Commit Graph

49 Commits

Author SHA1 Message Date
Paolo Bonzini 633f650254 optimize: optimize using nonzero bits
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>
2013-01-19 10:13:16 +00:00
Paolo Bonzini 3a9d8b179b optimize: track nonzero bits of registers
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>
2013-01-19 10:13:14 +00:00
Paolo Bonzini d193a14a2c optimize: only write to state when clearing optimizer data
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>
2013-01-19 10:13:13 +00:00
Evgeny Voevodin 92414b31e7 TCG: Use gen_opc_buf from context instead of global variable.
Signed-off-by: Evgeny Voevodin <e.voevodin@samsung.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2012-11-17 13:53:36 +00:00
Aurelien Jarno 7850527966 tcg: rework TCG helper flags
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>
2012-10-28 14:54:23 +01:00
Richard Henderson 1414968a6a tcg: Optimize mulu2
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>
2012-10-17 17:51:39 +02:00
Richard Henderson 212c328d61 tcg: Constant fold add2 and sub2
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-10-17 17:51:37 +02:00
Richard Henderson 6c4382f8f4 tcg: Do constant folding on double-word comparisons
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-10-17 17:51:35 +02:00
Richard Henderson 9519da7e39 tcg: Split out subroutines from do_constant_folding_cond
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>
2012-10-17 17:51:32 +02:00
Richard Henderson bc1473eff4 tcg: Optimize double-word comparisons against zero
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-10-17 17:32:29 +02:00
Richard Henderson 6e14e91b66 tcg: Use common code when failing to optimize
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>
2012-10-17 17:32:01 +02:00
Richard Henderson 0bfcb86538 tcg: Swap commutative double-word comparisons
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-10-17 17:31:57 +02:00
Richard Henderson 1e484e61e2 tcg: Canonicalize add2 operand ordering
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-10-17 17:31:53 +02:00
Richard Henderson 24c9ae4eba tcg: Split out swap_commutative as a subroutine
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>
2012-10-17 17:30:40 +02:00
Richard Henderson 0aed257f08 tcg: Add TCG_COND_NEVER, TCG_COND_ALWAYS
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>
2012-10-06 18:48:40 +02:00
Aurelien Jarno 7ef55fc919 tcg/optimize: add constant folding for deposit
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-22 15:10:21 +02:00
Aurelien Jarno c2b0e2fea2 tcg/optimize: prefer the "op a, a, b" form for commutative ops
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno b336ceb691 tcg/optimize: further optimize brcond/movcond/setcond
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno 3c94193e0b tcg/optimize: optimize "op r, a, a => movi r, 0"
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno 0aba1c7376 tcg/optimize: optimize "op r, a, a => mov r, a"
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno 1ff8c5418a tcg/optimize: do copy propagation for all operations
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno e590d4e6b3 tcg/optimize: rework copy progagation
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>
2012-09-22 15:10:21 +02:00
Aurelien Jarno b80bb016d8 tcg/optimize: check types in copy propagation
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>
2012-09-22 15:10:20 +02:00
Aurelien Jarno 48b56ce168 tcg/optimize: remove TCG_TEMP_ANY
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>
2012-09-22 15:10:20 +02:00
Richard Henderson 5d8f536300 tcg: Optimize two-address commutative operations
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>
2012-09-21 19:53:17 +02:00
Richard Henderson fa01a2084e tcg: Optimize movcond for constant comparisons
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-21 19:53:17 +02:00
Aurelien Jarno a255066039 tcg/optimize: fix end of basic block detection
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>
2012-09-19 21:53:46 +02:00
Aurelien Jarno d104bebd07 revert "TCG: fix copy propagation"
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>
2012-09-19 21:40:47 +02:00
Aurelien Jarno fedc0da251 tcg/optimize: fix if/else/break coding style
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>
2012-09-11 18:06:04 +02:00
Aurelien Jarno fbeaa26c4c tcg/optimize: add constant folding for brcond
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-11 18:06:03 +02:00
Aurelien Jarno f8dd19e5c7 tcg/optimize: add constant folding for setcond
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-11 18:06:01 +02:00
Aurelien Jarno 65a7cce17d tcg/optimize: swap brcond/setcond arguments when possible
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>
2012-09-11 18:05:59 +02:00
Aurelien Jarno 01ee5282ea tcg/optimize: simplify shift/rot r, 0, a => movi r, 0 cases
shift/rot r, 0, a is equivalent to movi r, 0.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-11 18:05:58 +02:00
Aurelien Jarno 61251c0c79 tcg/optimize: simplify and r, a, 0 cases
and r, a, 0 is equivalent to a movi r, 0.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-11 18:05:57 +02:00
Aurelien Jarno 38ee188b1b tcg/optimize: simplify or/xor r, a, 0 cases
or/xor r, a, 0 is equivalent to a mov r, a.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
2012-09-11 18:05:56 +02:00
Aurelien Jarno 56e4943825 tcg/optimize: split expression simplification
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>
2012-09-11 18:05:55 +02:00
Blue Swirl fe0de7aa5e TCG: improve optimizer debugging
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>
2011-08-28 07:17:27 +00:00
Richard Henderson cb25c80a9b tcg: Constant fold neg, andc, orc, eqv, nand, nor.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2011-08-21 18:52:25 +00:00
Richard Henderson 25c4d9cc84 tcg: Always define all of the TCGOpcode enum members.
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>
2011-08-21 18:52:24 +00:00
Richard Henderson 8399ad59e7 tcg: Add and use TCG_OPF_64BIT.
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>
2011-08-21 18:52:22 +00:00
Blue Swirl e31b0a7c05 TCG: fix copy propagation
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>
2011-08-07 09:33:20 +00:00
Blue Swirl 2ec00650f6 TCG: fix breakage by previous patch
Fix incorrect logic and typos in previous commit
1bfd07bdfe.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2011-07-30 18:54:23 +00:00
Blue Swirl 1bfd07bdfe TCG: fix breakage on some RISC hosts
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>
2011-07-30 12:21:33 +00:00
Kirill Batuzov a640f03178 Do constant folding for unary operations.
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>
2011-07-30 10:51:30 +00:00
Kirill Batuzov 55c0975c5b Do constant folding for shift operations.
Perform constant forlding for SHR, SHL, SAR, ROTR, ROTL operations.

Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2011-07-30 10:51:29 +00:00
Kirill Batuzov 9a81090b12 Do constant folding for boolean operations.
Perform constant folding for AND, OR, XOR operations.

Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2011-07-30 10:51:29 +00:00
Kirill Batuzov 53108fb574 Do constant folding for basic arithmetic operations.
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>
2011-07-30 10:51:28 +00:00
Kirill Batuzov 22613af4a6 Add copy and constant propagation.
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>
2011-07-30 10:51:27 +00:00
Kirill Batuzov 8f2e8c07a6 Add TCG optimizations stub
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>
2011-07-30 10:51:25 +00:00