From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mail-wm1-f51.google.com (mail-wm1-f51.google.com [209.85.128.51]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 746D635975 for ; Thu, 26 Jun 2025 15:16:19 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=209.85.128.51 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1750950981; cv=none; b=o82vDy3siY/O62G38so425gwXzCh1ZrzQFwAoTMG68C02k2OaA1/3tqFblxGk3dU5WRoyFOyB0Dqnk2H2pj9EAPTQvYEM5ifs42YJszmEKkS/GgdU51thw7QMem0fQlJEK8s75i8bklXkF53Z7llbSST+bem+5e9kTHniH3iGTY= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1750950981; c=relaxed/simple; bh=9znENFBWpJKE1Q1Cxka1I7osJnjGaAfYOcZFp5dZK0E=; h=Message-ID:Date:MIME-Version:Subject:To:Cc:References:From: In-Reply-To:Content-Type; b=JJQUnzxKwBo6HZjFjQ1GzNTH9PhOte1WSdocX0Z0HJnv8iAYpinD60wrA2A2MqMvFDn6J2sa3HddOy8VUhF1pIOkMBYnKVNbOuNg48zKS1Wfv/fbPU01PpQeldNVlFBDUbuyBy+ge1smiL71vCTKVN93vsHBW3Xf1savzzwG7hw= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=gmail.com; spf=pass smtp.mailfrom=gmail.com; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.b=RkafChEE; arc=none smtp.client-ip=209.85.128.51 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=gmail.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.b="RkafChEE" Received: by mail-wm1-f51.google.com with SMTP id 5b1f17b1804b1-451dbe494d6so11593355e9.1 for ; Thu, 26 Jun 2025 08:16:19 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20230601; t=1750950978; x=1751555778; darn=vger.kernel.org; h=content-transfer-encoding:in-reply-to:from:content-language :references:cc:to:subject:user-agent:mime-version:date:message-id :from:to:cc:subject:date:message-id:reply-to; bh=ZYbv5E+bbRXfoXv2F6AaS7viJ98wzUfV0UPEECnLnoo=; b=RkafChEEmq2eVPwk0VS5Vxsb3jrFv+D0FEDyU0t0F2qZMqX+UAj8h881DobHqErck6 1zUK+2F+xOKvVfJ4Eawipj/Em8iS7PyrY01v4TwOW20rdv2I2uqjZVVn1/+EXI6QbiPp UgfiA17VMuJ1XdDHcjL0cQsDWTbDHIhWynkDfulrgcFtaEZxKLDkuYQkXOoktpVffOni Ad9X3QZZdVNijKlFPl6e+R9kP2uHSQD6IBQ1k9KduRiA7DLxiER3SRLYam7DRmbGwA9m TVfy0uFvzvf6SiNjPKo4xptV6GALinZ0R4ha0SgTzQzTB55LVQkhHyHaR2jdKyPEHKCn XLbQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1750950978; x=1751555778; h=content-transfer-encoding:in-reply-to:from:content-language :references:cc:to:subject:user-agent:mime-version:date:message-id :x-gm-message-state:from:to:cc:subject:date:message-id:reply-to; bh=ZYbv5E+bbRXfoXv2F6AaS7viJ98wzUfV0UPEECnLnoo=; b=H9OXgPtj2sdA6pUx1YKaAGTVa2gjVvAJqm+gKOwiyoVkm49Im6qypbD8KHbeOqVPhp NaO/djDTHE8LnohjfWQ3piYna2XdZ6brdgyc/y3PlNkoVlWA2pjYf+qCDn/Fu815abYm 1J055DgmvdE5v54aZdDkEzobbgvM6Y9uT2odwi4QccCKQeIMmOShlo5jJKyVnzOAW6CN 03g3KoyGoTxDUyVlqI0Kkqd3CMpzn/r330Np/HnxvUTAAQjztkolnRwJxdAPh91Wkqyr kGTCotI5daeviXlmPhJ60IVW0ETu+C19V+hDAyMoySCMclljoOHXvq+LcdIghzgzlrAk nbCA== X-Forwarded-Encrypted: i=1; AJvYcCUkYXmXPI94O/utBhAPpmmSmTG8XmRB/a5YiYEkVsL6aC3fhHMZYG65BO3PHPFzsf9hmDrjhqLkRwW+IQDpJg==@vger.kernel.org X-Gm-Message-State: AOJu0YwlDkjdsX1Xa2jG6JSmwBXbC7ym6ragct9NgavlEJ9oR1XZDOYi 2itGPSOVhr15gWZoz4rXLLmqk3wnwH6wnqpr9RNhFKrXQMDt6RRNeiBK X-Gm-Gg: ASbGncvHTnXR7P3QuLwM3LzAZBnSUOctNFTTAwXfJwrrR2Q3FFZLpGw8E2lNL6qzC0J KZjs0FJXnQp5ecKVFidvt7eO4R+K2CLz7VLB4TAOSmxrljn9k1+xRwN6tLOBOOMeOgsyDBvVr0i