[media] fixp-arith: replace sin/cos table by a better precision one
The cos table used at fixp-arith.h has only 8 bits of precision.
That causes problems if it is reused on other drivers.
As some media drivers require a higher precision sin/cos
implementation, replace the current implementation by one that
will provide 32 bits precision.
The values generated by the new implementation matches the
32 bit precision of glibc's sin for an angle measured in
integer degrees.
It also provides support for fractional angles via linear
interpolation. On experimental calculus, when used a table
with a 0.001 degree angle, the maximum error for sin is
0.000038, which is likely good enough for practical purposes.
There are some logic there that seems to be specific to the
usage inside ff-memless.c. Move those logic to there, as they're
not needed elsewhere.
Cc: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
Signed-off-by: Prashant Laddha <prladdha@cisco.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
diff --git a/drivers/input/ff-memless.c b/drivers/input/ff-memless.c
index 74c0d8c6..fcc6c33 100644
--- a/drivers/input/ff-memless.c
+++ b/drivers/input/ff-memless.c
@@ -237,6 +237,18 @@
(force + new_force)) << 1;
}
+#define FRAC_N 8
+static inline s16 fixp_new16(s16 a)
+{
+ return ((s32)a) >> (16 - FRAC_N);
+}
+
+static inline s16 fixp_mult(s16 a, s16 b)
+{
+ a = ((s32)a * 0x100) / 0x7fff;
+ return ((s32)(a * b)) >> FRAC_N;
+}
+
/*
* Combine two effects and apply gain.
*/
@@ -247,7 +259,7 @@
struct ff_effect *new = state->effect;
unsigned int strong, weak, i;
int x, y;
- fixp_t level;
+ s16 level;
switch (new->type) {
case FF_CONSTANT:
@@ -255,8 +267,8 @@
level = fixp_new16(apply_envelope(state,
new->u.constant.level,
&new->u.constant.envelope));
- x = fixp_mult(fixp_sin(i), level) * gain / 0xffff;
- y = fixp_mult(-fixp_cos(i), level) * gain / 0xffff;
+ x = fixp_mult(fixp_sin16(i), level) * gain / 0xffff;
+ y = fixp_mult(-fixp_cos16(i), level) * gain / 0xffff;
/*
* here we abuse ff_ramp to hold x and y of constant force
* If in future any driver wants something else than x and y