qemu-stable
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v2 06/14] tcg/{i386, s390x}: Add earlyclobber to the op_add2'


From: Philippe Mathieu-Daudé
Subject: Re: [PATCH v2 06/14] tcg/{i386, s390x}: Add earlyclobber to the op_add2's first output
Date: Wed, 19 Jul 2023 14:42:40 +0200
User-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0) Gecko/20100101 Thunderbird/102.13.0

On 19/7/23 14:30, Ilya Leoshkevich wrote:
On Wed, 2023-07-19 at 14:08 +0200, Philippe Mathieu-Daudé wrote:
Hi Ilya,

On 19/7/23 11:44, Ilya Leoshkevich wrote:
i386 and s390x implementations of op_add2 require an earlyclobber,
which is currently missing. This breaks VCKSM in s390x guests.
E.g., on
x86_64 the following op:

      add2_i32 tmp2,tmp3,tmp2,tmp3,tmp3,tmp2   dead: 0 2 3 4 5
pref=none,0xffff

is translated to:

      addl     %ebx, %r12d
      adcl     %r12d, %ebx

Introduce a new C_N1_O1_I4 constraint, and make sure that
earlyclobber
of aliased outputs is honored.

Cc: qemu-stable@nongnu.org
Fixes: 82790a870992 ("tcg: Add markup for output requires new
register")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---
   tcg/i386/tcg-target-con-set.h  | 2 +-
   tcg/i386/tcg-target.c.inc      | 2 +-
   tcg/s390x/tcg-target-con-set.h | 5 ++---
   tcg/s390x/tcg-target.c.inc     | 4 ++--
   tcg/tcg.c                      | 8 +++++++-
   5 files changed, 13 insertions(+), 8 deletions(-)


diff --git a/tcg/tcg.c b/tcg/tcg.c
index 652e8ea6b93..ddfe9a96cb7 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -648,6 +648,7 @@ static void tcg_out_movext3(TCGContext *s,
const TCGMovExtend *i1,
   #define C_O2_I2(O1, O2, I1, I2)         C_PFX4(c_o2_i2_, O1, O2,
I1, I2),
   #define C_O2_I3(O1, O2, I1, I2, I3)     C_PFX5(c_o2_i3_, O1, O2,
I1, I2, I3),
   #define C_O2_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_o2_i4_, O1, O2,
I1, I2, I3, I4),
+#define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_n1_o1_i4_, O1,
O2, I1, I2, I3, I4),

No need for O2. Also can you place it earlier just after C_N1_I2?

Shouldn't it still be a 6-argument constraint?
INDEX_op_add2_i32 and friends take 6 arguments after all.

Oops sorry, you are right.



reply via email to

[Prev in Thread] Current Thread [Next in Thread]