From: Laurent Desnogues <laurent.desnogues@gmail.com>
To: malc <av1474@comtv.ru>
Cc: qemu-devel@nongnu.org, Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH 0/7] tcg: conditional set and move opcodes
Date: Thu, 17 Dec 2009 16:37:27 +0100 [thread overview]
Message-ID: <761ea48b0912170737i54f25561q468a4a193114a903@mail.gmail.com> (raw)
In-Reply-To: <Pine.LNX.4.64.0912171825330.4454@linmac.oyster.ru>
On Thu, Dec 17, 2009 at 4:32 PM, malc <av1474@comtv.ru> wrote:
[...]
>
> Some:
> a. It breaks tcg on PPC[1]:
>
> ...qemu/tcg/tcg.c:1378: tcg fatal error
What a surprise :-)
I can provide a similar patch for ARM (I already have one
for my own implementation of setcond), but I'll wait for this
patch series to stabilize first.
Laurent
> b. Documentation for movcond has a typo, t0 is assigned not t1
>
> c. Historically things like that were made conditional with
> a generic fallback (bswap, neg, not, rot, etc)
>
> d. Documentation for setcond2 is missing
>
> e. There's some indentation weirdness here and there and `git am'
> complains about added trailing whitespace
>
> It would also be interesting to learn what impact adding those two
> has on performance, any results?
>
> [..snip..]
>
> [1] With following i can run some i386 user tests on PPC32 (ls,
> openssl)
>
> diff --git a/tcg/ppc/tcg-target.c b/tcg/ppc/tcg-target.c
> index 07e6941..195af13 100644
> --- a/tcg/ppc/tcg-target.c
> +++ b/tcg/ppc/tcg-target.c
> @@ -316,6 +316,7 @@ static int tcg_target_const_match(tcg_target_long val,
> #define STH OPCD(44)
> #define STW OPCD(36)
>
> +#define ADDIC OPCD(12)
> #define ADDI OPCD(14)
> #define ADDIS OPCD(15)
> #define ORI OPCD(24)
> @@ -339,6 +340,7 @@ static int tcg_target_const_match(tcg_target_long val,
> #define CRANDC XO19(129)
> #define CRNAND XO19(225)
> #define CROR XO19(449)
> +#define CRNOR XO19( 33)
>
> #define EXTSB XO31(954)
> #define EXTSH XO31(922)
> @@ -365,6 +367,8 @@ static int tcg_target_const_match(tcg_target_long val,
> #define MTSPR XO31(467)
> #define SRAWI XO31(824)
> #define NEG XO31(104)
> +#define MFCR XO31( 19)
> +#define CNTLZW XO31( 26)
>
> #define LBZX XO31( 87)
> #define LHZX XO31(279)
> @@ -1073,6 +1077,95 @@ static void tcg_out_brcond (TCGContext *s, int cond,
> tcg_out_bc (s, tcg_to_bc[cond], label_index);
> }
>
> +static void tcg_out_setcond (TCGContext *s, int cond, TCGArg arg0,
> + TCGArg arg1, TCGArg arg2, int const_arg2)
> +{
> + int crop, sh;
> +
> + switch (cond) {
> + case TCG_COND_EQ:
> + if (const_arg2) {
> + if ((uint16_t) arg2 == arg2) {
> + tcg_out32 (s, XORI | RS (arg1) | RA (0) | arg2);
> + }
> + else {
> + tcg_out_movi (s, TCG_TYPE_I32, 0, arg2);
> + tcg_out32 (s, XOR | SAB (arg1, 0, 0));
> + }
> + }
> + else {
> + tcg_out32 (s, XOR | SAB (arg1, 0, arg2));
> + }
> + tcg_out32 (s, CNTLZW | RS (0) | RA (0));
> + tcg_out32 (s, (RLWINM
> + | RA (arg0)
> + | RS (0)
> + | SH (27)
> + | MB (5)
> + | ME (31)
> + )
> + );
> + return;
> +
> + case TCG_COND_NE:
> + if (const_arg2) {
> + if ((uint16_t) arg2 == arg2) {
> + tcg_out32 (s, XORI | RS (arg1) | RA (0) | arg2);
> + }
> + else {
> + tcg_out_movi (s, TCG_TYPE_I32, 0, arg2);
> + tcg_out32 (s, XOR | SAB (arg1, 0, 0));
> + }
> + }
> + else {
> + tcg_out32 (s, XOR | SAB (arg1, 0, arg2));
> + }
> +
> + tcg_out32 (s, ADDIC | RT (arg0) | RA (0) | 0xffff);
> + tcg_out32 (s, SUBFE | TAB (arg0, arg0, 0));
