditto.
* gas/testsuite/gas/i386/i386.exp: Run new tests.
This patch arranges for symbols defined in .sdata and .sbss to be
reported by nm with 'g' and 's' flags, for coff targets that support
.sdata and .sbss.
The assembler changes regarding SEC_SMALL_DATA are really just
documentation. As far as I'm aware, this won't change any assembler
output.
bfd/
* coff-alpha.c (alpha_ecoff_le_vec): Add SEC_SMALL_DATA to
applicable section flags.
* coff-mips.c (mips_ecoff_le_vec, mips_ecoff_be_vec): Likewise.
(mips_ecoff_bele_vec): Likewise.
* coffcode.h (sec_to_styp_flags): Set SEC_SMALL_DATA for .sdata
and .sbss sections.
* ecoff.c (_bfd_ecoff_new_section_hook): Likewise.
(_bfd_ecoff_styp_to_sec_flags): Likewise.
gas/
* config/tc-m32r.c (md_begin): Set SEC_SMALL_DATA on .scommon section.
* config/tc-mips.c (s_change_sec): Set SEC_SMALL_DATA for .sdata
and .sbss sections.
* config/tc-score.c: Delete !BFD_ASSEMBLER code throughout.
(s3_s_change_sec): Set SEC_SMALL_DATA for .sbss section.
(s3_s_score_lcomm): Likewise.
* config/tc-score7.c: Similarly.
* read.c (bss_alloc): Set SEC_SMALL_DATA for .sbss section.
When there is multi-labels on the same insn, the current code
will take care about the last one. it may cause that no sync
is added at the target.
Here we scan all labels with same value of
S_GET_VALUE(label_list->label)
by label_list->next.
2020-02-28 YunQiang Su <syq@debian.org>
PR gas/25539
* config/tc-mips.c (fix_loongson3_llsc): Compare label value
to handle multi-labels.
(has_label_name): New.
The implementation of the Arm CDE CX*A instructions doesn't allow a
condition code on the instructions but does allow the instruction in an
ITBlock.
The specification has been recently updated, though this is yet to be
published.
The instructions now should allow a suffix, and should behave in the
standard way around ITBlocks.
This patch removes the custom pred_instruction_type and handling for
these instructions, and uses the standard INSIDE_IT_INSN predication
type to describe the new instructions.
gas/ChangeLog:
2020-02-26 Matthew Malcomson <matthew.malcomson@arm.com>
* config/tc-arm.c (enum pred_instruction_type): Remove
NEUTRAL_IT_NO_VPT_INSN predication type.
(cxn_handle_predication): Modify to require condition suffixes.
(handle_pred_state): Remove NEUTRAL_IT_NO_VPT_INSN cases.
* testsuite/gas/arm/cde-scalar.s: Update test.
* testsuite/gas/arm/cde-warnings.l: Update test.
* testsuite/gas/arm/cde-warnings.s: Update test.
gettext("") returns the header entry with meta information, not the
empty string.
* config/tc-arm.c (reg_expected_msgs[REG_TYPE_RNB]): Don't use
N_() on empty string.
CSRRW and CSRRWI always write CSR. CSRRS, CSRRC, CSRRSI and CSRRCI write CSR
when RS1 isn't zero. The CSR is read only if the [11:10] bits of CSR address
is 0x3. The read-only CSR can not be written by the CSR instructions.
gas/
* config/tc-riscv.c (riscv_ip): New boolean insn_with_csr to indicate
we are assembling instruction with CSR. Call riscv_csr_read_only_check
after parsing all arguments.
(enum csr_insn_type): New enum is used to classify the CSR instruction.
(riscv_csr_insn_type, riscv_csr_read_only_check): New functions. These
are used to check if we write a read-only CSR by the CSR instruction.
* testsuite/gas/riscv/priv-reg-fail-read-only-01.s: New testcase. Test
all CSR for the read-only CSR checking.
* testsuite/gas/riscv/priv-reg-fail-read-only-01.d: Likewise.
