From mboxrd@z Thu Jan 1 00:00:00 1970 Path: news.gmane.org!not-for-mail From: Daniel Colascione Newsgroups: gmane.emacs.devel Subject: Set operations on bool-vectors Date: Fri, 20 Sep 2013 15:59:47 -0700 Message-ID: <523CD363.6020400@dancol.org> NNTP-Posting-Host: plane.gmane.org Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="6AsIIl4JhB7AE8Ksqq1g1P57LDgGhnHBf" X-Trace: ger.gmane.org 1379718080 14273 80.91.229.3 (20 Sep 2013 23:01:20 GMT) X-Complaints-To: usenet@ger.gmane.org NNTP-Posting-Date: Fri, 20 Sep 2013 23:01:20 +0000 (UTC) To: Emacs development discussions Original-X-From: emacs-devel-bounces+ged-emacs-devel=m.gmane.org@gnu.org Sat Sep 21 01:01:23 2013 Return-path: Envelope-to: ged-emacs-devel@m.gmane.org Original-Received: from lists.gnu.org ([208.118.235.17]) by plane.gmane.org with esmtp (Exim 4.69) (envelope-from ) id 1VN9hS-00078M-UB for ged-emacs-devel@m.gmane.org; Sat, 21 Sep 2013 01:01:23 +0200 Original-Received: from localhost ([::1]:58091 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VN9hS-000518-5c for ged-emacs-devel@m.gmane.org; Fri, 20 Sep 2013 19:01:22 -0400 Original-Received: from eggs.gnu.org ([2001:4830:134:3::10]:60792) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VN9hG-00050x-J5 for emacs-devel@gnu.org; Fri, 20 Sep 2013 19:01:19 -0400 Original-Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1VN9h8-00017I-1m for emacs-devel@gnu.org; Fri, 20 Sep 2013 19:01:10 -0400 Original-Received: from dancol.org ([2600:3c01::f03c:91ff:fedf:adf3]:44419) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1VN9h7-00013d-Lv for emacs-devel@gnu.org; Fri, 20 Sep 2013 19:01:01 -0400 Original-Received: from c-76-22-66-162.hsd1.wa.comcast.net ([76.22.66.162] helo=[192.168.1.52]) by dancol.org with esmtpsa (TLS1.0:DHE_RSA_CAMELLIA_256_CBC_SHA1:256) (Exim 4.80) (envelope-from ) id 1VN9h0-0002bO-TY for emacs-devel@gnu.org; Fri, 20 Sep 2013 16:00:55 -0700 User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:17.0) Gecko/20130801 Thunderbird/17.0.8 X-Enigmail-Version: 1.5.2 X-detected-operating-system: by eggs.gnu.org: Error: Malformed IPv6 address (bad octet value). X-Received-From: 2600:3c01::f03c:91ff:fedf:adf3 X-BeenThere: emacs-devel@gnu.org X-Mailman-Version: 2.1.14 Precedence: list List-Id: "Emacs development discussions." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: emacs-devel-bounces+ged-emacs-devel=m.gmane.org@gnu.org Original-Sender: emacs-devel-bounces+ged-emacs-devel=m.gmane.org@gnu.org Xref: news.gmane.org gmane.emacs.devel:163522 Archived-At: This is an OpenPGP/MIME signed message (RFC 4880 and 3156) --6AsIIl4JhB7AE8Ksqq1g1P57LDgGhnHBf Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable I've implemented built-in set operations on bool vectors. Many interesting algorithms boil down to applying set-valued functions to graphs until the graph no longer changes. The interface below is designed with this use in mind. In particular, these functions simultaneously compute the destination set and whether the destination set is different from its input. Operations take a destination bool vector and allocate memory only if explicitly asked to do so. These operations also allow callers to efficiently enumerate all the nil or t bits in a set. According to my benchmarks, these operations are about 1,000 times faster than the compiled elisp equivalents. =3D=3D=3D modified file 