From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:44713) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1V9zso-0001hZ-Dz for qemu-devel@nongnu.org; Thu, 15 Aug 2013 11:54:47 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1V9zsj-0000O5-5w for qemu-devel@nongnu.org; Thu, 15 Aug 2013 11:54:42 -0400 Received: from hall.aurel32.net ([2001:470:1f0b:4a8::1]:58991) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1V9zsi-0000Nr-V3 for qemu-devel@nongnu.org; Thu, 15 Aug 2013 11:54:37 -0400 Date: Thu, 15 Aug 2013 17:54:34 +0200 From: Aurelien Jarno Message-ID: <20130815155434.GA19174@ohm.aurel32.net> References: <1375726045-20797-1-git-send-email-rth@twiddle.net> <1375726045-20797-3-git-send-email-rth@twiddle.net> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: <1375726045-20797-3-git-send-email-rth@twiddle.net> Subject: Re: [Qemu-devel] [PATCH for-next 2/8] tcg-i386: Try pc-relative lea for constant formation List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Richard Henderson Cc: qemu-devel@nongnu.org On Mon, Aug 05, 2013 at 08:07:19AM -1000, Richard Henderson wrote: > Use a 7 byte lea before the ultimate 10 byte movq. > > Signed-off-by: Richard Henderson > --- > tcg/i386/tcg-target.c | 19 ++++++++++++++++--- > 1 file changed, 16 insertions(+), 3 deletions(-) > > diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c > index 841bd75..456bd9e 100644 > --- a/tcg/i386/tcg-target.c > +++ b/tcg/i386/tcg-target.c > @@ -541,19 +541,32 @@ static inline void tcg_out_mov(TCGContext *s, TCGType type, > static void tcg_out_movi(TCGContext *s, TCGType type, > TCGReg ret, tcg_target_long arg) > { > + tcg_target_long diff; > + > if (arg == 0) { > tgen_arithr(s, ARITH_XOR, ret, ret); > return; > } else if (arg == (uint32_t)arg || type == TCG_TYPE_I32) { > tcg_out_opc(s, OPC_MOVL_Iv + LOWREGMASK(ret), 0, ret, 0); > tcg_out32(s, arg); > + return; > } else if (arg == (int32_t)arg) { > tcg_out_modrm(s, OPC_MOVL_EvIz + P_REXW, 0, ret); > tcg_out32(s, arg); > - } else { > - tcg_out_opc(s, OPC_MOVL_Iv + P_REXW + LOWREGMASK(ret), 0, ret, 0); > - tcg_out64(s, arg); > + return; > } Now that all the else parts end up with a return, it would improve readability to remove them and keep only the ifs. > + > + /* Try a 7 byte pc-relative lea before the 10 byte movq. */ > + diff = arg - ((tcg_target_long)s->code_ptr + 7); > + if (diff == (int32_t)diff) { > + tcg_out_opc(s, OPC_LEA | P_REXW, ret, 0, 0); > + tcg_out8(s, (LOWREGMASK(ret) << 3) | 5); > + tcg_out32(s, diff); > + return; > + } > + > + tcg_out_opc(s, OPC_MOVL_Iv + P_REXW + LOWREGMASK(ret), 0, ret, 0); > + tcg_out64(s, arg); > } > > static inline void tcg_out_pushi(TCGContext *s, tcg_target_long val) Otherwise it looks good. -- Aurelien Jarno GPG: 1024D/F1BCDB73 aurelien@aurel32.net http://www.aurel32.net