X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fx86%2Farith.lisp;h=4731c2653c1db8ee441442369ee788633bf806ae;hb=a6b91f356da1b5ae2987f79db9bd137970512959;hp=2da36e3eeeeb3ab80e1b5a734e240fdde6027931;hpb=b61003dec6f5af2b03549439155676666186283e;p=sbcl.git

diff --git a/src/compiler/x86/arith.lisp b/src/compiler/x86/arith.lisp
index 2da36e3..4731c26 100644
--- a/src/compiler/x86/arith.lisp
+++ b/src/compiler/x86/arith.lisp
@@ -1,4 +1,4 @@
-;;;; the VM definition arithmetic VOPs for the x86
+;;;; the VM definition of arithmetic VOPs for the x86
 
 ;;;; This software is part of the SBCL system. See the README file for
 ;;;; more information.
@@ -46,13 +46,13 @@
 
 (define-vop (fast-lognot/fixnum fixnum-unop)
   (:translate lognot)
-  (:generator 2
+  (:generator 1
     (move res x)
     (inst xor res (fixnumize -1))))
 
 (define-vop (fast-lognot/signed signed-unop)
   (:translate lognot)
-  (:generator 1
+  (:generator 2
     (move res x)
     (inst not res)))
 
@@ -62,49 +62,49 @@
 
 (define-vop (fast-fixnum-binop fast-safe-arith-op)
   (:args (x :target r :scs (any-reg)
-	    :load-if (not (and (sc-is x control-stack)
-			       (sc-is y any-reg)
-			       (sc-is r control-stack)
-			       (location= x r))))
-	 (y :scs (any-reg control-stack)))
+            :load-if (not (and (sc-is x control-stack)
+                               (sc-is y any-reg)
+                               (sc-is r control-stack)
+                               (location= x r))))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:results (r :scs (any-reg) :from (:argument 0)
-	       :load-if (not (and (sc-is x control-stack)
-				  (sc-is y any-reg)
-				  (sc-is r control-stack)
-				  (location= x r)))))
+               :load-if (not (and (sc-is x control-stack)
+                                  (sc-is y any-reg)
+                                  (sc-is r control-stack)
+                                  (location= x r)))))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic"))
 
 (define-vop (fast-unsigned-binop fast-safe-arith-op)
   (:args (x :target r :scs (unsigned-reg)
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is y unsigned-reg)
-			       (sc-is r unsigned-stack)
-			       (location= x r))))
-	 (y :scs (unsigned-reg unsigned-stack)))
+            :load-if (not (and (sc-is x unsigned-stack)
+                               (sc-is y unsigned-reg)
+                               (sc-is r unsigned-stack)
+                               (location= x r))))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
   (:results (r :scs (unsigned-reg) :from (:argument 0)
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is y unsigned-reg)
-			       (sc-is r unsigned-stack)
-			       (location= x r)))))
+            :load-if (not (and (sc-is x unsigned-stack)
+                               (sc-is y unsigned-reg)
+                               (sc-is r unsigned-stack)
+                               (location= x r)))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic"))
 
 (define-vop (fast-signed-binop fast-safe-arith-op)
   (:args (x :target r :scs (signed-reg)
-	    :load-if (not (and (sc-is x signed-stack)
-			       (sc-is y signed-reg)
-			       (sc-is r signed-stack)
-			       (location= x r))))
-	 (y :scs (signed-reg signed-stack)))
+            :load-if (not (and (sc-is x signed-stack)
+                               (sc-is y signed-reg)
+                               (sc-is r signed-stack)
+                               (location= x r))))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types signed-num signed-num)
   (:results (r :scs (signed-reg) :from (:argument 0)
-	    :load-if (not (and (sc-is x signed-stack)
-			       (sc-is y signed-reg)
-			       (sc-is r signed-stack)
-			       (location= x r)))))
+            :load-if (not (and (sc-is x signed-stack)
+                               (sc-is y signed-reg)
+                               (sc-is r signed-stack)
+                               (location= x r)))))
   (:result-types signed-num)
   (:note "inline (signed-byte 32) arithmetic"))
 
@@ -113,7 +113,7 @@
   (:info y)
   (:arg-types tagged-num (:constant (signed-byte 30)))
   (:results (r :scs (any-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic"))
 
@@ -122,7 +122,7 @@
   (:info y)
   (:arg-types unsigned-num (:constant (unsigned-byte 32)))
   (:results (r :scs (unsigned-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic"))
 
@@ -131,54 +131,61 @@
   (:info y)
   (:arg-types signed-num (:constant (signed-byte 32)))
   (:results (r :scs (signed-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types signed-num)
   (:note "inline (signed-byte 32) arithmetic"))
 
 (macrolet ((define-binop (translate untagged-penalty op)
-	     `(progn
-		(define-vop (,(symbolicate "FAST-" translate "/FIXNUM=>FIXNUM")
-			     fast-fixnum-binop)
-		  (:translate ,translate)
-		  (:generator 2
-			      (move r x)
-			      (inst ,op r y)))
-		(define-vop (,(symbolicate 'fast- translate '-c/fixnum=>fixnum)
-			     fast-fixnum-binop-c)
-		  (:translate ,translate)
-		  (:generator 1
-		  (move r x)
-		  (inst ,op r (fixnumize y))))
-		(define-vop (,(symbolicate "FAST-" translate "/SIGNED=>SIGNED")
-			     fast-signed-binop)
-		  (:translate ,translate)
-		  (:generator ,(1+ untagged-penalty)
-		  (move r x)
-		  (inst ,op r y)))
-		(define-vop (,(symbolicate 'fast- translate '-c/signed=>signed)
-			     fast-signed-binop-c)
-		  (:translate ,translate)
-		  (:generator ,untagged-penalty
-		  (move r x)
-		  (inst ,op r y)))
-		(define-vop (,(symbolicate "FAST-"
-					   translate
-					   "/UNSIGNED=>UNSIGNED")
-		fast-unsigned-binop)
-		  (:translate ,translate)
-		  (:generator ,(1+ untagged-penalty)
-		  (move r x)
-		  (inst ,op r y)))
-		(define-vop (,(symbolicate 'fast-
-					   translate
-					   '-c/unsigned=>unsigned)
-			     fast-unsigned-binop-c)
-		  (:translate ,translate)
-		  (:generator ,untagged-penalty
-		  (move r x)
-		  (inst ,op r y))))))
-
-  ;;(define-binop + 4 add)
+             `(progn
+                (define-vop (,(symbolicate "FAST-" translate "/FIXNUM=>FIXNUM")
+                             fast-fixnum-binop)
+                  (:translate ,translate)
+                  (:generator 2
+                              (move r x)
+                              (inst ,op r y)))
+                (define-vop (,(symbolicate 'fast- translate '-c/fixnum=>fixnum)
+                             fast-fixnum-binop-c)
+                  (:translate ,translate)
+                  (:generator 1
+                  (move r x)
+                  (inst ,op r (fixnumize y))))
+                (define-vop (,(symbolicate "FAST-" translate "/SIGNED=>SIGNED")
+                             fast-signed-binop)
+                  (:translate ,translate)
+                  (:generator ,(1+ untagged-penalty)
+                  (move r x)
+                  (inst ,op r y)))
+                (define-vop (,(symbolicate 'fast- translate '-c/signed=>signed)
+                             fast-signed-binop-c)
+                  (:translate ,translate)
+                  (:generator ,untagged-penalty
+                  (move r x)
+                  (inst ,op r y)))
+                (define-vop (,(symbolicate "FAST-"
+                                           translate
+                                           "/UNSIGNED=>UNSIGNED")
+                fast-unsigned-binop)
+                  (:translate ,translate)
+                  (:generator ,(1+ untagged-penalty)
+                  (move r x)
+                  (inst ,op r y)))
+                (define-vop (,(symbolicate 'fast-
+                                           translate
+                                           '-c/unsigned=>unsigned)
+                             fast-unsigned-binop-c)
+                  (:translate ,translate)
+                  (:generator ,untagged-penalty
+                  (move r x)
+                  ,(if (eq translate 'logand)
+                       ;; for the -C/UNSIGNED=>UNSIGNED VOP, this case
+                       ;; is optimized away as an identity somewhere
+                       ;; along the lines.  However, this VOP is used in
+                       ;; -C/SIGNED=>UNSIGNED, below, when the
+                       ;; higher-level lisp code can't optimize away the
+                       ;; non-trivial identity.
+                       `(unless (= y #.(1- (ash 1 n-word-bits)))
+                          (inst ,op r y))
+                       `(inst ,op r y)))))))
   (define-binop - 4 sub)
   (define-binop logand 2 and)
   (define-binop logior 2 or)
@@ -189,26 +196,26 @@
 (define-vop (fast-+/fixnum=>fixnum fast-safe-arith-op)
   (:translate +)
   (:args (x :scs (any-reg) :target r
-	    :load-if (not (and (sc-is x control-stack)
-			       (sc-is y any-reg)
-			       (sc-is r control-stack)
-			       (location= x r))))
-	 (y :scs (any-reg control-stack)))
+            :load-if (not (and (sc-is x control-stack)
+                               (sc-is y any-reg)
+                               (sc-is r control-stack)
+                               (location= x r))))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:results (r :scs (any-reg) :from (:argument 0)
-	       :load-if (not (and (sc-is x control-stack)
-				  (sc-is y any-reg)
-				  (sc-is r control-stack)
-				  (location= x r)))))
+               :load-if (not (and (sc-is x control-stack)
+                                  (sc-is y any-reg)
+                                  (sc-is r control-stack)
+                                  (location= x r)))))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic")
   (:generator 2
     (cond ((and (sc-is x any-reg) (sc-is y any-reg) (sc-is r any-reg)
-		(not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :index y :scale 1)))
-	  (t
-	   (move r x)
-	   (inst add r y)))))
+                (not (location= x r)))
+           (inst lea r (make-ea :dword :base x :index y :scale 1)))
+          (t
+           (move r x)
+           (inst add r y)))))
 
 (define-vop (fast-+-c/fixnum=>fixnum fast-safe-arith-op)
   (:translate +)
@@ -216,65 +223,64 @@
   (:info y)
   (:arg-types tagged-num (:constant (signed-byte 30)))
   (:results (r :scs (any-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic")
   (:generator 1
     (cond ((and (sc-is x any-reg) (sc-is r any-reg) (not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :disp (fixnumize y))))
-	  (t
-	   (move r x)
-	   (inst add r (fixnumize y))))))
+           (inst lea r (make-ea :dword :base x :disp (fixnumize y))))
+          (t
+           (move r x)
+           (inst add r (fixnumize y))))))
 
