Load constants in no more than two insns, which turns
out to be faster than using the constant pool.
Suggested-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Merge maybe_out_small_movi, as it no longer has additional users.
Use is_const_p{16,32}.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Give 64-bit comparison second operand a signed 33-bit immediate.
This is the smallest superset of uint32_t and int32_t, as used
by CLGFI and CGFI respectively. The rest of the 33-bit space
can be loaded into TCG_TMP0. Drop use of the constant pool.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
There is an older form that produces per-byte results,
and a newer form that produces per-register results.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reuse code from movcond to conditionally copy a2 to dest,
based on the condition codes produced by FLOGR.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The new select instruction provides two separate register inputs,
whereas the old load-on-condition instruction overlaps one of the
register inputs with the destination.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Generalize movcond to support pre-computed conditions, and the same
set of arguments at all times. This will be assumed by a following
patch, which needs to reuse tgen_movcond_int.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Return both regular and inverted condition codes from tgen_cmp2.
This lets us choose after the fact which comparision we want.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This is andc, orc, nand, nor, eqv.
We can use nor for implementing not.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Let the register allocator handle such immediates by matching
only what one insn can achieve.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Drop support for sequential OR and XOR, as the serial dependency is
slower than loading the constant first. Let the register allocator
handle such immediates by matching only what one insn can achieve.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
There is only one instruction that is applicable
to a 32-bit immediate xor.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The MIE2 facility adds a 3-operand signed 64x64->128 multiply.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
There are multiple variations, with different fields.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
One has 3 register arguments; the other has 2 plus an m3 field.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Add one instead of dropping odd addresses to the constant pool.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The distinct-operands facility is bundled into facility 45,
along with load-on-condition. We are checking this at startup.
Remove the a0 == a1 checks for 64-bit sub, and, or, xor, as there
is no space savings for avoiding the distinct-operands insn.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The fast-bcr-serialization facility is bundled into facility 45,
along with load-on-condition. We are checking this at startup.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The general-instruction-extension facility was introduced in z196,
which itself was end-of-life in 2021. In addition, z196 is the
minimum CPU supported by our set of supported operating systems:
RHEL 7 (z196), SLES 12 (z196) and Ubuntu 16.04 (zEC12).
Check for facility number 45, which will be the consilidated check
for several facilities.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The general-instruction-extension facility was introduced in z10,
which itself was end-of-life in 2019.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The extended-immediate facility was introduced in z9-109,
which itself was end-of-life in 2017.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We are already assuming the existance of long-displacement, but were
not being explicit about it. This has been present since z990.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The size of a compiled TB is limited by the uint16_t used by
gen_insn_end_off[] -- there is no need for a 32-bit branch.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Since USE_REG_TB is removed, there is no need to load the
target TB address into a register.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This reverts 829e1376d9 ("tcg/s390: Introduce TCG_REG_TB"), and
several follow-up patches. The primary motivation is to reduce the
less-tested code paths, pre-z10. Secondarily, this allows the
unconditional use of TCG_TARGET_HAS_direct_jump, which might be more
important for performance than any slight increase in code size.
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
v4: Do not simplify tgen_ori, tgen_xori.
This eliminates an ifdef for TCI, and will be required for
expanding the call for TCGv_i128.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For 64-bit hosts that had TCG_TARGET_EXTEND_ARGS, set
TCG_TARGET_CALL_ARG_I32 to TCG_CALL_ARG_EXTEND.
Otherwise, use TCG_CALL_ARG_NORMAL.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
For 32-bit hosts when TCG_TARGET_CALL_ALIGN_ARGS was set, use
TCG_CALL_ARG_EVEN. For 64-bit hosts, TCG_TARGET_CALL_ALIGN_ARGS
was silently ignored, so always use TCG_CALL_ARG_NORMAL.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We are going to modify this code, so fix its style first to avoid:
ERROR: spaces required around that '*' (ctx:VxV)
#281: FILE: tcg/s390x/tcg-target.c.inc:1224:
+ uintptr_t mask = ~(0xffffull << i*16);
^
Reviewed-by: Wilfred Mallawa <wilfred.mallawa@wdc.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221130132654.76369-2-philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We copied the data from the general register input to the
vector register output, but have not yet replicated it.
We intended to fall through into the vector-vector case,
but failed to redirect the input register.
This is caught by an assertion failure in tcg_out_insn_VRIc,
which diagnosed the incorrect register class.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The operands are output in the wrong order: the tcg selector
argument is first, whereas the s390x selector argument is last.
Tested-by: Thomas Huth <thuth@redhat.com>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/898
Fixes: 9bca986df8 ("tcg/s390x: Implement TCG_TARGET_HAS_bitsel_vec")
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We've had placeholders for these opcodes for a while,
and should have support on ppc, s390x and avx512 hosts.
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Renaming defines for quad in their various forms so that their signedness is
now explicit.
Done using git grep as suggested by Philippe, with a bit of hand edition to
keep assignments aligned.
Signed-off-by: Frédéric Pétrot <frederic.petrot@univ-grenoble-alpes.fr>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 20220106210108.138226-2-frederic.petrot@univ-grenoble-alpes.fr
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Newly defined tcg_out_vec_op (34ef767609 tcg/s390x: Add host vector framework)
for s390x uses pointer argument definition.
This fails on gcc 11 as original declaration uses array argument:
In file included from ../tcg/tcg.c:430:
/builddir/build/BUILD/qemu-6.1.50/tcg/s390x/tcg-target.c.inc:2702:42: error: argument 5 of type 'const TCGArg *' {aka 'const long unsigned int *'} declared as a pointer [-Werror=array-parameter=]
2702 | const TCGArg *args, const int *const_args)
| ~~~~~~~~~~~~~~^~~~
../tcg/tcg.c:121:41: note: previously declared as an array 'const TCGArg[16]' {aka 'const long unsigned int[16]'}
121 | const TCGArg args[TCG_MAX_OP_ARGS],
| ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~
In file included from ../tcg/tcg.c:430:
/builddir/build/BUILD/qemu-6.1.50/tcg/s390x/tcg-target.c.inc:2702:59: error: argument 6 of type 'const int *' declared as a pointer [-Werror=array-parameter=]
2702 | const TCGArg *args, const int *const_args)
| ~~~~~~~~~~~^~~~~~~~~~
../tcg/tcg.c:122:38: note: previously declared as an array 'const int[16]'
122 | const int const_args[TCG_MAX_OP_ARGS]);
| ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~
Fixing argument type to pass build.
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Message-Id: <20211027085629.240704-1-mrezanin@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The unsigned saturations are handled via generic code
using min/max. The signed saturations are expanded using
double-sized arithmetic and a saturating pack.
Since all operations are done via expansion, do not
actually set TCG_TARGET_HAS_sat_vec.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
These logical and arithmetic operations are optional but trivial.
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Implementing add, sub, and, or, xor as the minimal set.
This allows us to actually enable vectors in query_s390_facilities.
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
