summaryrefslogtreecommitdiff
path: root/src/math
diff options
context:
space:
mode:
authorSzabolcs Nagy <nsz@port70.net>2013-11-24 01:06:38 +0000
committerSzabolcs Nagy <nsz@port70.net>2013-11-24 01:06:38 +0000
commit3fdf94ec5101fab63d0e8196d3e557641a0e19e2 (patch)
treea8684dad2f216505e97263187b768013648146d0 /src/math
parent10c8b7148b918938d8e681c5801b913dd56cb7e4 (diff)
downloadmusl-3fdf94ec5101fab63d0e8196d3e557641a0e19e2.tar.gz
math: clean up __rem_pio2
- remove the HAVE_EFFICIENT_IRINT case: fn is an exact integer, so it can be converted to int32_t a bit more efficiently than with a cast (the rounding mode change can be avoided), but musl does not support this case on any arch. - __rem_pio2: use double_t where possible - __rem_pio2f: use less assignments to avoid stores on i386 - use unsigned int bit manipulation (and union instead of macros) - use hexfloat literals instead of named constants
Diffstat (limited to 'src/math')
-rw-r--r--src/math/__rem_pio2.c72
-rw-r--r--src/math/__rem_pio2_large.c18
-rw-r--r--src/math/__rem_pio2f.c34
3 files changed, 53 insertions, 71 deletions
diff --git a/src/math/__rem_pio2.c b/src/math/__rem_pio2.c
index 9305be5f..5fafc4a8 100644
--- a/src/math/__rem_pio2.c
+++ b/src/math/__rem_pio2.c
@@ -29,7 +29,6 @@
* pio2_3t: pi/2 - (pio2_1+pio2_2+pio2_3)
*/
static const double
-two24 = 1.67772160000000000000e+07, /* 0x41700000, 0x00000000 */
invpio2 = 6.36619772367581382433e-01, /* 0x3FE45F30, 0x6DC9C883 */
pio2_1 = 1.57079632673412561417e+00, /* 0x3FF921FB, 0x54400000 */
pio2_1t = 6.07710050650619224932e-11, /* 0x3DD0B461, 0x1A626331 */
@@ -41,18 +40,19 @@ pio2_3t = 8.47842766036889956997e-32; /* 0x397B839A, 0x252049C1 */
/* caller must handle the case when reduction is not needed: |x| ~<= pi/4 */
int __rem_pio2(double x, double *y)
{
- double z,w,t,r,fn;
- double tx[3],ty[2];
- int32_t e0,i,j,nx,n,ix,hx;
- uint32_t low;
+ union {double f; uint64_t i;} u = {x};
+ double_t z,w,t,r;
+ double tx[3],ty[2],fn;
+ uint32_t ix;
+ int sign, n, ex, ey, i;
- GET_HIGH_WORD(hx,x);
- ix = hx & 0x7fffffff;
+ sign = u.i>>63;
+ ix = u.i>>32 & 0x7fffffff;
if (ix <= 0x400f6a7a) { /* |x| ~<= 5pi/4 */
if ((ix & 0xfffff) == 0x921fb) /* |x| ~= pi/2 or 2pi/2 */
goto medium; /* cancellation -- use medium case */
if (ix <= 0x4002d97c) { /* |x| ~<= 3pi/4 */
- if (hx > 0) {
+ if (!sign) {
z = x - pio2_1; /* one round good to 85 bits */
y[0] = z - pio2_1t;
y[1] = (z-y[0]) - pio2_1t;
@@ -64,7 +64,7 @@ int __rem_pio2(double x, double *y)
return -1;
}
} else {
- if (hx > 0) {
+ if (!sign) {
z = x - 2*pio2_1;
y[0] = z - 2*pio2_1t;
y[1] = (z-y[0]) - 2*pio2_1t;
@@ -81,7 +81,7 @@ int __rem_pio2(double x, double *y)
if (ix <= 0x4015fdbc) { /* |x| ~<= 7pi/4 */
