The commit d03b174a83 (target/ppc: simplify bcdadd/sub functions)
meant to simplify some of the code but it inadvertently altered the
way the CR6 field is set after the operation has overflowed.
The CR6 bits are set based on the *unbounded* result of the operation,
so we need to look at the result before returning from bcd_add_mag,
otherwise we will look at 0 when it overflows.
Consider the following subtraction:
v0 = 0x9999999999999999999999999999999c (maximum positive BCD value)
v1 = 0x0000000000000000000000000000001d (negative one BCD value)
bcdsub. v0,v0,v1,0
The Power ISA 2.07B says:
If the unbounded result is greater than zero, do the following.
If PS=0, the sign code of the result is set to 0b1100.
If PS=1, the sign code of the result is set to 0b1111.
If the operation overflows, CR field 6 is set to 0b0101. Otherwise,
CR field 6 is set to 0b0100.
POWER9 hardware:
vr0 = 0x0000000000000000000000000000000c (positive zero BCD value)
cr6 = 0b0101 (0x5) (positive, overflow)
QEMU:
vr0 = 0x0000000000000000000000000000000c (positive zero BCD value)
cr6 = 0b0011 (0x3) (zero, overflow) <--- wrong
This patch reverts the part of d03b174a83 that introduced the
problem and adds a test-case to avoid further regressions:
before:
$ make run-tcg-tests-ppc64le-linux-user
(...)
TEST bcdsub on ppc64le
bcdsub: qemu/tests/tcg/ppc64le/bcdsub.c:58: test_bcdsub_gt:
Assertion `(cr >> 4) == ((1 << 2) | (1 << 0))' failed.
Fixes: d03b174a83 (target/ppc: simplify bcdadd/sub functions)
Reported-by: Paul Clarke <pc@us.ibm.com>
Signed-off-by: Fabiano Rosas <farosas@linux.ibm.com>
Message-Id: <20210222194035.2723056-1-farosas@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Blink and you miss the cross TCG compiler stuff so lets display it
with the rest of the compiler information.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210222101455.12640-2-alex.bennee@linaro.org>
Signed-off-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <1612763186-18161-34-git-send-email-tsimpson@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Taylor Simpson <tsimpson@quicinc.com>
Message-Id: <1612763186-18161-33-git-send-email-tsimpson@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Enable multiarch tests for Hexagon
Modify tests/tcg/configure.sh
Add reference files to tests/tcg/hexagon
Signed-off-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <1612763186-18161-32-git-send-email-tsimpson@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
A duplicate insn is one that is appears to be executed twice in a row.
This is currently possible due to -icount and cpu_io_recompile()
causing a re-translation of a block. On it's own this won't trigger
any tests though.
The heuristics that the plugin use can't deal with the x86 rep
instruction which (validly) will look like executing the same
instruction several times. To avoid problems later we tweak the rules
for x86 to run the "inline" version of the plugin. This also has the
advantage of increasing coverage of the plugin code (see bugfix in
previous commit).
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210213130325.14781-15-alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210212184902.1251044-32-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
We were fudging TBI1 enabled to speed up the generated code.
Now that we've improved the code generation, remove this.
Also, tidy the comment to reflect the current code.
The pauth test was testing a kernel address (-1) and making
incorrect assumptions about TBI1; stick to userland addresses.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210212184902.1251044-23-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
/bin/true is missing on macOS, but simply "true" is available as a shell builtin.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20210128135627.2067003-1-sw@weilnetz.de>
Message-Id: <20210202134001.25738-12-alex.bennee@linaro.org>
Otherwise people won't know what they are missing.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20210202134001.25738-10-alex.bennee@linaro.org>
The current EXRL tests crash on real machines: we must not use r0 as a base
register for trt/trtr, otherwise the content gets ignored. Also, we must
not use r0 for exrl, otherwise it gets ignored.
Let's use the "a" constraint so we get a general purpose register != r0.
For op2, we can simply specify a memory operand directly via "Q" (Memory
reference without index register and with short displacement).
