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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Before we rearrange the TCG_COND enumeration, add a predicate for
the (single) use of comparisons vs TCGCond.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

The TCG jmp operation doesn't really make sense in the QEMU context, it
is unused, it is not implemented by some targets, and it is wrongly
implemented by some others.

This patch simply removes it.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Acked-by: NBlue Swirl <blauwirbel@gmail.com>
Acked-by: Stefan Weil<sw@weilnetz.de>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Support for helper functions with 5 arguments was missing
in the code generator and in the interpreter.

There is no need to pass the constant TCG_AREG0 from the
code generator to the interpreter. Remove that code for
the INDEX_op_qemu_st* opcodes.
Signed-off-by: NStefan Weil <sw@weilnetz.de>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

The movcond_i32 op has to be protected with TCG_TARGET_HAS_movcond_i32
to fix the build with -march < i686.

Thanks to Richard Henderson for the hint.
Reported-by: NAlex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

When movcond_i32 is available we can further reduce the generated
op count from 12 to 6, and the generated code size on i686 from
88 to 74 bytes.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Avoiding 64-bit arithmetic (outside of the compare) reduces the
generated op count from 15 to 12, and the generated code size on
i686 from 105 to 88 bytes.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Checking that we don't try for idx != [01] is trivial.  Checking
that we don't issue more than one of any index requires a tad
more data and some ifdefs protecting that new variable.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Given these are constants, checking once here means everything
after can assume they're correct.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Like the C assert macro, except only enabled for CONFIG_DEBUG_TCG,
and without having to set _NDEBUG and disable all other asserts at
the same time.

The use of __builtin_unreachable (when available) gives the compiler
the same information, which may (or may not) help it optimize better.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

For tcg_gen_concat_i32_i64 we only use deposit if the host supports it.
For tcg_gen_concat32_i64 even if the host does not, as we get identical
code before and after.

Note that this relies on the ANDI -> EXTU patch for the identity claim.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Note that xori_i64 failed to perform even the minimal
optimizations promised by the README.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Copy the same optimizations from ori_i32.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Note that andi_i64 failed to perform even the minimal
optimizations promised by the README.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

The README file documented the operand ordering of the tcg_gen_*
functions.  Since we're documenting opcodes here, use the true
operand ordering.
Signed-off-by: NRichard Henderson <rth@twiddle.net>
Cc: malc <av1474@comtv.ru>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Fix the MIPS32(R2) cpu detection so that it also works with
-march=octeon. Thanks to Andrew Pinski for the hint.

Cc: Andrew Pinski <apinski@cavium.com>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

This reverts commit ad49d1f7.

This commit was not supposed to be pushed.
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Thanks to Richard Henderson
Signed-off-by: Nmalc <av1474@comtv.ru>

The TCG targets no longer need individual implementations.

Since commit 6a18ae2d,
'flags' is no longer used in tcg_target_get_call_iarg_regs_count.

The remaining tcg_target_get_call_iarg_regs_count is trivial and only
called once. Therefore the patch eliminates it completely.
Signed-off-by: NStefan Weil <sw@weilnetz.de>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

32 bit x86 hosts don't need registers for helper function arguments
because they use the default stack based calling convention.

Removing the registers allows simpler code for function
tcg_target_get_call_iarg_regs_count.
Signed-off-by: NStefan Weil <sw@weilnetz.de>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

While 64 bit hosts use the first three registers which are also used
as function input parameters, 32 bit hosts use TCG_REG_EAX and
TCG_REG_EDX which are not used in parameter passing.

After defining new register macros for the registers used in L
constraint, the patch replaces most occurrences of
tcg_target_call_iarg_regs[0], tcg_target_call_iarg_regs[1] and
tcg_target_call_iarg_regs[2] by those new macros.

tcg_target_call_iarg_regs remains unchanged when it is used for input
arguments (only with 64 bit hosts) before tcg_out_calli.

A comment related to those registers was fixed, too.
Signed-off-by: NStefan Weil <sw@weilnetz.de>
[aurel32: build fix on i386, small optimization for i386 in the prologue]
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

TCG uses 6 registers for function arguments on 64 bit Linux hosts,
but only 4 registers on W64 hosts.

Commit 2999a0b2 increased the number
of arguments for some important helper functions from 4 to 5
which triggered a bug for W64 hosts: QEMU aborts when executing
helper_lcall_real in the guest's BIOS because function
tcg_target_get_call_iarg_regs_count always returned 6.

As W64 has only 4 registers for arguments, the 5th argument must be
passed on the stack using a correct stack offset.
Signed-off-by: NStefan Weil <sw@weilnetz.de>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

See
http://lists.nongnu.org/archive/html/qemu-devel/2012-09/msg03196.html
for the whole story.
Signed-off-by: NMax Filippov <jcmvbkbc@gmail.com>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Signed-off-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

Commit 25c4d9cc changed all TCGOpcode enums to be available, so we don't
need to #ifdef #endif the one that are available only on some targets.
This makes the code easier to read.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien 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: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

TCG_TEMP_ANY has no different meaning than TCG_TEMP_UNDEF, so use
the later instead.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

movcond operation can be implemented on MIPS32 Release 2 using the MOVN,
MOVZ, SLT and SLTU instructions.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

deposit operations can be optimized on MIPS32 Release 2 using the INS
instruction.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

rotr operations can be optimized on MIPS32 Release 2 using the ROTR and
ROTRV instructions. Also implemented rotl operations by subtracting the
shift from 32.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

bswap operations can be optimized on MIPS32 Release 2 using the ROTR,
WSBH and SEH instructions. We can't use the non-R2 code to implement the
ops due to registers constraints, so don't define the corresponding
TCG_TARGET_HAS_bswap* values.

Also bswap16* operations are supposed to be called with the 16 high bits
zeroed. This is the case everywhere (including for TCG by definition)
except when called from the store helper. Remove the AND instructions from
bswap16* and move it there.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>

MIPS has some conditional branch instructions when comparing with zero.
Use them.
Reviewed-by: NRichard Henderson <rth@twiddle.net>
Signed-off-by: NAurelien Jarno <aurelien@aurel32.net>