Use vrsub.vi to subtract from a constant.
Reviewed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Extend comparison results from mask registers to SEW-width elements,
following recommendations in The RISC-V SPEC Volume I (Version 20240411).
This aligns with TCG's cmp_vec behavior by expanding compare results to
full element width: all 1s for true, all 0s for false.
Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20241007025700.47259-7-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The RISC-V vector instruction set utilizes the LMUL field to group
multiple registers, enabling variable-length vector registers. This
implementation uses only the first register number of each group while
reserving the other register numbers within the group.
In TCG, each VEC_IR can have 3 types (TCG_TYPE_V64/128/256), and the
host runtime needs to adjust LMUL based on the type to use different
register groups.
This presents challenges for TCG's register allocation. Currently, we
avoid modifying the register allocation part of TCG and only expose the
minimum number of vector registers.
For example, when the host vlen is 64 bits and type is TCG_TYPE_V256, with
LMUL equal to 4, we use 4 vector registers as one register group. We can
use a maximum of 8 register groups, but the V0 register number is reserved
as a mask register, so we can effectively use at most 7 register groups.
Moreover, when type is smaller than TCG_TYPE_V256, only 7 registers are
forced to be used. This is because TCG cannot yet dynamically constrain
registers with type; likewise, when the host vlen is 128 bits and
TCG_TYPE_V256, we can use at most 15 registers.
There is not much pressure on vector register allocation in TCG now, so
using 7 registers is feasible and will not have a major impact on code
generation.
This patch:
1. Reserves vector register 0 for use as a mask register.
2. When using register groups, reserves the additional registers within
each group.
Signed-off-by: Huang Shiyuan <swung0x48@outlook.com>
Co-authored-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20241007025700.47259-3-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This pointer needs to be reset after overflow just like
code_buf and code_ptr.
Cc: qemu-stable@nongnu.org
Fixes: 57a269469d ("tcg: Infrastructure for managing constant pools")
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
32-bit TSTEQ and TSTNE is subject to the same constraints as
for 64-bit, but setcond_i32 and negsetcond_i32 were incorrectly
using TCG_CT_CONST ("i") instead of TCG_CT_CONST_CMP ("C").
Adjust the constraint and make tcg_target_const_match use the
same sequence as tgen_cmp2: first check if the constant is a
valid operand for TSTEQ/TSTNE, then accept everything for 32-bit
non-test comparisons, finally check if the constant is a valid
operand for 64-bit non-test comparisons.
Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
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 tcg_out_qemu_ldst_i128, we need a non-zero index register,
which we then use as a base register in several address modes.
Since we always have TCG_REG_TMP2 available, use that.
Cc: qemu-stable@nongnu.org
Fixes: 526cd4ec01 ("tcg/ppc: Support 128-bit load/store")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2597
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-By: Michael Tokarev <mjt@tls.msk.ru>
In the fallback when STDBRX is not available, avoid clobbering
TCG_REG_TMP1, which might be h.base, which is still in use.
Use TCG_REG_TMP2 instead.
Cc: qemu-stable@nongnu.org
Fixes: 01a112e2e9 ("tcg/ppc: Reorg tcg_out_tlb_read")
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-By: Michael Tokarev <mjt@tls.msk.ru>
This patch is part of a series that moves towards a consistent use of
g_assert_not_reached() rather than an ad hoc mix of different
assertion mechanisms.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-ID: <20240919044641.386068-25-pierrick.bouvier@linaro.org>
Signed-off-by: Thomas Huth <thuth@redhat.com>
These can be simplified to and/or/andc/orc,
avoiding the load of the constantinto a register.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Do not allow cmpsel_vec to be expanded early, so that we can
make the correct decision wrt the sense of the comparison.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
These can be simplified to and/or/andc/orc,
avoiding the load of the constantinto a register.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Do not allow cmpsel_vec to be expanded early, so that we can
make the correct decision wrt the sense of the comparison.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The avx512 vpblendm* instructions exactly implement cmpsel,
using a predicate input. Of course this matches nicely with
the avx512 predicate comparison instructions.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Extend tcg_out_evex_opc to handle the predicate and
zero-merging parameters of the evex prefix.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The sse/avx instruction set only has EQ and GT as direct comparisons.
Other signed comparisons can be generated from swapping and inversion.
However unsigned comparisons are not available and must be transformed
to signed comparisons by biasing the inputs.
The avx512 instruction set has a complete set of comparisons, with
results placed into a predicate register. We can produce the normal
cmp_vec result by using VPMOVM2*.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Place immediate values second in the comparison.
Place destination matches first in the true/false values.
All of this mirrors what we do for integer setcond and movcond.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Fold "x = cond ? y : y" to "x = y".
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Expand during output instead of during opcode generation.
Remove x86_vpblendvb_vec opcode, this this removes the only user.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Move most of expansion to opcode generation, leaving the
conversion of unsigned to signed to be done in the early phase.
Small inefficiencies, but not incorrect results, are introduced
until cmpsel_vec is converted in the next patch.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Helper function to handle setting of VEXL based
on the type of the operation.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Add declaration to tcg-internal.h, making it available for
use from tcg backend vector expanders.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The loop in the 32-bit case of the vector compare operation
was incorrectly incrementing by 8 bytes per iteration instead
of 4 bytes. This caused the function to process only half of
the intended elements.
