Make heap large chunk bit twiddling branchless (#4526)

* Make heap large chunk bit twiddling branchless

We use some bit twiddling operations for setting some flags on large
chunks in actor heaps. Before this commit, the functionality was
implemented with a ternary operator which would result in a branch.

This commit changes the logic to use bit shifts instead to avoid
the branching.

* use `#define` preprocessor macros

src/libponyrt/mem/heap.c

@@ -46,9 +46,12 @@ typedef struct small_chunk_t
 #define SMALL_CHUNK_SIZECLASS_BITMASK (uintptr_t)0x1C
 // TODO: figure out how to calculate the `2` from `SMALL_CHUNK_SIZECLASS_BITMASK` at compile time
 #define SMALL_CHUNK_SIZECLASS_SHIFT (uintptr_t)2
-#define LARGE_CHUNK_SLOT_BITMASK (uintptr_t)0x4
-#define LARGE_CHUNK_SHALLOW_BITMASK (uintptr_t)0x8
-#define LARGE_CHUNK_FINALISER_BITMASK (uintptr_t)0x10
+#define LARGE_CHUNK_SLOT_BITMASK_SHIFT_AMOUNT (uintptr_t)2
+#define LARGE_CHUNK_SLOT_BITMASK ((uintptr_t)0x1 << LARGE_CHUNK_SLOT_BITMASK_SHIFT_AMOUNT)
+#define LARGE_CHUNK_SHALLOW_BITMASK_SHIFT_AMOUNT (uintptr_t)3
+#define LARGE_CHUNK_SHALLOW_BITMASK ((uintptr_t)0x1 << LARGE_CHUNK_SHALLOW_BITMASK_SHIFT_AMOUNT)
+#define LARGE_CHUNK_FINALISER_BITMASK_SHIFT_AMOUNT (uintptr_t)4
+#define LARGE_CHUNK_FINALISER_BITMASK ((uintptr_t)0x1 << LARGE_CHUNK_FINALISER_BITMASK_SHIFT_AMOUNT)
 #define CHUNK_M_BITMASK ~(CHUNK_TYPE_BITMASK | CHUNK_NEEDS_TO_BE_CLEARED_BITMASK | SMALL_CHUNK_SIZECLASS_BITMASK | LARGE_CHUNK_SLOT_BITMASK | LARGE_CHUNK_SHALLOW_BITMASK | LARGE_CHUNK_FINALISER_BITMASK)
 
 enum
@@ -172,7 +175,10 @@ static void set_large_chunk_slot(large_chunk_t* chunk, uint32_t slot)
 {
   // `!!` to normalize to 1 or 0
   slot = !!slot;
-  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_SLOT_BITMASK) | (slot == 1 ? LARGE_CHUNK_SLOT_BITMASK : 0));
+  // left shift size to get bits in the right spot for OR'ing into `chunk->m`
+  slot = slot << LARGE_CHUNK_SLOT_BITMASK_SHIFT_AMOUNT;
+  pony_assert(slot == LARGE_CHUNK_SLOT_BITMASK || slot == 0);
+  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_SLOT_BITMASK) | slot);
 }
 
 static uint32_t get_large_chunk_shallow(large_chunk_t* chunk)
@@ -185,7 +191,10 @@ static void set_large_chunk_shallow(large_chunk_t* chunk, uint32_t shallow)
 {
   // `!!` to normalize to 1 or 0
   shallow = !!shallow;
-  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_SHALLOW_BITMASK) | (shallow == 1 ? LARGE_CHUNK_SHALLOW_BITMASK : 0));
+  // left shift size to get bits in the right spot for OR'ing into `chunk->m`
+  shallow = shallow << LARGE_CHUNK_SHALLOW_BITMASK_SHIFT_AMOUNT;
+  pony_assert(shallow == LARGE_CHUNK_SHALLOW_BITMASK || shallow == 0);
+  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_SHALLOW_BITMASK) | shallow);
 }
 
 static uint32_t get_large_chunk_finaliser(large_chunk_t* chunk)
@@ -198,7 +207,10 @@ static void set_large_chunk_finaliser(large_chunk_t* chunk, uint32_t finaliser)
 {
   // `!!` to normalize to 1 or 0
   finaliser = !!finaliser;
-  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_FINALISER_BITMASK) | (finaliser == 1 ? LARGE_CHUNK_FINALISER_BITMASK : 0));
+  // left shift size to get bits in the right spot for OR'ing into `chunk->m`
+  finaliser = finaliser << LARGE_CHUNK_FINALISER_BITMASK_SHIFT_AMOUNT;
+  pony_assert(finaliser == LARGE_CHUNK_FINALISER_BITMASK || finaliser == 0);
+  ((chunk_t*)chunk)->m = (char*)(((uintptr_t)((chunk_t*)chunk)->m & ~LARGE_CHUNK_FINALISER_BITMASK) | finaliser);
 }
 
 static size_t get_small_chunk_size(small_chunk_t* chunk)