The previous setup required ops and args to be completely sequential,
and was error prone when it came to both iteration and optimization.
Reviewed-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Richard Henderson <rth@twiddle.net>
With the "old" ldst ops we didn't know the real width of the
result of the load, but with the "new" ldst ops we do.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Since all backends have been converted, remove the compatibility code.
Acked-by: Claudio Fontana <claudio.fontana@huawei.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
For a 64-bit host, the high bits of a register after a 32-bit operation
are undefined. Adjust the temps mask for all 32-bit ops to reflect that.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
If either the high or low pair can be resolved, we can
simplify to either a constant or to a 32-bit comparison.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Avoid allocating a tcg temporary to hold the constant address,
and instead place it directly into the op_call arguments.
At the same time, convert to the newly introduced tcg_out_call
backend function, rather than invoking tcg_out_op for the call.
Signed-off-by: Richard Henderson <rth@twiddle.net>
By inspection, for a deposit(x, y, 0, 64), we'd have a shift of (1<<64)
and everything else falls apart. But we can reuse the existing deposit
logic to get this right.
Signed-off-by: Richard Henderson <rth@twiddle.net>
The TCG result would be undefined, but we can at least produce one
plausible result and avoid triggering the wrath of analysis tools.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Recognize 0 operand to andc, and -1 operands to and, orc, eqv.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Like we already do for SUB and XOR.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Given, of course, an appropriate constant. These could be generated
from the "canonical" operation for inversion on the guest, or via
other optimizations.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
The shl_i32 op might set some bits of the unused 32 high bits of the
mask. Fix that by clearing the unused 32 high bits for all 32-bit ops
except load/store which operate on tl values.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Known-zero bits optimization is a great idea that helps to generate more
optimized code. However the current implementation only works in very few
cases as the computed mask is not saved.
Fix this to make it really working.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
32-bit versions of sar and shr ops should not propagate known-zero bits
from the unused 32 high bits. For sar it could even lead to wrong code
being generated.
Cc: qemu-stable@nongnu.org
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Use them in places where mulu2 and muls2 are used.
Optimize mulx2 with dead low part to mulxh.
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Richard Henderson <rth@twiddle.net>
When setcond2 is rewritten into setcond, the state of the destination
temp should be reset, so that a copy of the previous value is not
used instead of the result.
Reported-by: Michael Tokarev <mjt@tls.msk.ru>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This adds two optimizations using the non-zero bit mask. In some cases
involving shifts or ANDs the value can become zero, and can thus be
optimized to a move of zero. Second, useless zero-extension or an
AND with constant can be detected that would only zero bits that are
already zero.
The main advantage of this optimization is that it turns zero-extensions
into moves, thus enabling much better copy propagation (around 1% code
reduction). Here is for example a "test $0xff0000,%ecx + je" before
optimization:
mov_i64 tmp0,rcx
movi_i64 tmp1,$0xff0000
discard cc_src
and_i64 cc_dst,tmp0,tmp1
movi_i32 cc_op,$0x1c
ext32u_i64 tmp0,cc_dst
movi_i64 tmp12,$0x0
brcond_i64 tmp0,tmp12,eq,$0x0
and after (without patch on the left, with on the right):
movi_i64 tmp1,$0xff0000 movi_i64 tmp1,$0xff0000
discard cc_src discard cc_src
and_i64 cc_dst,rcx,tmp1 and_i64 cc_dst,rcx,tmp1
movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c
ext32u_i64 tmp0,cc_dst
movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0
brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0
Other similar cases: "test %eax, %eax + jne" where eax is already 32-bit
(after optimization, without patch on the left, with on the right):
discard cc_src discard cc_src
mov_i64 cc_dst,rax mov_i64 cc_dst,rax
movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c
ext32u_i64 tmp0,cc_dst
movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0
brcond_i64 tmp0,tmp12,ne,$0x0 brcond_i64 rax,tmp12,ne,$0x0
"test $0x1, %dl + je":
movi_i64 tmp1,$0x1 movi_i64 tmp1,$0x1
discard cc_src discard cc_src
and_i64 cc_dst,rdx,tmp1 and_i64 cc_dst,rdx,tmp1
movi_i32 cc_op,$0x1a movi_i32 cc_op,$0x1a
ext8u_i64 tmp0,cc_dst
movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0
brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0
In some cases TCG even outsmarts GCC. :) Here the input code has
"and $0x2,%eax + movslq %eax,%rbx + test %rbx, %rbx" and the optimizer,
thanks to copy propagation, does the following:
movi_i64 tmp12,$0x2 movi_i64 tmp12,$0x2
and_i64 rax,rax,tmp12 and_i64 rax,rax,tmp12
mov_i64 cc_dst,rax mov_i64 cc_dst,rax
ext32s_i64 tmp0,rax -> nop
mov_i64 rbx,tmp0 -> mov_i64 rbx,cc_dst
and_i64 cc_dst,rbx,rbx -> nop
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Add a "mask" field to the tcg_temp_info struct. A bit that is zero
in "mask" will always be zero in the corresponding temporary.