Fixes: ad8c851d2e ("target/s390x: add EX support for TRT and TRTR")
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210111163845.18148-5-david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
As per POSIX specification of limits.h [1], OS libc may define
PAGE_SIZE in limits.h.
Self defined PAGE_SIZE is frequently used in tests, to prevent
collosion of definition, we give PAGE_SIZE definitons reasonable
prefixs.
[1]: https://pubs.opengroup.org/onlinepubs/7908799/xsh/limits.h.html
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210118063808.12471-7-jiaxun.yang@flygoat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
While GDB can work with any XML description given to it there is
special handling for SVE registers on the GDB side which makes the
users life a little better. The changes aren't that major and all the
registers save the $vg reported the same. All that changes is:
- report org.gnu.gdb.aarch64.sve
- use gdb nomenclature for names and types
- minor re-ordering of the types to match reference
- re-enable ieee_half (as we know gdb supports it now)
- $vg is now a 64 bit int
- check $vN and $zN aliasing in test
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Luis Machado <luis.machado@linaro.org>
Message-Id: <20210108224256.2321-11-alex.bennee@linaro.org>
This allows gdb to access the target’s auxiliary vector,
which can be helpful for telling system libraries important details
about the hardware, operating system, and process.
Signed-off-by: Lirong Yuan <yuanzi@google.com>
[AJB: minor tweaks to test case, update MAINTAINERS, restrict to Linux]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20200730193932.3654677-1-yuanzi@google.com>
Message-Id: <20210108224256.2321-7-alex.bennee@linaro.org>
This adds a new tests that allows us to test softmmu only features
including watchpoints. To do achieve this we need to:
- add _exit: labels to the boot codes
- write a memory.py test case
- plumb the test case into the build system
- tweak the run_test script to:
- re-direct output when asked
- use socket based connection for all tests
- add a small pause before connection
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20210108224256.2321-6-alex.bennee@linaro.org>
We won't attempt to run the test now it's gated on a newer version of
gdb.
This reverts commit a930cadd83.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20210108224256.2321-5-alex.bennee@linaro.org>
Hopefully this will guard against sloppy code getting into our tests.
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20201210190417.31673-9-alex.bennee@linaro.org>
Since commit efc6c07 ("configure: Add a test for the minimum compiler
version"), QEMU explicitely depends on GCC >= 4.8.
(clang >= 3.4 advertizes itself as GCC >= 4.2 compatible)
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20201210134752.780923-6-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Disable the run-gdbstub-sha1 test: it provokes an internal error
assertion failure in Ubuntu gdb 8.1.1-0ubuntu1 (Ubuntu gdb
8.1-0ubuntu3.2 also has this assert but we were previously skipping
this test because it doesn't support connection over local domain
sockets) :
timeout 60 /home/petmay01/linaro/qemu-for-merges/tests/guest-debug/run-test.py --gdb /usr/bin/gdb-multiar
/build/gdb-veKdC1/gdb-8.1.1/gdb/regcache.c:122: internal-error: void* init_regcache_descr(gdbarch*): Asser
A problem internal to GDB has been detected,
further debugging may prove unreliable.
This is a bug, please report it. For instructions, see:
<http://www.gnu.org/software/gdb/bugs/>.
Aborted (core dumped)
/home/petmay01/linaro/qemu-for-merges/tests/tcg/multiarch/Makefile.target:51: recipe for target 'run-gdbst
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20201214133702.24088-1-peter.maydell@linaro.org
For PDEP and PEXT, the mask is provided in the memory (mod+r/m)
operand, and therefore is loaded in s->T0 by gen_ldst_modrm.
The source is provided in the second source operand (VEX.vvvv)
and therefore is loaded in s->T1. Fix the order in which
they are passed to the helpers.
Reported-by: Lenard Szolnoki <blog@lenardszolnoki.com>
Analyzed-by: Lenard Szolnoki <blog@lenardszolnoki.com>
Fixes: https://bugs.launchpad.net/qemu/+bug/1605123
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The note test requires gcc 10 for -mbranch-protection=standard.
