For another patch I wanted to use a sufficiently benign option (simply
to be able to specify one, which certain test case invocations require),
and I stumbled across -Q in the --help output. Before realizing that
this is x86-specific anyway, I've tried and and ran into a mysterious
testsuite failure, until I further realized that other than the help
text suggests the option requires an argument. Correct the help text,
and make the implementation actually match what the comment there has
been describing (and what the help text now says).
Recently a patch was submitted for a Xen Project test harness binary to
override the compiler specified @object to @func (see [1]). In a reply I
suggested we shouldn't make ourselves dependent on currently unspecified
behavior of gas here: It accumulates all requests, and then
bfd/elf.c:swap_out_syms(), in an apparently ad hoc manner, prioritizes
certain flags over others.
Make the behavior predictable: Generally the last .type is what counts.
Exceptions are directives which set multiple bits (TLS, IFUNC, and
UNIQUE): Subsequent directives requesting just the more generic bit
(i.e. FUNC following IFUNC) won't clear the more specific one. Warn
about incompatible changes, except from/to STT_NOTYPE.
Also add a new target hook, which hppa wants to use right away afaict.
In the course of adding the warning I ran into two ld testsuite
failures. I can only assume that it was a copy-and-paste mistake that
lead to the same symbol having its type set twice.
[1] https://lists.xenproject.org/archives/html/xen-devel/2019-05/msg01980.html
For example, when provided with the (incorrect) instruction
st4 {v0.16b-v3.16b}[4],[x0]
currently assembler provides the following error message
"Error: comma expected between operands at operand 2 -- `st4 {v0.16b-v3.16b}[4],[x0]'".
This was due to the assembler consuming the {v0.16b-v3.16b} as the first operand leaving
[4],[x0] as what it believed to be the second operand.
The actual error is that the first operand should be of element type and not
vector type (as provided). The new diagnostic for this error is
"Error: expected element type rather than vector type at operand 1 -- `st4 {v0.16b-v3.16b}[4],[x0]'.
Added testcases to check for the correct diagnostic message as well as checking that
variations of the structural load/store by element instruction also generate the error
when they have the same problem.
* config/tc-aarch64.c (parse_operands): Add error check.
* testsuite/gas/aarch64/diagnostic.l: New test.
* testsuite/gas/aarch64/diagnostic.s: New test.
* testsuite/gas/aarch64/illegal.l: New tests.
* testsuite/gas/aarch64/illegal.s: New tests.
The entry for the FMOV alias of FCPY was missing C_SCAN_MOVPRFX.
(The entry for FCPY itself was OK.)
This was the only /m-predicated instruction I could see that was
missing the flag.
2019-07-02 Richard Sandiford <richard.sandiford@arm.com>
opcodes/
* aarch64-tbl.h (aarch64_opcode): Set C_SCAN_MOVPRFX for the
SVE FMOV alias of FCPY.
gas/
* testsuite/gas/aarch64/sve-movprfx_27.s,
* testsuite/gas/aarch64/sve-movprfx_27.d: New test.
SVE FCVTZS, FCVTZU, SCVTF and UCVTF need the same treatment as FCVT:
the register size used in a predicated MOVPRFX must be the wider of
the destination and source sizes.
Since I was adding a (supposedly) complete set of tests for converts,
it seemed more consistent to add a complete set of tests for shifts
as well, even though there's no bug to fix there.
2019-07-02 Richard Sandiford <richard.sandiford@arm.com>
opcodes/
* aarch64-tbl.h (aarch64_opcode_table): Add C_MAX_ELEM flags
to SVE fcvtzs, fcvtzu, scvtf and ucvtf entries.
gas/
* testsuite/gas/aarch64/sve-movprfx_26.s: Also test FCVTZS, FCVTZU,
SCVTF, UCVTF, LSR and ASR.
* testsuite/gas/aarch64/sve-movprfx_26.d: Update accordingly.
* testsuite/gas/aarch64/sve-movprfx_26.l: Likewise.
One of the MOVPRFX tests has:
output register of preceding `movprfx' used as input at operand 3 -- `cpy z1.d,p1/m,x1'
But X1 and Z1 are not the same register, so the instruction is
actually OK.
2019-07-02 Richard Sandiford <richard.sandiford@arm.com>
opcodes/
* aarch64-opc.c (verify_constraints): Skip GPRs when scanning the
registers in an instruction prefixed by MOVPRFX.
gas/
* testsuite/gas/aarch64/sve-movprfx_25.s: Allow CPY Z1.D.P1/M,X1
to be prefixed by MOVPRFX.
* testsuite/gas/aarch64/sve-movprfx_25.d: Update accordingly.
* testsuite/gas/aarch64/sve-movprfx_25.l: Likewise.
