[Technical Question]About the c64c target

[complyue]

I'm interested in the c64c target, after seeing:

koka/kklib/include/kklib.h

Line 418 in 9f400dd

const kk_addr_t heap_mid; // mid point of the reserved heap address space (or 0 if the heap is not compressed)

koka/kklib/include/kklib.h

Lines 926 to 929 in 9f400dd

	#if (KK_INTB_SIZE==4)
	// compress to 32-bit offsets, ctx->heap_mid contains the mid-point in the heap so we can do signed extension
	a = a - ctx->heap_mid;
	#else

koka/kklib/include/kklib.h

Lines 956 to 958 in 9f400dd

	#if (KK_INTB_SIZE == 4)
	a = a + ctx->heap_mid;
	#else

koka/kklib/src/init.c

Lines 295 to 307 in 9f400dd

	kk_assert_internal(arena != 0 && arena_start != NULL);
	mi_heap_t* heap = mi_heap_new_in_arena(arena);
	ctx = (kk_context_t*)mi_heap_zalloc(heap, sizeof(kk_context_t));
	kk_assign_const(kk_heap_t,ctx->heap) = heap;
	kk_assign_const(void*, ctx->heap_start) = arena_start;
	kk_addr_t arena_start_addr;
	#if KK_CHERI
	arena_start_addr = __builtin_cheri_address_get(arena_start);
	#else
	arena_start_addr = (kk_addr_t)arena_start;
	#endif
	kk_assign_const(kk_addr_t, ctx->heap_mid) = arena_start_addr + (kk_addr_t)(arena_size / 2);
	#elif defined(KK_MIMALLOC)

Looks like 32-bit "compressed" pointers (thus heaps) will be used on 64-bit systems by opting for the c64c target, I wonder how effective it turned out to be, are there benchmarks around this target?

Why the design decision to use a heap_mid in thread-local context?

---

I'm interested in this, mainly bcoz I'm sketching about "relative" pointers for zero-cost relocatable heaps, i.e. the memory cell would store a pointer value as relative to its own residential address. And the relative pointer values can obviously be "compressed" as well to 32-bit width, for heap-compression purpose.

Has this idea been explored by researchers yet? How would it compare to the already implemented heap_mid approach?

[complyue]

Also asked at https://langdev.stackexchange.com/q/3857/426

Got insightful answer there. It looks like heap-size-limit due to compressed bit-width of ptrs will still be there, but as Koka tends to allocate nested heaps per supported by mimalloc, it (relative compressed pointers) may be of more value as for Koka?