The mmap test uses PROT_BTI and does not require special compiler support.
Acked-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20201021173749.111103-13-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Unlike glibc, musl does not use transparent unions to hide
the different structures that overlap struct sockaddr.
Add an explicit cast to work around this.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Since we are now always doing out-of-tree builds, these gitignore
files should not be necessary anymore.
Message-Id: <20200919133637.72744-1-thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Commit 660f793093 was a bit overzealous
with respect to tests/tcg, which needed quiet-command and $(BUILD_DIR).
Reinstate quiet-command, and replace $(BUILD_DIR) with just the
current directory.
Reported-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We have an exploding complexity problem in the testing so lets just
move the more involved plugins into contrib. tests/plugins still exist
for the basic plugins that exercise the API. We restore the old
pre-meson style Makefile for contrib as it also doubles as a guide for
out-of-tree plugin builds.
While we are at it add some examples to the documentation and a
specific plugins build target.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200909112742.25730-11-alex.bennee@linaro.org>
Define anything that is missing as 0, so that flags & FE_FOO
is false for any missing FOO.
Tested-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This is optional in ISO C, and not all cpus provide it.
Tested-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Not attempting to use a single cross-compiler for both
big-endian and little-endian at this time.
Tested-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
DFPU doesn't have pre-increment FP load/store opcodes, it has
post-increment opcodes instead. Test increment opcodes present in the
current config.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
DFPU sets Invalid flag in FSR when at least one argument of FP
comparison opcodes is NaN, SNaN for most opcodes, any NaN for olt/ole.
Add checks for FSR and expected FSR values.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Test that madd doesn't do rounding after multiplication.
Test NaN propagation rules for FPU2000 and DFPU madd opcode.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Space for test results may be allocated in IRAM which is only
word-accessible. Use full 32-bit words to access test results.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
The binaries move to the root directory, e.g. qemu-system-i386 or
qemu-arm. This requires changes to qtests, CI, etc.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The definition of top_bit used in this function is one higher
than that used in the Arm ARM psuedo-code, which put the error
indication at top_bit - 1 at the wrong place, which meant that
it wasn't visible to Auth.
Fixing the definition of top_bit requires more changes, because
its most common use is for the count of bits in top_bit:bot_bit,
which would then need to be computed as top_bit - bot_bit + 1.
For now, prefer the minimal fix to the error indication alone.
Fixes: 63ff0ca94c
Reported-by: Derrick McKee <derrick.mckee@gmail.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200728195706.11087-1-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
[PMM: added comment about the divergence from the pseudocode]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
- tests/vm support for aarch64 VMs
- tests/tcg better cross-compiler detection
- update docker tooling to support registries
- update docker support for xtensa
- gitlab build docker images and store in registry
- gitlab use docker images for builds
- a number of skipIf updates to support move
- linux-user MAP_FIXED_NOREPLACE fix
- qht-bench compiler tweaks
- configure fix for secret keyring
- tsan fiber annotation clean-up
- doc updates for mttcg/icount/gdbstub
- fix cirrus to use brew bash for iotests
- revert virtio-gpu breakage
- fix LC_ALL to avoid sorting changes in iotests
-----BEGIN PGP SIGNATURE-----
iQEzBAABCgAdFiEEZoWumedRZ7yvyN81+9DbCVqeKkQFAl8J0yoACgkQ+9DbCVqe
KkSzTAf/Vn+9TU8Qt7nZvl7W4tz7Sy5K8EJGwj2RXx6CWWWLiFbsXurIM8Krw5Vc
RmvUxwa359b+J0lQpfeNDHYm1nM8RZLFlkG0a5bl0I8sW0EcPjBRtwNaGKXh2p0u
