Roll my eyes every time I see a "Considered Harmful" headline.
As for SIMD, it's a huge benefit when used in the right context. I applied it image processing and video compression algorithms in the past, with significant performance gains.
As for SIMD, it's a huge benefit when used in the right context. I applied it image processing and video compression algorithms in the past, with significant performance gains.
Me too.
> As for SIMD, it's a huge benefit when used in the right context.
Sure, but as the article points out, that benefit could be even huger when used on a "proper" vector architecture with veeeery wide vector registers that do not also double as not-very-wide scalar registers. "The SIMD instructions execute 10 to 20 times more instructions than RV32V because each SIMD loop does only 2 or 4 elements instead of 64 in the vector case."
I think adding SIMD instructions to x86 was a good trade-off at the time, but I also think the authors are correct that new ISAs designed now are better off with a vector architecture like they propose. In the end it's apples vs. oranges because the two contexts are not comparable.
It all started with the famous Edsger Dijkstra paper titled “Go To Statement Considered Harmful” [1], which lead to an endless series of subsequent papers later patterned after its title [2].
I agree with the sentiment that this title pattern is overused currently, to the point of being cliche — with perhaps a bit of presumptuousness as well, due to the implicit suggestion that the author’s claim of “Considered Harmful” will withstand the test of time as well as Dijkstra’s paper (though perhaps I’m reading too much into it, in this case).
Same. Please, when you write a title for your article, be specific about what you think is wrong. "X Considered Harmful" is lazy writing and devalues any substantive argument you make because it's become a cliche to the point where it's almost anti-clickbait.
GUILTY ADMISSION: a related trap I started falling into when writing content for work (email, documents, whatever) and I was in a hurry for a subject line or title was "Thoughts on X". That's right, "thoughts": I communed with the creator and distilled them from on high. Yeah, ain't nobody got time to read that.
I'm always dubious of claims like this. I think in general the architecture would need to turn it back into the highest SIMD code it supports at instruction decode.