 (define-vop (fast-+/signed=>signed fast-safe-arith-op)
   (:translate +)
   (:args (x :scs (signed-reg) :target r
-	    :load-if (not (and (sc-is x signed-stack)
-			       (sc-is y signed-reg)
-			       (sc-is r signed-stack)
-			       (location= x r))))
-	 (y :scs (signed-reg signed-stack)))
+            :load-if (not (and (sc-is x signed-stack)
+                               (sc-is y signed-reg)
+                               (sc-is r signed-stack)
+                               (location= x r))))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types signed-num signed-num)
   (:results (r :scs (signed-reg) :from (:argument 0)
-	       :load-if (not (and (sc-is x signed-stack)
-				  (sc-is y signed-reg)
-				  (location= x r)))))
+               :load-if (not (and (sc-is x signed-stack)
+                                  (sc-is y signed-reg)
+                                  (location= x r)))))
   (:result-types signed-num)
   (:note "inline (signed-byte 32) arithmetic")
   (:generator 5
     (cond ((and (sc-is x signed-reg) (sc-is y signed-reg) (sc-is r signed-reg)
-		(not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :index y :scale 1)))
-	  (t
-	   (move r x)
-	   (inst add r y)))))
-
+                (not (location= x r)))
+           (inst lea r (make-ea :dword :base x :index y :scale 1)))
+          (t
+           (move r x)
+           (inst add r y)))))
 
 ;;;; Special logand cases: (logand signed unsigned) => unsigned
 
 (define-vop (fast-logand/signed-unsigned=>unsigned
-	     fast-logand/unsigned=>unsigned)
+             fast-logand/unsigned=>unsigned)
   (:args (x :target r :scs (signed-reg)
-	    :load-if (not (and (sc-is x signed-stack)
-			       (sc-is y unsigned-reg)
-			       (sc-is r unsigned-stack)
-			       (location= x r))))
-	 (y :scs (unsigned-reg unsigned-stack)))
+            :load-if (not (and (sc-is x signed-stack)
+                               (sc-is y unsigned-reg)
+                               (sc-is r unsigned-stack)
+                               (location= x r))))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types signed-num unsigned-num))
 
 (define-vop (fast-logand-c/signed-unsigned=>unsigned
-	     fast-logand-c/unsigned=>unsigned)
+             fast-logand-c/unsigned=>unsigned)
   (:args (x :target r :scs (signed-reg signed-stack)))
   (:arg-types signed-num (:constant (unsigned-byte 32))))
 
 (define-vop (fast-logand/unsigned-signed=>unsigned
-	     fast-logand/unsigned=>unsigned)
+             fast-logand/unsigned=>unsigned)
   (:args (x :target r :scs (unsigned-reg)
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is y signed-reg)
-			       (sc-is r unsigned-stack)
-			       (location= x r))))
-	 (y :scs (signed-reg signed-stack)))
+            :load-if (not (and (sc-is x unsigned-stack)
+                               (sc-is y signed-reg)
+                               (sc-is r unsigned-stack)
+                               (location= x r))))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types unsigned-num signed-num))
 
 
@@ -284,42 +290,42 @@
   (:info y)
   (:arg-types signed-num (:constant (signed-byte 32)))
   (:results (r :scs (signed-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types signed-num)
   (:note "inline (signed-byte 32) arithmetic")
   (:generator 4
     (cond ((and (sc-is x signed-reg) (sc-is r signed-reg)
-		(not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :disp y)))
-	  (t
-	   (move r x)
-	   (if (= y 1)
-	       (inst inc r)
-	     (inst add r y))))))
+                (not (location= x r)))
+           (inst lea r (make-ea :dword :base x :disp y)))
+          (t
+           (move r x)
+           (if (= y 1)
+               (inst inc r)
+             (inst add r y))))))
 
 (define-vop (fast-+/unsigned=>unsigned fast-safe-arith-op)
   (:translate +)
   (:args (x :scs (unsigned-reg) :target r
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is y unsigned-reg)
-			       (sc-is r unsigned-stack)
-			       (location= x r))))
-	 (y :scs (unsigned-reg unsigned-stack)))
+            :load-if (not (and (sc-is x unsigned-stack)
+                               (sc-is y unsigned-reg)
+                               (sc-is r unsigned-stack)
+                               (location= x r))))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
   (:results (r :scs (unsigned-reg) :from (:argument 0)
-	       :load-if (not (and (sc-is x unsigned-stack)
-				  (sc-is y unsigned-reg)
-				  (sc-is r unsigned-stack)
-				  (location= x r)))))
+               :load-if (not (and (sc-is x unsigned-stack)
+                                  (sc-is y unsigned-reg)
+                                  (sc-is r unsigned-stack)
+                                  (location= x r)))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic")
   (:generator 5
     (cond ((and (sc-is x unsigned-reg) (sc-is y unsigned-reg)
-		(sc-is r unsigned-reg) (not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :index y :scale 1)))
-	  (t
-	   (move r x)
-	   (inst add r y)))))
+                (sc-is r unsigned-reg) (not (location= x r)))
+           (inst lea r (make-ea :dword :base x :index y :scale 1)))
+          (t
+           (move r x)
+           (inst add r y)))))
 
 (define-vop (fast-+-c/unsigned=>unsigned fast-safe-arith-op)
   (:translate +)
@@ -327,18 +333,18 @@
   (:info y)
   (:arg-types unsigned-num (:constant (unsigned-byte 32)))
   (:results (r :scs (unsigned-reg)
-	       :load-if (not (location= x r))))
+               :load-if (not (location= x r))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic")
   (:generator 4
     (cond ((and (sc-is x unsigned-reg) (sc-is r unsigned-reg)
-		(not (location= x r)))
-	   (inst lea r (make-ea :dword :base x :disp y)))
-	  (t
-	   (move r x)
-	   (if (= y 1)
-	       (inst inc r)
-	     (inst add r y))))))
+                (not (location= x r)))
+           (inst lea r (make-ea :dword :base x :disp y)))
+          (t
+           (move r x)
+           (if (= y 1)
+               (inst inc r)
+             (inst add r y))))))
 
 ;;;; multiplication and division
 
@@ -346,7 +352,7 @@
   (:translate *)
   ;; We need different loading characteristics.
   (:args (x :scs (any-reg) :target r)
-	 (y :scs (any-reg control-stack)))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:results (r :scs (any-reg) :from (:argument 0)))
   (:result-types tagged-num)
@@ -372,7 +378,7 @@
   (:translate *)
   ;; We need different loading characteristics.
   (:args (x :scs (signed-reg) :target r)
-	 (y :scs (signed-reg signed-stack)))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types signed-num signed-num)
   (:results (r :scs (signed-reg) :from (:argument 0)))
   (:result-types signed-num)
@@ -396,14 +402,14 @@
 (define-vop (fast-*/unsigned=>unsigned fast-safe-arith-op)
   (:translate *)
   (:args (x :scs (unsigned-reg) :target eax)
-	 (y :scs (unsigned-reg unsigned-stack)))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
-  (:temporary (:sc unsigned-reg :offset eax-offset :target result
-		   :from (:argument 0) :to :result) eax)
+  (:temporary (:sc unsigned-reg :offset eax-offset :target r
+                   :from (:argument 0) :to :result) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset
-		   :from :eval :to :result) edx)
+                   :from :eval :to :result) edx)
   (:ignore edx)
-  (:results (result :scs (unsigned-reg)))
+  (:results (r :scs (unsigned-reg)))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic")
   (:vop-var vop)
@@ -411,36 +417,36 @@
   (:generator 6
     (move eax x)
     (inst mul eax y)
-    (move result eax)))
+    (move r eax)))
 
 
 (define-vop (fast-truncate/fixnum=>fixnum fast-safe-arith-op)
   (:translate truncate)
   (:args (x :scs (any-reg) :target eax)
-	 (y :scs (any-reg control-stack)))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:temporary (:sc signed-reg :offset eax-offset :target quo
-		   :from (:argument 0) :to (:result 0)) eax)
+                   :from (:argument 0) :to (:result 0)) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :target rem
-		   :from (:argument 0) :to (:result 1)) edx)
+                   :from (:argument 0) :to (:result 1)) edx)
   (:results (quo :scs (any-reg))
-	    (rem :scs (any-reg)))
+            (rem :scs (any-reg)))
   (:result-types tagged-num tagged-num)
   (:note "inline fixnum arithmetic")
   (:vop-var vop)
   (:save-p :compute-only)
   (:generator 31
-    (let ((zero (generate-error-code vop division-by-zero-error x y)))
+    (let ((zero (generate-error-code vop 'division-by-zero-error x y)))
       (if (sc-is y any-reg)
-	  (inst test y y)  ; smaller instruction
-	  (inst cmp y 0))
+          (inst test y y)  ; smaller instruction
+          (inst cmp y 0))
       (inst jmp :eq zero))
     (move eax x)
     (inst cdq)
     (inst idiv eax y)
     (if (location= quo eax)
-	(inst shl eax 2)
-	(inst lea quo (make-ea :dword :index eax :scale 4)))
+        (inst shl eax 2)
+        (inst lea quo (make-ea :dword :index eax :scale 4)))
     (move rem edx)))
 
 (define-vop (fast-truncate-c/fixnum=>fixnum fast-safe-arith-op)
@@ -449,12 +455,12 @@
   (:info y)
   (:arg-types tagged-num (:constant (signed-byte 30)))
   (:temporary (:sc signed-reg :offset eax-offset :target quo
-		   :from :argument :to (:result 0)) eax)
+                   :from :argument :to (:result 0)) eax)
   (:temporary (:sc any-reg :offset edx-offset :target rem
-		   :from :eval :to (:result 1)) edx)
+                   :from :eval :to (:result 1)) edx)
   (:temporary (:sc any-reg :from :eval :to :result) y-arg)
   (:results (quo :scs (any-reg))
-	    (rem :scs (any-reg)))
+            (rem :scs (any-reg)))
   (:result-types tagged-num tagged-num)
   (:note "inline fixnum arithmetic")
   (:vop-var vop)
@@ -465,30 +471,30 @@
     (inst mov y-arg (fixnumize y))
     (inst idiv eax y-arg)
     (if (location= quo eax)
-	(inst shl eax 2)
-	(inst lea quo (make-ea :dword :index eax :scale 4)))
+        (inst shl eax 2)
+        (inst lea quo (make-ea :dword :index eax :scale 4)))
     (move rem edx)))
 
