All glibc malloc() implementations I've checked have a smallest
allocation size worth of 3 pointers, with an increment worth of 2
pointers. Hence mnemonics with multiple templates can be stored more
efficiently when maintaining the shared "name" field only in the actual
hash entry. (To express the shared nature, also convert "name" to by
pointer-to-const.)
While doing the conversation also pull out common code from the involved
if/else construct in expand_templates().
Register names are (including their nul terminators) on average almost 4
bytes long. Otoh no register name is longer than 8 bytes. Hence even for
32-bit builds using a pointer is only slightly more space efficient than
embedding the strings. A level of indirection can be also avoided by
embedding the names as an array of 8 characters directly in the arrays,
and the number of base relocations in libopcodes.so (or PIE builds of
statically linked executables) goes down as well.
To amortize for the otherwise reduced folding of string literals by the
linker, use att_names_seg[] in place of string literals in append_seg()
and OP_ESreg().
Register names are (including their nul terminators) on average almost 4
bytes long. Otoh no register name is longer than 7 bytes. Hence even for
32-bit builds using a pointer is only slightly more space efficient than
embedding the strings. A level of indirection can be also avoided by
embedding the names as an array of 8 characters directly in the struct,
and the number of base relocations in PIE builds of gas goes down as
well.
When generating the mnemonic string table we already set up an
identifier for the following entry in a number of cases. Re-use that on
the next loop iteration rather than re-doing allocation and conversion.
Compact the mnemonic string table such that the tails of longer
mnemonics are re-used for shorter ones, going beyond what compilers
would typically do, but matching what ELF linkers may do when processing
SHF_MERGE|SHF_STRINGS sections. This reduces table size by about 12.5%.
Using full pointers to reference the insn mnemonic strings is not very
efficient. With overall string size presently just slightly over 20k,
even a 16-bit value would suffice. Use "unsigned int" for now, as
there's no good use we could presently make of the otherwise saved 16
bits.
For 64-bit builds this reduces table size by 6.25% (prior to the recent
ISA extension additions it would have been 12.5%), with a similar effect
on cache occupation of table entries accessed. For PIE builds of gas
this also reduces the number of base relocations quite a bit (obviously
independent of bitness).
The newer update-copyright.py fixes file encoding too, removing cr/lf
on binutils/bfdtest2.c and ld/testsuite/ld-cygwin/exe-export.exp, and
embedded cr in binutils/testsuite/binutils-all/ar.exp string match.
TSXLDTRK takes RTM as a prereq. Additionally introduce an umbrella "tsx"
extension option covering both RTM and HLE, paralleling the "abm" one we
already have.
SEV-ES is an extension to SVME. SNP in turn is an extension to SEV-ES,
and yet in turn RMPQUERY is a SNP extension.
Note that cpu_arch[] has no SNP entry, so CPU_ANY_SNP_FLAGS remains
unused (just like CPU_SNP_FLAGS already is).
Like various other features AMX-TILE takes XSAVE as a prereq.
XSAVES, unconditionally using compacted format, in turn effectively
takes XSAVEC as a prereq (an SDM clarification to this effect is in the
works).
Like AVX512-FP16, several other extensions require wider than 16-bit
mask registers. As a result they take AVX512BW as a prereq, not (just)
AVX512F. Which in turn points out wrong expectations in the noavx512-1
testcase.
SSE itself takes FXSR as a prereq. Like AES, PCLMUL, and SHA both GFNI
and KL take SSE2 as a prereq, for operating on packed integers. And
while correcting KL also record it as a prereq to WIDEKL.
Getting both forward and reverse ISA dependencies right / consistent has
been a permanent source of mistakes. Reduce what needs specifying
manually to just the direct forward dependencies. Transitive forward
dependencies as well as reverse ones are now derived and hence cannot go
out of sync anymore (at least in the vast majority of cases; there are a
few special cases to still take care of manually). In the course of this
several CPU_ANY_*_FLAGS disappear, requiring adjustment to the
assembler's cpu_arch[].
