X-Git-Url: http://repo.macrolet.net/gitweb/?a=blobdiff_plain;f=src%2Fcompiler%2Fsparc%2Farith.lisp;h=2669bf6d607bdbcbce46287148748b9625a2a487;hb=1b650be8b800cf96e2c268ae317fb26d0bf36827;hp=515f49ed3bafe94040c600e6e64ec371e6e8f079;hpb=68fd2d2dd6f265669a8957accd8a33e62786a97e;p=sbcl.git diff --git a/src/compiler/sparc/arith.lisp b/src/compiler/sparc/arith.lisp index 515f49e..2669bf6 100644 --- a/src/compiler/sparc/arith.lisp +++ b/src/compiler/sparc/arith.lisp @@ -1,4 +1,4 @@ -;;;; the VM definition arithmetic VOPs for the Alpha +;;;; the VM definition arithmetic VOPs for the SPARC ;;;; This software is part of the SBCL system. See the README file for ;;;; more information. @@ -159,21 +159,20 @@ (define-vop (fast-logand/signed-unsigned=>unsigned fast-logand/unsigned=>unsigned) - (:args (x :target r :scs (signed-reg)) - (y :scs (unsigned-reg unsigned-stack))) + (:args (x :scs (signed-reg)) + (y :target r :scs (unsigned-reg))) (:arg-types signed-num unsigned-num)) (define-vop (fast-logand/unsigned-signed=>unsigned fast-logand/unsigned=>unsigned) (:args (x :target r :scs (unsigned-reg)) - (y :scs (signed-reg signed-stack))) + (y :scs (signed-reg))) (:arg-types unsigned-num signed-num)) ;;; Special case fixnum + and - that trap on overflow. Useful when we ;;; don't know that the output type is a fixnum. -;;; I (toy@rtp.ericsson.se) took these out. They don't seem to be -;;; used anywhere at all. +;;; I (Raymond Toy) took these out. They don't seem to be used anywhere at all. #+nil (progn (define-vop (+/fixnum fast-+/fixnum=>fixnum) @@ -228,8 +227,9 @@ (:temporary (:scs (signed-reg)) y-int) (:vop-var vop) (:save-p :compute-only) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 12 (let ((zero (generate-error-code vop division-by-zero-error x y))) (inst cmp y zero-tn) @@ -238,7 +238,7 @@ (inst sra r x 31) (inst wry r) ;; Remove tag bits so Q and R will be tagged correctly. - (inst sra y-int y fixnum-tag-bits) + (inst sra y-int y n-fixnum-tag-bits) (inst nop) (inst nop) @@ -262,14 +262,17 @@ (:temporary (:scs (signed-reg)) r) (:vop-var vop) (:save-p :compute-only) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 12 (let ((zero (generate-error-code vop division-by-zero-error x y))) (inst cmp y zero-tn) - (inst b :eq zero #!+:sparc-v9 :pn) + (if (member :sparc-v9 *backend-subfeatures*) + (inst b :eq zero :pn) + (inst b :eq zero)) ;; Extend the sign of X into the Y register - (inst sra r x 31) + (inst sra r x 31) (inst wry r) (inst nop) (inst nop) @@ -295,13 +298,16 @@ (:temporary (:scs (unsigned-reg)) r) (:vop-var vop) (:save-p :compute-only) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 8 (let ((zero (generate-error-code vop division-by-zero-error x y))) (inst cmp y zero-tn) - (inst b :eq zero #!+:sparc-v9 :pn) - (inst wry zero-tn) ; Clear out high part + (if (member :sparc-v9 *backend-subfeatures*) + (inst b :eq zero :pn) + (inst b :eq zero)) + (inst wry zero-tn) ; Clear out high part (inst nop) (inst nop) (inst nop) @@ -313,7 +319,6 @@ (unless (location= quo q) (inst move quo q))))) -#!+:sparc-v9 (define-vop (fast-v9-truncate/signed=>signed fast-safe-arith-op) (:translate truncate) (:args (x :scs (signed-reg)) @@ -327,13 +332,13 @@ (:temporary (:scs (signed-reg)) r) (:vop-var vop) (:save-p :compute-only) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 8 (let ((zero (generate-error-code vop division-by-zero-error x y))) (inst cmp y zero-tn) - (inst b :eq zero #!