[PATCH] RegAlloc: Account for a variable entry block frequency

[Manman Ren]

   unsigned calculateRegionSplitCost(LiveInterval &VirtReg,
                                     AllocationOrder &Order,
-                                    BlockFrequency &BestCost,
+                                    const BlockFrequency &AlternateCost,
 unsigned RAGreedy::calculateRegionSplitCost(LiveInterval &VirtReg,
                                             AllocationOrder &Order,
-                                            BlockFrequency &BestCost,
+                                            const BlockFrequency
&AlternateCost,
                                             unsigned &NumCands,
                                             bool IgnoreCSR) {
+  BlockFrequency BestCost = AlternateCost;
--> The above should be separated (is there a reason for these changes?)

+  uint64_t FixedEntry = 1 << 14;
+  if (ActualEntry < FixedEntry)
+    CSRCost *= BranchProbability(ActualEntry, FixedEntry);
+  else if (ActualEntry <= FixedEntry * 16)
+    // Invert the fraction and divide.
+    CSRCost /= BranchProbability(FixedEntry, ActualEntry);
+  else
+    // Can't use BranchProbability in general, since it takes 32-bit
numbers.
+    // This path is faster anyway, albeit less precise.
+    CSRCost = CSRCost.getFrequency() * (ActualEntry >> 14);
--> Are we trying to be precise when possible with this if statement? Why a
fixed 16 in the 2nd condition?

Thanks,
Manman


On Mon, Apr 7, 2014 at 11:25 AM, Duncan P. N. Exon Smith <
dexonsmith at apple.com> wrote:

> Until r197284, the entry frequency was constant -- i.e., set to 2^14.
> Although current ToT still has a constant entry frequency, since r197284
> that has been an implementation detail (which is soon going to change).
>
>   - r204690 made the wrong assumption for the CSRCost metric.  Adjust
>     callee-saved register cost based on entry frequency.
>
>   - r185393 made the wrong assumption (although it was valid at the
>     time).  Update SpillPlacement.cpp::Threshold to be relative to the
>     entry frequency.
>
> Since ToT still has 2^14 entry frequency, this should have no observable
> functionality change.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140407/38a7586a/attachment.html>