If you look at the set of hprof records generated by SA and hotspot for GC roots, they differ in a few areas. I'm uncertain why, or if this can lead to incorrect heap representation:

The following records are generated by both hotspot and SA:

HPROF_GC_ROOT_THREAD_OBJ   thread object
HPROF_GC_ROOT_JNI_GLOBAL     JNI global ref root
HPROF_GC_ROOT_JNI_LOCAL       JNI local ref

The following records are generated only by hotspot:

HPROF_GC_ROOT_STICKY_CLASS System class (HOTSPOT)
HPROF_GC_ROOT_JAVA_FRAME Java stack frame (HOTSPOT)
HPROF_GC_ROOT_NATIVE_STACK Native stack (HOTSPOT)

The following records are not generated by either SA or hotspot:

HPROF_GC_ROOT_UNKNOWN         unknown root
HPROF_GC_ROOT_MONITOR_USED    Busy monitor
HPROF_GC_ROOT_THREAD_BLOCK    Reference from thread block

Note there is also an HPROF_FRAME record that is generated by both and also an HPROF_TRACE record, which appears to be a stack trace referencing HPROF_FRAME records. These don't appear to be related to HPROF_GC_ROOT_JAVA_FRAME.

There are also HPROF_GC_CLASS_DUMP records generated by both. I'm not sure how these relate to HPROF_GC_ROOT_STICKY_CLASS, which is only generated by hotspot.