 (define-vop (fast-truncate/unsigned=>unsigned fast-safe-arith-op)
   (:translate truncate)
   (:args (x :scs (unsigned-reg) :target eax)
-	 (y :scs (unsigned-reg signed-stack)))
+         (y :scs (unsigned-reg signed-stack)))
   (:arg-types unsigned-num unsigned-num)
   (:temporary (:sc unsigned-reg :offset eax-offset :target quo
-		   :from (:argument 0) :to (:result 0)) eax)
+                   :from (:argument 0) :to (:result 0)) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :target rem
-		   :from (:argument 0) :to (:result 1)) edx)
+                   :from (:argument 0) :to (:result 1)) edx)
   (:results (quo :scs (unsigned-reg))
-	    (rem :scs (unsigned-reg)))
+            (rem :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic")
   (:vop-var vop)
   (:save-p :compute-only)
   (:generator 33
-    (let ((zero (generate-error-code vop division-by-zero-error x y)))
+    (let ((zero (generate-error-code vop 'division-by-zero-error x y)))
       (if (sc-is y unsigned-reg)
-	  (inst test y y)  ; smaller instruction
-	  (inst cmp y 0))
+          (inst test y y)  ; smaller instruction
+          (inst cmp y 0))
       (inst jmp :eq zero))
     (move eax x)
     (inst xor edx edx)
@@ -502,12 +508,12 @@
   (:info y)
   (:arg-types unsigned-num (:constant (unsigned-byte 32)))
   (:temporary (:sc unsigned-reg :offset eax-offset :target quo
-		   :from :argument :to (:result 0)) eax)
+                   :from :argument :to (:result 0)) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :target rem
-		   :from :eval :to (:result 1)) edx)
+                   :from :eval :to (:result 1)) edx)
   (:temporary (:sc unsigned-reg :from :eval :to :result) y-arg)
   (:results (quo :scs (unsigned-reg))
-	    (rem :scs (unsigned-reg)))
+            (rem :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:note "inline (unsigned-byte 32) arithmetic")
   (:vop-var vop)
@@ -523,23 +529,23 @@
 (define-vop (fast-truncate/signed=>signed fast-safe-arith-op)
   (:translate truncate)
   (:args (x :scs (signed-reg) :target eax)
-	 (y :scs (signed-reg signed-stack)))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types signed-num signed-num)
   (:temporary (:sc signed-reg :offset eax-offset :target quo
-		   :from (:argument 0) :to (:result 0)) eax)
+                   :from (:argument 0) :to (:result 0)) eax)
   (:temporary (:sc signed-reg :offset edx-offset :target rem
-		   :from (:argument 0) :to (:result 1)) edx)
+                   :from (:argument 0) :to (:result 1)) edx)
   (:results (quo :scs (signed-reg))
-	    (rem :scs (signed-reg)))
+            (rem :scs (signed-reg)))
   (:result-types signed-num signed-num)
   (:note "inline (signed-byte 32) arithmetic")
   (:vop-var vop)
   (:save-p :compute-only)
   (:generator 33
-    (let ((zero (generate-error-code vop division-by-zero-error x y)))
+    (let ((zero (generate-error-code vop 'division-by-zero-error x y)))
       (if (sc-is y signed-reg)
-	  (inst test y y)  ; smaller instruction
-	  (inst cmp y 0))
+          (inst test y y)  ; smaller instruction
+          (inst cmp y 0))
       (inst jmp :eq zero))
     (move eax x)
     (inst cdq)
@@ -553,12 +559,12 @@
   (:info y)
   (:arg-types signed-num (:constant (signed-byte 32)))
   (:temporary (:sc signed-reg :offset eax-offset :target quo
-		   :from :argument :to (:result 0)) eax)
+                   :from :argument :to (:result 0)) eax)
   (:temporary (:sc signed-reg :offset edx-offset :target rem
-		   :from :eval :to (:result 1)) edx)
+                   :from :eval :to (:result 1)) edx)
   (:temporary (:sc signed-reg :from :eval :to :result) y-arg)
   (:results (quo :scs (signed-reg))
-	    (rem :scs (signed-reg)))
+            (rem :scs (signed-reg)))
   (:result-types signed-num signed-num)
   (:note "inline (signed-byte 32) arithmetic")
   (:vop-var vop)
@@ -578,51 +584,54 @@
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (any-reg) :target result
-		 :load-if (not (and (sc-is number any-reg control-stack)
-				    (sc-is result any-reg control-stack)
-				    (location= number result)))))
+                 :load-if (not (and (sc-is number any-reg control-stack)
+                                    (sc-is result any-reg control-stack)
+                                    (location= number result)))))
   (:info amount)
   (:arg-types tagged-num (:constant integer))
   (:results (result :scs (any-reg)
-		    :load-if (not (and (sc-is number control-stack)
-				       (sc-is result control-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number control-stack)
+                                       (sc-is result control-stack)
+                                       (location= number result)))))
   (:result-types tagged-num)
   (:note "inline ASH")
   (:generator 2
     (cond ((and (= amount 1) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 2)))
-	  ((and (= amount 2) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 4)))
-	  ((and (= amount 3) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 8)))
-	  (t
-	   (move result number)
-	   (cond ((plusp amount)
-		  ;; We don't have to worry about overflow because of the
-		  ;; result type restriction.
-		  (inst shl result amount))
-		 (t
-		  ;; If the amount is greater than 31, only shift by 31. We
-		  ;; have to do this because the shift instructions only look
-		  ;; at the low five bits of the result.
-		  (inst sar result (min 31 (- amount)))
-		  ;; Fixnum correction.
-		  (inst and result #xfffffffc)))))))
+           (inst lea result (make-ea :dword :base number :index number)))
+          ((and (= amount 2) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 4)))
+          ((and (= amount 3) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 8)))
+          (t
+           (move result number)
+           (cond ((< -32 amount 32)
+                  ;; this code is used both in ASH and ASH-SMOD30, so
+                  ;; be careful
+                  (if (plusp amount)
+                      (inst shl result amount)
+                      (progn
+                        (inst sar result (- amount))
+                        (inst and result (lognot fixnum-tag-mask)))))
+                 ((plusp amount)
+                  (if (sc-is result any-reg)
+                      (inst xor result result)
+                      (inst mov result 0)))
+                 (t (inst sar result 31)
+                    (inst and result (lognot fixnum-tag-mask))))))))
 
 (define-vop (fast-ash-left/fixnum=>fixnum)
   (:translate ash)
   (:args (number :scs (any-reg) :target result
-		 :load-if (not (and (sc-is number control-stack)
-				    (sc-is result control-stack)
-				    (location= number result))))
-	 (amount :scs (unsigned-reg) :target ecx))
+                 :load-if (not (and (sc-is number control-stack)
+                                    (sc-is result control-stack)
+                                    (location= number result))))
+         (amount :scs (unsigned-reg) :target ecx))
   (:arg-types tagged-num positive-fixnum)
   (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
   (:results (result :scs (any-reg) :from (:argument 0)
-		    :load-if (not (and (sc-is number control-stack)
-				       (sc-is result control-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number control-stack)
+                                       (sc-is result control-stack)
+                                       (location= number result)))))
   (:result-types tagged-num)
   (:policy :fast-safe)
   (:note "inline ASH")
@@ -636,70 +645,76 @@
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (signed-reg) :target result
-		 :load-if (not (and (sc-is number signed-stack)
-				    (sc-is result signed-stack)
-				    (location= number result)))))
+                 :load-if (not (and (sc-is number signed-stack)
+                                    (sc-is result signed-stack)
+                                    (location= number result)))))
   (:info amount)
   (:arg-types signed-num (:constant integer))
   (:results (result :scs (signed-reg)
-		    :load-if (not (and (sc-is number signed-stack)
-				       (sc-is result signed-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number signed-stack)
+                                       (sc-is result signed-stack)
+                                       (location= number result)))))
   (:result-types signed-num)
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 2)))
-	  ((and (= amount 2) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 4)))
-	  ((and (= amount 3) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 8)))
-	  (t
-	   (move result number)
-	   (cond ((plusp amount) (inst shl result amount))
-		 (t (inst sar result (min 31 (- amount)))))))))
+           (inst lea result (make-ea :dword :base number :index number)))
+          ((and (= amount 2) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 4)))
+          ((and (= amount 3) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 8)))
+          (t
+           (move result number)
+           (cond ((plusp amount) (inst shl result amount))
+                 (t (inst sar result (min 31 (- amount)))))))))
 
 (define-vop (fast-ash-c/unsigned=>unsigned)
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (unsigned-reg) :target result
-		 :load-if (not (and (sc-is number unsigned-stack)
-				    (sc-is result unsigned-stack)
-				    (location= number result)))))
+                 :load-if (not (and (sc-is number unsigned-stack)
+                                    (sc-is result unsigned-stack)
+                                    (location= number result)))))
   (:info amount)
   (:arg-types unsigned-num (:constant integer))
   (:results (result :scs (unsigned-reg)
-		    :load-if (not (and (sc-is number unsigned-stack)
-				       (sc-is result unsigned-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number unsigned-stack)
+                                       (sc-is result unsigned-stack)
+                                       (location= number result)))))
   (:result-types unsigned-num)
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 2)))
-	  ((and (= amount 2) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 4)))
-	  ((and (= amount 3) (not (location= number result)))
-	   (inst lea result (make-ea :dword :index number :scale 8)))
-	  (t
-	   (move result number)
-	   (cond ((plusp amount) (inst shl result amount))
-		 ((< amount -31) (inst xor result result))
-		 (t (inst shr result (- amount))))))))
+           (inst lea result (make-ea :dword :base number :index number)))
+          ((and (= amount 2) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 4)))
+          ((and (= amount 3) (not (location= number result)))
+           (inst lea result (make-ea :dword :index number :scale 8)))
+          (t
+           (move result number)
+           (cond ((< -32 amount 32)
+                  ;; this code is used both in ASH and ASH-MOD32, so
+                  ;; be careful
+                  (if (plusp amount)
+                      (inst shl result amount)
+                      (inst shr result (- amount))))
+                 (t (if (sc-is result unsigned-reg)
+                        (inst xor result result)
+                        (inst mov result 0))))))))
 
 (define-vop (fast-ash-left/signed=>signed)
   (:translate ash)
   (:args (number :scs (signed-reg) :target result
-		 :load-if (not (and (sc-is number signed-stack)
-				    (sc-is result signed-stack)
-				    (location= number result))))
-	 (amount :scs (unsigned-reg) :target ecx))
+                 :load-if (not (and (sc-is number signed-stack)
+                                    (sc-is result signed-stack)
+                                    (location= number result))))
+         (amount :scs (unsigned-reg) :target ecx))
   (:arg-types signed-num positive-fixnum)
   (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
   (:results (result :scs (signed-reg) :from (:argument 0)
-		    :load-if (not (and (sc-is number signed-stack)
-				       (sc-is result signed-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number signed-stack)
+                                       (sc-is result signed-stack)
+                                       (location= number result)))))
   (:result-types signed-num)
   (:policy :fast-safe)
   (:note "inline ASH")
@@ -711,16 +726,16 @@
 (define-vop (fast-ash-left/unsigned=>unsigned)
   (:translate ash)
   (:args (number :scs (unsigned-reg) :target result
-		 :load-if (not (and (sc-is number unsigned-stack)
-				    (sc-is result unsigned-stack)
-				    (location= number result))))
-	 (amount :scs (unsigned-reg) :target ecx))
+                 :load-if (not (and (sc-is number unsigned-stack)
+                                    (sc-is result unsigned-stack)
+                                    (location= number result))))
+         (amount :scs (unsigned-reg) :target ecx))
   (:arg-types unsigned-num positive-fixnum)
   (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
   (:results (result :scs (unsigned-reg) :from (:argument 0)
-		    :load-if (not (and (sc-is number unsigned-stack)
-				       (sc-is result unsigned-stack)
-				       (location= number result)))))
+                    :load-if (not (and (sc-is number unsigned-stack)
+                                       (sc-is result unsigned-stack)
+                                       (location= number result)))))
   (:result-types unsigned-num)
   (:policy :fast-safe)
   (:note "inline ASH")
@@ -733,7 +748,7 @@
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (signed-reg) :target result)
-	 (amount :scs (signed-reg) :target ecx))
+         (amount :scs (signed-reg) :target ecx))
   (:arg-types signed-num signed-num)
   (:results (result :scs (signed-reg) :from (:argument 0)))
   (:result-types signed-num)
@@ -762,7 +777,7 @@
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (unsigned-reg) :target result)
-	 (amount :scs (signed-reg) :target ecx))
+         (amount :scs (signed-reg) :target ecx))
   (:arg-types unsigned-num signed-num)
   (:results (result :scs (unsigned-reg) :from (:argument 0)))
   (:result-types unsigned-num)
@@ -788,6 +803,82 @@
 
