According to Cortex-R5 r1p2 manual, register with opcode2=0 is
BTCM and with opcode2=1 is ATCM, - exactly the opposite from how
qemu labels them. Just swap the labels to avoid confusion, -
both registers are implemented as always-zero.
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241121171602.3273252-1-mjt@tls.msk.ru
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This path is reachable with plugins enabled, and provoked
with run-plugin-catch-syscalls-with-libinline.so.
Cc: qemu-stable@nongnu.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20241112141232.321354-1-richard.henderson@linaro.org>
FEAT_CMOW introduces support for controlling cache maintenance
instructions executed in EL0/1 and is mandatory from Armv8.8.
On real hardware, the main use for this feature is to prevent processes
from invalidating or flushing cache lines for addresses they only have
read permission, which can impact the performance of other processes.
QEMU implements all cache instructions as NOPs, and, according to rule
[1], which states that generating any Permission fault when a cache
instruction is implemented as a NOP is implementation-defined, no
Permission fault is generated for any cache instruction when it lacks
read and write permissions.
QEMU does not model any cache topology, so the PoU and PoC are before
any cache, and rules [2] apply. These rules state that generating any
MMU fault for cache instructions in this topology is also
implementation-defined. Therefore, for FEAT_CMOW, we do not generate any
MMU faults either, instead, we only advertise it in the feature
register.
[1] Rule R_HGLYG of section D8.14.3, Arm ARM K.a.
[2] Rules R_MZTNR and R_DNZYL of section D8.14.3, Arm ARM K.a.
Signed-off-by: Gustavo Romero <gustavo.romero@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241104142606.941638-1-gustavo.romero@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Our implementation of the indexed version of SVE SDOT/UDOT/USDOT got
the calculation of the inner loop terminator wrong. Although we
correctly account for the element size when we calculate the
terminator for the first iteration:
intptr_t segend = MIN(16 / sizeof(TYPED), opr_sz_n);
we don't do that when we move it forward after the first inner loop
completes. The intention is that we process the vector in 128-bit
segments, which for a 64-bit element size should mean (1, 2), (3, 4),
(5, 6), etc. This bug meant that we would iterate (1, 2), (3, 4, 5,
6), (7, 8, 9, 10) etc and apply the wrong indexed element to some of
the operations, and also index off the end of the vector.
You don't see this bug if the vector length is small enough that we
don't need to iterate the outer loop, i.e. if it is only 128 bits,
or if it is the 64-bit special case from AA32/AA64 AdvSIMD. If the
vector length is 256 bits then we calculate the right results for the
elements in the vector but do index off the end of the vector. Vector
lengths greater than 256 bits see wrong answers. The instructions
that produce 32-bit results behave correctly.
Fix the recalculation of 'segend' for subsequent iterations, and
restore a version of the comment that was lost in the refactor of
commit 7020ffd656 that explains why we only need to clamp segend to
opr_sz_n for the first iteration, not the later ones.
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2595
Fixes: 7020ffd656 ("target/arm: Macroize helper_gvec_{s,u}dot_idx_{b,h}")
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241101185544.2130972-1-peter.maydell@linaro.org
Our current usage of MMU indexes when EL3 is AArch32 is confused.
Architecturally, when EL3 is AArch32, all Secure code runs under the
Secure PL1&0 translation regime:
* code at EL3, which might be Mon, or SVC, or any of the
other privileged modes (PL1)
* code at EL0 (Secure PL0)
This is different from when EL3 is AArch64, in which case EL3 is its
own translation regime, and EL1 and EL0 (whether AArch32 or AArch64)
have their own regime.
We claimed to be mapping Secure PL1 to our ARMMMUIdx_EL3, but didn't
do anything special about Secure PL0, which meant it used the same
ARMMMUIdx_EL10_0 that NonSecure PL0 does. This resulted in a bug
where arm_sctlr() incorrectly picked the NonSecure SCTLR as the
controlling register when in Secure PL0, which meant we were
spuriously generating alignment faults because we were looking at the
wrong SCTLR control bits.
The use of ARMMMUIdx_EL3 for Secure PL1 also resulted in the bug that
we wouldn't honour the PAN bit for Secure PL1, because there's no
equivalent _PAN mmu index for it.
Fix this by adding two new MMU indexes:
* ARMMMUIdx_E30_0 is for Secure PL0
* ARMMMUIdx_E30_3_PAN is for Secure PL1 when PAN is enabled
The existing ARMMMUIdx_E3 is used to mean "Secure PL1 without PAN"
(and would be named ARMMMUIdx_E30_3 in an AArch32-centric scheme).
These extra two indexes bring us up to the maximum of 16 that the
core code can currently support.
