The allows immediates to be used for ORR and BIC,
as well as the trivial inversions, ORC and AND.
Backports commit 9e27f58b9902834dffc0d66d9eb62f78d9c2a632 from qemu
Use MOVI+ORR or MVNI+BIC in order to build some vector constants,
as opposed to dropping them to the constant pool. This includes
all 16-bit constants and a similar set of 32-bit constants.
Backports commit 02f3a5b4744885258758d07ebe09cf965de78bcf from qemu
The compliment of a subset of immediates can be computed
with a single instruction.
Backports commit 7e308e003e5b6ddd3130e09711e1d33693230696 from qemu
There are several sub-classes of vector immediate, and only MOVI
can use them all. This will enable usage of MVNI and ORRI, which
use progressively fewer sub-classes.
This patch adds no new functionality, merely splits the function
and moves part of the logic into tcg_out_dupi_vec.
Backports commit 984fdcee342473dfe797897758929dad654693c8 from qemu
The instruction set has 3 insns that perform the same operation,
only varying in which operand must overlap the destination. We
can represent the operation without overlap and choose based on
the operands seen.
Backports commit a9e434a5dc16f71ee156428619fc3c3765b68f26 from qemu
Using umin(a, b) == a as an expansion for TCG_COND_LEU is a
better alternative to (a - INT_MIN) <= (b - INT_MIN).
Backports commit ebcfb91abed8c0fb180a968b9004419c208dcc02 from qemu
We already had backend support for this feature. Expand the new
cmpsel opcode using vpblendb. The combination allows us to avoid
an extra NOT for some comparison codes.
Backports commit 904c5e19672778cc3349f4975437cfdf3371abb6 from qemu
If INDEX_op_foo is always expanded by tcg_expand_vec_op, then
there may be no reasonable set of constraints to return from
tcg_target_op_def for that opcode.
Let TCG_TARGET_HAS_foo be specified as -1 in that case. Thus a
boolean test for TCG_TARGET_HAS_foo is true, but we will not
assert within process_op_defs when no constraints are specified.
Compare this with tcg_can_emit_vec_op, which already uses this
tri-state indication.
Backports commit 25c012b4009256505be3430480954a0233de343e from qemu
This makes do_op3 match do_op2 in allowing for failure,
and thus fall back expansions.
Backports commit 17f79944ebeace8bf43047a33b7775ba5ed9070e from qemu
Perform a per-element conditional move. This combination operation is
easier to implement on some host vector units than plain cmp+bitsel.
Omit the usual gvec interface, as this is intended to be used by
target-specific gvec expansion call-backs.
Backports commit f75da2988eb2457fa23d006d573220c5c680ec4e from qemu
This operation performs d = (b & a) | (c & ~a), and is present
on a majority of host vector units. Include gvec expanders.
Backports commit 38dc12947ec9106237f9cdbd428792c985cd86ae from qemu
The paths through tcg_gen_dup_mem_vec and through MO_128 were
missing the check_size_align. The path through MO_128 was also
missing the expand_clr. This last was not visible because the
only user is ARM SVE, which would set oprsz == maxsz, and not
require the clear.
Fix by adding the check_size_align and using do_dup directly
instead of duplicating the check in tcg_gen_gvec_dup_{i32,i64}.
Backports commit 532ba368a13712724137228b5e7e9435994d25e1 from qemu
The VBROADCASTSD instruction only allows %ymm registers as destination.
Rather than forcing VEX.L and writing to the entire 256-bit register,
revert to using MOVDDUP with an %xmm register. This is sufficient for
an avx1 host since we do not support TCG_TYPE_V256 for that case.
Also fix the 32-bit avx2, which should have used VPBROADCASTW.
Fixes: 1e262b49b533
Backports commit 7b60ef3264e9627ac6efb34e9a6130647e9b55c0 from qemu
Remove a function of the same name from target/arm/.
Use a branchless implementation of abs gleaned from gcc.
Backports commit ff1f11f7f8710a768f9313f24bd7f509d3db27e5 from qemu
Allow expansion either via shift by scalar or by replicating
the scalar for shift by vector.
Backports commit b4578cd91cda4cef1c413304353ca6dc5b957b60 from qemu
The gvec expanders perform a modulo on the shift count. If the target
requires alternate behaviour, then it cannot use the generic gvec
expanders anyway, and will have to have its own custom code.
Backports commit 5ee5c14cacda27e904cd6b0d9e7ffe1acff42838 from qemu
Allow the backend to expand dup from memory directly, instead of
forcing the value into a temp first. This is especially important
if integer/vector register moves do not exist.
Note that officially tcg_out_dupm_vec is allowed to fail.
If it did, we could fix this up relatively easily:
VECE == 32/64:
Load the value into a vector register, then dup.
Both of these must work.
VECE == 8/16:
If the value happens to be at an offset such that an aligned
load would place the desired value in the least significant
end of the register, go ahead and load w/garbage in high bits.
