diff options
Diffstat (limited to 'soft-fp')
| -rw-r--r-- | soft-fp/op-1.h | 218 | ||||
| -rw-r--r-- | soft-fp/op-2.h | 298 | ||||
| -rw-r--r-- | soft-fp/op-4.h | 574 | ||||
| -rw-r--r-- | soft-fp/op-8.h | 146 |
4 files changed, 726 insertions, 510 deletions
diff --git a/soft-fp/op-1.h b/soft-fp/op-1.h index 33682cf..3547e74 100644 --- a/soft-fp/op-1.h +++ b/soft-fp/op-1.h @@ -92,27 +92,28 @@ * normalize the data. */ -#define _FP_UNPACK_RAW_1(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs _flo; \ - _flo.flt = (val); \ - \ - X##_f = _flo.bits.frac; \ - X##_e = _flo.bits.exp; \ - X##_s = _flo.bits.sign; \ - } \ +#define _FP_UNPACK_RAW_1(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs _FP_UNPACK_RAW_1_flo; \ + _FP_UNPACK_RAW_1_flo.flt = (val); \ + \ + X##_f = _FP_UNPACK_RAW_1_flo.bits.frac; \ + X##_e = _FP_UNPACK_RAW_1_flo.bits.exp; \ + X##_s = _FP_UNPACK_RAW_1_flo.bits.sign; \ + } \ while (0) -#define _FP_UNPACK_RAW_1_P(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - X##_f = _flo->bits.frac; \ - X##_e = _flo->bits.exp; \ - X##_s = _flo->bits.sign; \ - } \ +#define _FP_UNPACK_RAW_1_P(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs *_FP_UNPACK_RAW_1_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + X##_f = _FP_UNPACK_RAW_1_P_flo->bits.frac; \ + X##_e = _FP_UNPACK_RAW_1_P_flo->bits.exp; \ + X##_s = _FP_UNPACK_RAW_1_P_flo->bits.sign; \ + } \ while (0) /* @@ -122,25 +123,26 @@ #define _FP_PACK_RAW_1(fs, val, X) \ do \ { \ - union _FP_UNION_##fs _flo; \ + union _FP_UNION_##fs _FP_PACK_RAW_1_flo; \ \ - _flo.bits.frac = X##_f; \ - _flo.bits.exp = X##_e; \ - _flo.bits.sign = X##_s; \ + _FP_PACK_RAW_1_flo.bits.frac = X##_f; \ + _FP_PACK_RAW_1_flo.bits.exp = X##_e; \ + _FP_PACK_RAW_1_flo.bits.sign = X##_s; \ \ - (val) = _flo.flt; \ + (val) = _FP_PACK_RAW_1_flo.flt; \ } \ while (0) -#define _FP_PACK_RAW_1_P(fs, val, X) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - _flo->bits.frac = X##_f; \ - _flo->bits.exp = X##_e; \ - _flo->bits.sign = X##_s; \ - } \ +#define _FP_PACK_RAW_1_P(fs, val, X) \ + do \ + { \ + union _FP_UNION_##fs *_FP_PACK_RAW_1_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + _FP_PACK_RAW_1_P_flo->bits.frac = X##_f; \ + _FP_PACK_RAW_1_P_flo->bits.exp = X##_e; \ + _FP_PACK_RAW_1_P_flo->bits.sign = X##_s; \ + } \ while (0) @@ -181,13 +183,14 @@ #define _FP_MUL_MEAT_1_wide(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_2 (_Z); \ - _FP_MUL_MEAT_DW_1_wide (wfracbits, _Z, X, Y, doit); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_1_wide_Z); \ + _FP_MUL_MEAT_DW_1_wide (wfracbits, _FP_MUL_MEAT_1_wide_Z, \ + X, Y, doit); \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_2 (_Z, wfracbits-1, 2*wfracbits); \ - R##_f = _Z_f0; \ + _FP_FRAC_SRS_2 (_FP_MUL_MEAT_1_wide_Z, wfracbits-1, 2*wfracbits); \ + R##_f = _FP_MUL_MEAT_1_wide_Z_f0; \ } \ while (0) @@ -196,40 +199,49 @@ #define _FP_MUL_MEAT_DW_1_hard(wfracbits, R, X, Y) \ do \ { \ - _FP_W_TYPE _xh, _xl, _yh, _yl; \ - _FP_FRAC_DECL_2 (_a); \ + _FP_W_TYPE _FP_MUL_MEAT_DW_1_hard_xh, _FP_MUL_MEAT_DW_1_hard_xl; \ + _FP_W_TYPE _FP_MUL_MEAT_DW_1_hard_yh, _FP_MUL_MEAT_DW_1_hard_yl; \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_1_hard_a); \ \ /* split the words in half */ \ - _xh = X##_f >> (_FP_W_TYPE_SIZE/2); \ - _xl = X##_f & (((_FP_W_TYPE) 1 << (_FP_W_TYPE_SIZE/2)) - 1); \ - _yh = Y##_f >> (_FP_W_TYPE_SIZE/2); \ - _yl = Y##_f & (((_FP_W_TYPE) 1 << (_FP_W_TYPE_SIZE/2)) - 1); \ + _FP_MUL_MEAT_DW_1_hard_xh = X##_f >> (_FP_W_TYPE_SIZE/2); \ + _FP_MUL_MEAT_DW_1_hard_xl \ + = X##_f & (((_FP_W_TYPE) 1 << (_FP_W_TYPE_SIZE/2)) - 1); \ + _FP_MUL_MEAT_DW_1_hard_yh = Y##_f >> (_FP_W_TYPE_SIZE/2); \ + _FP_MUL_MEAT_DW_1_hard_yl \ + = Y##_f & (((_FP_W_TYPE) 1 << (_FP_W_TYPE_SIZE/2)) - 1); \ \ /* multiply the pieces */ \ - R##_f0 = _xl * _yl; \ - _a_f0 = _xh * _yl; \ - _a_f1 = _xl * _yh; \ - R##_f1 = _xh * _yh; \ + R##_f0 = _FP_MUL_MEAT_DW_1_hard_xl * _FP_MUL_MEAT_DW_1_hard_yl; \ + _FP_MUL_MEAT_DW_1_hard_a_f0 \ + = _FP_MUL_MEAT_DW_1_hard_xh * _FP_MUL_MEAT_DW_1_hard_yl; \ + _FP_MUL_MEAT_DW_1_hard_a_f1 \ + = _FP_MUL_MEAT_DW_1_hard_xl * _FP_MUL_MEAT_DW_1_hard_yh; \ + R##_f1 = _FP_MUL_MEAT_DW_1_hard_xh * _FP_MUL_MEAT_DW_1_hard_yh; \ \ /* reassemble into two full words */ \ - if ((_a_f0 += _a_f1) < _a_f1) \ + if ((_FP_MUL_MEAT_DW_1_hard_a_f0 += _FP_MUL_MEAT_DW_1_hard_a_f1) \ + < _FP_MUL_MEAT_DW_1_hard_a_f1) \ R##_f1 += (_FP_W_TYPE) 1 << (_FP_W_TYPE_SIZE/2); \ - _a_f1 = _a_f0 >> (_FP_W_TYPE_SIZE/2); \ - _a_f0 = _a_f0 << (_FP_W_TYPE_SIZE/2); \ - _FP_FRAC_ADD_2 (R, R, _a); \ + _FP_MUL_MEAT_DW_1_hard_a_f1 \ + = _FP_MUL_MEAT_DW_1_hard_a_f0 >> (_FP_W_TYPE_SIZE/2); \ + _FP_MUL_MEAT_DW_1_hard_a_f0 \ + = _FP_MUL_MEAT_DW_1_hard_a_f0 << (_FP_W_TYPE_SIZE/2); \ + _FP_FRAC_ADD_2 (R, R, _FP_MUL_MEAT_DW_1_hard_a); \ } \ while (0) -#define _FP_MUL_MEAT_1_hard(wfracbits, R, X, Y) \ - do \ - { \ - _FP_FRAC_DECL_2 (_z); \ - _FP_MUL_MEAT_DW_1_hard (wfracbits, _z, X, Y); \ - \ - /* normalize */ \ - _FP_FRAC_SRS_2 (_z, wfracbits - 1, 2*wfracbits); \ - R##_f = _z_f0; \ - } \ +#define _FP_MUL_MEAT_1_hard(wfracbits, R, X, Y) \ + do \ + { \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_1_hard_z); \ + _FP_MUL_MEAT_DW_1_hard (wfracbits, _FP_MUL_MEAT_1_hard_z, X, Y); \ + \ + /* normalize */ \ + _FP_FRAC_SRS_2 (_FP_MUL_MEAT_1_hard_z, \ + wfracbits - 1, 2*wfracbits); \ + R##_f = _FP_MUL_MEAT_1_hard_z_f0; \ + } \ while (0) @@ -242,16 +254,16 @@ C primitives or _FP_DIV_HELP_ldiv for the ISO function. Which you choose will depend on what the compiler does with divrem4. */ -#define _FP_DIV_MEAT_1_imm(fs, R, X, Y, doit) \ - do \ - { \ - _FP_W_TYPE _q, _r; \ - X##_f <<= (X##_f < Y##_f \ - ? R##_e--, _FP_WFRACBITS_##fs \ - : _FP_WFRACBITS_##fs - 1); \ - doit (_q, _r, X##_f, Y##_f); \ - R##_f = _q | (_r != 0); \ - } \ +#define _FP_DIV_MEAT_1_imm(fs, R, X, Y, doit) \ + do \ + { \ + _FP_W_TYPE _FP_DIV_MEAT_1_imm_q, _FP_DIV_MEAT_1_imm_r; \ + X##_f <<= (X##_f < Y##_f \ + ? R##_e--, _FP_WFRACBITS_##fs \ + : _FP_WFRACBITS_##fs - 1); \ + doit (_FP_DIV_MEAT_1_imm_q, _FP_DIV_MEAT_1_imm_r, X##_f, Y##_f); \ + R##_f = _FP_DIV_MEAT_1_imm_q | (_FP_DIV_MEAT_1_imm_r != 0); \ + } \ while (0) /* GCC's longlong.h defines a 2W / 1W => (1W,1W) primitive udiv_qrnnd @@ -262,47 +274,59 @@ #define _FP_DIV_MEAT_1_udiv_norm(fs, R, X, Y) \ do \ { \ - _FP_W_TYPE _nh, _nl, _q, _r, _y; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_norm_nh; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_norm_nl; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_norm_q; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_norm_r; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_norm_y; \ \ /* Normalize Y -- i.e. make the most significant bit set. */ \ - _y = Y##_f << _FP_WFRACXBITS_##fs; \ + _FP_DIV_MEAT_1_udiv_norm_y = Y##_f << _FP_WFRACXBITS_##fs; \ \ /* Shift X op correspondingly high, that is, up one full word. */ \ if (X##_f < Y##_f) \ { \ R##_e--; \ - _nl = 0; \ - _nh = X##_f; \ + _FP_DIV_MEAT_1_udiv_norm_nl = 0; \ + _FP_DIV_MEAT_1_udiv_norm_nh = X##_f; \ } \ else \ { \ - _nl = X##_f << (_FP_W_TYPE_SIZE - 1); \ - _nh = X##_f >> 1; \ + _FP_DIV_MEAT_1_udiv_norm_nl = X##_f << (_FP_W_TYPE_SIZE - 1); \ + _FP_DIV_MEAT_1_udiv_norm_nh = X##_f >> 1; \ } \ \ - udiv_qrnnd (_q, _r, _nh, _nl, _y); \ - R##_f = _q | (_r != 0); \ + udiv_qrnnd (_FP_DIV_MEAT_1_udiv_norm_q, \ + _FP_DIV_MEAT_1_udiv_norm_r, \ + _FP_DIV_MEAT_1_udiv_norm_nh, \ + _FP_DIV_MEAT_1_udiv_norm_nl, \ + _FP_DIV_MEAT_1_udiv_norm_y); \ + R##_f = (_FP_DIV_MEAT_1_udiv_norm_q \ + | (_FP_DIV_MEAT_1_udiv_norm_r != 0)); \ } \ while (0) -#define _FP_DIV_MEAT_1_udiv(fs, R, X, Y) \ - do \ - { \ - _FP_W_TYPE _nh, _nl, _q, _r; \ - if (X##_f < Y##_f) \ - { \ - R##_e--; \ - _nl = X##_f << _FP_WFRACBITS_##fs; \ - _nh = X##_f >> _FP_WFRACXBITS_##fs; \ - } \ - else \ - { \ - _nl = X##_f << (_FP_WFRACBITS_##fs - 1); \ - _nh = X##_f >> (_FP_WFRACXBITS_##fs + 1); \ - } \ - udiv_qrnnd (_q, _r, _nh, _nl, Y##_f); \ - R##_f = _q | (_r != 0); \ - } \ +#define _FP_DIV_MEAT_1_udiv(fs, R, X, Y) \ + do \ + { \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_nh, _FP_DIV_MEAT_1_udiv_nl; \ + _FP_W_TYPE _FP_DIV_MEAT_1_udiv_q, _FP_DIV_MEAT_1_udiv_r; \ + if (X##_f < Y##_f) \ + { \ + R##_e--; \ + _FP_DIV_MEAT_1_udiv_nl = X##_f << _FP_WFRACBITS_##fs; \ + _FP_DIV_MEAT_1_udiv_nh = X##_f >> _FP_WFRACXBITS_##fs; \ + } \ + else \ + { \ + _FP_DIV_MEAT_1_udiv_nl = X##_f << (_FP_WFRACBITS_##fs - 1); \ + _FP_DIV_MEAT_1_udiv_nh = X##_f >> (_FP_WFRACXBITS_##fs + 1); \ + } \ + udiv_qrnnd (_FP_DIV_MEAT_1_udiv_q, _FP_DIV_MEAT_1_udiv_r, \ + _FP_DIV_MEAT_1_udiv_nh, _FP_DIV_MEAT_1_udiv_nl, \ + Y##_f); \ + R##_f = _FP_DIV_MEAT_1_udiv_q | (_FP_DIV_MEAT_1_udiv_r != 0); \ + } \ while (0) diff --git a/soft-fp/op-2.h b/soft-fp/op-2.h index 160990f..4ea2a00 100644 --- a/soft-fp/op-2.h +++ b/soft-fp/op-2.h @@ -182,12 +182,12 @@ (rh = xh - yh - ((rl = xl - yl) > xl)) # endif # ifndef __FP_FRAC_DEC_2 -# define __FP_FRAC_DEC_2(xh, xl, yh, yl) \ - do \ - { \ - UWtype _t = xl; \ - xh -= yh + ((xl -= yl) > _t); \ - } \ +# define __FP_FRAC_DEC_2(xh, xl, yh, yl) \ + do \ + { \ + UWtype __FP_FRAC_DEC_2_t = xl; \ + xh -= yh + ((xl -= yl) > __FP_FRAC_DEC_2_t); \ + } \ while (0) # endif @@ -210,29 +210,30 @@ * normalize the data. */ -#define _FP_UNPACK_RAW_2(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs _flo; \ - _flo.flt = (val); \ - \ - X##_f0 = _flo.bits.frac0; \ - X##_f1 = _flo.bits.frac1; \ - X##_e = _flo.bits.exp; \ - X##_s = _flo.bits.sign; \ - } \ +#define _FP_UNPACK_RAW_2(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs _FP_UNPACK_RAW_2_flo; \ + _FP_UNPACK_RAW_2_flo.flt = (val); \ + \ + X##_f0 = _FP_UNPACK_RAW_2_flo.bits.frac0; \ + X##_f1 = _FP_UNPACK_RAW_2_flo.bits.frac1; \ + X##_e = _FP_UNPACK_RAW_2_flo.bits.exp; \ + X##_s = _FP_UNPACK_RAW_2_flo.bits.sign; \ + } \ while (0) -#define _FP_UNPACK_RAW_2_P(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - X##_f0 = _flo->bits.frac0; \ - X##_f1 = _flo->bits.frac1; \ - X##_e = _flo->bits.exp; \ - X##_s = _flo->bits.sign; \ - } \ +#define _FP_UNPACK_RAW_2_P(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs *_FP_UNPACK_RAW_2_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + X##_f0 = _FP_UNPACK_RAW_2_P_flo->bits.frac0; \ + X##_f1 = _FP_UNPACK_RAW_2_P_flo->bits.frac1; \ + X##_e = _FP_UNPACK_RAW_2_P_flo->bits.exp; \ + X##_s = _FP_UNPACK_RAW_2_P_flo->bits.sign; \ + } \ while (0) @@ -243,27 +244,28 @@ #define _FP_PACK_RAW_2(fs, val, X) \ do \ { \ - union _FP_UNION_##fs _flo; \ + union _FP_UNION_##fs _FP_PACK_RAW_2_flo; \ \ - _flo.bits.frac0 = X##_f0; \ - _flo.bits.frac1 = X##_f1; \ - _flo.bits.exp = X##_e; \ - _flo.bits.sign = X##_s; \ + _FP_PACK_RAW_2_flo.bits.frac0 = X##_f0; \ + _FP_PACK_RAW_2_flo.bits.frac1 = X##_f1; \ + _FP_PACK_RAW_2_flo.bits.exp = X##_e; \ + _FP_PACK_RAW_2_flo.bits.sign = X##_s; \ \ - (val) = _flo.flt; \ + (val) = _FP_PACK_RAW_2_flo.flt; \ } \ while (0) -#define _FP_PACK_RAW_2_P(fs, val, X) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - _flo->bits.frac0 = X##_f0; \ - _flo->bits.frac1 = X##_f1; \ - _flo->bits.exp = X##_e; \ - _flo->bits.sign = X##_s; \ - } \ +#define _FP_PACK_RAW_2_P(fs, val, X) \ + do \ + { \ + union _FP_UNION_##fs *_FP_PACK_RAW_2_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + _FP_PACK_RAW_2_P_flo->bits.frac0 = X##_f0; \ + _FP_PACK_RAW_2_P_flo->bits.frac1 = X##_f1; \ + _FP_PACK_RAW_2_P_flo->bits.exp = X##_e; \ + _FP_PACK_RAW_2_P_flo->bits.sign = X##_s; \ + } \ while (0) @@ -276,20 +278,28 @@ #define _FP_MUL_MEAT_DW_2_wide(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_2 (_b); \ - _FP_FRAC_DECL_2 (_c); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_2_wide_b); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_2_wide_c); \ \ - doit (_FP_FRAC_WORD_4 (R, 1), _FP_FRAC_WORD_4 (R, 0), X##_f0, Y##_f0); \ - doit (_b_f1, _b_f0, X##_f0, Y##_f1); \ - doit (_c_f1, _c_f0, X##_f1, Y##_f0); \ - doit (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), X##_f1, Y##_f1); \ + doit (_FP_FRAC_WORD_4 (R, 1), _FP_FRAC_WORD_4 (R, 0), \ + X##_f0, Y##_f0); \ + doit (_FP_MUL_MEAT_DW_2_wide_b_f1, _FP_MUL_MEAT_DW_2_wide_b_f0, \ + X##_f0, Y##_f1); \ + doit (_FP_MUL_MEAT_DW_2_wide_c_f1, _FP_MUL_MEAT_DW_2_wide_c_f0, \ + X##_f1, Y##_f0); \ + doit (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ + X##_f1, Y##_f1); \ \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _FP_FRAC_WORD_4 (R, 1), 0, _b_f1, _b_f0, \ + _FP_FRAC_WORD_4 (R, 1), 0, \ + _FP_MUL_MEAT_DW_2_wide_b_f1, \ + _FP_MUL_MEAT_DW_2_wide_b_f0, \ _FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ _FP_FRAC_WORD_4 (R, 1)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _FP_FRAC_WORD_4 (R, 1), 0, _c_f1, _c_f0, \ + _FP_FRAC_WORD_4 (R, 1), 0, \ + _FP_MUL_MEAT_DW_2_wide_c_f1, \ + _FP_MUL_MEAT_DW_2_wide_c_f0, \ _FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ _FP_FRAC_WORD_4 (R, 1)); \ } \ @@ -298,16 +308,17 @@ #define _FP_MUL_MEAT_2_wide(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_4 (_z); \ + _FP_FRAC_DECL_4 (_FP_MUL_MEAT_2_wide_z); \ \ - _FP_MUL_MEAT_DW_2_wide (wfracbits, _z, X, Y, doit); \ + _FP_MUL_MEAT_DW_2_wide (wfracbits, _FP_MUL_MEAT_2_wide_z, \ + X, Y, doit); \ \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_4 (_z, wfracbits-1, 2*wfracbits); \ - R##_f0 = _FP_FRAC_WORD_4 (_z, 0); \ - R##_f1 = _FP_FRAC_WORD_4 (_z, 1); \ + _FP_FRAC_SRS_4 (_FP_MUL_MEAT_2_wide_z, wfracbits-1, 2*wfracbits); \ + R##_f0 = _FP_FRAC_WORD_4 (_FP_MUL_MEAT_2_wide_z, 0); \ + R##_f1 = _FP_FRAC_WORD_4 (_FP_MUL_MEAT_2_wide_z, 1); \ } \ while (0) @@ -318,35 +329,51 @@ #define _FP_MUL_MEAT_DW_2_wide_3mul(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_2 (_b); \ - _FP_FRAC_DECL_2 (_c); \ - _FP_W_TYPE _d; \ - int _c1, _c2; \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_2_wide_3mul_b); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_2_wide_3mul_c); \ + _FP_W_TYPE _FP_MUL_MEAT_DW_2_wide_3mul_d; \ + int _FP_MUL_MEAT_DW_2_wide_3mul_c1; \ + int _FP_MUL_MEAT_DW_2_wide_3mul_c2; \ \ - _b_f0 = X##_f0 + X##_f1; \ - _c1 = _b_f0 < X##_f0; \ - _b_f1 = Y##_f0 + Y##_f1; \ - _c2 = _b_f1 < Y##_f0; \ - doit (_d, _FP_FRAC_WORD_4 (R, 0), X##_f0, Y##_f0); \ - doit (_FP_FRAC_WORD_4 (R, 2), _FP_FRAC_WORD_4 (R, 1), _b_f0, _b_f1); \ - doit (_c_f1, _c_f0, X##_f1, Y##_f1); \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f0 = X##_f0 + X##_f1; \ + _FP_MUL_MEAT_DW_2_wide_3mul_c1 \ + = _FP_MUL_MEAT_DW_2_wide_3mul_b_f0 < X##_f0; \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f1 = Y##_f0 + Y##_f1; \ + _FP_MUL_MEAT_DW_2_wide_3mul_c2 \ + = _FP_MUL_MEAT_DW_2_wide_3mul_b_f1 < Y##_f0; \ + doit (_FP_MUL_MEAT_DW_2_wide_3mul_d, _FP_FRAC_WORD_4 (R, 0), \ + X##_f0, Y##_f0); \ + doit (_FP_FRAC_WORD_4 (R, 2), _FP_FRAC_WORD_4 (R, 1), \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f0, \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f1); \ + doit (_FP_MUL_MEAT_DW_2_wide_3mul_c_f1, \ + _FP_MUL_MEAT_DW_2_wide_3mul_c_f0, X##_f1, Y##_f1); \ \ - _b_f0 &= -_c2; \ - _b_f1 &= -_c1; \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f0 \ + &= -_FP_MUL_MEAT_DW_2_wide_3mul_c2; \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f1 \ + &= -_FP_MUL_MEAT_DW_2_wide_3mul_c1; \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _FP_FRAC_WORD_4 (R, 1), (_c1 & _c2), 0, _d, \ + _FP_FRAC_WORD_4 (R, 1), \ + (_FP_MUL_MEAT_DW_2_wide_3mul_c1 \ + & _FP_MUL_MEAT_DW_2_wide_3mul_c2), 0, \ + _FP_MUL_MEAT_DW_2_wide_3mul_d, \ 0, _FP_FRAC_WORD_4 (R, 2), _FP_FRAC_WORD_4 (R, 1)); \ __FP_FRAC_ADDI_2 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _b_f0); \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f0); \ __FP_FRAC_ADDI_2 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _b_f1); \ + _FP_MUL_MEAT_DW_2_wide_3mul_b_f1); \ __FP_FRAC_DEC_3 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ _FP_FRAC_WORD_4 (R, 1), \ - 0, _d, _FP_FRAC_WORD_4 (R, 0)); \ + 0, _FP_MUL_MEAT_DW_2_wide_3mul_d, \ + _FP_FRAC_WORD_4 (R, 0)); \ __FP_FRAC_DEC_3 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _FP_FRAC_WORD_4 (R, 1), 0, _c_f1, _c_f0); \ + _FP_FRAC_WORD_4 (R, 1), 0, \ + _FP_MUL_MEAT_DW_2_wide_3mul_c_f1, \ + _FP_MUL_MEAT_DW_2_wide_3mul_c_f0); \ __FP_FRAC_ADD_2 (_FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2), \ - _c_f1, _c_f0, \ + _FP_MUL_MEAT_DW_2_wide_3mul_c_f1, \ + _FP_MUL_MEAT_DW_2_wide_3mul_c_f0, \ _FP_FRAC_WORD_4 (R, 3), _FP_FRAC_WORD_4 (R, 2)); \ } \ while (0) @@ -354,45 +381,50 @@ #define _FP_MUL_MEAT_2_wide_3mul(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_4 (_z); \ + _FP_FRAC_DECL_4 (_FP_MUL_MEAT_2_wide_3mul_z); \ \ - _FP_MUL_MEAT_DW_2_wide_3mul (wfracbits, _z, X, Y, doit); \ + _FP_MUL_MEAT_DW_2_wide_3mul (wfracbits, \ + _FP_MUL_MEAT_2_wide_3mul_z, \ + X, Y, doit); \ \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_4 (_z, wfracbits-1, 2*wfracbits); \ - R##_f0 = _FP_FRAC_WORD_4 (_z, 0); \ - R##_f1 = _FP_FRAC_WORD_4 (_z, 1); \ + _FP_FRAC_SRS_4 (_FP_MUL_MEAT_2_wide_3mul_z, \ + wfracbits-1, 2*wfracbits); \ + R##_f0 = _FP_FRAC_WORD_4 (_FP_MUL_MEAT_2_wide_3mul_z, 0); \ + R##_f1 = _FP_FRAC_WORD_4 (_FP_MUL_MEAT_2_wide_3mul_z, 1); \ } \ while (0) #define _FP_MUL_MEAT_DW_2_gmp(wfracbits, R, X, Y) \ do \ { \ - _FP_W_TYPE _x[2], _y[2]; \ - _x[0] = X##_f0; \ - _x[1] = X##_f1; \ - _y[0] = Y##_f0; \ - _y[1] = Y##_f1; \ + _FP_W_TYPE _FP_MUL_MEAT_DW_2_gmp_x[2]; \ + _FP_W_TYPE _FP_MUL_MEAT_DW_2_gmp_y[2]; \ + _FP_MUL_MEAT_DW_2_gmp_x[0] = X##_f0; \ + _FP_MUL_MEAT_DW_2_gmp_x[1] = X##_f1; \ + _FP_MUL_MEAT_DW_2_gmp_y[0] = Y##_f0; \ + _FP_MUL_MEAT_DW_2_gmp_y[1] = Y##_f1; \ \ - mpn_mul_n (R##_f, _x, _y, 2); \ + mpn_mul_n (R##_f, _FP_MUL_MEAT_DW_2_gmp_x, \ + _FP_MUL_MEAT_DW_2_gmp_y, 2); \ } \ while (0) #define _FP_MUL_MEAT_2_gmp(wfracbits, R, X, Y) \ do \ { \ - _FP_FRAC_DECL_4 (_z); \ + _FP_FRAC_DECL_4 (_FP_MUL_MEAT_2_gmp_z); \ \ - _FP_MUL_MEAT_DW_2_gmp (wfracbits, _z, X, Y); \ + _FP_MUL_MEAT_DW_2_gmp (wfracbits, _FP_MUL_MEAT_2_gmp_z, X, Y); \ \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_4 (_z, wfracbits-1, 2*wfracbits); \ - R##_f0 = _z_f[0]; \ - R##_f1 = _z_f[1]; \ + _FP_FRAC_SRS_4 (_FP_MUL_MEAT_2_gmp_z, wfracbits-1, 2*wfracbits); \ + R##_f0 = _FP_MUL_MEAT_2_gmp_z_f[0]; \ + R##_f1 = _FP_MUL_MEAT_2_gmp_z_f[1]; \ } \ while (0) @@ -507,67 +539,91 @@ #define _FP_DIV_MEAT_2_udiv(fs, R, X, Y) \ do \ { \ - _FP_W_TYPE _n_f2, _n_f1, _n_f0, _r_f1, _r_f0, _m_f1, _m_f0; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_n_f2; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_n_f1; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_n_f0; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_r_f1; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_r_f0; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_m_f1; \ + _FP_W_TYPE _FP_DIV_MEAT_2_udiv_m_f0; \ if (_FP_FRAC_GE_2 (X, Y)) \ { \ - _n_f2 = X##_f1 >> 1; \ - _n_f1 = X##_f1 << (_FP_W_TYPE_SIZE - 1) | X##_f0 >> 1; \ - _n_f0 = X##_f0 << (_FP_W_TYPE_SIZE - 1); \ + _FP_DIV_MEAT_2_udiv_n_f2 = X##_f1 >> 1; \ + _FP_DIV_MEAT_2_udiv_n_f1 \ + = X##_f1 << (_FP_W_TYPE_SIZE - 1) | X##_f0 >> 1; \ + _FP_DIV_MEAT_2_udiv_n_f0 \ + = X##_f0 << (_FP_W_TYPE_SIZE - 1); \ } \ else \ { \ R##_e--; \ - _n_f2 = X##_f1; \ - _n_f1 = X##_f0; \ - _n_f0 = 0; \ + _FP_DIV_MEAT_2_udiv_n_f2 = X##_f1; \ + _FP_DIV_MEAT_2_udiv_n_f1 = X##_f0; \ + _FP_DIV_MEAT_2_udiv_n_f0 = 0; \ } \ \ /* Normalize, i.e. make the most significant bit of the \ denominator set. */ \ _FP_FRAC_SLL_2 (Y, _FP_WFRACXBITS_##fs); \ \ - udiv_qrnnd (R##_f1, _r_f1, _n_f2, _n_f1, Y##_f1); \ - umul_ppmm (_m_f1, _m_f0, R##_f1, Y##_f0); \ - _r_f0 = _n_f0; \ - if (_FP_FRAC_GT_2 (_m, _r)) \ + udiv_qrnnd (R##_f1, _FP_DIV_MEAT_2_udiv_r_f1, \ + _FP_DIV_MEAT_2_udiv_n_f2, _FP_DIV_MEAT_2_udiv_n_f1, \ + Y##_f1); \ + umul_ppmm (_FP_DIV_MEAT_2_udiv_m_f1, _FP_DIV_MEAT_2_udiv_m_f0, \ + R##_f1, Y##_f0); \ + _FP_DIV_MEAT_2_udiv_r_f0 = _FP_DIV_MEAT_2_udiv_n_f0; \ + if (_FP_FRAC_GT_2 (_FP_DIV_MEAT_2_udiv_m, _FP_DIV_MEAT_2_udiv_r)) \ { \ R##_f1--; \ - _FP_FRAC_ADD_2 (_r, Y, _r); \ - if (_FP_FRAC_GE_2 (_r, Y) && _FP_FRAC_GT_2 (_m, _r)) \ + _FP_FRAC_ADD_2 (_FP_DIV_MEAT_2_udiv_r, Y, \ + _FP_DIV_MEAT_2_udiv_r); \ + if (_FP_FRAC_GE_2 (_FP_DIV_MEAT_2_udiv_r, Y) \ + && _FP_FRAC_GT_2 (_FP_DIV_MEAT_2_udiv_m, \ + _FP_DIV_MEAT_2_udiv_r)) \ { \ R##_f1--; \ - _FP_FRAC_ADD_2 (_r, Y, _r); \ + _FP_FRAC_ADD_2 (_FP_DIV_MEAT_2_udiv_r, Y, \ + _FP_DIV_MEAT_2_udiv_r); \ } \ } \ - _FP_FRAC_DEC_2 (_r, _m); \ + _FP_FRAC_DEC_2 (_FP_DIV_MEAT_2_udiv_r, _FP_DIV_MEAT_2_udiv_m); \ \ - if (_r_f1 == Y##_f1) \ + if (_FP_DIV_MEAT_2_udiv_r_f1 == Y##_f1) \ { \ /* This is a special case, not an optimization \ - (_r/Y##_f1 would not fit into UWtype). \ - As _r is guaranteed to be < Y, R##_f0 can be either \ - (UWtype)-1 or (UWtype)-2. But as we know what kind \ - of bits it is (sticky, guard, round), we don't care. \ - We also don't care what the reminder is, because the \ - guard bit will be set anyway. -jj */ \ + (_FP_DIV_MEAT_2_udiv_r/Y##_f1 would not fit into UWtype). \ + As _FP_DIV_MEAT_2_udiv_r is guaranteed to be < Y, \ + R##_f0 can be either (UWtype)-1 or (UWtype)-2. But as we \ + know what kind of bits it is (sticky, guard, round), \ + we don't care. We also don't care what the reminder is, \ + because the guard bit will be set anyway. -jj */ \ R##_f0 = -1; \ } \ else \ { \ - udiv_qrnnd (R##_f0, _r_f1, _r_f1, _r_f0, Y##_f1); \ - umul_ppmm (_m_f1, _m_f0, R##_f0, Y##_f0); \ - _r_f0 = 0; \ - if (_FP_FRAC_GT_2 (_m, _r)) \ + udiv_qrnnd (R##_f0, _FP_DIV_MEAT_2_udiv_r_f1, \ + _FP_DIV_MEAT_2_udiv_r_f1, \ + _FP_DIV_MEAT_2_udiv_r_f0, Y##_f1); \ + umul_ppmm (_FP_DIV_MEAT_2_udiv_m_f1, \ + _FP_DIV_MEAT_2_udiv_m_f0, R##_f0, Y##_f0); \ + _FP_DIV_MEAT_2_udiv_r_f0 = 0; \ + if (_FP_FRAC_GT_2 (_FP_DIV_MEAT_2_udiv_m, \ + _FP_DIV_MEAT_2_udiv_r)) \ { \ R##_f0--; \ - _FP_FRAC_ADD_2 (_r, Y, _r); \ - if (_FP_FRAC_GE_2 (_r, Y) && _FP_FRAC_GT_2 (_m, _r)) \ + _FP_FRAC_ADD_2 (_FP_DIV_MEAT_2_udiv_r, Y, \ + _FP_DIV_MEAT_2_udiv_r); \ + if (_FP_FRAC_GE_2 (_FP_DIV_MEAT_2_udiv_r, Y) \ + && _FP_FRAC_GT_2 (_FP_DIV_MEAT_2_udiv_m, \ + _FP_DIV_MEAT_2_udiv_r)) \ { \ R##_f0--; \ - _FP_FRAC_ADD_2 (_r, Y, _r); \ + _FP_FRAC_ADD_2 (_FP_DIV_MEAT_2_udiv_r, Y, \ + _FP_DIV_MEAT_2_udiv_r); \ } \ } \ - if (!_FP_FRAC_EQ_2 (_r, _m)) \ + if (!_FP_FRAC_EQ_2 (_FP_DIV_MEAT_2_udiv_r, \ + _FP_DIV_MEAT_2_udiv_m)) \ R##_f0 |= _FP_WORK_STICKY; \ } \ } \ diff --git a/soft-fp/op-4.h b/soft-fp/op-4.h index a80bdb2..3acf96c 100644 --- a/soft-fp/op-4.h +++ b/soft-fp/op-4.h @@ -39,49 +39,67 @@ #define _FP_FRAC_LOW_4(X) (X##_f[0]) #define _FP_FRAC_WORD_4(X, w) (X##_f[w]) -#define _FP_FRAC_SLL_4(X, N) \ - do \ - { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _up = (N) % _FP_W_TYPE_SIZE; \ - _down = _FP_W_TYPE_SIZE - _up; \ - if (!_up) \ - for (_i = 3; _i >= _skip; --_i) \ - X##_f[_i] = X##_f[_i-_skip]; \ - else \ - { \ - for (_i = 3; _i > _skip; --_i) \ - X##_f[_i] = (X##_f[_i-_skip] << _up \ - | X##_f[_i-_skip-1] >> _down); \ - X##_f[_i--] = X##_f[0] << _up; \ - } \ - for (; _i >= 0; --_i) \ - X##_f[_i] = 0; \ - } \ +#define _FP_FRAC_SLL_4(X, N) \ + do \ + { \ + _FP_I_TYPE _FP_FRAC_SLL_4_up, _FP_FRAC_SLL_4_down; \ + _FP_I_TYPE _FP_FRAC_SLL_4_skip, _FP_FRAC_SLL_4_i; \ + _FP_FRAC_SLL_4_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SLL_4_up = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SLL_4_down = _FP_W_TYPE_SIZE - _FP_FRAC_SLL_4_up; \ + if (!_FP_FRAC_SLL_4_up) \ + for (_FP_FRAC_SLL_4_i = 3; \ + _FP_FRAC_SLL_4_i >= _FP_FRAC_SLL_4_skip; \ + --_FP_FRAC_SLL_4_i) \ + X##_f[_FP_FRAC_SLL_4_i] \ + = X##_f[_FP_FRAC_SLL_4_i-_FP_FRAC_SLL_4_skip]; \ + else \ + { \ + for (_FP_FRAC_SLL_4_i = 3; \ + _FP_FRAC_SLL_4_i > _FP_FRAC_SLL_4_skip; \ + --_FP_FRAC_SLL_4_i) \ + X##_f[_FP_FRAC_SLL_4_i] \ + = ((X##_f[_FP_FRAC_SLL_4_i-_FP_FRAC_SLL_4_skip] \ + << _FP_FRAC_SLL_4_up) \ + | (X##_f[_FP_FRAC_SLL_4_i-_FP_FRAC_SLL_4_skip-1] \ + >> _FP_FRAC_SLL_4_down)); \ + X##_f[_FP_FRAC_SLL_4_i--] = X##_f[0] << _FP_FRAC_SLL_4_up; \ + } \ + for (; _FP_FRAC_SLL_4_i >= 0; --_FP_FRAC_SLL_4_i) \ + X##_f[_FP_FRAC_SLL_4_i] = 0; \ + } \ while (0) /* This one was broken too */ -#define _FP_FRAC_SRL_4(X, N) \ - do \ - { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _down = (N) % _FP_W_TYPE_SIZE; \ - _up = _FP_W_TYPE_SIZE - _down; \ - if (!_down) \ - for (_i = 0; _i <= 3-_skip; ++_i) \ - X##_f[_i] = X##_f[_i+_skip]; \ - else \ - { \ - for (_i = 0; _i < 3-_skip; ++_i) \ - X##_f[_i] = (X##_f[_i+_skip] >> _down \ - | X##_f[_i+_skip+1] << _up); \ - X##_f[_i++] = X##_f[3] >> _down; \ - } \ - for (; _i < 4; ++_i) \ - X##_f[_i] = 0; \ - } \ +#define _FP_FRAC_SRL_4(X, N) \ + do \ + { \ + _FP_I_TYPE _FP_FRAC_SRL_4_up, _FP_FRAC_SRL_4_down; \ + _FP_I_TYPE _FP_FRAC_SRL_4_skip, _FP_FRAC_SRL_4_i; \ + _FP_FRAC_SRL_4_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRL_4_down = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRL_4_up = _FP_W_TYPE_SIZE - _FP_FRAC_SRL_4_down; \ + if (!_FP_FRAC_SRL_4_down) \ + for (_FP_FRAC_SRL_4_i = 0; \ + _FP_FRAC_SRL_4_i <= 3-_FP_FRAC_SRL_4_skip; \ + ++_FP_FRAC_SRL_4_i) \ + X##_f[_FP_FRAC_SRL_4_i] \ + = X##_f[_FP_FRAC_SRL_4_i+_FP_FRAC_SRL_4_skip]; \ + else \ + { \ + for (_FP_FRAC_SRL_4_i = 0; \ + _FP_FRAC_SRL_4_i < 3-_FP_FRAC_SRL_4_skip; \ + ++_FP_FRAC_SRL_4_i) \ + X##_f[_FP_FRAC_SRL_4_i] \ + = ((X##_f[_FP_FRAC_SRL_4_i+_FP_FRAC_SRL_4_skip] \ + >> _FP_FRAC_SRL_4_down) \ + | (X##_f[_FP_FRAC_SRL_4_i+_FP_FRAC_SRL_4_skip+1] \ + << _FP_FRAC_SRL_4_up)); \ + X##_f[_FP_FRAC_SRL_4_i++] = X##_f[3] >> _FP_FRAC_SRL_4_down; \ + } \ + for (; _FP_FRAC_SRL_4_i < 4; ++_FP_FRAC_SRL_4_i) \ + X##_f[_FP_FRAC_SRL_4_i] = 0; \ + } \ while (0) @@ -90,40 +108,53 @@ * but that if any of the bits that fall off the right hand side * were one then we always set the LSbit. */ -#define _FP_FRAC_SRST_4(X, S, N, size) \ - do \ - { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _FP_W_TYPE _s; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _down = (N) % _FP_W_TYPE_SIZE; \ - _up = _FP_W_TYPE_SIZE - _down; \ - for (_s = _i = 0; _i < _skip; ++_i) \ - _s |= X##_f[_i]; \ - if (!