Improve performance of newSmallArray#

This:
- Hoists part of the condition outside of the initialization loop in
  `stg_newSmallArrayzh`.
- Annotates one of the unlikely branches as unlikely, also in
  `stg_newSmallArrayzh`.
- Adds a couple of annotations to `allocateMightFail` indicating which
  branches are likely to be taken.

Together this gives about 5% improvement.
Signed-off-by: Michal Terepeta <michal.terepeta@gmail.com>

includes/Rts.h

@@ -58,7 +58,13 @@ extern "C" {
 #if __GNUC__ >= 4
 #define RTS_UNLIKELY(p) __builtin_expect((p),0)
 #else
-#define RTS_UNLIKELY(p) p
+#define RTS_UNLIKELY(p) (p)
+#endif
+
+#if __GNUC__ >= 4
+#define RTS_LIKELY(p) __builtin_expect(!!(p), 1)
+#else
+#define RTS_LIKELY(p) (p)
 #endif
 
 /* __builtin_unreachable is supported since GNU C 4.5 */

rts/PrimOps.cmm

@@ -403,7 +403,7 @@ stg_newSmallArrayzh ( W_ n /* words */, gcptr init )
 
     words = BYTES_TO_WDS(SIZEOF_StgSmallMutArrPtrs) + n;
     ("ptr" arr) = ccall allocateMightFail(MyCapability() "ptr",words);
-    if (arr == NULL) {
+    if (arr == NULL) (likely: False) {
         jump stg_raisezh(base_GHCziIOziException_heapOverflow_closure);
     }
     TICK_ALLOC_PRIM(SIZEOF_StgSmallMutArrPtrs, WDS(n), 0);
@@ -413,8 +413,11 @@ stg_newSmallArrayzh ( W_ n /* words */, gcptr init )
 
     // Initialise all elements of the array with the value in R2
     p = arr + SIZEOF_StgSmallMutArrPtrs;
+    // Avoid the shift for `WDS(n)` in the inner loop
+    W_ limit;
+    limit = arr + SIZEOF_StgSmallMutArrPtrs + WDS(n);
   for:
-    if (p < arr + SIZEOF_StgSmallMutArrPtrs + WDS(n)) (likely: True) {
+    if (p < limit) (likely: True) {
         W_[p] = init;
         p = p + WDS(1);
         goto for;

rts/sm/Storage.c

@@ -856,7 +856,7 @@ allocateMightFail (Capability *cap, W_ n)
     bdescr *bd;
     StgPtr p;
 
-    if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
+    if (RTS_UNLIKELY(n >= LARGE_OBJECT_THRESHOLD/sizeof(W_))) {
         // The largest number of words such that
         // the computation of req_blocks will not overflow.
         W_ max_words = (HS_WORD_MAX & ~(BLOCK_SIZE-1)) / sizeof(W_);
@@ -897,7 +897,7 @@ allocateMightFail (Capability *cap, W_ n)
 
     accountAllocation(cap, n);
     bd = cap->r.rCurrentAlloc;
-    if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
+    if (RTS_UNLIKELY(bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W)) {
 
         if (bd) finishedNurseryBlock(cap,bd);