if (ix == 0x4012d97c) /* |x| ~= 3pi/2 */
goto medium;
- if (hx > 0) {
+ if (!sign) {
z = x - 3*pio2_1;
y[0] = z - 3*pio2_1t;
y[1] = (z-y[0]) - 3*pio2_1t;
@@ -95,7 +95,7 @@ int __rem_pio2(double x, double *y)
} else {
if (ix == 0x401921fb) /* |x| ~= 4pi/2 */
goto medium;
- if (hx > 0) {
+ if (!sign) {
z = x - 4*pio2_1;
y[0] = z - 4*pio2_1t;
y[1] = (z-y[0]) - 4*pio2_1t;
@@ -109,32 +109,26 @@ int __rem_pio2(double x, double *y)
}
}
if (ix < 0x413921fb) { /* |x| ~< 2^20*(pi/2), medium size */
- uint32_t high;
medium:
- /* Use a specialized rint() to get fn. Assume round-to-nearest. */
+ /* rint(x/(pi/2)), Assume round-to-nearest. */
fn = x*invpio2 + 0x1.8p52;
fn = fn - 0x1.8p52;
-// FIXME
-#ifdef HAVE_EFFICIENT_IRINT
- n = irint(fn);
-#else
n = (int32_t)fn;
-#endif
r = x - fn*pio2_1;
w = fn*pio2_1t; /* 1st round, good to 85 bits */
- j = ix>>20;
y[0] = r - w;
- GET_HIGH_WORD(high,y[0]);
- i = j - ((high>>20)&0x7ff);
- if (i > 16) { /* 2nd round, good to 118 bits */
+ u.f = y[0];
+ ey = u.i>>52 & 0x7ff;
+ ex = ix>>20;
+ if (ex - ey > 16) { /* 2nd round, good to 118 bits */
t = r;
w = fn*pio2_2;
r = t - w;
w = fn*pio2_2t - ((t-r)-w);
y[0] = r - w;
- GET_HIGH_WORD(high,y[0]);
- i = j - ((high>>20)&0x7ff);
- if (i > 49) { /* 3rd round, good to 151 bits, covers all cases */
+ u.f = y[0];
+ ey = u.i>>52 & 0x7ff;
+ if (ex - ey > 49) { /* 3rd round, good to 151 bits, covers all cases */
t = r;
w = fn*pio2_3;
r = t - w;
@@ -142,7 +136,7 @@ medium:
y[0] = r - w;
}
}
- y[1] = (r-y[0]) - w;
+ y[1] = (r - y[0]) - w;
return n;
}
/*
@@ -152,19 +146,21 @@ medium:
y[0] = y[1] = x - x;
return 0;
}
- /* set z = scalbn(|x|,ilogb(x)-23) */
- GET_LOW_WORD(low,x);
- e0 = (ix>>20) - 1046; /* e0 = ilogb(z)-23; */
- INSERT_WORDS(z, ix - ((int32_t)(e0<<20)), low);
- for (i=0; i<2; i++) {
- tx[i] = (double)((int32_t)(z));
- z = (z-tx[i])*two24;
+ /* set z = scalbn(|x|,-ilogb(x)+23) */
+ u.f = x;
+ u.i &= (uint64_t)-1>>12;
+ u.i |= (uint64_t)(0x3ff + 23)<<52;
+ z = u.f;
+ for (i=0; i < 2; i++) {
+ tx[i] = (double)(int32_t)z;
+ z = (z-tx[i])*0x1p24;
}
- tx[2] = z;
- nx = 3;
- while (tx[nx-1] == 0.0) nx--; /* skip zero term */
- n = __rem_pio2_large(tx,ty,e0,nx,1);
- if (hx < 0) {
+ tx[i] = z;
+ /* skip zero terms, first term is non-zero */
+ while (tx[i] == 0.0)
+ i--;
+ n = __rem_pio2_large(tx,ty,(int)(ix>>20)-(0x3ff+23),i+1,1);
+ if (sign) {
y[0] = -ty[0];
y[1] = -ty[1];
return -n;
diff --git a/src/math/__rem_pio2_large.c b/src/math/__rem_pio2_large.c
index bb2dc43f..958f28c2 100644
--- a/src/math/__rem_pio2_large.c
+++ b/src/math/__rem_pio2_large.c
@@ -270,10 +270,6 @@ static const double PIo2[] = {
2.16741683877804819444e-51, /* 0x3569F31D, 0x00000000 */
};
-static const double