     DONE))
 
+(in-package "SB!C")
+
+(defknown %lea (integer integer (member 1 2 4 8) (signed-byte 32))
+  integer
+  (foldable flushable movable))
+
+(defoptimizer (%lea derive-type) ((base index scale disp))
+  (when (and (constant-lvar-p scale)
+             (constant-lvar-p disp))
+    (let ((scale (lvar-value scale))
+          (disp (lvar-value disp))
+          (base-type (lvar-type base))
+          (index-type (lvar-type index)))
+      (when (and (numeric-type-p base-type)
+                 (numeric-type-p index-type))
+        (let ((base-lo (numeric-type-low base-type))
+              (base-hi (numeric-type-high base-type))
+              (index-lo (numeric-type-low index-type))
+              (index-hi (numeric-type-high index-type)))
+          (make-numeric-type :class 'integer
+                             :complexp :real
+                             :low (when (and base-lo index-lo)
+                                    (+ base-lo (* index-lo scale) disp))
+                             :high (when (and base-hi index-hi)
+                                     (+ base-hi (* index-hi scale) disp))))))))
+
+(defun %lea (base index scale disp)
+  (+ base (* index scale) disp))
+
+(in-package "SB!VM")
+
+(define-vop (%lea/unsigned=>unsigned)
+  (:translate %lea)
+  (:policy :fast-safe)
+  (:args (base :scs (unsigned-reg))
+         (index :scs (unsigned-reg)))
+  (:info scale disp)
+  (:arg-types unsigned-num unsigned-num
+              (:constant (member 1 2 4 8))
+              (:constant (signed-byte 32)))
+  (:results (r :scs (unsigned-reg)))
+  (:result-types unsigned-num)
+  (:generator 5
+    (inst lea r (make-ea :dword :base base :index index
+                         :scale scale :disp disp))))
+
+(define-vop (%lea/signed=>signed)
+  (:translate %lea)
+  (:policy :fast-safe)
+  (:args (base :scs (signed-reg))
+         (index :scs (signed-reg)))
+  (:info scale disp)
+  (:arg-types signed-num signed-num
+              (:constant (member 1 2 4 8))
+              (:constant (signed-byte 32)))
+  (:results (r :scs (signed-reg)))
+  (:result-types signed-num)
+  (:generator 4
+    (inst lea r (make-ea :dword :base base :index index
+                         :scale scale :disp disp))))
+
+(define-vop (%lea/fixnum=>fixnum)
+  (:translate %lea)
+  (:policy :fast-safe)
+  (:args (base :scs (any-reg))
+         (index :scs (any-reg)))
+  (:info scale disp)
+  (:arg-types tagged-num tagged-num
+              (:constant (member 1 2 4 8))
+              (:constant (signed-byte 32)))
+  (:results (r :scs (any-reg)))
+  (:result-types tagged-num)
+  (:generator 3
+    (inst lea r (make-ea :dword :base base :index index
+                         :scale scale :disp disp))))
+
 ;;; FIXME: before making knowledge of this too public, it needs to be
 ;;; fixed so that it's actually _faster_ than the non-CMOV version; at
 ;;; least on my Celeron-XXX laptop, this version is marginally slower
@@ -796,7 +887,7 @@
   (:translate ash)
   (:policy :fast-safe)
   (:args (number :scs (unsigned-reg) :target result)
-	 (amount :scs (signed-reg) :target ecx))
+         (amount :scs (signed-reg) :target ecx))
   (:arg-types unsigned-num signed-num)
   (:results (result :scs (unsigned-reg) :from (:argument 0)))
   (:result-types unsigned-num)
@@ -815,14 +906,13 @@
     (inst cmp ecx 31)
     (inst cmov :nbe result zero)
     (inst jmp done)
-    
+
     POSITIVE
     ;; The result-type ensures us that this shift will not overflow.
     (inst shl result :cl)
 
     DONE))
 
-;;; Note: documentation for this function is wrong - rtfm
 (define-vop (signed-byte-32-len)
   (:translate integer-length)
   (:note "inline (signed-byte 32) integer-length")
@@ -833,7 +923,9 @@
   (:result-types unsigned-num)
   (:generator 28
     (move res arg)
-    (inst cmp res 0)
+    (if (sc-is res unsigned-reg)
+        (inst test res res)
+        (inst cmp res 0))
     (inst jmp :ge POS)
     (inst not res)
     POS
@@ -866,43 +958,53 @@
   (:translate logcount)
   (:note "inline (unsigned-byte 32) logcount")
   (:policy :fast-safe)
-  (:args (arg :scs (unsigned-reg)))
+  (:args (arg :scs (unsigned-reg) :target result))
   (:arg-types unsigned-num)
   (:results (result :scs (unsigned-reg)))
   (:result-types positive-fixnum)
-  (:temporary (:sc unsigned-reg :from (:argument 0)) temp)
-  (:generator 30
+  (:temporary (:sc unsigned-reg) temp)
+  (:generator 14
+    ;; See the comments below for how the algorithm works. The tricks
+    ;; used can be found for example in AMD's software optimization
+    ;; guide or at "http://www.hackersdelight.org/HDcode/pop.cc" in the
+    ;; function "pop1".
+    ;; Calculate 2-bit sums. Note that the value of a two-digit binary
+    ;; number is the sum of the right digit and twice the left digit.
+    ;; Thus we can calculate the sum of the two digits by shifting the
+    ;; left digit to the right position and doing a two-bit subtraction.
+    ;; This subtraction will never create a borrow and thus can be made
+    ;; on all 16 2-digit numbers at once.
     (move result arg)
-
-    (inst mov temp result)
-    (inst shr temp 1)
+    (move temp arg)
+    (inst shr result 1)
     (inst and result #x55555555)
-    (inst and temp #x55555555)
-    (inst add result temp)
-
-    (inst mov temp result)
+    (inst sub temp result)
+    ;; Calculate 4-bit sums by straightforward shift, mask and add.
+    ;; Note that we shift the source operand of the MOV and not its
+    ;; destination so that the SHR and the MOV can execute in the same
+    ;; clock cycle.
+    (inst mov result temp)
     (inst shr temp 2)
     (inst and result #x33333333)
     (inst and temp #x33333333)
     (inst add result temp)
-
+    ;; Calculate 8-bit sums. Since each sum is at most 8, which fits
+    ;; into 4 bits, we can apply the mask after the addition, saving one
+    ;; instruction.
     (inst mov temp result)
-    (inst shr temp 4)
-    (inst and result #x0f0f0f0f)
-    (inst and temp #x0f0f0f0f)
+    (inst shr result 4)
     (inst add result temp)
-
+    (inst and result #x0f0f0f0f)
+    ;; Calculate the two 16-bit sums and the 32-bit sum. No masking is
+    ;; necessary inbetween since the final sum is at most 32 which fits
+    ;; into 6 bits.
     (inst mov temp result)
-    (inst shr temp 8)
-    (inst and result #x00ff00ff)
-    (inst and temp #x00ff00ff)
+    (inst shr result 8)
     (inst add result temp)
-
     (inst mov temp result)
-    (inst shr temp 16)
-    (inst and result #x0000ffff)
-    (inst and temp #x0000ffff)
-    (inst add result temp)))
+    (inst shr result 16)
+    (inst add result temp)
+    (inst and result #xff)))
 
 ;;;; binary conditional VOPs
 
@@ -915,9 +1017,9 @@
 
 (define-vop (fast-conditional/fixnum fast-conditional)
   (:args (x :scs (any-reg)
-	    :load-if (not (and (sc-is x control-stack)
-			       (sc-is y any-reg))))
-	 (y :scs (any-reg control-stack)))
+            :load-if (not (and (sc-is x control-stack)
+                               (sc-is y any-reg))))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:note "inline fixnum comparison"))
 
@@ -928,9 +1030,9 @@
 
 (define-vop (fast-conditional/signed fast-conditional)
   (:args (x :scs (signed-reg)
-	    :load-if (not (and (sc-is x signed-stack)
-			       (sc-is y signed-reg))))
-	 (y :scs (signed-reg signed-stack)))
+            :load-if (not (and (sc-is x signed-stack)
+                               (sc-is y signed-reg))))
+         (y :scs (signed-reg signed-stack)))
   (:arg-types signed-num signed-num)
   (:note "inline (signed-byte 32) comparison"))
 
@@ -941,9 +1043,9 @@
 
 (define-vop (fast-conditional/unsigned fast-conditional)
   (:args (x :scs (unsigned-reg)
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is y unsigned-reg))))
-	 (y :scs (unsigned-reg unsigned-stack)))
+            :load-if (not (and (sc-is x unsigned-stack)
+                               (sc-is y unsigned-reg))))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
   (:note "inline (unsigned-byte 32) comparison"))
 
@@ -952,35 +1054,97 @@
   (:arg-types unsigned-num (:constant (unsigned-byte 32)))
   (:info target not-p y))
 
+(macrolet ((define-logtest-vops ()
+             `(progn
+               ,@(loop for suffix in '(/fixnum -c/fixnum
+                                       /signed -c/signed
+                                       /unsigned -c/unsigned)
+                       for cost in '(4 3 6 5 6 5)
+                       collect
+                       `(define-vop (,(symbolicate "FAST-LOGTEST" suffix)
+                                     ,(symbolicate "FAST-CONDITIONAL" suffix))
+                         (:translate logtest)
+                         (:generator ,cost
+                          (emit-optimized-test-inst x
+                                                    ,(if (eq suffix '-c/fixnum)
+                                                         '(fixnumize y)
+                                                         'y))
+                          (inst jmp (if not-p :e :ne) target)))))))
+  (define-logtest-vops))
+
+(defknown %logbitp (integer unsigned-byte) boolean
+  (movable foldable flushable always-translatable))
+
+;;; only for constant folding within the compiler
+(defun %logbitp (integer index)
+  (logbitp index integer))
+
+;;; too much work to do the non-constant case (maybe?)
+(define-vop (fast-logbitp-c/fixnum fast-conditional-c/fixnum)
+  (:translate %logbitp)
+  (:arg-types tagged-num (:constant (integer 0 29)))
+  (:generator 4
+    (inst bt x (+ y n-fixnum-tag-bits))
+    (inst jmp (if not-p :nc :c) target)))
+
+(define-vop (fast-logbitp/signed fast-conditional/signed)
+  (:args (x :scs (signed-reg signed-stack))
+         (y :scs (signed-reg)))
+  (:translate %logbitp)
+  (:generator 6
+    (inst bt x y)
+    (inst jmp (if not-p :nc :c) target)))
+
+(define-vop (fast-logbitp-c/signed fast-conditional-c/signed)
+  (:translate %logbitp)
+  (:arg-types signed-num (:constant (integer 0 31)))
+  (:generator 5
+    (inst bt x y)
+    (inst jmp (if not-p :nc :c) target)))
+
+(define-vop (fast-logbitp/unsigned fast-conditional/unsigned)
+  (:args (x :scs (unsigned-reg unsigned-stack))
+         (y :scs (unsigned-reg)))
+  (:translate %logbitp)
+  (:generator 6
+    (inst bt x y)
+    (inst jmp (if not-p :nc :c) target)))
+
+(define-vop (fast-logbitp-c/unsigned fast-conditional-c/unsigned)
+  (:translate %logbitp)
+  (:arg-types unsigned-num (:constant (integer 0 31)))
+  (:generator 5
+    (inst bt x y)
+    (inst jmp (if not-p :nc :c) target)))
 