Load the value w/INDEX_op_ld{8,16}_i32.
Attempt a move directly to vector reg, which may fail.
Store the value into the backing store for OTS.
Load the value into the vector reg w/TCG_TYPE_I32, which must work.
Duplicate from the vector reg into itself, which must work.
All of which is well and good, except that all supported
hosts can support dupm for all vece, so all of the failure
paths would be dead code and untestable.
Backports commit 37ee55a081b7863ffab2151068dd1b2f11376914 from qemu
The LD1R instruction does all the work. Note that the only
useful addressing mode is a base register with no offset.
Backports commit f23e5e15edfd49d5dd72cab2ed2d85ac354b2eeb from qemu
This case is similar to INDEX_op_mov_* in that we need to do
different things depending on the current location of the source.
Backports commit bab1671f0fa928fd678a22f934739f06fd5fd035 from qemu
The i386 backend already has these functions, and the aarch64 backend
could easily split out one. Nothing is done with these functions yet,
but this will aid register allocation of INDEX_op_dup_vec in a later patch.
Adjust the aarch64 tcg_out_dupi_vec signature to match the new interface.
Backports commit e7632cfa8b76cdbbc1c76e8737338ef5844e7d60 from qemu
PowerPC Altivec does not support direct moves between vector registers
and general registers. So when tcg_out_mov fails, we can use the
backing memory for the temporary to perform the move.
Backports commit 240c08d0998f402c325fce489de0d14831048128 from qemu
This patch merely changes the interface, aborting on all failures,
of which there are currently none.
Backports commit 78113e83e0007e869c9f0cb4c0497a77538988e3 from qemu
We have a function that takes an additional condition parameter
over the standard backend interface. It already takes care of
eliding no-op moves.
Backports commit c16f52b2c5d91c36e121795bd3b386cea0b7573c from qemu
The only fixed_reg is cpu_env, and it should not be modified
during any TB. Therefore code that tries to special-case moves
into a fixed_reg is dead. Remove it.
Backports commit d63e3b6e694ad6c887be135dddb9cd4893f1a844 from qemu
Replace the single opcode in .opc with a null-terminated
array in .opt_opc. We still require that all opcodes be
used with the same .vece.
Validate the contents of this list with CONFIG_DEBUG_TCG.
All tcg_gen_*_vec functions will check any list active
during .fniv expansion. Swap the active list in and out
as we expand other opcodes, or take control away from the
front-end function.
Convert all existing vector aware front ends.
Backports commit 53229a7703eeb2bbe101a19a33ef22aaf960c65b from qemu
PowerPC Altivec does not support add and subtract of 64-bit elements.
Prepare for that configuration by not assuming the operation is
universally supported.
Backports commit ce27c5d1a38e93da38653af71fb468c5eded4c7b from qemu
Use tcg_can_emit_vec_op instead of just TCG_TARGET_HAS_neg_vec,
so that we check the type and vece for the actual operation.
Backports commit ac383dde33405106469d04a78de1d76f1a730cb1 from qemu
Let's add tcg_gen_gvec_3i(), similar to tcg_gen_gvec_2i(), however
without introducing "gen_helper_gvec_3i *fnoi", as it isn't needed
for now.
Backports commit e1227bb6e59173117f094a6a13b998587b45c928 from qemu
Thereby decoupling the resulting translated code from the current state
of the system.
The tb->cflags field is not passed to tcg generation functions. So
we add a field to TCGContext, storing there a copy of tb->cflags.
Most architectures have <= 32 registers, which results in a 4-byte hole
in TCGContext. Use this hole for the new field.
Backports commit e82d5a2460b0e176128027651ff9b104e4bdf5cc from qemu
This will not necessarily restrict the size of the TB, since for v7
the majority of constant pool usage is for calls from the out-of-line
ldst code, which is already at the end of the TB. But this does
allow us to save one insn per reference on the off-chance.
Backports commit b4b82d7e9caff7ccca5c621817b5a4b8e95eb9b1 from qemu
There is no point in coding for a 2GB offset when the max TB size
is already limited to 64k. If we further restrict to 32k then we
can eliminate the extra ADDIS instruction.
Backports commit a7cdaf710f2aaaf0be855a338dd67463d4bb99e2 from qemu
If the TB generates too much code, such that backend relocations
overflow, try again with a smaller TB. In support of this, move
relocation processing from a random place within tcg_out_op, in
the handling of branch opcodes, to a new function at the end of
tcg_gen_code.
This is not a complete solution, as there are additional relocs
generated for out-of-line ldst handling and constant pools.
Backports commit 7ecd02a06f8f4c0bbf872ecc15e37035b7e1df5f from qemu
If a TB generates too much code, try again with fewer insns.
Fixes: https://bugs.launchpad.net/bugs/1824853
Backports commit 6e6c4efed995d9eca6ae0cfdb2252df830262f50 from qemu