X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fx86-64%2Farith.lisp;h=0e0be28edea74f98dcf3974c72967a7f6895b8a3;hb=179812c3ad5dad69239c625ec929a7d486cf568f;hp=304590315fde2ecefede6d5e12f4404910b6dbef;hpb=0d871fd7a98fc4af92a8b942a1154761466ad8c9;p=sbcl.git

diff --git a/src/compiler/x86-64/arith.lisp b/src/compiler/x86-64/arith.lisp
index 3045903..0e0be28 100644
--- a/src/compiler/x86-64/arith.lisp
+++ b/src/compiler/x86-64/arith.lisp
@@ -1,4 +1,4 @@
-;;;; the VM definition of arithmetic VOPs for the x86
+;;;; the VM definition of arithmetic VOPs for the x86-64
 
 ;;;; 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)))
 
@@ -109,31 +109,38 @@
   (:note "inline (signed-byte 64) arithmetic"))
 
 (define-vop (fast-fixnum-binop-c fast-safe-arith-op)
-  (:args (x :target r :scs (any-reg control-stack)))
+  (:args (x :target r :scs (any-reg)
+            :load-if (or (not (typep y '(signed-byte 29)))
+                         (not (sc-is x any-reg control-stack)))))
   (:info y)
-  (:arg-types tagged-num (:constant (signed-byte 29)))
+  (:arg-types tagged-num (:constant fixnum))
   (:results (r :scs (any-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(signed-byte 29))))))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic"))
 
-;; 31 not 64 because it's hard work loading 64 bit constants, and since
-;; sign-extension of immediates causes problems with 32.
 (define-vop (fast-unsigned-binop-c fast-safe-arith-op)
-  (:args (x :target r :scs (unsigned-reg unsigned-stack)))
+  (:args (x :target r :scs (unsigned-reg)
+            :load-if (or (not (typep y '(unsigned-byte 31)))
+                         (not (sc-is x unsigned-reg unsigned-stack)))))
   (:info y)
-  (:arg-types unsigned-num (:constant (unsigned-byte 31)))
+  (:arg-types unsigned-num (:constant (unsigned-byte 64)))
   (:results (r :scs (unsigned-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(unsigned-byte 31))))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 64) arithmetic"))
 
 (define-vop (fast-signed-binop-c fast-safe-arith-op)
-  (:args (x :target r :scs (signed-reg signed-stack)))
+  (:args (x :target r :scs (signed-reg)
+            :load-if (or (not (typep y '(signed-byte 32)))
+                         (not (sc-is x signed-reg signed-stack)))))
   (:info y)
-  (:arg-types signed-num (:constant (signed-byte 32)))
+  (:arg-types signed-num (:constant (signed-byte 64)))
   (:results (r :scs (signed-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(signed-byte 32))))))
   (:result-types signed-num)
   (:note "inline (signed-byte 64) arithmetic"))
 
@@ -150,7 +157,9 @@
                   (:translate ,translate)
                   (:generator 1
                   (move r x)
-                  (inst ,op r (fixnumize y))))
+                  (inst ,op r (if (typep y '(signed-byte 29))
+                                  (fixnumize y)
+                                  (register-inline-constant :qword (fixnumize y))))))
                 (define-vop (,(symbolicate "FAST-" translate "/SIGNED=>SIGNED")
                              fast-signed-binop)
                   (:translate ,translate)
@@ -162,7 +171,9 @@
                   (:translate ,translate)
                   (:generator ,untagged-penalty
                   (move r x)
-                  (inst ,op r y)))
+                  (inst ,op r (if (typep y '(signed-byte 32))
+                                  y
+                                  (register-inline-constant :qword y)))))
                 (define-vop (,(symbolicate "FAST-"
                                            translate
                                            "/UNSIGNED=>UNSIGNED")
@@ -178,7 +189,9 @@
                   (:translate ,translate)
                   (:generator ,untagged-penalty
                   (move r x)
-                  (inst ,op r y))))))
+                  (inst ,op r (if (typep y '(unsigned-byte 31))
+                                  y
+                                  (register-inline-constant :qword y))))))))
 
   ;;(define-binop + 4 add)
   (define-binop - 4 sub)
@@ -214,19 +227,26 @@
 
 (define-vop (fast-+-c/fixnum=>fixnum fast-safe-arith-op)
   (:translate +)
-  (:args (x :target r :scs (any-reg control-stack)))
+  (:args (x :target r :scs (any-reg)
+            :load-if (or (not (typep y '(signed-byte 29)))
+                         (not (sc-is x any-reg control-stack)))))
   (:info y)
-  (:arg-types tagged-num (:constant (signed-byte 29)))
+  (:arg-types tagged-num (:constant fixnum))
   (:results (r :scs (any-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(signed-byte 29))))))
   (: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)))
+    (cond ((and (sc-is x any-reg) (sc-is r any-reg) (not (location= x r))
+                (typep y '(signed-byte 29)))
            (inst lea r (make-ea :qword :base x :disp (fixnumize y))))
+          ((typep y '(signed-byte 29))
+           (move r x)
+           (inst add r (fixnumize y)))
           (t
            (move r x)
-           (inst add r (fixnumize y))))))
+           (inst add r (register-inline-constant :qword (fixnumize y)))))))
 
 (define-vop (fast-+/signed=>signed fast-safe-arith-op)
   (:translate +)
@@ -266,8 +286,10 @@
 
