Files
calc/zmath.c
Landon Curt Noll ff90bc0e3a add E_STRING to error, errno, strerror, change multiple E_STRING's
While help/errstr has been added, the errstr builtin function is
not yet written.  In anticipation of the new errstr builtin the
rest of the calc error system has been updated to associated errsym
E_STRING's with errnum error codes and errmsg error messages.

Minor improvements to help/rand.

The verify_error_table() function that does a verification
the error_table[] array and setup private_error_alias[] array
is now called by libcalc_call_me_first().

Fix comment about wrong include file in have_sys_mount.h.

Removed unused booltostr() and strtobool() macros from bool.h.

Moved define of math_error(char *, ...) from zmath.h to errtbl.h.
The errtbl.h include file, unless ERRCODE_SRC is defined
also includes attribute.h and errsym.h.

Group calc error related builtin support functions together in func.c.

Make switch indenting in func.c consistent.

Passing an invalid argument to error(), errno() or strerror() will
set errno AND throw a math error.  Before errno would be set and
an error value was returned.  Before there was no way to tell if
the error value was a result of the arg or if an error detected.

Added E_STRING to error([errnum | "E_STRING"]) builtin function.
Added E_STRING to errno([errnum | "E_STRING"]) builtin function.
Added E_STRING to strerror([errnum | "E_STRING"]) builtin function.
Calling these functions with an E_STRING errsym is the same as calling
them with the matching errnum code.

Standardized on calc computation error related E_STRING strings
where there are a set of related codes.  Changed "E_...digits" into
"E_..._digits".  For example, E_FPUTC1 became E_FPUTC_1, E_FPUTC2
became E_FPUTC_2, and E_FPUTC3 became E_FPUTC_3.  In a few cases
such as E_APPR became E_APPR_1, because there was a E_APPR2 (which
became E_APPR_2) and E_APPR3 (which became E_APPR_3).  To other
special cases, E_ILOG10 became E_IBASE10_LOG and E_ILOG2 became
E_IBASE2_LOG because E_ILOG10 and E_ILOG2 are both independent calc
computation error related E_STRING strings.  Now related sets of
E_STRING strings end in _ (underscore) followed by digits.

The following is the list of E_STRING strings changes:

    E_APPR ==> E_APPR_1
    E_ROUND ==> E_ROUND_1
    E_SQRT ==> E_SQRT_1
    E_ROOT ==> E_ROOT_1
    E_SHIFT ==> E_SHIFT_1
    E_SCALE ==> E_SCALE_1
    E_POWI ==> E_POWI_1
    E_POWER ==> E_POWER_1
    E_QUO ==> E_QUO_1
    E_MOD ==> E_MOD_1
    E_ABS ==> E_ABS_1
    E_APPR2 ==> E_APPR_2
    E_APPR3 ==> E_APPR_3
    E_ROUND2 ==> E_ROUND_2
    E_ROUND3 ==> E_ROUND_3
    E_BROUND2 ==> E_BROUND_2
    E_BROUND3 ==> E_BROUND_3
    E_SQRT2 ==> E_SQRT_2
    E_SQRT3 ==> E_SQRT_3
    E_ROOT2 ==> E_ROOT_2
    E_ROOT3 ==> E_ROOT_3
    E_SHIFT2 ==> E_SHIFT_2
    E_SCALE2 ==> E_SCALE_2
    E_POWI2 ==> E_POWI_2
    E_POWER2 ==> E_POWER_2
    E_POWER3 ==> E_POWER_3
    E_QUO2 ==> E_QUO_2
    E_QUO3 ==> E_QUO_3
    E_MOD2 ==> E_MOD_2
    E_MOD3 ==> E_MOD_3
    E_ABS2 ==> E_ABS_2
    E_EXP1 ==> E_EXP_1
    E_EXP2 ==> E_EXP_2
    E_FPUTC1 ==> E_FPUTC_1
    E_FPUTC2 ==> E_FPUTC_2
    E_FPUTC3 ==> E_FPUTC_3
    E_FGETC1 ==> E_FGETC_1
    E_FGETC2 ==> E_FGETC_2
    E_FOPEN1 ==> E_FOPEN_1
    E_FOPEN2 ==> E_FOPEN_2
    E_FREOPEN1 ==> E_FREOPEN_1
    E_FREOPEN2 ==> E_FREOPEN_2
    E_FREOPEN3 ==> E_FREOPEN_3
    E_FCLOSE1 ==> E_FCLOSE_1
    E_FPUTS1 ==> E_FPUTS_1
    E_FPUTS2 ==> E_FPUTS_2
    E_FPUTS3 ==> E_FPUTS_3
    E_FGETS1 ==> E_FGETS_1
    E_FGETS2 ==> E_FGETS_2
    E_FPUTSTR1 ==> E_FPUTSTR_1
    E_FPUTSTR2 ==> E_FPUTSTR_2
    E_FPUTSTR3 ==> E_FPUTSTR_3
    E_FGETSTR1 ==> E_FGETSTR_1
    E_FGETSTR2 ==> E_FGETSTR_2
    E_FGETLINE1 ==> E_FGETLINE_1
    E_FGETLINE2 ==> E_FGETLINE_2
    E_FGETFIELD1 ==> E_FGETFIELD_1
    E_FGETFIELD2 ==> E_FGETFIELD_2
    E_REWIND1 ==> E_REWIND_1
    E_PRINTF1 ==> E_PRINTF_1
    E_PRINTF2 ==> E_PRINTF_2
    E_FPRINTF1 ==> E_FPRINTF_1
    E_FPRINTF2 ==> E_FPRINTF_2
    E_FPRINTF3 ==> E_FPRINTF_3
    E_STRPRINTF1 ==> E_STRPRINTF_1
    E_STRPRINTF2 ==> E_STRPRINTF_2
    E_FSCAN1 ==> E_FSCAN_1
    E_FSCAN2 ==> E_FSCAN_2
    E_FSCANF1 ==> E_FSCANF_1
    E_FSCANF2 ==> E_FSCANF_2
    E_FSCANF3 ==> E_FSCANF_3
    E_FSCANF4 ==> E_FSCANF_4
    E_STRSCANF1 ==> E_STRSCANF_1
    E_STRSCANF2 ==> E_STRSCANF_2
    E_STRSCANF3 ==> E_STRSCANF_3
    E_STRSCANF4 ==> E_STRSCANF_4
    E_SCANF1 ==> E_SCANF_1
    E_SCANF2 ==> E_SCANF_2
    E_SCANF3 ==> E_SCANF_3
    E_FTELL1 ==> E_FTELL_1
    E_FTELL2 ==> E_FTELL_2
    E_FSEEK1 ==> E_FSEEK_1
    E_FSEEK2 ==> E_FSEEK_2
    E_FSEEK3 ==> E_FSEEK_3
    E_FSIZE1 ==> E_FSIZE_1
    E_FSIZE2 ==> E_FSIZE_2
    E_FEOF1 ==> E_FEOF_1
    E_FEOF2 ==> E_FEOF_2
    E_FERROR1 ==> E_FERROR_1
    E_FERROR2 ==> E_FERROR_2
    E_UNGETC1 ==> E_UNGETC_1
    E_UNGETC2 ==> E_UNGETC_2
    E_UNGETC3 ==> E_UNGETC_3
    E_ISATTY1 ==> E_ISATTY_1
    E_ISATTY2 ==> E_ISATTY_2
    E_ACCESS1 ==> E_ACCESS_1
    E_ACCESS2 ==> E_ACCESS_2