 (macrolet ((define-conditional-vop (tran cond unsigned not-cond not-unsigned)
-	     `(progn
-		,@(mapcar
-		   (lambda (suffix cost signed)
-		     `(define-vop (;; FIXME: These could be done more
-				   ;; cleanly with SYMBOLICATE.
-				   ,(intern (format nil "~:@(FAST-IF-~A~A~)"
-						    tran suffix))
-				   ,(intern
-				     (format nil "~:@(FAST-CONDITIONAL~A~)"
-					     suffix)))
-			(:translate ,tran)
-			(:generator ,cost
-				    (inst cmp x
-					  ,(if (eq suffix '-c/fixnum)
-					       '(fixnumize y)
-					       'y))
-				    (inst jmp (if not-p
-						  ,(if signed
-						       not-cond
-						       not-unsigned)
-						  ,(if signed
-						       cond
-						       unsigned))
-					  target))))
-		   '(/fixnum -c/fixnum /signed -c/signed /unsigned -c/unsigned)
-		   '(4 3 6 5 6 5)
-		   '(t t t t nil nil)))))
+             `(progn
+                ,@(mapcar
+                   (lambda (suffix cost signed)
+                     `(define-vop (;; FIXME: These could be done more
+                                   ;; cleanly with SYMBOLICATE.
+                                   ,(intern (format nil "~:@(FAST-IF-~A~A~)"
+                                                    tran suffix))
+                                   ,(intern
+                                     (format nil "~:@(FAST-CONDITIONAL~A~)"
+                                             suffix)))
+                        (:translate ,tran)
+                        (:generator ,cost
+                                    (inst cmp x
+                                          ,(if (eq suffix '-c/fixnum)
+                                               '(fixnumize y)
+                                               'y))
+                                    (inst jmp (if not-p
+                                                  ,(if signed
+                                                       not-cond
+                                                       not-unsigned)
+                                                  ,(if signed
+                                                       cond
+                                                       unsigned))
+                                          target))))
+                   '(/fixnum -c/fixnum /signed -c/signed /unsigned -c/unsigned)
+                   '(4 3 6 5 6 5)
+                   '(t t t t nil nil)))))
 
   (define-conditional-vop < :l :b :ge :ae)
   (define-conditional-vop > :g :a :le :be))
@@ -995,9 +1159,9 @@
   (:translate eql)
   (:generator 5
     (cond ((and (sc-is x signed-reg) (zerop y))
-	   (inst test x x))  ; smaller instruction
-	  (t
-	   (inst cmp x y)))
+           (inst test x x))  ; smaller instruction
+          (t
+           (inst cmp x y)))
     (inst jmp (if not-p :ne :e) target)))
 
 (define-vop (fast-if-eql/unsigned fast-conditional/unsigned)
@@ -1010,9 +1174,9 @@
   (:translate eql)
   (:generator 5
     (cond ((and (sc-is x unsigned-reg) (zerop y))
-	   (inst test x x))  ; smaller instruction
-	  (t
-	   (inst cmp x y)))
+           (inst test x x))  ; smaller instruction
+          (t
+           (inst cmp x y)))
     (inst jmp (if not-p :ne :e) target)))
 
 ;;; EQL/FIXNUM is funny because the first arg can be of any type, not just a
@@ -1026,9 +1190,9 @@
 
 (define-vop (fast-eql/fixnum fast-conditional)
   (:args (x :scs (any-reg)
-	    :load-if (not (and (sc-is x control-stack)
-			       (sc-is y any-reg))))
-	 (y :scs (any-reg control-stack)))
+            :load-if (not (and (sc-is x control-stack)
+                               (sc-is y any-reg))))
+         (y :scs (any-reg control-stack)))
   (:arg-types tagged-num tagged-num)
   (:note "inline fixnum comparison")
   (:translate eql)
@@ -1037,9 +1201,9 @@
     (inst jmp (if not-p :ne :e) target)))
 (define-vop (generic-eql/fixnum fast-eql/fixnum)
   (:args (x :scs (any-reg descriptor-reg)
-	    :load-if (not (and (sc-is x control-stack)
-			       (sc-is y any-reg))))
-	 (y :scs (any-reg control-stack)))
+            :load-if (not (and (sc-is x control-stack)
+                               (sc-is y any-reg))))
+         (y :scs (any-reg control-stack)))
   (:arg-types * tagged-num)
   (:variant-cost 7))
 
@@ -1050,9 +1214,9 @@
   (:translate eql)
   (:generator 2
     (cond ((and (sc-is x any-reg) (zerop y))
-	   (inst test x x))  ; smaller instruction
-	  (t
-	   (inst cmp x (fixnumize y))))
+           (inst test x x))  ; smaller instruction
+          (t
+           (inst cmp x (fixnumize y))))
     (inst jmp (if not-p :ne :e) target)))
 (define-vop (generic-eql-c/fixnum fast-eql-c/fixnum)
   (:args (x :scs (any-reg descriptor-reg control-stack)))
@@ -1064,8 +1228,8 @@
 (define-vop (merge-bits)
   (:translate merge-bits)
   (:args (shift :scs (signed-reg unsigned-reg) :target ecx)
-	 (prev :scs (unsigned-reg) :target result)
-	 (next :scs (unsigned-reg)))
+         (prev :scs (unsigned-reg) :target result)
+         (next :scs (unsigned-reg)))
   (:arg-types tagged-num unsigned-num unsigned-num)
   (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 0)) ecx)
   (:results (result :scs (unsigned-reg) :from (:argument 1)))
@@ -1076,44 +1240,11 @@
     (move result prev)
     (inst shrd result next :cl)))
 
-(define-source-transform 32bit-logical-not (x)
-  `(logand (lognot (the (unsigned-byte 32) ,x)) #.(1- (ash 1 32))))
-
-(deftransform 32bit-logical-and ((x y))
-  '(logand x y))
-
-(define-source-transform 32bit-logical-nand (x y)
-  `(32bit-logical-not (32bit-logical-and ,x ,y)))
-
-(deftransform 32bit-logical-or ((x y))
-  '(logior x y))
-
-(define-source-transform 32bit-logical-nor (x y)
-  `(32bit-logical-not (32bit-logical-or ,x ,y)))
-
-(deftransform 32bit-logical-xor ((x y))
-  '(logxor x y))
-
-(define-source-transform 32bit-logical-eqv (x y)
-  `(32bit-logical-not (32bit-logical-xor ,x ,y)))
-
-(define-source-transform 32bit-logical-orc1 (x y)
-  `(32bit-logical-or (32bit-logical-not ,x) ,y))
-
-(define-source-transform 32bit-logical-orc2 (x y)
-  `(32bit-logical-or ,x (32bit-logical-not ,y)))
-
-(define-source-transform 32bit-logical-andc1 (x y)
-  `(32bit-logical-and (32bit-logical-not ,x) ,y))
-
-(define-source-transform 32bit-logical-andc2 (x y)
-  `(32bit-logical-and ,x (32bit-logical-not ,y)))
-
 ;;; Only the lower 5 bits of the shift amount are significant.
 (define-vop (shift-towards-someplace)
   (:policy :fast-safe)
   (:args (num :scs (unsigned-reg) :target r)
-	 (amount :scs (signed-reg) :target ecx))
+         (amount :scs (signed-reg) :target ecx))
   (:arg-types unsigned-num tagged-num)
   (:temporary (:sc signed-reg :offset ecx-offset :from (:argument 1)) ecx)
   (:results (r :scs (unsigned-reg) :from (:argument 0)))
@@ -1136,51 +1267,177 @@
     (inst shl r :cl)))
 
 ;;;; Modular functions
+(defmacro define-mod-binop ((name prototype) function)
+  `(define-vop (,name ,prototype)
+       (:args (x :target r :scs (unsigned-reg signed-reg)
+                 :load-if (not (and (or (sc-is x unsigned-stack)
+                                        (sc-is x signed-stack))
+                                    (or (sc-is y unsigned-reg)
+                                        (sc-is y signed-reg))
+                                    (or (sc-is r unsigned-stack)
+                                        (sc-is r signed-stack))
+                                    (location= x r))))
+              (y :scs (unsigned-reg signed-reg unsigned-stack signed-stack)))
+     (:arg-types untagged-num untagged-num)
+     (:results (r :scs (unsigned-reg signed-reg) :from (:argument 0)
+                  :load-if (not (and (or (sc-is x unsigned-stack)
+                                         (sc-is x signed-stack))
+                                     (or (sc-is y unsigned-reg)
+                                         (sc-is y unsigned-reg))
+                                     (or (sc-is r unsigned-stack)
+                                         (sc-is r unsigned-stack))
+                                     (location= x r)))))
+     (:result-types unsigned-num)
+     (:translate ,function)))
+(defmacro define-mod-binop-c ((name prototype) function)
+  `(define-vop (,name ,prototype)
+       (:args (x :target r :scs (unsigned-reg signed-reg)
+                 :load-if (not (and (or (sc-is x unsigned-stack)
+                                        (sc-is x signed-stack))
+                                    (or (sc-is r unsigned-stack)
+                                        (sc-is r signed-stack))
+                                    (location= x r)))))
+     (:info y)
+     (:arg-types untagged-num (:constant (or (unsigned-byte 32) (signed-byte 32))))
+     (:results (r :scs (unsigned-reg signed-reg) :from (:argument 0)
+                  :load-if (not (and (or (sc-is x unsigned-stack)
+                                         (sc-is x signed-stack))
+                                     (or (sc-is r unsigned-stack)
+                                         (sc-is r unsigned-stack))
+                                     (location= x r)))))
+     (:result-types unsigned-num)
+     (:translate ,function)))
+
+(macrolet ((def (name -c-p)
+             (let ((fun32 (intern (format nil "~S-MOD32" name)))
+                   (vopu (intern (format nil "FAST-~S/UNSIGNED=>UNSIGNED" name)))
+                   (vopcu (intern (format nil "FAST-~S-C/UNSIGNED=>UNSIGNED" name)))
+                   (vopf (intern (format nil "FAST-~S/FIXNUM=>FIXNUM" name)))
+                   (vopcf (intern (format nil "FAST-~S-C/FIXNUM=>FIXNUM" name)))
+                   (vop32u (intern (format nil "FAST-~S-MOD32/WORD=>UNSIGNED" name)))
+                   (vop32f (intern (format nil "FAST-~S-MOD32/FIXNUM=>FIXNUM" name)))
+                   (vop32cu (intern (format nil "FAST-~S-MOD32-C/WORD=>UNSIGNED" name)))
+                   (vop32cf (intern (format nil "FAST-~S-MOD32-C/FIXNUM=>FIXNUM" name)))
+                   (sfun30 (intern (format nil "~S-SMOD30" name)))
+                   (svop30f (intern (format nil "FAST-~S-SMOD30/FIXNUM=>FIXNUM" name)))
+                   (svop30cf (intern (format nil "FAST-~S-SMOD30-C/FIXNUM=>FIXNUM" name))))
+               `(progn
+                  (define-modular-fun ,fun32 (x y) ,name :untagged nil 32)
+                  (define-modular-fun ,sfun30 (x y) ,name :tagged t 30)
+                  (define-mod-binop (,vop32u ,vopu) ,fun32)
+                  (define-vop (,vop32f ,vopf) (:translate ,fun32))
+                  (define-vop (,svop30f ,vopf) (:translate ,sfun30))
+                  ,@(when -c-p
+                      `((define-mod-binop-c (,vop32cu ,vopcu) ,fun32)
+                        (define-vop (,svop30cf ,vopcf) (:translate ,sfun30))))))))
+  (def + t)
+  (def - t)
+  ;; (no -C variant as x86 MUL instruction doesn't take an immediate)
+  (def * nil))
+
+
+(define-vop (fast-ash-left-mod32-c/unsigned=>unsigned
+             fast-ash-c/unsigned=>unsigned)
+  (:translate ash-left-mod32))
+
+(define-vop (fast-ash-left-mod32/unsigned=>unsigned
+             fast-ash-left/unsigned=>unsigned))
+(deftransform ash-left-mod32 ((integer count)
+                              ((unsigned-byte 32) (unsigned-byte 5)))
+  (when (sb!c::constant-lvar-p count)
+    (sb!c::give-up-ir1-transform))
+  '(%primitive fast-ash-left-mod32/unsigned=>unsigned integer count))
+
+(define-vop (fast-ash-left-smod30-c/fixnum=>fixnum
+             fast-ash-c/fixnum=>fixnum)
+  (:translate ash-left-smod30))
+
+(define-vop (fast-ash-left-smod30/fixnum=>fixnum
+             fast-ash-left/fixnum=>fixnum))
+(deftransform ash-left-smod30 ((integer count)
+                               ((signed-byte 30) (unsigned-byte 5)))
+  (when (sb!c::constant-lvar-p count)
+    (sb!c::give-up-ir1-transform))
+  '(%primitive fast-ash-left-smod30/fixnum=>fixnum integer count))
+
+(in-package "SB!C")
 