 (define-vop (fast-logand-c/signed-unsigned=>unsigned
              fast-logand-c/unsigned=>unsigned)
-  (:args (x :target r :scs (signed-reg signed-stack)))
-  (:arg-types signed-num (:constant (unsigned-byte 31))))
+  (:args (x :target r :scs (signed-reg)
+            :load-if (or (not (typep y '(unsigned-byte 31)))
+                         (not (sc-is r signed-reg signed-stack)))))
+  (:arg-types signed-num (:constant (unsigned-byte 64))))
 
 (define-vop (fast-logand/unsigned-signed=>unsigned
              fast-logand/unsigned=>unsigned)
@@ -282,22 +304,29 @@
 
 (define-vop (fast-+-c/signed=>signed fast-safe-arith-op)
   (:translate +)
-  (:args (x :target r :scs (signed-reg signed-stack)))
+  (:args (x :target r :scs (signed-reg)
+            :load-if (or (not (typep y '(signed-byte 32)))
+                         (not (sc-is r signed-reg signed-stack)))))
   (:info y)
-  (:arg-types signed-num (:constant (signed-byte 32)))
+  (:arg-types signed-num (:constant (signed-byte 64)))
   (:results (r :scs (signed-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(signed-byte 32))))))
   (:result-types signed-num)
   (:note "inline (signed-byte 64) arithmetic")
   (:generator 4
     (cond ((and (sc-is x signed-reg) (sc-is r signed-reg)
-                (not (location= x r)))
+                (not (location= x r))
+                (typep y '(signed-byte 32)))
            (inst lea r (make-ea :qword :base x :disp y)))
           (t
            (move r x)
-           (if (= y 1)
-               (inst inc r)
-             (inst add r y))))))
+           (cond ((= y 1)
+                  (inst inc r))
+                 ((typep y '(signed-byte 32))
+                  (inst add r y))
+                 (t
+                  (inst add r (register-inline-constant :qword y))))))))
 
 (define-vop (fast-+/unsigned=>unsigned fast-safe-arith-op)
   (:translate +)
@@ -325,22 +354,29 @@
 
 (define-vop (fast-+-c/unsigned=>unsigned fast-safe-arith-op)
   (:translate +)
-  (:args (x :target r :scs (unsigned-reg unsigned-stack)))
+  (:args (x :target r :scs (unsigned-reg)
+            :load-if (or (not (typep y '(unsigned-byte 31)))
+                         (not (sc-is x unsigned-reg unsigned-stack)))))
   (:info y)
-  (:arg-types unsigned-num (:constant (unsigned-byte 31)))
+  (:arg-types unsigned-num (:constant (unsigned-byte 64)))
   (:results (r :scs (unsigned-reg)
-               :load-if (not (location= x r))))
+               :load-if (or (not (location= x r))
+                            (not (typep y '(unsigned-byte 31))))))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 64) arithmetic")
   (:generator 4
     (cond ((and (sc-is x unsigned-reg) (sc-is r unsigned-reg)
-                (not (location= x r)))
+                (not (location= x r))
+                (typep y '(unsigned-byte 31)))
            (inst lea r (make-ea :qword :base x :disp y)))
           (t
            (move r x)
-           (if (= y 1)
-               (inst inc r)
-             (inst add r y))))))
+           (cond ((= y 1)
+                  (inst inc r))
+                 ((typep y '(unsigned-byte 31))
+                  (inst add r y))
+                 (t
+                  (inst add r (register-inline-constant :qword y))))))))
 
 ;;;; multiplication and division
 
@@ -355,20 +391,26 @@
   (:note "inline fixnum arithmetic")
   (:generator 4
     (move r x)
-    (inst sar r 3)
+    (inst sar r n-fixnum-tag-bits)
     (inst imul r y)))
 
 (define-vop (fast-*-c/fixnum=>fixnum fast-safe-arith-op)
   (:translate *)
   ;; We need different loading characteristics.
-  (:args (x :scs (any-reg control-stack)))
+  (:args (x :scs (any-reg)
+            :load-if (or (not (typep y '(signed-byte 32)))
+                         (not (sc-is x any-reg control-stack)))))
   (:info y)
-  (:arg-types tagged-num (:constant (signed-byte 29)))
+  (:arg-types tagged-num (:constant fixnum))
   (:results (r :scs (any-reg)))
   (:result-types tagged-num)
   (:note "inline fixnum arithmetic")
   (:generator 3
-    (inst imul r x y)))
+    (cond ((typep y '(signed-byte 32))
+           (inst imul r x y))
+          (t
+           (move r x)
+           (inst imul r (register-inline-constant :qword y))))))
 
 (define-vop (fast-*/signed=>signed fast-safe-arith-op)
   (:translate *)
@@ -386,26 +428,32 @@
 (define-vop (fast-*-c/signed=>signed fast-safe-arith-op)
   (:translate *)
   ;; We need different loading characteristics.
-  (:args (x :scs (signed-reg signed-stack)))
+  (:args (x :scs (signed-reg)
+            :load-if (or (not (typep y '(signed-byte 32)))
+                         (not (sc-is x signed-reg signed-stack)))))
   (:info y)
-  (:arg-types signed-num (:constant (signed-byte 32)))
+  (:arg-types signed-num (:constant (signed-byte 64)))
   (:results (r :scs (signed-reg)))
   (:result-types signed-num)
   (:note "inline (signed-byte 64) arithmetic")
   (:generator 4
-    (inst imul r x y)))
+    (cond ((typep y '(signed-byte 32))
+           (inst imul r x y))
+          (t
+           (move r x)
+           (inst imul r (register-inline-constant :qword y))))))
 