+:sparc-v9 :pn) + (inst b :eq zero :pn) ;; Sign extend the numbers, just in case. - (inst sra x 0) + (inst sra x 0) (inst sra y 0) (inst sdivx q x y) ;; Compute remainder @@ -355,13 +360,13 @@ (:temporary (:scs (unsigned-reg)) r) (:vop-var vop) (:save-p :compute-only) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 8 (let ((zero (generate-error-code vop division-by-zero-error x y))) (inst cmp y zero-tn) - (inst b :eq zero #!+:sparc-v9 :pn) + (inst b :eq zero :pn) ;; Zap the higher 32 bits, just in case - (inst srl x 0) + (inst srl x 0) (inst srl y 0) (inst udivx q x y) ;; Compute remainder @@ -372,64 +377,71 @@ ;;; Shifting -(macrolet - ((frob (name sc-type type shift-right-inst) - `(define-vop (,name) - (:note "inline ASH") - (:args (number :scs (,sc-type) :to :save) - (amount :scs (signed-reg immediate))) - (:arg-types ,type signed-num) - (:results (result :scs (,sc-type))) - (:result-types ,type) - (:translate ash) - (:policy :fast-safe) - (:temporary (:sc non-descriptor-reg) ndesc) - (:generator 5 - (sc-case amount - #!+:sparc-v9 - (signed-reg - (let ((done (gen-label)) - (positive (gen-label))) - (inst cmp amount) - (inst b :ge positive) - (inst neg ndesc amount) - ;; ndesc = max(-amount, 31) - (inst cmp ndesc 31) - (inst cmove :ge ndesc 31) - (inst b done) - (inst ,shift-right-inst result number ndesc) - (emit-label positive) - ;; The result-type assures us that this shift will not - ;; overflow. - (inst sll result number amount) - ;; We want a right shift of the appropriate size. - (emit-label done))) - #!-:sparc-v9 - (signed-reg - (let ((positive (gen-label)) - (done (gen-label))) - (inst cmp amount) - (inst b :ge positive) - (inst neg ndesc amount) - (inst cmp ndesc 31) - (inst b :le done) - (inst ,shift-right-inst result number ndesc) - (inst b done) - (inst ,shift-right-inst result number 31) - - (emit-label positive) - ;; The result-type assures us that this shift will not overflow. - (inst sll result number amount) - - (emit-label done))) - (immediate - (let ((amount (tn-value amount))) - (if (minusp amount) - (let ((amount (min 31 (- amount)))) - (inst ,shift-right-inst result number amount)) - (inst sll result number amount))))))))) - (frob fast-ash/signed=>signed signed-reg signed-num sra) - (frob fast-ash/unsigned=>unsigned unsigned-reg unsigned-num srl)) +(define-vop (fast-ash/signed=>signed) + (:note "inline ASH") + (:args (number :scs (signed-reg) :to :save) + (amount :scs (signed-reg immediate) :to :save)) + (:arg-types signed-num signed-num) + (:results (result :scs (signed-reg))) + (:result-types signed-num) + (:translate ash) + (:policy :fast-safe) + (:temporary (:sc non-descriptor-reg) ndesc) + (:generator 5 + (sc-case amount + (signed-reg + (let ((done (gen-label))) + (inst cmp amount) + (inst b :ge done) + ;; The result-type assures us that this shift will not + ;; overflow. + (inst sll result number amount) + (inst neg ndesc amount) + (inst cmp ndesc 31) + (if (member :sparc-v9 *backend-subfeatures*) + (progn + (inst cmove :ge ndesc 31) + (inst sra result number ndesc)) + (progn + (inst b :le done) + (inst sra result number ndesc) + (inst sra result number 31))) + (emit-label done))) + (immediate + (bug "IMMEDIATE case in ASH VOP; should have been transformed"))))) + +(define-vop (fast-ash/unsigned=>unsigned) + (:note "inline ASH") + (:args (number :scs (unsigned-reg) :to :save) + (amount :scs (signed-reg immediate) :to :save)) + (:arg-types unsigned-num signed-num) + (:results (result :scs (unsigned-reg))) + (:result-types unsigned-num) + (:translate ash) + (:policy :fast-safe) + (:temporary (:sc non-descriptor-reg) ndesc) + (:generator 5 + (sc-case amount + (signed-reg + (let ((done (gen-label))) + (inst cmp amount) + (inst b :ge done) + ;; The result-type assures us that this shift will not + ;; overflow. + (inst sll result number amount) + (inst neg ndesc amount) + (inst cmp ndesc 32) + (if (member :sparc-v9 *backend-subfeatures*) + (progn + (inst srl result number ndesc) + (inst cmove :ge result zero-tn)) + (progn + (inst b :lt done) + (inst srl result number ndesc) + (move result zero-tn))) + (emit-label done))) + (immediate + (bug "IMMEDIATE case in ASH VOP; should have been transformed"))))) ;; Some special cases where we know we want a left shift. Just do the ;; shift, instead of checking for the sign of the shift. @@ -446,7 +458,7 @@ (:policy :fast-safe) (:generator ,cost ;; The result-type assures us that this shift will not - ;; overflow. And for fixnum's, the zero bits that get + ;; overflow. And for fixnums, the zero bits that get ;; shifted in are just fine for the fixnum tag. (sc-case amount ((signed-reg unsigned-reg) @@ -575,49 +587,83 @@ (define-vop (fast-v8-*/fixnum=>fixnum fast-fixnum-binop) (:temporary (:scs (non-descriptor-reg)) temp) (:translate *) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 2 ;; The cost here should be less than the cost for ;; */signed=>signed. Why? A fixnum product using signed=>signed ;; has to convert both args to signed-nums. But using this, we ;; don't have to and that saves an instruction. - (inst sra temp y fixnum-tag-bits) + (inst sra temp y n-fixnum-tag-bits) (inst smul r x temp))) +(define-vop (fast-v8-*-c/fixnum=>fixnum fast-safe-arith-op) + (:args (x :target r :scs (any-reg zero))) + (:info y) + (:arg-types tagged-num + (:constant (and (signed-byte 13) (not (integer 0 0))))) + (:results (r :scs (any-reg))) + (:result-types tagged-num) + (:note "inline fixnum arithmetic") + (:translate *) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) + (:generator 1 + (inst smul r x y))) + (define-vop (fast-v8-*/signed=>signed fast-signed-binop) (:translate *) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 3 (inst smul r x y))) +(define-vop (fast-v8-*-c/signed=>signed fast-signed-binop-c) + (:translate *) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) + (:generator 2 + (inst smul r x y))) + (define-vop (fast-v8-*/unsigned=>unsigned fast-unsigned-binop) (:translate *) - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) (:generator 3 (inst umul r x y))) +(define-vop (fast-v8-*-c/unsigned=>unsigned fast-unsigned-binop-c) + (:translate *) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (and (member :sparc-v9 *backend-subfeatures*) + (not (member :sparc-64 *backend-subfeatures*))))) + (:generator 2 + (inst umul r x y))) + ;; The smul and umul instructions are deprecated on the Sparc V9. Use ;; mulx instead. (define-vop (fast-v9-*/fixnum=>fixnum fast-fixnum-binop) (:temporary (:scs (non-descriptor-reg)) temp) (:translate *) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 4 - (inst sra temp y fixnum-tag-bits) + (inst sra temp y n-fixnum-tag-bits) (inst mulx r x temp))) (define-vop (fast-v9-*/signed=>signed fast-signed-binop) (:translate *) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 3 (inst mulx r x y))) (define-vop (fast-v9-*/unsigned=>unsigned fast-unsigned-binop) (:translate *) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 3 (inst mulx r x y))) @@ -631,14 +677,6 @@ (:affected) (:policy :fast-safe)) -(deftype integer-with-a-bite-out (s bite) - (cond ((eq s '*) 'integer) - ((and (integerp s) (> s 1)) - (let ((bound (ash 1 (1- s)))) - `(integer ,(- bound) ,(- bound bite 1)))) - (t - (error "Bad size specified for SIGNED-BYTE type specifier: ~S." s)))) - (define-vop (fast-conditional/fixnum fast-conditional) (:args (x :scs (any-reg zero)) (y :scs (any-reg zero))) @@ -880,7 +918,7 @@ (:args (digit :scs (unsigned-reg))) (:arg-types unsigned-num) (:results (result :scs (descriptor-reg))) - (:guard #!