Ht136h6+a3erK/H2xYl7tU4AyD5UpZ1vFAowCEhlhOMB9aOyPef83SjWN/ePe0F2kNHX5IHE3HN /MUj/66c7u7jvlPYf8tbBB58L88qVyXS0zUw8S/LQLjVpMcQYJIolVbGPDD3ySdlSWJl1/PZtu+ x9fQdW2AkOOZCpac9oL2L0c/hmuhCcD922UVwg1HDcAtOZs0yUzMqT1ZIC+Y/+oNtFMkdGs/rcr j/6NwqdONpXdsCxlgWJMVeGHxLE/3xBvyFYa2FvKrmv82g0KJuFeOUUdr+PAU84KSq2jAV16qP3 4Fh6AOlUMZRKXpTgeTOybuQYzpsUjaqvNJe6PrRq5cJ8iF6FsbuhuUzuZU X-Google-Smtp-Source: AGHT+IHkN6kst/2U0cI5ffj5ZypcrsRm6UjSAT4J4Pq42ykGcoxP+OW0hFReQqb4pCc9CMehOs7PNw== X-Received: by 2002:a05:600c:5306:b0:441:d2d8:bd8b with SMTP id 5b1f17b1804b1-45381aaa03fmr73036515e9.8.1750950977009; Thu, 26 Jun 2025 08:16:17 -0700 (PDT) Received: from ?IPV6:2a02:8388:e103:2700:ebc0:27a3:cde8:9846? (2a02-8388-e103-2700-ebc0-27a3-cde8-9846.cable.dynamic.v6.surfer.at. [2a02:8388:e103:2700:ebc0:27a3:cde8:9846]) by smtp.gmail.com with ESMTPSA id 5b1f17b1804b1-4538a3a72dasm22725015e9.16.2025.06.26.08.16.16 (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Thu, 26 Jun 2025 08:16:16 -0700 (PDT) Message-ID: <5f01c181-515e-4cff-bf8a-81c620886cd2@gmail.com> Date: Thu, 26 Jun 2025 17:16:15 +0200 Precedence: bulk X-Mailing-List: rust-for-linux@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [PATCH] drm/panic: Use a decimal fifo to avoid u64 by u64 divide To: Jocelyn Falempe Cc: Miguel Ojeda , Christian Schrefl , Arnd Bergmann , Russell King , Simona Vetter , Paolo Bonzini , David Airlie , Maxime Ripard , Maarten Lankhorst , Javier Martinez Canillas , Thomas Zimmermann , Linux ARM , dri-devel@lists.freedesktop.org, rust-for-linux References: <20250418165059.560503-1-jfalempe@redhat.com> <80d092a3-0210-48d8-97da-f9175260a5fc@redhat.com> Content-Language: en-US From: Andrei Lalaev In-Reply-To: <80d092a3-0210-48d8-97da-f9175260a5fc@redhat.com> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit On 26.06.25 16:19, Jocelyn Falempe wrote: > On 25/06/2025 00:18, Jocelyn Falempe wrote: >> On 24/06/2025 20:55, Andrei Lalaev wrote: >>> On 18.04.25 18:48, Jocelyn Falempe wrote: >>>> On 32bits ARM, u64/u64 is not supported [1], so change the algorithm >>>> to use a simple fifo with decimal digits as u8 instead. >>>> This is slower but should compile on all architecture. >>>> >>>> Link: https://lore.kernel.org/dri-devel/ CANiq72ke45eOwckMhWHvmwxc03dxr4rnxxKvx+HvWdBLopZfrQ@mail.gmail.com/ [1] >>>> Signed-off-by: Jocelyn Falempe >>>> --- >>>>   drivers/gpu/drm/drm_panic_qr.rs | 71 ++++++++++++++++++++++----------- >>>>   1 file changed, 48 insertions(+), 23 deletions(-) >>>> >>>> diff --git a/drivers/gpu/drm/drm_panic_qr.rs b/drivers/gpu/drm/ drm_panic_qr.rs >>>> index 6025a705530e..dd55b1cb764d 100644 >>>> --- a/drivers/gpu/drm/drm_panic_qr.rs >>>> +++ b/drivers/gpu/drm/drm_panic_qr.rs >>>> @@ -366,8 +366,48 @@ fn iter(&self) -> SegmentIterator<'_> { >>>>           SegmentIterator { >>>>               segment: self, >>>>               offset: 0, >>>> -            carry: 0, >>>> -            carry_len: 0, >>>> +            decfifo: Default::default(), >>>> +        } >>>> +    } >>>> +} >>>> + >>>> +/// Max fifo size is 17 (max push) + 2 (max remaining) >>>> +const MAX_FIFO_SIZE: usize = 19; >>>> + >>>> +/// A