 (define-vop (fast-*/unsigned=>unsigned fast-safe-arith-op)
   (:translate *)
   (:args (x :scs (unsigned-reg) :target eax)
          (y :scs (unsigned-reg unsigned-stack)))
   (:arg-types unsigned-num unsigned-num)
-  (:temporary (:sc unsigned-reg :offset eax-offset :target result
+  (: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)
   (:ignore edx)
-  (:results (result :scs (unsigned-reg)))
+  (:results (r :scs (unsigned-reg)))
   (:result-types unsigned-num)
   (:note "inline (unsigned-byte 64) arithmetic")
   (:vop-var vop)
@@ -413,7 +461,27 @@
   (:generator 6
     (move eax x)
     (inst mul eax y)
-    (move result eax)))
+    (move r eax)))
+
+(define-vop (fast-*-c/unsigned=>unsigned fast-safe-arith-op)
+  (:translate *)
+  (:args (x :scs (unsigned-reg) :target eax))
+  (:info y)
+  (:arg-types unsigned-num (:constant (unsigned-byte 64)))
+  (: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)
+  (:ignore edx)
+  (:results (r :scs (unsigned-reg)))
+  (:result-types unsigned-num)
+  (:note "inline (unsigned-byte 64) arithmetic")
+  (:vop-var vop)
+  (:save-p :compute-only)
+  (:generator 6
+    (move eax x)
+    (inst mul eax (register-inline-constant :qword y))
+    (move r eax)))
 
 
 (define-vop (fast-truncate/fixnum=>fixnum fast-safe-arith-op)
@@ -432,7 +500,7 @@
   (: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))
@@ -441,15 +509,16 @@
     (inst cqo)
     (inst idiv eax y)
     (if (location= quo eax)
-        (inst shl eax 3)
-        (inst lea quo (make-ea :qword :index eax :scale 8)))
+        (inst shl eax n-fixnum-tag-bits)
+        (inst lea quo (make-ea :qword :index eax
+                               :scale (ash 1 n-fixnum-tag-bits))))
     (move rem edx)))
 
 (define-vop (fast-truncate-c/fixnum=>fixnum fast-safe-arith-op)
   (:translate truncate)
   (:args (x :scs (any-reg) :target eax))
   (:info y)
-  (:arg-types tagged-num (:constant (signed-byte 29)))
+  (:arg-types tagged-num (:constant fixnum))
   (:temporary (:sc signed-reg :offset eax-offset :target quo
                    :from :argument :to (:result 0)) eax)
   (:temporary (:sc any-reg :offset edx-offset :target rem
@@ -464,11 +533,14 @@
   (:generator 30
     (move eax x)
     (inst cqo)
-    (inst mov y-arg (fixnumize y))
+    (if (typep y '(signed-byte 29))
+        (inst mov y-arg (fixnumize y))
+        (setf y-arg (register-inline-constant :qword (fixnumize y))))
     (inst idiv eax y-arg)
     (if (location= quo eax)
-        (inst shl eax 3)
-        (inst lea quo (make-ea :qword :index eax :scale 8)))
+        (inst shl eax n-fixnum-tag-bits)
+        (inst lea quo (make-ea :qword :index eax
+                               :scale (ash 1 n-fixnum-tag-bits))))
     (move rem edx)))
 
 (define-vop (fast-truncate/unsigned=>unsigned fast-safe-arith-op)
@@ -487,7 +559,7 @@
   (: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))
@@ -502,7 +574,7 @@
   (:translate truncate)
   (:args (x :scs (unsigned-reg) :target eax))
   (:info y)
-  (:arg-types unsigned-num (:constant (unsigned-byte 31)))
+  (:arg-types unsigned-num (:constant (unsigned-byte 64)))
   (:temporary (:sc unsigned-reg :offset eax-offset :target quo
                    :from :argument :to (:result 0)) eax)
   (:temporary (:sc unsigned-reg :offset edx-offset :target rem
@@ -517,7 +589,9 @@
   (:generator 32
     (move eax x)
     (inst xor edx edx)
-    (inst mov y-arg y)
+    (if (typep y '(unsigned-byte 31))
+        (inst mov y-arg y)
+        (setf y-arg (register-inline-constant :qword y)))
     (inst div eax y-arg)
     (move quo eax)
     (move rem edx)))
@@ -538,7 +612,7 @@
   (: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))
@@ -553,7 +627,7 @@
   (:translate truncate)
   (:args (x :scs (signed-reg) :target eax))
   (:info y)
-  (:arg-types signed-num (:constant (signed-byte 32)))
+  (:arg-types signed-num (:constant (signed-byte 64)))
   (:temporary (:sc signed-reg :offset eax-offset :target quo
                    :from :argument :to (:result 0)) eax)
   (:temporary (:sc signed-reg :offset edx-offset :target rem
@@ -568,7 +642,9 @@
   (:generator 32
     (move eax x)
     (inst cqo)
-    (inst mov y-arg y)
+    (if (typep y '(signed-byte 32))
+        (inst mov y-arg y)
+        (setf y-arg (register-inline-constant :qword y)))
     (inst idiv eax y-arg)
     (move quo eax)
     (move rem edx)))
@@ -591,28 +667,36 @@
                                        (location= number result)))))
   (:result-types tagged-num)
   (:note "inline ASH")
+  (:variant nil)
+  (:variant-vars modularp)
   (:generator 2
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :qword :index number :scale 2)))
+           (inst lea result (make-ea :qword :base number :index number)))
           ((and (= amount 2) (not (location= number result)))
            (inst lea result (make-ea :qword :index number :scale 4)))
           ((and (= amount 3) (not (location= number result)))
            (inst lea result (make-ea :qword :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
-                  ;; Since the shift instructions take the shift amount
-                  ;; modulo 64 we must special case amounts of 64 and more.
-                  ;; Because fixnums have only 61 bits, the result is 0 or
-                  ;; -1 for all amounts of 60 or more, so use this as the
-                  ;; limit instead.
-                  (inst sar result (min (- n-word-bits n-fixnum-tag-bits 1)
-                                        (- amount)))
-                  (inst and result (lognot fixnum-tag-mask))))))))
+           (cond ((< -64 amount 64)
+                  ;; this code is used both in ASH and ASH-MODFX, so
+                  ;; be careful
+                  (if (plusp amount)
+                      (inst shl result amount)
+                      (progn
+                        (inst sar result (- amount))
+                        (inst and result (lognot fixnum-tag-mask)))))
+                 ;; shifting left (zero fill)
+                 ((plusp amount)
+                  (unless modularp
+                    (aver (not "Impossible: fixnum ASH should not be called with
+constant shift greater than word length")))
+                  (if (sc-is result any-reg)
+                      (zeroize result)
+                      (inst mov result 0)))
+                 ;; shifting right (sign fill)
+                 (t (inst sar result 63)
+                    (inst and result (lognot fixnum-tag-mask))))))))
 