_down) \ - for (_i = 0; _i <= 3-_skip; ++_i) \ - X##_f[_i] = X##_f[_i+_skip]; \ - else \ - { \ - _s |= X##_f[_i] << _up; \ - for (_i = 0; _i < 3-_skip; ++_i) \ - X##_f[_i] = (X##_f[_i+_skip] >> _down \ - | X##_f[_i+_skip+1] << _up); \ - X##_f[_i++] = X##_f[3] >> _down; \ - } \ - for (; _i < 4; ++_i) \ - X##_f[_i] = 0; \ - S = (_s != 0); \ - } \ +#define _FP_FRAC_SRST_4(X, S, N, size) \ + do \ + { \ + _FP_I_TYPE _FP_FRAC_SRST_4_up, _FP_FRAC_SRST_4_down; \ + _FP_I_TYPE _FP_FRAC_SRST_4_skip, _FP_FRAC_SRST_4_i; \ + _FP_W_TYPE _FP_FRAC_SRST_4_s; \ + _FP_FRAC_SRST_4_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRST_4_down = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRST_4_up = _FP_W_TYPE_SIZE - _FP_FRAC_SRST_4_down; \ + for (_FP_FRAC_SRST_4_s = _FP_FRAC_SRST_4_i = 0; \ + _FP_FRAC_SRST_4_i < _FP_FRAC_SRST_4_skip; \ + ++_FP_FRAC_SRST_4_i) \ + _FP_FRAC_SRST_4_s |= X##_f[_FP_FRAC_SRST_4_i]; \ + if (!_FP_FRAC_SRST_4_down) \ + for (_FP_FRAC_SRST_4_i = 0; \ + _FP_FRAC_SRST_4_i <= 3-_FP_FRAC_SRST_4_skip; \ + ++_FP_FRAC_SRST_4_i) \ + X##_f[_FP_FRAC_SRST_4_i] \ + = X##_f[_FP_FRAC_SRST_4_i+_FP_FRAC_SRST_4_skip]; \ + else \ + { \ + _FP_FRAC_SRST_4_s \ + |= X##_f[_FP_FRAC_SRST_4_i] << _FP_FRAC_SRST_4_up; \ + for (_FP_FRAC_SRST_4_i = 0; \ + _FP_FRAC_SRST_4_i < 3-_FP_FRAC_SRST_4_skip; \ + ++_FP_FRAC_SRST_4_i) \ + X##_f[_FP_FRAC_SRST_4_i] \ + = ((X##_f[_FP_FRAC_SRST_4_i+_FP_FRAC_SRST_4_skip] \ + >> _FP_FRAC_SRST_4_down) \ + | (X##_f[_FP_FRAC_SRST_4_i+_FP_FRAC_SRST_4_skip+1] \ + << _FP_FRAC_SRST_4_up)); \ + X##_f[_FP_FRAC_SRST_4_i++] \ + = X##_f[3] >> _FP_FRAC_SRST_4_down; \ + } \ + for (; _FP_FRAC_SRST_4_i < 4; ++_FP_FRAC_SRST_4_i) \ + X##_f[_FP_FRAC_SRST_4_i] = 0; \ + S = (_FP_FRAC_SRST_4_s != 0); \ + } \ while (0) -#define _FP_FRAC_SRS_4(X, N, size) \ - do \ - { \ - int _sticky; \ - _FP_FRAC_SRST_4 (X, _sticky, N, size); \ - X##_f[0] |= _sticky; \ - } \ +#define _FP_FRAC_SRS_4(X, N, size) \ + do \ + { \ + int _FP_FRAC_SRS_4_sticky; \ + _FP_FRAC_SRST_4 (X, _FP_FRAC_SRS_4_sticky, N, size); \ + X##_f[0] |= _FP_FRAC_SRS_4_sticky; \ + } \ while (0) #define _FP_FRAC_ADD_4(R, X, Y) \ @@ -201,60 +232,62 @@ while (0) -#define _FP_UNPACK_RAW_4(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs _flo; \ - _flo.flt = (val); \ - X##_f[0] = _flo.bits.frac0; \ - X##_f[1] = _flo.bits.frac1; \ - X##_f[2] = _flo.bits.frac2; \ - X##_f[3] = _flo.bits.frac3; \ - X##_e = _flo.bits.exp; \ - X##_s = _flo.bits.sign; \ - } \ +#define _FP_UNPACK_RAW_4(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs _FP_UNPACK_RAW_4_flo; \ + _FP_UNPACK_RAW_4_flo.flt = (val); \ + X##_f[0] = _FP_UNPACK_RAW_4_flo.bits.frac0; \ + X##_f[1] = _FP_UNPACK_RAW_4_flo.bits.frac1; \ + X##_f[2] = _FP_UNPACK_RAW_4_flo.bits.frac2; \ + X##_f[3] = _FP_UNPACK_RAW_4_flo.bits.frac3; \ + X##_e = _FP_UNPACK_RAW_4_flo.bits.exp; \ + X##_s = _FP_UNPACK_RAW_4_flo.bits.sign; \ + } \ while (0) -#define _FP_UNPACK_RAW_4_P(fs, X, val) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - X##_f[0] = _flo->bits.frac0; \ - X##_f[1] = _flo->bits.frac1; \ - X##_f[2] = _flo->bits.frac2; \ - X##_f[3] = _flo->bits.frac3; \ - X##_e = _flo->bits.exp; \ - X##_s = _flo->bits.sign; \ - } \ +#define _FP_UNPACK_RAW_4_P(fs, X, val) \ + do \ + { \ + union _FP_UNION_##fs *_FP_UNPACK_RAW_4_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + X##_f[0] = _FP_UNPACK_RAW_4_P_flo->bits.frac0; \ + X##_f[1] = _FP_UNPACK_RAW_4_P_flo->bits.frac1; \ + X##_f[2] = _FP_UNPACK_RAW_4_P_flo->bits.frac2; \ + X##_f[3] = _FP_UNPACK_RAW_4_P_flo->bits.frac3; \ + X##_e = _FP_UNPACK_RAW_4_P_flo->bits.exp; \ + X##_s = _FP_UNPACK_RAW_4_P_flo->bits.sign; \ + } \ while (0) #define _FP_PACK_RAW_4(fs, val, X) \ do \ { \ - union _FP_UNION_##fs _flo; \ - _flo.bits.frac0 = X##_f[0]; \ - _flo.bits.frac1 = X##_f[1]; \ - _flo.bits.frac2 = X##_f[2]; \ - _flo.bits.frac3 = X##_f[3]; \ - _flo.bits.exp = X##_e; \ - _flo.bits.sign = X##_s; \ - (val) = _flo.flt; \ + union _FP_UNION_##fs _FP_PACK_RAW_4_flo; \ + _FP_PACK_RAW_4_flo.bits.frac0 = X##_f[0]; \ + _FP_PACK_RAW_4_flo.bits.frac1 = X##_f[1]; \ + _FP_PACK_RAW_4_flo.bits.frac2 = X##_f[2]; \ + _FP_PACK_RAW_4_flo.bits.frac3 = X##_f[3]; \ + _FP_PACK_RAW_4_flo.bits.exp = X##_e; \ + _FP_PACK_RAW_4_flo.bits.sign = X##_s; \ + (val) = _FP_PACK_RAW_4_flo.flt; \ } \ while (0) -#define _FP_PACK_RAW_4_P(fs, val, X) \ - do \ - { \ - union _FP_UNION_##fs *_flo = (union _FP_UNION_##fs *) (val); \ - \ - _flo->bits.frac0 = X##_f[0]; \ - _flo->bits.frac1 = X##_f[1]; \ - _flo->bits.frac2 = X##_f[2]; \ - _flo->bits.frac3 = X##_f[3]; \ - _flo->bits.exp = X##_e; \ - _flo->bits.sign = X##_s; \ - } \ +#define _FP_PACK_RAW_4_P(fs, val, X) \ + do \ + { \ + union _FP_UNION_##fs *_FP_PACK_RAW_4_P_flo \ + = (union _FP_UNION_##fs *) (val); \ + \ + _FP_PACK_RAW_4_P_flo->bits.frac0 = X##_f[0]; \ + _FP_PACK_RAW_4_P_flo->bits.frac1 = X##_f[1]; \ + _FP_PACK_RAW_4_P_flo->bits.frac2 = X##_f[2]; \ + _FP_PACK_RAW_4_P_flo->bits.frac3 = X##_f[3]; \ + _FP_PACK_RAW_4_P_flo->bits.exp = X##_e; \ + _FP_PACK_RAW_4_P_flo->bits.sign = X##_s; \ + } \ while (0) /* @@ -266,81 +299,126 @@ #define _FP_MUL_MEAT_DW_4_wide(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_2 (_b); \ - _FP_FRAC_DECL_2 (_c); \ - _FP_FRAC_DECL_2 (_d); \ - _FP_FRAC_DECL_2 (_e); \ - _FP_FRAC_DECL_2 (_f); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_4_wide_b); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_4_wide_c); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_4_wide_d); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_4_wide_e); \ + _FP_FRAC_DECL_2 (_FP_MUL_MEAT_DW_4_wide_f); \ \ - doit (_FP_FRAC_WORD_8 (R, 1), _FP_FRAC_WORD_8 (R, 0), X##_f[0], Y##_f[0]); \ - doit (_b_f1, _b_f0, X##_f[0], Y##_f[1]); \ - doit (_c_f1, _c_f0, X##_f[1], Y##_f[0]); \ - doit (_d_f1, _d_f0, X##_f[1], Y##_f[1]); \ - doit (_e_f1, _e_f0, X##_f[0], Y##_f[2]); \ - doit (_f_f1, _f_f0, X##_f[2], Y##_f[0]); \ + doit (_FP_FRAC_WORD_8 (R, 1), _FP_FRAC_WORD_8 (R, 0), \ + X##_f[0], Y##_f[0]); \ + doit (_FP_MUL_MEAT_DW_4_wide_b_f1, _FP_MUL_MEAT_DW_4_wide_b_f0, \ + X##_f[0], Y##_f[1]); \ + doit (_FP_MUL_MEAT_DW_4_wide_c_f1, _FP_MUL_MEAT_DW_4_wide_c_f0, \ + X##_f[1], Y##_f[0]); \ + doit (_FP_MUL_MEAT_DW_4_wide_d_f1, _FP_MUL_MEAT_DW_4_wide_d_f0, \ + X##_f[1], Y##_f[1]); \ + doit (_FP_MUL_MEAT_DW_4_wide_e_f1, _FP_MUL_MEAT_DW_4_wide_e_f0, \ + X##_f[0], Y##_f[2]); \ + doit (_FP_MUL_MEAT_DW_4_wide_f_f1, _FP_MUL_MEAT_DW_4_wide_f_f0, \ + X##_f[2], Y##_f[0]); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 3), _FP_FRAC_WORD_8 (R, 2), \ - _FP_FRAC_WORD_8 (R, 1), 0, _b_f1, _b_f0, \ + _FP_FRAC_WORD_8 (R, 1), 0, \ + _FP_MUL_MEAT_DW_4_wide_b_f1, \ + _FP_MUL_MEAT_DW_4_wide_b_f0, \ 0, 0, _FP_FRAC_WORD_8 (R, 1)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 3), _FP_FRAC_WORD_8 (R, 2), \ - _FP_FRAC_WORD_8 (R, 1), 0, _c_f1, _c_f0, \ + _FP_FRAC_WORD_8 (R, 1), 0, \ + _FP_MUL_MEAT_DW_4_wide_c_f1, \ + _FP_MUL_MEAT_DW_4_wide_c_f0, \ _FP_FRAC_WORD_8 (R, 3), _FP_FRAC_WORD_8 (R, 2), \ _FP_FRAC_WORD_8 (R, 1)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3), \ - _FP_FRAC_WORD_8 (R, 2), 0, _d_f1, _d_f0, \ + _FP_FRAC_WORD_8 (R, 2), 0, \ + _FP_MUL_MEAT_DW_4_wide_d_f1, \ + _FP_MUL_MEAT_DW_4_wide_d_f0, \ 0, _FP_FRAC_WORD_8 (R, 3), _FP_FRAC_WORD_8 (R, 2)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3), \ - _FP_FRAC_WORD_8 (R, 2), 0, _e_f1, _e_f0, \ + _FP_FRAC_WORD_8 (R, 2), 0, \ + _FP_MUL_MEAT_DW_4_wide_e_f1, \ + _FP_MUL_MEAT_DW_4_wide_e_f0, \ _FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3), \ _FP_FRAC_WORD_8 (R, 2)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3), \ - _FP_FRAC_WORD_8 (R, 2), 0, _f_f1, _f_f0, \ + _FP_FRAC_WORD_8 (R, 2), 0, \ + _FP_MUL_MEAT_DW_4_wide_f_f1, \ + _FP_MUL_MEAT_DW_4_wide_f_f0, \ _FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3), \ _FP_FRAC_WORD_8 (R, 2)); \ - doit (_b_f1, _b_f0, X##_f[0], Y##_f[3]); \ - doit (_c_f1, _c_f0, X##_f[3], Y##_f[0]); \ - doit (_d_f1, _d_f0, X##_f[1], Y##_f[2]); \ - doit (_e_f1, _e_f0, X##_f[2], Y##_f[1]); \ + doit (_FP_MUL_MEAT_DW_4_wide_b_f1, \ + _FP_MUL_MEAT_DW_4_wide_b_f0, X##_f[0], Y##_f[3]); \ + doit (_FP_MUL_MEAT_DW_4_wide_c_f1, \ + _FP_MUL_MEAT_DW_4_wide_c_f0, X##_f[3], Y##_f[0]); \ + doit (_FP_MUL_MEAT_DW_4_wide_d_f1, _FP_MUL_MEAT_DW_4_wide_d_f0, \ + X##_f[1], Y##_f[2]); \ + doit (_FP_MUL_MEAT_DW_4_wide_e_f1, _FP_MUL_MEAT_DW_4_wide_e_f0, \ + X##_f[2], Y##_f[1]); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ - _FP_FRAC_WORD_8 (R, 3), 0, _b_f1, _b_f0, \ + _FP_FRAC_WORD_8 (R, 3), 0, \ + _FP_MUL_MEAT_DW_4_wide_b_f1, \ + _FP_MUL_MEAT_DW_4_wide_b_f0, \ 0, _FP_FRAC_WORD_8 (R, 4), _FP_FRAC_WORD_8 (R, 3)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ - _FP_FRAC_WORD_8 (R, 3), 0, _c_f1, _c_f0, \ + _FP_FRAC_WORD_8 (R, 3), 0, \ + _FP_MUL_MEAT_DW_4_wide_c_f1, \ + _FP_MUL_MEAT_DW_4_wide_c_f0, \ _FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ _FP_FRAC_WORD_8 (R, 3)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ - _FP_FRAC_WORD_8 (R, 3), 0, _d_f1, _d_f0, \ + _FP_FRAC_WORD_8 (R, 3), 0, \ + _FP_MUL_MEAT_DW_4_wide_d_f1, \ + _FP_MUL_MEAT_DW_4_wide_d_f0, \ _FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ _FP_FRAC_WORD_8 (R, 3)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ - _FP_FRAC_WORD_8 (R, 3), 0, _e_f1, _e_f0, \ + _FP_FRAC_WORD_8 (R, 3), 0, \ + _FP_MUL_MEAT_DW_4_wide_e_f1, \ + _FP_MUL_MEAT_DW_4_wide_e_f0, \ _FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4), \ _FP_FRAC_WORD_8 (R, 3)); \ - doit (_b_f1, _b_f0, X##_f[2], Y##_f[2]); \ - doit (_c_f1, _c_f0, X##_f[1], Y##_f[3]); \ - doit (_d_f1, _d_f0, X##_f[3], Y##_f[1]); \ - doit (_e_f1, _e_f0, X##_f[2], Y##_f[3]); \ - doit (_f_f1, _f_f0, X##_f[3], Y##_f[2]); \ + doit (_FP_MUL_MEAT_DW_4_wide_b_f1, _FP_MUL_MEAT_DW_4_wide_b_f0, \ + X##_f[2], Y##_f[2]); \ + doit (_FP_MUL_MEAT_DW_4_wide_c_f1, _FP_MUL_MEAT_DW_4_wide_c_f0, \ + X##_f[1], Y##_f[3]); \ + doit (_FP_MUL_MEAT_DW_4_wide_d_f1, _FP_MUL_MEAT_DW_4_wide_d_f0, \ + X##_f[3], Y##_f[1]); \ + doit (_FP_MUL_MEAT_DW_4_wide_e_f1, _FP_MUL_MEAT_DW_4_wide_e_f0, \ + X##_f[2], Y##_f[3]); \ + doit (_FP_MUL_MEAT_DW_4_wide_f_f1, _FP_MUL_MEAT_DW_4_wide_f_f0, \ + X##_f[3], Y##_f[2]); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5), \ - _FP_FRAC_WORD_8 (R, 4), 0, _b_f1, _b_f0, \ + _FP_FRAC_WORD_8 (R, 4), 0, \ + _FP_MUL_MEAT_DW_4_wide_b_f1, \ + _FP_MUL_MEAT_DW_4_wide_b_f0, \ 0, _FP_FRAC_WORD_8 (R, 5), _FP_FRAC_WORD_8 (R, 4)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5), \ - _FP_FRAC_WORD_8 (R, 4), 0, _c_f1, _c_f0, \ + _FP_FRAC_WORD_8 (R, 4), 0, \ + _FP_MUL_MEAT_DW_4_wide_c_f1, \ + _FP_MUL_MEAT_DW_4_wide_c_f0, \ _FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5), \ _FP_FRAC_WORD_8 (R, 4)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5), \ - _FP_FRAC_WORD_8 (R, 4), 0, _d_f1, _d_f0, \ + _FP_FRAC_WORD_8 (R, 4), 0, \ + _FP_MUL_MEAT_DW_4_wide_d_f1, \ + _FP_MUL_MEAT_DW_4_wide_d_f0, \ _FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5), \ _FP_FRAC_WORD_8 (R, 4)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 7), _FP_FRAC_WORD_8 (R, 6), \ - _FP_FRAC_WORD_8 (R, 5), 0, _e_f1, _e_f0, \ + _FP_FRAC_WORD_8 (R, 5), 0, \ + _FP_MUL_MEAT_DW_4_wide_e_f1, \ + _FP_MUL_MEAT_DW_4_wide_e_f0, \ 0, _FP_FRAC_WORD_8 (R, 6), _FP_FRAC_WORD_8 (R, 5)); \ __FP_FRAC_ADD_3 (_FP_FRAC_WORD_8 (R, 7), _FP_FRAC_WORD_8 (R, 6), \ - _FP_FRAC_WORD_8 (R, 5), 0, _f_f1, _f_f0, \ + _FP_FRAC_WORD_8 (R, 5), 0, \ + _FP_MUL_MEAT_DW_4_wide_f_f1, \ + _FP_MUL_MEAT_DW_4_wide_f_f0, \ _FP_FRAC_WORD_8 (R, 7), _FP_FRAC_WORD_8 (R, 6), \ _FP_FRAC_WORD_8 (R, 5)); \ - doit (_b_f1, _b_f0, X##_f[3], Y##_f[3]); \ + doit (_FP_MUL_MEAT_DW_4_wide_b_f1, _FP_MUL_MEAT_DW_4_wide_b_f0, \ + X##_f[3], Y##_f[3]); \ __FP_FRAC_ADD_2 (_FP_FRAC_WORD_8 (R, 7), _FP_FRAC_WORD_8 (R, 6), \ - _b_f1, _b_f0, \ + _FP_MUL_MEAT_DW_4_wide_b_f1, \ + _FP_MUL_MEAT_DW_4_wide_b_f0, \ _FP_FRAC_WORD_8 (R, 7), _FP_FRAC_WORD_8 (R, 6)); \ } \ while (0) @@ -348,16 +426,19 @@ #define _FP_MUL_MEAT_4_wide(wfracbits, R, X, Y, doit) \ do \ { \ - _FP_FRAC_DECL_8 (_z); \ + _FP_FRAC_DECL_8 (_FP_MUL_MEAT_4_wide_z); \ \ - _FP_MUL_MEAT_DW_4_wide (wfracbits, _z, X, Y, doit); \ + _FP_MUL_MEAT_DW_4_wide (wfracbits, _FP_MUL_MEAT_4_wide_z, \ + X, Y, doit); \ \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_8 (_z, wfracbits-1, 2*wfracbits); \ - __FP_FRAC_SET_4 (R, _FP_FRAC_WORD_8 (_z, 3), _FP_FRAC_WORD_8 (_z, 2), \ - _FP_FRAC_WORD_8 (_z, 1), _FP_FRAC_WORD_8 (_z, 0)); \ + _FP_FRAC_SRS_8 (_FP_MUL_MEAT_4_wide_z, wfracbits-1, 2*wfracbits); \ + __FP_FRAC_SET_4 (R, _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_wide_z, 3), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_wide_z, 2), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_wide_z, 1), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_wide_z, 0)); \ } \ while (0) @@ -371,16 +452,18 @@ #define _FP_MUL_MEAT_4_gmp(wfracbits, R, X, Y) \ do \ { \ - _FP_FRAC_DECL_8 (_z); \ + _FP_FRAC_DECL_8 (_FP_MUL_MEAT_4_gmp_z); \ \ - _FP_MUL_MEAT_DW_4_gmp (wfracbits, _z, X, Y); \ + _FP_MUL_MEAT_DW_4_gmp (wfracbits, _FP_MUL_MEAT_4_gmp_z, X, Y); \ \ /* Normalize since we know