Added SCE SIMD math library in Extras/simdmathlibrary
The upcoming vectormath that will used to speed up the SPU version of Extras/BulletMultiThreaded depends on this.
This commit is contained in:
ejcoumans
356
Extras/simdmathlibrary/spu/remquod2.c
Normal file
356
Extras/simdmathlibrary/spu/remquod2.c
Normal file
@@ -0,0 +1,356 @@
|
||||
/* remquod2 -
|
||||
Copyright (C) 2006, 2007 Sony Computer Entertainment Inc.
|
||||
All rights reserved.
|
||||
|
||||
Redistribution and use in source and binary forms,
|
||||
with or without modification, are permitted provided that the
|
||||
following conditions are met:
|
||||
* Redistributions of source code must retain the above copyright
|
||||
notice, this list of conditions and the following disclaimer.
|
||||
* Redistributions in binary form must reproduce the above copyright
|
||||
notice, this list of conditions and the following disclaimer in the
|
||||
documentation and/or other materials provided with the distribution.
|
||||
* Neither the name of the Sony Computer Entertainment Inc nor the names
|
||||
of its contributors may be used to endorse or promote products derived
|
||||
from this software without specific prior written permission.
|
||||
|
||||
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
||||
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
||||
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
|
||||
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
||||
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
||||
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
||||
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
||||
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
||||
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
||||
POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
#include <simdmath.h>
|
||||
#include <spu_intrinsics.h>
|
||||
|
||||
/*
|
||||
* This function returns the same vector double result as remainderd2().
|
||||
* In addition a vector signed long long is storedin *pquo,
|
||||
* that contains the corresponding element values whose sign is
|
||||
* the sign of xi / yi and whose magnitude is congruent modulo 2n to
|
||||
* the magnitude of the integral quotient of xi / yi, where n is
|
||||
* an implementation-defined integer greater than or equal to 3.
|
||||
*/
|
||||
|
||||
static inline vec_uint4 _sub_d_(vec_uint4 aa, vec_uint4 bb);
|
||||
static inline vec_uint4 _vec_gt64_half(vec_uint4 aa, vec_uint4 bb);
|
||||
static inline vec_uint4 _vec_gt64(vec_uint4 aa, vec_uint4 bb);
|
||||
static inline vec_uint4 _twice(vec_uint4 aa);
|
||||
|
||||
vector double
|
||||
remquod2(vector double x, vector double yy, vector signed long long *quo)
|
||||
{
|
||||
vec_uchar16 splat_hi = ((vec_uchar16){ 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11});
|
||||
vec_int4 quotient, quotient0;
|
||||
vec_uint4 y_hi;
|
||||
vec_uint4 abs_x, abs_yy, abs_2x, abs_8y, abs_4y, abs_2y;
|
||||
vec_uint4 bias;
|
||||
vec_uint4 nan_out, not_ge, quo_pos, overflow;
|
||||
vec_uint4 result;
|
||||
vec_uint4 half_smax = spu_splats((unsigned int)0x7FEFFFFF);
|
||||
vec_uint4 sign_mask = (vec_uint4)(spu_splats(0x8000000000000000ULL));
|
||||
vec_uint4 exp_mask = (vec_uint4)(spu_splats(0x7FF0000000000000ULL));
|
||||