u2RS2QAi6A9AvYT4ZREYlBM+o9WzbxCEQm4s8fr6WEJCQfxBnb5/bGiEjWR64e8C
j9Kvou+zAKfVizbQMtu+mwqjsoPtcS1b3vVcO7anhNuUsuaEKkS0dFWzWvw3lwJR
STIYnb8Y/eJ1yKr0hPH2qtWv3n6yhlYvYmpUCH6AwshGMUoeFEzR2VoWS6yZPGG6
na6XA3UW5R9AxIDfkCJ5ueeo8t9xMQ==
=HRWa
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/stsquad/tags/pull-testing-and-misc-110720-2' into staging
Testing and misc build updates:
- tests/vm support for aarch64 VMs
- tests/tcg better cross-compiler detection
- update docker tooling to support registries
- update docker support for xtensa
- gitlab build docker images and store in registry
- gitlab use docker images for builds
- a number of skipIf updates to support move
- linux-user MAP_FIXED_NOREPLACE fix
- qht-bench compiler tweaks
- configure fix for secret keyring
- tsan fiber annotation clean-up
- doc updates for mttcg/icount/gdbstub
- fix cirrus to use brew bash for iotests
- revert virtio-gpu breakage
- fix LC_ALL to avoid sorting changes in iotests
# gpg: Signature made Sat 11 Jul 2020 15:56:42 BST
# gpg: using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <alex.bennee@linaro.org>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8 DF35 FBD0 DB09 5A9E 2A44
* remotes/stsquad/tags/pull-testing-and-misc-110720-2: (50 commits)
iotests: Set LC_ALL=C for sort
Revert "vga: build virtio-gpu as module"
tests: fix "make check-qtest" for modular builds
.cirrus.yml: add bash to the brew packages
tests/docker: update toolchain set in debian-xtensa-cross
tests/docker: fall back more gracefully when pull fails
docs: Add to gdbstub documentation the PhyMemMode
docs/devel: add some notes on tcg-icount for developers
docs/devel: convert and update MTTCG design document
tests/qht-bench: Adjust threshold computation
tests/qht-bench: Adjust testing rate by -1
travis.yml: Test also the other targets on s390x
shippable: pull images from registry instead of building
testing: add check-build target
containers.yml: build with docker.py tooling
gitlab: limit re-builds of the containers
tests: improve performance of device-introspect-test
gitlab: add avocado asset caching
gitlab: enable check-tcg for linux-user tests
linux-user/elfload: use MAP_FIXED_NOREPLACE in pgb_reserved_va
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Switch to the prebuilt xtensa toolchains release 2020.07.
Drop csp toolchain as the csp core is not a part of QEMU.
Add de233_fpu and dsp3400 toolchains to enable DFPU and FPU2000 tests.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
[AJB: fix path in configure.sh]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200708082347.27318-1-jcmvbkbc@gmail.com>
Message-Id: <20200709141327.14631-13-alex.bennee@linaro.org>
We were missing a bunch of compilers which we could use if they were
locally installed. The defaults are based on Debian as they seem to be
the best distro for well distributed cross-build compilers.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200701135652.1366-32-alex.bennee@linaro.org>
We've been misusing the tag naming scheme for some time by overloading
the post : section with the image type. Really it should be saved for
the revision of that particular build. Move the details to the other
side so we have:
qemu/image-name
with the implied :latest version added by the tooling.
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200701135652.1366-18-alex.bennee@linaro.org>
The SSE instruction implementations all fail to raise the expected
IEEE floating-point exceptions because they do nothing to convert the
exception state from the softfloat machinery into the exception flags
in MXCSR.
Fix this by adding such conversions. Unlike for x87, emulated SSE
floating-point operations might be optimized using hardware floating
point on the host, and so a different approach is taken that is
compatible with such optimizations. The required invariant is that
all exceptions set in env->sse_status (other than "denormal operand",
for which the SSE semantics are different from those in the softfloat
code) are ones that are set in the MXCSR; the emulated MXCSR is
updated lazily when code reads MXCSR, while when code sets MXCSR, the
exceptions in env->sse_status are set accordingly.
A few instructions do not raise all the exceptions that would be
raised by the softfloat code, and those instructions are made to save
and restore the softfloat exception state accordingly.
