Mesa (master): vc4: Improve the accuracy of SIN and COS.

[Eric Anholt]

Module: Mesa
Branch: master
Commit: 5bc91b6e322354d0964c07375c9a3a28b0083a38
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=5bc91b6e322354d0964c07375c9a3a28b0083a38

Author: Eric Anholt <eric at anholt.net>
Date:   Mon Oct 13 14:11:28 2014 +0100

vc4: Improve the accuracy of SIN and COS.

This gets them to pass glsl-sin/cos.  There was an obvious problem that I
was using the FRC code on the scaled input value, which means that we had
a range in [0, 1], while our taylor is most accurate across [-0.5, 0.5].
We can just slide things over, but that means flipping the sign of the
coefficients.  After that, it was just a matter of stuffing more
coefficients in.

---

 src/gallium/drivers/vc4/vc4_program.c |   28 +++++++++++++++++-----------
 1 file changed, 17 insertions(+), 11 deletions(-)

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 455b5d7..1bbdba5 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -766,10 +766,11 @@ tgsi_to_qir_sin(struct vc4_compile *c,
                 enum qop op, struct qreg *src, int i)
 {
         float coeff[] = {
-                2.0 * M_PI,
-                -pow(2.0 * M_PI, 3) / (3 * 2 * 1),
-                pow(2.0 * M_PI, 5) / (5 * 4 * 3 * 2 * 1),
-                -pow(2.0 * M_PI, 7) / (7 * 6 * 5 * 4 * 3 * 2 * 1),
+                -2.0 * M_PI,
+                pow(2.0 * M_PI, 3) / (3 * 2 * 1),
+                -pow(2.0 * M_PI, 5) / (5 * 4 * 3 * 2 * 1),
+                pow(2.0 * M_PI, 7) / (7 * 6 * 5 * 4 * 3 * 2 * 1),
+                -pow(2.0 * M_PI, 9) / (9 * 8 * 7 * 6 * 5 * 4 * 3 * 2 * 1),
         };
 
         struct qreg scaled_x =
@@ -777,8 +778,9 @@ tgsi_to_qir_sin(struct vc4_compile *c,
                          src[0 * 4 + 0],
                          qir_uniform_f(c, 1.0f / (M_PI * 2.0f)));
 
-
-        struct qreg x = tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0);
+        struct qreg x = qir_FADD(c,
+                                 tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0),
+                                 qir_uniform_f(c, -0.5));
         struct qreg x2 = qir_FMUL(c, x, x);
         struct qreg sum = qir_FMUL(c, x, qir_uniform_f(c, coeff[0]));
         for (int i = 1; i < ARRAY_SIZE(coeff); i++) {
@@ -799,16 +801,20 @@ tgsi_to_qir_cos(struct vc4_compile *c,
                 enum qop op, struct qreg *src, int i)
 {
         float coeff[] = {
-                1.0f,
-                -pow(2.0 * M_PI, 2) / (2 * 1),
-                pow(2.0 * M_PI, 4) / (4 * 3 * 2 * 1),
-                -pow(2.0 * M_PI, 6) / (6 * 5 * 4 * 3 * 2 * 1),
+                -1.0f,
+                pow(2.0 * M_PI, 2) / (2 * 1),
+                -pow(2.0 * M_PI, 4) / (4 * 3 * 2 * 1),
+                pow(2.0 * M_PI, 6) / (6 * 5 * 4 * 3 * 2 * 1),
+                -pow(2.0 * M_PI, 8) / (8 * 7 * 6 * 5 * 4 * 3 * 2 * 1),
+                pow(2.0 * M_PI, 10) / (10 * 9 * 8 * 7 * 6 * 5 * 4 * 3 * 2 * 1),
         };
 
         struct qreg scaled_x =
                 qir_FMUL(c, src[0 * 4 + 0],
                          qir_uniform_f(c, 1.0f / (M_PI * 2.0f)));
-        struct qreg x_frac = tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0);
+        struct qreg x_frac = qir_FADD(c,
+                                      tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0),
+                                      qir_uniform_f(c, -0.5));
 
         struct qreg sum = qir_uniform_f(c, coeff[0]);
         struct qreg x2 = qir_FMUL(c, x_frac, x_frac);