    E_SEARCH1 ==> E_SEARCH_1
    E_SEARCH2 ==> E_SEARCH_2
    E_SEARCH3 ==> E_SEARCH_3
    E_SEARCH4 ==> E_SEARCH_4
    E_SEARCH5 ==> E_SEARCH_5
    E_SEARCH6 ==> E_SEARCH_6
    E_RSEARCH1 ==> E_RSEARCH_1
    E_RSEARCH2 ==> E_RSEARCH_2
    E_RSEARCH3 ==> E_RSEARCH_3
    E_RSEARCH4 ==> E_RSEARCH_4
    E_RSEARCH5 ==> E_RSEARCH_5
    E_RSEARCH6 ==> E_RSEARCH_6
    E_REWIND2 ==> E_REWIND_2
    E_STRERROR1 ==> E_STRERROR_1
    E_STRERROR2 ==> E_STRERROR_2
    E_COS1 ==> E_COS_1
    E_COS2 ==> E_COS_2
    E_SIN1 ==> E_SIN_1
    E_SIN2 ==> E_SIN_2
    E_EVAL2 ==> E_EVAL_2
    E_ARG1 ==> E_ARG_1
    E_ARG2 ==> E_ARG_2
    E_POLAR1 ==> E_POLAR_1
    E_POLAR2 ==> E_POLAR_2
    E_MATFILL1 ==> E_MATFILL_1
    E_MATFILL2 ==> E_MATFILL_2
    E_MATTRANS1 ==> E_MATTRANS_1
    E_MATTRANS2 ==> E_MATTRANS_2
    E_DET1 ==> E_DET_1
    E_DET2 ==> E_DET_2
    E_DET3 ==> E_DET_3
    E_MATMIN1 ==> E_MATMIN_1
    E_MATMIN2 ==> E_MATMIN_2
    E_MATMIN3 ==> E_MATMIN_3
    E_MATMAX1 ==> E_MATMAX_1
    E_MATMAX2 ==> E_MATMAX_2
    E_MATMAX3 ==> E_MATMAX_3
    E_CP1 ==> E_CP_1
    E_CP2 ==> E_CP_2
    E_CP3 ==> E_CP_3
    E_DP1 ==> E_DP_1
    E_DP2 ==> E_DP_2
    E_DP3 ==> E_DP_3
    E_SUBSTR1 ==> E_SUBSTR_1
    E_SUBSTR2 ==> E_SUBSTR_2
    E_INSERT1 ==> E_INSERT_1
    E_INSERT2 ==> E_INSERT_2
    E_DELETE1 ==> E_DELETE_1
    E_DELETE2 ==> E_DELETE_2
    E_LN1 ==> E_LN_1
    E_LN2 ==> E_LN_2
    E_ERROR1 ==> E_ERROR_1
    E_ERROR2 ==> E_ERROR_2
    E_EVAL3 ==> E_EVAL_3
    E_EVAL4 ==> E_EVAL_4
    E_RM1 ==> E_RM_1
    E_RM2 ==> E_RM_2
    E_BLK1 ==> E_BLK_1
    E_BLK2 ==> E_BLK_2
    E_BLK3 ==> E_BLK_3
    E_BLK4 ==> E_BLK_4
    E_BLKFREE1 ==> E_BLKFREE_1
    E_BLKFREE2 ==> E_BLKFREE_2
    E_BLKFREE3 ==> E_BLKFREE_3
    E_BLKFREE4 ==> E_BLKFREE_4
    E_BLKFREE5 ==> E_BLKFREE_5
    E_BLOCKS1 ==> E_BLOCKS_1
    E_BLOCKS2 ==> E_BLOCKS_2
    E_COPY1 ==> E_COPY_01
    E_COPY2 ==> E_COPY_02
    E_COPY3 ==> E_COPY_03
    E_COPY4 ==> E_COPY_04
    E_COPY5 ==> E_COPY_05
    E_COPY6 ==> E_COPY_06
    E_COPY7 ==> E_COPY_07
    E_COPY8 ==> E_COPY_08
    E_COPY9 ==> E_COPY_09
    E_COPY10 ==> E_COPY_10
    E_COPY11 ==> E_COPY_11
    E_COPY12 ==> E_COPY_12
    E_COPY13 ==> E_COPY_13
    E_COPY14 ==> E_COPY_14
    E_COPY15 ==> E_COPY_15
    E_COPY16 ==> E_COPY_16
    E_COPY17 ==> E_COPY_17
    E_COPYF1 ==> E_COPYF_1
    E_COPYF2 ==> E_COPYF_2
    E_COPYF3 ==> E_COPYF_3
    E_COPYF4 ==> E_COPYF_4
    E_PROTECT1 ==> E_PROTECT_1
    E_PROTECT2 ==> E_PROTECT_2
    E_PROTECT3 ==> E_PROTECT_3
    E_MATFILL3 ==> E_MATFILL_3
    E_MATFILL4 ==> E_MATFILL_4
    E_MATTRACE1 ==> E_MATTRACE_1
    E_MATTRACE2 ==> E_MATTRACE_2
    E_MATTRACE3 ==> E_MATTRACE_3
    E_TAN1 ==> E_TAN_1
    E_TAN2 ==> E_TAN_2
    E_COT1 ==> E_COT_1
    E_COT2 ==> E_COT_2
    E_SEC1 ==> E_SEC_1
    E_SEC2 ==> E_SEC_2
    E_CSC1 ==> E_CSC_1
    E_CSC2 ==> E_CSC_2
    E_SINH1 ==> E_SINH_1
    E_SINH2 ==> E_SINH_2
    E_COSH1 ==> E_COSH_1
    E_COSH2 ==> E_COSH_2
    E_TANH1 ==> E_TANH_1
    E_TANH2 ==> E_TANH_2
    E_COTH1 ==> E_COTH_1
    E_COTH2 ==> E_COTH_2
    E_SECH1 ==> E_SECH_1
    E_SECH2 ==> E_SECH_2
    E_CSCH1 ==> E_CSCH_1
    E_CSCH2 ==> E_CSCH_2
    E_ASIN1 ==> E_ASIN_1
    E_ASIN2 ==> E_ASIN_2
    E_ACOS1 ==> E_ACOS_1
    E_ACOS2 ==> E_ACOS_2
    E_ATAN1 ==> E_ATAN_1
    E_ATAN2 ==> E_ATAN_2
    E_ACOT1 ==> E_ACOT_1
    E_ACOT2 ==> E_ACOT_2
    E_ASEC1 ==> E_ASEC_1
    E_ASEC2 ==> E_ASEC_2
    E_ACSC1 ==> E_ACSC_1
    E_ACSC2 ==> E_ACSC_2
    E_ASINH1 ==> E_ASINH_1
    E_ASINH2 ==> E_ASINH_2
    E_ACOSH1 ==> E_ACOSH_1
    E_ACOSH2 ==> E_ACOSH_2
    E_ATANH1 ==> E_ATANH_1
    E_ATANH2 ==> E_ATANH_2
    E_ACOTH1 ==> E_ACOTH_1
    E_ACOTH2 ==> E_ACOTH_2
    E_ASECH1 ==> E_ASECH_1
    E_ASECH2 ==> E_ASECH_2
    E_ACSCH1 ==> E_ACSCH_1
    E_ACSCH2 ==> E_ACSCH_2
    E_GD1 ==> E_GD_1
    E_GD2 ==> E_GD_2
    E_AGD1 ==> E_AGD_1
    E_AGD2 ==> E_AGD_2
    E_BIT1 ==> E_BIT_1
    E_BIT2 ==> E_BIT_2
    E_SETBIT1 ==> E_SETBIT_1
    E_SETBIT2 ==> E_SETBIT_2
    E_SETBIT3 ==> E_SETBIT_3
    E_SEG1 ==> E_SEG_1
    E_SEG2 ==> E_SEG_2
    E_SEG3 ==> E_SEG_3
    E_HIGHBIT1 ==> E_HIGHBIT_1
    E_HIGHBIT2 ==> E_HIGHBIT_2
    E_LOWBIT1 ==> E_LOWBIT_1
    E_LOWBIT2 ==> E_LOWBIT_2
    E_HEAD1 ==> E_HEAD_1
    E_HEAD2 ==> E_HEAD_2
    E_TAIL1 ==> E_TAIL_1
    E_TAIL2 ==> E_TAIL_2
    E_XOR1 ==> E_XOR_1
    E_XOR2 ==> E_XOR_2
    E_INDICES1 ==> E_INDICES_1
    E_INDICES2 ==> E_INDICES_2
    E_EXP3 ==> E_EXP_3
    E_SINH3 ==> E_SINH_3
    E_COSH3 ==> E_COSH_3
    E_SIN3 ==> E_SIN_3
    E_COS3 ==> E_COS_3
    E_GD3 ==> E_GD_3
    E_AGD3 ==> E_AGD_3
    E_POWER4 ==> E_POWER_4
    E_ROOT4 ==> E_ROOT_4
    E_DGT1 ==> E_DGT_1
    E_DGT2 ==> E_DGT_2
    E_DGT3 ==> E_DGT_3
    E_PLCS1 ==> E_PLCS_1
    E_PLCS2 ==> E_PLCS_2
    E_DGTS1 ==> E_DGTS_1
    E_DGTS2 ==> E_DGTS_2
    E_ILOG10 ==> E_IBASE10_LOG
    E_ILOG2 ==> E_IBASE2_LOG
    E_COMB1 ==> E_COMB_1
    E_COMB2 ==> E_COMB_2
    E_ASSIGN1 ==> E_ASSIGN_1
    E_ASSIGN2 ==> E_ASSIGN_2
    E_ASSIGN3 ==> E_ASSIGN_3
    E_ASSIGN4 ==> E_ASSIGN_4