* testsuite/gas/riscv/priv-reg-fail-read-only-01.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-read-only-02.s: New testcase. Test
all CSR instructions for the read-only CSR checking.
* testsuite/gas/riscv/priv-reg-fail-read-only-02.d: Likewise.
* testsuite/gas/riscv/priv-reg-fail-read-only-02.l: Likewise.
Add new .option `csr-check/no-csr-check` and GAS option `-mcsr-check
/-mno-csr-check` to enbale/disable the CSR checking. Disable the CSR
checking by default.
gas/
* config/tc-riscv.c: Add new .option and GAS options to enbale/disable
the CSR checking. We disable the CSR checking by default.
(reg_lookup_internal): Check the `riscv_opts.csr_check`
before we doing the CSR checking.
* doc/c-riscv.texi: Add description for the new .option and assembler
options.
* testsuite/gas/riscv/priv-reg-fail-fext.d: Add `-mcsr-check` to enable
the CSR checking.
* testsuite/gas/riscv/priv-reg-fail-rv32-only.d: Likewise.
According to the riscv privilege spec, some CSR are only valid when rv32 or
the specific extension is set. We extend the DECLARE_CSR and DECLARE_CSR_ALIAS
to record more informaton we need, and then check whether the CSR is valid
according to these information. We report warning message when the CSR is
invalid, so we have a choice between error and warning by --fatal-warnings
option. Also, a --no-warn/-W option is used to turn the warnings off, if
people don't want the warnings.
gas/
* config/tc-riscv.c (enum riscv_csr_class): New enum. Used to decide
whether or not this CSR is legal in the current ISA string.
(struct riscv_csr_extra): New structure to hold all extra information
of CSR.
(riscv_init_csr_hash): New function. According to the DECLARE_CSR and
DECLARE_CSR_ALIAS, insert CSR extra information into csr_extra_hash.
Call hash_reg_name to insert CSR address into reg_names_hash.
(md_begin): Call riscv_init_csr_hashes for each DECLARE_CSR.
(reg_csr_lookup_internal, riscv_csr_class_check): New functions.
Decide whether the CSR is valid according to the csr_extra_hash.
(init_opcode_hash): Update 'if (hash_error != NULL)' as hash_error is
not a boolean. This is same as riscv_init_csr_hash, so keep the
consistent usage.
* testsuite/gas/riscv/csr-dw-regnums.d: Add -march=rv32if option.
* testsuite/gas/riscv/priv-reg.d: Add f-ext by -march option.
* testsuite/gas/riscv/priv-reg-fail-fext.d: New testcase. The source
file is `priv-reg.s`, and the ISA is rv32i without f-ext, so the
f-ext CSR are not allowed.
* testsuite/gas/riscv/priv-reg-fail-fext.l: Likewise.
* testsuite/gas/riscv/priv-reg-fail-rv32-only.d: New testcase. The
source file is `priv-reg.s`, and the ISA is rv64if, so the
rv32-only CSR are not allowed.
* testsuite/gas/riscv/priv-reg-fail-rv32-only.l: Likewise.
include/
* opcode/riscv-opc.h: Extend DECLARE_CSR and DECLARE_CSR_ALIAS to
record riscv_csr_class.
opcodes/
* riscv-dis.c (print_insn_args): Updated since the DECLARE_CSR is changed.
gdb/
* riscv-tdep.c: Updated since the DECLARE_CSR is changed.
* riscv-tdep.h: Likewise.
* features/riscv/rebuild-csr-xml.sh: Generate the 64bit-csr.xml without
rv32-only CSR.
* features/riscv/64bit-csr.xml: Regernated.
binutils/
* dwarf.c: Updated since the DECLARE_CSR is changed.
This allows pdp11 to handle 32-bit fixups that can be applied by gas,
and improves the error message emitted when the required reloc isn't
available.
* config/tc-pdp11.c (md_apply_fix): Handle BFD_RELOC_32.
(tc_gen_reloc): Only give a BAD_CASE assertion on pcrel relocs.
