X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fx86%2Farith.lisp;h=4731c2653c1db8ee441442369ee788633bf806ae;hb=a6b91f356da1b5ae2987f79db9bd137970512959;hp=b19fcb3470dcad72fbac68fc3470e4912f9eb6d7;hpb=1ef1aac7257895837ff23dfcc05c0fceac43e659;p=sbcl.git

diff --git a/src/compiler/x86/arith.lisp b/src/compiler/x86/arith.lisp
index b19fcb3..4731c26 100644
--- a/src/compiler/x86/arith.lisp
+++ b/src/compiler/x86/arith.lisp
@@ -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)))
 
@@ -436,7 +436,7 @@
   (: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))
@@ -491,7 +491,7 @@
   (: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))
@@ -542,7 +542,7 @@
   (: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))
@@ -597,24 +597,27 @@
   (:note "inline ASH")
   (:generator 2
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :dword :index number :scale 2)))
+           (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)
-                  ;; 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 (lognot fixnum-tag-mask))))))))
+           (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)
@@ -655,7 +658,7 @@
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :dword :index number :scale 2)))
+           (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)))
@@ -682,7 +685,7 @@
   (:note "inline ASH")
   (:generator 3
     (cond ((and (= amount 1) (not (location= number result)))
-           (inst lea result (make-ea :dword :index number :scale 2)))
+           (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)))
@@ -920,7 +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
@@ -953,43 +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
 
@@ -1050,15 +1065,17 @@
                                      ,(symbolicate "FAST-CONDITIONAL" suffix))
                          (:translate logtest)
                          (:generator ,cost
-                          (inst test x ,(if (eq suffix '-c/fixnum)
-                                            '(fixnumize y)
-                                            'y))
+                          (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))
+  (movable foldable flushable always-translatable))
 
+;;; only for constant folding within the compiler
 (defun %logbitp (integer index)
   (logbitp index integer))
 
@@ -1305,8 +1322,8 @@
                    (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 :unsigned 32)
-                  (define-modular-fun ,sfun30 (x y) ,name :signed 30)
+                  (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))
@@ -1352,19 +1369,19 @@
   (signed-byte 30)
   (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 32)
              (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-mod32))
-(define-modular-fun-optimizer %lea ((base index scale disp) :signed :width width)
+(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 :signed width)
-    (cut-to-width index :signed width)
+    (cut-to-width base :tagged width t)
+    (cut-to-width index :tagged width t)
     'sb!vm::%lea-smod30))
 
 #+sb-xc-host
@@ -1398,7 +1415,7 @@
   (:translate %lea-smod30))
 
 ;;; logical operations
-(define-modular-fun lognot-mod32 (x) lognot :unsigned 32)
+(define-modular-fun lognot-mod32 (x) lognot :untagged nil 32)
 (define-vop (lognot-mod32/word=>unsigned)
   (:translate lognot-mod32)
   (:args (x :scs (unsigned-reg signed-reg unsigned-stack signed-stack) :target r
@@ -1421,20 +1438,6 @@
     (move r x)
     (inst not r)))
 
-(define-modular-fun logxor-mod32 (x y) logxor :unsigned 32)
-(define-mod-binop (fast-logxor-mod32/word=>unsigned
-                   fast-logxor/unsigned=>unsigned)
-    logxor-mod32)
-(define-mod-binop-c (fast-logxor-mod32-c/word=>unsigned
-                     fast-logxor-c/unsigned=>unsigned)
-    logxor-mod32)
-(define-vop (fast-logxor-mod32/fixnum=>fixnum
-             fast-logxor/fixnum=>fixnum)
-  (:translate logxor-mod32))
-(define-vop (fast-logxor-mod32-c/fixnum=>fixnum
-             fast-logxor-c/fixnum=>fixnum)
-  (:translate logxor-mod32))
-
 (define-source-transform logeqv (&rest args)
   (if (oddp (length args))
       `(logxor ,@args)
@@ -1464,7 +1467,9 @@
 
 (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)
 
@@ -1654,11 +1659,25 @@
                     :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 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)
   (:generator 1
@@ -1710,10 +1729,7 @@
   (: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.
@@ -1722,25 +1738,15 @@
     (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