    E_ASSIGN5 ==> E_ASSIGN_5
    E_ASSIGN6 ==> E_ASSIGN_6
    E_ASSIGN7 ==> E_ASSIGN_7
    E_ASSIGN8 ==> E_ASSIGN_8
    E_ASSIGN9 ==> E_ASSIGN_9
    E_SWAP1 ==> E_SWAP_1
    E_SWAP2 ==> E_SWAP_2
    E_SWAP3 ==> E_SWAP_3
    E_QUOMOD1 ==> E_QUOMOD_1
    E_QUOMOD2 ==> E_QUOMOD_2
    E_QUOMOD3 ==> E_QUOMOD_3
    E_PREINC1 ==> E_PREINC_1
    E_PREINC2 ==> E_PREINC_2
    E_PREINC3 ==> E_PREINC_3
    E_PREDEC1 ==> E_PREDEC_1
    E_PREDEC2 ==> E_PREDEC_2
    E_PREDEC3 ==> E_PREDEC_3
    E_POSTINC1 ==> E_POSTINC_1
    E_POSTINC2 ==> E_POSTINC_2
    E_POSTINC3 ==> E_POSTINC_3
    E_POSTDEC1 ==> E_POSTDEC_1
    E_POSTDEC2 ==> E_POSTDEC_2
    E_POSTDEC3 ==> E_POSTDEC_3
    E_INIT1 ==> E_INIT_01
    E_INIT2 ==> E_INIT_02
    E_INIT3 ==> E_INIT_03
    E_INIT4 ==> E_INIT_04
    E_INIT5 ==> E_INIT_05
    E_INIT6 ==> E_INIT_06
    E_INIT7 ==> E_INIT_07
    E_INIT8 ==> E_INIT_08
    E_INIT9 ==> E_INIT_09
    E_INIT10 ==> E_INIT_10
    E_LIST1 ==> E_LIST_1
    E_LIST2 ==> E_LIST_2
    E_LIST3 ==> E_LIST_3
    E_LIST4 ==> E_LIST_4
    E_LIST5 ==> E_LIST_5
    E_LIST6 ==> E_LIST_6
    E_MODIFY1 ==> E_MODIFY_1
    E_MODIFY2 ==> E_MODIFY_2
    E_MODIFY3 ==> E_MODIFY_3
    E_MODIFY4 ==> E_MODIFY_4
    E_MODIFY5 ==> E_MODIFY_5
    E_FPATHOPEN1 ==> E_FPATHOPEN_1
    E_FPATHOPEN2 ==> E_FPATHOPEN_2
    E_LOG1 ==> E_LOG_1
    E_LOG2 ==> E_LOG_2
    E_LOG3 ==> E_LOG_3
    E_FGETFILE1 ==> E_FGETFILE_1
    E_FGETFILE2 ==> E_FGETFILE_2
    E_FGETFILE3 ==> E_FGETFILE_3
    E_TAN3 ==> E_TAN_3
    E_TAN4 ==> E_TAN_4
    E_COT3 ==> E_COT_3
    E_COT4 ==> E_COT_4
    E_SEC3 ==> E_SEC_3
    E_CSC3 ==> E_CSC_3
    E_TANH3 ==> E_TANH_3
    E_TANH4 ==> E_TANH_4
    E_COTH3 ==> E_COTH_3
    E_COTH4 ==> E_COTH_4
    E_SECH3 ==> E_SECH_3
    E_CSCH3 ==> E_CSCH_3
    E_ASIN3 ==> E_ASIN_3
    E_ACOS3 ==> E_ACOS_3
    E_ASINH3 ==> E_ASINH_3
    E_ACOSH3 ==> E_ACOSH_3
    E_ATAN3 ==> E_ATAN_3
    E_ACOT3 ==> E_ACOT_3
    E_ASEC3 ==> E_ASEC_3
    E_ACSC3 ==> E_ACSC_3
    E_ATANH3 ==> E_ATANH_3
    E_ACOTH3 ==> E_ACOTH_3
    E_ASECH3 ==> E_ASECH_3
    E_ACSCH3 ==> E_ACSCH_3
    E_D2R1 ==> E_D2R_1
    E_D2R2 ==> E_D2R_2
    E_R2D1 ==> E_R2D_1
    E_R2D2 ==> E_R2D_2
    E_G2R1 ==> E_G2R_1
    E_G2R2 ==> E_G2R_2
    E_R2G1 ==> E_R2G_1
    E_R2G2 ==> E_R2G_2
    E_D2G1 ==> E_D2G_1
    E_G2D1 ==> E_G2D_1
    E_D2DMS1 ==> E_D2DMS_1
    E_D2DMS2 ==> E_D2DMS_2
    E_D2DMS3 ==> E_D2DMS_3
    E_D2DMS4 ==> E_D2DMS_4
    E_D2DM1 ==> E_D2DM_1
    E_D2DM2 ==> E_D2DM_2
    E_D2DM3 ==> E_D2DM_3
    E_D2DM4 ==> E_D2DM_4
    E_G2GMS1 ==> E_G2GMS_1
    E_G2GMS2 ==> E_G2GMS_2
    E_G2GMS3 ==> E_G2GMS_3
    E_G2GMS4 ==> E_G2GMS_4
    E_G2GM1 ==> E_G2GM_1
    E_G2GM2 ==> E_G2GM_2
    E_G2GM3 ==> E_G2GM_3
    E_G2GM4 ==> E_G2GM_4
    E_H2HMS1 ==> E_H2HMS_1
    E_H2HMS2 ==> E_H2HMS_2
    E_H2HMS3 ==> E_H2HMS_3
    E_H2HMS4 ==> E_H2HMS_4
    E_H2HM1 ==> E_H2HM_1
    E_H2HM2 ==> E_H2HM_2
    E_H2HM3 ==> E_H2HM_3
    E_H2HM4 ==> E_H2HM_4
    E_DMS2D1 ==> E_DMS2D_1
    E_DMS2D2 ==> E_DMS2D_2
    E_DM2D1 ==> E_DM2D_1
    E_DM2D2 ==> E_DM2D_2
    E_GMS2G1 ==> E_GMS2G_1
    E_GMS2G2 ==> E_GMS2G_2
    E_GM2G1 ==> E_GM2G_1
    E_GM2G2 ==> E_GM2G_2
    E_HMS2H1 ==> E_HMS2H_1
    E_HMS2H2 ==> E_HMS2H_2
    E_HM2H1 ==> E_HM2H_1
    E_HM2H2 ==> E_HM2H_2
    E_VERSIN1 ==> E_VERSIN_1
    E_VERSIN2 ==> E_VERSIN_2
    E_VERSIN3 ==> E_VERSIN_3
    E_AVERSIN1 ==> E_AVERSIN_1
    E_AVERSIN2 ==> E_AVERSIN_2
    E_AVERSIN3 ==> E_AVERSIN_3
    E_COVERSIN1 ==> E_COVERSIN_1
    E_COVERSIN2 ==> E_COVERSIN_2
    E_COVERSIN3 ==> E_COVERSIN_3
    E_ACOVERSIN1 ==> E_ACOVERSIN_1
    E_ACOVERSIN2 ==> E_ACOVERSIN_2
    E_ACOVERSIN3 ==> E_ACOVERSIN_3
    E_VERCOS1 ==> E_VERCOS_1
    E_VERCOS2 ==> E_VERCOS_2
    E_VERCOS3 ==> E_VERCOS_3
    E_AVERCOS1 ==> E_AVERCOS_1
    E_AVERCOS2 ==> E_AVERCOS_2
    E_AVERCOS3 ==> E_AVERCOS_3
    E_COVERCOS1 ==> E_COVERCOS_1
    E_COVERCOS2 ==> E_COVERCOS_2
    E_COVERCOS3 ==> E_COVERCOS_3
    E_ACOVERCOS1 ==> E_ACOVERCOS_1
    E_ACOVERCOS2 ==> E_ACOVERCOS_2
    E_ACOVERCOS3 ==> E_ACOVERCOS_3
    E_TAN5 ==> E_TAN_5
    E_COT5 ==> E_COT_5
    E_COT6 ==> E_COT_6
    E_SEC5 ==> E_SEC_5
    E_CSC5 ==> E_CSC_5
    E_CSC6 ==> E_CSC_6
2023-09-19 18:34:21 -07:00

2166 lines
39 KiB
C

/*
* zmath - extended precision integral arithmetic primitives
*
* Copyright (C) 1999-2007,2021-2023 David I. Bell, Landon Curt Noll and Ernest Bowen
*
* Primary author: David I. Bell
*
* Calc is open software; you can redistribute it and/or modify it under
* the terms of the version 2.1 of the GNU Lesser General Public License
* as published by the Free Software Foundation.
*
* Calc is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
* or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General
* Public License for more details.
*
* A copy of version 2.1 of the GNU Lesser General Public License is
* distributed with calc under the filename COPYING-LGPL. You should have
* received a copy with calc; if not, write to Free Software Foundation, Inc.
