;;;; implementation-dependent transforms ;;;; This software is part of the SBCL system. See the README file for ;;;; more information. ;;;; ;;;; This software is derived from the CMU CL system, which was ;;;; written at Carnegie Mellon University and released into the ;;;; public domain. The software is in the public domain and is ;;;; provided with absolutely no warranty. See the COPYING and CREDITS ;;;; files for more information. (in-package "SB!C") ;;; We need to define these predicates, since the TYPEP source ;;; transform picks whichever predicate was defined last when there ;;; are multiple predicates for equivalent types. (define-source-transform short-float-p (x) `(single-float-p ,x)) #!-long-float (define-source-transform long-float-p (x) `(double-float-p ,x)) (define-source-transform compiled-function-p (x) `(functionp ,x)) (define-source-transform char-int (x) `(char-code ,x)) (deftransform abs ((x) (rational)) '(if (< x 0) (- x) x)) ;;; The layout is stored in slot 0. (define-source-transform %instance-layout (x) `(truly-the layout (%instance-ref ,x 0))) (define-source-transform %set-instance-layout (x val) `(%instance-set ,x 0 (the layout ,val))) (define-source-transform %funcallable-instance-layout (x) `(truly-the layout (%funcallable-instance-info ,x 0))) (define-source-transform %set-funcallable-instance-layout (x val) `(setf (%funcallable-instance-info ,x 0) (the layout ,val))) ;;;; character support ;;; In our implementation there are really only BASE-CHARs. #+nil (define-source-transform characterp (obj) `(base-char-p ,obj)) ;;;; simplifying HAIRY-DATA-VECTOR-REF and HAIRY-DATA-VECTOR-SET (deftransform hairy-data-vector-ref ((string index) (simple-string t)) (let ((ctype (lvar-type string))) (if (array-type-p ctype) ;; the other transform will kick in, so that's OK (give-up-ir1-transform) `(etypecase string ((simple-array character (*)) (data-vector-ref string index)) #!+sb-unicode ((simple-array base-char (*)) (data-vector-ref string index)) ((simple-array nil (*)) (data-vector-ref string index)))))) (deftransform hairy-data-vector-ref ((array index) (array t) *) "avoid runtime dispatch on array element type" (let ((element-ctype (extract-upgraded-element-type array)) (declared-element-ctype (extract-declared-element-type array))) (declare (type ctype element-ctype)) (when (eq *wild-type* element-ctype) (give-up-ir1-transform "Upgraded element type of array is not known at compile time.")) ;; (The expansion here is basically a degenerate case of ;; WITH-ARRAY-DATA. Since WITH-ARRAY-DATA is implemented as a ;; macro, and macros aren't expanded in transform output, we have ;; to hand-expand it ourselves.) (let ((element-type-specifier (type-specifier element-ctype))) `(multiple-value-bind (array index) (%data-vector-and-index array index) (declare (type (simple-array ,element-type-specifier 1) array)) ,(let ((bare-form '(data-vector-ref array index))) (if (type= element-ctype declared-element-ctype) bare-form `(the ,(type-specifier declared-element-ctype) ,bare-form))))))) ;;; Transform multi-dimensional array to one dimensional data vector ;;; access. (deftransform data-vector-ref ((array index) (simple-array t)) (let ((array-type (lvar-type array))) (unless (array-type-p array-type) (give-up-ir1-transform)) (let ((dims (array-type-dimensions array-type))) (when (or (atom dims) (= (length dims) 1)) (give-up-ir1-transform)) (let ((el-type (array-type-specialized-element-type array-type)) (total-size (if (member '* dims) '* (reduce #'* dims)))) `(data-vector-ref (truly-the (simple-array ,(type-specifier el-type) (,total-size)) (%array-data-vector array)) index))))) (deftransform hairy-data-vector-set ((string index new-value) (simple-string t t)) (let ((ctype (lvar-type string))) (if (array-type-p ctype) ;; the other transform will kick in, so that's OK (give-up-ir1-transform) `(etypecase string ((simple-array character (*)) (data-vector-set string index new-value)) #!