Zero bits in the mask can be produced from moves of immediates,
zero-extensions, ANDs with constants, shifts; they can then be
be propagated by logical operations, shifts, sign-extensions,
negations, deposit operations, and conditional moves. Other
operations will just reset the mask to all-ones, i.e. unknown.
[rth: s/target_ulong/tcg_target_ulong/]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
The next patch will add to the TCG optimizer a field that should be
non-zero in the default case. Thus, replace the memset of the
temps array with a loop. Only the state field has to be up-to-date,
because others are not used except if the state is TCG_TEMP_COPY
or TCG_TEMP_CONST.
[rth: Extracted the loop to a function.]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
The current helper flags, TCG_CALL_CONST and TCG_CALL_PURE might be
confusing and doesn't provide enough granularity for some helpers (FP
helpers for example).
This patch changes them into the following helpers flags:
- TCG_CALL_NO_READ_GLOBALS means that the helper does not read globals,
either directly or via an exception. They will not be saved to their
canonical location before calling the helper.
- TCG_CALL_NO_WRITE_GLOBALS means that the helper does not modify any
globals. They will only be saved to their canonical locations before
calling helpers, but they won't be reloaded afterwise.
- TCG_CALL_NO_SIDE_EFFECTS means that the call to the function is
removed if the return value is not used.
It provides convenience flags, to avoid helper definitions longer than
80 characters. It also provides compatibility flags, and updates the
documentation.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Like add2, do operand ordering, constant folding, and dead operand
elimination. The latter happens about 15% of all mulu2 during an
x86_64 bios boot.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
We can re-use these for implementing double-word folding.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This saves a whole lot of repetitive code sequences.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reduces code duplication and prefers
movcond d, c1, c2, const, s
to
movcond d, c1, c2, s, const
It also prefers
add r, r, c
over
add r, c, r
when both inputs are known constants. This doesn't matter for true add, as
we will fully constant fold that. But it matters for a follow-on patch using
this routine for add2 which may not be fully foldable.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
There are several cases that can be handled easier inside both
translators and code generators if we have out-of-band values
for conditions. It's easy enough to handle ALWAYS and NEVER in
the natural way inside the tcg middle-end.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The "op a, a, b" form is better handled on non-RISC host than the "op
a, b, a" form, so swap the arguments to this form when possible, and
when b is not a constant.