Nothing is done about "denormal operand"; setting that (only for the
case when input denormals are *not* flushed to zero, the opposite of
the logic in the softfloat code for such an exception) will require
custom code for relevant instructions, or else architecture-specific
conditionals in the softfloat code for when to set such an exception
together with custom code for various SSE conversion and rounding
instructions that do not set that exception.
Nothing is done about trapping exceptions (for which there is minimal
and largely broken support in QEMU's emulation in the x87 case and no
support at all in the SSE case).
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006252358000.3832@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The x87 fpatan emulation is currently based around conversion to
double. This is inherently unsuitable for a good emulation of any
floatx80 operation. Reimplement using the soft-float operations, as
for other such instructions.
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006230000340.24721@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The x87 fyl2x emulation is currently based around conversion to
double. This is inherently unsuitable for a good emulation of any
floatx80 operation. Reimplement using the soft-float operations,
building on top of the reimplementation of fyl2xp1 and factoring out
code to be shared between the two instructions.
The included test assumes that the result in round-to-nearest mode
should always be one of the two closest floating-point numbers to the
mathematically exact result (including that it should be exact, in the
exact cases which cover more cases than for fyl2xp1).
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006172321530.20587@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The x87 fyl2xp1 emulation is currently based around conversion to
double. This is inherently unsuitable for a good emulation of any
floatx80 operation, even before considering that it is a particularly
naive implementation using double (adding 1 then using log rather than
attempting a better emulation using log1p).
Reimplement using the soft-float operations, as was done for f2xm1; as
in that case, m68k has related operations but not exactly this one and
it seemed safest to implement directly rather than reusing the m68k
code to avoid accumulation of errors.
A test is included with many randomly generated inputs. The
assumption of the test is that the result in round-to-nearest mode
should always be one of the two closest floating-point numbers to the
mathematical value of y * log2(x + 1); the implementation aims to do
somewhat better than that (about 70 correct bits before rounding). I
haven't investigated how accurate hardware is.
Intel manuals describe a narrower range of valid arguments to this
instruction than AMD manuals. The implementation accepts the wider
range (it's needed anyway for the core code to be reusable in a
subsequent patch reimplementing fyl2x), but the test only has inputs
in the narrower range so that it's valid on hardware that may reject
or produce poor results for inputs outside that range.
Code in the previous implementation that sets C2 for some out-of-range
arguments is not carried forward to the new implementation; C2 is
undefined for this instruction and I suspect that code was just
cut-and-pasted from the trigonometric instructions (fcos, fptan, fsin,
fsincos) where C2 *is* defined to be set for out-of-range arguments.
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006172320190.20587@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The x87 f2xm1 emulation is currently based around conversion to
double. This is inherently unsuitable for a good emulation of any
floatx80 operation, even before considering that it is a particularly
naive implementation using double (computing with pow and then
subtracting 1 rather than attempting a better emulation using expm1).
Reimplement using the soft-float operations, including additions and
multiplications with higher precision where appropriate to limit
accumulation of errors. I considered reusing some of the m68k code
for transcendental operations, but the instructions don't generally
correspond exactly to x87 operations (for example, m68k has 2^x and
e^x - 1, but not 2^x - 1); to avoid possible accumulation of errors
from applying multiple such operations each rounding to floatx80
precision, I wrote a direct implementation of 2^x - 1 instead. It
would be possible in principle to make the implementation more
efficient by doing the intermediate operations directly with
significands, signs and exponents and not packing / unpacking floatx80
format for each operation, but that would make it significantly more
complicated and it's not clear that's worthwhile; the m68k emulation
doesn't try to do that.
A test is included with many randomly generated inputs. The
assumption of the test is that the result in round-to-nearest mode
should always be one of the two closest floating-point numbers to the
mathematical value of 2^x - 1; the implementation aims to do somewhat
better than that (about 70 correct bits before rounding). I haven't
investigated how accurate hardware is.
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2006112341010.18393@digraph.polyomino.org.uk>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
