^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 1) /* Software floating-point emulation.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 2) Basic one-word fraction declaration and manipulation.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 3) Copyright (C) 1997,1998,1999 Free Software Foundation, Inc.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 4) This file is part of the GNU C Library.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 5) Contributed by Richard Henderson (rth@cygnus.com),
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 6) Jakub Jelinek (jj@ultra.linux.cz),
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 7) David S. Miller (davem@redhat.com) and
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 8) Peter Maydell (pmaydell@chiark.greenend.org.uk).
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 9)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 10) The GNU C Library is free software; you can redistribute it and/or
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 11) modify it under the terms of the GNU Library General Public License as
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 12) published by the Free Software Foundation; either version 2 of the
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 13) License, or (at your option) any later version.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 14)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 15) The GNU C Library is distributed in the hope that it will be useful,
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 16) but WITHOUT ANY WARRANTY; without even the implied warranty of
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 17) MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 18) Library General Public License for more details.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 19)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 20) You should have received a copy of the GNU Library General Public
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 21) License along with the GNU C Library; see the file COPYING.LIB. If
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 22) not, write to the Free Software Foundation, Inc.,
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 23) 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 24)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 25) #ifndef __MATH_EMU_OP_1_H__
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 26) #define __MATH_EMU_OP_1_H__
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 27)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 28) #define _FP_FRAC_DECL_1(X) _FP_W_TYPE X##_f=0
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 29) #define _FP_FRAC_COPY_1(D,S) (D##_f = S##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 30) #define _FP_FRAC_SET_1(X,I) (X##_f = I)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 31) #define _FP_FRAC_HIGH_1(X) (X##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 32) #define _FP_FRAC_LOW_1(X) (X##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 33) #define _FP_FRAC_WORD_1(X,w) (X##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 34)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 35) #define _FP_FRAC_ADDI_1(X,I) (X##_f += I)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 36) #define _FP_FRAC_SLL_1(X,N) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 37) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 38) if (__builtin_constant_p(N) && (N) == 1) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 39) X##_f += X##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 40) else \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 41) X##_f <<= (N); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 42) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 43) #define _FP_FRAC_SRL_1(X,N) (X##_f >>= N)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 44)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 45) /* Right shift with sticky-lsb. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 46) #define _FP_FRAC_SRS_1(X,N,sz) __FP_FRAC_SRS_1(X##_f, N, sz)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 47)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 48) #define __FP_FRAC_SRS_1(X,N,sz) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 49) (X = (X >> (N) | (__builtin_constant_p(N) && (N) == 1 \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 50) ? X & 1 : (X << (_FP_W_TYPE_SIZE - (N))) != 0)))
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 51)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 52) #define _FP_FRAC_ADD_1(R,X,Y) (R##_f = X##_f + Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 53) #define _FP_FRAC_SUB_1(R,X,Y) (R##_f = X##_f - Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 54) #define _FP_FRAC_DEC_1(X,Y) (X##_f -= Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 55) #define _FP_FRAC_CLZ_1(z, X) __FP_CLZ(z, X##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 56)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 57) /* Predicates */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 58) #define _FP_FRAC_NEGP_1(X) ((_FP_WS_TYPE)X##_f < 0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 59) #define _FP_FRAC_ZEROP_1(X) (X##_f == 0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 60) #define _FP_FRAC_OVERP_1(fs,X) (X##_f & _FP_OVERFLOW_##fs)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 61) #define _FP_FRAC_CLEAR_OVERP_1(fs,X) (X##_f &= ~_FP_OVERFLOW_##fs)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 62) #define _FP_FRAC_EQ_1(X, Y) (X##_f == Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 63) #define _FP_FRAC_GE_1(X, Y) (X##_f >= Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 64) #define _FP_FRAC_GT_1(X, Y) (X##_f > Y##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 65)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 66) #define _FP_ZEROFRAC_1 0
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 67) #define _FP_MINFRAC_1 1
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 68) #define _FP_MAXFRAC_1 (~(_FP_WS_TYPE)0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 69)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 70) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 71) * Unpack the raw bits of a native fp value. Do not classify or
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 72) * normalize the data.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 73) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 74)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 75) #define _FP_UNPACK_RAW_1(fs, X, val) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 76) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 77) union _FP_UNION_##fs _flo; _flo.flt = (val); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 78) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 79) X##_f = _flo.bits.frac; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 80) X##_e = _flo.bits.exp; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 81) X##_s = _flo.bits.sign; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 82) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 83)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 84) #define _FP_UNPACK_RAW_1_P(fs, X, val) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 85) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 86) union _FP_UNION_##fs *_flo = \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 87) (union _FP_UNION_##fs *)(val); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 88) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 89) X##_f = _flo->bits.frac; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 90) X##_e = _flo->bits.exp; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 91) X##_s = _flo->bits.sign; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 92) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 93)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 94) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 95) * Repack the raw bits of a native fp value.