qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Aurelien Jarno <aurelien@aurel32.net>
To: Richard Henderson <rth@twiddle.net>
Cc: qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [PATCH 03/22] tcg-i386: Tidy ext8u and ext16u operations.
Date: Wed, 19 May 2010 08:47:13 +0200	[thread overview]
Message-ID: <20100519064713.GC25432@ohm.aurel32.net> (raw)
In-Reply-To: <d3bd14117b5ecc639dcad07f997c86581cc03bf8.1272479073.git.rth@twiddle.net>

Hi,

On Tue, Apr 13, 2010 at 03:59:20PM -0700, Richard Henderson wrote:
> Define OPC_MOVZBL and OPC_MOVZWL.  Factor opcode emission to
> separate functions.  Don't restrict the input register to the
> low 4 "q" registers; emit an AND instead if needed.

I am fine about the cleaning part, but I don't know what to think about
the constraints change.

The reg allocator is able to issue move if needed, so the only
improvement this patch is for doing a ext8u on both "q" registers.

OTOH the reg allocator knows this situation and will try to avoid this
situation during the allocation. Cheating on the reg allocator might
have some wrong effects, especially after your patch "Allocate
call-saved registers first". I am thinking of the scenario where the
value is in memory (which is likely to be the case given the limited 
number of registers), it will be likely loaded in a "r" register (they
are now at the top priority), and then ext8u will be called, which will 
issue "mov" + "and" instructions instead of a "movzbl" instruction.

If there are still cases to optimize I think it should be done in the
reg allocator instead, so it could benefit all ops and all targets. I
have started to play on that with Laurent Desnogues, I have a few more
ideas how to improve it, but unfortunately I don't have time to code
them.

All of that is purely theoretical. Do you know how does it behave in 
practice?

> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
>  tcg/i386/tcg-target.c |   68 ++++++++++++++++++++++++++++++------------------
>  1 files changed, 42 insertions(+), 26 deletions(-)
> 
> diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c
> index 359f81b..2cc1191 100644
> --- a/tcg/i386/tcg-target.c
> +++ b/tcg/i386/tcg-target.c
> @@ -161,6 +161,11 @@ static inline int tcg_target_const_match(tcg_target_long val,
>          return 0;
>  }
>  
> +#define P_EXT   0x100 /* 0x0f opcode prefix */
> +
> +#define OPC_MOVZBL	(0xb6 | P_EXT)
> +#define OPC_MOVZWL	(0xb7 | P_EXT)
> +
>  #define ARITH_ADD 0
>  #define ARITH_OR  1
>  #define ARITH_ADC 2
> @@ -194,8 +199,6 @@ static inline int tcg_target_const_match(tcg_target_long val,
>  #define JCC_JLE 0xe
>  #define JCC_JG  0xf
>  
> -#define P_EXT   0x100 /* 0x0f opcode prefix */
> -
>  static const uint8_t tcg_cond_to_jcc[10] = {
>      [TCG_COND_EQ] = JCC_JE,
>      [TCG_COND_NE] = JCC_JNE,
> @@ -288,6 +291,27 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, int arg,
>      tcg_out_modrm_offset(s, 0x89, arg, arg1, arg2);
>  }
>  
> +static void tcg_out_ext8u(TCGContext *s, int dest, int src)
> +{
> +    if (src >= 4) {
> +        tcg_out_mov(s, dest, src);
> +        if (dest >= 4) {
> +            tcg_out_modrm(s, 0x81, ARITH_AND, dest);
> +            tcg_out32(s, 0xff);
> +            return;
> +        }
> +        src = dest;
> +    }
> +    /* movzbl */
> +    tcg_out_modrm(s, OPC_MOVZBL, dest, src);
> +}
> +
> +static inline void tcg_out_ext16u(TCGContext *s, int dest, int src)
> +{
> +    /* movzwl */
> +    tcg_out_modrm(s, OPC_MOVZWL, dest, src);
> +}
> +
>  static inline void tgen_arithi(TCGContext *s, int c, int r0, int32_t val, int cf)
>  {
>      if (!cf && ((c == ARITH_ADD && val == 1) || (c == ARITH_SUB && val == -1))) {
> @@ -300,11 +324,9 @@ static inline void tgen_arithi(TCGContext *s, int c, int r0, int32_t val, int cf
>          tcg_out_modrm(s, 0x83, c, r0);
>          tcg_out8(s, val);
>      } else if (c == ARITH_AND && val == 0xffu && r0 < 4) {
> -        /* movzbl */
> -        tcg_out_modrm(s, 0xb6 | P_EXT, r0, r0);
> +        tcg_out_ext8u(s, r0, r0);
>      } else if (c == ARITH_AND && val == 0xffffu) {
> -        /* movzwl */
> -        tcg_out_modrm(s, 0xb7 | P_EXT, r0, r0);
> +        tcg_out_ext16u(s, r0, r0);
>      } else {
>          tcg_out_modrm(s, 0x81, c, r0);
>          tcg_out32(s, val);
> @@ -645,12 +667,10 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
>          tcg_out_modrm(s, 0xbf | P_EXT, data_reg, TCG_REG_EAX);
>          break;
>      case 0:
> -        /* movzbl */
> -        tcg_out_modrm(s, 0xb6 | P_EXT, data_reg, TCG_REG_EAX);
> +        tcg_out_ext8u(s, data_reg, TCG_REG_EAX);
>          break;
>      case 1:
> -        /* movzwl */
> -        tcg_out_modrm(s, 0xb7 | P_EXT, data_reg, TCG_REG_EAX);
> +        tcg_out_ext16u(s, data_reg, TCG_REG_EAX);
>          break;
>      case 2:
>      default:
> @@ -690,7 +710,7 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
>      switch(opc) {
>      case 0:
>          /* movzbl */
> -        tcg_out_modrm_offset(s, 0xb6 | P_EXT, data_reg, r0, GUEST_BASE);
> +        tcg_out_modrm_offset(s, OPC_MOVZBL, data_reg, r0, GUEST_BASE);
>          break;
>      case 0 | 4:
>          /* movsbl */
> @@ -698,7 +718,7 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
>          break;
>      case 1:
>          /* movzwl */
> -        tcg_out_modrm_offset(s, 0xb7 | P_EXT, data_reg, r0, GUEST_BASE);
> +        tcg_out_modrm_offset(s, OPC_MOVZWL, data_reg, r0, GUEST_BASE);
>          if (bswap) {
>              /* rolw $8, data_reg */
>              tcg_out8(s, 0x66); 
> @@ -850,12 +870,10 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args,
>      } else {
>          switch(opc) {
>          case 0:
> -            /* movzbl */
> -            tcg_out_modrm(s, 0xb6 | P_EXT, TCG_REG_EDX, data_reg);
> +            tcg_out_ext8u(s, TCG_REG_EDX, data_reg);
>              break;
>          case 1:
> -            /* movzwl */
> -            tcg_out_modrm(s, 0xb7 | P_EXT, TCG_REG_EDX, data_reg);
> +            tcg_out_ext16u(s, TCG_REG_EDX, data_reg);
>              break;
>          case 2:
>              tcg_out_mov(s, TCG_REG_EDX, data_reg);
> @@ -881,12 +899,10 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args,
>          tcg_out_mov(s, TCG_REG_EDX, addr_reg2);
>          switch(opc) {
>          case 0:
> -            /* movzbl */
> -            tcg_out_modrm(s, 0xb6 | P_EXT, TCG_REG_ECX, data_reg);
> +            tcg_out_ext8u(s, TCG_REG_ECX, data_reg);
>              break;
>          case 1:
> -            /* movzwl */
> -            tcg_out_modrm(s, 0xb7 | P_EXT, TCG_REG_ECX, data_reg);
> +            tcg_out_ext16u(s, TCG_REG_ECX, data_reg);
>              break;
>          case 2:
>              tcg_out_mov(s, TCG_REG_ECX, data_reg);
> @@ -1022,7 +1038,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
>          break;
>      case INDEX_op_ld8u_i32:
>          /* movzbl */
> -        tcg_out_modrm_offset(s, 0xb6 | P_EXT, args[0], args[1], args[2]);
> +        tcg_out_modrm_offset(s, OPC_MOVZBL, args[0], args[1], args[2]);
>          break;
>      case INDEX_op_ld8s_i32:
>          /* movsbl */
> @@ -1030,7 +1046,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
>          break;
>      case INDEX_op_ld16u_i32:
>          /* movzwl */
> -        tcg_out_modrm_offset(s, 0xb7 | P_EXT, args[0], args[1], args[2]);
> +        tcg_out_modrm_offset(s, OPC_MOVZWL, args[0], args[1], args[2]);
>          break;
>      case INDEX_op_ld16s_i32:
>          /* movswl */
> @@ -1177,10 +1193,10 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
>          tcg_out_modrm(s, 0xbf | P_EXT, args[0], args[1]);
>          break;
>      case INDEX_op_ext8u_i32:
> -        tcg_out_modrm(s, 0xb6 | P_EXT, args[0], args[1]);
> +        tcg_out_ext8u(s, args[0], args[1]);
>          break;
>      case INDEX_op_ext16u_i32:
> -        tcg_out_modrm(s, 0xb7 | P_EXT, args[0], args[1]);
> +        tcg_out_ext16u(s, args[0], args[1]);
>          break;
>  
>      case INDEX_op_setcond_i32:
> @@ -1275,8 +1291,8 @@ static const TCGTargetOpDef x86_op_defs[] = {
>  
>      { INDEX_op_ext8s_i32, { "r", "q" } },
>      { INDEX_op_ext16s_i32, { "r", "r" } },
> -    { INDEX_op_ext8u_i32, { "r", "q"} },
> -    { INDEX_op_ext16u_i32, { "r", "r"} },
> +    { INDEX_op_ext8u_i32, { "r", "r" } },
> +    { INDEX_op_ext16u_i32, { "r", "r" } },
>  
>      { INDEX_op_setcond_i32, { "q", "r", "ri" } },
>      { INDEX_op_setcond2_i32, { "r", "r", "r", "ri", "ri" } },
> -- 
> 1.6.6.1
> 
> 
> 
> 