vec_uint4 val_nan = (vec_uint4)(spu_splats(0x7FF8000000000000ULL));
|
||||
vec_uint4 vec_zero = spu_splats((unsigned int)0);
|
||||
vec_uint4 is_zeroy;
|
||||
|
||||
// cut sign
|
||||
abs_x = spu_andc((vec_uint4)x, sign_mask);
|
||||
abs_yy = spu_andc((vec_uint4)yy, sign_mask);
|
||||
y_hi = spu_shuffle(abs_yy, abs_yy, splat_hi);
|
||||
|
||||
quo_pos = spu_cmpgt((vec_int4)spu_and((vec_uint4)spu_xor(x, yy), sign_mask), -1);
|
||||
quo_pos = spu_shuffle(quo_pos, quo_pos, splat_hi);
|
||||
|
||||
// check nan out
|
||||
is_zeroy = spu_cmpeq(abs_yy, vec_zero);
|
||||
is_zeroy = spu_and(is_zeroy, spu_rlqwbyte(is_zeroy, 4));
|
||||
nan_out = _vec_gt64_half(abs_yy, exp_mask); // y > 7FF00000
|
||||
nan_out = spu_or(nan_out, spu_cmpgt(abs_x, half_smax)); // x >= 7FF0000000000000
|
||||
nan_out = spu_or(nan_out, is_zeroy); // y = 0
|
||||
nan_out = spu_shuffle(nan_out, nan_out, splat_hi);
|
||||
|
||||
|
||||
// make y x8
|
||||
abs_2y = _twice(abs_yy); // 2 x y
|
||||
abs_4y = _twice(abs_2y); // 4 x y
|
||||
abs_8y = _twice(abs_4y); // 2 x y
|
||||
|
||||
/*
|
||||
* use fmodd2 function
|
||||
*/
|
||||
// get remainder of y x8
|
||||
// result = (vec_uint4)_fmodd2( x, (vec_double2)abs_8y);
|
||||
{
|
||||
vec_double2 y = (vec_double2)abs_8y;
|
||||
|
||||
int shiftx0, shiftx1, shifty0, shifty1;
|
||||
vec_uchar16 swap_words = ((vec_uchar16){ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11});
|
||||
vec_uchar16 propagate = ((vec_uchar16){ 4,5,6,7, 192,192,192,192, 12,13,14,15, 192,192,192,192});
|
||||
// vec_uchar16 splat_hi = ((vec_uchar16){ 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11});
|
||||
vec_int4 n, shift;
|
||||
vec_uint4 exp_x, exp_y;
|
||||
// , sign;
|
||||
// vec_uint4 abs_x, abs_y;
|
||||
vec_uint4 abs_y;
|
||||
vec_uint4 mant_x, mant_x0, mant_x1;
|
||||
vec_uint4 mant_y, mant_y0, mant_y1;
|
||||
vec_uint4 mant_0, mant_1;
|
||||
vec_uint4 mant_r, mant_l;
|
||||
// vec_uint4 result;
|
||||
vec_uint4 result0, resultx;
|
||||
vec_uint4 zero_x, zero_y;
|
||||
vec_uint4 denorm_x, denorm_y;
|
||||
vec_uint4 cnt, cnt_x, cnt_y;
|
||||
vec_uint4 shift_x, shift_y;
|
||||
vec_uint4 adj_x, adj_y;
|
||||
vec_uint4 z, borrow, mask;
|
||||
vec_uint4 lsb = (vec_uint4)(spu_splats(0x0000000000000001ULL));
|
||||
// vec_uint4 sign_mask = (vec_uint4)(spu_splats(0x8000000000000000ULL));
|
||||
vec_uint4 implied_1 = (vec_uint4)(spu_splats(0x0010000000000000ULL));
|
||||
vec_uint4 mant_mask = (vec_uint4)(spu_splats(0x000FFFFFFFFFFFFFULL));
|
||||
// vec_uint4 exp_mask = (vec_uint4)(spu_splats(0x7FF0000000000000ULL));
|
||||
vec_uint4 merge_sel = ((vec_uint4){0,0,-1,-1});
|
||||
// vec_uint4 vec_zero = spu_splats((unsigned int)0);
|
||||
|
||||
// sign = spu_and( (vec_uint4)x, sign_mask);
|
||||
// abs_x = spu_andc((vec_uint4)x, sign_mask);
|
||||
abs_y = spu_andc((vec_uint4)y, sign_mask);
|
||||
exp_x = spu_rlmask(abs_x, -20);
|
||||
exp_y = spu_rlmask(abs_y, -20);
|
||||
// get shift count for denorm
|
||||
cnt_x = spu_cntlz(abs_x);
|
||||
cnt_y = spu_cntlz(abs_y);
|
||||
cnt_x = spu_add(cnt_x, spu_sel( vec_zero, spu_rlqwbyte(cnt_x, 4), spu_cmpeq(cnt_x, 32)));
|
||||
cnt_y = spu_add(cnt_y, spu_sel( vec_zero, spu_rlqwbyte(cnt_y, 4), spu_cmpeq(cnt_y, 32)));
|
||||
|
||||
zero_x = spu_cmpgt(cnt_x, 63); // zero ?