'src/alloc.c' --- src/alloc.c 2013-09-04 22:34:04 +0000 +++ src/alloc.c 2013-09-19 22:24:27 +0000 @@ -2003,6 +2003,29 @@ return val; } +static +ptrdiff_t +bool_vector_payload_bytes (EMACS_INT length) +{ + EMACS_INT exact_needed_bytes; + EMACS_INT needed_bytes; + + exact_needed_bytes =3D (length + CHAR_BIT - 1) / CHAR_BIT; + needed_bytes =3D (exact_needed_bytes + sizeof (size_t) - 1) + / sizeof (size_t); + + needed_bytes *=3D sizeof (size_t); + + if (needed_bytes =3D=3D 0) + { + /* Always allocate at least one machine word of payload so that + bool-vector operations in data.c don't need a special case + for empty vectors. */ + needed_bytes =3D sizeof (size_t); + } + + return needed_bytes; +} DEFUN ("make-bool-vector", Fmake_bool_vector, Smake_bool_vector, 2, 2, 0= , doc: /* Return a new bool-vector of length LENGTH, using INIT for= each element. @@ -2011,37 +2034,39 @@ { register Lisp_Object val; struct Lisp_Bool_Vector *p; - ptrdiff_t length_in_chars; - EMACS_INT length_in_elts; - int bits_per_value; - int extra_bool_elts =3D ((bool_header_size - header_size + word_size -= 1) - / word_size); + EMACS_INT exact_payload_bytes; + EMACS_INT total_payload_bytes; + EMACS_INT needed_elements; CHECK_NATNUM (length); - bits_per_value =3D sizeof (EMACS_INT) * BOOL_VECTOR_BITS_PER_CHAR; - - length_in_elts =3D (XFASTINT (length) + bits_per_value - 1) / bits_per= _value; - - val =3D Fmake_vector (make_number (length_in_elts + extra_bool_elts), = Qnil); - - /* No Lisp_Object to trace in there. */ + exact_payload_bytes =3D (XFASTINT (length) + CHAR_BIT - 1) / CHAR_BIT;= + total_payload_bytes =3D bool_vector_payload_bytes (XFASTINT (length));= + + eassert (exact_payload_bytes <=3D total_payload_bytes); + + needed_elements =3D ((bool_header_size - header_size + total_payload_b= ytes) + + word_size - 1) / word_size; + + p =3D (struct Lisp_Bool_Vector* ) allocate_vector(needed_elements); + XSETVECTOR (val, p); XSETPVECTYPESIZE (XVECTOR (val), PVEC_BOOL_VECTOR, 0, 0); - p =3D XBOOL_VECTOR (val); p->size =3D XFASTINT (length); - - length_in_chars =3D ((XFASTINT (length) + BOOL_VECTOR_BITS_PER_CHAR - = 1) - / BOOL_VECTOR_BITS_PER_CHAR); - if (length_in_chars) + if (exact_payload_bytes) { - memset (p->data, ! NILP (init) ? -1 : 0, length_in_chars); + memset (p->data, ! NILP (init) ? -1 : 0, exact_payload_bytes); /* Clear any extraneous bits in the last byte. */ - p->data[length_in_chars - 1] + p->data[exact_payload_bytes - 1] &=3D (1 << ((XFASTINT (length) - 1) % BOOL_VECTOR_BITS_PER_CHAR + 1)) -= 1; } + /* Clear padding at the end. */ + memset (p->data + exact_payload_bytes, + 0, + total_payload_bytes - exact_payload_bytes); + return val; } @@ -2785,10 +2810,10 @@ if (size & PSEUDOVECTOR_FLAG) { if (PSEUDOVECTOR_TYPEP (&v->header, PVEC_BOOL_VECTOR)) - size =3D (bool_header_size - + (((struct Lisp_Bool_Vector *) v)->size - + BOOL_VECTOR_BITS_PER_CHAR - 1) - / BOOL_VECTOR_BITS_PER_CHAR); + { + struct Lisp_Bool_Vector *bv =3D (struct Lisp_Bool_Vector *) v;= + size =3D bool_header_size + bool_vector_payload_bytes (bv->siz= e); + } else size =3D (header_size + ((size & PSEUDOVECTOR_SIZE_MASK) @@ -2895,10 +2920,11 @@ pseudovector type grows beyond VBLOCK_BYTES_MAX. */ eassert (PSEUDOVECTOR_TYPEP (&vector->header, PVEC_BOOL_VECTOR));= - total_vector_slots - +=3D (bool_header_size - + ((b->size + BOOL_VECTOR_BITS_PER_CHAR - 