diff options
Diffstat (limited to 'intern/cycles/util/util_transform.h')
-rw-r--r-- | intern/cycles/util/util_transform.h | 187 |
1 files changed, 50 insertions, 137 deletions
diff --git a/intern/cycles/util/util_transform.h b/intern/cycles/util/util_transform.h index ac0804a7227..987f4dac777 100644 --- a/intern/cycles/util/util_transform.h +++ b/intern/cycles/util/util_transform.h @@ -26,10 +26,10 @@ CCL_NAMESPACE_BEGIN -/* Data Types */ +/* Affine transformation, stored as 4x3 matrix. */ typedef struct Transform { - float4 x, y, z, w; /* rows */ + float4 x, y, z; #ifndef __KERNEL_GPU__ float4 operator[](int i) const { return *(&x + i); } @@ -37,32 +37,16 @@ typedef struct Transform { #endif } Transform; -/* transform decomposed in rotation/translation/scale. we use the same data +/* Transform decomposed in rotation/translation/scale. we use the same data * structure as Transform, and tightly pack decomposition into it. first the * rotation (4), then translation (3), then 3x3 scale matrix (9). */ -typedef struct ccl_may_alias MotionTransform { - Transform pre; - Transform mid; - Transform post; -} MotionTransform; - -typedef struct PerspectiveMotionTransform { - Transform pre; - Transform post; -} PerspectiveMotionTransform; +typedef struct DecomposedTransform { + float4 x, y, z, w; +} DecomposedTransform; /* Functions */ -ccl_device_inline float3 transform_perspective(const Transform *t, const float3 a) -{ - float4 b = make_float4(a.x, a.y, a.z, 1.0f); - float3 c = make_float3(dot(t->x, b), dot(t->y, b), dot(t->z, b)); - float w = dot(t->w, b); - - return (w != 0.0f)? c/w: make_float3(0.0f, 0.0f, 0.0f); -} - ccl_device_inline float3 transform_point(const Transform *t, const float3 a) { /* TODO(sergey): Disabled for now, causes crashes in certain cases. */ @@ -73,7 +57,7 @@ ccl_device_inline float3 transform_point(const Transform *t, const float3 a) x = _mm_loadu_ps(&t->x.x); y = _mm_loadu_ps(&t->y.x); z = _mm_loadu_ps(&t->z.x); - w = _mm_loadu_ps(&t->w.x); + w = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f); _MM_TRANSPOSE4_PS(x, y, z, w); @@ -129,29 +113,15 @@ ccl_device_inline float3 transform_direction_transposed(const Transform *t, cons return make_float3(dot(x, a), dot(y, a), dot(z, a)); } -ccl_device_inline Transform transform_transpose(const Transform a) -{ - Transform t; - - t.x.x = a.x.x; t.x.y = a.y.x; t.x.z = a.z.x; t.x.w = a.w.x; - t.y.x = a.x.y; t.y.y = a.y.y; t.y.z = a.z.y; t.y.w = a.w.y; - t.z.x = a.x.z; t.z.y = a.y.z; t.z.z = a.z.z; t.z.w = a.w.z; - t.w.x = a.x.w; t.w.y = a.y.w; t.w.z = a.z.w; t.w.w = a.w.w; - - return t; -} - ccl_device_inline Transform make_transform(float a, float b, float c, float d, float e, float f, float g, float h, - float i, float j, float k, float l, - float m, float n, float o, float p) + float i, float j, float k, float l) { Transform t; t.x.x = a; t.x.y = b; t.x.z = c; t.x.w = d; t.y.x = e; t.y.y = f; t.y.z = g; t.y.w = h; t.z.x = i; t.z.y = j; t.z.z = k; t.z.w = l; - t.w.x = m; t.w.y = n; t.w.z = o; t.w.w = p; return t; } @@ -165,21 +135,22 @@ ccl_device_inline Transform make_transform_frame(float3 N) const float3 dy = normalize(cross(N, dx)); return make_transform(dx.x, dx.y, dx.z, 