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
* Under source code control: 1990/02/15 01:48:28
* File existed as early as: before 1990
*
* Share and enjoy! :-) http://www.isthe.com/chongo/tech/comp/calc/
*/
#include <stdio.h>
#include "int.h"
#include "alloc.h"
#include "zmath.h"
#include "errtbl.h"
#include "banned.h" /* include after system header <> includes */
HALF _zeroval_[] = { 0 };
ZVALUE _zero_ = { _zeroval_, 1, 0};
HALF _oneval_[] = { 1 };
ZVALUE _one_ = { _oneval_, 1, 0 };
ZVALUE _neg_one_ = { _oneval_, 1, 1 };
HALF _twoval_[] = { 2 };
ZVALUE _two_ = { _twoval_, 1, 0 };
HALF _tenval_[] = { 10 };
ZVALUE _ten_ = { _tenval_, 1, 0 };
HALF _sqbaseval_[] = { 0, 1 };
ZVALUE _sqbase_ = { _sqbaseval_, 2, 0 };
HALF _pow4baseval_[] = { 0, 0, 1 };
ZVALUE _pow4base_ = { _pow4baseval_, 4, 0 };
HALF _pow8baseval_[] = { 0, 0, 0, 0, 1 };
ZVALUE _pow8base_ = { _pow8baseval_, 4, 0 };
/*
* 2^64 as a ZVALUE
*/
#if BASEB == 32
ZVALUE _b32_ = { _sqbaseval_, 2, 0 };
ZVALUE _b64_ = { _pow4baseval_, 3, 0 };
#elif BASEB == 16
ZVALUE _b32_ = { _pow4baseval_, 3, 0 };
ZVALUE _b64_ = { _pow8baseval_, 5, 0 };
#else
-=@=- BASEB not 16 or 32 -=@=-
#endif
/*
* ZVALUE - values that should not be freed
*/
HALF *half_tbl[] = {
_zeroval_,
_oneval_,
_twoval_,
_tenval_,
_sqbaseval_,
_pow4baseval_,
_pow8baseval_,
NULL /* must be last */
};
/*
* highhalf[i] - masks off the upper i bits of a HALF
* rhighhalf[i] - masks off the upper BASEB-i bits of a HALF
* lowhalf[i] - masks off the lower i bits of a HALF
* rlowhalf[i] - masks off the lower BASEB-i bits of a HALF
* bitmask[i] - (1 << i) for 0 <= i <= BASEB*2
*
* NOTE: In all cases 0 <= i <= BASEB
*/
HALF highhalf[BASEB+1] = {
#if BASEB == 32
0x00000000,
0x80000000, 0xC0000000, 0xE0000000, 0xF0000000,
0xF8000000, 0xFC000000, 0xFE000000, 0xFF000000,
0xFF800000, 0xFFC00000, 0xFFE00000, 0xFFF00000,
0xFFF80000, 0xFFFC0000, 0xFFFE0000, 0xFFFF0000,
0xFFFF8000, 0xFFFFC000, 0xFFFFE000, 0xFFFFF000,
0xFFFFF800, 0xFFFFFC00, 0xFFFFFE00, 0xFFFFFF00,
0xFFFFFF80, 0xFFFFFFC0, 0xFFFFFFE0, 0xFFFFFFF0,
0xFFFFFFF8, 0xFFFFFFFC, 0xFFFFFFFE, 0xFFFFFFFF
#elif BASEB == 16
0x0000,
0x8000, 0xC000, 0xE000, 0xF000,
0xF800, 0xFC00, 0xFE00, 0xFF00,
0xFF80, 0xFFC0, 0xFFE0, 0xFFF0,
0xFFF8, 0xFFFC, 0xFFFE, 0xFFFF
#else
-=@=- BASEB not 16 or 32 -=@=-
#endif
};
HALF rhighhalf[BASEB+1] = {
#if BASEB == 32
0xFFFFFFFF, 0xFFFFFFFE, 0xFFFFFFFC, 0xFFFFFFF8,
0xFFFFFFF0, 0xFFFFFFE0, 0xFFFFFFC0, 0xFFFFFF80,
0xFFFFFF00, 0xFFFFFE00, 0xFFFFFC00, 0xFFFFF800,
0xFFFFF000, 0xFFFFE000, 0xFFFFC000, 0xFFFF8000,
0xFFFF0000, 0xFFFE0000, 0xFFFC0000, 0xFFF80000,
0xFFF00000, 0xFFE00000, 0xFFC00000, 0xFF800000,
0xFF000000, 0xFE000000, 0xFC000000, 0xF8000000,
0xF0000000, 0xE0000000, 0xC0000000, 0x80000000,
0x00000000
#elif BASEB == 16
0xFFFF, 0xFFFE, 0xFFFC, 0xFFF8,
0xFFF0, 0xFFE0, 0xFFC0, 0xFF80,
0xFF00, 0xFE00, 0xFC00, 0xF800,
0xF000, 0xE000, 0xC000, 0x8000,
0x0000
#else
-=@=- BASEB not 16 or 32 -=@=-
#endif
};
HALF lowhalf[BASEB+1] = {
0x0,
0x1, 0x3, 0x7, 0xF,
0x1F, 0x3F, 0x7F, 0xFF,
0x1FF, 0x3FF, 0x7FF, 0xFFF,
0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF
#if BASEB == 32
,0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF,
0x1FFFFF, 0x3FFFFF, 0x7FFFFF, 0xFFFFFF,
0x1FFFFFF, 0x3FFFFFF, 0x7FFFFFF, 0xFFFFFFF,
0x1FFFFFFF, 0x3FFFFFFF, 0x7FFFFFFF, 0xFFFFFFFF
#endif
};
HALF rlowhalf[BASEB+1] = {
#if BASEB == 32
0xFFFFFFFF, 0x7FFFFFFF, 0x3FFFFFFF, 0x1FFFFFFF,
0xFFFFFFF, 0x7FFFFFF, 0x3FFFFFF, 0x1FFFFFF,
0xFFFFFF, 0x7FFFFF, 0x3FFFFF, 0x1FFFFF,
0xFFFFF, 0x7FFFF, 0x3FFFF, 0x1FFFF,
#endif
0xFFFF, 0x7FFF, 0x3FFF, 0x1FFF,
0xFFF, 0x7FF, 0x3FF, 0x1FF,
0xFF, 0x7F, 0x3F, 0x1F,
0xF, 0x7, 0x3, 0x1,
0x0
};
HALF bitmask[(2*BASEB)+1] = {
#if BASEB == 32
0x00000001, 0x00000002, 0x00000004, 0x00000008,
0x00000010, 0x00000020, 0x00000040, 0x00000080,
0x00000100, 0x00000200, 0x00000400, 0x00000800,
0x00001000, 0x00002000, 0x00004000, 0x00008000,
0x00010000, 0x00020000, 0x00040000, 0x00080000,
0x00100000, 0x00200000, 0x00400000, 0x00800000,
0x01000000, 0x02000000, 0x04000000, 0x08000000,
0x10000000, 0x20000000, 0x40000000, 0x80000000,
0x00000001, 0x00000002, 0x00000004, 0x00000008,
0x00000010, 0x00000020, 0x00000040, 0x00000080,
0x00000100, 0x00000200, 0x00000400, 0x00000800,
0x00001000, 0x00002000, 0x00004000, 0x00008000,
0x00010000, 0x00020000, 0x00040000, 0x00080000,
0x00100000, 0x00200000, 0x00400000, 0x00800000,
0x01000000, 0x02000000, 0x04000000, 0x08000000,
0x10000000, 0x20000000, 0x40000000, 0x80000000,
0x00000001
#elif BASEB == 16
0x0001, 0x0002, 0x0004, 0x0008,
0x0010, 0x0020, 0x0040, 0x0080,
0x0100, 0x0200, 0x0400, 0x0800,
0x1000, 0x2000, 0x4000, 0x8000,
0x0001, 0x0002, 0x0004, 0x0008,
0x0010, 0x0020, 0x0040, 0x0080,
0x0100, 0x0200, 0x0400, 0x0800,
0x1000, 0x2000, 0x4000, 0x8000,
0x0001
#else
-=@=- BASEB not 16 or 32 -=@=-
#endif
}; /* actual rotation thru 8 cycles */
bool _math_abort_; /* nonzero to abort calculations */
/*
* popcnt - popcnt[x] number of 1 bits in 0 <= x < 256
*/
char popcnt[256] = {
0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8
};
#ifdef ALLOCTEST
STATIC long nalloc = 0;
STATIC long nfree = 0;
#endif
HALF *
alloc(LEN len)
{
HALF *hp;
if (_math_abort_) {
math_error("Calculation aborted");
not_reached();
}
hp = (HALF *) malloc((len+1) * sizeof(HALF));
if (hp == 0) {
math_error("Not enough memory");
not_reached();
}
#ifdef ALLOCTEST
++nalloc;
#endif
return hp;
}
/*
* is_const - determine if a HALF array is an pre-allocated array
*
* given:
* h pointer to the beginning of the HALF array
*
* returns:
* true - h is found in the half_tbl array
* false - is is not found in the half_tbl array
*/
int
is_const(HALF *h)
{
HALF **h_p; /* half_tbl array pointer */
/* firewall */
if (h == NULL) {
math_error("%s: h NULL", __func__);
not_reached();
}
/* search the half_tbl for h */
for (h_p = &half_tbl[0]; *h_p != NULL; ++h_p) {
if (h == *h_p) {
return true; /* found in the half_tbl array */
}
}
/* not found in the half_tbl array */
return false;
}
#ifdef ALLOCTEST
void
freeh(HALF *h)
{
/* firewall */
if (h == NULL) {
math_error("%s: h NULL", __func__);
not_reached();
}
/* free h is not a constant */
if (!is_const(h)) {
free(h);
++nfree;
}
}
void
allocStat(void)
{
fprintf(stderr, "nalloc: %ld nfree: %ld kept: %ld\n",
nalloc, nfree, nalloc - nfree);
}
#endif
/*
* Convert a normal integer to a number.
*/
void
itoz(long i, ZVALUE *res)
{
long diddle, len;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
res->len = 1;
res->sign = 0;
diddle = 0;
if (i == 0) {
res->v = _zeroval_;
return;
}
if (i < 0) {
res->sign = 1;
i = -i;
if (i < 0) { /* fix most negative number */
diddle = 1;
i--;
}
}
if (i == 1) {
res->v = _oneval_;
return;
}
len = 1 + (((FULL) i) >= BASE);
res->len = (LEN)len;
res->v = alloc((LEN)len);
res->v[0] = (HALF) (i + diddle);
if (len == 2)
res->v[1] = (HALF) (i / BASE);
}
/*
* Convert a number to a normal integer, as far as possible.
* If the number is out of range, the largest number is returned.
*/
long
ztoi(ZVALUE z)
{
long i;
if (zgtmaxlong(z)) {
i = MAXLONG;
return (z.sign ? -i : i);
}
i = ztolong(z);
return (z.sign ? -i : i);
}
/*
* Convert a normal unsigned integer to a number.