This commit:
* adds the new MMU index handling to the various places
where we deal in MMU index values
* adds assertions that we aren't AArch32 EL3 in a couple of
places that currently use the E10 indexes, to document why
they don't also need to handle the E30 indexes
* documents in a comment why regime_has_2_ranges() doesn't need
updating
Notes for backporting: this commit depends on the preceding revert of
4c2c04746932; that revert and this commit should probably be
backported to everywhere that we originally backported 4c2c047469.
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2326
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2588
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241101142845.1712482-3-peter.maydell@linaro.org
This reverts commit 4c2c047469.
This commit tried to fix a problem with our usage of MMU indexes when
EL3 is AArch32, using what it described as a "more complicated
approach" where we share the same MMU index values for Secure PL1&0
and NonSecure PL1&0. In theory this should work, but the change
didn't account for (at least) two things:
(1) The design change means we need to flush the TLBs at any point
where the CPU state flips from one to the other. We already flush
the TLB when SCR.NS is changed, but we don't flush the TLB when we
take an exception from NS PL1&0 into Mon or when we return from Mon
to NS PL1&0, and the commit didn't add any code to do that.
(2) The ATS12NS* address translate instructions allow Mon code (which
is Secure) to do a stage 1+2 page table walk for NS. I thought this
was OK because do_ats_write() does a page table walk which doesn't
use the TLBs, so because it can pass both the MMU index and also an
ARMSecuritySpace argument we can tell the table walk that we want NS
stage1+2, not S. But that means that all the code within the ptw
that needs to find e.g. the regime EL cannot do so only with an
mmu_idx -- all these functions like regime_sctlr(), regime_el(), etc
would need to pass both an mmu_idx and the security_space, so they
can tell whether this is a translation regime controlled by EL1 or
EL3 (and so whether to look at SCTLR.S or SCTLR.NS, etc).
In particular, because regime_el() wasn't updated to look at the
ARMSecuritySpace it would return 1 even when the CPU was in Monitor
mode (and the controlling EL is 3). This meant that page table walks
in Monitor mode would look at the wrong SCTLR, TCR, etc and would
generally fault when they should not.
Rather than trying to make the complicated changes needed to rescue
the design of 4c2c047469, we revert it in order to instead take the
route that that commit describes as "the most straightforward" fix,
where we add new MMU indexes EL30_0, EL30_3, EL30_3_PAN to correspond
to "Secure PL1&0 at PL0", "Secure PL1&0 at PL1", and "Secure PL1&0 at
PL1 with PAN".
This revert will re-expose the "spurious alignment faults in
Secure PL0" issue #2326; we'll fix it again in the next commit.
Cc: qemu-stable@nongnu.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Thomas Huth <thuth@redhat.com>
Message-id: 20241101142845.1712482-2-peter.maydell@linaro.org
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Pass the stage size to step function callback, otherwise do_setm
would hang when size is larger then page size because stage size
would underflow. This fix changes do_setm to be more inline with
do_setp.
Cc: qemu-stable@nongnu.org
Fixes: 0e92818887 ("target/arm: Implement the SET* instructions")
Signed-off-by: Ido Plat <ido.plat1@ibm.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20241025024909.799989-1-ido.plat1@ibm.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Fill in the tlb_fill_align hook. Handle alignment not due to
memory type, since that's no longer handled by generic code.
Pass memop to get_phys_addr.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Zero is the safe do-nothing value for callers to use.
Reviewed-by: Helge Deller <deller@gmx.de>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rename to use "memop_" prefix, like other functions
that operate on MemOp.
Reviewed-by: Helge Deller <deller@gmx.de>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
in many cases, <zlib.h> is only included for crc32 function,
and in some of them, there's a comment saying that, but in
a different way. In one place (hw/net/rtl8139.c), there was
another #include added between the comment and <zlib.h> include.
Make all such comments to be on the same line as #include, make
it consistent, and also add a few missing comments, including
hw/nvram/mac_nvram.c which uses adler32 instead.
There's no code changes.
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
The Neoverse-V1 TRM is a bit confused about the layout of the
ID_AA64ISAR1_EL1 register, and so its table 3-6 has the wrong value
for this ID register. Trust instead section 3.2.74's list of which
fields are set.
This means that we stop incorrectly reporting FEAT_XS as present, and
now report the presence of FEAT_BF16.
Cc: qemu-stable@nongnu.org
Reported-by: Marcin Juszkiewicz <marcin.juszkiewicz@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240917161337.3012188-1-peter.maydell@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-30-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-29-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-28-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
While these functions really do return a 32-bit value,
widening the return type means that we need do less
marshalling between TCG types.
Remove NeonGenNarrowEnvFn typedef; add NeonGenOne64OpEnvFn.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20240912024114.1097832-27-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-26-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This includes SHL and SLI.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-25-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This includes SSHR, USHR, SSRA, USRA, SRSHR, URSHR,
SRSRA, URSRA, SRI.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-24-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-23-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
There isn't a lot of commonality along the different paths of
handle_shri_with_rndacc. Split them out to separate functions,
which will be usable during the decodetree conversion.