PR 24748
* write.c (create_note_reloc): Add desc2_offset parameter. Change
name of offset parameter to note_offset. Only use desc2_offset
when placing addend into REL reloc's address space.
(maybe_generate_build_notes): Update parameters passed to
create_note_reloc.
I had mistakenly given all variants of the new SVE2 instructions
pmull{t,b} a dependency on the feature +sve2-aes.
Only the variant specifying .Q -> .D sizes should have that
restriction.
This patch fixes that mistake and updates the testsuite to have extra
tests (matching the given set of tests per line in aarch64-tbl.h that
the rest of the SVE2 tests follow).
We also add a line in the documentation of the command line to clarify
how to enable `pmull{t,b}` of this larger size. This is needed because
all other instructions gated under the `sve2-aes` architecture extension
are marked in the instruction documentation by an `HaveSVE2AES` check
while pmull{t,b} is gated under the `HaveSVE2PMULL128` check.
Regtested targeting aarch64-linux.
gas/ChangeLog:
2019-07-01 Matthew Malcomson <matthew.malcomson@arm.com>
* testsuite/gas/aarch64/illegal-sve2-aes.d: Update tests.
* testsuite/gas/aarch64/illegal-sve2.l: Update tests.
* doc/c-aarch64.texi: Add special note of pmull{t,b}
instructions under the sve2-aes architecture extension.
* testsuite/gas/aarch64/illegal-sve2.s: Add small size
pmull{t,b} instructions.
* testsuite/gas/aarch64/sve2.d: Add small size pmull{t,b}
disassembly.
* testsuite/gas/aarch64/sve2.s: Add small size pmull{t,b}
instructions.
include/ChangeLog:
2019-07-01 Matthew Malcomson <matthew.malcomson@arm.com>
* opcode/aarch64.h (enum aarch64_insn_class): sve_size_013
renamed to sve_size_13.
opcodes/ChangeLog:
2019-07-01 Matthew Malcomson <matthew.malcomson@arm.com>
* aarch64-asm.c (aarch64_encode_variant_using_iclass): Use new
sve_size_13 icode to account for variant behaviour of
pmull{t,b}.
* aarch64-dis-2.c: Regenerate.
* aarch64-dis.c (aarch64_decode_variant_using_iclass): Use new
sve_size_13 icode to account for variant behaviour of
pmull{t,b}.
* aarch64-tbl.h (OP_SVE_VVV_HD_BS): Add new qualifier.
(OP_SVE_VVV_Q_D): Add new qualifier.
(OP_SVE_VVV_QHD_DBS): Remove now unused qualifier.
(struct aarch64_opcode): Split pmull{t,b} into those requiring
AES and those not.
It is pretty wasteful to have a per-operand flag which is used in
exactly 4 cases. It can be relatively easily replaced, and by doing so
I've actually found some dead code to remove at the same time (there's
no case of ImmExt set at the same time as Vec_Imm4).
In quite a few cases ImmExt gets used when there's not really any
immediate, but rather a degenerate ModR/M byte. ENCL{S,U} show how this
case is supposed to be dealt with. Eliminate most abuses, leaving in
place (for now) only ones where process_immext() is involved.
It seems to be not uncommon for people to use AND or OR in this form for
just setting the status flags. TEST, which doesn't write to any
register other than EFLAGS, ought to be preferred. Make the change only
for -O2 and above though, at least for now.
When they're in the 0F opcode space, swapping their source operands may
allow switching from 3-byte to 2-byte VEX prefix encoding. Note that NaN
behavior precludes us doing so for many packed and scalar floating point
insns; such an optimization would need to be done by the compiler
instead in this case, when it knows that NaN-s have undefined behavior
anyway.
While for explicitly specified AVX/AVX2 insns the optimization (for now
at least) gets done only for -O2 and -Os, it is utilized by default in
SSE2AVX mode, as there we're re-writing the programmer's specified insns
anyway.
Rather than introducing a new attribute flag, the change re-uses one
which so far was meaningful only for EVEX-encoded insns.
This implication allows to simplify some conditionals, thus slightly
improving performance. This change also paves the way for re-using
StaticRounding for non-EVEX insns.
As long as there's no write mask as well as no broadcast, and as long
as the scaled Disp8 wouldn't result in a shorter EVEX encoding, encode
VPAND{D,Q}, VPANDN{D,Q}, VPOR{D,Q}, and VPXOR{D,Q} acting on only the
lower 16 XMM/YMM registers using their VEX equivalents with -O1.
Also take the opportunity and avoid looping twice over all operands
when dealing with memory-with-displacement ones.
While the ISA extensions doc suggests them to be made available just
like the SDM does for the PCLMULQDQ ISA extension, these weren't added
when supposrt for the new extension was introduced.