|
||||
zero_y = spu_cmpgt(cnt_y, 63); // zero ?
|
||||
result0 = spu_or(zero_x, zero_y);
|
||||
result0 = spu_shuffle(result0, result0, splat_hi);
|
||||
|
||||
// 0 - (cnt_x - 11) = 11 - cnt_x
|
||||
shift_x= spu_add(cnt_x, -11);
|
||||
shift_y= spu_add(cnt_y, -11);
|
||||
cnt_x = spu_sub(11, cnt_x);
|
||||
cnt_y = spu_sub(11, cnt_y);
|
||||
|
||||
// count to normalize
|
||||
adj_x = spu_sel(spu_add(exp_x, -1), cnt_x, spu_cmpeq(exp_x, 0));
|
||||
adj_y = spu_sel(spu_add(exp_y, -1), cnt_y, spu_cmpeq(exp_y, 0));
|
||||
adj_x = spu_shuffle(adj_x, adj_x, splat_hi);
|
||||
adj_y = spu_shuffle(adj_y, adj_y, splat_hi);
|
||||
|
||||
// for denorm
|
||||
shiftx0 = spu_extract(shift_x, 0);
|
||||
shiftx1 = spu_extract(shift_x, 2);
|
||||
shifty0 = spu_extract(shift_y, 0);
|
||||
shifty1 = spu_extract(shift_y, 2);
|
||||
mant_x0 = spu_slqwbytebc( spu_slqw(spu_and(abs_x,((vec_uint4){-1,-1,0,0})),shiftx0), shiftx0);
|
||||
mant_y0 = spu_slqwbytebc( spu_slqw(spu_and(abs_y,((vec_uint4){-1,-1,0,0})),shifty0), shifty0);
|
||||
mant_x1 = spu_slqwbytebc( spu_slqw(abs_x,shiftx1), shiftx1);
|
||||
mant_y1 = spu_slqwbytebc( spu_slqw(abs_y,shifty1), shifty1);
|
||||
mant_x = spu_sel(mant_x0, mant_x1, merge_sel);
|
||||
mant_y = spu_sel(mant_y0, mant_y1, merge_sel);
|
||||
|
||||
denorm_x = spu_cmpgt((vec_int4)vec_zero, (vec_int4)adj_x);
|
||||
denorm_y = spu_cmpgt((vec_int4)vec_zero, (vec_int4)adj_y);
|
||||
mant_x = spu_sel(spu_and(abs_x, mant_mask), mant_x, denorm_x);
|
||||
mant_y = spu_sel(spu_and(abs_y, mant_mask), mant_y, denorm_y);
|
||||
mant_x = spu_or(mant_x, implied_1); // hidden bit
|
||||
mant_y = spu_or(mant_y, implied_1); // hidden bit
|
||||
|
||||
// x < y ?
