* [Qemu-devel] Fwd: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst
[not found] <dh3mgdbdpd5d06hw5i0nqofq.1411427440387@email.android.com>
@ 2014-09-23 18:17 ` Richard Henderson
2014-09-23 18:42 ` Paolo Bonzini
0 siblings, 1 reply; 2+ messages in thread
From: Richard Henderson @ 2014-09-23 18:17 UTC (permalink / raw)
To: qemu-devel
[-- Attachment #1: Type: text/plain, Size: 698 bytes --]
Ug. Failure to operate the mail program on my phone properly;
the following went to Paolo only instead of the list too.
r~
-------- Forwarded Message --------
Subject: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst
Date: Mon, 22 Sep 2014 16:10:40 -0700
From: Richard Henderson <rth@twiddle.net>
To: Paolo Bonzini <pbonzini@redhat.com>
We probably average less than 10 memory operations per TB, but we reserve space
for 640 of them.
Probably what ought to be fixed is tcg's entire notion of preallocated arrays.
Obstacks, or similar pooled memory, may well work much better.
I can certainly drop this patch, but Claudio was the only one to comment when
it was posted three weeks ago.
r~
[-- Attachment #2: Attached Message Part --]
[-- Type: text/plain, Size: 46 bytes --]
Paolo Bonzini <pbonzini@redhat.com> wrote:
[-- Attachment #3: Attached Message Part --]
[-- Type: text/plain, Size: 1277 bytes --]
Il 22/09/2014 22:57, Richard Henderson ha scritto:
> +QEMU_BUILD_BUG_ON(TCG_TARGET_NB_REGS > 32);
> +QEMU_BUILD_BUG_ON(NB_MMU_MODES > 8);
> +
> typedef struct TCGLabelQemuLdst {
> - bool is_ld:1; /* qemu_ld: true, qemu_st: false */
> - TCGMemOp opc:4;
> - TCGReg addrlo_reg; /* reg index for low word of guest virtual addr */
> - TCGReg addrhi_reg; /* reg index for high word of guest virtual addr */
> - TCGReg datalo_reg; /* reg index for low word to be loaded or stored */
> - TCGReg datahi_reg; /* reg index for high word to be loaded or stored */
> - int mem_index; /* soft MMU memory index */
> + TCGMemOp opc : 4;
> + bool is_ld : 1; /* qemu_ld: true, qemu_st: false */
> + TCGReg addrlo_reg : 5; /* reg index for low word of guest virtual addr */
> + TCGReg addrhi_reg : 5; /* reg index for high word of guest virtual addr */
> + TCGReg datalo_reg : 5; /* reg index for low word to be loaded or stored */
> + TCGReg datahi_reg : 5; /* reg index for high word to be loaded or stored */
> + unsigned mem_index : 3; /* soft MMU memory index */
> + /* 4 bits unused in 32-bit word */
Why? Are there more than 10 or so loads in the typical tb?
Paolo
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [Qemu-devel] Fwd: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst
2014-09-23 18:17 ` [Qemu-devel] Fwd: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst Richard Henderson
@ 2014-09-23 18:42 ` Paolo Bonzini
0 siblings, 0 replies; 2+ messages in thread
From: Paolo Bonzini @ 2014-09-23 18:42 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Il 23/09/2014 20:17, Richard Henderson ha scritto:
> Ug. Failure to operate the mail program on my phone properly;
> the following went to Paolo only instead of the list too.
And my reply too, which was basically "np, just asking".
Paolo
>
> -------- Forwarded Message --------
> Subject: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst
> Date: Mon, 22 Sep 2014 16:10:40 -0700
> From: Richard Henderson <rth@twiddle.net>
> To: Paolo Bonzini <pbonzini@redhat.com>
>
> We probably average less than 10 memory operations per TB, but we reserve space
> for 640 of them.
>
> Probably what ought to be fixed is tcg's entire notion of preallocated arrays.
> Obstacks, or similar pooled memory, may well work much better.
>
> I can certainly drop this patch, but Claudio was the only one to comment when
> it was posted three weeks ago.
>
>
> r~
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2014-09-23 18:42 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <dh3mgdbdpd5d06hw5i0nqofq.1411427440387@email.android.com>
2014-09-23 18:17 ` [Qemu-devel] Fwd: Re: [PULL 07/11] tcg: Compress TCGLabelQemuLdst Richard Henderson
2014-09-23 18:42 ` Paolo Bonzini
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).