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 96) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 97)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 98) #define _FP_PACK_RAW_1(fs, val, X) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 99) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 100) union _FP_UNION_##fs _flo; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 101) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 102) _flo.bits.frac = X##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 103) _flo.bits.exp = X##_e; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 104) _flo.bits.sign = X##_s; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 105) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 106) (val) = _flo.flt; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 107) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 108)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 109) #define _FP_PACK_RAW_1_P(fs, val, X) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 110) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 111) union _FP_UNION_##fs *_flo = \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 112) (union _FP_UNION_##fs *)(val); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 113) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 114) _flo->bits.frac = X##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 115) _flo->bits.exp = X##_e; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 116) _flo->bits.sign = X##_s; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 117) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 118)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 119)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 120) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 121) * Multiplication algorithms:
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 122) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 123)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 124) /* Basic. Assuming the host word size is >= 2*FRACBITS, we can do the
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 125) multiplication immediately. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 126)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 127) #define _FP_MUL_MEAT_1_imm(wfracbits, R, X, Y) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 128) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 129) R##_f = X##_f * Y##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 130) /* Normalize since we know where the msb of the multiplicands \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 131) were (bit B), we know that the msb of the of the product is \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 132) at either 2B or 2B-1. */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 133) _FP_FRAC_SRS_1(R, wfracbits-1, 2*wfracbits); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 134) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 135)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 136) /* Given a 1W * 1W => 2W primitive, do the extended multiplication. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 137)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 138) #define _FP_MUL_MEAT_1_wide(wfracbits, R, X, Y, doit) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 139) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 140) _FP_W_TYPE _Z_f0, _Z_f1; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 141) doit(_Z_f1, _Z_f0, X##_f, Y##_f); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 142) /* Normalize since we know where the msb of the multiplicands \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 143) were (bit B), we know that the msb of the of the product is \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 144) at either 2B or 2B-1. */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 145) _FP_FRAC_SRS_2(_Z, wfracbits-1, 2*wfracbits); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 146) R##_f = _Z_f0; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 147) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 148)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 149) /* Finally, a simple widening multiply algorithm. What fun! */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 150)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 151) #define _FP_MUL_MEAT_1_hard(wfracbits, R, X, Y) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 152) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 153) _FP_W_TYPE _xh, _xl, _yh, _yl, _z_f0, _z_f1, _a_f0, _a_f1; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 154) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 155) /* split the words in half */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 156) _xh = X##_f >> (_FP_W_TYPE_SIZE/2); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 157) _xl = X##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 158) _yh = Y##_f >> (_FP_W_TYPE_SIZE/2); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 159) _yl = Y##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 160) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 161) /* multiply the pieces */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 162) _z_f0 = _xl * _yl; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 163) _a_f0 = _xh * _yl; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 164) _a_f1 = _xl * _yh; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 165) _z_f1 = _xh * _yh; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 166) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 167) /* reassemble into two full words */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 168) if ((_a_f0 += _a_f1) < _a_f1) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 169) _z_f1 += (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 170) _a_f1 = _a_f0 >> (_FP_W_TYPE_SIZE/2); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 171) _a_f0 = _a_f0 << (_FP_W_TYPE_SIZE/2); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 172) _FP_FRAC_ADD_2(_z, _z, _a); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 173) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 174) /* normalize */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 175) _FP_FRAC_SRS_2(_z, wfracbits - 1, 2*wfracbits); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 176) R##_f = _z_f0; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 177) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 178)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 179)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 180) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 181) * Division algorithms:
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 182) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 183)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 184) /* Basic. Assuming the host word size is >= 2*FRACBITS, we can do the
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 185) division immediately. Give this macro either _FP_DIV_HELP_imm for
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 186) C primitives or _FP_DIV_HELP_ldiv for the ISO function. Which you
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 187) choose will depend on what the compiler does with divrem4. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 188)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 189) #define _FP_DIV_MEAT_1_imm(fs, R, X, Y, doit) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 190) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 191) _FP_W_TYPE _q, _r; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 192) X##_f <<= (X##_f < Y##_f \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 193) ? R##_e--, _FP_WFRACBITS_##fs \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 194) : _FP_WFRACBITS_##fs - 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 195) doit(_q, _r, X##_f, Y##_f); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 196) R##_f = _q | (_r != 0); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 197) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 198)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 199) /* GCC's longlong.h defines a 2W / 1W => (1W,1W) primitive udiv_qrnnd