|
||||
resultx = _vec_gt64(abs_y, abs_x);
|
||||
|
||||
n = spu_sub((vec_int4)adj_x, (vec_int4)adj_y);
|
||||
mask = spu_cmpgt(n, 0);
|
||||
mask = spu_andc(mask, resultx);
|
||||
|
||||
while (spu_extract(spu_gather(mask), 0)) {
|
||||
borrow = spu_genb(mant_x, mant_y);
|
||||
borrow = spu_shuffle(borrow, borrow, propagate);
|
||||
z = spu_subx(mant_x, mant_y, borrow);
|
||||
|
||||
result0 = spu_or(spu_and(spu_cmpeq(spu_or(z, spu_shuffle(z, z, swap_words)), 0), mask), result0);
|
||||
|
||||
mant_x = spu_sel(mant_x,
|
||||
spu_sel(spu_slqw(mant_x, 1), spu_andc(spu_slqw(z, 1), lsb), spu_cmpgt((vec_int4)spu_shuffle(z, z, splat_hi), -1)),
|
||||
mask);
|
||||
|
||||
n = spu_add(n, -1);
|
||||
mask = spu_cmpgt(n, 0);
|
||||
}
|
||||
|
||||
borrow = spu_genb(mant_x, mant_y);
|
||||
borrow = spu_shuffle(borrow, borrow, propagate);
|
||||
z = spu_subx(mant_x, mant_y, borrow);
|
||||
mant_x = spu_sel(mant_x, z, spu_cmpgt((vec_int4)spu_shuffle(z, z, splat_hi), -1));
|
||||
result0 = spu_or(spu_cmpeq(spu_or(mant_x, spu_shuffle(mant_x, mant_x, swap_words)), 0), result0);
|
||||
|
||||
// bring back to original range
|
||||
mant_0 = spu_and(mant_x, ((vec_uint4){0x001FFFFF,-1,0,0}));
|
||||
mant_1 = spu_and(mant_x, ((vec_uint4){0,0,0x001FFFFF,-1}));
|
||||
|
||||
// for adj_y < 0 exp max=1
|
||||
shiftx0 = spu_extract(adj_y, 0);
|
||||
shiftx1 = spu_extract(adj_y, 2);
|
||||
mant_x0 = spu_rlmaskqwbytebc(spu_rlmaskqw(mant_0, shiftx0), 7 + shiftx0);
|
||||
mant_x1 = spu_rlmaskqwbytebc(spu_rlmaskqw(mant_1, shiftx1), 7 + shiftx1);
|
||||
mant_r = spu_sel(mant_x0, mant_x1, merge_sel);
|
||||
|
||||
// for adj_y >= 0
|
||||
cnt = spu_cntlz(mant_x);
|
||||
cnt = spu_add(cnt, spu_sel( vec_zero, spu_rlqwbyte(cnt, 4), spu_cmpeq(cnt, 32)));
|
||||
cnt = spu_add(cnt, -11);
|
||||
cnt = spu_sel(vec_zero, cnt, spu_cmpgt(cnt, 0)); // for exp >= 1
|
||||
shift = (vec_int4)spu_sel(cnt, adj_y, spu_cmpgt(cnt, adj_y));
|
||||
shiftx0 = spu_extract(shift, 0);
|
||||
shiftx1 = spu_extract(shift, 2);
|
||||
mant_x0 = spu_slqwbytebc(spu_slqw(mant_0, shiftx0), shiftx0);
|
||||
mant_x1 = spu_slqwbytebc(spu_slqw(mant_1, shiftx1), shiftx1);
|
||||
mant_l = spu_sel(mant_x0, mant_x1, merge_sel);
|
||||
cnt = spu_sub(adj_y, (vec_uint4)shift);
|
||||
mant_l = spu_add(mant_l, spu_and(spu_rl(cnt,20), exp_mask));
|
||||
|
||||
result = spu_sel(mant_l, mant_r, denorm_y);
|
||||
result = spu_sel(result, vec_zero, result0); // reminder 0
|
||||
result = spu_sel(result, abs_x, resultx); // x < y
|
||||
// result = spu_xor(result, sign); // set sign
|
||||
|
||||
// return ((vec_double2)result);
|
||||
}
|
||||
|
||||
// if y (x8->exp+3 7FF-7FC) overflow
|
||||