*/
void
utoz(FULL i, ZVALUE *res)
{
long len;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
res->len = 1;
res->sign = 0;
if (i == 0) {
res->v = _zeroval_;
return;
}
if (i == 1) {
res->v = _oneval_;
return;
}
len = 1 + (((FULL) i) >= BASE);
res->len = (LEN)len;
res->v = alloc((LEN)len);
res->v[0] = (HALF)i;
if (len == 2)
res->v[1] = (HALF) (i / BASE);
}
/*
* Convert a normal signed integer to a number.
*/
void
stoz(SFULL i, ZVALUE *res)
{
long diddle, len;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
res->len = 1;
res->sign = 0;
diddle = 0;
if (i == 0) {
res->v = _zeroval_;
return;
}
if (i < 0) {
res->sign = 1;
i = -i;
if (i < 0) { /* fix most negative number */
diddle = 1;
i--;
}
}
if (i == 1) {
res->v = _oneval_;
return;
}
len = 1 + (((FULL) i) >= BASE);
res->len = (LEN)len;
res->v = alloc((LEN)len);
res->v[0] = (HALF) (i + diddle);
if (len == 2)
res->v[1] = (HALF) (i / BASE);
}
/*
* Convert a number to a unsigned normal integer, as far as possible.
* If the number is out of range, the largest number is returned.
* The absolute value of z is converted.
*/
FULL
ztou(ZVALUE z)
{
if (z.len > 2) {
return MAXUFULL;
}
return ztofull(z);
}
/*
* Convert a number to a signed normal integer, as far as possible.
*
* If the number is too large to fit into an integer, than the largest
* positive or largest negative integer is returned depending on the sign.
*/
SFULL
ztos(ZVALUE z)
{
FULL val; /* absolute value of the return value */
/* negative value processing */
if (z.sign) {
if (z.len > 2) {
return MINSFULL;
}
val = ztofull(z);
if (val > TOPFULL) {
return MINSFULL;
}
return (SFULL)(-val);
}
/* positive value processing */
if (z.len > 2) {
return (SFULL)MAXFULL;
}
val = ztofull(z);
if (val > MAXFULL) {
return (SFULL)MAXFULL;
}
return (SFULL)val;
}
/*
* Make a copy of an integer value
*/
void
zcopy(ZVALUE z, ZVALUE *res)
{
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
/* make copy */
res->sign = z.sign;
res->len = z.len;
if (zisabsleone(z)) { /* zero or plus or minus one are easy */
res->v = (z.v[0] ? _oneval_ : _zeroval_);
return;
}
res->v = alloc(z.len);
zcopyval(z, *res);
}
/*
* Add together two integers.
*/
void
zadd(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
ZVALUE dest;
HALF *p1, *p2, *pd;
long len;
FULL carry;
SIUNION sival;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (z1.sign && !z2.sign) {
z1.sign = 0;
zsub(z2, z1, res);
return;
}
if (z2.sign && !z1.sign) {
z2.sign = 0;
zsub(z1, z2, res);
return;
}
if (z2.len > z1.len) {
pd = z1.v; z1.v = z2.v; z2.v = pd;
len = z1.len; z1.len = z2.len; z2.len = (LEN)len;
}
dest.len = z1.len + 1;
dest.v = alloc(dest.len);
dest.sign = z1.sign;
carry = 0;
pd = dest.v;
p1 = z1.v;
p2 = z2.v;
len = z2.len;
while (len--) {
sival.ivalue = ((FULL) *p1++) + ((FULL) *p2++) + carry;
/* ignore Saber-C warning #112 - get ushort from uint */
/* OK to ignore on name zadd`sival */
*pd++ = sival.silow;
carry = sival.sihigh;
}
len = z1.len - z2.len;
while (len--) {
sival.ivalue = ((FULL) *p1++) + carry;
*pd++ = sival.silow;
carry = sival.sihigh;
}
*pd = (HALF)carry;
zquicktrim(dest);
*res = dest;
}
/*
* Subtract two integers.
*/
void
zsub(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
register HALF *h1, *h2, *hd;
long len1, len2;
FULL carry;
SIUNION sival;
ZVALUE dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (z1.sign != z2.sign) {
z2.sign = z1.sign;
zadd(z1, z2, res);
return;
}
len1 = z1.len;
len2 = z2.len;
if (len1 == len2) {
h1 = z1.v + len1;
h2 = z2.v + len2;
while ((len1 > 0) && ((FULL)*--h1 == (FULL)*--h2)) {
len1--;
}
if (len1 == 0) {
*res = _zero_;
return;
}
len2 = len1;
carry = ((FULL)*h1 < (FULL)*h2);
} else {
carry = (len1 < len2);
}
dest.sign = z1.sign;
h1 = z1.v;
h2 = z2.v;
if (carry) {
carry = len1;
len1 = len2;
len2 = (long)carry;
h1 = z2.v;
h2 = z1.v;
dest.sign = !dest.sign;
}
hd = alloc((LEN)len1);
dest.v = hd;
dest.len = (LEN)len1;
len1 -= len2;
carry = 0;
while (--len2 >= 0) {
/* ignore Saber-C warning #112 - get ushort from uint */
/* OK to ignore on name zsub`sival */
sival.ivalue = (BASE1 - ((FULL) *h1++)) + *h2++ + carry;
*hd++ = (HALF)(BASE1 - sival.silow);
carry = sival.sihigh;
}
while (--len1 >= 0) {
sival.ivalue = (BASE1 - ((FULL) *h1++)) + carry;
*hd++ = (HALF)(BASE1 - sival.silow);
carry = sival.sihigh;
}
if (hd[-1] == 0)
ztrim(&dest);
*res = dest;
}
/*
* Multiply an integer by a small number.
*/
void
zmuli(ZVALUE z, long n, ZVALUE *res)
{
register HALF *h1, *sd;
FULL low;
FULL high;
FULL carry;
long len;
SIUNION sival;
ZVALUE dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if ((n == 0) || ziszero(z)) {
*res = _zero_;
return;
}
if (n < 0) {
n = -n;
z.sign = !z.sign;
}
if (n == 1) {
zcopy(z, res);
return;
}
#if LONG_BITS > BASEB
low = ((FULL) n) & BASE1;
high = ((FULL) n) >> BASEB;
#else
low = (FULL)n;
high = 0;
#endif
dest.len = z.len + 2;
dest.v = alloc(dest.len);
dest.sign = z.sign;
/*
* Multiply by the low digit.
*/
h1 = z.v;
sd = dest.v;
len = z.len;
carry = 0;
while (len--) {
/* ignore Saber-C warning #112 - get ushort from uint */
/* OK to ignore on name zmuli`sival */
sival.ivalue = ((FULL) *h1++) * low + carry;
*sd++ = sival.silow;
carry = sival.sihigh;
}
*sd = (HALF)carry;
/*
* If there was only one digit, then we are all done except
* for trimming the number if there was no last carry.
*/
if (high == 0) {
dest.len--;
if (carry == 0)
dest.len--;
*res = dest;
return;
}
/*
* Need to multiply by the high digit and add it into the
* previous value. Clear the final word of rubbish first.
*/
*(++sd) = 0;
h1 = z.v;
sd = dest.v + 1;
len = z.len;
carry = 0;
while (len--) {
sival.ivalue = ((FULL) *h1++) * high + ((FULL) *sd) + carry;
*sd++ = sival.silow;
carry = sival.sihigh;
}
*sd = (HALF)carry;
zquicktrim(dest);
*res = dest;
}
/*
* Divide two numbers by their greatest common divisor.
* This is useful for reducing the numerator and denominator of
* a fraction to its lowest terms.
*/
void
zreduce(ZVALUE z1, ZVALUE z2, ZVALUE *z1res, ZVALUE *z2res)
{
ZVALUE tmp;
/* firewall */
if (z1res == NULL) {
math_error("%s: z1res NULL", __func__);
not_reached();
}
if (z2res == NULL) {
math_error("%s: z2res NULL", __func__);
not_reached();
}
if (zisabsleone(z1) || zisabsleone(z2))
tmp = _one_;
else
zgcd(z1, z2, &tmp);
if (zisunit(tmp)) {
zcopy(z1, z1res);
zcopy(z2, z2res);
} else {
zequo(z1, tmp, z1res);
zequo(z2, tmp, z2res);
}
zfree(tmp);
}
/*
* Compute the quotient and remainder for division of an integer by an
* integer, rounding criteria determined by rnd. Returns the sign of
* the remainder.