Cc: qemu-stable@nongnu.org
Fixes: 9622c697d1 (tcg: Add gvec compare with immediate and scalar operand)
Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20240904142739.854-2-zhiwei_liu@linux.alibaba.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
The use of tcg_last_op does not interact well with
TCGContext.emit_before_op, resulting in the label
being linked to something other than the branch op.
In this case it is easier to simply collect the emitted
branch op and pass it directly to add_as_label_use.
Reported-by: Elisha Hollander <just4now666666@gmail.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
TCGOp to be propagated further in the next patch.
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Different code paths handle memory accesses:
- tcg generated code
- load/store helpers
- atomic helpers
This value is saved in cpu->neg.plugin_mem_value_{high,low}. Values are
written only for accessed word size (upper bits are not set).
Atomic operations are doing read/write at the same time, so we generate
two memory callbacks instead of one, to allow plugins to access distinct
values.
For now, we can have access only up to 128 bits, thus split this in two
64 bits words. When QEMU will support wider operations, we'll be able to
reconsider this.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-Id: <20240724194708.1843704-2-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240916085400.1046925-5-alex.bennee@linaro.org>
Ensure the code structure is the same for matching constraints
and emitting code, lest we allow constants that cannot be
trivially tested.
Cc: qemu-stable@nongnu.org
Fixes: ad788aebba ("tcg/ppc: Support TCG_COND_TST{EQ,NE}")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2487
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <44328324-af73-4439-9d2b-d414e0e13dd7@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Argument ordering for setcond2 is:
output, a_low, a_high, b_low, b_high, cond
The test is supposed to be against b_low, not a_high.
Cc: qemu-stable@nongnu.org
Fixes: ceb9ee06b7 ("tcg/optimize: Handle TCG_COND_TST{EQ,NE}")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2413
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240701024623.1265028-1-richard.henderson@linaro.org>
Move detection code out of tcg, similar to other hosts.
Reviewed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The non-standard .fa library suffix breaks the link source
de-duplication done by Meson so drop it.
The lack of link source de-duplication causes AddressSanitizer to
complain ODR violations, and makes GNU ld abort when combined with
clang's LTO.
Fortunately, the non-standard suffix is not necessary anymore for
two reasons.
First, the non-standard suffix was necessary for fork-fuzzing.
Meson wraps all standard-suffixed libraries with --start-group and
--end-group. This made a fork-fuzz.ld linker script wrapped as well and
broke builds. Commit d2e6f9272d ("fuzz: remove fork-fuzzing
scaffolding") dropped fork-fuzzing so we can now restore the standard
suffix.
Second, the libraries are not even built anymore, because it is
possible to just use the object files directly via extract_all_objects().
The occurences of the suffix were detected and removed by performing
a tree-wide search with 'fa' and .fa (note the quotes and dot).
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-ID: <20240524-xkb-v4-4-2de564e5c859@daynix.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We used to request declare_dependency() to link_whole static libraries.
If a static library is a thin archive, GNU ld keeps all object files
referenced by the archive open, and sometimes exceeds the open file limit.
Another problem with link_whole is that suboptimal handling of nested
dependencies.
link_whole by itself does not propagate dependencies. In particular,
gnutls, a dependency of crypto, is not propagated to its users, and we
currently workaround the issue by declaring gnutls as a dependency for
each crypto user. On the other hand, if you write something like
libfoo = static_library('foo', 'foo.c', dependencies: gnutls)
foo = declare_dependency(link_whole: libfoo)
libbar = static_library('bar', 'bar.c', dependencies: foo)
bar = declare_dependency(link_whole: libbar, dependencies: foo)
executable('prog', sources: files('prog.c'), dependencies: [foo, bar])
hoping to propagate the gnutls dependency into bar.c, you'll see a
linking failure for "prog", because the foo.c.o object file is included in
libbar.a and therefore it is linked twice into "prog": once from libfoo.a
and once from libbar.a. Here Meson does not see the duplication, it
just asks the linker to link all of libfoo.a and libbar.a into "prog".
Instead of using link_whole, extract objects included in static libraries
and pass them to declare_dependency(); and then the dependencies can be
added as well so that they are propagated, because object files on the
linker command line are always deduplicated.
This requires Meson 1.1.0 or later.
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-ID: <20240524-objects-v1-1-07cbbe96166b@daynix.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit 45ccdbcb24.
The x86-64 instruction set can now be tuned down to x86-64 v1
or i386 Pentium Pro.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Simplify the logic for two-part, 32-bit pc-relative addresses.
Rather than assume all such fit in int32_t, do some arithmetic
and assert a result, do some arithmetic first and then check
to see if the pieces are in range.
Cc: qemu-stable@nongnu.org
Fixes: dacc51720d ("tcg/loongarch64: Implement tcg_out_mov and tcg_out_movi")
Reviewed-by: Song Gao <gaosong@loongson.cn>
Reported-by: Song Gao <gaosong@loongson.cn>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Song Gao <gaosong@loongson.cn>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