1) - / BOOL_VECTOR_BITS_PER_CHAR)) / word_size; + total_vector_slots +=3D (bool_header_size + + bool_vector_payload_bytes (b->siz= e) + + word_size + - 1) + / word_size; } else total_vector_slots =3D=3D=3D modified file 'src/data.c' --- src/data.c 2013-09-11 07:20:20 +0000 +++ src/data.c 2013-09-20 22:31:51 +0000 @@ -54,6 +54,7 @@ static Lisp_Object Qnatnump; Lisp_Object Qstringp, Qarrayp, Qsequencep, Qbufferp; Lisp_Object Qchar_or_string_p, Qmarkerp, Qinteger_or_marker_p, Qvectorp;= +Lisp_Object Qbool_vector_p; Lisp_Object Qbuffer_or_string_p; static Lisp_Object Qkeywordp, Qboundp; Lisp_Object Qfboundp; @@ -2956,6 +2957,384 @@ return make_number (order); } +/* Because we round up the BOOL_VECTOR allocate size to word_size + units, we can safely read past the "end" of the vector in the + operations below. These extra bits are always zero. Also, we + always BOOL_VECTORS with at least one size_t of storage so that we + don't have to special-case empty bit vectors. */ + +#if (SIZE_MAX >> 32) & 1 +# define BITS_PER_SIZE_T 64 +#else +# define BITS_PER_SIZE_T 32 +#endif + +static inline +size_t +bool_vector_spare_mask (EMACS_INT nr_bits) +{ + return (((size_t) 1) << (nr_bits % BITS_PER_SIZE_T)) - 1; +} + +static inline +EMACS_INT +popcount_size_t(size_t val) +{ + EMACS_INT count; + +#if defined __GNUC__ && BITS_PER_SIZE_T =3D=3D 64 + count =3D __builtin_popcountll (val); +#elif defined __GNUC__ && BITS_PER_SIZE_T =3D=3D 32 + count =3D __builtin_popcount (val); +#elif defined __MSC_VER && BITS_PER_SIZE_T =3D=3D 64 +# pragma intrinsic __popcnt64 + count =3D __popcnt64 (val); +#elif defined __MSC_VER && BITS_PER_SIZE_T =3D=3D 32 +# pragma intrinsic __popcnt + count =3D __popcnt (val); +#else + { + EMACS_INT j; + count =3D 0; + for (j =3D 0; j < BITS_PER_SIZE_T; ++j) + count +=3D !!((((size_t) 1) << j) & val); + } +#endif + + return count; +} + +enum bool_vector_op { bool_vector_exclusive_or, + bool_vector_union, + bool_vector_intersection, + bool_vector_set_difference, + bool_vector_subsetp }; + +static inline +Lisp_Object +bool_vector_binop_driver (Lisp_Object a, + Lisp_Object b, + Lisp_Object c, + enum bool_vector_op op) +{ + EMACS_INT nr_bits; + size_t *adata, *bdata, *cdata; + EMACS_INT i; + size_t changed =3D 0; + size_t mword; + + CHECK_BOOL_VECTOR (b); + CHECK_BOOL_VECTOR (c); + + nr_bits =3D min (XBOOL_VECTOR (b)->size, + XBOOL_VECTOR (c)->size); + + if (NILP (a)) + { + a =3D Fmake_bool_vector (make_number (nr_bits), Qnil); + changed =3D 1; + } + else + { + CHECK_BOOL_VECTOR (a); + nr_bits =3D min (nr_bits, XBOOL_VECTOR (a)->size); + } + + adata =3D (size_t*) XBOOL_VECTOR (a)->data; + bdata =3D (size_t*) XBOOL_VECTOR (b)->data; + cdata =3D (size_t*) XBOOL_VECTOR (c)->data; + i =3D 0; + do + { + if (op =3D=3D bool_vector_exclusive_or) + mword =3D bdata[i] ^ cdata[i]; + else if (op =3D=3D bool_vector_union || op =3D=3D bool_vector_subs= etp) + mword =3D bdata[i] | cdata[i]; + else if (op =3D=3D bool_vector_intersection) + mword =3D bdata[i] & cdata[i]; + else if (op =3D=3D bool_vector_set_difference) + mword =3D bdata[i] &~ cdata[i]; + else + abort (); + + changed |=3D adata[i] ^ mword; + + if (op !