Note that to retain the correct reverse dependency of AVX512F wrt
AVX512-VP2INTERSECT, the latter has the previously missing AVX512F
prereq added.
Note further that to avoid adding the following undue prereqs:
* ATHLON, K8, and AMDFAM10 gain CMOV and FXSR,
* IAMCU gains 387,
auxiliary table entries (including a colon-separated modifier) are
introduced in addition to the ones representing from converting the old
table.
To maintain forward-only dependencies between AVX (XOP) and SSE* (SSE4a)
(i.e. "nosse" not disabling AVX), reverse dependency tracking is
artifically suppressed.
As a side effect disabling of SSE or SSE2 will now also disable AES,
PCLMUL, and SHA (respective elements were missing from
CPU_ANY_SSE2_FLAGS).
While originally indeed used for register size checking only, the
attribute has been used for memory operand size checking as well already
for quite a while, with more such uses recently having been added.
Having a "None" field in the vast majority of entries is needlessly
cluttering the overall table. Instead of this being a separate field,
use a representation matching that of Intel SDM and AMD PM for the main
use of the field: Append the value after a / as the separator.
PR gas/29524
Having templates with a suffix explicitly present has always been
quirky. After prior adjustment all that's left to also eliminate the
anomaly from move-with-sign-extend is to consolidate the insn templates
and to make may_need_pass2() cope (plus extend testsuite coverage).
The need for them on the operand-less string insns has gone away with
the removal of maybe_adjust_templates() and associated logic. Since
i386_index_check() needs adjustment then anyway, take the opportunity
and also simplify it, possible again as a result of said removal (plus
the opcode template adjustments done here).
Having templates with a suffix explicitly present has always been
quirky. Introduce a 2nd matching pass in case the 1st one couldn't find
a suitable template _and_ didn't itself already need to trim off a
suffix to find a match at all. This requires error reporting adjustments
(albeit luckily fewer than I was afraid might be necessary), as errors
previously reported during matching now need deferring until after the
2nd pass (because, obviously, we must not emit any error if the 2nd pass
succeeds). While also related to PR gas/29524, it was requested that
move-with-sign-extend be left as broken as it always was.
PR gas/29525
Note that with the dropped CMPSD and MOVSD Intel Syntax string insn
templates taking operands, mixed IsString/non-IsString template groups
(with memory operands) cannot occur anymore. With that
maybe_adjust_templates() becomes unnecessary (and is hence being
removed).
PR gas/29526
Note further that while the additions to the intel16 testcase aren't
really proper Intel syntax, we've been permitting all of those except
for the MOVD variant. The test therefore is to avoid re-introducing such
an inconsistency.
This reverts the disassembler parts of 859aa2c86d ("x86: Allow 16-bit
register source for LAR and LSL"), adjusting testcases as necessary.
That change was itself a partial revert of c9f5b96bda ("x86: correct
handling of LAR and LSL"), without actually saying so. While the earlier
commit was properly agreed upon, the partial revert was not, and hence
should not have been committed. This is even more so that the revert
part of that change wasn't even necessary to address PR gas/29844.
Speed up gas startup by avoiding runtime allocation of the instances of
type "templates". At the same time cut the memory requirement to just
very little over half (not even accounting for any overhead
notes_alloc() may incur) by reusing the "end" slot of a preceding entry
for the "start" slot of the subsequent one.
Now that the table is local to gas, ARRAY_SIZE() can be used to
determine the end of the table. Re-arrange the processing loop in
md_begin() accordingly, at the same time folding the two calls to
notes_alloc() into just one.
Remove the now empty i386-opc.c. To compensate, tie table generation in
opcodes/ to the building of i386-dis.o, despite the file not really
depending on the generated data.
Unlike many other architectures, x86 does not share an opcode table
between assembly and disassembly. Any consumer of libopcodes would only
ever access one of the two. Since gas is the only consumer of the
assembly data, move it there. While doing so mark respective entities
"static" in i386-gen (we may want to do away with i386_regtab_size
altogether).
This also shrinks the number of relocations to be processed for
libopcodes.so by about 30%.
