From mboxrd@z Thu Jan 1 00:00:00 1970 From: cotulla@yandex.ua Subject: Re: [DISCUSSION] Hexagon code inside kernel Date: Sat, 16 Feb 2013 16:39:49 +0400 Message-ID: <1163031361018389@web26d.yandex.ru> References: <1093431360938494@web22d.yandex.ru> <1433341360978513@web2h.yandex.ru> Mime-Version: 1.0 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=yandex.ru; s=mail; t=1361018389; bh=RvUemFU9dQ0hfmjrUXp0Ji8QkP2DjMdLlceED/UNWMY=; h=From:To:In-Reply-To:References:Subject:Date; b=jGBiSNuudDGknljvcUdVMhSNe9+la600FjZUfBa93MmIpVydmX9YYYK3wzFzABPGu gk5FB+mLgrclsUojolMgTnUAZx9RIe4YtKbVuTxV2nd1pGMAwuq+5owW8EwjOFRZlT 6DdViKQas9O87PPuRcpyqgK0huguJ9Isq0Mb4wGQ= In-Reply-To: Sender: linux-hexagon-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="windows-1252" To: linux-hexagon@vger.kernel.org Hi, > =9AFor the qdsp6v3 the effective clock rate was 300MHz per core, so y= es. > =9AIt might be even slower for v2, not sure. =9A(the chip clock rate = is 1.8 > =9AGHz, there are 6 interleaved cores, so 1.8/6 =3D 300 =9AThe power = savings > =9Aare not from the clock rate, but from the tiny transistor count. T= he > =9Aperformance efficiency is from keeping all of those transistors > =9Aconstantly wiggling, which is what the interleaved pipeline does.) Hm, I thought the maximum clock rate is 595.2 Mhz? Or 1.8 is another clock? But by changing this clock rate I can get different Q6 performance. > =9ANoo .. in fact, that slow clock rate meant it's more or less the s= ame > =9Aspeed as DRAM. =9ASo low penalty for cache miss. =9ABut there's al= so a bus > =9Abetween here and there, and I think it was selected for simplicity= not > =9Aspeed... I thought that it can be bus arbiter setting, but other masters like MO= DEM and APPS were almost idle during test. I think in that case Q6 can take whole bandwidth or is it still limited= somehow? > =9AOh, well, no, the VM wasn't created for linux, it was created for = all > =9Athe other junk they try to run on there (eg audio processing, nois= e > =9Acancellation, etc.), so that there would be a forward migration pa= th > =9Afor that software with each new chip generation. =9AWhether those = other > =9Asoftware parts actually used the thing, and in what time-frame or > =9Arelease .. hey .. typical big-company inter-divisional politics, > =9Abickering, etc. =9AThe truly daring wanted all those other parts t= o run > =9Aon the linux kernel. =9AThe old-school was like "hell no we don't = need > =9Ano stinkin OS, we'll hand code everything in assembly" which sound= s > =9Agreat until you have ten support staff helping them kludge up a re= ally > =9Areally badly designed home-grown linker/loader/scheduler/irq-handl= er. > =9ASometimes, really smart people just don't get it. Well, sometimes such code becomes too large and slow, while doesn't pro= vide universal solution for all cases. As well as more code =3D=3D more bugs inside and harder to follow logic= inside.=20 > =9ADon't know v2. But v3 had a 'real' MMU Hm, are you sure in that?=20 I had never seen any usage of it. As well as binutils registers definit= ion doesn't include any suitable registers for that.=20 > =9AGood, because the bootloader was going to be the other issue. Yes, in my case it's working :) But another guys who also want participate in this project with MSM8960= /APQ8064 they still can't run any unsigned code on Q6. In modern phones it's often locked from changes :( > =9AI'd done the patches for glibc (yes, they're publicly available on > =9Asome website, don't know if they got merged or not), got 98% of th= e > =9Amany hundreds of glibc unit tests to pass, including most or all o= f > =9Athe thread tests including TLS. Someone had bootstrapped hundreds = of > =9A.debs and both python and perl passed 100% of their tests. =9AI'm = sure > =9Ano one cares, but even guile worked, and I was about to start fidd= ling > =9Awith haskell :-) Good to hear that. Good job! So userspace support is rather good in common. Maybe we will try to compile and run Android on that 6 cores pff hardwa= re threads :D -Cotulla