Simplify 64-bit rounding operations to not require double-word arithmetic.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-22-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
We always pass the same value for round; compute it
within common code.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-21-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-20-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Combine the right shift with the extension via
the tcg extract operations.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-19-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This includes SHL and SLI.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-18-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This includes SSHR, USHR, SSRA, USRA, SRSHR, URSHR, SRSRA, URSRA, SRI.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-17-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-16-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Handle the two special cases within these new
functions instead of higher in the call stack.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-15-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-14-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-13-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-12-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-11-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Use simple shift and add instead of ctpop, ctz, shift and mask.
Unlike SVE, there is no predicate to disable elements.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-8-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The extract2 tcg op performs the same operation
as the do_ext64 function.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Instead of cmp+and or cmp+andc, use cmpsel. This will
be better for hosts that use predicate registers for cmp.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Instead of cmp+and or cmp+andc, use cmpsel. This will
be better for hosts that use predicate registers for cmp.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-4-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Instead of copying a constant into a temporary with dupi,
use a vector constant directly.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Instead of copying a constant into a temporary with dupi,
use a vector constant directly.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240912024114.1097832-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This patch allows for easier manipulation of the cache description
register, CCSIDR. Which is helpful for testing as well. Currently,
numbers get hard-coded and might be prone to errors.
Therefore, this patch adds a wrapper for different types of CPUs
available in tcg to decribe caches. One function `make_ccsidr` supports
two cases by carrying a parameter as FORMAT that can be LEGACY and
CCIDX which determines the specification of the register.
For CCSIDR register, 32 bit version follows specification [1].
Conversely, 64 bit version follows specification [2].
[1] B4.1.19, ARM Architecture Reference Manual ARMv7-A and ARMv7-R
edition, https://developer.arm.com/documentation/ddi0406
[2] D23.2.29, ARM Architecture Reference Manual for A-profile Architecture,
https://developer.arm.com/documentation/ddi0487/latest/
Signed-off-by: Alireza Sanaee <alireza.sanaee@huawei.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20240903144550.280-1-alireza.sanaee@huawei.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
In vfp.decode we have the names of the VFNMA and VFNMS instructions
the wrong way around. The architecture says that bit 6 is the 'op'
bit, which is 1 for VFNMA and 0 for VFNMS, but we label these two
lines of decode the other way around. This doesn't cause any
user-visible problem because in the handling of these functions in
translate-vfp.c we give VFNMA the behaviour specified for VFNMS and
vice-versa, but it's confusing when reading the code.
Switch the names of the VFP VFNMA and VFNMS instructions in
the decode file and flip the behaviour also.
NB: the instructions VFMA and VFMS *are* decoded with op=0 for
VFMA and op=1 for VFMS; the confusion probably arose because
we assumed VFNMA and VFNMS to be the same way around.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2536
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20240830152156.2046590-1-peter.maydell@linaro.org
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Now that we've implemented the required behaviour for FEAT_EBF16, we
can enable it for the "max" CPU type, list it in our documentation,
and delete a TODO comment about it being missing.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Implement the FPCR.EBF=1 semantics for bfdotadd() operations:
* is_ebf() sets up fpst and fpst_odd
* bfdotadd_ebf() implements the fused paired-multiply-and-add
operation that we need
The paired-multiply-and-add is similar to f16_dotadd() and
we use the same trick here as in that function, but the inputs
here are bfloat16 rather than float16.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
We use bfdotadd() in four callsites for various helper functions. Currently
this all assumes that we have the FPCR.EBF=0 semantics. For FPCR.EBF=1
we will need to:
* call a different routine to bfdotadd() because we need to do a
fused multiply-add rather than separate multiply and add steps
* use a different float_status that honours the FPCR rounding mode
and denormal-flushing fields
* pass in an extra float_status that has been set up to perform
round-to-odd rounding
To prepare for this, refactor all the callsites so that instead of
for (...) {
x = bfdotadd(...);
}
they are:
float_status fpst, fpst_odd;
if (is_ebf(env, &fpst, &fpst_odd)) {
for (...) {
x = bfdotadd_ebf(..., &fpst, &fpst_odd);
}
} else {
for (...) {
x = bfdotadd(..., &fpst);
}
}
For the moment the is_ebf() function always returns false, sets up
fpst for EBF=0 semantics and never sets up fpst_odd; bfdotadd_ebf()
will assert if called. We'll fill in the handling for EBF=1 in the
next commit.
This change should be a zero-behaviour-change refactor.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Pass the env pointer through to the gvec_bfmmla helper,
so we can use it to add support for FEAT_EBF16.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Matt Borgerson