+sb-unicode ((simple-array base-char (*)) (data-vector-set string index new-value)) ((simple-array nil (*)) (data-vector-set string index new-value)))))) (deftransform hairy-data-vector-set ((array index new-value) (array t t) *) "avoid runtime dispatch on array element type" (let ((element-ctype (extract-upgraded-element-type array)) (declared-element-ctype (extract-declared-element-type array))) (declare (type ctype element-ctype)) (when (eq *wild-type* element-ctype) (give-up-ir1-transform "Upgraded element type of array is not known at compile time.")) (let ((element-type-specifier (type-specifier element-ctype))) `(multiple-value-bind (array index) (%data-vector-and-index array index) (declare (type (simple-array ,element-type-specifier 1) array) (type ,element-type-specifier new-value)) ,(if (type= element-ctype declared-element-ctype) '(data-vector-set array index new-value) `(truly-the ,(type-specifier declared-element-ctype) (data-vector-set array index (the ,(type-specifier declared-element-ctype) new-value)))))))) (deftransform data-vector-set ((array index new-value) (simple-array t t)) (let ((array-type (lvar-type array))) (unless (array-type-p array-type) (give-up-ir1-transform)) (let ((dims (array-type-dimensions array-type))) (when (or (atom dims) (= (length dims) 1)) (give-up-ir1-transform)) (let ((el-type (array-type-specialized-element-type array-type)) (total-size (if (member '* dims) '* (reduce #'* dims)))) `(data-vector-set (truly-the (simple-array ,(type-specifier el-type) (,total-size)) (%array-data-vector array)) index new-value))))) (defoptimizer (%data-vector-and-index derive-type) ((array index)) (let ((atype (lvar-type array))) (when (array-type-p atype) (values-specifier-type `(values (simple-array ,(type-specifier (array-type-specialized-element-type atype)) (*)) index))))) (deftransform %data-vector-and-index ((%array %index) (simple-array t) *) ;; KLUDGE: why the percent signs? Well, ARRAY and INDEX are ;; respectively exported from the CL and SB!INT packages, which ;; means that they're visible to all sorts of things. If the ;; compiler can prove that the call to ARRAY-HEADER-P, below, either ;; returns T or NIL, it will delete the irrelevant branch. However, ;; user code might have got here with a variable named CL:ARRAY, and ;; quite often compiler code with a variable named SB!INT:INDEX, so ;; this can generate code deletion notes for innocuous user code: ;; (DEFUN F (ARRAY I) (DECLARE (SIMPLE-VECTOR ARRAY)) (AREF ARRAY I)) ;; -- CSR, 2003-04-01 ;; We do this solely for the -OR-GIVE-UP side effect, since we want ;; to know that the type can be figured out in the end before we ;; proceed, but we don't care yet what the type will turn out to be. (upgraded-element-type-specifier-or-give-up %array) '(if (array-header-p %array) (values (%array-data-vector %array) %index) (values %array %index))) ;;; transforms for getting at simple arrays of (UNSIGNED-BYTE N) when (< N 8) ;;; ;;; FIXME: In CMU CL, these were commented out with #+NIL. Why? Should ;;; we fix them or should we delete them? (Perhaps these definitions ;;; predate the various DATA-VECTOR-REF-FOO VOPs which have ;;; (:TRANSLATE DATA-VECTOR-REF), and are redundant now?) #+nil (macrolet ((frob (type bits) (let ((elements-per-word (truncate sb!vm:n-word-bits bits))) `(progn (deftransform data-vector-ref ((vector index) (,type *)) `(multiple-value-bind (word bit) (floor index ,',elements-per-word) (ldb ,(ecase sb!vm:target-byte-order (:little-endian '(byte ,bits (* bit ,bits))) (:big-endian '(byte ,bits (- sb!vm:n-word-bits (* (1+ bit) ,bits))))) (%raw-bits vector (+ word sb!vm:vector-data-offset))))) (deftransform data-vector-set ((vector index new-value) (,type * *)) `(multiple-value-bind (word bit) (floor index ,',elements-per-word) (setf (ldb ,(ecase sb!vm:target-byte-order (:little-endian '(byte ,bits (* bit ,bits))) (:big-endian '(byte ,bits (- sb!vm:n-word-bits (* (1+ bit) ,bits))))) (%raw-bits vector (+ word sb!vm:vector-data-offset))) new-value))))))) (frob simple-bit-vector 1) (frob (simple-array (unsigned-byte 2) (*)) 2) (frob (simple-array (unsigned-byte 4) (*)) 4)) ;;;; BIT-VECTOR hackery ;;; SIMPLE-BIT-VECTOR bit-array operations are transformed to a word ;;; loop that does 32 bits at a time. ;;; ;;; FIXME: This is a lot of repeatedly macroexpanded code. It should ;;; be a function call instead. (macrolet ((def (bitfun wordfun) `(deftransform ,bitfun ((bit-array-1 bit-array-2 result-bit-array) (simple-bit-vector simple-bit-vector simple-bit-vector) * :node node :policy (>= speed space)) `(progn ,@(unless (policy node (zerop safety)) '((unless (= (length bit-array-1) (length bit-array-2) (length result-bit-array)) (error "Argument and/or result bit arrays are not the same length:~ ~% ~S~% ~S ~% ~S" bit-array-1 bit-array-2 result-bit-array)))) (let ((length (length result-bit-array))) (if (= length 0) ;; We avoid doing anything to 0-length ;; bit-vectors, or rather, the memory that ;; follows them. Other divisible-by-32 cases ;; are handled by the (1- length), below. ;; CSR, 2002-04-24 result-bit-array (do ((index sb!vm:vector-data-offset (1+ index)) (end-1 (+ sb!vm:vector-data-offset ;; bit-vectors of length 1-32 ;; need precisely one (SETF ;; %RAW-BITS), done here in the ;; epilogue. - CSR, 2002-04-24 (truncate (truly-the index (1- length)) sb!vm:n-word-bits)))) ((= index end-1) (setf (%raw-bits result-bit-array index) (,',wordfun (%raw-bits bit-array-1 index) (%raw-bits bit-array-2 index))) result-bit-array) (declare (optimize (speed 3) (safety 0)) (type index index end-1)) (setf (%raw-bits result-bit-array index) (,',wordfun (%raw-bits bit-array-1 index) (%raw-bits bit-array-2 index)))))))))) (def bit-and word-logical-and) (def bit-ior word-logical-or) (def bit-xor word-logical-xor) (def bit-eqv word-logical-eqv) (def bit-nand word-logical-nand) (def bit-nor word-logical-nor) (def bit-andc1 word-logical-andc1) (def bit-andc2 word-logical-andc2) (def bit-orc1 word-logical-orc1) (def bit-orc2 word-logical-orc2)) (deftransform bit-not ((bit-array result-bit-array) (simple-bit-vector simple-bit-vector) * :node node :policy (>= speed space)) `(progn ,@(unless (policy node (zerop safety)) '((unless (= (length bit-array) (length result-bit-array)) (error "Argument and result bit arrays are not the same length:~ ~% ~S~% ~S" bit-array result-bit-array)))) (let ((length (length result-bit-array))) (if (= length 0) ;; We avoid doing anything to 0-length bit-vectors, or rather, ;; the