X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=OPTIMIZATIONS;h=c79f922af56f7b7a374b69f8cb15c589f7ed8c11;hb=04553847c4e0235ec0a78e96204ff08b86fc6cd7;hp=d8d018f3f164a5a9d57343e495ce1b445627ca77;hpb=d04b59670ab69405c4115ea3caac99fd62a4b7ab;p=sbcl.git

diff --git a/OPTIMIZATIONS b/OPTIMIZATIONS
index d8d018f..c79f922 100644
--- a/OPTIMIZATIONS
+++ b/OPTIMIZATIONS
@@ -218,3 +218,50 @@ SBCL cannot derive upper bound for I and uses generic arithmetic here:
 
 (So the constraint propagator or a possible future SSA-convertor
 should know the connection between an NLE and its CLEANUP.)
+--------------------------------------------------------------------------------
+#27
+Initialization of stack-allocated arrays is inefficient: we always
+fill the vector with zeroes, even when it is not needed (as for
+platforms with conservative GC or for arrays of unboxed objectes) and
+is performed later explicitely.
+--------------------------------------------------------------------------------
+#28
+a. Accessing raw slots in structure instances is more inefficient than
+it could be; if we placed raw slots before the header word, we would
+not need to do arithmetic at runtime to access them.  (But beware:
+this would complicate handling of the interior pointer).
+
+b. (Also note that raw slots are currently disabled on HPPA)
+--------------------------------------------------------------------------------
+#29
+Python is overly zealous when converting high-level CL functions, such
+as MIN/MAX, LOGBITP, and LOGTEST, to low-level CL functions.  Reducing
+Python's aggressiveness would make it easier to effect changes such as
+
+x86-64:
+* direct MIN/MAX on {SINGLE,DOUBLE}-FLOATs ({MIN,MAX}S{S,D})
+
+x86{,-64}:
+* direct LOGBITP on word-sized integers and fixnums (BT + JC)
+
+x86{,-64}/PPC:
+* branch-free MIN/MAX on word-sized integers and fixnums
+* efficient LOGTESTs on word-sized integers and fixnums (TEST / AND.)
+
+PPC:
+* efficient LDB on word-sized integers and fixnums (RLWINM)
+
+etc., etc.
+
+The "easier" part claimed above would come about because the functions
+would be available for :TRANSLATE through a VOP or similar, whereas with
+the current architecture, one would have to pattern-match IR1.  While
+IR1 pattern-matching would be useful in other contexts, it seems better
+here to attempt the direct :TRANSLATE route.
+
+I (NJF) don't know how to implement such architecture-specific
+optimizations whilst keeping the high->low transformations for other
+architectures.  Certainly adding #!+/- magic in compiler/*.lisp could be
+done, but such a solution is somewhat inelegant.  Moving the relevant
+DEFTRANSFORMs to the architecture-specific compiler/* areas is also
+possible, but that would duplicate quite a bit of code.