Also make sure the 64-bit non-AVX512 test actually tests VEX encodings,
not EVEX ones.
For example, this code is invalid:
smc #0x6951
The code would previously check for and encode for up to 16 bit immediate values, however
this immediate should instead be only a 4 bit value
(as documented herehttps://static.docs.arm.com/ddi0406/c/DDI0406C_C_arm_architecture_reference_manual.pdf ).
Fixed this by adding range checks in the relevant areas and also removing code that would
encode more than the first 4 bits of the immediate (code that is now redundant, as any immediate operand
larger than 0xF would error now anyway).
gas * config/tc-arm.c (do_smc): Add range check for immediate operand.
(do_t_smc): Add range check for immediate operand. Remove
obsolete immediate encoding.
(md_apply_fix): Fix range check. Remove obsolete immediate encoding.
* testsuite/gas/arm/arch6zk.d: Fix test.
* testsuite/gas/arm/arch6zk.s: Fix test.
* testsuite/gas/arm/smc-bad.d: New test.
* testsuite/gas/arm/smc-bad.l: New test.
* testsuite/gas/arm/smc-bad.s: New test.
* testsuite/gas/arm/thumb32.d: Fix test.
* testsuite/gas/arm/thumb32.s: Fix test.
These encodings aren't valid in real and VM86 modes, but they are very
well usable in 16-bit protected mode.
A few adjustments in the disassembler tables are needed where Ev or Gv
were wrongly used. Additionally an adjustment is needed to avoid
printing "addr32" when that's already recognizable by the use of %eiz.
Furthermore the Iq operand template was wrong for XOP:0Ah encoding
insns: They're having a uniform 32-bit immediate. Drop Iq and introduce
Id instead.
Clone a few existing test cases to exercise assembler and disassembler.
Noticed by a customer while looking at a tangentially related problem. The
gas testsuite for xstormy16 has two scripts that have a typo on the first
line, they are missing the !. They also use shell syntax that doesn't work
on a system where /bin/sh is dash. So I fixed the typo, changed the shell
to bash, and made them executable, so that they now work when run directly
even if /bin/sh is dash.
gas/
* testsuite/gas/xstormy16/allinsn.sh: Change first line to
#!/bin/bash and make it executable.
* testsuite/gas/xstormy16/gcc.sh: Likewise.
When MTHC1 instruction is paired with MTC1 to write a value to a
64-bit FPR, the MTC1 must be executed first, because the semantic
definition of MTC1 is not aware that software will be using an MTHC1
to complete the operation, and sets the upper half of the 64-bit FPR
to an UNPREDICTABLE value[1].
Fix the order of MTHC1 and MTC1 instructions in LI macro expansion.
Modify the expansions to exploit moves from $zero directly by-passing
the use of $AT, where ever possible.
[1] "MIPS Architecture for Programmers Volume II-A: The MIPS32
Instruction Set Manual", Wave Computing, Inc., Document
Number: MD00086, Revision 5.04, December 11, 2013, Section 3.2
"Alphabetical List of Instructions", pp. 217.
gas/
* config/tc-mips.c (macro) <M_LI>: Re-order MTHC1 with
respect to MTC1 and use $0 for either part where possible.
* testsuite/gas/mips/li-d.s: Add test cases for non-zero
words in double precision constants.
* testsuite/gas/mips/li-d.d: Update reference output.
* testsuite/gas/mips/micromips@isa-override-1.d: Likewise.
* testsuite/gas/mips/mips32r2@isa-override-1.d: Likewise.
* testsuite/gas/mips/mips64r2@isa-override-1.d: Likewise.
For quite some time we've been using combinations of bits for
specifying various registers in operands and templates. I think it was
Alan who had indicated that likely the debug printing would need
adjustment as a result. Here we go.
Accumulator handling for GPRs gets changed to match that for FPU regs.
For this to work, OPERAND_TYPE_ACC{32,64} get repurposed, with their
original uses replaced by direct checks of the two bits of interest,
which is cheaper than operand_type_equal() invocations.
For SIMD registers nothing similar appears to be needed, as respective
operands get stripped from the (copy of the) template before pt() is
reached.
The type change on pi() is to silence a compiler diagnostic. Arguably
its other parameter could also be const-qualified.
MOVNTI was wrongly assembled with a 66h prefix. Add IgnoreSize to
address this. It and the scalar to/from integer conversion insns also
were also wrongly using Ev / Gv, leading to 16-bit register names being
printed when 32-bit ones were meant.
Clone the 32-bit SSE2 test to cover both assembler and disassembler.
This patch corrects ppc rs_align_code handling to choose the alignment
nops based on the machine in force at the alignment directive rather
than the machine at the end of file.
* config/tc-ppc.h (ppc_nop_select): Declare.
(NOP_OPCODE): Define.