This patch fixes an ancient wart in aout support, in that text and
data section sizes are rounded up for alignment rather that just the
corresponding header sizes. Changing section sizes could conceivably
result in buffer overflows if section contents were held in memory.
Also, keeping the original section sizes allows this PR to be fixed
nicely.
bfd/
PR 25569
* aoutx.h (adjust_o_magic, adjust_z_magic, adjust_n_magic): Use
"text", "data" and "bss" section pointer vars. Don't update
section size, just exec header sizes.
(adjust_sizes_and_vmas): Don't update text section size. Set
initial exec header a_text. Print exec headers sizes.
* pdp11.c (adjust_o_magic, adjust_z_magic, adjust_n_magic),
(adjust_sizes_and_vmas): Similarly. Formatting.
(final_link): Correct final file extension.
gas/
PR 25569
* config/obj-aout.c (obj_aout_frob_file_before_fix): Don't loop
on section size adjustment, instead perform another write if
exec header size is larger than section size.
* config/tc-i386.c (output_insn): Mark cvtpi2ps and cvtpi2pd
with GNU_PROPERTY_X86_FEATURE_2_MMX.
* testsuite/gas/i386/i386.exp: Run property-3 and
x86-64-property-3.
* testsuite/gas/i386/property-3.d: New file.
* testsuite/gas/i386/property-3.s: Likewise.
* testsuite/gas/i386/x86-64-property-3.d: Likewise.
AMD ABM has 2 instructions: popcnt and lzcnt. ABM CPUID feature bit has
been reused for lzcnt and a POPCNT CPUID feature bit is added for popcnt
which used to be the part of SSE4.2. This patch removes CpuABM and adds
CpuPOPCNT. It changes ABM to enable both lzcnt and popcnt, changes SSE4.2
to also enable popcnt.
gas/
* config/tc-i386.c (cpu_arch): Add .popcnt.
* doc/c-i386.texi: Remove abm and .abm. Add popcnt and .popcnt.
Add a tab before @samp{.sse4a}.
opcodes/
* i386-gen.c (cpu_flag_init): Replace CpuABM with
CpuLZCNT|CpuPOPCNT. Add CpuPOPCNT to CPU_SSE4_2_FLAGS. Add
CPU_POPCNT_FLAGS.
(cpu_flags): Remove CpuABM. Add CpuPOPCNT.
* i386-opc.h (CpuABM): Removed.
(CpuPOPCNT): New.
(i386_cpu_flags): Remove cpuabm. Add cpupopcnt.
* i386-opc.tbl: Replace CpuABM|CpuSSE4_2 with CpuPOPCNT on
popcnt. Remove CpuABM from lzcnt.
* i386-init.h: Regenerated.
* i386-tbl.h: Likewise.
There's no need to have separate Cpu64 and CpuNo64 templates: There
already is special logic handling the attribute, and all that's needed
is rejecting 16-bit address registers in 64-bit mode. Suppress suffix
guessing and group all involved logic together, outside of suffix
processing (arguably it doesn't even belong in process_suffix()).
Also, since no AddrPrefixOpReg template permits any suffixes, move the
No_*Suf specifiers for them to a central place. Along with this drop
the no longer relevant NoRex64 from there.
Generally, the documentation doesn't allow for any explicit operands
to be specified with MONITOR/MWAIT. To permit the more legible
overriding of the address size via specifying operands, the option is
being retained even in Intel mode, but operand swapping is being
suppressed by this patch. This is both because it makes no sense here
(all of the operands are inputs) and because, as a result, old gcc
(prior to 4.8) actually expects it this way with -mintel-syntax (and
hence gets fixed by this change rather than, as claimed by a reply in
the bug report, broken).
Conversions which shrink element size and which have a memory source
can't be disambiguated between their 128- and 256-bit variants by
looking at the register operand. "operand size mismatch", however, is a
pretty misleading diagnostic. Generalize the logic introduced for
VFPCLASSP{S,D} such that, with suitable similar adjustments to the
respective templates, it'll cover these cases too.