-(define-modular-fun +-mod32 (x y) + 32)
-(define-vop (fast-+-mod32/unsigned=>unsigned fast-+/unsigned=>unsigned)
-  (:translate +-mod32))
-(define-vop (fast-+-mod32-c/unsigned=>unsigned fast-+-c/unsigned=>unsigned)
-  (:translate +-mod32))
-(define-modular-fun --mod32 (x y) - 32)
-(define-vop (fast---mod32/unsigned=>unsigned fast--/unsigned=>unsigned)
-  (:translate --mod32))
-(define-vop (fast---mod32-c/unsigned=>unsigned fast---c/unsigned=>unsigned)
-  (:translate --mod32))
-
-(defknown ash-left-constant-mod32 (integer (integer 0)) (unsigned-byte 32)
+(defknown sb!vm::%lea-mod32 (integer integer (member 1 2 4 8) (signed-byte 32))
+  (unsigned-byte 32)
   (foldable flushable movable))
-(define-vop (fast-ash-left-constant-mod32/unsigned=>unsigned
-	     fast-ash-c/unsigned=>unsigned)
-  (:translate ash-left-constant-mod32))
+(defknown sb!vm::%lea-smod30 (integer integer (member 1 2 4 8) (signed-byte 32))
+  (signed-byte 30)
+  (foldable flushable movable))
+
+(define-modular-fun-optimizer %lea ((base index scale disp) :untagged nil :width width)
+  (when (and (<= width 32)
+             (constant-lvar-p scale)
+             (constant-lvar-p disp))
+    (cut-to-width base :untagged width nil)
+    (cut-to-width index :untagged width nil)
+    'sb!vm::%lea-mod32))
+(define-modular-fun-optimizer %lea ((base index scale disp) :tagged t :width width)
+  (when (and (<= width 30)
+             (constant-lvar-p scale)
+             (constant-lvar-p disp))
+    (cut-to-width base :tagged width t)
+    (cut-to-width index :tagged width t)
+    'sb!vm::%lea-smod30))
+
+#+sb-xc-host
+(progn
+  (defun sb!vm::%lea-mod32 (base index scale disp)
+    (ldb (byte 32 0) (%lea base index scale disp)))
+  (defun sb!vm::%lea-smod30 (base index scale disp)
+    (mask-signed-field 30 (%lea base index scale disp))))
+#-sb-xc-host
+(progn
+  (defun sb!vm::%lea-mod32 (base index scale disp)
+    (let ((base (logand base #xffffffff))
+          (index (logand index #xffffffff)))
+      ;; can't use modular version of %LEA, as we only have VOPs for
+      ;; constant SCALE and DISP.
+      (ldb (byte 32 0) (+ base (* index scale) disp))))
+  (defun sb!vm::%lea-smod30 (base index scale disp)
+    (let ((base (mask-signed-field 30 base))
+          (index (mask-signed-field 30 index)))
+      ;; can't use modular version of %LEA, as we only have VOPs for
+      ;; constant SCALE and DISP.
+      (mask-signed-field 30 (+ base (* index scale) disp)))))
+
+(in-package "SB!VM")
+
+(define-vop (%lea-mod32/unsigned=>unsigned
+             %lea/unsigned=>unsigned)
+  (:translate %lea-mod32))
+(define-vop (%lea-smod30/fixnum=>fixnum
+             %lea/fixnum=>fixnum)
+  (:translate %lea-smod30))
 
 ;;; logical operations
-(define-modular-fun lognot-mod32 (x) lognot 32)
-(define-vop (lognot-mod32/unsigned=>unsigned)
+(define-modular-fun lognot-mod32 (x) lognot :untagged nil 32)
+(define-vop (lognot-mod32/word=>unsigned)
   (:translate lognot-mod32)
-  (:args (x :scs (unsigned-reg unsigned-stack) :target r
-	    :load-if (not (and (sc-is x unsigned-stack)
-			       (sc-is r unsigned-stack)
-			       (location= x r)))))
+  (:args (x :scs (unsigned-reg signed-reg unsigned-stack signed-stack) :target r
+            :load-if (not (and (or (sc-is x unsigned-stack)
+                                   (sc-is x signed-stack))
+                               (or (sc-is r unsigned-stack)
+                                   (sc-is r signed-stack))
+                               (location= x r)))))
   (:arg-types unsigned-num)
   (:results (r :scs (unsigned-reg)
-	       :load-if (not (and (sc-is x unsigned-stack)
-				  (sc-is r unsigned-stack)
-				  (location= x r)))))
+               :load-if (not (and (or (sc-is x unsigned-stack)
+                                      (sc-is x signed-stack))
+                                  (or (sc-is r unsigned-stack)
+                                      (sc-is r signed-stack))
+                                  (sc-is r unsigned-stack)
+                                  (location= x r)))))
   (:result-types unsigned-num)
   (:policy :fast-safe)
   (:generator 1
     (move r x)
     (inst not r)))
 
-(define-modular-fun logxor-mod32 (x y) logxor 32)
-(define-vop (fast-logxor-mod32/unsigned=>unsigned
-             fast-logxor/unsigned=>unsigned)
-  (:translate logxor-mod32))
-(define-vop (fast-logxor-mod32-c/unsigned=>unsigned
-             fast-logxor-c/unsigned=>unsigned)
-  (:translate logxor-mod32))
-
 (define-source-transform logeqv (&rest args)
   (if (oddp (length args))
       `(logxor ,@args)
@@ -1201,21 +1458,23 @@
 ;;;; bignum stuff
 
 (define-vop (bignum-length get-header-data)
-  (:translate sb!bignum::%bignum-length)
+  (:translate sb!bignum:%bignum-length)
   (:policy :fast-safe))
 
 (define-vop (bignum-set-length set-header-data)
-  (:translate sb!bignum::%bignum-set-length)
+  (:translate sb!bignum:%bignum-set-length)
   (:policy :fast-safe))
 
 (define-full-reffer bignum-ref * bignum-digits-offset other-pointer-lowtag
-  (unsigned-reg) unsigned-num sb!bignum::%bignum-ref)
-
+  (unsigned-reg) unsigned-num sb!bignum:%bignum-ref)
+(define-full-reffer+offset bignum-ref-with-offset *
+  bignum-digits-offset other-pointer-lowtag
+  (unsigned-reg) unsigned-num sb!bignum:%bignum-ref-with-offset)
 (define-full-setter bignum-set * bignum-digits-offset other-pointer-lowtag
-  (unsigned-reg) unsigned-num sb!bignum::%bignum-set)
+  (unsigned-reg) unsigned-num sb!bignum:%bignum-set)
 
 (define-vop (digit-0-or-plus)
-  (:translate sb!bignum::%digit-0-or-plusp)
+  (:translate sb!bignum:%digit-0-or-plusp)
   (:policy :fast-safe)
   (:args (digit :scs (unsigned-reg)))
   (:arg-types unsigned-num)
@@ -1227,19 +1486,19 @@
 
 
 ;;; For add and sub with carry the sc of carry argument is any-reg so
-;;; the it may be passed as a fixnum or word and thus may be 0, 1, or
+;;; that it may be passed as a fixnum or word and thus may be 0, 1, or
 ;;; 4. This is easy to deal with and may save a fixnum-word
 ;;; conversion.
 (define-vop (add-w/carry)
-  (:translate sb!bignum::%add-with-carry)
+  (:translate sb!bignum:%add-with-carry)
   (:policy :fast-safe)
   (:args (a :scs (unsigned-reg) :target result)
-	 (b :scs (unsigned-reg unsigned-stack) :to :eval)
-	 (c :scs (any-reg) :target temp))
+         (b :scs (unsigned-reg unsigned-stack) :to :eval)
+         (c :scs (any-reg) :target temp))
   (:arg-types unsigned-num unsigned-num positive-fixnum)
   (:temporary (:sc any-reg :from (:argument 2) :to :eval) temp)
   (:results (result :scs (unsigned-reg) :from (:argument 0))
-	    (carry :scs (unsigned-reg)))
+            (carry :scs (unsigned-reg)))
   (:result-types unsigned-num positive-fixnum)
   (:generator 4
     (move result a)
@@ -1249,40 +1508,39 @@
     (inst mov carry 0)
     (inst adc carry carry)))
 
-;;; Note: the borrow is the oppostite of the x86 convention - 1 for no
-;;; borrow and 0 for a borrow.
+;;; Note: the borrow is 1 for no borrow and 0 for a borrow, the opposite
+;;; of the x86 convention.
 (define-vop (sub-w/borrow)
-  (:translate sb!bignum::%subtract-with-borrow)
+  (:translate sb!bignum:%subtract-with-borrow)
   (:policy :fast-safe)
   (:args (a :scs (unsigned-reg) :to :eval :target result)
-	 (b :scs (unsigned-reg unsigned-stack) :to :result)
-	 (c :scs (any-reg control-stack)))
+         (b :scs (unsigned-reg unsigned-stack) :to :result)
+         (c :scs (any-reg control-stack)))
   (:arg-types unsigned-num unsigned-num positive-fixnum)
   (:results (result :scs (unsigned-reg) :from :eval)
-	    (borrow :scs (unsigned-reg)))
+            (borrow :scs (unsigned-reg)))
   (:result-types unsigned-num positive-fixnum)
   (:generator 5
     (inst cmp c 1) ; Set the carry flag to 1 if c=0 else to 0
     (move result a)
     (inst sbb result b)
-    (inst mov borrow 0)
-    (inst adc borrow borrow)
-    (inst xor borrow 1)))
+    (inst mov borrow 1)
+    (inst sbb borrow 0)))
 
