Uploaded March 2012 | Updated September 2026, 2 weeks ago
In this tutorial we're going to look at 10 of the basic SSE data movement instructions. There's many more but these 10 will allow us to move pretty much whatever we want with SSE registers. Most of this tutorial is spent on looking at a very interesting topic, data alignment. Some of the SSE data movement instructions require the memory operands to be aligned (alignment never affects registers since registers have no address!). It's very important to realise that the data and code segments are the same to the CPU, you can align instructions in the code segment as easily as definitions in the data segment. Try placing "align (16)" just prior to the label at the top of a loop and you might get a free performance boost!
We also look a little alignment in the code segment. Basically you want to use the Aligned SSE instructions to move data from and to memory if you can since they are quicker. There's SSE instructions for moving integers, single precision floats (real4) and double precision floats (real8).
There's only a very small difference between moving 4 real4's or 2 real8's into or out of SSE registers but it's good to maintain the correct instructions as per your data size since there's a small performance penalty for using a register against what the CPU expects.
In this tutorial we're going to look at 10 of the basic SSE data movement instructions. There's many more but these 10 will allow us to move pretty much whatever we want with SSE registers. Most of this tutorial is spent on looking at a very interesting topic, data alignment. Some of the SSE data movement instructions require the memory operands to be aligned (alignment never affects registers since registers have no address!). It's very important to realise that the data and code segments are the same to the CPU, you can align instructions in the code segment as easily as definitions in the data segment. Try placing "align (16)" just prior to the label at the top of a loop and you might get a free performance boost!
We also look a little alignment in the code segment. Basically you want to use the Aligned SSE instructions to move data from and to memory if you can since they are quicker. There's SSE instructions for moving integers, single precision floats (real4) and double precision floats (real8).
There's only a very small difference between moving 4 real4's or 2 real8's into or out of SSE registers but it's good to maintain the correct instructions as per your data size since there's a small performance penalty for using a register against what the CPU expects.
