From: Richard Henderson <richard.henderson@linaro.org>
To: Paolo Bonzini <pbonzini@redhat.com>, qemu-devel@nongnu.org
Cc: zhao1.liu@intel.com
Subject: Re: [PATCH v2 05/25] target/i386: cleanup cc_op changes for REP/REPZ/REPNZ
Date: Mon, 6 May 2024 09:07:38 -0700 [thread overview]
Message-ID: <e1019e1e-3678-4882-9cfb-f4747d80b9a3@linaro.org> (raw)
In-Reply-To: <20240506080957.10005-6-pbonzini@redhat.com>
On 5/6/24 01:09, Paolo Bonzini wrote:
> gen_update_cc_op must be called before control flow splits. Do it
> where the jump on ECX!=0 is translated.
>
> On the other hand, remove the call before gen_jcc1, which takes care of
> it already, and explain why REPZ/REPNZ need not use CC_OP_DYNAMIC---the
> translation block ends before any control-flow-dependent cc_op could
> be observed.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> target/i386/tcg/translate.c | 13 +++++++++----
> 1 file changed, 9 insertions(+), 4 deletions(-)
>
> diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c
> index 3f1d2858fc9..6b766f5dd3f 100644
> --- a/target/i386/tcg/translate.c
> +++ b/target/i386/tcg/translate.c
> @@ -1242,11 +1242,15 @@ static inline void gen_jcc1(DisasContext *s, int b, TCGLabel *l1)
> }
>
> /* XXX: does not work with gdbstub "ice" single step - not a
> - serious problem */
> + serious problem. The caller can jump to the returned label
> + to stop the REP but, if the flags have changed, it has to call
> + gen_update_cc_op before doing so. */
> static TCGLabel *gen_jz_ecx_string(DisasContext *s)
> {
> TCGLabel *l1 = gen_new_label();
> TCGLabel *l2 = gen_new_label();
> +
> + gen_update_cc_op(s);
> gen_op_jnz_ecx(s, l1);
> gen_set_label(l2);
> gen_jmp_rel_csize(s, 0, 1);
> @@ -1342,7 +1346,6 @@ static void gen_repz(DisasContext *s, MemOp ot,
> void (*fn)(DisasContext *s, MemOp ot))
> {
> TCGLabel *l2;
> - gen_update_cc_op(s);
> l2 = gen_jz_ecx_string(s);
> fn(s, ot);
> gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
Ok.
> @@ -1364,11 +1367,13 @@ static void gen_repz2(DisasContext *s, MemOp ot, int nz,
> void (*fn)(DisasContext *s, MemOp ot))
> {
> TCGLabel *l2;
> - gen_update_cc_op(s);
> l2 = gen_jz_ecx_string(s);
> + /*
> + * Only one iteration is done at a time, so there is
> + * no control flow junction here and cc_op is never dynamic.
> + */
> fn(s, ot);
> gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
> - gen_update_cc_op(s);
> gen_jcc1(s, (JCC_Z << 1) | (nz ^ 1), l2);
> if (s->repz_opt) {
> gen_op_jz_ecx(s, l2);
Ok, but only because gen_jcc1 does the gen_update_cc_op. The comment is neither correct
nor necessary.
The reason to write cc_op before branches instead of junctions is to avoid having *two*
writes of cc_op on either side of the branch.
r~
next prev parent reply other threads:[~2024-05-06 16:09 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-06 8:09 [PATCH v2 00/25] target/i386: convert 1-byte opcodes to new decoder Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 01/25] target/i386: use TSTEQ/TSTNE to test low bits Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 02/25] target/i386: use TSTEQ/TSTNE to check flags Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 03/25] target/i386: remove mask from CCPrepare Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 04/25] target/i386: cc_op is not dynamic in gen_jcc1 Paolo Bonzini
2024-05-06 15:53 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 05/25] target/i386: cleanup cc_op changes for REP/REPZ/REPNZ Paolo Bonzini
2024-05-06 16:07 ` Richard Henderson [this message]
2024-05-06 16:31 ` Paolo Bonzini
2024-05-06 16:39 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 06/25] target/i386: pull cc_op update to callers of gen_jmp_rel{, _csize} Paolo Bonzini
2024-05-06 16:12 ` [PATCH v2 06/25] target/i386: pull cc_op update to callers of gen_jmp_rel{,_csize} Richard Henderson
2024-05-06 8:09 ` [PATCH v2 07/25] target/i386: extend cc_* when using them to compute flags Paolo Bonzini
2024-05-06 16:16 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 08/25] target/i386: do not use s->T0 and s->T1 as scratch registers for CCPrepare Paolo Bonzini
2024-05-06 16:18 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 09/25] target/i386: clarify the "reg" argument of functions returning CCPrepare Paolo Bonzini
2024-05-06 16:19 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 10/25] target/i386: cleanup *gen_eob* Paolo Bonzini
2024-05-06 16:21 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 11/25] target/i386: reintroduce debugging mechanism Paolo Bonzini
2024-05-06 16:23 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 12/25] target/i386: move 00-5F opcodes to new decoder Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 13/25] target/i386: extract gen_far_call/jmp, reordering temporaries Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 14/25] target/i386: allow instructions with more than one immediate Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 15/25] target/i386: move 60-BF opcodes to new decoder Paolo Bonzini
2024-05-06 16:44 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 16/25] target/i386: generalize gen_movl_seg_T0 Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 17/25] target/i386: move C0-FF opcodes to new decoder (except for x87) Paolo Bonzini
2024-05-06 16:56 ` Richard Henderson
2024-05-06 8:09 ` [PATCH v2 18/25] target/i386: merge and enlarge a few ranges for call to disas_insn_new Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 19/25] target/i386: move remaining conditional operations to new decoder Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 20/25] target/i386: move BSWAP " Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 21/25] target/i386: port extensions of one-byte opcodes " Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 22/25] target/i386: remove now-converted opcodes from old decoder Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 23/25] target/i386: decode x87 instructions in a separate function Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 24/25] target/i386: split legacy decoder into " Paolo Bonzini
2024-05-06 8:09 ` [PATCH v2 25/25] target/i386: remove duplicate prefix decoding Paolo Bonzini
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=e1019e1e-3678-4882-9cfb-f4747d80b9a3@linaro.org \
--to=richard.henderson@linaro.org \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=zhao1.liu@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).