memory that follows them. Other divisible-by ;; n-word-bits cases are handled by the (1- length), below. ;; CSR, 2002-04-24 result-bit-array (do ((index sb!vm:vector-data-offset (1+ index)) (end-1 (+ sb!vm:vector-data-offset ;; bit-vectors of length 1 to n-word-bits need ;; precisely one (SETF %RAW-BITS), done here in ;; the epilogue. - CSR, 2002-04-24 (truncate (truly-the index (1- length)) sb!vm:n-word-bits)))) ((= index end-1) (setf (%raw-bits result-bit-array index) (word-logical-not (%raw-bits bit-array index))) result-bit-array) (declare (optimize (speed 3) (safety 0)) (type index index end-1)) (setf (%raw-bits result-bit-array index) (word-logical-not (%raw-bits bit-array index)))))))) (deftransform bit-vector-= ((x y) (simple-bit-vector simple-bit-vector)) `(and (= (length x) (length y)) (let ((length (length x))) (or (= length 0) (do* ((i sb!vm:vector-data-offset (+ i 1)) (end-1 (+ sb!vm:vector-data-offset (floor (1- length) sb!vm:n-word-bits)))) ((= i end-1) (let* ((extra (1+ (mod (1- length) sb!vm:n-word-bits))) (mask (ash #.(1- (ash 1 sb!vm:n-word-bits)) (- extra sb!vm:n-word-bits))) (numx (logand (ash mask ,(ecase sb!c:*backend-byte-order* (:little-endian 0) (:big-endian '(- sb!vm:n-word-bits extra)))) (%raw-bits x i))) (numy (logand (ash mask ,(ecase sb!c:*backend-byte-order* (:little-endian 0) (:big-endian '(- sb!vm:n-word-bits extra)))) (%raw-bits y i)))) (declare (type (integer 1 #.sb!vm:n-word-bits) extra) (type sb!vm:word mask numx numy)) (= numx numy))) (declare (type index i end-1)) (let ((numx (%raw-bits x i)) (numy (%raw-bits y i))) (declare (type sb!vm:word numx numy)) (unless (= numx numy) (return nil)))))))) (deftransform count ((item sequence) (bit simple-bit-vector) * :policy (>= speed space)) `(let ((length (length sequence))) (if (zerop length) 0 (do ((index sb!vm:vector-data-offset (1+ index)) (count 0) (end-1 (+ sb!vm:vector-data-offset (truncate (truly-the index (1- length)) sb!vm:n-word-bits)))) ((= index end-1) (let* ((extra (1+ (mod (1- length) sb!vm:n-word-bits))) (mask (ash #.(1- (ash 1 sb!vm:n-word-bits)) (- extra sb!vm:n-word-bits))) (bits (logand (ash mask ,(ecase sb!c:*backend-byte-order* (:little-endian 0) (:big-endian '(- sb!vm:n-word-bits extra)))) (%raw-bits sequence index)))) (declare (type (integer 1 #.sb!vm:n-word-bits) extra)) (declare (type sb!vm:word mask bits)) ;; could consider LOGNOT for the zero case instead of ;; doing the subtraction... (incf count ,(if (constant-lvar-p item) (if (zerop (lvar-value item)) '(- extra (logcount bits)) '(logcount bits)) '(if (zerop item) (- extra (logcount bits)) (logcount bits)))))) (declare (type index index count end-1) (optimize (speed 3) (safety 0))) (incf count ,(if (constant-lvar-p item) (if (zerop (lvar-value item)) '(- sb!vm:n-word-bits (logcount (%raw-bits sequence index))) '(logcount (%raw-bits sequence index))) '(if (zerop item) (- sb!vm:n-word-bits (logcount (%raw-bits sequence index))) (logcount (%raw-bits sequence index))))))))) (deftransform fill ((sequence item) (simple-bit-vector bit) * :policy (>= speed space)) (let ((value (if (constant-lvar-p item) (if (= (lvar-value item) 0) 0 #.(1- (ash 1 sb!vm:n-word-bits))) `(if (= item 0) 0 #.