For VCVTNEPS2BF16 also fold the two previously separate AVX512VL
templates to achieve the intended effect. This is then also accompanied
by a respective addition to the inval-avx512f testcase.
commit 7deea9aad8 changed nosse4 to include CpuSSE4a. But AMD SSE4a is
a superset of SSE3 and Intel SSE4 is a superset of SSSE3. Disable Intel
SSE4 shouldn't disable AMD SSE4a. This patch restores nosse4. It also
adds .sse4a and nosse4a.
gas/
* config/tc-i386.c (cpu_arch): Add .sse4a and nosse4a. Restore
nosse4.
* doc/c-i386.texi: Document sse4a and nosse4a.
opcodes/
* i386-gen.c (cpu_flag_init): Add CPU_ANY_SSE4A_FLAGS. Remove
CPU_ANY_SSE4_FLAGS.
For these to get treatment consistent with other operand size checking
the special logic shouldn't live in md_assemble(), but process_suffix().
And there's more logic involved than simply zapping the suffix.
Note however that MOVS[BW]* and MOVZ[BW]* still won't be fully
consistent, due to the objection to fold MOVS* templates just like was
done for MOVZ* in c07315e0c6 ("x86: allow suffix-less movzw and 64-bit
movzb").
Note further that it is against my own intentions to have MOVSX/MOVZX
silently default to a byte source in AT&T mode. This should happen only
when the destination register is a 16-bit one. In all other cases there
is an ambiguity, and the user should be warned. But it was explicitly
requested for this to be done in a way inconsistent with everything
else.
Note finally that the assembler change points out (and this patch fixes)
a wrong Intel syntax test introduced by bc31405ebb ("x86-64: Properly
encode and decode movsxd"): When source code specifies a 16-bit
destination register, disassembly expectations shouldn't have been to
find a 32-bit one.
Since we already suppress the prefix altogether when it's the default
one for the chosen addressing mode, let's do so also when instruction
prefix and override specified with the memory operand match. (Note that
insn prefix specified segment overrides never get discarded.)
For one both possible forms should be warned about. And then, to guard
against future surprises, qualify the original opcode check by excluding
VEX/EVEX-like templates.
Since PLT entry isn't needed for branch to local symbol, we can resolve
R_386_PLT32/R_X86_64_PLT32 relocation aganst local symbol to section,
similar to R_386_PC32/R_X86_64_PC32.
2020-02-13 Fangrui Song <maskray@google.com>
H.J. Lu <hongjiu.lu@intel.com>
PR gas/25551
* config/tc-i386.c (tc_i386_fix_adjustable): Don't check
BFD_RELOC_386_PLT32 nor BFD_RELOC_X86_64_PLT32.
* testsuite/gas/i386/i386.exp: Run relax-5 and x86-64-relax-4.
* testsuite/gas/i386/relax-5.d: New file.
* testsuite/gas/i386/relax-5.s: Likewise.
* testsuite/gas/i386/x86-64-relax-4.d: Likewise.
* testsuite/gas/i386/x86-64-relax-4.s: Likewise.
With AVX512VL disabled (e.g. when writing code for the Knights family
of processors) these insns aren't ambiguous when used with a memory
source, and hence should be accepted without suffix or operand size
specifier. When AVX512VL is enabled, to be consistent with this as
well as other ambiguous operand size handling it would seem better to
just warn about the ambiguity in AT&T mode, and still default to 512-bit
operands (on the assumption that the code may have been written without
AVX512VL in mind yet), but it was requested to leave AT&T syntax mode
alone here.
AMD and Intel differ in their handling of far indirect branches as well
as LFS/LGS/LSS: AMD CPUs ignore REX.W while Intel ones honors it. (Note
how the latter three were hybrids so far, while far branches were fully
AMD-like.)
Along the lines of be4c5e58bd ("x86: Always disallow double word suffix
with word general register") also adjust check_{byte,word}_reg(), to make
overall behavior consistent again in this regard.
The diagnostics issued by check_*_reg() are pretty AT&T-centric. Re-use
logic already used for SIMD memory operand size checking also for ones
where GPRs would alternatively also be allowed. (There's certainly room
for further improvement here.)