*/
long
zdiv(ZVALUE z1, ZVALUE z2, ZVALUE *quo, ZVALUE *rem, long rnd)
{
register HALF *a, *b;
HALF s, u;
HALF *A, *B, *a1, *b0;
FULL f, g, h, x;
bool adjust, onebit;
LEN m, n, len, i, p, j1, j2, k;
long t, val;
/* firewall */
if (quo == NULL) {
math_error("%s: quo NULL", __func__);
not_reached();
}
if (rem == NULL) {
math_error("%s: rem NULL", __func__);
not_reached();
}
if (ziszero(z2)) {
math_error("Division by zero in zdiv");
not_reached();
}
m = z1.len;
n = z2.len;
B = z2.v;
s = 0;
if (m < n) {
A = alloc(n + 1);
memcpy(A, z1.v, m * sizeof(HALF));
for (i = m; i <= n; i++)
A[i] = 0;
a1 = A + n;
len = 1;
goto done;
}
A = alloc(m + 2);
memcpy(A, z1.v, m * sizeof(HALF));
A[m] = 0;
A[m + 1] = 0;
len = m - n + 1; /* quotient length will be len or len +/- 1 */
a1 = A + n; /* start of digits for quotient */
b0 = B;
p = n;
while (!*b0) { /* b0: working start for divisor */
++b0;
--p;
}
if (p == 1) {
u = *b0;
if (u == 1) {
for (; m >= n; m--)
A[m] = A[m - 1];
A[m] = 0;
goto done;
}
f = 0;
a = A + m;
i = len;
while (i--) {
f = f << BASEB | *--a;
a[1] = (HALF)(f / u);
f = f % u;
}
*a = (HALF)f;
m = n;
goto done;
}
f = B[n - 1];
k = 1;
while (f >>= 1)
k++; /* k: number of bits in top divisor digit */
j1 = BASEB - k;
j2 = BASEB + j1;
h = j1 ? ((FULL) B[n - 1] << j1 | B[n - 2] >> k) : B[n-1];
onebit = (bool)((B[n - 2] >> (k - 1)) & 1);
m++;
while (m > n) {
m--;
f = (FULL) A[m] << j2 | (FULL) A[m - 1] << j1;
if (j1) f |= A[m - 2] >> k;
if (s) f = ~f;
x = f / h;
if (x) {
if (onebit && x > TOPHALF + f % h)
x--;
a = A + m - p;
b = b0;
u = 0;
i = p;
if (s) {
while (i--) {
f = (FULL) *a + u + x * *b++;
*a++ = (HALF) f;
u = (HALF) (f >> BASEB);
}
s = *a + u;
A[m] = (HALF) (~x + !s);
} else {
while (i--) {
f = (FULL) *a - u - x * *b++;
*a++ = (HALF) f;
u = -(HALF)(f >> BASEB);
}
s = *a - u;
A[m] = (HALF)(x + s);
}
}
else
A[m] = s;
}
done: while (m > 0 && A[m - 1] == 0)
m--;
if (m == 0 && s == 0) {
*rem = _zero_;
val = 0;
if (a1[len - 1] == 0)
len--;
if (len == 0) {
*quo = _zero_;
} else {
quo->len = len;
quo->v = alloc(len);
memcpy(quo->v, a1, len * sizeof(HALF));
quo->sign = z1.sign ^ z2.sign;
}
freeh(A);
return val;
}
if (rnd & 8)
adjust = (((*a1 ^ rnd) & 1) ? true : false);
else
adjust = (((rnd & 1) ^ z1.sign ^ z2.sign) ? true : false);
if (rnd & 2)
adjust ^= z1.sign ^ z2.sign;
if (rnd & 4)
adjust ^= z2.sign;
if (rnd & 16) { /* round-off case */
a = A + n;
b = B + n;
i = n + 1;
f = g = 0;
t = -1;
if (s) {
while (--i > 0) {
g = (FULL) *--a + (*--b >> 1 | f);
f = *b & 1 ? TOPHALF : 0;
if (g != BASE1)
break;
}
if (g == BASE && f == 0) {
while ((--i > 0) && ((*--a | *--b) == 0));
t = (i > 0);
} else if (g >= BASE) {
t = 1;
}
} else {
while (--i > 0) {
g = (FULL) *--a - (*--b >> 1 | f);
f = *b & 1 ? TOPHALF : 0;
if (g != 0)
break;
}
if (g > 0 && g < BASE)
t = 1;
else if (g == 0 && f == 0)
t = 0;
}
if (t)
adjust = (t > 0);
}
if (adjust) {
i = len;
a = a1;
while (i > 0 && *a == BASE1) {
i--;
*a++ = 0;
}
(*a)++;
if (i == 0)
len++;
}
if (s && adjust) {
i = 0;
while (A[i] == 0)
i++;
A[i] = -A[i];
while (++i < n)
A[i] = ~A[i];
m = n;
while (A[m - 1] == 0)
m--;
}
if (!s && adjust) {
a = A;
b = B;
i = n;
u = 0;
while (i--) {
f = (FULL) *b++ - *a - (FULL) u;
*a++ = (HALF) f;
u = -(HALF)(f >> BASEB);
}
m = n;
while (A[m - 1] == 0)
m--;
}
if (s && !adjust) {
a = A;
b = B;
i = n;
f = 0;
while (i--) {
f = (FULL) *b++ + *a + (f >> BASEB);
*a++ = (HALF) f;
}
m = n;
while (A[m-1] == 0)
m--;
}
rem->len = m;
rem->v = alloc(m);
memcpy(rem->v, A, m * sizeof(HALF));
rem->sign = z1.sign ^ adjust;
val = rem->sign ? -1 : 1;
if (a1[len - 1] == 0)
len--;
if (len == 0) {
*quo = _zero_;
} else {
quo->len = len;
quo->v = alloc(len);
memcpy(quo->v, a1, len * sizeof(HALF));
quo->sign = z1.sign ^ z2.sign;
}
freeh(A);
return val;
}
/*
* Compute and store at a specified location the integer quotient
* of two integers, the type of rounding being determined by rnd.
* Returns the sign of z1/z2 - calculated quotient.
*/
long
zquo(ZVALUE z1, ZVALUE z2, ZVALUE *res, long rnd)
{
ZVALUE tmp;
long val;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
val = zdiv(z1, z2, res, &tmp, rnd);
if (z2.sign)
val = -val;
zfree(tmp);
return val;
}
/*
* Compute and store at a specified location the remainder for
* division of an integer by an integer, the type of rounding
* used being determined by rnd. Returns the sign of the remainder.
*/
long
zmod(ZVALUE z1, ZVALUE z2, ZVALUE *res, long rnd)
{
ZVALUE tmp;
long val;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
val = zdiv(z1, z2, &tmp, res, rnd);
zfree(tmp);
return val;
}
/*
* Computes the quotient z1/z2 on the assumption that this is exact.
* There is no thorough check on the exactness of the division
* so this should not be called unless z1/z2 is an integer.
*/
void
zequo(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
LEN i, j, k, len, m, n, o, p;
HALF *a, *a0, *A, *b, *B, u, v, w, x;
FULL f, g;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (ziszero(z1)) {
*res = _zero_;
return;
}
if (ziszero(z2)) {
math_error("Division by zero");
not_reached();
}
if (zisone(z2)) {
zcopy(z1, res);
return;
}
if (zhighbit(z1) < zhighbit(z2)) {
math_error("Bad call to zequo");
not_reached();
}
B = z2.v;
o = 0;
while (!*B) {
++B;
++o;
}
m = z1.len - o;
n = z2.len - o;
len = m - n + 1; /* Maximum length of quotient */
v = *B;
A = alloc(len+1);
memcpy(A, z1.v + o, len * sizeof(HALF));
A[len] = 0;
if (n == 1) {
if (v > 1) {
a = A + len;
i = len;
f = 0;
while (i--) {
f = f << BASEB | *--a;
*a = (HALF)(f / v);
f %= v;
}
}
} else {
k = 0;
while (!(v & 1)) {
k++;
v >>= 1;
}
j = BASEB - k;
if (n > 1 && k > 0)
v |= B[1] << j;
u = v - 1;
w = x = 1;
while (u) { /* To find w = inverse of v modulo BASE */
do {
v <<= 1;
x <<= 1;
}
while (!(u & x));
u += v;
w |= x;
}
a0 = A;
p = len;
while (p > 1) {
if (!*a0) {
while (!*++a0 && p > 1)
p--;
--a0;
}
if (p == 1)
break;
x = k ? w * (*a0 >> k | a0[1] << j) : w * *a0;
g = x;
if (x) {
a = a0;
b = B;
u = 0;
i = n;
if (i > p)
i = p;
while (i--) {
f = (FULL) *a - g * *b++ - (FULL) u;
*a++ = (HALF)f;
u = -(HALF)(f >> BASEB);
}
if (u && p > n) {
i = p - n;
while (u && i--) {
f = (FULL) *a - u;
*a++ = (HALF) f;
u = -(HALF)(f >> BASEB);
}
}
}
*a0++ = x;
p--;
}
if (k == 0) {
*a0 = w * *a0;
} else {
u = (HALF)(w * *a0) >> k;
x = (HALF)(((FULL) z1.v[z1.len - 1] << BASEB
| z1.v[z1.len - 2])
/((FULL) B[n-1] << BASEB | B[n-2]));
if ((x ^ u) & 1) x--;
*a0 = x;
}
}
if (!A[len - 1]) len--;
res->v = A;
res->len = len;
res->sign = z1.sign != z2.sign;
}
/*
* Return the quotient and remainder of an integer divided by a small
* number. A nonzero remainder is only meaningful when both numbers
* are positive.
*/
long
zdivi(ZVALUE z, long n, ZVALUE *res)
{
HALF *h1, *sd;
FULL val;
HALF divval[2];
ZVALUE div;
ZVALUE dest;
LEN len;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (n == 0) {
math_error("Division by zero");
not_reached();
}
if (ziszero(z)) {
*res = _zero_;
return 0;
}
if (n < 0) {
n = -n;
z.sign = !z.sign;
}
if (n == 1) {
zcopy(z, res);
return 0;
}
/*
* If the division is by a large number, then call the normal
* divide routine.
