At instruction level, the execution of instructions is an iterative process, so like for any other iterative process parallelism or pipelining or both parallelism and pipelining may be used.
Most modern CPUs use both parallelism and pipelining in the execution of instructions. If you have e.g. a stream of multiply instructions, the CPU may have for example 2 multiply pipelines, where each multiply pipeline has 4 stages. The incoming multiply instruction stream is divided into 2 substreams, which are dispatched in parallel to the 2 multiply pipelines, so 2 multiply instructions are initiated in each clock cycle, which makes the example CPU superscalar. The multiply instructions are completed after 4 clock cycles, which is their latency, when they exit the pipeline. Thus, in the example CPU, 8 multiply instructions are simultaneously executed in each clock cycle, in various stages of the 2 parallel pipelines.
Whenever you have independent iterations, e.g. what looks like a "for" loop in the source program, where there are no dependencies between distinct executions of the loop body, the iterations can be executed in 4 different ways, sequentially, interleaved, pipelined or in parallel.
The last 3 ways can provide an acceleration in comparison with the sequential execution of the iteration. In the last 2 ways there is simultaneous execution of multiple iterations or multiple parts of an iteration.
For parallel execution of the iteration (like in OpenMP "parallel for" or like in NVIDIA CUDA), a thread must be created for each iteration execution and all threads are launched to be executed in parallel by multiple hardware execution units.
For pipelined execution of the iteration, the iteration body is partitioned in multiple consecutive blocks that use as input data the output data of the previous block (which may need adding additional storage variables, to separate output from input for each block), then a thread must be created for each such block that implements a part of the iteration, and then all such threads are launched to be executed in parallel by multiple hardware execution units.
These 2 ways of organizing simultaneous work, pipelined execution and parallel execution are applicable to any kind of iterative process. Two of the most important such iterative processes are the execution of a stream of instructions and the implementation of an array operation, which performs some operation on all the elements of an array. In both cases one can use a combination of pipelining and parallelism to achieve maximum speed. For these 2 cases, sometimes the terms "instruction-level parallelism and pipelining" and "data-level parallelism and pipelining" are used. The second of these 2 terms is misleading, because not data are executed in parallel or pipelined, but the iterations that process data are executed in parallel or pipelined. For any case where pipelining may be used, it is important to recognize which is the iterative process that can be implemented in this way. For unrelated tasks a.k.a. processes a.k.a. threads, only 3 ways of execution are available: sequential, interleaved and in parallel. The 4th way of execution, pipelined, is available only for iterations, where the difference between iterations and unrelated tasks is that each iteration executes the same program (i.e. the loop body, when the iteration is written with the sequential loop syntax).