where the msb of the multiplicands \ were (bit B), we know that the msb of the of the product is \ at either 2B or 2B-1. */ \ - _FP_FRAC_SRS_8 (_z, wfracbits-1, 2*wfracbits); \ - __FP_FRAC_SET_4 (R, _FP_FRAC_WORD_8 (_z, 3), _FP_FRAC_WORD_8 (_z, 2), \ - _FP_FRAC_WORD_8 (_z, 1), _FP_FRAC_WORD_8 (_z, 0)); \ + _FP_FRAC_SRS_8 (_FP_MUL_MEAT_4_gmp_z, wfracbits-1, 2*wfracbits); \ + __FP_FRAC_SET_4 (R, _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_gmp_z, 3), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_gmp_z, 2), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_gmp_z, 1), \ + _FP_FRAC_WORD_8 (_FP_MUL_MEAT_4_gmp_z, 0)); \ } \ while (0) @@ -391,12 +474,12 @@ #define umul_ppppmnnn(p3, p2, p1, p0, m, n2, n1, n0) \ do \ { \ - UWtype _t; \ + UWtype umul_ppppmnnn_t; \ umul_ppmm (p1, p0, m, n0); \ - umul_ppmm (p2, _t, m, n1); \ - __FP_FRAC_ADDI_2 (p2, p1, _t); \ - umul_ppmm (p3, _t, m, n2); \ - __FP_FRAC_ADDI_2 (p3, p2, _t); \ + umul_ppmm (p2, umul_ppppmnnn_t, m, n1); \ + __FP_FRAC_ADDI_2 (p2, p1, umul_ppppmnnn_t); \ + umul_ppmm (p3, umul_ppppmnnn_t, m, n2); \ + __FP_FRAC_ADDI_2 (p3, p2, umul_ppppmnnn_t); \ } \ while (0) @@ -407,13 +490,14 @@ #define _FP_DIV_MEAT_4_udiv(fs, R, X, Y) \ do \ { \ - int _i; \ - _FP_FRAC_DECL_4 (_n); \ - _FP_FRAC_DECL_4 (_m); \ - _FP_FRAC_SET_4 (_n, _FP_ZEROFRAC_4); \ + int _FP_DIV_MEAT_4_udiv_i; \ + _FP_FRAC_DECL_4 (_FP_DIV_MEAT_4_udiv_n); \ + _FP_FRAC_DECL_4 (_FP_DIV_MEAT_4_udiv_m); \ + _FP_FRAC_SET_4 (_FP_DIV_MEAT_4_udiv_n, _FP_ZEROFRAC_4); \ if (_FP_FRAC_GE_4 (X, Y)) \ { \ - _n_f[3] = X##_f[0] << (_FP_W_TYPE_SIZE - 1); \ + _FP_DIV_MEAT_4_udiv_n_f[3] \ + = X##_f[0] << (_FP_W_TYPE_SIZE - 1); \ _FP_FRAC_SRL_4 (X, 1); \ } \ else \ @@ -423,49 +507,58 @@ denominator set. */ \ _FP_FRAC_SLL_4 (Y, _FP_WFRACXBITS_##fs); \ \ - for (_i = 3; ; _i--) \ + for (_FP_DIV_MEAT_4_udiv_i = 3; ; _FP_DIV_MEAT_4_udiv_i--) \ { \ if (X##_f[3] == Y##_f[3]) \ { \ /* This is a special case, not an optimization \ (X##_f[3]/Y##_f[3] would not fit into UWtype). \ - As X## is guaranteed to be < Y, R##_f[_i] can be either \ + As X## is guaranteed to be < Y, \ + R##_f[_FP_DIV_MEAT_4_udiv_i] can be either \ (UWtype)-1 or (UWtype)-2. */ \ - R##_f[_i] = -1; \ - if (!_i) \ + R##_f[_FP_DIV_MEAT_4_udiv_i] = -1; \ + if (!_FP_DIV_MEAT_4_udiv_i) \ break; \ __FP_FRAC_SUB_4 (X##_f[3], X##_f[2], X##_f[1], X##_f[0], \ Y##_f[2], Y##_f[1], Y##_f[0], 0, \ - X##_f[2], X##_f[1], X##_f[0], _n_f[_i]); \ + X##_f[2], X##_f[1], X##_f[0], \ + _FP_DIV_MEAT_4_udiv_n_f[_FP_DIV_MEAT_4_udiv_i]); \ _FP_FRAC_SUB_4 (X, Y, X); \ if (X##_f[3] > Y##_f[3]) \ { \ - R##_f[_i] = -2; \ + R##_f[_FP_DIV_MEAT_4_udiv_i] = -2; \ _FP_FRAC_ADD_4 (X, Y, X); \ } \ } \ else \ { \ - udiv_qrnnd (R##_f[_i], X##_f[3], X##_f[3], X##_f[2], Y##_f[3]); \ - umul_ppppmnnn (_m_f[3], _m_f[2], _m_f[1], _m_f[0], \ - R##_f[_i], Y##_f[2], Y##_f[1], Y##_f[0]); \ + udiv_qrnnd (R##_f[_FP_DIV_MEAT_4_udiv_i], \ + X##_f[3], X##_f[3], X##_f[2], Y##_f[3]); \ + umul_ppppmnnn (_FP_DIV_MEAT_4_udiv_m_f[3], \ + _FP_DIV_MEAT_4_udiv_m_f[2], \ + _FP_DIV_MEAT_4_udiv_m_f[1], \ + _FP_DIV_MEAT_4_udiv_m_f[0], \ + R##_f[_FP_DIV_MEAT_4_udiv_i], \ + Y##_f[2], Y##_f[1], Y##_f[0]); \ X##_f[2] = X##_f[1]; \ X##_f[1] = X##_f[0]; \ - X##_f[0] = _n_f[_i]; \ - if (_FP_FRAC_GT_4 (_m, X)) \ + X##_f[0] \ + = _FP_DIV_MEAT_4_udiv_n_f[_FP_DIV_MEAT_4_udiv_i]; \ + if (_FP_FRAC_GT_4 (_FP_DIV_MEAT_4_udiv_m, X)) \ { \ - R##_f[_i]--; \ + R##_f[_FP_DIV_MEAT_4_udiv_i]--; \ _FP_FRAC_ADD_4 (X, Y, X); \ - if (_FP_FRAC_GE_4 (X, Y) && _FP_FRAC_GT_4 (_m, X)) \ + if (_FP_FRAC_GE_4 (X, Y) \ + && _FP_FRAC_GT_4 (_FP_DIV_MEAT_4_udiv_m, X)) \ { \ - R##_f[_i]--; \ + R##_f[_FP_DIV_MEAT_4_udiv_i]--; \ _FP_FRAC_ADD_4 (X, Y, X); \ } \ } \ - _FP_FRAC_DEC_4 (X, _m); \ - if (!_i) \ + _FP_FRAC_DEC_4 (X, _FP_DIV_MEAT_4_udiv_m); \ + if (!_FP_DIV_MEAT_4_udiv_i) \ { \ - if (!_FP_FRAC_EQ_4 (X, _m)) \ + if (!_FP_FRAC_EQ_4 (X, _FP_DIV_MEAT_4_udiv_m)) \ R##_f[0] |= _FP_WORK_STICKY; \ break; \ } \ @@ -591,18 +684,19 @@ # define __FP_FRAC_ADD_4(r3, r2, r1, r0, x3, x2, x1, x0, y3, y2, y1, y0) \ do \ { \ - _FP_W_TYPE _c1, _c2, _c3; \ + _FP_W_TYPE __FP_FRAC_ADD_4_c1, __FP_FRAC_ADD_4_c2; \ + _FP_W_TYPE __FP_FRAC_ADD_4_c3; \ r0 = x0 + y0; \ - _c1 = r0 < x0; \ + __FP_FRAC_ADD_4_c1 = r0 < x0; \ r1 = x1 + y1; \ - _c2 = r1 < x1; \ - r1 += _c1; \ - _c2 |= r1 < _c1; \ + __FP_FRAC_ADD_4_c2 = r1 < x1; \ + r1 += __FP_FRAC_ADD_4_c1; \ + __FP_FRAC_ADD_4_c2 |= r1 < __FP_FRAC_ADD_4_c1; \ r2 = x2 + y2; \ - _c3 = r2 < x2; \ - r2 += _c2; \ - _c3 |= r2 < _c2; \ - r3 = x3 + y3 + _c3; \ + __FP_FRAC_ADD_4_c3 = r2 < x2; \ + r2 += __FP_FRAC_ADD_4_c2; \ + __FP_FRAC_ADD_4_c3 |= r2 < __FP_FRAC_ADD_4_c2; \ + r3 = x3 + y3 + __FP_FRAC_ADD_4_c3; \ } \ while (0) #endif @@ -611,14 +705,14 @@ # define __FP_FRAC_SUB_3(r2, r1, r0, x2, x1, x0, y2, y1, y0) \ do \ { \ - _FP_W_TYPE _c1, _c2; \ + _FP_W_TYPE __FP_FRAC_SUB_3_c1, __FP_FRAC_SUB_3_c2; \ r0 = x0 - y0; \ - _c1 = r0 > x0; \ + __FP_FRAC_SUB_3_c1 = r0 > x0; \ r1 = x1 - y1; \ - _c2 = r1 > x1; \ - r1 -= _c1; \ - _c2 |= _c1 && (y1 == x1); \ - r2 = x2 - y2 - _c2; \ + __FP_FRAC_SUB_3_c2 = r1 > x1; \ + r1 -= __FP_FRAC_SUB_3_c1; \ + __FP_FRAC_SUB_3_c2 |= __FP_FRAC_SUB_3_c1 && (y1 == x1); \ + r2 = x2 - y2 - __FP_FRAC_SUB_3_c2; \ } \ while (0) #endif @@ -627,18 +721,19 @@ # define __FP_FRAC_SUB_4(r3, r2, r1, r0, x3, x2, x1, x0, y3, y2, y1, y0) \ do \ { \ - _FP_W_TYPE _c1, _c2, _c3; \ + _FP_W_TYPE __FP_FRAC_SUB_4_c1, __FP_FRAC_SUB_4_c2; \ + _FP_W_TYPE __FP_FRAC_SUB_4_c3; \ r0 = x0 - y0; \ - _c1 = r0 > x0; \ + __FP_FRAC_SUB_4_c1 = r0 > x0; \ r1 = x1 - y1; \ - _c2 = r1 > x1; \ - r1 -= _c1; \ - _c2 |= _c1 && (y1 == x1); \ + __FP_FRAC_SUB_4_c2 = r1 > x1; \ + r1 -= __FP_FRAC_SUB_4_c1; \ + __FP_FRAC_SUB_4_c2 |= __FP_FRAC_SUB_4_c1 && (y1 == x1); \ r2 = x2 - y2; \ - _c3 = r2 > x2; \ - r2 -= _c2; \ - _c3 |= _c2 && (y2 == x2); \ - r3 = x3 - y3 - _c3; \ + __FP_FRAC_SUB_4_c3 = r2 > x2; \ + r2 -= __FP_FRAC_SUB_4_c2; \ + __FP_FRAC_SUB_4_c3 |= __FP_FRAC_SUB_4_c2 && (y2 == x2); \ + r3 = x3 - y3 - __FP_FRAC_SUB_4_c3; \ } \ while (0) #endif @@ -647,36 +742,47 @@ # define __FP_FRAC_DEC_3(x2, x1, x0, y2, y1, y0) \ do \ { \ - UWtype _t0, _t1, _t2; \ - _t0 = x0, _t1 = x1, _t2 = x2; \ - __FP_FRAC_SUB_3 (x2, x1, x0, _t2, _t1, _t0, y2, y1, y0); \ + UWtype __FP_FRAC_DEC_3_t0, __FP_FRAC_DEC_3_t1; \ + UWtype __FP_FRAC_DEC_3_t2; \ + __FP_FRAC_DEC_3_t0 = x0; \ + __FP_FRAC_DEC_3_t1 = x1; \ + __FP_FRAC_DEC_3_t2 = x2; \ + __FP_FRAC_SUB_3 (x2, x1, x0, __FP_FRAC_DEC_3_t2, \ + __FP_FRAC_DEC_3_t1, __FP_FRAC_DEC_3_t0, \ + y2, y1, y0); \ } \ while (0) #endif #ifndef __FP_FRAC_DEC_4 -# define __FP_FRAC_DEC_4(x3, x2, x1, x0, y3, y2, y1, y0) \ - do \ - { \ - UWtype _t0, _t1, _t2, _t3; \ - _t0 = x0, _t1 = x1, _t2 = x2, _t3 = x3; \ - __FP_FRAC_SUB_4 (x3, x2, x1, x0, _t3, _t2, _t1, _t0, y3, y2, y1, y0); \ - } \ +# define __FP_FRAC_DEC_4(x3, x2, x1, x0, y3, y2, y1, y0) \ + do \ + { \ + UWtype __FP_FRAC_DEC_4_t0, __FP_FRAC_DEC_4_t1; \ + UWtype __FP_FRAC_DEC_4_t2, __FP_FRAC_DEC_4_t3; \ + __FP_FRAC_DEC_4_t0 = x0; \ + __FP_FRAC_DEC_4_t1 = x1; \ + __FP_FRAC_DEC_4_t2 = x2; \ + __FP_FRAC_DEC_4_t3 = x3; \ + __FP_FRAC_SUB_4 (x3, x2, x1, x0, __FP_FRAC_DEC_4_t3, \ + __FP_FRAC_DEC_4_t2, __FP_FRAC_DEC_4_t1, \ + __FP_FRAC_DEC_4_t0, y3, y2, y1, y0); \ + } \ while (0) #endif #ifndef __FP_FRAC_ADDI_4 -# define __FP_FRAC_ADDI_4(x3, x2, x1, x0, i) \ - do \ - { \ - UWtype _t; \ - _t = ((x0 += i) < i); \ - x1 += _t; \ - _t = (x1 < _t); \ - x2 += _t; \ - _t = (x2 < _t); \ - x3 += _t; \ - } \ +# define __FP_FRAC_ADDI_4(x3, x2, x1, x0, i) \ + do \ + { \ + UWtype __FP_FRAC_ADDI_4_t; \ + __FP_FRAC_ADDI_4_t = ((x0 += i) < i); \ + x1 += __FP_FRAC_ADDI_4_t; \ + __FP_FRAC_ADDI_4_t = (x1 < __FP_FRAC_ADDI_4_t); \ + x2 += __FP_FRAC_ADDI_4_t; \ + __FP_FRAC_ADDI_4_t = (x2 < __FP_FRAC_ADDI_4_t); \ + x3 += __FP_FRAC_ADDI_4_t; \ + } \ while (0) #endif diff --git a/soft-fp/op-8.h b/soft-fp/op-8.h index a50cd7b..c966ee2 100644 --- a/soft-fp/op-8.h +++ b/soft-fp/op-8.h @@ -36,48 +36,66 @@ #define _FP_FRAC_LOW_8(X) (X##_f[0]) #define _FP_FRAC_WORD_8(X, w) (X##_f[w]) -#define _FP_FRAC_SLL_8(X, N) \ - do \ - { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _up = (N) % _FP_W_TYPE_SIZE; \ - _down = _FP_W_TYPE_SIZE - _up; \ - if (!_up) \ - for (_i = 7; _i >= _skip; --_i) \ - X##_f[_i] = X##_f[_i-_skip]; \ - else \ - { \ - for (_i = 7; _i > _skip; --_i) \ - X##_f[_i] = (X##_f[_i-_skip] << _up \ - | X##_f[_i-_skip-1] >> _down); \ - X##_f[_i--] = X##_f[0] << _up; \ - } \ - for (; _i >= 0; --_i) \ - X##_f[_i] = 0; \ - } \ +#define _FP_FRAC_SLL_8(X, N) \ + do \ + { \ + _FP_I_TYPE _FP_FRAC_SLL_8_up, _FP_FRAC_SLL_8_down; \ + _FP_I_TYPE _FP_FRAC_SLL_8_skip, _FP_FRAC_SLL_8_i; \ + _FP_FRAC_SLL_8_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SLL_8_up = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SLL_8_down = _FP_W_TYPE_SIZE - _FP_FRAC_SLL_8_up; \ + if (!_FP_FRAC_SLL_8_up) \ + for (_FP_FRAC_SLL_8_i = 7; \ + _FP_FRAC_SLL_8_i >= _FP_FRAC_SLL_8_skip; \ + --_FP_FRAC_SLL_8_i) \ + X##_f[_FP_FRAC_SLL_8_i] \ + = X##_f[_FP_FRAC_SLL_8_i-_FP_FRAC_SLL_8_skip]; \ + else \ + { \ + for (_FP_FRAC_SLL_8_i = 7; \ + _FP_FRAC_SLL_8_i > _FP_FRAC_SLL_8_skip; \ + --_FP_FRAC_SLL_8_i) \ + X##_f[_FP_FRAC_SLL_8_i] \ + = ((X##_f[_FP_FRAC_SLL_8_i-_FP_FRAC_SLL_8_skip] \ + << _FP_FRAC_SLL_8_up) \ + | (X##_f[_FP_FRAC_SLL_8_i-_FP_FRAC_SLL_8_skip-1] \ + >> _FP_FRAC_SLL_8_down)); \ + X##_f[_FP_FRAC_SLL_8_i--] = X##_f[0] << _FP_FRAC_SLL_8_up; \ + } \ + for (; _FP_FRAC_SLL_8_i >= 0; --_FP_FRAC_SLL_8_i) \ + X##_f[_FP_FRAC_SLL_8_i] = 0; \ + } \ while (0) -#define _FP_FRAC_SRL_8(X, N) \ - do \ - { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _down = (N) % _FP_W_TYPE_SIZE; \ - _up = _FP_W_TYPE_SIZE - _down; \ - if (!_down) \ - for (_i = 0; _i <= 7-_skip; ++_i) \ - X##_f[_i] = X##_f[_i+_skip]; \ - else \ - { \ - for (_i = 0; _i < 7-_skip; ++_i) \ - X##_f[_i] = (X##_f[_i+_skip] >> _down \ - | X##_f[_i+_skip+1] << _up); \ - X##_f[_i++] = X##_f[7] >> _down; \ - } \ - for (; _i < 8; ++_i) \ - X##_f[_i] = 0; \ - } \ +#define _FP_FRAC_SRL_8(X, N) \ + do \ + { \ + _FP_I_TYPE _FP_FRAC_SRL_8_up, _FP_FRAC_SRL_8_down; \ + _FP_I_TYPE _FP_FRAC_SRL_8_skip, _FP_FRAC_SRL_8_i; \ + _FP_FRAC_SRL_8_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRL_8_down = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRL_8_up = _FP_W_TYPE_SIZE - _FP_FRAC_SRL_8_down; \ + if (!_FP_FRAC_SRL_8_down) \ + for (_FP_FRAC_SRL_8_i = 0; \ + _FP_FRAC_SRL_8_i <= 7-_FP_FRAC_SRL_8_skip; \ + ++_FP_FRAC_SRL_8_i) \ + X##_f[_FP_FRAC_SRL_8_i] \ + = X##_f[_FP_FRAC_SRL_8_i+_FP_FRAC_SRL_8_skip]; \ + else \ + { \ + for (_FP_FRAC_SRL_8_i = 0; \ + _FP_FRAC_SRL_8_i < 7-_FP_FRAC_SRL_8_skip; \ + ++_FP_FRAC_SRL_8_i) \ + X##_f[_FP_FRAC_SRL_8_i] \ + = ((X##_f[_FP_FRAC_SRL_8_i+_FP_FRAC_SRL_8_skip] \ + >> _FP_FRAC_SRL_8_down) \ + | (X##_f[_FP_FRAC_SRL_8_i+_FP_FRAC_SRL_8_skip+1] \ + << _FP_FRAC_SRL_8_up)); \ + X##_f[_FP_FRAC_SRL_8_i++] = X##_f[7] >> _FP_FRAC_SRL_8_down; \ + } \ + for (; _FP_FRAC_SRL_8_i < 8; ++_FP_FRAC_SRL_8_i) \ + X##_f[_FP_FRAC_SRL_8_i] = 0; \ + } \ while (0) @@ -89,28 +107,40 @@ #define _FP_FRAC_SRS_8(X, N, size) \ do \ { \ - _FP_I_TYPE _up, _down, _skip, _i; \ - _FP_W_TYPE _s; \ - _skip = (N) / _FP_W_TYPE_SIZE; \ - _down = (N) % _FP_W_TYPE_SIZE; \ - _up = _FP_W_TYPE_SIZE - _down; \ - for (_s = _i = 0; _i < _skip; ++_i) \ - _s |= X##_f[_i]; \ - if (!_down) \ - for (_i = 0; _i <= 7-_skip; ++_i) \ - X##_f[_i] = X##_f[_i+_skip]; \ + _FP_I_TYPE _FP_FRAC_SRS_8_up, _FP_FRAC_SRS_8_down; \ + _FP_I_TYPE _FP_FRAC_SRS_8_skip, _FP_FRAC_SRS_8_i; \ + _FP_W_TYPE _FP_FRAC_SRS_8_s; \ + _FP_FRAC_SRS_8_skip = (N) / _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRS_8_down = (N) % _FP_W_TYPE_SIZE; \ + _FP_FRAC_SRS_8_up = _FP_W_TYPE_SIZE - _FP_FRAC_SRS_8_down; \ + for (_FP_FRAC_SRS_8_s = _FP_FRAC_SRS_8_i = 0; \ + _FP_FRAC_SRS_8_i < _FP_FRAC_SRS_8_skip; \ + ++_FP_FRAC_SRS_8_i) \ + _FP_FRAC_SRS_8_s |= X##_f[_FP_FRAC_SRS_8_i]; \ + if (!_FP_FRAC_SRS_8_down) \ + for (_FP_FRAC_SRS_8_i = 0; \ + _FP_FRAC_SRS_8_i <= 7-_FP_FRAC_SRS_8_skip; \ + ++_FP_FRAC_SRS_8_i) \ + X##_f[_FP_FRAC_SRS_8_i] \ + = X##_f[_FP_FRAC_SRS_8_i+_FP_FRAC_SRS_8_skip]; \ else \ { \ - _s |= X##_f[_i] << _up; \ - for (_i = 0; _i < 7-_skip; ++_i) \ - X##_f[_i] = (X##_f[_i+_skip] >> _down \ - | X##_f[_i+_skip+1] << _up); \ - X##_f[_i++] = X##_f[7] >> _down; \ + _FP_FRAC_SRS_8_s \ + |= X##_f[_FP_FRAC_SRS_8_i] << _FP_FRAC_SRS_8_up; \ + for (_FP_FRAC_SRS_8_i = 0; \ + _FP_FRAC_SRS_8_i < 7-_FP_FRAC_SRS_8_skip; \ + ++_FP_FRAC_SRS_8_i) \ + X##_f[_FP_FRAC_SRS_8_i] \ + = ((X##_f[_FP_FRAC_SRS_8_i+_FP_FRAC_SRS_8_skip] \ + >> _FP_FRAC_SRS_8_down) \ + | (X##_f[_FP_FRAC_SRS_8_i+_FP_FRAC_SRS_8_skip+1] \ + << _FP_FRAC_SRS_8_up)); \ + X##_f[_FP_FRAC_SRS_8_i++] = X##_f[7] >> _FP_FRAC_SRS_8_down; \ } \ - for (; _i < 8; ++_i) \ - X##_f[_i] = 0; \ + for (; _FP_FRAC_SRS_8_i < 8; ++_FP_FRAC_SRS_8_i) \ + X##_f[_FP_FRAC_SRS_8_i] = 0; \ /* don't fix the LSB until the very end when we're sure f[0] is \ stable */ \ - X##_f[0] |= (_s != 0); \ + X##_f[0] |= (_FP_FRAC_SRS_8_s != 0); \ } \ while (0) |