*/
if (n & ~BASE1) {
div.sign = 0;
div.v = divval;
divval[0] = (HALF) n;
#if LONG_BITS > BASEB
divval[1] = (HALF)(((FULL) n) >> BASEB);
div.len = 2;
#else
div.len = 1;
#endif
zdiv(z, div, res, &dest, 0);
n = ztolong(dest);
zfree(dest);
return n;
}
/*
* Division is by a small number, so we can be quick about it.
*/
len = z.len;
dest.sign = z.sign;
dest.len = len;
dest.v = alloc(len);
h1 = z.v + len;
sd = dest.v + len;
val = 0;
while (len--) {
val = ((val << BASEB) + ((FULL) *--h1));
*--sd = (HALF)(val / n);
val %= n;
}
zquicktrim(dest);
*res = dest;
return (long)val;
}
/*
* Calculate the mod of an integer by a small number.
* This is only defined for positive moduli.
*/
long
zmodi(ZVALUE z, long n)
{
register HALF *h1;
FULL val;
HALF divval[2];
ZVALUE div;
ZVALUE temp;
long len;
if (n == 0) {
math_error("Division by zero");
not_reached();
}
if (n < 0) {
math_error("Non-positive modulus");
not_reached();
}
if (ziszero(z) || (n == 1))
return 0;
if (zisone(z))
return 1;
/*
* If the modulus is by a large number, then call the normal
* modulo routine.
*/
if (n & ~BASE1) {
div.sign = 0;
div.v = divval;
divval[0] = (HALF) n;
#if LONG_BITS > BASEB
divval[1] = (HALF)(((FULL) n) >> BASEB);
div.len = 2;
#else
div.len = 1;
#endif
zmod(z, div, &temp, 0);
n = ztolong(temp);
zfree(temp);
return n;
}
/*
* The modulus is by a small number, so we can do this quickly.
*/
len = z.len;
h1 = z.v + len;
val = 0;
while (len-- > 0)
val = ((val << BASEB) + ((FULL) *--h1)) % n;
if (val && z.sign)
val = n - val;
return (long)val;
}
/*
* Return whether or not one number exactly divides another one.
* Returns true if division occurs with no remainder.
* z1 is the number to be divided by z2.
*/
bool
zdivides(ZVALUE z1, ZVALUE z2)
{
LEN i, j, k, m, n;
HALF u, v, w, x;
HALF *a, *a0, *A, *b, *B, *c, *d;
FULL f;
bool ans;
if (zisunit(z2) || ziszero(z1)) return true;
if (ziszero(z2)) return false;
m = z1.len;
n = z2.len;
if (m < n) return false;
c = z1.v;
d = z2.v;
while (!*d) {
if (*c++) return false;
d++;
m--;
n--;
}
j = 0;
u = *c;
v = *d;
while (!(v & 1)) { /* Counting and checking zero bits */
if (u & 1) return false;
u >>= 1;
v >>= 1;
j++;
}
if (n == 1 && v == 1) return true;
if (!*c) { /* Removing any further zeros */
while(!*++c)
m--;
c--;
}
if (m < n) return false;
if (j) {
B = alloc((LEN)n); /* Array for shifted z2 */
d += n;
b = B + n;
i = n;
f = 0;
while(i--) {
f = f << BASEB | *--d;
*--b = (HALF)(f >> j);
}
if (!B[n - 1]) n--;
}
else B = d;
u = *B;
v = x = 1;
w = 0;
while (x) { /* Finding minv(*B, BASE) */
if (v & x) {
v -= x * u;
w |= x;
}
x <<= 1;
}
A = alloc((LEN)(m + 2)); /* Main working array */
memcpy(A, c, m * sizeof(HALF));
A[m + 1] = 1;
A[m] = 0;
k = m - n + 1; /* Length of presumed quotient */
a0 = A;
while (k--) {
if (*a0) {
x = w * *a0;
a = a0;
b = B;
i = n;
u = 0;
while (i--) {
f = (FULL) *a - (FULL) x * *b++ - u;
*a++ = (HALF)f;
u = -(HALF)(f >> BASEB);
}
f = (FULL) *a - u;
*a++ = (HALF)f;
u = -(HALF)(f >> BASEB);
if (u) {
while (*a == 0) *a++ = BASE1;
(*a)--;
}
}
a0++;
}
ans = false;
if (A[m + 1]) {
a = A + m;
while (--n && !*--a);
if (!n) ans = true;
}
freeh(A);
if (j) freeh(B);
return ans;
}
/*
* Compute the bitwise OR of two integers
*/
void
zor(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
register HALF *sp, *dp;
long len;
ZVALUE bz, lz, dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (z1.len >= z2.len) {
bz = z1;
lz = z2;
} else {
bz = z2;
lz = z1;
}
dest.len = bz.len;
dest.v = alloc(dest.len);
dest.sign = 0;
zcopyval(bz, dest);
len = lz.len;
sp = lz.v;
dp = dest.v;
while (len--)
*dp++ |= *sp++;
*res = dest;
}
/*
* Compute the bitwise AND of two integers
*/
void
zand(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
HALF *h1, *h2, *hd;
LEN len;
ZVALUE dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
len = ((z1.len <= z2.len) ? z1.len : z2.len);
h1 = &z1.v[len-1];
h2 = &z2.v[len-1];
while ((len > 1) && ((*h1 & *h2) == 0)) {
h1--;
h2--;
len--;
}
dest.len = len;
dest.v = alloc(len);
dest.sign = 0;
h1 = z1.v;
h2 = z2.v;
hd = dest.v;
while (len--)
*hd++ = (*h1++ & *h2++);
*res = dest;
}
/*
* Compute the bitwise XOR of two integers.
*/
void
zxor(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
HALF *dp, *h1, *h2;
LEN len, j, k;
ZVALUE dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
h1 = z1.v;
h2 = z2.v;
len = z1.len;
j = z2.len;
if (z1.len < z2.len) {
len = z2.len;
j = z1.len;
h1 = z2.v;
h2 = z1.v;
} else if (z1.len == z2.len) {
while (len > 1 && z1.v[len-1] == z2.v[len-1])
len--;
j = len;
}
k = len - j;
dest.len = len;
dest.v = alloc(len);
dest.sign = 0;
dp = dest.v;
while (j-- > 0)
*dp++ = *h1++ ^ *h2++;
while (k-- > 0)
*dp++ = *h1++;
*res = dest;
}
/*
* Compute the bitwise AND NOT of two integers.
*/
void
zandnot(ZVALUE z1, ZVALUE z2, ZVALUE *res)
{
HALF *dp, *h1, *h2;
LEN len, j, k;
ZVALUE dest;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
len = z1.len;
if (z2.len >= len) {
while (len > 1 && (z1.v[len-1] & ~z2.v[len-1]) == 0)
len--;
j = len;
k = 0;
} else {
j = z2.len;
k = len - z2.len;
}
dest.len = len;
dest.v = alloc(len);
dest.sign = 0;
dp = dest.v;
h1 = z1.v;
h2 = z2.v;
while (j-- > 0)
*dp++ = *h1++ & ~*h2++;
while (k-- > 0)
*dp++ = *h1++;
*res = dest;
}
/*
* Shift a number left (or right) by the specified number of bits.
* Positive shift means to the left. When shifting right, rightmost
* bits are lost. The sign of the number is preserved.
*/
void
zshift(ZVALUE z, long n, ZVALUE *res)
{
ZVALUE ans;
LEN hc; /* number of halfwords shift is by */
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (ziszero(z)) {
*res = _zero_;
return;
}
if (n == 0) {
zcopy(z, res);
return;
}
/*
* If shift value is negative, then shift right.
* Check for large shifts, and handle word-sized shifts quickly.
*/
if (n < 0) {
n = -n;
if ((n < 0) || (n >= (z.len * BASEB))) {
*res = _zero_;
return;
}
hc = (LEN)(n / BASEB);
n %= BASEB;
z.v += hc;
z.len -= hc;
ans.len = z.len;
ans.v = alloc(ans.len);
ans.sign = z.sign;
zcopyval(z, ans);
if (n > 0) {
zshiftr(ans, n);
ztrim(&ans);
}
if (ziszero(ans)) {
zfree(ans);
ans = _zero_;
}
*res = ans;
return;
}
/*
* Shift value is positive, so shift leftwards.
* Check specially for a shift of the value 1, since this is common.
* Also handle word-sized shifts quickly.
*/
if (zisunit(z)) {
zbitvalue(n, res);
res->sign = z.sign;
return;
}
hc = (LEN)(n / BASEB);
n %= BASEB;
ans.len = z.len + hc + 1;
ans.v = alloc(ans.len);
ans.sign = z.sign;
if (hc > 0)
memset((char *) ans.v, 0, hc * sizeof(HALF));
memcpy((char *) (ans.v + hc),
(char *) z.v, z.len * sizeof(HALF));
ans.v[ans.len - 1] = 0;
if (n > 0) {
ans.v += hc;
ans.len -= hc;
zshiftl(ans, n);
ans.v -= hc;
ans.len += hc;
}
ztrim(&ans);
*res = ans;
}
/*
* Return the position of the lowest bit which is set in the binary
* representation of a number (counting from zero). This is the highest
* power of two which evenly divides the number.