 
 (define-vop (bignum-mult-and-add-3-arg)
-  (:translate sb!bignum::%multiply-and-add)
+  (:translate sb!bignum:%multiply-and-add)
   (:policy :fast-safe)
   (:args (x :scs (unsigned-reg) :target eax)
-	 (y :scs (unsigned-reg unsigned-stack))
-	 (carry-in :scs (unsigned-reg unsigned-stack)))
+         (y :scs (unsigned-reg unsigned-stack))
+         (carry-in :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num unsigned-num)
   (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
-		   :to (:result 1) :target lo) eax)
+                   :to (:result 1) :target lo) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
-		   :to (:result 0) :target hi) edx)
+                   :to (:result 0) :target hi) edx)
   (:results (hi :scs (unsigned-reg))
-	    (lo :scs (unsigned-reg)))
+            (lo :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:generator 20
     (move eax x)
@@ -1293,19 +1551,19 @@
     (move lo eax)))
 
 (define-vop (bignum-mult-and-add-4-arg)
-  (:translate sb!bignum::%multiply-and-add)
+  (:translate sb!bignum:%multiply-and-add)
   (:policy :fast-safe)
   (:args (x :scs (unsigned-reg) :target eax)
-	 (y :scs (unsigned-reg unsigned-stack))
-	 (prev :scs (unsigned-reg unsigned-stack))
-	 (carry-in :scs (unsigned-reg unsigned-stack)))
+         (y :scs (unsigned-reg unsigned-stack))
+         (prev :scs (unsigned-reg unsigned-stack))
+         (carry-in :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num unsigned-num unsigned-num)
   (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
-		   :to (:result 1) :target lo) eax)
+                   :to (:result 1) :target lo) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
-		   :to (:result 0) :target hi) edx)
+                   :to (:result 0) :target hi) edx)
   (:results (hi :scs (unsigned-reg))
-	    (lo :scs (unsigned-reg)))
+            (lo :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:generator 20
     (move eax x)
@@ -1319,17 +1577,17 @@
 
 
 (define-vop (bignum-mult)
-  (:translate sb!bignum::%multiply)
+  (:translate sb!bignum:%multiply)
   (:policy :fast-safe)
   (:args (x :scs (unsigned-reg) :target eax)
-	 (y :scs (unsigned-reg unsigned-stack)))
+         (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
   (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0)
-		   :to (:result 1) :target lo) eax)
+                   :to (:result 1) :target lo) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
-		   :to (:result 0) :target hi) edx)
+                   :to (:result 0) :target hi) edx)
   (:results (hi :scs (unsigned-reg))
-	    (lo :scs (unsigned-reg)))
+            (lo :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:generator 20
     (move eax x)
@@ -1337,36 +1595,36 @@
     (move hi edx)
     (move lo eax)))
 
-(define-vop (bignum-lognot lognot-mod32/unsigned=>unsigned)
-  (:translate sb!bignum::%lognot))
+(define-vop (bignum-lognot lognot-mod32/word=>unsigned)
+  (:translate sb!bignum:%lognot))
 
 (define-vop (fixnum-to-digit)
-  (:translate sb!bignum::%fixnum-to-digit)
+  (:translate sb!bignum:%fixnum-to-digit)
   (:policy :fast-safe)
   (:args (fixnum :scs (any-reg control-stack) :target digit))
   (:arg-types tagged-num)
   (:results (digit :scs (unsigned-reg)
-		   :load-if (not (and (sc-is fixnum control-stack)
-				      (sc-is digit unsigned-stack)
-				      (location= fixnum digit)))))
+                   :load-if (not (and (sc-is fixnum control-stack)
+                                      (sc-is digit unsigned-stack)
+                                      (location= fixnum digit)))))
   (:result-types unsigned-num)
   (:generator 1
     (move digit fixnum)
     (inst sar digit 2)))
 
 (define-vop (bignum-floor)
-  (:translate sb!bignum::%floor)
+  (:translate sb!bignum:%floor)
   (:policy :fast-safe)
   (:args (div-high :scs (unsigned-reg) :target edx)
-	 (div-low :scs (unsigned-reg) :target eax)
-	 (divisor :scs (unsigned-reg unsigned-stack)))
+         (div-low :scs (unsigned-reg) :target eax)
+         (divisor :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num unsigned-num)
   (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 1)
-		   :to (:result 0) :target quo) eax)
+                   :to (:result 0) :target quo) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 0)
-		   :to (:result 1) :target rem) edx)
+                   :to (:result 1) :target rem) edx)
   (:results (quo :scs (unsigned-reg))
-	    (rem :scs (unsigned-reg)))
+            (rem :scs (unsigned-reg)))
   (:result-types unsigned-num unsigned-num)
   (:generator 300
     (move edx div-high)
@@ -1376,14 +1634,14 @@
     (move rem edx)))
 
 (define-vop (signify-digit)
-  (:translate sb!bignum::%fixnum-digit-with-correct-sign)
+  (:translate sb!bignum:%fixnum-digit-with-correct-sign)
   (:policy :fast-safe)
   (:args (digit :scs (unsigned-reg unsigned-stack) :target res))
   (:arg-types unsigned-num)
   (:results (res :scs (any-reg signed-reg)
-		 :load-if (not (and (sc-is digit unsigned-stack)
-				    (sc-is res control-stack signed-stack)
-				    (location= digit res)))))
+                 :load-if (not (and (sc-is digit unsigned-stack)
+                                    (sc-is res control-stack signed-stack)
+                                    (location= digit res)))))
   (:result-types signed-num)
   (:generator 1
     (move res digit)
@@ -1391,30 +1649,44 @@
       (inst shl res 2))))
 
 (define-vop (digit-ashr)
-  (:translate sb!bignum::%ashr)
+  (:translate sb!bignum:%ashr)
   (:policy :fast-safe)
   (:args (digit :scs (unsigned-reg unsigned-stack) :target result)
-	 (count :scs (unsigned-reg) :target ecx))
+         (count :scs (unsigned-reg) :target ecx))
   (:arg-types unsigned-num positive-fixnum)
   (:temporary (:sc unsigned-reg :offset ecx-offset :from (:argument 1)) ecx)
   (:results (result :scs (unsigned-reg) :from (:argument 0)
-		    :load-if (not (and (sc-is result unsigned-stack)
-				       (location= digit result)))))
+                    :load-if (not (and (sc-is result unsigned-stack)
+                                       (location= digit result)))))
   (:result-types unsigned-num)
-  (:generator 1
+  (:generator 2
     (move result digit)
     (move ecx count)
     (inst sar result :cl)))
 
+(define-vop (digit-ashr/c)
+  (:translate sb!bignum:%ashr)
+  (:policy :fast-safe)
+  (:args (digit :scs (unsigned-reg unsigned-stack) :target result))
+  (:arg-types unsigned-num (:constant (integer 0 31)))
+  (:info count)
+  (:results (result :scs (unsigned-reg) :from (:argument 0)
+                    :load-if (not (and (sc-is result unsigned-stack)
+                                       (location= digit result)))))
+  (:result-types unsigned-num)
+  (:generator 1
+    (move result digit)
+    (inst sar result count)))
+
 (define-vop (digit-lshr digit-ashr)
-  (:translate sb!bignum::%digit-logical-shift-right)
+  (:translate sb!bignum:%digit-logical-shift-right)
   (:generator 1
     (move result digit)
     (move ecx count)
     (inst shr result :cl)))
 
 (define-vop (digit-ashl digit-ashr)
-  (:translate sb!bignum::%ashl)
+  (:translate sb!bignum:%ashl)
   (:generator 1
     (move result digit)
     (move ecx count)
@@ -1453,41 +1725,28 @@
   (:arg-types simple-array-unsigned-byte-32)
   (:temporary (:sc unsigned-reg :from (:eval 0) :to :result) k)
   (:temporary (:sc unsigned-reg :offset eax-offset
-		   :from (:eval 0) :to :result) tmp)
+                   :from (:eval 0) :to :result) tmp)
   (:results (y :scs (unsigned-reg) :from (:eval 0)))
   (:result-types unsigned-num)
   (:generator 50
-    (inst mov k (make-ea :dword :base state
-			 :disp (- (* (+ 2 vector-data-offset)
-				     n-word-bytes)
-				  other-pointer-lowtag)))
+    (loadw k state (+ 2 vector-data-offset) other-pointer-lowtag)
     (inst cmp k 624)
     (inst jmp :ne no-update)
-    (inst mov tmp state)	; The state is passed in EAX.
+    (inst mov tmp state)        ; The state is passed in EAX.
     (inst call (make-fixup 'random-mt19937-update :assembly-routine))
     ;; Restore k, and set to 0.
     (inst xor k k)
     NO-UPDATE
     ;; y = ptgfsr[k++];
-    (inst mov y (make-ea :dword :base state :index k :scale 4
-			 :disp (- (* (+ 3 vector-data-offset)
-				     n-word-bytes)
-				  other-pointer-lowtag)))
+    (inst mov y (make-ea-for-vector-data state :index k :offset 3))
     ;; y ^= (y >> 11);
     (inst shr y 11)
-    (inst xor y (make-ea :dword :base state :index k :scale 4
-			 :disp (- (* (+ 3 vector-data-offset)
-				     n-word-bytes)
-				  other-pointer-lowtag)))
+    (inst xor y (make-ea-for-vector-data state :index k :offset 3))
     ;; y ^= (y << 7) & #x9d2c5680
     (inst mov tmp y)
     (inst inc k)
     (inst shl tmp 7)
-    (inst mov (make-ea :dword :base state
-		       :disp (- (* (+ 2 vector-data-offset)
-				   n-word-bytes)
-				other-pointer-lowtag))
-	  k)
+    (storew k state (+ 2 vector-data-offset) other-pointer-lowtag)
     (inst and tmp #x9d2c5680)
     (inst xor y tmp)
     ;; y ^= (y << 15) & #xefc60000
@@ -1502,175 +1761,127 @@
 
 (in-package "SB!C")
 