// abs_x = spu_sel(spu_andc(result, sign_mask), abs_x, spu_cmpgt(y_hi, spu_splats((unsigned int)0x7FBFFFFF)));
|
||||
abs_x = spu_sel(result, abs_x, spu_cmpgt(y_hi, spu_splats((unsigned int)0x7FBFFFFF)));
|
||||
|
||||
/* if (x >= 4*y)
|
||||
* x -= 4*y
|
||||
* quotient = 4
|
||||
* else
|
||||
* quotient = 0
|
||||
*/
|
||||
overflow = spu_cmpgt(y_hi, spu_splats((unsigned int)0x7FCFFFFF));
|
||||
|
||||
not_ge = _vec_gt64(abs_4y, abs_x);
|
||||
not_ge = spu_or(not_ge, overflow);
|
||||
abs_x = spu_sel(_sub_d_(abs_x, abs_4y), abs_x, not_ge);
|
||||
quotient = spu_andc(spu_splats((int)4), (vec_int4)not_ge);
|
||||
|
||||
/* if (x >= 2*y
|
||||
* x -= 2*y
|
||||
* quotient += 2
|
||||
*/
|
||||
overflow = spu_cmpgt(y_hi, spu_splats((unsigned int)0x7FDFFFFF));
|
||||
|
||||
not_ge = _vec_gt64(abs_2y, abs_x); // abs_2y > abs_x
|
||||
not_ge = spu_or(not_ge, overflow);
|
||||
|
||||
abs_x = spu_sel(_sub_d_(abs_x, abs_2y), abs_x, not_ge);
|
||||
quotient = spu_sel(spu_add(quotient, 2), quotient, not_ge);
|
||||
|
||||
/* if (2*x > y)
|
||||
* x -= y
|
||||
* if (2*x >= y) x -= y
|
||||
*/
|
||||
overflow = spu_cmpgt(y_hi, spu_splats((unsigned int)0x7FEFFFFF));
|
||||
// make x2
|
||||
abs_2x = _twice(abs_x); // 2 x x
|
||||
|
||||
bias = _vec_gt64(abs_2x, abs_yy); // abs_2x > abs_yy
|
||||
bias = spu_andc(bias, overflow);
|
||||
|
||||
abs_x = spu_sel(abs_x, _sub_d_(abs_x, abs_yy), bias);
|
||||
quotient = spu_sub(quotient, (vec_int4)bias);
|
||||
|
||||
overflow = spu_or(overflow, spu_shuffle(spu_rlmaska(abs_x, -31), vec_zero, splat_hi)); // minous
|
||||
|
||||
// make x2
|
||||
abs_2x = _twice(spu_andc(abs_x, sign_mask)); // 2 x x unsupport minous
|
||||
bias = spu_andc(bias, spu_rlmaska(_sub_d_(abs_2x, abs_yy), -31));
|
||||
bias = spu_andc(spu_shuffle(bias, bias, splat_hi), overflow);
|
||||
abs_x = spu_sel(abs_x, _sub_d_(abs_x, abs_yy), bias);
|
||||
quotient = spu_sub(quotient, (vec_int4)bias);
|
||||
|
||||
/* select final answer
|
||||
*/
|
||||
result = spu_xor(abs_x, spu_and((vec_uint4)x, sign_mask)); // set sign
|
||||
result = spu_sel(result, val_nan, nan_out); // if nan
|
||||
|
||||
quotient = spu_and(quotient, ((vec_int4){0,7,0,7})); // limit to 3bit
|
||||
quotient0 = spu_subx( (vec_int4)vec_zero, quotient, spu_rlqwbyte(spu_genb((vec_int4)vec_zero,quotient),4));
|
||||
quotient = spu_sel(quotient0, quotient, quo_pos);
|
||||
|
||||
*quo = (vec_llong2)quotient;
|
||||
|
||||
return ((vec_double2)result);
|
||||
}
|
||||
|
||||
/*
|
||||
* subtraction function in limited confdition
|
||||
*/
|
||||
static inline vec_uint4 _sub_d_(vec_uint4 aa, vec_uint4 bb)
|
||||
{
|
||||
// which is bigger input aa or bb
|
||||
vec_uint4 is_bigb = _vec_gt64(bb, aa); // bb > aa
|
||||
|
||||
// need denorm calc ?