* config/tc-ppc.c (ppc_elf_end, ppc_xcoff_end): Zero ppc_cpu.
(ppc_nop_encoding_for_rs_align_code): New enum.
(ppc_nop_select): New function.
(ppc_handle_align): Don't use ppc_cpu here. Get nop type from frag.
* testsuite/gas/ppc/groupnop.d,
* testsuite/gas/ppc/groupnop.s: New test.
* testsuite/gas/ppc/ppc.exp: Run it.
If VEX.vvvv and EVEX.vvvv are reserved, they must be all 1s, which are
all 0s in inverted form. Add check for unused VEX.vvvv and EVEX.vvvv
when disassembling VEX and EVEX instructions.
gas/
PR binutils/24626
* testsuite/gas/i386/disassem.s: Add tests for reserved VEX.vvvv
and EVEX.vvvv.
* testsuite/gas/i386/x86-64-disassem.s: Likewise.
* testsuite/gas/i386/disassem.d: Updated.
* testsuite/gas/i386/x86-64-disassem.d: Likewise.
opcodes/
PR binutils/24626
* i386-dis.c (print_insn): Check for unused VEX.vvvv and
EVEX.vvvv when disassembling VEX and EVEX instructions.
(OP_VEX): Set vex.register_specifier to 0 after readding
vex.register_specifier.
(OP_Vex_2src_1): Likewise.
(OP_Vex_2src_2): Likewise.
(OP_LWP_E): Likewise.
(OP_EX_Vex): Don't check vex.register_specifier.
(OP_XMM_Vex): Likewise.
This fixes a bug reported on the riscv.org sw-dev mailing list. This
rejects "lui x1,symbol", as a symbol should only be accepted here when
used inside %hi(). Without the fix, this gets assembled as "lui x1,0"
with no relocation which is clearly wrong.
gas/
* config/tc-riscv.c (riscv_ip) <'u'>: Move O_constant check inside if
statement. Delete O_symbol and O_constant check after if statement.
* testsuite/gas/riscv/auipc-parsing.s: Test lui with missing %hi.
* testsuite/gas/riscv/auipc-parsing.l: Update.
For AVX512 instructions with Disp8ShiftVL and Broadcast, we may need to
add CheckRegSize to check if broadcast matches the destination register
size.
gas/
PR gas/24625
* testsuite/gas/i386/inval-avx512f.s: Add tests for AVX512_BF16
instructions with invalid broadcast.
* testsuite/gas/i386/x86-64-inval-avx512f.s: Likewise.
* testsuite/gas/i386/inval-avx512f.l: Updated.
* testsuite/gas/i386/x86-64-inval-avx512f.l: Likewise.
opcodes/
PR gas/24625
* i386-opc.tbl: Add CheckRegSize to AVX512_BF16 instructions with
Disp8ShiftVL.
* i386-tbl.h: Regenerated.
Targets that lack ppc64 support were failing the new prefix-reloc
test. This patch adds some test infrastructure to deal with that, and
changes the powerpc gas usage info so that "-a64" is omitted when
unsupported.
I've been meaning to break up the usage message for a long time;
While doing so causes translators some work now, it should make it
easier next time a new powerpc option is added.
* config/tc-ppc.c (is_ppc64_target): New function.
(md_show_usage): Split up usage message. Don't show -a64 when
unsupported.
testsuite/gas/ppc/ppc.exp (supports_ppc64): New.
(prefix-reloc): Only run for ppc64.
Allow st_other values such as STO_AARCH64_VARIANT_PCS to be set for alias
symbols independently. This is needed for ifunc symbols which are
aliased to the resolver using .set and don't expect resolver attributes
to override the ifunc symbol attributes. This means .variant_pcs must be
added explicitly to aliases.
gas/ChangeLog:
* config/tc-aarch64.c (aarch64_elf_copy_symbol_attributes): Define.
* config/tc-aarch64.h (aarch64_elf_copy_symbol_attributes): Declare.
(OBJ_COPY_SYMBOL_ATTRIBUTES): Define.
* testsuite/gas/aarch64/symbol-variant_pcs-3.d: New test.
* testsuite/gas/aarch64/symbol-variant_pcs-3.s: New test.
In ELF objects the specified symbol is marked with STO_AARCH64_VARIANT_PCS.
gas/ChangeLog:
* config/tc-aarch64.c (s_variant_pcs): New function.
* doc/c-aarch64.texi: Document .variant_pcs.
* testsuite/gas/aarch64/symbol-variant_pcs-1.d: New test.
* testsuite/gas/aarch64/symbol-variant_pcs-1.s: New test.
* testsuite/gas/aarch64/symbol-variant_pcs-2.d: New test.
* testsuite/gas/aarch64/symbol-variant_pcs-2.s: New test.
