For ARM I implemented several optimizations to avoid put-into-buffer overhead, see comments above
putInt() in IonAssemblerBufferWithConstantPools.h. Subsequently, assembly time on ARM was not much of a factor any more, and profiling did not show
hasSpaceForInsts() to take significant time on that platform.
It is possible that the situation is different on ARM64 for other reasons, or that
hasSpaceForInsts() has since grown more complex again and takes more time now, or that I did poor profiling work, but it would be nice to have profiling data to support this bug.
As I wrote in bug 1443082, the vixl assembler is (based on profiling evidence anyway, and for webassembly compilation) a very poor performer, and optimizing that first might be important to get meaningful profiles for