0.8.12.33:

[sbcl.git] / OPTIMIZATIONS

diff --git a/OPTIMIZATIONS b/OPTIMIZATIONS
index c3c07e8..0ca48de 100644 (file)
--- a/OPTIMIZATIONS
+++ b/OPTIMIZATIONS
@@ -1,3 +1,4 @@
+#1
 (defun mysl (s)
     (declare (simple-string s))
     (declare (optimize (speed 3) (safety 0) (debug 0)))
@@ -10,14 +11,12 @@
 
 * On X86 I is represented as a tagged integer.
 
-* EQL uses "CMP reg,reg" instead of "CMP reg,im". This causes
-  allocation of an extra register and an extra move.
-
 * Unnecessary move:
   3: SLOT S!11[EDX] {SB-C::VECTOR-LENGTH 1 7} => t23[EAX]
   4: MOVE t23[EAX] => t24[EBX]
 
 --------------------------------------------------------------------------------
+#2
 (defun quux (v)
   (declare (optimize (speed 3) (safety 0) (space 2) (debug 0)))
   (declare (type (simple-array double-float 1) v))
@@ -34,6 +33,7 @@
   and emits costy MOVE ... => FR1.
 
 --------------------------------------------------------------------------------
+#3
 (defun bar (n)
   (declare (optimize (speed 3) (safety 0) (space 2))
            (type fixnum n))
@@ -43,6 +43,7 @@
 
 * IR1 does not optimize away (MAKE-LIST N).
 --------------------------------------------------------------------------------
+#4
 (defun bar (v1 v2)
   (declare (optimize (speed 3) (safety 0) (space 2))
            (type (simple-array base-char 1) v1 v2))
@@ -61,10 +62,7 @@ VOP DATA-VECTOR-SET/SIMPLE-STRING V2!14[EDI] t32[EAX] t30[S2]>t33[CL]
 
 * And why two moves?
 --------------------------------------------------------------------------------
-(loop repeat 1.5)
-
-uses generic arithmetic
---------------------------------------------------------------------------------
+#6
 09:49:05 <jtra> I have found a case in those where suboptimal code is
   generate with nested loops, it might be moderately easy to fix that
 09:49:28 <jtra> see
@@ -74,7 +72,6 @@ uses generic arithmetic
   memory location for iteration variable
 
 ;;; -*- mode: lisp -*-
-;;; $Id$
 ;;; http://www.bagley.org/~doug/shootout/
 ;;; from Friedrich Dominicus
 
@@ -93,19 +90,14 @@ uses generic arithmetic
                 (incf x)))))))
    (format t "~A~%" x)))
 --------------------------------------------------------------------------------
-(defun foo (x)
-  (declare (optimize speed (debug 0)))
-  (if (< x 0) x (foo (1- x))))
-
-SBCL generates a full call of FOO (but CMUCL does not).
---------------------------------------------------------------------------------
+#8
 (defun foo (d)
   (declare (optimize (speed 3) (safety 0) (debug 0)))
   (declare (type (double-float 0d0 1d0) d))
   (loop for i fixnum from 1 to 5
-     for x1 double-float = (sin d) ;;; !!!
-     do (loop for j fixnum from 1 to 4
-             sum x1 double-float)))
+        for x1 double-float = (sin d) ;;; !!!
+        do (loop for j fixnum from 1 to 4
+                 sum x1 double-float)))
 
 Without the marked declaration Python will use boxed representation for X1.
 
@@ -120,6 +112,7 @@ The initial binding is effectless, and without it X is of type
 DOUBLE-FLOAT. Unhopefully, IR1 does not optimize away effectless
 SETs/bindings, and IR2 does not perform type inference.
 --------------------------------------------------------------------------------
+#9 "Multi-path constant folding"
 (defun foo (x)
   (if (= (cond ((irgh x) 0)
                ((buh x) 1)
@@ -135,3 +128,60 @@ This code could be optimized to
         ((buh x) :no)
         (t :no)))
 --------------------------------------------------------------------------------
+#11
+(inverted variant of #9)
+
+(lambda (x)
+  (let ((y (sap-alien x c-string)))
+    (list (alien-sap y)
+          (alien-sap y))))
+
+It could be optimized to
+
+(lambda (x) (list x x))
+
+(if Y were used only once, the current compiler would optimize it)
+--------------------------------------------------------------------------------
+#12
+(typep (truly-the (simple-array * (*)) x) 'simple-vector)
+
+tests lowtag.
+--------------------------------------------------------------------------------
+#13
+FAST-+/FIXNUM and similar should accept unboxed arguments in interests
+of representation selection. Problem: inter-TN dependencies.
+--------------------------------------------------------------------------------
+#14
+The derived type of (/ (THE (DOUBLE-FLOAT (0D0)) X) (THE (DOUBLE-FLOAT
+1D0) Y)) is (DOUBLE-FLOAT 0.0d0). While it might be reasonable, it is
+better to derive (OR (MEMBER 0.0d0) (DOUBLE-FLOAT (0.0d0))).
+--------------------------------------------------------------------------------
+#15
+On the alpha, the system is reluctant to refer directly to a constant bignum,
+preferring to load a large constant through a slow sequence of instructions,
+then cons up a bignum for it:
+
+(LAMBDA (A)
+  (DECLARE (OPTIMIZE (SAFETY 1) (SPEED 3) (DEBUG 1))
+           (TYPE (INTEGER -10000 10000) A)
+           (IGNORABLE A))
+  (CASE A
+    ((89 125 16) (ASH A (MIN 18 -706)))
+    (T (DPB -3 (BYTE 30 30) -1))))
+--------------------------------------------------------------------------------
+#16
+(do ((i 0 (1+ i)))
+    ((= i (the (integer 0 100) n)))
+  ...)
+
+It is commonly expected for Python to derive (FIXNUMP I). (If ``='' is
+replaced with ``>='', Python will do.)
+--------------------------------------------------------------------------------
+#17 
+Type tests for (ARRAY BIT), (ARRAY T) and similar go through full
+%TYPEP, even though it is relatively simple to establish the arrayness
+of an object and also to obtain the element type of an array.  As of
+sbcl-0.8.12.30, this affects at least DUMP-OBJECT through
+COMPOUND-OBJECT-P, and (LABELS MAYBE-EMIT-MAKE-LOAD-FORMS GROVEL)
+through TYPEP UNBOXED-ARRAY, within the compiler itself.
+--------------------------------------------------------------------------------