simple Decimal digit FIFO >>>> +#[derive(Default)] >>>> +struct DecFifo { >>>> +    decimals: [u8; MAX_FIFO_SIZE], >>>> +    len: usize, >>>> +} >>>> + >>>> +impl DecFifo { >>>> +    fn push(&mut self, data: u64, len: usize) { >>>> +        let mut chunk = data; >>>> +        for i in (0..self.len).rev() { >>>> +            self.decimals[i + len] = self.decimals[i]; >>>> +        } >>>> +        for i in 0..len { >>>> +            self.decimals[i] = (chunk % 10) as u8; >>>> +            chunk /= 10; >>>> +        } >>>> +        self.len += len; >>>> +    } >>>> + >>>> +    /// Pop 3 decimal digits from the FIFO >>>> +    fn pop3(&mut self) -> Option<(u16, usize)> { >>>> +        if self.len == 0 { >>>> +            None >>>> +        } else { >>>> +            let poplen = 3.min(self.len); >>>> +            self.len -= poplen; >>>> +            let mut out = 0; >>>> +            let mut exp = 1; >>>> +            for i in 0..poplen { >>>> +                out += self.decimals[self.len + i] as u16 * exp; >>>> +                exp *= 10; >>>> +            } >>>> +            Some((out, NUM_CHARS_BITS[poplen])) >>>>           } >>>>       } >>>>   } >>>> @@ -375,8 +415,7 @@ fn iter(&self) -> SegmentIterator<'_> { >>>>   struct SegmentIterator<'a> { >>>>       segment: &'a Segment<'a>, >>>>       offset: usize, >>>> -    carry: u64, >>>> -    carry_len: usize, >>>> +    decfifo: DecFifo, >>>>   } >>>>   impl Iterator for SegmentIterator<'_> { >>>> @@ -394,31 +433,17 @@ fn next(&mut self) -> Option { >>>>                   } >>>>               } >>>>               Segment::Numeric(data) => { >>>> -                if self.carry_len < 3 && self.offset < data.len() { >>>> -                    // If there are less than 3 decimal digits in the carry, >>>> -                    // take the next 7 bytes of input, and add them to the carry. >>>> +                if self.decfifo.len < 3 && self.offset < data.len() { >>>> +                    // If there are less than 3 decimal digits in the fifo, >>>> +                    // take the next 7 bytes of input, and push them to the fifo. >>>>                       let mut buf = [0u8; 8]; >>>>                       let len = 7.min(data.len() - self.offset); >>>>                       buf[..len].copy_from_slice(&data[self.offset..self.offset + len]); >>>>                       let chunk = u64::from_le_bytes(buf); >>>> -                    let pow = u64::pow(10, BYTES_TO_DIGITS[len] as u32); >>>> -                    self.carry = chunk + self.carry * pow; >>>> +                    self.decfifo.push(chunk, BYTES_TO_DIGITS[len]); >>>>                       self.offset += len; >>>> -                    self.carry_len += BYTES_TO_DIGITS[len]; >>>> -                } >>>> -                match self.carry_len { >>>> -                    0 => None, >>>> -                    len => { >>>> -                        // take the next 3 decimal digits of the carry >>>> -                        // and return 10bits of numeric data. >>>> -                        let out_len = 3.min(len); >>>> -                        self.carry_len -= out_len; >>>> -                        