=3D bool_vector_subsetp) + adata[i] =3D mword; + + i +=3D 1; + } + while (i < (nr_bits + BITS_PER_SIZE_T - 1) / BITS_PER_SIZE_T); + + return changed ? a : Qnil; +} + +/* Compute the number of trailing zero bits in val. If val is zero, + return the number of bits in val. */ +static inline +EMACS_INT +count_trailing_zero_bits (size_t val) +{ + if (val =3D=3D 0) + return CHAR_BIT * sizeof (val); + +#if __GNUC__ && BITS_PER_SIZE_T =3D=3D 64 + return __builtin_ctzll (val); +#elif __GNUC__ && BITS_PER_SIZE_T =3D=3D 32 + return __builtin_ctz (val); +#elif __MSC_VER && BITS_PER_SIZE_T =3D=3D 64 +# pragma intrinsic _BitScanForward64 + { + unsigned long result; + _BitScanForward64 (&result, val); + return (EMACS_INT) result; + } +#elif __MSC_VER && BITS_PER_SIZE_T =3D=3D 32 +# pragma intrinsic _BitScanForward + { + unsigned long result; + _BitScanForward (&result, val); + return (EMACS_INT) result; + } +#else + { + EMACS_INT count; + + count =3D 0; + for(val =3D ~val; val & 1; val >>=3D 1) + ++count; + + return count; + } +#endif +} + +static inline +size_t +size_t_to_host_endian (size_t val) +{ +#ifdef WORDS_BIGENDIAN +# if BITS_PER_SIZE_T =3D=3D 64 + return swap64 (val); +# else + return swap32 (val); +# endif +#else + return val; +#endif +} + +DEFUN ("bool-vector-exclusive-or", Fbool_vector_exclusive_or, + Sbool_vector_exclusive_or, 3, 3, 0, + doc: /* Compute A =3D B ^ C, bitwise exclusive or. +A, B, and C must be bool vectors. A is +any bool vector except C. If A is nil, allocate a new bool +vector in which to store the result. +Return the destination vector if it changed or nil otherwise. */) + (Lisp_Object a, Lisp_Object b, Lisp_Object c) +{ + return bool_vector_binop_driver (a, b, c, bool_vector_exclusive_or); +} + +DEFUN ("bool-vector-union", Fbool_vector_union, + Sbool_vector_union, 3, 3, 0, + doc: /* Compute A =3D B | C, bitwise or. +A, B, and C must be bool vectors. A is +any bool vector except C. If A is nil, allocate a new bool +vector in which to store the result. +Return the destination vector if it changed and nil otherwise. */) + (Lisp_Object a, Lisp_Object b, Lisp_Object c) +{ + return bool_vector_binop_driver (a, b, c, bool_vector_union); +} + +DEFUN ("bool-vector-intersection", Fbool_vector_intersection, + Sbool_vector_intersection, 3, 3, 0, + doc: /* Compute A =3D B & C, bitwise and. +A, B, and C must be bool vectors. A is +any bool vector except C. If A is nil, allocate a new bool +vector in which to store the result. +Return the destination vector if it changed and nil otherwise. */) + (Lisp_Object a, Lisp_Object b, Lisp_Object c) +{ + return bool_vector_binop_driver (a, b, c, bool_vector_intersection); +} + +DEFUN ("bool-vector-set-difference", Fbool_vector_set_difference, + Sbool_vector_set_difference, 3, 3, 0, + doc: /* Compute A =3D B &~ C. +A, B, and C must be bool vectors. A is +any bool vector except C. If A is nil, allocate a new bool +vector in which to store the result. +Return the destination vector if it changed and nil otherwise. */) + (Lisp_Object a, Lisp_Object b, Lisp_Object c) +{ + return bool_vector_binop_driver (a, b, c, bool_vector_set_difference);= +} + +DEFUN ("bool-vector-subsetp", Fbool_vector_subsetp, + Sbool_vector_subsetp, 2, 2, 0, + doc: /* Check whether all bits in A are set in B. +A and B must be bool vectors. Return a generalize boolean. */) + (Lisp_Object a, Lisp_Object b) +{ + /* Like bool_vector_union, but doesn't modify b. */ + return bool_vector_binop_driver (b, a, b, bool_vector_subsetp); +} + +DEFUN ("bool-vector-not", Fbool_vector_not, + Sbool_vector_not, 2, 2, 0, + doc: /* Compute A =3D ~B. +A and B must be bool