I wanted to learn how the ARM architecture works, any directions?

For better or worse, "ARM" is actually a family of "vaguely similar" architectures, and the architecture is for the CPU and some "close" peripherals (interrupt controller, sysTick timer)  That means that you pretty much need to pick a particular chip if you want to study things in the same depth that you'd get from studying the ATmega328p datasheet.

Most of the ARM chips aimed at "8-bit microcontroller replacement" are likely to have a "Cortex-M0plus" "core" (CM0) (smallest, cheapest, most power-efficient, simplest.)  The detailed architecture (including all the instructions) is described in the "ARMv6m Architecture Reference Manual": https://developer.arm.com/documentation/ddi0419/c/  I should probably also recommend Joseph Yiu's book: https://www.amazon.com/Definitive-Guide-Cortex-Cortex-M0-Processors/dp/0128032774  The documentation tends to be scattered through multiple documents, and is sometimes of the annoying form "this architecture supports the same thing as the bigger architecture, except for xxx"  (I think it's gotten better than it used to be, though.)  Some vendor chip-specific documentation will have information about the CPU (SAM3X8) while others simply refer to the ARM documentation (SAMD.)  Grr.

The ARM ARM does not describe any of the vendor-dependent peripherals; for them you need to refer to the vendors' data sheets or user guides or whatever they call them.

As for your doubts:

1. CPUs more advanced than the Cortex-M0+ add instructions (and variations of instructions.)  That means that assembly language written for a CM0 will probably run even on the more advanced CPUs, but not vis versa.  For example, the CM0 does not have a "divide" instruction.  Also, any assembly language that accesses vendor peripherals is probably not portable.
2. Complete lists of instructions are in the ARM ARM I mentioned earlier.  ARM instruction code (especially when compressed to "THUMB", as is done on the CM0) is ... not as simple as "these are the opcodes."  For example, there are three different "add" instruction formats depending on whether you're adding a register, a 3-bit (!) immediate constant, or an 8-bit immediate constant.
3. A CM0 CPU does not have floating point instructions.  An M4 ("Cortex M4F" - an ARMv7m architecture) CPU optionally does have floating point, which can be 32bit only, or include double-precision 64bit floats.

For better or worse, "ARM" is actually a family of "vaguely similar" architectures, and the architecture is for the CPU and some "close" peripherals (interrupt controller, sysTick timer)  That means that you pretty much need to pick a particular chip if you want to study things in the same depth that you'd get from studying the ATmega328p datasheet.

Most of the ARM chips aimed at "8-bit microcontroller replacement" are likely to have a "Cortex-M0plus" "core" (CM0) (smallest, cheapest, most power-efficient, simplest.)  The detailed architecture (including all the instructions) is described in the "ARMv6m Architecture Reference Manual": https://developer.arm.com/documentation/ddi0419/c/  I should probably also recommend Joseph Yiu's book: https://www.amazon.com/Definitive-Guide-Cortex-Cortex-M0-Processors/dp/0128032774  The documentation tends to be scattered through multiple documents, and is sometimes of the annoying form "this architecture supports the same thing as the bigger architecture, except for xxx"  (I think it's gotten better than it used to be, though.)  Some vendor chip-specific documentation will have information about the CPU (SAM3X8) while others simply refer to the ARM documentation (SAMD.)  Grr.

The ARM ARM does not describe any of the vendor-dependent peripherals; for them you need to refer to the vendors' data sheets or user guides or whatever they call them.

As for your doubts:

1. CPUs more advanced than the Cortex-M0+ add instructions (and variations of instructions.)  That means that assembly language written for a CM0 will probably run even on the more advanced CPUs, but not vis versa.  For example, the CM0 does not have a "divide" instruction.  Also, any assembly language that accesses vendor peripherals is probably not portable.
2. Complete lists of instructions are in the ARM ARM I mentioned earlier.  ARM instruction code (especially when compressed to "THUMB", as is done on the CM0) is ... not as simple as "these are the opcodes."  For example, there are three different "add" instruction formats depending on whether you're adding a register, a 3-bit (!) immediate constant, or an 8-bit immediate constant.
3. A CM0 CPU does not have floating point instructions.  An M4 ("Cortex M4F" - an ARMv7m architecture) CPU optionally does have floating point, which can be 32bit only, or include double-precision 64bit floats.