For all the xh_mode usage in table, they are all using %XH, which will
print "{bad}" while EVEX.W=1. This makes this vex.w check unnecessary.
opcodes/ChangeLog:
* i386-dis.c (OP_E_memory): Remove vex.w check for xh_mode.
This commit adds disassembler styling support for MIPS. After this
commit objdump and GDB will style disassembler output.
This is a pretty straight forward change, we switch to use the
disassemble_info::fprintf_styled_func callback, and pass an
appropriate style through as needed. No additional tricks were
needed (compared to say i386, or ARM).
Tested by running all of the objdump commands used by the gas
testsuite and manually inspecting the styled output, everything looks
reasonable, though I'm not a MIPS expert, so it is possible that I've
missed some corner cases. Worst case though is that something will be
styled incorrectly, the actual content should be unchanged.
All the gas, ld, and binutils tests still pass for me.
While working on disassembler styling for MIPS, I noticed that
undefined instructions are printed by the disassembler as raw number
with no assembler directive prefix (e.g. without .word or .short).
I think adding something like .word, or .short, helps to make it
clearer the size of the value that is being displayed, and is inline
with what many of the other libopcode disassemblers do.
In this commit I've added the .word and .short directives, and updated
all the tests that I spotted that failed as a result.
Since LAR and LSL only access 16 bits of the source operand, regardless
of operand size, allow 16-bit register source for LAR and LSL, and always
disassemble LAR and LSL with 16-bit source operand.
gas/
PR gas/29844
* testsuite/gas/i386/i386.s: Add tests for LAR and LSL.
* testsuite/gas/i386/x86_64.s: Likewise.
* testsuite/gas/i386/intelbad.s: Remove "lar/lsl eax, ax".
* testsuite/gas/i386/i386-intel.d: Updated.
* testsuite/gas/i386/i386.d: Likewise.
* testsuite/gas/i386/intel-intel.d: Likewise.
* testsuite/gas/i386/intel.d: Likewise.
* testsuite/gas/i386/intelbad.l: Likewise.
* testsuite/gas/i386/x86_64-intel.d: Likewise.
* testsuite/gas/i386/x86_64.d: Likewise.
opcodes/
PR gas/29844
* i386-dis.c (MOD_0F02): Removed.
(MOD_0F03): Likewise.
(dis386_twobyte): Restore larS and lslS.
(mod_table): Remove MOD_0F02 and MOD_0F03.
* i386-opc.tbl: Allow 16-bit register source for LAR and LSL.
* i386-tbl.h: Regenerated.
With the general use of C99 there's no need anymore to have i386-gen
produce these. For more frequently used ones introduce local #define-s,
while others are simply spelled out directly. While doing this move
some static constants into more narrow scopes.
Note that as a "side effect" this corrects type_names[]'es imm8s entry.
Leverage the C (commutative) attribute to also reduce the number of XCHG
and TEST templates we have. This way the reg <-> r/m (and reg <-> reg for
XCHG) forms can also be folded into a single template each, utilizing D.
i386-init.h and i386-tbl.h are generated files. There is nothing to
translate. Remove them from HFILES (POTFILES).
* Makefile.am (HFILES): Remove i386-init.h and i386-tbl.h.
* Makefile.in: Regenerated.
* po/POTFILES.in: Likewise.
With the removal of its use for FPU insns the suffix is now finally
properly misnamed. Drop its use altogether, replacing it by a separate
boolean instead.
As a comment near the top of match_template() already says: We really
only need this pseudo-suffix for far branch handling. Stop "deriving" it
for floating point insns. (Don't bother renaming the now properly
misnamed LONG_DOUBLE_MNEM_SUFFIX, to e.g. FAR_BRANCH_SUFFIX - it's going
to disappear anyway.)
At the very least a comment in process_operands() is stale. Beyond that
there are effectively two options:
1) It is possible that FADDP and FMULP were mistakenly not marked as
being in need of dealing with the compiler anomaly, and hence the
respective templates weren't removed at the time when they should
have been.
2) It is also possible that there are indeed uses known beyond compiler
generated output for these two commutative opcodes, and hence the
templates need to stay.
