qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <rth@twiddle.net>
To: Aurelien Jarno <aurelien@aurel32.net>
Cc: qemu-devel@nongnu.org, agraf@suse.de
Subject: Re: [Qemu-devel] [PATCH 07/35] tcg: Optionally sign-extend 32-bit arguments for 64-bit host.
Date: Mon, 14 Jun 2010 15:20:45 -0700	[thread overview]
Message-ID: <4C16AB3D.3000700@twiddle.net> (raw)
In-Reply-To: <20100610102250.GG26968@volta.aurel32.net>

On 06/10/2010 03:22 AM, Aurelien Jarno wrote:
>> -                  0, GET_TCGV_I32(ret), 2, args);
>> +                  (is_signed ? 0x2a : 0x00), GET_TCGV_I32(ret), 2, args);
> 
> Wouldn't it be better to actually pass the whole flag to
> tcg_gen_helper32(), so that we can in the future also support mixed
> signedness in arguments? Also doing it here looks like a bit like a
> magic constant.

I've fixed this.

>> +#if defined(TCG_TARGET_EXTEND_ARGS) && TCG_TARGET_REG_BITS == 64
>> +    for (i = 0; i < nargs; ++i) {
>> +        int is_64bit = sizemask & (1 << (i+1)*2);
>> +        int is_signed = sizemask & (2 << (i+1)*2);
>> +        if (!is_64bit) {
>> +            TCGv_i64 temp = tcg_temp_new_i64();
>> +            TCGv_i64 orig = MAKE_TCGV_I64(args[i]);
>> +            if (is_signed) {
>> +                tcg_gen_ext32s_i64(temp, orig);
>> +            } else {
>> +                tcg_gen_ext32u_i64(temp, orig);
>> +            }
>> +            args[i] = GET_TCGV_I64(temp);
>> +        }
>> +    }
>> +#endif /* TCG_TARGET_EXTEND_ARGS */
>> +
> 
> This part allocates a lot of temp variables, that will probably generate
> a lot of register spills during the code generation.
> 
> As we do that for all arguments anyway, wouldn't it be possible to do
> the extension in place? The value in the register is changed, but that
> should not have any effect as it is ignored anyway in other
> instructions.

It is *not* possible to do the extension in-place.  At least not without
changing the format of the INDEX_op_call opcode.

With the extension done during opcode generation, like this, ORIG has been
marked TCG_TYPE_I32 and TEMP gets marked TCG_TYPE_I64.  This matters when
it comes time to copy the arguments into place.  If we somehow extended 
ORIG in-place, we'd still use the wrong instruction to copy the value into
the argument list -- either tcg_out_mov or tcg_out_st would get the wrong
TYPE argument.

If we try to do the extension later, e.g. while copying the value into the
argument list, we'd need to have SIZEMASK available.  To do that, we'd need
to save SIZEMASK into INDEX_op_call's argument list somehow.  That, I think,
is a more invasive change.


r~

  parent reply	other threads:[~2010-06-14 22:21 UTC|newest]