0.0f, dy.x, dy.y, dy.z, 0.0f, - N.x , N.y, N.z, 0.0f, - 0.0f, 0.0f, 0.0f, 1.0f); + N.x , N.y, N.z, 0.0f); } #ifndef __KERNEL_GPU__ ccl_device_inline Transform operator*(const Transform a, const Transform b) { - Transform c = transform_transpose(b); - Transform t; + float4 c_x = make_float4(b.x.x, b.y.x, b.z.x, 0.0f); + float4 c_y = make_float4(b.x.y, b.y.y, b.z.y, 0.0f); + float4 c_z = make_float4(b.x.z, b.y.z, b.z.z, 0.0f); + float4 c_w = make_float4(b.x.w, b.y.w, b.z.w, 1.0f); - t.x = make_float4(dot(a.x, c.x), dot(a.x, c.y), dot(a.x, c.z), dot(a.x, c.w)); - t.y = make_float4(dot(a.y, c.x), dot(a.y, c.y), dot(a.y, c.z), dot(a.y, c.w)); - t.z = make_float4(dot(a.z, c.x), dot(a.z, c.y), dot(a.z, c.z), dot(a.z, c.w)); - t.w = make_float4(dot(a.w, c.x), dot(a.w, c.y), dot(a.w, c.z), dot(a.w, c.w)); + Transform t; + t.x = make_float4(dot(a.x, c_x), dot(a.x, c_y), dot(a.x, c_z), dot(a.x, c_w)); + t.y = make_float4(dot(a.y, c_x), dot(a.y, c_y), dot(a.y, c_z), dot(a.y, c_w)); + t.z = make_float4(dot(a.z, c_x), dot(a.z, c_y), dot(a.z, c_z), dot(a.z, c_w)); return t; } @@ -189,7 +160,6 @@ ccl_device_inline void print_transform(const char *label, const Transform& t) print_float4(label, t.x); print_float4(label, t.y); print_float4(label, t.z); - print_float4(label, t.w); printf("\n"); } @@ -198,8 +168,7 @@ ccl_device_inline Transform transform_translate(float3 t) return make_transform( 1, 0, 0, t.x, 0, 1, 0, t.y, - 0, 0, 1, t.z, - 0, 0, 0, 1); + 0, 0, 1, t.z); } ccl_device_inline Transform transform_translate(float x, float y, float z) @@ -212,8 +181,7 @@ ccl_device_inline Transform transform_scale(float3 s) return make_transform( s.x, 0, 0, 0, 0, s.y, 0, 0, - 0, 0, s.z, 0, - 0, 0, 0, 1); + 0, 0, s.z, 0); } ccl_device_inline Transform transform_scale(float x, float y, float z) @@ -221,21 +189,6 @@ ccl_device_inline Transform transform_scale(float x, float y, float z) return transform_scale(make_float3(x, y, z)); } -ccl_device_inline Transform transform_perspective(float fov, float n, float f) -{ - Transform persp = make_transform( - 1, 0, 0, 0, - 0, 1, 0, 0, - 0, 0, f / (f - n), -f*n / (f - n), - 0, 0, 1, 0); - - float inv_angle = 1.0f/tanf(0.5f*fov); - - Transform scale = transform_scale(inv_angle, inv_angle, 1); - - return scale * persp; -} - ccl_device_inline Transform transform_rotate(float angle, float3 axis) { float s = sinf(angle); @@ -258,9 +211,7 @@ ccl_device_inline Transform transform_rotate(float angle, float3 axis) axis.z*axis.x*t - s*axis.y, axis.z*axis.y*t + s*axis.x, axis.z*axis.z*t + c, - 0.0f, - - 0.0f, 0.0f, 0.0f, 1.0f); + 0.0f); } /* Euler is assumed to be in XYZ order. */ @@ -272,12 +223,6 @@ ccl_device_inline Transform transform_euler(float3 euler) transform_rotate(euler.x, make_float3(1.0f, 0.0f, 0.0f)); } -ccl_device_inline Transform transform_orthographic(float znear, float zfar) -{ - return transform_scale(1.0f, 1.0f, 1.0f / (zfar-znear)) * - transform_translate(0.0f, 0.0f, -znear); -} - ccl_device_inline Transform transform_identity() { return transform_scale(1.0f, 1.0f, 1.0f); @@ -306,20 +251,20 @@ ccl_device_inline void transform_set_column(Transform *t, int column, float3 val } Transform transform_inverse(const