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

  reply	other threads:[~2010-05-19 17:37 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-04-28 18:24 [Qemu-devel] [PATCH 00/22] tcg-i386 cleanup and improvement, v2 Richard Henderson
2010-04-13 22:23 ` [Qemu-devel] [PATCH 01/22] tcg-i386: Allocate call-saved registers first Richard Henderson
2010-05-19  6:46   ` Aurelien Jarno
2010-04-13 22:26 ` [Qemu-devel] [PATCH 02/22] tcg-i386: Tidy initialization of tcg_target_call_clobber_regs Richard Henderson
2010-05-19  6:46   ` Aurelien Jarno
2010-04-13 22:59 ` [Qemu-devel] [PATCH 03/22] tcg-i386: Tidy ext8u and ext16u operations Richard Henderson
2010-05-19  6:47   ` Aurelien Jarno [this message]
2010-05-19 18:31     ` Richard Henderson
2010-05-20 13:39       ` Aurelien Jarno
2010-05-20 14:04         ` Aurelien Jarno
2010-05-20 14:40           ` Richard Henderson
2010-05-20 18:50             ` Aurelien Jarno
2010-04-13 23:13 ` [Qemu-devel] [PATCH 04/22] tcg-i386: Tidy ext8s and ext16s operations Richard Henderson
2010-05-20 18:52   ` Aurelien Jarno
2010-04-14 14:58 ` [Qemu-devel] [PATCH 07/22] tcg-i386: Tidy move operations Richard Henderson
2010-04-14 15:06 ` [Qemu-devel] [PATCH 08/22] tcg-i386: Eliminate extra move from qemu_ld64 Richard Henderson
2010-04-14 15:26 ` [Qemu-devel] [PATCH 09/22] tcg-i386: Tidy jumps Richard Henderson
2010-04-14 15:38 ` [Qemu-devel] [PATCH 10/22] tcg-i386: Tidy immediate arithmetic operations Richard Henderson
2010-05-21  9:38   ` Aurelien Jarno
2010-04-14 17:16 ` [Qemu-devel] [PATCH 11/22] tcg-i386: Tidy non-immediate " Richard Henderson
2010-05-21  9:38   ` Aurelien Jarno
2010-04-14 17:20 ` [Qemu-devel] [PATCH 12/22] tcg-i386: Tidy movi Richard Henderson
2010-05-21  9:38   ` Aurelien Jarno
2010-04-14 17:59 ` [Qemu-devel] [PATCH 13/22] tcg-i386: Tidy push/pop Richard Henderson
2010-05-21  9:38   ` Aurelien Jarno
2010-04-14 18:02 ` [Qemu-devel] [PATCH 14/22] tcg-i386: Tidy calls Richard Henderson
2010-05-21  9:40   ` Aurelien Jarno
2010-04-14 18:04 ` [Qemu-devel] [PATCH 15/22] tcg-i386: Tidy ret Richard Henderson
2010-05-21  9:40   ` Aurelien Jarno
2010-04-14 18:07 ` [Qemu-devel] [PATCH 16/22] tcg-i386: Tidy setcc Richard Henderson
2010-05-21  9:40   ` Aurelien Jarno
2010-04-14 18:22 ` [Qemu-devel] [PATCH 17/22] tcg-i386: Tidy unary arithmetic Richard Henderson
2010-05-21  9:41   ` Aurelien Jarno
2010-04-14 18:29 ` [Qemu-devel] [PATCH 18/22] tcg-i386: Tidy multiply Richard Henderson
2010-05-21  9:41   ` Aurelien Jarno
2010-04-14 18:32 ` [Qemu-devel] [PATCH 19/22] tcg-i386: Tidy xchg Richard Henderson
2010-05-21  9:42   ` Aurelien Jarno
2010-04-14 19:08 ` [Qemu-devel] [PATCH 20/22] tcg-i386: Tidy lea Richard Henderson
2010-05-21  9:43   ` Aurelien Jarno
2010-04-14 20:29 ` [Qemu-devel] [PATCH 21/22] tcg-i386: Use lea for three-operand add Richard Henderson
2010-05-21  9:44   ` Aurelien Jarno
2010-04-28 17:31 ` [Qemu-devel] [PATCH 05/22] tcg-i386: Tidy bswap operations Richard Henderson
2010-04-28 17:38 ` [Qemu-devel] [PATCH 06/22] tcg-i386: Tidy shift operations Richard Henderson
2010-04-28 18:23 ` [Qemu-devel] [PATCH 22/22] tcg-i386: Tidy data16 prefixes Richard Henderson
2010-05-21  9:45   ` Aurelien Jarno
2010-05-17 18:26 ` [Qemu-devel] [PATCH 00/22] tcg-i386 cleanup and improvement, v2 Richard Henderson
2010-05-17 19:54   ` Aurelien Jarno
2010-05-21  9:46 ` Aurelien Jarno

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=20100519064713.GC25432@ohm.aurel32.net \
    --to=aurelien@aurel32.net \
    --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).