AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-12-28 05:21:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

aarch64: Improve vecmath sin routines

Browse Source 
* Update ULP comment reflecting a new observed max in [-pi/2, pi/2]
* Use the same polynomial in AdvSIMD and SVE, rather than FTRIG instructions
* Improve register use near special-case branch

Also use overloaded intrinsics for SVE.
This commit is contained in:
Joe Ramsay 2023-10-05 10:31:38 +01:00 committed by Szabolcs Nagy
parent 820948edd9
commit 8014d1e832
3 changed files with 86 additions and 72 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
sysdeps/aarch64/fpu/sin_advsimd.c
View File

@ -24,7 +24,6 @@ static const struct data
float64x2_t poly[7]; float64x2_t poly[7];
float64x2_t range_val, inv_pi, shift, pi_1, pi_2, pi_3; float64x2_t range_val, inv_pi, shift, pi_1, pi_2, pi_3;
} data = { } data = {
/* Worst-case error is 2.8 ulp in [-pi/2, pi/2]. */
.poly = { V2 (-0x1.555555555547bp-3), V2 (0x1.1111111108a4dp-7), .poly = { V2 (-0x1.555555555547bp-3), V2 (0x1.1111111108a4dp-7),
V2 (-0x1.a01a019936f27p-13), V2 (0x1.71de37a97d93ep-19), V2 (-0x1.a01a019936f27p-13), V2 (0x1.71de37a97d93ep-19),
V2 (-0x1.ae633919987c6p-26), V2 (0x1.60e277ae07cecp-33), V2 (-0x1.ae633919987c6p-26), V2 (0x1.60e277ae07cecp-33),
@ -52,6 +51,15 @@ special_case (float64x2_t x, float64x2_t y, uint64x2_t odd, uint64x2_t cmp)
return v_call_f64 (sin, x, y, cmp); return v_call_f64 (sin, x, y, cmp);
} }
/* Vector (AdvSIMD) sin approximation.
Maximum observed error in [-pi/2, pi/2], where argument is not reduced,
is 2.87 ULP:
_ZGVnN2v_sin (0x1.921d5c6a07142p+0) got 0x1.fffffffa7dc02p-1
want 0x1.fffffffa7dc05p-1
Maximum observed error in the entire non-special domain ([-2^23, 2^23])
is 3.22 ULP:
_ZGVnN2v_sin (0x1.5702447b6f17bp+22) got 0x1.ffdcd125c84fbp-3
want 0x1.ffdcd125c84f8p-3. */
float64x2_t VPCS_ATTR V_NAME_D1 (sin) (float64x2_t x) float64x2_t VPCS_ATTR V_NAME_D1 (sin) (float64x2_t x)
{ {
const struct data *d = ptr_barrier (&data); const struct data *d = ptr_barrier (&data);