vectors. +If A is nil, allocate a new bool vector in which to store the result. +Return the destination vector. */) + (Lisp_Object a, Lisp_Object b) +{ + EMACS_INT nr_bits; + size_t *adata, *bdata; + EMACS_INT i; + size_t mword; + + CHECK_BOOL_VECTOR (b); + nr_bits =3D XBOOL_VECTOR (b)->size; + + if (NILP (a)) + a =3D Fmake_bool_vector (make_number (nr_bits), Qnil); + else + { + CHECK_BOOL_VECTOR (a); + nr_bits =3D min (nr_bits, XBOOL_VECTOR (a)->size); + } + + adata =3D (size_t*) XBOOL_VECTOR (a)->data; + bdata =3D (size_t*) XBOOL_VECTOR (b)->data; + i =3D 0; + do + { + adata[i] =3D ~bdata[i]; + i +=3D 1; + } + while (i < nr_bits / BITS_PER_SIZE_T); + + if (nr_bits % BITS_PER_SIZE_T) + { + mword =3D size_t_to_host_endian (bdata[i]); + mword =3D ~mword; + mword &=3D bool_vector_spare_mask (nr_bits); + adata[i] =3D size_t_to_host_endian (bdata[i]); + } + + return a; +} + +DEFUN ("bool-vector-count-matches", Fbool_vector_count_matches, + Sbool_vector_count_matches, 2, 2, 0, + doc: /* Count how many elements in A equal B. +A must be a bool vector. B is a generalized bool. */) + (Lisp_Object a, Lisp_Object b) +{ + EMACS_INT count; + EMACS_INT nr_bits; + size_t *adata; + size_t match; + EMACS_INT i; + + CHECK_BOOL_VECTOR (a); + + nr_bits =3D XBOOL_VECTOR (a)->size; + count =3D 0; + match =3D NILP (b) ? (size_t) -1 : 0; + adata =3D (size_t*) XBOOL_VECTOR (a)->data; + + for(i =3D 0; i < nr_bits / BITS_PER_SIZE_T; ++i) + count +=3D popcount_size_t (adata[i] ^ match); + + /* Mask out trailing parts of final mword. */ + if (nr_bits % BITS_PER_SIZE_T) + { + size_t mword =3D adata[i] ^ match; + mword =3D size_t_to_host_endian (mword); + count +=3D popcount_size_t (mword & bool_vector_spare_mask (nr_bit= s)); + } + + return make_number (count); +} + +DEFUN ("bool-vector-count-matches-at", + Fbool_vector_count_matches_at, + Sbool_vector_count_matches_at, 3, 3, 0, + doc: /* Count how many consecutive elements in A equal B at i. +A must be a bool vector. B is a generalized boolean. i is an +index into the vector.*/) + (Lisp_Object a, Lisp_Object b, Lisp_Object i) +{ + EMACS_INT count; + EMACS_INT nr_bits; + EMACS_INT offset; + size_t *adata; + size_t twiddle; + size_t mword; /* Machine word. */ + EMACS_INT pos; + EMACS_INT nr_words; + + CHECK_BOOL_VECTOR (a); + CHECK_NATNUM (i); + + nr_bits =3D XBOOL_VECTOR (a)->size; + if (XFASTINT (i) > nr_bits) /* Allow one past the end for convenience = */ + args_out_of_range (a, i); + + adata =3D (size_t*) XBOOL_VECTOR (a)->data; + nr_words =3D (nr_bits + BITS_PER_SIZE_T - 1) / BITS_PER_SIZE_T; + pos =3D XFASTINT (i) / BITS_PER_SIZE_T; + offset =3D XFASTINT (i) % BITS_PER_SIZE_T; + count =3D 0; + + /* By XORing with twiddle, we transform the problem of "count + consecutive equal values" into "count the zero bits". The latter + operation usually has hardware support. */ + twiddle =3D NILP (b) ? 0 : (size_t) -1; + + /* Scan the remainder of the mword at the current offset. */ + if (pos < nr_words && offset !