(1- (ash 1 sb!vm:n-word-bits)))))) `(let ((length (length sequence)) (value ,value)) (if (= length 0) sequence (do ((index sb!vm:vector-data-offset (1+ index)) (end-1 (+ sb!vm:vector-data-offset ;; bit-vectors of length 1 to n-word-bits need ;; precisely one (SETF %RAW-BITS), done here ;; in the epilogue. - CSR, 2002-04-24 (truncate (truly-the index (1- length)) sb!vm:n-word-bits)))) ((= index end-1) (setf (%raw-bits sequence index) value) sequence) (declare (optimize (speed 3) (safety 0)) (type index index end-1)) (setf (%raw-bits sequence index) value)))))) (deftransform fill ((sequence item) (simple-base-string base-char) * :policy (>= speed space)) (let ((value (if (constant-lvar-p item) (let* ((char (lvar-value item)) (code (sb!xc:char-code char)) (accum 0)) (dotimes (i sb!vm:n-word-bytes accum) (setf accum (logior accum (ash code (* 8 i)))))) `(let ((code (sb!xc:char-code item))) (logior ,@(loop for i from 0 below sb!vm:n-word-bytes collect `(ash code ,(* 8 i)))))))) `(let ((length (length sequence)) (value ,value)) (multiple-value-bind (times rem) (truncate length sb!vm:n-word-bytes) (do ((index sb!vm:vector-data-offset (1+ index)) (end (+ times sb!vm:vector-data-offset))) ((= index end) (let ((place (* times sb!vm:n-word-bytes))) (declare (fixnum place)) (dotimes (j rem sequence) (declare (index j)) (setf (schar sequence (the index (+ place j))) item)))) (declare (optimize (speed 3) (safety 0)) (type index index)) (setf (%raw-bits sequence index) value)))))) ;;;; %BYTE-BLT ;;; FIXME: The old CMU CL code used various COPY-TO/FROM-SYSTEM-AREA ;;; stuff (with all the associated bit-index cruft and overflow ;;; issues) even for byte moves. In SBCL, we're converting to byte ;;; moves as problems are discovered with the old code, and this is ;;; currently (ca. sbcl-0.6.12.30) the main interface for code in ;;; SB!KERNEL and SB!SYS (e.g. i/o code). It's not clear that it's the ;;; ideal interface, though, and it probably deserves some thought. (deftransform %byte-blt ((src src-start dst dst-start dst-end) ((or (simple-unboxed-array (*)) system-area-pointer) index (or (simple-unboxed-array (*)) system-area-pointer) index index)) ;; FIXME: CMU CL had a hairier implementation of this (back when it ;; was still called (%PRIMITIVE BYTE-BLT). It had the small problem ;; that it didn't work for large (>16M) values of SRC-START or ;; DST-START. However, it might have been more efficient. In ;; particular, I don't really know how much the foreign function ;; call costs us here. My guess is that if the overhead is ;; acceptable for SQRT and COS, it's acceptable here, but this ;; should probably be checked. -- WHN '(flet ((sapify (thing) (etypecase thing (system-area-pointer thing) ;; FIXME: The code here rather relies on the simple ;; unboxed array here having byte-sized entries. That ;; should be asserted explicitly, I just haven't found ;; a concise way of doing it. (It would be nice to ;; declare it in the DEFKNOWN too.) ((simple-unboxed-array (*)) (vector-sap thing))))) (declare (inline sapify)) (without-gcing (memmove (sap+ (sapify dst) dst-start) (sap+ (sapify src) src-start) (- dst-end dst-start))) (values))) ;;;; transforms for EQL of floating point values (deftransform eql ((x y) (single-float single-float)) '(= (single-float-bits x) (single-float-bits y))) (deftransform eql ((x y) (double-float double-float)) '(and (= (double-float-low-bits x) (double-float-low-bits y)) (= (double-float-high-bits x) (double-float-high-bits y)))) ;;;; modular functions (define-good-modular-fun logand :unsigned) (define-good-modular-fun logior :unsigned) ;;; FIXME: XOR? ANDC1, ANDC2? -- CSR, 2003-09-16 (macrolet ((def (name class width) (let ((type (ecase class (:unsigned 'unsigned-byte) (:signed 'signed-byte)))) `(progn (defknown ,name (integer (integer 0)) (,type ,width) (foldable flushable movable)) (define-modular-fun-optimizer ash ((integer count) ,class :width width) (when (and (<= width ,width) (or (and (constant-lvar-p count) (plusp (lvar-value count))) (csubtypep (lvar-type count) (specifier-type '(and unsigned-byte fixnum))))) (cut-to-width integer ,class width) ',name)) (setf (gethash ',name (modular-class-versions (find-modular-class ',class))) `(ash ,',width)))))) ;; This should really be dependent on SB!VM:N-WORD-BITS, but since we ;; don't have a true Alpha64 port yet, we'll have to stick to ;; SB!VM:N-MACHINE-WORD-BITS for the time being. --njf, 2004-08-14 #!+#.(cl:if (cl:= 32 sb!vm:n-machine-word-bits) '(and) '(or)) (progn #!+x86 (def sb!vm::ash-left-smod30 :signed 30) (def sb!vm::ash-left-mod32 :unsigned 32)) #!+#.(cl:if (cl:= 64 sb!vm:n-machine-word-bits) '(and) '(or)) (progn #!+x86-64 (def sb!vm::ash-left-smod61 :signed 61) (def sb!vm::ash-left-mod64 :unsigned 64))) ;;;; word-wise logical operations ;;; These transforms assume the presence of modular arithmetic to ;;; generate efficient code. (define-source-transform word-logical-not (x) `(logand (lognot (the sb!vm:word ,x)) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-and ((x y)) '(logand x y)) (deftransform word-logical-nand ((x y)) '(logand (lognand x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-or ((x y)) '(logior x y)) (deftransform word-logical-nor ((x y)) '(logand (lognor x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-xor ((x y)) '(logxor x y)) (deftransform word-logical-eqv ((x y)) '(logand (logeqv x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-orc1 ((x y)) '(logand (logorc1 x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-orc2 ((x y)) '(logand (logorc2 x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-andc1 ((x y)) '(logand (logandc1 x y) #.(1- (ash 1 sb!vm:n-word-bits)))) (deftransform word-logical-andc2 ((x y)) '(logand (logandc2 x y) #.(1- (ash 1 sb!vm:n-word-bits)))) ;;; There are two different ways the multiplier can be recoded. The ;;; more obvious is to shift X by the correct amount for each bit set ;;; in Y and to sum the results. But if there is a string of bits that ;;; are all set, you can add X shifted by one more then the bit ;;; position of the first set bit and subtract X shifted by the bit ;;; position of the last set bit. We can't use this second method when ;;; the high order bit is bit 31 because shifting by 32 doesn't work ;;; too well. (defun ub32-strength-reduce-constant-multiply (arg num) (declare (type (unsigned-byte 32) num)) (let ((adds 0) (shifts 0) (result nil) first-one) (labels ((add (next-factor) (setf result (if result (progn (incf adds) `(+ ,result ,next-factor)) next-factor)))) (declare (inline add)) (dotimes (bitpos 32) (if first-one (when (not (logbitp bitpos num)) (add (if (= (1+ first-one) bitpos) ;; There is only a single bit in the string. (progn (incf shifts) `(ash ,arg ,first-one)) ;; There are at least two. (progn (incf adds) (incf shifts 2) `(- (ash ,arg ,bitpos) (ash ,arg ,first-one))))) (setf first-one nil)) (when (logbitp bitpos num) (setf first-one bitpos)))) (when first-one (cond ((= first-one 31)) ((= first-one 30) (incf shifts) (add `(ash ,arg 30))) (t (incf shifts 2) (incf adds) (add `(- (ash ,arg 31) (ash ,arg ,first-one))))) (incf shifts) (add `(ash ,arg 31)))) (values (if (plusp adds) `(logand ,result #.(1- (ash 1 32))) ; using modular arithmetic result) adds shifts)))