X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;ds=sidebyside;f=OPTIMIZATIONS;h=1f8746f14400565d7246281ee0710e2f5721de78;hb=192731b882467a9014b64b1eb097c8000441da31;hp=a53233acfffed439424660d2696194b5cf6a6e52;hpb=f601ededaa926a19ca60f95e8bc6f7f0d966f9d1;p=sbcl.git diff --git a/OPTIMIZATIONS b/OPTIMIZATIONS index a53233a..1f8746f 100644 --- a/OPTIMIZATIONS +++ b/OPTIMIZATIONS @@ -11,9 +11,6 @@ * 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] @@ -65,11 +62,6 @@ VOP DATA-VECTOR-SET/SIMPLE-STRING V2!14[EDI] t32[EAX] t30[S2]>t33[CL] * And why two moves? -------------------------------------------------------------------------------- -#5 -(loop repeat 1.5) - -uses generic arithmetic --------------------------------------------------------------------------------- #6 09:49:05 I have found a case in those where suboptimal code is generate with nested loops, it might be moderately easy to fix that @@ -98,21 +90,14 @@ uses generic arithmetic (incf x))))))) (format t "~A~%" x))) -------------------------------------------------------------------------------- -#7 -(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. @@ -166,3 +151,42 @@ tests lowtag. 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. +-------------------------------------------------------------------------------- +#18 +(lambda (x) (declare (null x)) (sxhash x)) goes through SYMBOL-HASH +rather than either constant-folding or manipulating NIL-VALUE or +NULL-TN directly. +--------------------------------------------------------------------------------