If you need more control, what you can do is to access the lower-level hipsycl::algorithms library directly. These library functions mostly map 1:1 to the standard parallel algorithms, but their signature is slightly different:

• They have sycl::queue& as first argument
• They return sycl::event
• Some of them have some additional arguments to manage scratch memory that the algorithms might need.

Note that at some point in the future the hipsycl::algorithms namespace will be renamed to acpp::algorithms or similar. So maybe it is a good idea for you to introduce an alias here so that you only need to change one place in your code ;)

Currently I would strongly recommend using these algorithms exclusively with in-order queues. They are untested with out-of-order queues, and the signature of many algorithms currently lacks dependency events. So it is currently difficult to specify asynchronous dependencies.

Note that when you use custom queues and submit to your own set of runtime-selected devices, you can no longer rely on the automatic C++ memory management interception from --acpp-stdpar to work. Replacing C++ memory allocations with SYCL memory management as --acpp-stdpar does requires deciding on one particular device (or at least backend) as soon as the application starts up and starts allocating memory.
So you will have to do all memory allocations using explicit SYCL functionality, like in normal SYCL programs.