let pow = u64::pow(10, self.carry_len as u32); >>>> -                        let out = (self.carry / pow) as u16; >>>> -                        self.carry %= pow; >>>> -                        Some((out, NUM_CHARS_BITS[out_len])) >>>> -                    } >>>>                   } >>>> +                self.decfifo.pop3() >>>>               } >>>>           } >>>>       } >>>> >>>> base-commit: 74757ad1c105c8fc00b4cac0b7918fe3262cdb18 >>> >>> Hi Jocelyn, >>> >>> Apologies for reviving this old thread, but I'm still encountering >>> the same issue with the latest master (78f4e737a53e). >>> >>> When compiling this module for ARM32 (multi_v7_defconfig), >>> I get the following error: >>> >>>      ld.lld: error: undefined symbol: __aeabi_uldivmod >>>      >>> referenced by drm_panic_qr.rs:392 (drivers/gpu/drm/ drm_panic_qr.rs:392) >>>      >>>               drivers/gpu/drm/drm_panic_qr.o: (::next) in archive vmlinux.a >>>      >>> referenced by drm_panic_qr.rs:392 (drivers/gpu/drm/ drm_panic_qr.rs:392) >>>      >>>               drivers/gpu/drm/drm_panic_qr.o: (::next) in archive vmlinux.a >>>      >>> referenced by drm_panic_qr.rs:392 (drivers/gpu/drm/ drm_panic_qr.rs:392) >>>      >>>               drivers/gpu/drm/drm_panic_qr.o: (::next) in archive vmlinux.a >>>      >>> referenced 14 more times >>>      >>> did you mean: __aeabi_uidivmod >>>      >>> defined in: vmlinux.a(arch/arm/lib/lib1funcs.o) >>> >>> Since no one else has reported this in two months, I’m wondering >>> if this might be a configuration issue on my end. >> >> Ok, that's surprising, the lines 391 and 392 are: >> >> self.decimals[i] = (chunk % 10) as u8; >> chunk /= 10; >> >> So the compiler should be smart enough to do that without using a division. >> I will try to reproduce, and see if I can fix that. > > I reproduced the issues, it looks like clang doesn't do the optimization on ARM32: > > https://github.com/llvm/llvm-project/issues/37280 > > So I've made a quick test with the following changes, and it builds: > > > diff --git a/drivers/gpu/drm/drm_panic_qr.rs b/drivers/gpu/drm/drm_panic_qr.rs > index dd55b1cb764d..57bd3c6465bb 100644 > --- a/drivers/gpu/drm/drm_panic_qr.rs > +++ b/drivers/gpu/drm/drm_panic_qr.rs > @@ -381,6 +381,20 @@ struct DecFifo { >      len: usize, >  } > > +fn div10(val: u64) -> u64 > +{ > +    let val_h = val >> 32; > +    let val_l = val & 0xFFFFFFFF; > +    let b_h: u64 = 0x66666666; > +    let b_l: u64 = 0x66666667; > + > +    let tmp1 = val_h * b_l + ((val_l * b_l) >> 32); > +    let tmp2 = val_l * b_h + (tmp1 & 0xffffffff); > +    let tmp3 = val_h * b_h + (tmp1 >> 32) + (tmp2 >> 32); > + > +    tmp3 >> 2 > +} > + >  impl DecFifo { >      fn push(&mut self, data: u64, len: usize) { >          let mut chunk = data; > @@ -389,7 +403,7 @@ fn push(&mut self, data: u64, len: usize) { >          } >          for i in 0..len { >              self.decimals[i] = (chunk % 10) as u8; > -            chunk /= 10; > +            chunk = div10(chunk); >          } >          self.len += len; >      } > > > Best regards, > Compiles for me too on clang 20.1.6. Thanks a lot! -- Best regards, Andrei Lalaev