|
||||
vec_uint4 norm_a, norm_b;
|
||||
norm_a = spu_cmpgt(aa, (vec_uint4)(spu_splats(0x000FFFFFFFFFFFFFULL)));
|
||||
norm_b = spu_cmpgt(bb, (vec_uint4)(spu_splats(0x000FFFFFFFFFFFFFULL)));
|
||||
norm_a = spu_and(norm_a, norm_b);
|
||||
norm_a = spu_shuffle(norm_a, norm_a,((vec_uchar16){ 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11}));
|
||||
|
||||
// calc (aa - bb) and (bb - aa)
|
||||
vec_uint4 res_a, res_b, res;
|
||||
vec_uint4 borrow_a, borrow_b;
|
||||
vec_uchar16 mask_b = ((vec_uchar16){4,5,6,7,192,192,192,192,12,13,14,15,192,192,192,192});
|
||||
borrow_a = spu_genb(aa, bb);
|
||||
borrow_b = spu_genb(bb, aa);
|
||||
borrow_a = spu_shuffle(borrow_a, borrow_a, mask_b);
|
||||
borrow_b = spu_shuffle(borrow_b, borrow_b, mask_b);
|
||||
res_a = spu_subx(aa, bb, borrow_a);
|
||||
res_b = spu_subx(bb, aa, borrow_b);
|
||||
res_b = spu_or(res_b, ((vec_uint4){0x80000000,0,0x80000000,0})); // set sign
|
||||
|
||||
res = spu_sel(res_a, res_b, is_bigb); // select (aa - bb) or (bb - aa)
|
||||
// select normal calc or special
|
||||
res = spu_sel(res, (vec_uint4)spu_sub((vec_double2)aa, (vec_double2)bb), norm_a);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
|
||||
/*
|
||||
* extend spu_cmpgt function to 64bit data
|
||||
*/
|
||||
static inline vec_uint4 _vec_gt64_half(vec_uint4 aa, vec_uint4 bb)
|
||||
{
|
||||
vec_uint4 gt = spu_cmpgt(aa, bb); // aa > bb
|
||||
vec_uint4 eq = spu_cmpeq(aa, bb); // aa = bb
|
||||
return spu_or(gt, spu_and(eq, spu_rlqwbyte(gt, 4))); // only higher is right
|
||||
}
|
||||
static inline vec_uint4 _vec_gt64(vec_uint4 aa, vec_uint4 bb)
|
||||
{
|
||||
vec_uint4 gt_hi = _vec_gt64_half(aa, bb); // only higher is right
|
||||
return spu_shuffle(gt_hi, gt_hi, ((vec_uchar16){ 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11}));
|
||||
}
|
||||
|
||||
/*
|
||||
* double formated x2
|
||||
*/
|
||||
static inline vec_uint4 _twice(vec_uint4 aa)
|
||||
{
|
||||
vec_uint4 norm = spu_cmpgt(aa, (vec_uint4)(spu_splats(0x000FFFFFFFFFFFFFULL))); // exp > 0
|
||||
norm = spu_shuffle(norm, norm, ((vec_uchar16){ 0,1,2,3,0,1,2,3, 8,9,10,11, 8,9,10,11}));
|
||||
|
||||
// if denorm or zero << 1 , if norm exp + 1
|
||||
return spu_sel(spu_slqw(aa, 1), spu_add(aa, (vec_uint4)(spu_splats(0x0010000000000000ULL))), norm); // x2
|
||||
}
|
||||
Reference in New Issue
Block a user