 (define-vop (fast-ash-left/fixnum=>fixnum)
   (:translate ash)
@@ -653,7 +737,7 @@
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :qword :index number :scale 2)))
+           (inst lea result (make-ea :qword :base number :index number)))
           ((and (= amount 2) (not (location= number result)))
            (inst lea result (make-ea :qword :index number :scale 4)))
           ((and (= amount 3) (not (location= number result)))
@@ -680,7 +764,7 @@
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :qword :index number :scale 2)))
+           (inst lea result (make-ea :qword :base number :index number)))
           ((and (= amount 2) (not (location= number result)))
            (inst lea result (make-ea :qword :index number :scale 4)))
           ((and (= amount 3) (not (location= number result)))
@@ -694,7 +778,7 @@
                       (inst shl result amount)
                       (inst shr result (- amount))))
                  (t (if (sc-is result unsigned-reg)
-                        (inst xor result result)
+                        (zeroize result)
                         (inst mov result 0))))))))
 
 (define-vop (fast-ash-left/signed=>signed)
@@ -786,7 +870,7 @@
     (inst neg ecx)
     (inst cmp ecx 63)
     (inst jmp :be OKAY)
-    (inst xor result result)
+    (zeroize result)
     (inst jmp DONE)
     OKAY
     (inst shr result :cl)
@@ -896,7 +980,7 @@
     (inst or ecx ecx)
     (inst jmp :ns POSITIVE)
     (inst neg ecx)
-    (inst xor zero zero)
+    (zeroize zero)
     (inst shr result :cl)
     (inst cmp ecx 63)
     (inst cmov :nbe result zero)
@@ -918,7 +1002,9 @@
   (: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
@@ -927,7 +1013,7 @@
     (inst inc res)
     (inst jmp DONE)
     ZERO
-    (inst xor res res)
+    (zeroize res)
     DONE))
 
 (define-vop (unsigned-byte-64-len)
@@ -944,95 +1030,69 @@
     (inst inc res)
     (inst jmp DONE)
     ZERO
-    (inst xor res res)
+    (zeroize res)
     DONE))
 
-
 (define-vop (unsigned-byte-64-count)
   (:translate logcount)
   (:note "inline (unsigned-byte 64) 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)
-  (:temporary (:sc unsigned-reg :from (:argument 0)) t1)
-  (:generator 60
+  (:temporary (:sc unsigned-reg) temp)
+  (:temporary (:sc unsigned-reg) mask)
+  (: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", for 32-bit words. The extension to 64 bits is
+    ;; straightforward.
+    ;; 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 32 2-digit numbers at once.
     (move result arg)
-    (move t1 arg)
-
-    (inst mov temp result)
-    (inst shr temp 1)
-    (inst and result #x55555555)        ; note these masks will restrict the
-    (inst and temp #x55555555)          ; count to the lower half of arg
-    (inst add result temp)
-
-    (inst mov temp result)
+    (move temp arg)
+    (inst shr result 1)
+    (inst mov mask #x5555555555555555)
+    (inst and result mask)
+    (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)
-
-    (inst mov temp result)
-    (inst shr temp 4)
-    (inst and result #x0f0f0f0f)
-    (inst and temp #x0f0f0f0f)
+    (inst mov mask #x3333333333333333)
+    (inst and result mask)
+    (inst and temp mask)
     (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 8)
-    (inst and result #x00ff00ff)
-    (inst and temp #x00ff00ff)
+    (inst shr result 4)
     (inst add result temp)
-
-    (inst mov temp result)
-    (inst shr temp 16)
-    (inst and result #x0000ffff)
-    (inst and temp #x0000ffff)
-    (inst add result temp)
-
-    ;;; now do the upper half
-    (inst shr t1 32)
-
-    (inst mov temp t1)
-    (inst shr temp 1)
-    (inst and t1 #x55555555)
-    (inst and temp #x55555555)
-    (inst add t1 temp)
-
-    (inst mov temp t1)
-    (inst shr temp 2)
-    (inst and t1 #x33333333)
-    (inst and temp #x33333333)
-    (inst add t1 temp)
-
-    (inst mov temp t1)
-    (inst shr temp 4)
-    (inst and t1 #x0f0f0f0f)
-    (inst and temp #x0f0f0f0f)
-    (inst add t1 temp)
-
-    (inst mov temp t1)
-    (inst shr temp 8)
-    (inst and t1 #x00ff00ff)
-    (inst and temp #x00ff00ff)
-    (inst add t1 temp)
-
-    (inst mov temp t1)
-    (inst shr temp 16)
-    (inst and t1 #x0000ffff)
-    (inst and temp #x0000ffff)
-    (inst add t1 temp)
-    (inst add result t1)))
-
-
+    (inst mov mask #x0f0f0f0f0f0f0f0f)
+    (inst and result mask)
+    ;; Add all 8 bytes at once by multiplying with #256r11111111.
+    ;; We need to calculate only the lower 8 bytes of the product.
+    ;; Of these the most significant byte contains the final result.
+    ;; Note that there can be no overflow from one byte to the next
+    ;; as the sum is at most 64 which needs only 7 bits.
+    (inst mov mask #x0101010101010101)
+    (inst imul result mask)
+    (inst shr result 56)))
 
