The x86 instruction set architecture (ISA) doesn't seem to be going anywhere. We've had people declaring the impending ascent of Arm or RISC-V for years now, and the market hasn't budged. By and large, people still want to buy Intel and AMD systems running x86 Windows. So, maybe what's needed isn't replacement but reform, and maybe that should be in the form of SHRINK.
That's not a new proposal—the International Symposium on Computer Architecture (ISCA) paper is dated June 2015, coming out of the 42nd annual symposium—but it might be new to you. It was to me. It was brought to my attention by LaurieWired on X, who points out that just 12 of the 1,000+ mnemonics (instructions) on modern CPUs make up 89% of compiled C/C++ code.
And even if we don't go down the C/C++ route (as some suggest that this might not cover all instruction bases), the SHRINK research paper shows there are lots of unused instructions over time accounting for all bases: "We consider instructions that belong to vector extensions in separate categories [than unused], because albeit unused, they are still under adoption and their UISes are likely to be used in the future."
The apparent over-abundance of instructions is often seen as both a boon and a downside to the x86 architecture. It's very versatile and can do a lot—thus why most consumer apps are built for it and why most home PCs use it—but it doesn't do it in the most efficient way. That's led competitors such as Arm to step in with seemingly more efficient chips, albeit chips that can struggle with app compatibility.
The solution to x86's problem of large swathes of mostly redundant and unused instructions, according to the researchers who made the paper, could be SHRINK, which "provides a method for recycling instructions (i.e., removing unused and/or infrequently-used instructions and strategically reassigning their encodings to more frequently used instructions), and a mechanism to emulate the removed instructions."
The interesting word there is "emulate." It seems like the idea is that a large chunk of instructions—the lesser-used ones—can be removed from the actual instruction set and instead emulated in software when needed.
The researchers "conclude that at least 40% of the x86 ISA, even after excluding multimedia extensions, could be emulated with minor performance overhead in the analyzed execution traces."
Other x86 debloating attempts have been made, such as Intel x86s, which sought to remove 16-bit and 32-bit instruction support. This was terminated just last year, though. LaurieWired points out that a big part of the reason why it's hard to debloat x86 is just because "a lot of it is IP / legal hell." The x86 ISA isn't just an Intel product, remember: It's largely a cross-licensed affair between Intel and AMD. I wouldn't fancy working through that mire.
And I suppose there's an argument that debloat isn't needed, too. While many like to point out the efficiency of Arm chips (for example) compared to x86 ones, we've seen in recent years that some x86 chips can be mighty efficient, too. Just look at Intel's latest Lunar Lake mobile chips.
Still, as my programmer friend would likely remind me, redundant and unused code isn't a good look. Maybe some SHRINKage wouldn't be a bad thing.
1. Best overall:
AMD Ryzen 7 9800X3D
2. Best budget:
Intel Core i5 13400F
3. Best mid-range:
AMD Ryzen 7 9700X
4. Best high-end:
AMD Ryzen 9 9950X3D
5. Best AM4 upgrade:
AMD Ryzen 7 5700X3D
6. Best CPU graphics:
AMD Ryzen 7 8700G
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