Thread overview: 75+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-06-04 19:14 [Qemu-devel] [PATCH 00/35] S390 TCG target, version 2 Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 01/35] tcg-s390: Adjust compilation flags Richard Henderson
2010-06-09 22:53   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 02/35] s390x: Avoid _llseek Richard Henderson
2010-06-09 22:54   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 03/35] s390x: Don't use a linker script for user-only Richard Henderson
2010-06-09 22:54   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 04/35] tcg-s390: Compute is_write in cpu_signal_handler Richard Henderson
2010-06-09 22:54   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 05/35] tcg-s390: Icache flush is a no-op Richard Henderson
2010-06-09 22:55   ` Aurelien Jarno
2010-06-10 22:04     ` Richard Henderson
2010-06-11  6:46       ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 06/35] tcg-s390: Allocate the code_gen_buffer near the main program Richard Henderson
2010-06-09 22:59   ` Aurelien Jarno
2010-06-10 22:05     ` Richard Henderson
2010-06-11  7:31       ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 07/35] tcg: Optionally sign-extend 32-bit arguments for 64-bit host Richard Henderson
2010-06-10 10:22   ` Aurelien Jarno
2010-06-10 22:08     ` Richard Henderson
2010-06-14 22:20     ` Richard Henderson [this message]
2010-06-04 19:14 ` [Qemu-devel] [PATCH 08/35] s390: Update disassembler to the last GPLv2 from binutils Richard Henderson
2010-06-09 22:47   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 09/35] s390: Disassemble some general-instruction-extension insns Richard Henderson
2010-06-09 22:47   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 10/35] tcg-s390: New TCG target Richard Henderson
2010-06-10 10:24   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 11/35] tcg-s390: Tidy unimplemented opcodes Richard Henderson
2010-06-10 10:24   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 12/35] tcg-s390: Define TCG_TMP0 Richard Henderson
2010-06-10 10:25   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 13/35] tcg-s390: Tidy regset initialization; use R14 as temporary Richard Henderson
2010-06-10 10:26   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 14/35] tcg-s390: Rearrange register allocation order Richard Henderson
2010-06-10 10:26   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 15/35] tcg-s390: Query instruction extensions that are installed Richard Henderson
2010-06-10 10:28   ` Aurelien Jarno
2010-06-10 22:19     ` Richard Henderson
2010-06-11  8:06       ` Aurelien Jarno
2010-06-11 13:07         ` Richard Henderson
2010-06-12 11:57           ` Aurelien Jarno
2010-06-11 13:13         ` Richard Henderson
2010-06-13 10:49           ` Aurelien Jarno
2010-06-13 16:02             ` Richard Henderson
2010-06-13 16:44               ` Aurelien Jarno
2010-06-13 22:23                 ` Alexander Graf
2010-06-14 16:20                   ` Richard Henderson
2010-06-14 17:39                     ` Alexander Graf
2010-06-04 19:14 ` [Qemu-devel] [PATCH 16/35] tcg-s390: Re-implement tcg_out_movi Richard Henderson
2010-06-12 12:04   ` Aurelien Jarno
2010-06-13 23:19     ` Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 17/35] tcg-s390: Implement sign and zero-extension operations Richard Henderson
2010-06-12 12:32   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 18/35] tcg-s390: Implement bswap operations Richard Henderson
2010-06-12 12:32   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 19/35] tcg-s390: Implement rotates Richard Henderson
2010-06-12 12:33   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 20/35] tcg-s390: Use LOAD COMPLIMENT for negate Richard Henderson
2010-06-12 12:33   ` Aurelien Jarno
2010-06-04 19:14 ` [Qemu-devel] [PATCH 21/35] tcg-s390: Use the ADD IMMEDIATE instructions Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 22/35] tcg-s390: Use the AND " Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 23/35] tcg-s390: Use the OR " Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 24/35] tcg-s390: Use the XOR " Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 25/35] tcg-s390: Use the MULTIPLY " Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 26/35] tcg-s390: Tidy goto_tb Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 27/35] tcg-s390: Rearrange qemu_ld/st to avoid register copy Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 28/35] tcg-s390: Tidy tcg_prepare_qemu_ldst Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 29/35] tcg-s390: Tidy user qemu_ld/st Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 30/35] tcg-s390: Implement GUEST_BASE Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 31/35] tcg-s390: Use 16-bit branches for forward jumps Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 32/35] tcg-s390: Use the LOAD AND TEST instruction for compares Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 33/35] tcg-s390: Use the COMPARE IMMEDIATE instrucions " Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 34/35] tcg-s390: Use COMPARE AND BRANCH instructions Richard Henderson
2010-06-04 19:14 ` [Qemu-devel] [PATCH 35/35] tcg-s390: Enable compile in 32-bit mode Richard Henderson
2010-06-08 13:11 ` [Qemu-devel] Re: [PATCH 00/35] S390 TCG target, version 2 Alexander Graf

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=4C16AB3D.3000700@twiddle.net \
    --to=rth@twiddle.net \
    --cc=agraf@suse.de \
    --cc=aurelien@aurel32.net \
    --cc=qemu-devel@nongnu.org \
    /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).