 ;;;; binary conditional VOPs
 
 (define-vop (fast-conditional)
-  (:conditional)
-  (:info target not-p)
+  (:conditional :e)
+  (:info)
   (:effects)
   (:affected)
   (:policy :fast-safe))
@@ -1049,9 +1109,11 @@
   (:note "inline fixnum comparison"))
 
 (define-vop (fast-conditional-c/fixnum fast-conditional/fixnum)
-  (:args (x :scs (any-reg control-stack)))
-  (:arg-types tagged-num (:constant (signed-byte 29)))
-  (:info target not-p y))
+  (:args (x :scs (any-reg)
+            :load-if (or (not (typep y '(signed-byte 29)))
+                         (not (sc-is x any-reg control-stack)))))
+  (:arg-types tagged-num (:constant fixnum))
+  (:info y))
 
 (define-vop (fast-conditional/signed fast-conditional)
   (:args (x :scs (signed-reg)
@@ -1062,9 +1124,11 @@
   (:note "inline (signed-byte 64) comparison"))
 
 (define-vop (fast-conditional-c/signed fast-conditional/signed)
-  (:args (x :scs (signed-reg signed-stack)))
-  (:arg-types signed-num (:constant (signed-byte 31)))
-  (:info target not-p y))
+  (:args (x :scs (signed-reg)
+            :load-if (or (not (typep y '(signed-byte 32)))
+                         (not (sc-is x signed-reg signed-stack)))))
+  (:arg-types signed-num (:constant (signed-byte 64)))
+  (:info y))
 
 (define-vop (fast-conditional/unsigned fast-conditional)
   (:args (x :scs (unsigned-reg)
@@ -1075,9 +1139,11 @@
   (:note "inline (unsigned-byte 64) comparison"))
 
 (define-vop (fast-conditional-c/unsigned fast-conditional/unsigned)
-  (:args (x :scs (unsigned-reg unsigned-stack)))
-  (:arg-types unsigned-num (:constant (unsigned-byte 31)))
-  (:info target not-p y))
+  (:args (x :scs (unsigned-reg)
+            :load-if (or (not (typep y '(unsigned-byte 31)))
+                         (not (sc-is x unsigned-reg unsigned-stack)))))
+  (:arg-types unsigned-num (:constant (unsigned-byte 64)))
+  (:info y))
 
 (macrolet ((define-conditional-vop (tran cond unsigned not-cond not-unsigned)
              `(progn
@@ -1091,19 +1157,26 @@
                                      (format nil "~:@(FAST-CONDITIONAL~A~)"
                                              suffix)))
                         (:translate ,tran)
+                        (:conditional ,(if signed cond unsigned))
                         (: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))))
+                                          ,(case suffix
+                                             (-c/fixnum
+                                                `(if (typep y '(signed-byte 29))
+                                                     (fixnumize y)
+                                                     (register-inline-constant
+                                                      :qword (fixnumize y))))
+                                             (-c/signed
+                                                `(if (typep y '(signed-byte 32))
+                                                     y
+                                                     (register-inline-constant
+                                                      :qword y)))
+                                             (-c/unsigned
+                                                `(if (typep y '(unsigned-byte 31))
+                                                     y
+                                                     (register-inline-constant
+                                                      :qword y)))
+                                             (t 'y))))))
                    '(/fixnum -c/fixnum /signed -c/signed /unsigned -c/unsigned)
 ;                  '(/fixnum  /signed  /unsigned)
                    '(4 3 6 5 6 5)
@@ -1115,32 +1188,32 @@
 (define-vop (fast-if-eql/signed fast-conditional/signed)
   (:translate eql)
   (:generator 6
-    (inst cmp x y)
-    (inst jmp (if not-p :ne :e) target)))
+    (inst cmp x y)))
 
 (define-vop (fast-if-eql-c/signed fast-conditional-c/signed)
   (:translate eql)
   (:generator 5
     (cond ((and (sc-is x signed-reg) (zerop y))
            (inst test x x))  ; smaller instruction
+          ((typep y '(signed-byte 32))
+           (inst cmp x y))
           (t
-           (inst cmp x y)))
-    (inst jmp (if not-p :ne :e) target)))
+           (inst cmp x (register-inline-constant :qword y))))))
 
 (define-vop (fast-if-eql/unsigned fast-conditional/unsigned)
   (:translate eql)
   (:generator 6
-    (inst cmp x y)
-    (inst jmp (if not-p :ne :e) target)))
+    (inst cmp x y)))
 
 (define-vop (fast-if-eql-c/unsigned fast-conditional-c/unsigned)
   (:translate eql)
   (:generator 5
     (cond ((and (sc-is x unsigned-reg) (zerop y))
            (inst test x x))  ; smaller instruction
+          ((typep y '(unsigned-byte 31))
+           (inst cmp x y))
           (t
-           (inst cmp x y)))
-    (inst jmp (if not-p :ne :e) target)))
+           (inst cmp x (register-inline-constant :qword y))))))
 