> + return;
> +
> + case TCG_COND_LTU:
> + case TCG_COND_LT:
> + sh = 29;
> + crop = 0;
> + break;
> +
> + case TCG_COND_GEU:
> + case TCG_COND_GE:
> + sh = 31;
> + crop = CRNOR | BT (7, CR_EQ) | BA (7, CR_LT) | BB (7, CR_LT);
> + break;
> +
> + case TCG_COND_LEU:
> + case TCG_COND_LE:
> + sh = 31;
> + crop = CRNOR | BT (7, CR_EQ) | BA (7, CR_GT) | BB (7, CR_GT);
> + break;
> +
> + case TCG_COND_GTU:
> + case TCG_COND_GT:
> + sh = 30;
> + crop = 0;
> + break;
> +
> + default:
> + tcg_abort ();
> + }
> +
> + tcg_out_cmp (s, cond, arg1, arg2, const_arg2, 7);
> + tcg_out32 (s, MFCR | RT (0));
> + if (crop) tcg_out32 (s, crop);
> + tcg_out32 (s, (RLWINM
> + | RA (arg0)
> + | RS (0)
> + | SH (sh)
> + | MB (31)
> + | ME (31)
> + )
> + );
> +}
> +
> /* XXX: we implement it at the target level to avoid having to
> handle cross basic blocks temporaries */
> static void tcg_out_brcond2 (TCGContext *s, const TCGArg *args,
> @@ -1496,6 +1589,10 @@ static void tcg_out_op(TCGContext *s, int opc, const TCGArg *args,
> tcg_out32 (s, EXTSH | RS (args[1]) | RA (args[0]));
> break;
>
> + case INDEX_op_setcond_i32:
> + tcg_out_setcond(s, args[3], args[0], args[1], args[2], const_args[2]);
> + break;
> +
> default:
> tcg_dump_ops (s, stderr);
> tcg_abort ();
> @@ -1544,6 +1641,8 @@ static const TCGTargetOpDef ppc_op_defs[] = {
>
> { INDEX_op_neg_i32, { "r", "r" } },
>
> + { INDEX_op_setcond_i32, { "r", "r", "ri" } },
> +
> #if TARGET_LONG_BITS == 32
> { INDEX_op_qemu_ld8u, { "r", "L" } },
> { INDEX_op_qemu_ld8s, { "r", "L" } },
>
> --
> mailto:av1474@comtv.ru
>
>
>
next prev parent reply other threads:[~2009-12-17 15:37 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-12-17 1:19 [Qemu-devel] [PATCH 0/7] tcg: conditional set and move opcodes Richard Henderson
2009-12-16 0:34 ` [Qemu-devel] [PATCH 1/7] tcg: Generic support for conditional set and conditional move Richard Henderson
2009-12-16 0:35 ` [Qemu-devel] [PATCH 2/7] tcg-amd64: Implement setcond and movcond Richard Henderson
2009-12-16 0:36 ` [Qemu-devel] [PATCH 3/7] target-alpha: Use setcond/movcond in integer compares and cmoves Richard Henderson
2009-12-16 23:17 ` [Qemu-devel] [PATCH 4/7] tcg-i386: Implement setcond, movcond, setcond2 Richard Henderson
2009-12-16 23:26 ` [Qemu-devel] [PATCH 5/7] tcg-sparc: Implement setcond, movcond, setcond2, brcond2 Richard Henderson
2009-12-19 10:31 ` Blue Swirl
2009-12-19 17:47 ` Richard Henderson
2009-12-19 21:25 ` Blue Swirl
2009-12-19 22:52 ` Richard Henderson
2009-12-20 11:06 ` Blue Swirl
2009-12-16 23:28 ` [Qemu-devel] [PATCH 6/7] target-i386: Use setcond and movcond Richard Henderson
2009-12-16 23:29 ` [Qemu-devel] [PATCH 7/7] target-mips: " Richard Henderson
2009-12-17 15:32 ` [Qemu-devel] [PATCH 0/7] tcg: conditional set and move opcodes malc
2009-12-17 15:37 ` Laurent Desnogues [this message]
2009-12-17 17:07 ` Richard Henderson
2009-12-17 17:47 ` malc
2009-12-17 18:09 ` Richard Henderson
2009-12-17 17:48 ` Richard Henderson
2009-12-18 15:40 ` malc
2009-12-18 16:05 ` Richard Henderson
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=761ea48b0912170737i54f25561q468a4a193114a903@mail.gmail.com \
--to=laurent.desnogues@gmail.com \
--cc=av1474@comtv.ru \
--cc=qemu-devel@nongnu.org \
--cc=rth@twiddle.net \
/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 NNTP newsgroup(s).