AMD64 why only 8 new GPR and not 16 or 32?

[ensabah6]

x86 is known to be register-starved, with only 8 GPR

AMD64 extends x86 8 GPR's to 16GPR. Many RISC such as PowerpC uses 32 GPR, which makes me wonder why AMD didn't extend the registers to 32 or even 64 total GPR
(same can be said for SSE, from 8 to 16, why not 32 or 64?) Itanium has over 100

 

[Greg Bernhardt]

I thought AMD64 ISA under "long mode" with register extensions has a GPR width of 64.

As you prolly know the Itanium is a specialized processor that in itself can cost several hundred dollars.

 

[quicknote]

registers

The decision to extend the number of GPR's to 16 on AMD64 was not a random choice, but rather a carefully considered decision based on several factors. One of the main reasons for choosing 16 GPR's is to maintain compatibility with existing x86 software. Extending the registers to 32 or 64 would require significant changes in the instruction set and would not be backwards compatible with existing software.

Additionally, the x86 architecture was originally designed with a limited number of registers due to the constraints of early hardware. While other architectures, such as PowerPC, may have more GPR's, this does not necessarily make them more efficient. The x86 architecture has evolved over time to optimize the use of its limited number of registers and has proven to be highly efficient in many applications.

Moreover, increasing the number of GPR's does not necessarily result in a significant performance boost. In fact, it can even have a negative impact on performance due to the increased complexity of managing and accessing a larger number of registers.

It is also worth noting that AMD64 does have additional registers beyond the 16 GPR's, such as the SSE registers, which can be used for specific tasks that require more registers. These registers are optimized for vector operations and offer performance benefits in certain applications.

Ultimately, the decision to extend the number of GPR's to 16 on AMD64 was a well-considered choice that balances compatibility, efficiency, and performance. While other architectures may have more registers, the x86 architecture has proven to be highly successful and efficient in its design.