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 200) that may be useful in this situation. This first is for a primitive
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 201) that requires normalization, the second for one that does not. Look
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 202) for UDIV_NEEDS_NORMALIZATION to tell which your machine needs. */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 203)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 204) #define _FP_DIV_MEAT_1_udiv_norm(fs, R, X, Y) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 205) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 206) _FP_W_TYPE _nh, _nl, _q, _r, _y; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 207) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 208) /* Normalize Y -- i.e. make the most significant bit set. */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 209) _y = Y##_f << _FP_WFRACXBITS_##fs; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 210) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 211) /* Shift X op correspondingly high, that is, up one full word. */ \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 212) if (X##_f < Y##_f) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 213) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 214) R##_e--; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 215) _nl = 0; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 216) _nh = X##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 217) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 218) else \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 219) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 220) _nl = X##_f << (_FP_W_TYPE_SIZE - 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 221) _nh = X##_f >> 1; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 222) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 223) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 224) udiv_qrnnd(_q, _r, _nh, _nl, _y); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 225) R##_f = _q | (_r != 0); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 226) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 227)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 228) #define _FP_DIV_MEAT_1_udiv(fs, R, X, Y) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 229) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 230) _FP_W_TYPE _nh, _nl, _q, _r; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 231) if (X##_f < Y##_f) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 232) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 233) R##_e--; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 234) _nl = X##_f << _FP_WFRACBITS_##fs; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 235) _nh = X##_f >> _FP_WFRACXBITS_##fs; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 236) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 237) else \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 238) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 239) _nl = X##_f << (_FP_WFRACBITS_##fs - 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 240) _nh = X##_f >> (_FP_WFRACXBITS_##fs + 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 241) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 242) udiv_qrnnd(_q, _r, _nh, _nl, Y##_f); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 243) R##_f = _q | (_r != 0); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 244) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 245)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 246)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 247) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 248) * Square root algorithms:
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 249) * We have just one right now, maybe Newton approximation
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 250) * should be added for those machines where division is fast.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 251) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 252)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 253) #define _FP_SQRT_MEAT_1(R, S, T, X, q) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 254) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 255) while (q != _FP_WORK_ROUND) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 256) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 257) T##_f = S##_f + q; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 258) if (T##_f <= X##_f) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 259) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 260) S##_f = T##_f + q; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 261) X##_f -= T##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 262) R##_f += q; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 263) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 264) _FP_FRAC_SLL_1(X, 1); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 265) q >>= 1; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 266) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 267) if (X##_f) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 268) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 269) if (S##_f < X##_f) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 270) R##_f |= _FP_WORK_ROUND; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 271) R##_f |= _FP_WORK_STICKY; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 272) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 273) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 274)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 275) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 276) * Assembly/disassembly for converting to/from integral types.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 277) * No shifting or overflow handled here.
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 278) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 279)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 280) #define _FP_FRAC_ASSEMBLE_1(r, X, rsize) (r = X##_f)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 281) #define _FP_FRAC_DISASSEMBLE_1(X, r, rsize) (X##_f = r)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 282)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 283)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 284) /*
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 285) * Convert FP values between word sizes
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 286) */
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 287)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 288) #define _FP_FRAC_CONV_1_1(dfs, sfs, D, S) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 289) do { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 290) D##_f = S##_f; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 291) if (_FP_WFRACBITS_##sfs > _FP_WFRACBITS_##dfs) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 292) { \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 293) if (S##_c != FP_CLS_NAN) \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 294) _FP_FRAC_SRS_1(D, (_FP_WFRACBITS_##sfs-_FP_WFRACBITS_##dfs), \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 295) _FP_WFRACBITS_##sfs); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 296) else \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 297) _FP_FRAC_SRL_1(D, (_FP_WFRACBITS_##sfs-_FP_WFRACBITS_##dfs)); \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 298) } \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 299) else \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 300) D##_f <<= _FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs; \
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 301) } while (0)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 302)
^1da177e4c3f4 (Linus Torvalds 2005-04-16 15:20:36 -0700 303) #endif /* __MATH_EMU_OP_1_H__ */
VisionFive2 Linux kernel
StarFive Tech Linux Kernel for VisionFive (JH7110) boards (mirror)
More than 9999 Commits
32 Branches
54 Tags