This reduces the number of generated instructions by a tiny bit.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
When both argument of brcond/movcond/setcond are the same or when one
of the two values is a constant equal to zero, it's possible to do
further optimizations.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Now that it's possible to detect copies, we can optimize the case
the "op r, a, a => movi r, 0". This helps in the computation of
overflow flags when one of the two args is 0.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Now that we can easily detect all copies, we can optimize the
"op r, a, a => mov r, a" case a bit more.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
It is possible to due copy propagation for all operations, even the one
that have side effects or clobber arguments (it only concerns input
arguments). That said, the call operation should be handled differently
due to the variable number of arguments.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The copy propagation pass tries to keep track what is a copy of what
and what has copy of what, and in addition it keep a circular list of
of all the copies. Unfortunately this doesn't fully work: a mov from
a temp which has a state "COPY" changed it into a state "HAS_COPY".
Later when this temp is used again, it is considered has not having
copy and thus no propagation is done.
This patch fixes that by removing the hiearchy between copies, and thus
only keeping a "COPY" state both meaning "is a copy" and "has a copy".
The decision of which copy to use is deferred to the actual temp
replacement. At this stage there is not one best choice to do, but only
better choices than others. For doing the best choice the operation
would have to be parsed in reversed to know if a temp is going to be
used later or not. That what is done by the liveness analysis. At this
stage it is known that globals will be always live, that local temps
will be dead at the end of the translation block, and that the temps
will be dead at the end of the basic block. This means that this stage
should try to replace temps by local temps or globals and local temps
by globals.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The copy propagation doesn't check the types of the temps during copy
propagation. However TCG is using the mov_i32 for the i64 to i32
conversion and thus the two are not equivalent.
With this patch tcg_opt_gen_mov() doesn't consider two temps of
different type as copies anymore.
So far it seems the optimization was not aggressive enough to trigger
this bug, but it will be triggered later in this series once the copy
propagation is improved.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
TCG_TEMP_ANY has no different meaning than TCG_TEMP_UNDEF, so use
the later instead.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
While swapping constants to the second operand, swap
sources matching destinations to the first operand.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Commit e31b0a7c05 fixed copy propagation on
32-bit host by restricting the copy between different types. This was the
wrong fix.
The real problem is that the all temps states should be reset at the end
of a basic block. This was done by adding such operations in the switch,
but brcond2 was forgotten (that's why the crash was only observed on 32-bit
hosts).
Fix that by looking at the TCG_OPF_BB_END instead. We need to keep the case
for op_set_label as temps might be modified through another path.
Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Given the copy propagation breakage on 32-bit hosts has been fixed
commit e31b0a7c05 can be reverted.
Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
optimizer.c contains some cases were the break is appearing in both the
if and the else parts. Fix that by moving it to the outer part. Also
move some common code there.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
brcond and setcond ops are not commutative, but it's easy to compute the
new condition after swapping the arguments. Try to always put the constant
argument in second position like for commutative ops, to help backends to
generate better code.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Split expression simplification in multiple parts so that a given op
can appear multiple times. This patch should not change anything.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Use enum TCGOpcode instead of plain old int so that the name of
current op can be seen in GDB. Add a default case to switch
so that GCC does not complain about unhandled enum cases.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
By always defining these symbols, we can eliminate a lot of ifdefs.
To allow this to be checked reliably, the semantics of the
TCG_TARGET_HAS_* macros must be changed from def/undef to true/false.
This allows even more ifdefs to be removed, converting them into
C if statements.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
This allows the simplification of the op_bits function from
tcg/optimize.c.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Copy propagation introduced in 22613af4a6
considered only global registers. However, register temps and stack
allocated locals must be handled differently because register temps
don't survive across brcond.
Fix by propagating only within same class of temps.
Tested-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Fix breakage by a640f03178
and 55c0975c5b.
Some TCG targets don't implement all TCG ops, so make
optimizing those conditional.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Perform constant folding for NOT and EXT{8,16,32}{S,U} operations.
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Perform actual constant folding for ADD, SUB and MUL operations.
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Make tcg_constant_folding do copy and constant propagation. It is a
preparational work before actual constant folding.
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Added file tcg/optimize.c to hold TCG optimizations. Function tcg_optimize
is called from tcg_gen_code_common. It calls other functions performing
specific optimizations. Stub for constant folding was added.
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
