diff options
author | Ville Syrjälä <ville.syrjala@linux.intel.com> | 2018-06-06 16:28:31 +0300 |
---|---|---|
committer | Ville Syrjälä <ville.syrjala@linux.intel.com> | 2018-06-06 16:28:31 +0300 |
commit | c07af1f4713b4d014de8fd0d401f9e3ed639f6d6 (patch) | |
tree | d5310540eccbcaef5be515a4e80665ac5bf62818 /lib/igt_matrix.h | |
parent | c8f1ae58e1b7da17af4722a5ce5a9cd8b9a34059 (diff) |
lib/igt_matrix: Unroll and inline igt_matrix_transform()
Using the current igt_matrix for NV12 conversion ends up being
about 4x as slow as the current non-igt_matrix based code. Unrolling
and inlining igt_matrix_transform() improves that factor to ~1.5x.
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Diffstat (limited to 'lib/igt_matrix.h')
-rw-r--r-- | lib/igt_matrix.h | 43 |
1 files changed, 41 insertions, 2 deletions
diff --git a/lib/igt_matrix.h b/lib/igt_matrix.h index 33acb815..7a2b9ad8 100644 --- a/lib/igt_matrix.h +++ b/lib/igt_matrix.h @@ -44,13 +44,52 @@ struct igt_mat4 { float d[16]; }; +#define m(row, col) ((col) * 4 + (row)) + void igt_matrix_print(const struct igt_mat4 *m); struct igt_mat4 igt_matrix_identity(void); struct igt_mat4 igt_matrix_scale(float x, float y, float z); struct igt_mat4 igt_matrix_translate(float x, float y, float z); -struct igt_vec4 igt_matrix_transform(const struct igt_mat4 *m, - const struct igt_vec4 *v); struct igt_mat4 igt_matrix_multiply(const struct igt_mat4 *a, const struct igt_mat4 *b); +/** + * igt_matrix_transform: + * + * Transform the vector @v by the matrix @m. @m is on the left, + * @v on the right. + * + * Returns: + * The transformed vector. + */ +static inline struct igt_vec4 +igt_matrix_transform(const struct igt_mat4 *m, + const struct igt_vec4 *v) +{ + struct igt_vec4 ret = { + .d = { m->d[m(0, 0)] * v->d[0] + + m->d[m(0, 1)] * v->d[1] + + m->d[m(0, 2)] * v->d[2] + + m->d[m(0, 3)] * v->d[3], + + m->d[m(1, 0)] * v->d[0] + + m->d[m(1, 1)] * v->d[1] + + m->d[m(1, 2)] * v->d[2] + + m->d[m(1, 3)] * v->d[3], + + m->d[m(2, 0)] * v->d[0] + + m->d[m(2, 1)] * v->d[1] + + m->d[m(2, 2)] * v->d[2] + + m->d[m(2, 3)] * v->d[3], + + m->d[m(3, 0)] * v->d[0] + + m->d[m(3, 1)] * v->d[1] + + m->d[m(3, 2)] * v->d[2] + + m->d[m(3, 3)] * v->d[3], + }, + }; + + return ret; +} + #endif /* __IGT_MATRIX_H__ */ |