104
sysdeps/aarch64/fpu/sin_sve.c
View File

@ -21,20 +21,22 @@
static const struct data static const struct data
{ {
double inv_pi, half_pi, inv_pi_over_2, pi_over_2_1, pi_over_2_2, pi_over_2_3, double inv_pi, pi_1, pi_2, pi_3, shift, range_val;
shift; double poly[7];
} data = { } data = {
/* Polynomial coefficients are hard-wired in the FTMAD instruction. */ .poly = { -0x1.555555555547bp-3, 0x1.1111111108a4dp-7, -0x1.a01a019936f27p-13,
0x1.71de37a97d93ep-19, -0x1.ae633919987c6p-26,
0x1.60e277ae07cecp-33, -0x1.9e9540300a1p-41, },
.inv_pi = 0x1.45f306dc9c883p-2, .inv_pi = 0x1.45f306dc9c883p-2,
.half_pi = 0x1.921fb54442d18p+0, .pi_1 = 0x1.921fb54442d18p+1,
.inv_pi_over_2 = 0x1.45f306dc9c882p-1, .pi_2 = 0x1.1a62633145c06p-53,
.pi_over_2_1 = 0x1.921fb50000000p+0, .pi_3 = 0x1.c1cd129024e09p-106,
.pi_over_2_2 = 0x1.110b460000000p-26, .shift = 0x1.8p52,
.pi_over_2_3 = 0x1.1a62633145c07p-54, .range_val = 0x1p23,
.shift = 0x1.8p52
}; };
#define RangeVal 0x4160000000000000 /* asuint64 (0x1p23). */ #define C(i) sv_f64 (d->poly[i])
static svfloat64_t NOINLINE static svfloat64_t NOINLINE
special_case (svfloat64_t x, svfloat64_t y, svbool_t cmp) special_case (svfloat64_t x, svfloat64_t y, svbool_t cmp)
@ -42,56 +44,58 @@ special_case (svfloat64_t x, svfloat64_t y, svbool_t cmp)
return sv_call_f64 (sin, x, y, cmp); return sv_call_f64 (sin, x, y, cmp);
} }
/* A fast SVE implementation of sin based on trigonometric /* A fast SVE implementation of sin.
instructions (FTMAD, FTSSEL, FTSMUL). Maximum observed error in [-pi/2, pi/2], where argument is not reduced,
Maximum observed error in 2.52 ULP: is 2.87 ULP:
SV_NAME_D1 (sin)(0x1.2d2b00df69661p+19) got 0x1.10ace8f3e786bp-40 _ZGVsMxv_sin (0x1.921d5c6a07142p+0) got 0x1.fffffffa7dc02p-1
want 0x1.10ace8f3e7868p-40. */ want 0x1.fffffffa7dc05p-1
Maximum observed error in the entire non-special domain ([-2^23, 2^23])
is 3.22 ULP:
_ZGVsMxv_sin (0x1.5702447b6f17bp+22) got 0x1.ffdcd125c84fbp-3
want 0x1.ffdcd125c84f8p-3. */
svfloat64_t SV_NAME_D1 (sin) (svfloat64_t x, const svbool_t pg) svfloat64_t SV_NAME_D1 (sin) (svfloat64_t x, const svbool_t pg)
{ {
const struct data *d = ptr_barrier (&data); const struct data *d = ptr_barrier (&data);
svfloat64_t r = svabs_f64_x (pg, x); /* Load some values in quad-word chunks to minimise memory access. */
svuint64_t sign const svbool_t ptrue = svptrue_b64 ();
= sveor_u64_x (pg, svreinterpret_u64_f64 (x), svreinterpret_u64_f64 (r)); svfloat64_t shift = sv_f64 (d->shift);
svbool_t cmp = svcmpge_n_u64 (pg, svreinterpret_u64_f64 (r), RangeVal); svfloat64_t inv_pi_and_pi1 = svld1rq (ptrue, &d->inv_pi);
svfloat64_t pi2_and_pi3 = svld1rq (ptrue, &d->pi_2);
/* Load first two pio2-related constants to one vector. */ /* n = rint(|x|/pi). */
svfloat64_t invpio2_and_pio2_1 svfloat64_t n = svmla_lane (shift, x, inv_pi_and_pi1, 0);
= svld1rq_f64 (svptrue_b64 (), &d->inv_pi_over_2); svuint64_t odd = svlsl_x (pg, svreinterpret_u64 (n), 63);
n = svsub_x (pg, n, shift);
/* n = rint(|x|/(pi/2)). */ /* r = |x| - n*(pi/2) (range reduction into -pi/2 .. pi/2). */
svfloat64_t q = svmla_lane_f64 (sv_f64 (d->shift), r, invpio2_and_pio2_1, 0); svfloat64_t r = x;
svfloat64_t n = svsub_n_f64_x (pg, q, d->shift); r = svmls_lane (r, n, inv_pi_and_pi1, 1);
r = svmls_lane (r, n, pi2_and_pi3, 0);
/* r = |x| - n*(pi/2) (range reduction into -pi/4 .. pi/4). */ r = svmls_lane (r, n, pi2_and_pi3, 1);
r = svmls_lane_f64 (r, n, invpio2_and_pio2_1, 1);
r = svmls_n_f64_x (pg, r, n, d->pi_over_2_2);
r = svmls_n_f64_x (pg, r, n, d->pi_over_2_3);
/* Final multiplicative factor: 1.0 or x depending on bit #0 of q. */
svfloat64_t f = svtssel_f64 (r, svreinterpret_u64_f64 (q));
/* sin(r) poly approx. */ /* sin(r) poly approx. */
svfloat64_t r2 = svtsmul_f64 (r, svreinterpret_u64_f64 (q)); svfloat64_t r2 = svmul_x (pg, r, r);
svfloat64_t y = sv_f64 (0.0); svfloat64_t r3 = svmul_x (pg, r2, r);
y = svtmad_f64 (y, r2, 7); svfloat64_t r4 = svmul_x (pg, r2, r2);
y = svtmad_f64 (y, r2, 6);
y = svtmad_f64 (y, r2, 5);
y = svtmad_f64 (y, r2, 4);
y = svtmad_f64 (y, r2, 3);
y = svtmad_f64 (y, r2, 2);
y = svtmad_f64 (y, r2, 1);
y = svtmad_f64 (y, r2, 0);
/* Apply factor. */ svfloat64_t t1 = svmla_x (pg, C (4), C (5), r2);
y = svmul_f64_x (pg, f, y); svfloat64_t t2 = svmla_x (pg, C (2), C (3), r2);
svfloat64_t t3 = svmla_x (pg, C (0), C (1), r2);
/* sign = y^sign. */ svfloat64_t y = svmla_x (pg, t1, C (6), r4);
y = svreinterpret_f64_u64 ( y = svmla_x (pg, t2, y, r4);
sveor_u64_x (pg, svreinterpret_u64_f64 (y), sign)); y = svmla_x (pg, t3, y, r4);
y = svmla_x (pg, r, y, r3);
svbool_t cmp = svacle (pg, x, d->range_val);
cmp = svnot_z (pg, cmp);
if (__glibc_unlikely (svptest_any (pg, cmp))) if (__glibc_unlikely (svptest_any (pg, cmp)))
return special_case (x, y, cmp); return special_case (x,
return y; svreinterpret_f64 (sveor_z (
svnot_z (pg, cmp), svreinterpret_u64 (y), odd)),
cmp);
/* Copy sign. */
return svreinterpret_f64 (sveor_z (pg, svreinterpret_u64 (y), odd));
} }