-:sparc-v9 t #!+:sparc-v9 nil) + (:guard (not (member :sparc-v9 *backend-subfeatures*))) (:generator 3 (let ((done (gen-label))) (inst cmp digit) @@ -895,7 +933,7 @@ (:args (digit :scs (unsigned-reg))) (:arg-types unsigned-num) (:results (result :scs (descriptor-reg))) - (:guard #!+:sparc-v9 t #!-:sparc-v9 nil) + (:guard (member :sparc-v9 *backend-subfeatures*)) (:generator 3 (inst cmp digit) (load-symbol result t) @@ -959,46 +997,46 @@ (type (or tn (signed-byte 13)) multiplicand)) ;; It seems that emit-multiply is only used to do an unsigned ;; multiply, so the code only does an unsigned multiply. - #!+:sparc-64 - (progn - ;; Take advantage of V9's 64-bit multiplier. - ;; - ;; Make sure the multiplier and multiplicand are really - ;; unsigned 64-bit numbers. - (inst srl multiplier 0) - (inst srl multiplicand 0) + (cond + ((member :sparc-64 *backend-subfeatures*) + ;; Take advantage of V9's 64-bit multiplier. + ;; + ;; Make sure the multiplier and multiplicand are really + ;; unsigned 64-bit numbers. + (inst srl multiplier 0) + (inst srl multiplicand 0) - ;; Multiply the two numbers and put the result in - ;; result-high. Copy the low 32-bits to result-low. Then - ;; shift result-high so the high 32-bits end up in the low - ;; 32-bits. - (inst mulx result-high multiplier multiplicand) - (inst move result-low result-high) - (inst srax result-high 32)) - #!+(and (not :sparc-64) (or :sparc-v8 :sparc-v9)) - (progn - ;; V8 has a multiply instruction. This should also work for - ;; the V9, but umul and the Y register is deprecated on the - ;; V9. - (inst umul result-low multiplier multiplicand) - (inst rdy result-high)) - #!+(and (not :sparc-64) (not (or :sparc-v8 :sparc-v9))) - (let ((label (gen-label))) - (inst wry multiplier) - (inst andcc result-high zero-tn) - ;; Note: we can't use the Y register until three insts - ;; after it's written. - (inst nop) - (inst nop) - (dotimes (i 32) - (inst mulscc result-high multiplicand)) - (inst mulscc result-high zero-tn) - (inst cmp multiplicand) - (inst b :ge label) - (inst nop) - (inst add result-high multiplier) - (emit-label label) - (inst rdy result-low))) + ;; Multiply the two numbers and put the result in + ;; result-high. Copy the low 32-bits to result-low. Then + ;; shift result-high so the high 32-bits end up in the low + ;; 32-bits. + (inst mulx result-high multiplier multiplicand) + (inst move result-low result-high) + (inst srax result-high 32)) + ((or (member :sparc-v8 *backend-subfeatures*) + (member :sparc-v9 *backend-subfeatures*)) + ;; V8 has a multiply instruction. This should also work for + ;; the V9, but umul and the Y register is deprecated on the + ;; V9. + (inst umul result-low multiplier multiplicand) + (inst rdy result-high)) + (t + (let ((label (gen-label))) + (inst wry multiplier) + (inst andcc result-high zero-tn) + ;; Note: we can't use the Y register until three insts + ;; after it's written. + (inst nop) + (inst nop) + (dotimes (i 32) + (inst mulscc result-high multiplicand)) + (inst