=3D 0) + { + mword =3D size_t_to_host_endian (adata[pos]); + mword ^=3D twiddle; + mword >>=3D offset; + count =3D count_trailing_zero_bits (mword); + count =3D min (count, BITS_PER_SIZE_T - offset); + pos +=3D 1; + if (count + offset < BITS_PER_SIZE_T) + return make_number (count); + } + + /* Scan whole words until we either reach the end of the vector or + find an mword that doesn't completely match. twiddle is + endian-independent. */ + while (pos < nr_words && adata[pos] =3D=3D twiddle) + { + count +=3D BITS_PER_SIZE_T; + ++pos; + } + + if (pos < nr_words) + { + /* If we stopped because of a mismatch, see how many bits match + in the current mword. */ + mword =3D size_t_to_host_endian (adata[pos]); + mword ^=3D twiddle; + count +=3D count_trailing_zero_bits (mword); + } + else if (nr_bits % BITS_PER_SIZE_T !=3D 0) + { + /* If we hit the end, we might have overshot our count. Reduce + the total by the number of spare bits at the end of the + vector. */ + count -=3D BITS_PER_SIZE_T - nr_bits % BITS_PER_SIZE_T; + } + + return make_number (count); +} =0C void @@ -3005,6 +3384,7 @@ DEFSYM (Qsequencep, "sequencep"); DEFSYM (Qbufferp, "bufferp"); DEFSYM (Qvectorp, "vectorp"); + DEFSYM (Qbool_vector_p, "bool-vector-p"); DEFSYM (Qchar_or_string_p, "char-or-string-p"); DEFSYM (Qmarkerp, "markerp"); DEFSYM (Qbuffer_or_string_p, "buffer-or-string-p"); @@ -3222,6 +3602,15 @@ defsubr (&Ssubr_arity); defsubr (&Ssubr_name); + defsubr (&Sbool_vector_exclusive_or); + defsubr (&Sbool_vector_union); + defsubr (&Sbool_vector_intersection); + defsubr (&Sbool_vector_set_difference); + defsubr (&Sbool_vector_not); + defsubr (&Sbool_vector_subsetp); + defsubr (&Sbool_vector_count_matches); + defsubr (&Sbool_vector_count_matches_at); + set_symbol_function (Qwholenump, XSYMBOL (Qnatnump)->function); DEFVAR_LISP ("most-positive-fixnum", Vmost_positive_fixnum, =3D=3D=3D modified file 'src/lisp.h' --- src/lisp.h 2013-09-11 05:03:23 +0000 +++ src/lisp.h 2013-09-20 22:52:47 +0000 @@ -733,6 +733,7 @@ extern Lisp_Object Qarrayp, Qbufferp, Qbuffer_or_string_p, Qchar_table_p= ; extern Lisp_Object Qconsp, Qfloatp, Qintegerp, Qlambda, Qlistp, Qmarkerp= , Qnil; extern Lisp_Object Qnumberp, Qstringp, Qsymbolp, Qvectorp; +extern Lisp_Object Qbool_vector_p; extern Lisp_Object Qvector_or_char_table_p, Qwholenump; extern Lisp_Object Qwindow; extern Lisp_Object Ffboundp (Lisp_Object); @@ -2362,6 +2363,11 @@ CHECK_TYPE (VECTORP (x), Qvectorp, x); } LISP_INLINE void +CHECK_BOOL_VECTOR (Lisp_Object x) +{ + CHECK_TYPE (BOOL_VECTOR_P (x), Qbool_vector_p, x); +} +LISP_INLINE void CHECK_VECTOR_OR_STRING (Lisp_Object x) { CHECK_TYPE (VECTORP (x) || STRINGP (x), Qarrayp, x); @@ -4350,6 +4356,33 @@ return 0; } +LISP_INLINE +uint16_t +swap16 (uint16_t val) +{ + return (val << 8) | (val & 0xFF); +} + +LISP_INLINE +uint32_t +swap32 (uint32_t val) +{ + uint32_t low =3D swap16 (val & 0xFFFF); + uint32_t high =3D swap16 (val >> 16); + return (low << 16) | high; +} + +#ifdef UINT64_MAX +LISP_INLINE +uint64_t +swap64 (uint64_t val) +{ + uint64_t low =3D swap32 (val & 0xFFFFFFFF); + uint64_t high =3D swap32 (val >> 32); + return (low << 32) | high; +} +#endif + INLINE_HEADER_END #endif /* EMACS_LISP_H */ =3D=3D=3D modified file 'src/xsettings.c' --- src/xsettings.c 2013-09-17 07:06:42 +0000 +++ src/xsettings.c 2013-09-19 22:24:27 +0000 @@ -336,9 +336,6 @@ XUngrabServer (dpy); } -#define SWAP32(nr) (((nr) << 24) | (((nr) << 8) & 0xff0000) \ - | (((nr) >> 8) & 0xff00) | ((nr) >> 24)) -#define SWAP16(nr) (((nr) << 8) | ((nr) >> 8)) #define PAD(nr) (((nr) + 3) & ~3) /* Parse xsettings and extract those that deal with Xft. @@ -408,7 +405,7 @@ if (bytes < 12) return BadLength; memcpy (&n_settings, prop+8, 4); - if (my_bo !=3D that_bo) n_settings =3D SWAP32 (n_settings); + if (my_bo !