44
sysdeps/aarch64/fpu/sinf_sve.c
View File

@ -23,7 +23,7 @@ static const struct data
{ {
float poly[4]; float poly[4];
/* Pi-related values to be loaded as one quad-word and used with /* Pi-related values to be loaded as one quad-word and used with
svmla_lane_f32. */ svmla_lane. */
float negpi1, negpi2, negpi3, invpi; float negpi1, negpi2, negpi3, invpi;
float shift; float shift;
} data = { } data = {
@ -57,40 +57,42 @@ svfloat32_t SV_NAME_F1 (sin) (svfloat32_t x, const svbool_t pg)
{ {
const struct data *d = ptr_barrier (&data); const struct data *d = ptr_barrier (&data);
svfloat32_t ax = svabs_f32_x (pg, x); svfloat32_t ax = svabs_x (pg, x);
svuint32_t sign = sveor_u32_x (pg, svreinterpret_u32_f32 (x), svuint32_t sign
svreinterpret_u32_f32 (ax)); = sveor_x (pg, svreinterpret_u32 (x), svreinterpret_u32 (ax));
svbool_t cmp = svcmpge_n_u32 (pg, svreinterpret_u32_f32 (ax), RangeVal); svbool_t cmp = svcmpge (pg, svreinterpret_u32 (ax), RangeVal);
/* pi_vals are a quad-word of helper values - the first 3 elements contain /* pi_vals are a quad-word of helper values - the first 3 elements contain
-pi in extended precision, the last contains 1 / pi. */ -pi in extended precision, the last contains 1 / pi. */
svfloat32_t pi_vals = svld1rq_f32 (svptrue_b32 (), &d->negpi1); svfloat32_t pi_vals = svld1rq (svptrue_b32 (), &d->negpi1);
/* n = rint(|x|/pi). */ /* n = rint(|x|/pi). */
svfloat32_t n = svmla_lane_f32 (sv_f32 (d->shift), ax, pi_vals, 3); svfloat32_t n = svmla_lane (sv_f32 (d->shift), ax, pi_vals, 3);
svuint32_t odd = svlsl_n_u32_x (pg, svreinterpret_u32_f32 (n), 31); svuint32_t odd = svlsl_x (pg, svreinterpret_u32 (n), 31);
n = svsub_n_f32_x (pg, n, d->shift); n = svsub_x (pg, n, d->shift);
/* r = |x| - n*pi (range reduction into -pi/2 .. pi/2). */ /* r = |x| - n*pi (range reduction into -pi/2 .. pi/2). */
svfloat32_t r; svfloat32_t r;
r = svmla_lane_f32 (ax, n, pi_vals, 0); r = svmla_lane (ax, n, pi_vals, 0);
r = svmla_lane_f32 (r, n, pi_vals, 1); r = svmla_lane (r, n, pi_vals, 1);
r = svmla_lane_f32 (r, n, pi_vals, 2); r = svmla_lane (r, n, pi_vals, 2);
/* sin(r) approx using a degree 9 polynomial from the Taylor series /* sin(r) approx using a degree 9 polynomial from the Taylor series
expansion. Note that only the odd terms of this are non-zero. */ expansion. Note that only the odd terms of this are non-zero. */
svfloat32_t r2 = svmul_f32_x (pg, r, r); svfloat32_t r2 = svmul_x (pg, r, r);
svfloat32_t y; svfloat32_t y;
y = svmla_f32_x (pg, C (2), r2, C (3)); y = svmla_x (pg, C (2), r2, C (3));
y = svmla_f32_x (pg, C (1), r2, y); y = svmla_x (pg, C (1), r2, y);
y = svmla_f32_x (pg, C (0), r2, y); y = svmla_x (pg, C (0), r2, y);
y = svmla_f32_x (pg, r, r, svmul_f32_x (pg, y, r2)); y = svmla_x (pg, r, r, svmul_x (pg, y, r2));
/* sign = y^sign^odd. */ /* sign = y^sign^odd. */
y = svreinterpret_f32_u32 (sveor_u32_x (pg, svreinterpret_u32_f32 (y), sign = sveor_x (pg, sign, odd);
sveor_u32_x (pg, sign, odd)));
if (__glibc_unlikely (svptest_any (pg, cmp))) if (__glibc_unlikely (svptest_any (pg, cmp)))
return special_case (x, y, cmp); return special_case (x,
return y; svreinterpret_f32 (sveor_x (
svnot_z (pg, cmp), svreinterpret_u32 (y), sign)),
cmp);
return svreinterpret_f32 (sveor_x (pg, svreinterpret_u32 (y), sign));
} }