To be on the safe side assume 2: Update the comment and fold the
templates into their "normal" ones (utilizing D), adjusting consuming
code accordingly.
For FMULP also add a comment paralleling a similar one FADDP has.
There are just 4 templates using it, which can be easily identified by
other means, as D is set only on a very limited number of FPU templates.
Also move the respective conditional out of the code path taken by all
"reverse match" insns (it probably should have been this way already
before, to avoid the one conditional in the common case).
With this the templates which had FloatR dropped no longer differ from
their AT&T syntax + mnemonic counterparts - the only difference is now
which of the two would be recognized. For this, however, we don't need
two templates - we can simply arrange the condition for setting
Opcode_FloatR accordingly.
Commit bb996692bd ("RISC-V/gas: allow generating up to 176-bit
instructions with .insn") tried to start supporting long instructions but
it was insufficient.
On the disassembler, correct ".byte" output was limited to the first 64-bits
of an instruction. After that, zeroes are incorrectly printed.
Note that, it only happens on ".byte" output (instruction part) and not on
hexdump (data) part. For example, before this commit, hexdump and ".byte"
produces different values:
Assembly:
.insn 22, 0xfedcba98765432100123456789abcdef55aa33cc607f
objdump output example (before the fix):
10: 607f 33cc 55aa cdef .byte 0x7f, 0x60, 0xcc, 0x33, 0xaa, 0x55, 0xef, 0xcd, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
18: 89ab 4567 0123 3210
20: 7654 ba98 fedc
Note that, after 0xcd (after first 64-bits of the target instruction), all
".byte" values are incorrectly printed as zero while hexdump prints correct
instruction bits.
To resolve this, this commit adds "packet" argument to support dumping
instructions longer than 64-bits (to print correct instruction bits on
".byte"). This commit will be tested on the separate commit.
Assembly:
.insn 22, 0xfedcba98765432100123456789abcdef55aa33cc607f
objdump output example (after the fix):
10: 607f 33cc 55aa cdef .byte 0x7f, 0x60, 0xcc, 0x33, 0xaa, 0x55, 0xef, 0xcd, 0xab, 0x89, 0x67, 0x45, 0x23, 0x01, 0x10, 0x32, 0x54, 0x76, 0x98, 0xba, 0xdc, 0xfe
18: 89ab 4567 0123 3210
20: 7654 ba98 fedc
opcodes/ChangeLog:
* riscv-dis.c (riscv_disassemble_insn): Print unknown instruction
using the new argument packet.
(riscv_disassemble_data): Add unused argument packet.
(print_insn_riscv): Pass packet to the disassemble function.
First of all make operand_type_register_match() apply to all sized
operands, i.e. in Intel Syntax also to respective memory ones. This
addresses gas wrongly accepting certain SIMD insns where register and
memory operand sizes should match but don't. This apparently has
affected all templates with one memory-only operand and one or more
register ones, both permitting at least two sizes, due to CheckRegSize
not taking effect.
Then also add CheckRegSize to a couple of non-SIMD templates matching
that same pattern of memory-only vs register operands. This replaces
bogus (for Intel Syntax) diagnostics referring to a wrong suffix (when
none was used at all) by "type mismatch" ones, just like already emitted
for insns where the template allows a register operand alongside a
memory one at any particular position.
This also is a prereq to limiting (ideally eliminating in the long run)
suffix "derivation" in Intel Syntax mode.
While making the code adjustment also flip order of checks to do the
cheaper one first in both cases.
To properly and predictably determine operand size encoding (operand
size or REX.W prefixes), consistent operand sizes need to be specified.
Add CheckRegSize where this was previously missing.
Both uniformly only ever take 16-bit memory operands while at the same
time requiring matching (in size) register operands, which then also
should disassemble that way. This in particular requires splitting each
of the templates for the assembler and separating decode of the
register and memory forms in the disassembler.
Jan Beulich
Max Filippov
Nick Clifton
Alan Modra
Peter Bergner
Haochen Jiang
Andrew Burgess
H.J. Lu
Tsukasa OI