=3D that_bo) n_settings =3D swap32 (n_settings); bytes_parsed =3D 12; memset (settings, 0, sizeof (*settings)); @@ -430,7 +427,7 @@ memcpy (&nlen, prop+bytes_parsed, 2); bytes_parsed +=3D 2; - if (my_bo !=3D that_bo) nlen =3D SWAP16 (nlen); + if (my_bo !=3D that_bo) nlen =3D swap16 (nlen); if (bytes_parsed+nlen > bytes) return BadLength; to_cpy =3D nlen > 127 ? 127 : nlen; memcpy (name, prop+bytes_parsed, to_cpy); @@ -457,7 +454,7 @@ if (want_this) { memcpy (&ival, prop+bytes_parsed, 4); - if (my_bo !=3D that_bo) ival =3D SWAP32 (ival); + if (my_bo !=3D that_bo) ival =3D swap32 (ival); } bytes_parsed +=3D 4; break; @@ -466,7 +463,7 @@ if (bytes_parsed+4 > bytes) return BadLength; memcpy (&vlen, prop+bytes_parsed, 4); bytes_parsed +=3D 4; - if (my_bo !=3D that_bo) vlen =3D SWAP32 (vlen); + if (my_bo !=3D that_bo) vlen =3D swap32 (vlen); if (want_this) { to_cpy =3D vlen > 127 ? 127 : vlen; =3D=3D=3D modified file 'test/automated/data-tests.el' --- test/automated/data-tests.el 2013-09-11 05:03:23 +0000 +++ test/automated/data-tests.el 2013-09-20 22:46:42 +0000 @@ -21,6 +21,9 @@ ;;; Code: +(require 'cl-lib) +(eval-when-compile (require 'cl)) + (ert-deftest data-tests-=3D () (should-error (=3D)) (should (=3D 1)) @@ -71,5 +74,166 @@ ;; Short circuits before getting to bad arg (should-not (>=3D 8 9 'foo))) +;; Bool vector tests. Compactly represent bool vectors as hex +;; strings. + +(ert-deftest bool-vector-count-matches-all-0-nil () + (cl-loop for sz in '(0 45 1 64 9 344) + do (let* ((bv (make-bool-vector sz nil))) + (should + (eql + (bool-vector-count-matches bv nil) + sz))))) + +(ert-deftest bool-vector-count-matches-all-0-t () + (cl-loop for sz in '(0 45 1 64 9 344) + do (let* ((bv (make-bool-vector sz nil))) + (should + (eql + (bool-vector-count-matches bv t) + 0))))) + +(ert-deftest bool-vector-count-matches-1-nil () + (let* ((bv (make-bool-vector 45 nil))) + (aset bv 40 t) + (aset bv 0 t) + (should + (eql + (bool-vector-count-matches bv t) + 2))) + ) + +(ert-deftest bool-vector-count-matches-1-t () + (let* ((bv (make-bool-vector 45 nil))) + (aset bv 40 t) + (aset bv 0 t) + (should + (eql + (bool-vector-count-matches bv nil) + 43)))) + +(defun mock-bool-vector-count-matches-at (a b i) + (loop for i from i below (length a) + while (eq (aref a i) b) + sum 1)) + +(defun test-bool-vector-bv-from-hex-string (desc) + (let (bv nchars nibbles) + (dolist (c (string-to-list desc)) + (push (string-to-number + (char-to-string c) + 16) + nibbles)) + (setf bv (make-bool-vector (* 4 (length nibbles)) nil)) + (let ((i 0)) + (dolist (n (nreverse nibbles)) + (dotimes (_ 4) + (aset bv i (> (logand 1 n) 0)) + (incf i) + (setf n (lsh n -1))))) + bv)) + +(defun test-bool-vector-to-hex-string (bv) + (let (nibbles (v (cl-coerce bv 'list))) + (while v + (push (logior + (lsh (if (nth 0 v) 1 0) 0) + (lsh (if (nth 1 v) 1 0) 1) + (lsh (if (nth 2 v) 1 0) 2) + (lsh (if (nth 3 v) 1 0) 3)) + nibbles) + (setf v (nthcdr 4 v))) + (mapconcat (lambda (n) (format "%X" n)) + (nreverse nibbles) + ""))) + +(defun test-bool-vector-count-matches-at-tc (desc) + "Run a test case for bool-vector-count-matches-at. +DESC is a string describing the test. It is a sequence of +hexadecimal digits describing the bool vector. We exhaustively +test all counts at all possible positions in the vector by +comparing the subr with a much slower lisp implementation." + (let ((bv (test-bool-vector-bv-from-hex-string desc))) + (loop + for lf in '(nil t) + do (loop + for pos from 0 upto (length bv) + for cnt =3D (mock-bool-vector-count-matches-at bv lf pos) + for rcnt =3D (bool-vector-count-matches-at bv lf pos) + unless (eql cnt rcnt) + do (error "FAILED testcase %S %3S %3S %3S" + pos lf cnt rcnt))))) + +(defconst bool-vector-test-vectors +'("" + "0" + "F" + "0F" + "F0" + "00000000000000000000000000000FFFFF0000000" + "44a50234053fba3340000023444a50234053fba33400000234" + "12341234123456123412346001234123412345612341234600" + "44a50234053fba33400000234" + "1234123412345612341234600" + "44a50234053fba33400000234" + "1234123412345612341234600" + "44a502340" + "123412341" + "0000000000000000000000000" + "FFFFFFFFFFFFFFFF1")) + +(ert-deftest bool-vector-count-matches-at () + (mapc #'test-bool-vector-count-matches-at-tc + bool-vector-test-vectors)) + +(defun test-bool-vector-apply-mock-op (mock a b c) + "Compute (slowly) the correct result of a bool-vector set operation." + (let (changed nv) + (assert (eql (length b) (length c))) + (if a (setf nv a) + (setf a (make-bool-vector (length b) nil)) + (setf changed t)) + + (loop for i below (length b) + for mockr =3D (funcall mock + (if (aref b i) 1 0) + (if (aref c i) 1 0)) + for r =3D (not (=3D 0 mockr)) + do (progn + (unless (eq (aref a i) r) + (setf changed t)) + (setf (aref a i) r))) + (if changed a))) + +(defun test-bool-vector-binop (mock real) + "Test a binary set operation." + (loop for s1 in bool-vector-test-vectors + for bv1 =3D (test-bool-vector-bv-from-hex-string s1) + for vecs2 =3D (cl-remove-if-not + (lambda (x) (eql (length x) (length s1))) + bool-vector-test-vectors) + do (loop for s2 in vecs2 + for bv2 =3D (test-bool-vector-bv-from-hex-string s2) + for mock-result =3D (test-bool-vector-apply-mock-op + mock nil bv1 bv2) + for real-result =3D (funcall real nil bv1 bv2) + do (progn + (should (equal mock-result real-result)))))) + +(ert-deftest bool-vector-intersection-op () + (test-bool-vector-binop + #'logand + #'bool-vector-intersection)) + +(ert-deftest bool-vector-union-op () + (test-bool-vector-binop + #'logior + #'bool-vector-union)) + +(ert-deftest bool-vector-xor-op () + (test-bool-vector-binop + #'logxor + #'bool-vector-exclusive-or)) + + ;;; data-tests.el ends here - --6AsIIl4JhB7AE8Ksqq1g1P57LDgGhnHBf Content-Type: application/pgp-signature; name="signature.asc" Content-Description: OpenPGP digital signature Content-Disposition: attachment; filename="signature.asc" -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.14 (Darwin) iQIcBAEBAgAGBQJSPNNoAAoJEMAaIROpHW7I8BkP/1Z14N3RONPWvFYlSoGqKv+n Ks/6DeX+WypkXVR8IIBTyT1b+dyPgitDFyrIV4Qg4D8QCfAq3n02/1ok7MJbSAPE Qdm0KK0/1tEZmr15r0fGXzSOJzB01V6kM0ubFlfJ3IeQj5H7Vw/B5mPT5R0QYHyP ChkhNJDXUMGzyGU0zcGL25vGf50/STVuXaR/GpUGisjAy7EMQ4gmeAy52Z4pNHrn 4kgi3hE4qdsKOWFMuGBYfB2W3nuztDAAfBz8eXM7wMO5eIHCr1hTp1KdMOvr2dNh N0cBcByVoZ/QEc/OpgLIv2Zz+QoV0BndD42IRHuhpmt0vk0RgVZPDFrYtAl5MS/a mzBn8hjuqSH1HpAlG4uDnZ+k//iS4m75os3QKuhxp8oFafUcZPAH2a/uwG+H8+77 qOQpsQDmaBXUT6vYOJaU7hDK6RwkCTSyE/lSD+5cU+3lc+dlb9fxf4mSElNXAPFi GtRgwkDln3vEPPHbXxxco8fCynku311STMZzvEn/3l/1tOFhaWPCagmx1UYJFayf tw5TOI51AcW+7RuyLC+7h2oTWGKcpYTOlNZzQghe48+numOSZngvtj4BMTpylkTw CIK4ivsBq9P9QQrLbKGM/LfknZuMUt/Qf+lYQ8h0a4iPG3XhaWxSDZvYoKaNcIHu eTDmLFAeJdE4NKaan7PV =SkyW -----END PGP SIGNATURE----- --6AsIIl4JhB7AE8Ksqq1g1P57LDgGhnHBf--