 ;;; EQL/FIXNUM is funny because the first arg can be of any type, not just a
 ;;; known fixnum.
@@ -1160,8 +1233,8 @@
   (:note "inline fixnum comparison")
   (:translate eql)
   (:generator 4
-    (inst cmp x y)
-    (inst jmp (if not-p :ne :e) target)))
+    (inst cmp x y)))
+
 (define-vop (generic-eql/fixnum fast-eql/fixnum)
   (:args (x :scs (any-reg descriptor-reg)
             :load-if (not (and (sc-is x control-stack)
@@ -1170,41 +1243,28 @@
   (:arg-types * tagged-num)
   (:variant-cost 7))
 
-
 (define-vop (fast-eql-c/fixnum fast-conditional/fixnum)
-  (:args (x :scs (any-reg control-stack)))
-  (:arg-types tagged-num (:constant (signed-byte 29)))
-  (:info target not-p y)
+  (:args (x :scs (any-reg)
+            :load-if (or (not (typep y '(signed-byte 29)))
+                         (not (sc-is x any-reg descriptor-reg control-stack)))))
+  (:arg-types tagged-num (:constant fixnum))
+  (:info y)
   (:translate eql)
   (:generator 2
-    (cond ((and (sc-is x any-reg) (zerop y))
+    (cond ((and (sc-is x any-reg descriptor-reg) (zerop y))
            (inst test x x))  ; smaller instruction
+          ((typep y '(signed-byte 29))
+           (inst cmp x (fixnumize y)))
           (t
-           (inst cmp x (fixnumize y))))
-    (inst jmp (if not-p :ne :e) target)))
+           (inst cmp x (register-inline-constant :qword (fixnumize y)))))))
 
 (define-vop (generic-eql-c/fixnum fast-eql-c/fixnum)
-  (:args (x :scs (any-reg descriptor-reg control-stack)))
-  (:arg-types * (:constant (signed-byte 29)))
+  (:args (x :scs (any-reg descriptor-reg)))
+  (:arg-types * (:constant fixnum))
   (:variant-cost 6))
 
 ;;;; 32-bit logical operations
 
-(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)))
-  (: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)))
-  (:result-types unsigned-num)
-  (:policy :fast-safe)
-  (:generator 4
-    (move ecx shift)
-    (move result prev)
-    (inst shrd result next :cl)))
-
 ;;; Only the lower 6 bits of the shift amount are significant.
 (define-vop (shift-towards-someplace)
   (:policy :fast-safe)
@@ -1233,32 +1293,74 @@
 
 ;;;; 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)
+                                    (typep y '(signed-byte 32))))))
+     (:info y)
+     (:arg-types untagged-num (:constant (or (unsigned-byte 64) (signed-byte 64))))
+     (: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 ((fun64 (intern (format nil "~S-MOD64" 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)))
-                   (vop64u (intern (format nil "FAST-~S-MOD64/UNSIGNED=>UNSIGNED" name)))
+                   (vop64u (intern (format nil "FAST-~S-MOD64/WORD=>UNSIGNED" name)))
                    (vop64f (intern (format nil "FAST-~S-MOD64/FIXNUM=>FIXNUM" name)))
-                   (vop64cu (intern (format nil "FAST-~S-MOD64-C/UNSIGNED=>UNSIGNED" name)))
+                   (vop64cu (intern (format nil "FAST-~S-MOD64-C/WORD=>UNSIGNED" name)))
                    (vop64cf (intern (format nil "FAST-~S-MOD64-C/FIXNUM=>FIXNUM" name)))
-                   (sfun61 (intern (format nil "~S-SMOD61" name)))
-                   (svop61f (intern (format nil "FAST-~S-SMOD61/FIXNUM=>FIXNUM" name)))
-                   (svop61cf (intern (format nil "FAST-~S-SMOD61-C/FIXNUM=>FIXNUM" name))))
+                   (funfx (intern (format nil "~S-MODFX" name)))
+                   (vopfxf (intern (format nil "FAST-~S-MODFX/FIXNUM=>FIXNUM" name)))
+                   (vopfxcf (intern (format nil "FAST-~S-MODFX-C/FIXNUM=>FIXNUM" name))))
                `(progn
-                  (define-modular-fun ,fun64 (x y) ,name :unsigned 64)
-                  (define-modular-fun ,sfun61 (x y) ,name :signed 61)
-                  (define-vop (,vop64u ,vopu) (:translate ,fun64))
+                  (define-modular-fun ,fun64 (x y) ,name :untagged nil 64)
+                  (define-modular-fun ,funfx (x y) ,name :tagged t
+                                      #.(- n-word-bits n-fixnum-tag-bits))
+                  (define-mod-binop (,vop64u ,vopu) ,fun64)
                   (define-vop (,vop64f ,vopf) (:translate ,fun64))
-                  (define-vop (,svop61f ,vopf) (:translate ,sfun61))
+                  (define-vop (,vopfxf ,vopf) (:translate ,funfx))
                   ,@(when -c-p
-                      `((define-vop (,vop64cu ,vopcu) (:translate ,fun64))
-                        (define-vop (,svop61cf ,vopcf) (:translate ,sfun61))))))))
+                      `((define-mod-binop-c (,vop64cu ,vopcu) ,fun64)
+                        (define-vop (,vopfxcf ,vopcf) (:translate ,funfx))))))))
   (def + t)
   (def - t)
-  ;; (no -C variant as x86 MUL instruction doesn't take an immediate)
-  (def * nil))
+  (def * t))
 
