Bitwise Shift Operations

[matsp]

If it needs tons of operations for stuff that x86 does with one opcode, isn't it then a very bad assembly language for learning purposes? If you'd leave the float/xmm operations out then x86 itself is a very simple assembly language. Why would someone implement an assembly language without a processor anyway?

I am just curious.

No, because a small set of instructions teach you more about how you do things in assembler, rather than becoming a "single operation".

Yes, all modern processors can do multiply, all manners of shifts, and most do divide in one instruction too. But that's not necessarily going to teach you a great deal.

--
Mats

[maxorator]

If you need to finish an application in x86 assembly, then you won't be feeling you have too many different instructions, trust me.

[matsp]

If you need to finish an application in x86 assembly, then you won't be feeling you have too many different instructions, trust me.

That's certainly not what I said.

But if you have to FIRST teach a bunch of students 46 different basic instructions [each with 6-10 different addressing modes], you may spend most of the first two weeks just going through what each instruction does.

But don't you think there really is a few too many ways of doing the same thing in x86:
For most things, you can left-shift a number 4 bits by:

rol eax, 4

lsl eax, 4

lea eax, [eax*4]
lea eax, [eax*4]

imul eax, 16

And those are only the most obvious different variants (and not covering MMX/SSE alternatives). The rotate/shift operations are obviously the fastest, and most obvious.

And to the original poster: adding a number to itself should have the same result as a shift - it's a more complex operation in hardware, but aside from that, it's equivalent.

--
Mats

[laserlight]

And to the original poster: adding a number to itself should have the same result as a shift - it's a more complex operation in hardware, but aside from that, it's equivalent.

The original poster already understood that and stated as such in the original post. The real question is about right shifting.

[maxorator]

But don't you think there really is a few too many ways of doing the same thing in x86:
For most things, you can left-shift a number 4 bits by:

rol eax, 4

lsl eax, 4

lea eax, [eax*4]
lea eax, [eax*4]

imul eax, 16

Well there's an optimal way to do everything. rol is the best for normal shifting, imul is good for normal multiplication (yeah, you can multiplicate with other numbers than 1,2,4,8,16... too), lea is good for putting simple calculations to few operations and so on. Every opcode is very useful when used in the right place.