*/
long
zlowbit(ZVALUE z)
{
register HALF *zp;
long n;
HALF dataval;
HALF *bitval;
n = 0;
for (zp = z.v; *zp == 0; zp++)
if (++n >= z.len)
return 0;
dataval = *zp;
bitval = bitmask;
/* ignore Saber-C warning #530 about empty while statement */
/* OK to ignore in proc zlowbit */
while ((*(bitval++) & dataval) == 0) {
}
return (n*BASEB)+(bitval-bitmask-1);
}
/*
* Return the position of the highest bit which is set in the binary
* representation of a number (counting from zero). This is the highest power
* of two which is less than or equal to the number (which is assumed nonzero).
*/
LEN
zhighbit(ZVALUE z)
{
HALF dataval;
HALF *bitval;
dataval = z.v[z.len-1];
if (dataval == 0) {
return 0;
}
bitval = bitmask+BASEB;
if (dataval) {
/* ignore Saber-C warning #530 about empty while statement */
/* OK to ignore in proc zhighbit */
while ((*(--bitval) & dataval) == 0) {
}
}
return (z.len*BASEB)+(LEN)(bitval-bitmask-BASEB);
}
/*
* Return whether or not the specified bit number is set in a number.
* Rightmost bit of a number is bit 0.
*/
bool
zisset(ZVALUE z, long n)
{
if ((n < 0) || ((n / BASEB) >= z.len))
return false;
return ((z.v[n / BASEB] & (((HALF) 1) << (n % BASEB))) != 0);
}
/*
* Check whether or not a number has exactly one bit set, and
* thus is an exact power of two. Returns true if so.
*/
bool
zisonebit(ZVALUE z)
{
register HALF *hp;
register LEN len;
if (ziszero(z) || zisneg(z))
return false;
hp = z.v;
len = z.len;
while (len > 4) {
len -= 4;
if (*hp++ || *hp++ || *hp++ || *hp++)
return false;
}
while (--len > 0) {
if (*hp++)
return false;
}
return ((*hp & -*hp) == *hp); /* NEEDS 2'S COMPLEMENT */
}
/*
* Check whether or not a number has all of its bits set below some
* bit position, and thus is one less than an exact power of two.
* Returns true if so.
*/
bool
zisallbits(ZVALUE z)
{
register HALF *hp;
register LEN len;
HALF digit;
if (ziszero(z) || zisneg(z))
return false;
hp = z.v;
len = z.len;
while (len > 4) {
len -= 4;
if ((*hp++ != BASE1) || (*hp++ != BASE1) ||
(*hp++ != BASE1) || (*hp++ != BASE1))
return false;
}
while (--len > 0) {
if (*hp++ != BASE1)
return false;
}
digit = (HALF)(*hp + 1);
return ((digit & -digit) == digit); /* NEEDS 2'S COMPLEMENT */
}
/*
* Return the number whose binary representation contains only one bit which
* is in the specified position (counting from zero). This is equivalent
* to raising two to the given power.
*/
void
zbitvalue(long n, ZVALUE *res)
{
ZVALUE z;
/* firewall */
if (res == NULL) {
math_error("%s: res NULL", __func__);
not_reached();
}
if (n < 0) n = 0;
z.sign = 0;
z.len = (LEN)((n / BASEB) + 1);
z.v = alloc(z.len);
zclearval(z);
z.v[z.len-1] = (((HALF) 1) << (n % BASEB));
*res = z;
}
/*
* Compare a number against zero.
* Returns the sgn function of the number (-1, 0, or 1).
*/
FLAG
ztest(ZVALUE z)
{
register int sign;
register HALF *h;
register long len;
sign = 1;
if (z.sign)
sign = -sign;
h = z.v;
len = z.len;
while (len--)
if (*h++)
return sign;
return 0;
}
/*
* Return the sign of z1 - z2, i.e. 1 if the first integer is greater,
* 0 if they are equal, -1 otherwise.
*/
FLAG
zrel(ZVALUE z1, ZVALUE z2)
{
HALF *h1, *h2;
LEN len;
int sign;
sign = 1;
if (z1.sign < z2.sign)
return 1;
if (z2.sign < z1.sign)
return -1;
if (z2.sign)
sign = -1;
if (z1.len != z2.len)
return (z1.len > z2.len) ? sign : -sign;
len = z1.len;
h1 = z1.v + len;
h2 = z2.v + len;
while (len > 0) {
if (*--h1 != *--h2)
break;
len--;
}
if (len > 0)
return (*h1 > *h2) ? sign : -sign;
return 0;
}
/*
* Return the sign of abs(z1) - abs(z2), i.e. 1 if the first integer
* has greater absolute value, 0 is they have equal absolute value,
* -1 otherwise.
*/
FLAG
zabsrel(ZVALUE z1, ZVALUE z2)
{
HALF *h1, *h2;
LEN len;
if (z1.len != z2.len)
return (z1.len > z2.len) ? 1 : -1;
len = z1.len;
h1 = z1.v + len;
h2 = z2.v + len;
while (len > 0) {
if (*--h1 != *--h2)
break;
len--;
}
if (len > 0)
return (*h1 > *h2) ? 1 : -1;
return 0;
}
/*
* Compare two numbers to see if they are equal or not.
* Returns true if they differ.
*/
bool
zcmp(ZVALUE z1, ZVALUE z2)
{
register HALF *h1, *h2;
register long len;
if ((z1.sign != z2.sign) || (z1.len != z2.len) || (*z1.v != *z2.v))
return true;
len = z1.len;
h1 = z1.v;
h2 = z2.v;
while (--len > 0) {
if (*++h1 != *++h2)
return true;
}
return false;
}
/*
* Utility to calculate the gcd of two FULL integers.
*/
FULL
uugcd(FULL f1, FULL f2)
{
FULL temp;
while (f1) {
temp = f2 % f1;
f2 = f1;
f1 = temp;
}
return (FULL) f2;
}
/*
* Utility to calculate the gcd of two small integers.
*/
long
iigcd(long i1, long i2)
{
FULL f1, f2, temp;
f1 = (FULL) ((i1 >= 0) ? i1 : -i1);
f2 = (FULL) ((i2 >= 0) ? i2 : -i2);
while (f1) {
temp = f2 % f1;
f2 = f1;
f1 = temp;
}
return (long) f2;
}
void
ztrim(ZVALUE *z)
{
HALF *h;
LEN len;
/* firewall */
if (z == NULL) {
math_error("%s: z NULL", __func__);
not_reached();
}
h = z->v + z->len - 1;
len = z->len;
while (*h == 0 && len > 1) {
--h;
--len;
}
z->len = len;
}
/*
* Utility routine to shift right.
*
* NOTE: The ZVALUE length is not adjusted instead, the value is
* zero padded from the left. One may need to call ztrim()
* or use zshift() instead.
*/
void
zshiftr(ZVALUE z, long n)
{
register HALF *h, *lim;
FULL mask, maskt;
long len;
if (n >= BASEB) {
len = n / BASEB;
h = z.v;
lim = z.v + z.len - len;
while (h < lim) {
h[0] = h[len];
++h;
}
n -= BASEB * len;
lim = z.v + z.len;
while (h < lim)
*h++ = 0;
}
if (n) {
len = z.len;
h = z.v + len;
mask = 0;
while (len--) {
maskt = (((FULL) *--h) << (BASEB - n)) & BASE1;
*h = ((*h >> n) | (HALF)mask);
mask = maskt;
}
}
}
/*
* Utility routine to shift left.
*
* NOTE: The ZVALUE length is not adjusted. The bits in the upper
* HALF are simply tossed. You may want to use zshift() instead.
*/
void
zshiftl(ZVALUE z, long n)
{
register HALF *h;
FULL mask, i;
long len;
if (n >= BASEB) {
len = n / BASEB;
h = z.v + z.len - 1;
while (!*h)
--h;
while (h >= z.v) {
h[len] = h[0];
--h;
}
n -= BASEB * len;
while (len)
h[len--] = 0;
}
if (n > 0) {
len = z.len;
h = z.v;
mask = 0;
while (len--) {
i = (((FULL) *h) << n) | mask;
if (i > BASE1) {
mask = i >> BASEB;
i &= BASE1;
} else {
mask = 0;
}
*h = (HALF) i;
++h;
}
}
}
/*
* popcnt - count the number of 0 or 1 bits in an integer
*
* We ignore all 0 bits above the highest bit.
*/
long
zpopcnt(ZVALUE z, int bitval)
{
long cnt = 0; /* number of times found */
HALF h; /* HALF to count */
int i;
/*
* count 1's
*/
if (bitval) {
/*
* count each HALF
*/
for (i=0; i < z.len; ++i) {
/* count each octet */
for (h = z.v[i]; h; h >>= 8) {
cnt += (long)popcnt[h & 0xff];
}
}
/*
* count 0's
*/
} else {
/*
* count each HALF up until the last
*/
for (i=0; i < z.len-1; ++i) {
/* count each octet */
cnt += BASEB;
for (h = z.v[i]; h; h >>= 8) {
cnt -= (long)popcnt[h & 0xff];
}
}
/*
* count the last octet up until the highest 1 bit
*/
for (h = z.v[z.len-1]; h; h>>=1) {
/* count each 0 bit */
if ((h & 0x1) == 0) {
++cnt;
}
}
}
/*
* return count
*/
return cnt;
}