 (define-vop (fast-ash-left-mod64-c/unsigned=>unsigned
              fast-ash-c/unsigned=>unsigned)
@@ -1271,47 +1373,49 @@
     (sb!c::give-up-ir1-transform))
   '(%primitive fast-ash-left-mod64/unsigned=>unsigned integer count))
 
-(define-vop (fast-ash-left-smod61-c/fixnum=>fixnum
+(define-vop (fast-ash-left-modfx-c/fixnum=>fixnum
              fast-ash-c/fixnum=>fixnum)
-  (:translate ash-left-smod61))
-(define-vop (fast-ash-left-smod61/fixnum=>fixnum
+  (:variant :modular)
+  (:translate ash-left-modfx))
+(define-vop (fast-ash-left-modfx/fixnum=>fixnum
              fast-ash-left/fixnum=>fixnum))
-(deftransform ash-left-smod61 ((integer count)
-                               ((signed-byte 61) (unsigned-byte 6)))
+(deftransform ash-left-modfx ((integer count)
+                              (fixnum (unsigned-byte 6)))
   (when (sb!c::constant-lvar-p count)
     (sb!c::give-up-ir1-transform))
-  '(%primitive fast-ash-left-smod61/fixnum=>fixnum integer count))
+  '(%primitive fast-ash-left-modfx/fixnum=>fixnum integer count))
 
 (in-package "SB!C")
 
 (defknown sb!vm::%lea-mod64 (integer integer (member 1 2 4 8) (signed-byte 64))
   (unsigned-byte 64)
   (foldable flushable movable))
-(defknown sb!vm::%lea-smod61 (integer integer (member 1 2 4 8) (signed-byte 64))
-  (signed-byte 61)
+(defknown sb!vm::%lea-modfx (integer integer (member 1 2 4 8) (signed-byte 64))
+  fixnum
   (foldable flushable movable))
 
-(define-modular-fun-optimizer %lea ((base index scale disp) :unsigned :width width)
+(define-modular-fun-optimizer %lea ((base index scale disp) :untagged nil :width width)
   (when (and (<= width 64)
              (constant-lvar-p scale)
              (constant-lvar-p disp))
-    (cut-to-width base :unsigned width)
-    (cut-to-width index :unsigned width)
+    (cut-to-width base :untagged width nil)
+    (cut-to-width index :untagged width nil)
     'sb!vm::%lea-mod64))
-(define-modular-fun-optimizer %lea ((base index scale disp) :signed :width width)
-  (when (and (<= width 61)
+(define-modular-fun-optimizer %lea ((base index scale disp) :tagged t :width width)
+  (when (and (<= width (- sb!vm:n-word-bits sb!vm:n-fixnum-tag-bits))
              (constant-lvar-p scale)
              (constant-lvar-p disp))
-    (cut-to-width base :signed width)
-    (cut-to-width index :signed width)
-    'sb!vm::%lea-smod61))
+    (cut-to-width base :tagged width t)
+    (cut-to-width index :tagged width t)
+    'sb!vm::%lea-modfx))
 
 #+sb-xc-host
 (progn
   (defun sb!vm::%lea-mod64 (base index scale disp)
     (ldb (byte 64 0) (%lea base index scale disp)))
-  (defun sb!vm::%lea-smod61 (base index scale disp)
-    (mask-signed-field 61 (%lea base index scale disp))))
+  (defun sb!vm::%lea-modfx (base index scale disp)
+    (mask-signed-field (- sb!vm:n-word-bits sb!vm:n-fixnum-tag-bits)
+                       (%lea base index scale disp))))
 #-sb-xc-host
 (progn
   (defun sb!vm::%lea-mod64 (base index scale disp)
@@ -1320,24 +1424,25 @@
       ;; can't use modular version of %LEA, as we only have VOPs for
       ;; constant SCALE and DISP.
       (ldb (byte 64 0) (+ base (* index scale) disp))))
-  (defun sb!vm::%lea-smod61 (base index scale disp)
-    (let ((base (mask-signed-field 61 base))
-          (index (mask-signed-field 61 index)))
+  (defun sb!vm::%lea-modfx (base index scale disp)
+    (let* ((fixnum-width (- sb!vm:n-word-bits sb!vm:n-fixnum-tag-bits))
+           (base (mask-signed-field fixnum-width base))
+           (index (mask-signed-field fixnum-width index)))
       ;; can't use modular version of %LEA, as we only have VOPs for
       ;; constant SCALE and DISP.
-      (mask-signed-field 61 (+ base (* index scale) disp)))))
+      (mask-signed-field fixnum-width (+ base (* index scale) disp)))))
 
 (in-package "SB!VM")
 
 (define-vop (%lea-mod64/unsigned=>unsigned
              %lea/unsigned=>unsigned)
   (:translate %lea-mod64))
-(define-vop (%lea-smod61/fixnum=>fixnum
+(define-vop (%lea-modfx/fixnum=>fixnum
              %lea/fixnum=>fixnum)
-  (:translate %lea-smod61))
+  (:translate %lea-modfx))
 
 ;;; logical operations
-(define-modular-fun lognot-mod64 (x) lognot :unsigned 64)
+(define-modular-fun lognot-mod64 (x) lognot :untagged nil 64)
 (define-vop (lognot-mod64/unsigned=>unsigned)
   (:translate lognot-mod64)
   (:args (x :scs (unsigned-reg unsigned-stack) :target r
@@ -1355,20 +1460,6 @@
     (move r x)
     (inst not r)))
 
