On Fri, Dec 21, 2018 at 01:07:21PM -0800, Matthew Wilcox wrote: > On Fri, Dec 21, 2018 at 11:46:16PM +0300, Cyrill Gorcunov wrote: > > Cast to unsigned char is needed in any case. And as far as I remember > > we've been using this multiplication trick for a really long time > > in x86 land. I'm out of sources right now but it should be somewhere > > in assembly libs. > > x86 isn't the only CPU. Some CPUs have slow multiplies but fast shifts. This is x86-64 patch, not some generic code. > Also loading 0x0101010101010101 into a register may be inefficient on > some CPUs. It is pretty efficient on x86-64. Moreover the self dependents as a |= a << b is a source for data hazards inside cpu engine. Anyway i'm not going to insist, just wanted to remind about such trick. Up to you what to choose.