-(defknown %lea ((or (signed-byte 32) (unsigned-byte 32))
-		(or (signed-byte 32) (unsigned-byte 32))
-		(member 1 2 4 8) (signed-byte 32))
-  (or (signed-byte 32) (unsigned-byte 32))
-  (foldable flushable))
-
-(defoptimizer (%lea derive-type) ((base index scale disp))
-  (when (and (constant-lvar-p scale)
-	     (constant-lvar-p disp))
-    (let ((scale (lvar-value scale))
-	  (disp (lvar-value disp))
-	  (base-type (lvar-type base))
-	  (index-type (lvar-type index)))
-      (when (and (numeric-type-p base-type)
-		 (numeric-type-p index-type))
-	(let ((base-lo (numeric-type-low base-type))
-	      (base-hi (numeric-type-high base-type))
-	      (index-lo (numeric-type-low index-type))
-	      (index-hi (numeric-type-high index-type)))
-	  (make-numeric-type :class 'integer
-			     :complexp :real
-			     :low (when (and base-lo index-lo)
-				    (+ base-lo (* index-lo scale) disp))
-			     :high (when (and base-hi index-hi)
-				     (+ base-hi (* index-hi scale) disp))))))))
-
-(defun %lea (base index scale disp)
-  (+ base (* index scale) disp))
-
-(in-package "SB!VM")
-
-(define-vop (%lea/unsigned=>unsigned)
-  (:translate %lea)
-  (:policy :fast-safe)
-  (:args (base :scs (unsigned-reg))
-	 (index :scs (unsigned-reg)))
-  (:info scale disp)
-  (:arg-types unsigned-num unsigned-num
-	      (:constant (member 1 2 4 8))
-	      (:constant (signed-byte 32)))
-  (:results (r :scs (unsigned-reg)))
-  (:result-types unsigned-num)
-  (:generator 5
-    (inst lea r (make-ea :dword :base base :index index
-			 :scale scale :disp disp))))
-
-(define-vop (%lea/signed=>signed)
-  (:translate %lea)
-  (:policy :fast-safe)
-  (:args (base :scs (signed-reg))
-	 (index :scs (signed-reg)))
-  (:info scale disp)
-  (:arg-types signed-num signed-num
-	      (:constant (member 1 2 4 8))
-	      (:constant (signed-byte 32)))
-  (:results (r :scs (signed-reg)))
-  (:result-types signed-num)
-  (:generator 4
-    (inst lea r (make-ea :dword :base base :index index
-			 :scale scale :disp disp))))
-
-(define-vop (%lea/fixnum=>fixnum)
-  (:translate %lea)
-  (:policy :fast-safe)
-  (:args (base :scs (any-reg))
-	 (index :scs (any-reg)))
-  (:info scale disp)
-  (:arg-types tagged-num tagged-num
-	      (:constant (member 1 2 4 8))
-	      (:constant (signed-byte 32)))
-  (:results (r :scs (any-reg)))
-  (:result-types tagged-num)
-  (:generator 3
-    (inst lea r (make-ea :dword :base base :index index
-			 :scale scale :disp disp))))
-
-(in-package "SB!C")
+(defun mask-result (class width result)
+  (ecase class
+    (:unsigned
+     `(logand ,result ,(1- (ash 1 width))))
+    (:signed
+     `(mask-signed-field ,width ,result))))
 
 ;;; This is essentially a straight implementation of the algorithm in
 ;;; "Strength Reduction of Multiplications by Integer Constants",
 ;;; Youfeng Wu, ACM SIGPLAN Notices, Vol. 30, No.2, February 1995.
-(defun basic-decompose-multiplication (arg num n-bits condensed)
+(defun basic-decompose-multiplication (class width arg num n-bits condensed)
   (case (aref condensed 0)
     (0
      (let ((tmp (min 3 (aref condensed 1))))
        (decf (aref condensed 1) tmp)
-       `(logand #xffffffff
-	 (%lea ,arg
-	       ,(decompose-multiplication
-		 arg (ash (1- num) (- tmp)) (1- n-bits) (subseq condensed 1))
-	       ,(ash 1 tmp) 0))))
+       (mask-result class width
+                    `(%lea ,arg
+                           ,(decompose-multiplication class width
+                             arg (ash (1- num) (- tmp)) (1- n-bits) (subseq condensed 1))
+                           ,(ash 1 tmp) 0))))
     ((1 2 3)
      (let ((r0 (aref condensed 0)))
        (incf (aref condensed 1) r0)
-       `(logand #xffffffff
-	 (%lea ,(decompose-multiplication
-		 arg (- num (ash 1 r0)) (1- n-bits) (subseq condensed 1))
-	       ,arg
-	       ,(ash 1 r0) 0))))
+       (mask-result class width
+                    `(%lea ,(decompose-multiplication class width
+                             arg (- num (ash 1 r0)) (1- n-bits) (subseq condensed 1))
+                           ,arg
+                           ,(ash 1 r0) 0))))
     (t (let ((r0 (aref condensed 0)))
-	 (setf (aref condensed 0) 0)
-	 `(logand #xffffffff
-	   (ash ,(decompose-multiplication
-		  arg (ash num (- r0)) n-bits condensed)
-	        ,r0))))))
+         (setf (aref condensed 0) 0)
+         (mask-result class width
+                      `(ash ,(decompose-multiplication class width
+                              arg (ash num (- r0)) n-bits condensed)
+                            ,r0))))))
 
-(defun decompose-multiplication (arg num n-bits condensed)
+(defun decompose-multiplication (class width arg num n-bits condensed)
   (cond
     ((= n-bits 0) 0)
     ((= num 1) arg)
     ((= n-bits 1)
-     `(logand #xffffffff (ash ,arg ,(1- (integer-length num)))))
+     (mask-result class width `(ash ,arg ,(1- (integer-length num)))))
     ((let ((max 0) (end 0))
        (loop for i from 2 to (length condensed)
-	     for j = (reduce #'+ (subseq condensed 0 i))
-	     when (and (> (- (* 2 i) 3 j) max)
-		       (< (+ (ash 1 (1+ j))
-			     (ash (ldb (byte (- 32 (1+ j)) (1+ j)) num)
-				  (1+ j)))
-			  (ash 1 32)))
-	       do (setq max (- (* 2 i) 3 j)
-			end i))
+             for j = (reduce #'+ (subseq condensed 0 i))
+             when (and (> (- (* 2 i) 3 j) max)
+                       (< (+ (ash 1 (1+ j))
+                             (ash (ldb (byte (- 32 (1+ j)) (1+ j)) num)
+                                  (1+ j)))
+                          (ash 1 32)))
+               do (setq max (- (* 2 i) 3 j)
+                        end i))
        (when (> max 0)
-	 (let ((j (reduce #'+ (subseq condensed 0 end))))
-	   (let ((n2 (+ (ash 1 (1+ j))
-			(ash (ldb (byte (- 32 (1+ j)) (1+ j)) num) (1+ j))))
-		 (n1 (1+ (ldb (byte (1+ j) 0) (lognot num)))))
-	   `(logand #xffffffff
-	     (- ,(optimize-multiply arg n2) ,(optimize-multiply arg n1))))))))
+         (let ((j (reduce #'+ (subseq condensed 0 end))))
+           (let ((n2 (+ (ash 1 (1+ j))
+                        (ash (ldb (byte (- 32 (1+ j)) (1+ j)) num) (1+ j))))
+                 (n1 (1+ (ldb (byte (1+ j) 0) (lognot num)))))
+           (mask-result class width
+                        `(- ,(optimize-multiply class width arg n2)
+                            ,(optimize-multiply  class width arg n1))))))))
     ((dolist (i '(9 5 3))
        (when (integerp (/ num i))
-	 (when (< (logcount (/ num i)) (logcount num))
-	   (let ((x (gensym)))
-	     (return `(let ((,x ,(optimize-multiply arg (/ num i))))
-		       (logand #xffffffff
-			(%lea ,x ,x (1- ,i) 0)))))))))
-    (t (basic-decompose-multiplication arg num n-bits condensed))))
-	   
-(defun optimize-multiply (arg x)
+         (when (< (logcount (/ num i)) (logcount num))
+           (let ((x (gensym)))
+             (return `(let ((,x ,(optimize-multiply class width arg (/ num i))))
+                       ,(mask-result class width
+                                     `(%lea ,x ,x (1- ,i) 0)))))))))
+    (t (basic-decompose-multiplication class width arg num n-bits condensed))))
+
+(defun optimize-multiply (class width arg x)
   (let* ((n-bits (logcount x))
-	 (condensed (make-array n-bits)))
+         (condensed (make-array n-bits)))
     (let ((count 0) (bit 0))
       (dotimes (i 32)
-	(cond ((logbitp i x)
-	       (setf (aref condensed bit) count)
-	       (setf count 1)
-	       (incf bit))
-	      (t (incf count)))))
-    (decompose-multiplication arg x n-bits condensed)))
+        (cond ((logbitp i x)
+               (setf (aref condensed bit) count)
+               (setf count 1)
+               (incf bit))
+              (t (incf count)))))
+    (decompose-multiplication class width arg x n-bits condensed)))
+
+(defun *-transformer (class width y)
+  (cond
+    ((= y (ash 1 (integer-length y)))
+     ;; there's a generic transform for y = 2^k
+     (give-up-ir1-transform))
+    ((member y '(3 5 9))
+     ;; we can do these multiplications directly using LEA
+     `(%lea x x ,(1- y) 0))
+    ((member :pentium4 *backend-subfeatures*)
+     ;; the pentium4's multiply unit is reportedly very good
+     (give-up-ir1-transform))
+    ;; FIXME: should make this more fine-grained.  If nothing else,
+    ;; there should probably be a cutoff of about 9 instructions on
+    ;; pentium-class machines.
+    (t (optimize-multiply class width 'x y))))
 
 (deftransform * ((x y)
-		 ((unsigned-byte 32) (constant-arg (unsigned-byte 32)))
-		 (unsigned-byte 32))
+                 ((unsigned-byte 32) (constant-arg (unsigned-byte 32)))
+                 (unsigned-byte 32))
+  "recode as leas, shifts and adds"
+  (let ((y (lvar-value y)))
+    (*-transformer :unsigned 32 y)))
+(deftransform sb!vm::*-mod32
+    ((x y) ((unsigned-byte 32) (constant-arg (unsigned-byte 32)))
+     (unsigned-byte 32))
+  "recode as leas, shifts and adds"
+  (let ((y (lvar-value y)))
+    (*-transformer :unsigned 32 y)))
+
+(deftransform * ((x y)
+                 ((signed-byte 30) (constant-arg (unsigned-byte 32)))
+                 (signed-byte 30))
+  "recode as leas, shifts and adds"
+  (let ((y (lvar-value y)))
+    (*-transformer :signed 30 y)))
+(deftransform sb!vm::*-smod30
+    ((x y) ((signed-byte 30) (constant-arg (unsigned-byte 32)))
+     (signed-byte 30))
   "recode as leas, shifts and adds"
   (let ((y (lvar-value y)))
-    (cond
-      ((= y (ash 1 (integer-length y)))
-       ;; there's a generic transform for y = 2^k
-       (give-up-ir1-transform))
-      ((member y '(3 5 9))
-       ;; we can do these multiplications directly using LEA
-       `(%lea x x ,(1- y) 0))
-      ((member :pentium4 *backend-subfeatures*)
-       ;; the pentium4's multiply unit is reportedly very good
-       (give-up-ir1-transform))
-      ;; FIXME: should make this more fine-grained.  If nothing else,
-      ;; there should probably be a cutoff of about 9 instructions on
-      ;; pentium-class machines.
-      (t (optimize-multiply 'x y)))))
+    (*-transformer :signed 30 y)))
 
 ;;; FIXME: we should also be able to write an optimizer or two to
 ;;; convert (+ (* x 2) 17), (- (* x 9) 5) to a %LEA.