PTX,... does comparing a bit either a 0 or 1 take 64 bits?

Is it faster to compare a bit than an int?

I wish to try something new along with learning ASM.


GPUs are 32-bit machines, with minimal extensions to facilitate 64-bit addressing. A general-purpose register in the GPU comprises 32 bits. 64-bit integer operations are emulated using pairs of registers for the operands.

Therefore, a 32-bit ‘int’ is the most appropriate integer data type, whether at the level of the high-level language (a C++ variant) or at PTX level. Unless there is a demonstrated need for an integer to be some other width integer type, or an unsigned type, integer data is best represented as ‘int’.

Note that by standard C++ semantics, any operations on integer data narrower than ‘int’ has to be widened to ‘int’ first before being incorporated into an expression.

Predicates are stored in what are essentially 1-bit registers, and the GPU has some instructions to perform simple logic (such as AND, OR) on predicates. This capability is frequently used for the implementation of branches with compound branch conditions.

Yes, thanks.
Though your implementation needs verifying Intellectually and Copyright acquired.
£5,000,000,000 for Copyright use… per processor version.

How are people getting on with multiplication? Does your shift register memory buffer look good?
There is a lot involved, classic designs.
It’s buffer size vs cost and speed.
I’m here only by… not personal to positive technologist’s.