-two24 = 1.67772160000000000000e+07, /* 0x41700000, 0x00000000 */
-twon24 = 5.96046447753906250000e-08; /* 0x3E700000, 0x00000000 */
-
int __rem_pio2_large(double *x, double *y, int e0, int nx, int prec)
{
int32_t jz,jx,jv,jp,jk,carry,n,iq[20],i,j,k,m,q0,ih;
@@ -304,8 +300,8 @@ int __rem_pio2_large(double *x, double *y, int e0, int nx, int prec)
recompute:
/* distill q[] into iq[] reversingly */
for (i=0,j=jz,z=q[jz]; j>0; i++,j--) {
- fw = (double)((int32_t)(twon24* z));
- iq[i] = (int32_t)(z-two24*fw);
+ fw = (double)(int32_t)(0x1p-24*z);
+ iq[i] = (int32_t)(z - 0x1p24*fw);
z = q[j-1]+fw;
}
@@ -330,7 +326,7 @@ recompute:
if (carry == 0) {
if (j != 0) {
carry = 1;
- iq[i] = 0x1000000- j;
+ iq[i] = 0x1000000 - j;
}
} else
iq[i] = 0xffffff - j;
@@ -378,9 +374,9 @@ recompute:
}
} else { /* break z into 24-bit if necessary */
z = scalbn(z,-q0);
- if (z >= two24) {
- fw = (double)((int32_t)(twon24*z));
- iq[jz] = (int32_t)(z-two24*fw);
+ if (z >= 0x1p24) {
+ fw = (double)(int32_t)(0x1p-24*z);
+ iq[jz] = (int32_t)(z - 0x1p24*fw);
jz += 1;
q0 += 24;
iq[jz] = (int32_t)fw;
@@ -392,7 +388,7 @@ recompute:
fw = scalbn(1.0,q0);
for (i=jz; i>=0; i--) {
q[i] = fw*(double)iq[i];
- fw *= twon24;
+ fw *= 0x1p-24;
}
/* compute PIo2[0,...,jp]*q[jz,...,0] */
diff --git a/src/math/__rem_pio2f.c b/src/math/__rem_pio2f.c
index 5bdeb529..838e1fca 100644
--- a/src/math/__rem_pio2f.c
+++ b/src/math/__rem_pio2f.c
@@ -34,42 +34,32 @@ pio2_1t = 1.58932547735281966916e-08; /* 0x3E5110b4, 0x611A6263 */
int __rem_pio2f(float x, double *y)
{
- double w,r,fn;
- double tx[1],ty[1];
- float z;
- int32_t e0,n,ix,hx;
+ union {float f; uint32_t i;} u = {x};
+ double tx[1],ty[1],fn;
+ uint32_t ix;
+ int n, sign, e0;
- GET_FLOAT_WORD(hx, x);
- ix = hx & 0x7fffffff;
+ ix = u.i & 0x7fffffff;
/* 25+53 bit pi is good enough for medium size */
if (ix < 0x4dc90fdb) { /* |x| ~< 2^28*(pi/2), medium size */
/* Use a specialized rint() to get fn. Assume round-to-nearest. */
fn = x*invpio2 + 0x1.8p52;
fn = fn - 0x1.8p52;
-// FIXME
-#ifdef HAVE_EFFICIENT_IRINT
- n = irint(fn);
-#else
n = (int32_t)fn;
-#endif
- r = x - fn*pio2_1;
- w = fn*pio2_1t;
- *y = r - w;
+ *y = x - fn*pio2_1 - fn*pio2_1t;
return n;
}
- /*
- * all other (large) arguments
- */
if(ix>=0x7f800000) { /* x is inf or NaN */
*y = x-x;
return 0;
}
- /* set z = scalbn(|x|,ilogb(|x|)-23) */
- e0 = (ix>>23) - 150; /* e0 = ilogb(|x|)-23; */
- SET_FLOAT_WORD(z, ix - ((int32_t)(e0<<23)));
- tx[0] = z;
+ /* scale x into [2^23, 2^24-1] */
+ sign = u.i>>31;
+ e0 = (ix>>23) - (0x7f+23); /* e0 = ilogb(|x|)-23, positive */
+ u.i = ix - (e0<<23);
+ tx[0] = u.f;
n = __rem_pio2_large(tx,ty,e0,1,0);
- if (hx < 0) {
+ if (sign) {
*y = -ty[0];
return -n;
}