mulscc result-high zero-tn) + (inst cmp multiplicand) + (inst b :ge label) + (inst nop) + (inst add result-high multiplier) + (emit-label label) + (inst rdy result-low))))) (define-vop (bignum-mult-and-add-3-arg) (:translate sb!bignum::%multiply-and-add) @@ -1063,7 +1101,7 @@ (:results (digit :scs (unsigned-reg))) (:result-types unsigned-num) (:generator 1 - (inst sra digit fixnum fixnum-tag-bits))) + (inst sra digit fixnum n-fixnum-tag-bits))) (define-vop (bignum-floor) (:translate sb!bignum::%floor) @@ -1075,8 +1113,6 @@ (:results (quo :scs (unsigned-reg) :from (:argument 1)) (rem :scs (unsigned-reg) :from (:argument 0))) (:result-types unsigned-num unsigned-num) - (:guard #!+(not (or :sparc-v8 :sparc-v9)) t - #!-(not (or :sparc-v8 :sparc-v9)) nil) (:generator 300 (move rem div-high) (move quo div-low) @@ -1104,8 +1140,8 @@ (:temporary (:scs (unsigned-reg) :target quo) q) ;; This vop is for a v8 or v9, provided we're also not using ;; sparc-64, for which there a special sparc-64 vop. - (:guard #!+(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) t - #!-(or :sparc-v8 (and :sparc-v9 (not :sparc-64))) nil) + (:guard (or (member :sparc-v8 *backend-subfeatures*) + (member :sparc-v9 *backend-subfeatures*))) (:generator 15 (inst wry div-high) (inst nop) @@ -1131,7 +1167,7 @@ (:results (quo :scs (unsigned-reg)) (rem :scs (unsigned-reg))) (:result-types unsigned-num unsigned-num) - (:guard #!+:sparc-64 t #!-:sparc-64 nil) + (:guard (member :sparc-64 *backend-subfeatures*)) (:generator 5 ;; Set dividend to be div-high and div-low (inst sllx dividend div-high 32) @@ -1152,7 +1188,7 @@ (:generator 1 (sc-case res (any-reg - (inst sll res digit fixnum-tag-bits)) + (inst sll res digit n-fixnum-tag-bits)) (signed-reg (move res digit))))) @@ -1205,21 +1241,39 @@ ;; Need these so constant folding works with the deftransform. -(defun ash-right-signed (num shift) - (declare (type (signed-byte #.sb!vm:n-word-bits) num) - (type (integer 0 #.(1- sb!vm:n-word-bits)) shift)) - (ash-right-signed num shift)) +;; FIXME KLUDGE ew yuk. +#-sb-xc-host +(progn + (defun ash-right-signed (num shift) + (ash-right-signed num shift)) + + (defun ash-right-unsigned (num shuft) + (ash-right-unsigned num shift))) + +(in-package "SB!C") -(defun ash-right-unsigned (num shift) - (declare (type (unsigned-byte #.sb!vm:n-word-bits) num) - (type (integer 0 #.(1- sb!vm:n-word-bits)) shift)) - (ash-right-unsigned num shift)) +(deftransform * ((x y) + ((unsigned-byte 32) (constant-arg (unsigned-byte 32))) + (unsigned-byte 32)) + "recode as shifts and adds" + (let ((y (continuation-value y))) + (multiple-value-bind (result adds shifts) + (ub32-strength-reduce-constant-multiply 'x y) + (cond + ;; we assume, perhaps foolishly, that good SPARCs don't have an + ;; issue with multiplications. (Remember that there's a + ;; different transform for converting x*2^k to a shift). + ((member :sparc-64 *backend-subfeatures*) (give-up-ir1-transform)) + ((or (member :sparc-v9 *backend-subfeatures*) + (member :sparc-v8 *backend-subfeatures*)) + ;; breakeven point as measured by Raymond Toy + (when (> (+ adds shifts) 9) + (give-up-ir1-transform)))) + (or result 0)))) ;; If we can prove that we have a right shift, just do the right shift ;; instead of calling the inline ASH which has to check for the ;; direction of the shift at run-time. -(in-package "SB!C") - (deftransform ash ((num shift) (integer integer)) (let ((num-type (continuation-type num)) (shift-type (continuation-type shift)))