-(define-modular-fun logxor-mod64 (x y) logxor :unsigned 64)
-(define-vop (fast-logxor-mod64/unsigned=>unsigned
-             fast-logxor/unsigned=>unsigned)
-  (:translate logxor-mod64))
-(define-vop (fast-logxor-mod64-c/unsigned=>unsigned
-             fast-logxor-c/unsigned=>unsigned)
-  (:translate logxor-mod64))
-(define-vop (fast-logxor-mod64/fixnum=>fixnum
-             fast-logxor/fixnum=>fixnum)
-  (:translate logxor-mod64))
-(define-vop (fast-logxor-mod64-c/fixnum=>fixnum
-             fast-logxor-c/fixnum=>fixnum)
-  (:translate logxor-mod64))
-
 (define-source-transform logeqv (&rest args)
   (if (oddp (length args))
       `(logxor ,@args)
@@ -1398,7 +1489,9 @@
 
 (define-full-reffer bignum-ref * bignum-digits-offset other-pointer-lowtag
   (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)
 
@@ -1407,16 +1500,14 @@
   (:policy :fast-safe)
   (:args (digit :scs (unsigned-reg)))
   (:arg-types unsigned-num)
-  (:conditional)
-  (:info target not-p)
+  (:conditional :ns)
   (:generator 3
-    (inst or digit digit)
-    (inst jmp (if not-p :s :ns) target)))
+    (inst or digit digit)))
 
 
 ;;; 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
-;;; 4. This is easy to deal with and may save a fixnum-word
+;;; that it may be passed as a fixnum or word and thus may be 0, 1, or
+;;; 8. This is easy to deal with and may save a fixnum-word
 ;;; conversion.
 (define-vop (add-w/carry)
   (:translate sb!bignum:%add-with-carry)
@@ -1437,8 +1528,8 @@
     (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 x86-64 convention.
 (define-vop (sub-w/borrow)
   (:translate sb!bignum:%subtract-with-borrow)
   (:policy :fast-safe)
@@ -1453,9 +1544,8 @@
     (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)
@@ -1525,6 +1615,42 @@
     (move hi edx)
     (move lo eax)))
 
+#!+multiply-high-vops
+(define-vop (mulhi)
+  (:translate sb!kernel:%multiply-high)
+  (:policy :fast-safe)
+  (:args (x :scs (unsigned-reg) :target eax)
+         (y :scs (unsigned-reg unsigned-stack)))
+  (:arg-types unsigned-num unsigned-num)
+  (:temporary (:sc unsigned-reg :offset eax-offset :from (:argument 0))
+              eax)
+  (:temporary (:sc unsigned-reg :offset edx-offset :from (:argument 1)
+                   :to (:result 0) :target hi) edx)
+  (:results (hi :scs (unsigned-reg)))
+  (:result-types unsigned-num)
+  (:generator 20
+    (move eax x)
+    (inst mul eax y)
+    (move hi edx)))
+
+#!+multiply-high-vops
+(define-vop (mulhi/fx)
+  (:translate sb!kernel:%multiply-high)
+  (:policy :fast-safe)
+  (:args (x :scs (any-reg) :target eax)
+         (y :scs (unsigned-reg unsigned-stack)))
+  (:arg-types positive-fixnum unsigned-num)
+  (:temporary (:sc any-reg :offset eax-offset :from (:argument 0)) eax)
+  (:temporary (:sc any-reg :offset edx-offset :from (:argument 1)
+                   :to (:result 0) :target hi) edx)
+  (:results (hi :scs (any-reg)))
+  (:result-types positive-fixnum)
+  (:generator 15
+    (move eax x)
+    (inst mul eax y)
+    (move hi edx)
+    (inst and hi (lognot fixnum-tag-mask))))
+
 (define-vop (bignum-lognot lognot-mod64/unsigned=>unsigned)
   (:translate sb!bignum:%lognot))
 
@@ -1540,10 +1666,10 @@
   (:result-types unsigned-num)
   (:generator 1
     (move digit fixnum)
-    (inst sar digit 3)))
+    (inst sar digit n-fixnum-tag-bits)))
 
 (define-vop (bignum-floor)
-  (:translate sb!bignum:%floor)
+  (:translate sb!bignum:%bigfloor)
   (:policy :fast-safe)
   (:args (div-high :scs (unsigned-reg) :target edx)
          (div-low :scs (unsigned-reg) :target eax)
@@ -1576,7 +1702,7 @@
   (:generator 1
     (move res digit)
     (when (sc-is res any-reg control-stack)
-      (inst shl res 3))))
+      (inst shl res n-fixnum-tag-bits))))
 
 (define-vop (digit-ashr)
   (:translate sb!bignum:%ashr)
@@ -1589,11 +1715,25 @@
                     :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 63)))
+  (: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)
   (:generator 1
@@ -1651,14 +1791,14 @@
     (*-transformer y)))
 
 (deftransform * ((x y)
-                 ((signed-byte 61) (constant-arg (unsigned-byte 64)))
-                 (signed-byte 61))
+                 (fixnum (constant-arg (unsigned-byte 64)))
+                 fixnum)
   "recode as leas, shifts and adds"
   (let ((y (lvar-value y)))
     (*-transformer y)))
-(deftransform sb!vm::*-smod61
-    ((x y) ((signed-byte 61) (constant-arg (unsigned-byte 64)))
-     (signed-byte 61))
+(deftransform sb!vm::*-modfx
+    ((x y) (fixnum (constant-arg (unsigned-byte 64)))
+     fixnum)
   "recode as leas, shifts and adds"
   (let ((y (lvar-value y)))
     (*-transformer y)))