Transform& a); +Transform transform_transposed_inverse(const Transform& a); ccl_device_inline bool transform_uniform_scale(const Transform& tfm, float& scale) { /* the epsilon here is quite arbitrary, but this function is only used for - * surface area and bump, where we except it to not be so sensitive */ - Transform ttfm = transform_transpose(tfm); + * surface area and bump, where we expect it to not be so sensitive */ float eps = 1e-6f; float sx = len_squared(float4_to_float3(tfm.x)); float sy = len_squared(float4_to_float3(tfm.y)); float sz = len_squared(float4_to_float3(tfm.z)); - float stx = len_squared(float4_to_float3(ttfm.x)); - float sty = len_squared(float4_to_float3(ttfm.y)); - float stz = len_squared(float4_to_float3(ttfm.z)); + float stx = len_squared(transform_get_column(&tfm, 0)); + float sty = len_squared(transform_get_column(&tfm, 1)); + float stz = len_squared(transform_get_column(&tfm, 2)); if(fabsf(sx - sy) < eps && fabsf(sx - sz) < eps && fabsf(sx - stx) < eps && fabsf(sx - sty) < eps && @@ -357,7 +302,6 @@ ccl_device_inline Transform transform_empty() return make_transform( 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0); } @@ -414,12 +358,11 @@ ccl_device_inline Transform transform_quick_inverse(Transform M) R.x = make_float4(Rx.x, Rx.y, Rx.z, dot(Rx, T)); R.y = make_float4(Ry.x, Ry.y, Ry.z, dot(Ry, T)); R.z = make_float4(Rz.x, Rz.y, Rz.z, dot(Rz, T)); - R.w = make_float4(0.0f, 0.0f, 0.0f, 1.0f); return R; } -ccl_device_inline void transform_compose(Transform *tfm, const Transform *decomp) +ccl_device_inline void transform_compose(Transform *tfm, const DecomposedTransform *decomp) { /* rotation */ float q0, q1, q2, q3, qda, qdb, qdc, qaa, qab, qac, qbb, qbc, qcc; @@ -452,60 +395,30 @@ ccl_device_inline void transform_compose(Transform *tfm, const Transform *decomp tfm->x = make_float4(dot(rotation_x, scale_x), dot(rotation_x, scale_y), dot(rotation_x, scale_z), decomp->y.x); tfm->y = make_float4(dot(rotation_y, scale_x), dot(rotation_y, scale_y), dot(rotation_y, scale_z), decomp->y.y); tfm->z = make_float4(dot(rotation_z, scale_x), dot(rotation_z, scale_y), dot(rotation_z, scale_z), decomp->y.z); - tfm->w = make_float4(0.0f, 0.0f, 0.0f, 1.0f); } -/* Disabled for now, need arc-length parametrization for constant speed motion. - * #define CURVED_MOTION_INTERPOLATE */ - -ccl_device void transform_motion_interpolate(Transform *tfm, const MotionTransform *motion, float t) +/* Interpolate from array of decomposed transforms. */ +ccl_device void transform_motion_array_interpolate(Transform *tfm, + const ccl_global DecomposedTransform *motion, + uint numsteps, + float time) { - /* possible optimization: is it worth it adding a check to skip scaling? - * it's probably quite uncommon to have scaling objects. or can we skip - * just shearing perhaps? */ - Transform decomp; - -#ifdef CURVED_MOTION_INTERPOLATE - /* 3 point bezier curve interpolation for position */ - float3 Ppre = float4_to_float3(motion->pre.y); - float3 Pmid = float4_to_float3(motion->mid.y); - float3 Ppost = float4_to_float3(motion->post.y); - - float3 Pcontrol = 2.0f*Pmid - 0.5f*(Ppre + Ppost); - float3 P = Ppre*t*t + Pcontrol*2.0f*t*(1.0f - t) + Ppost*(1.0f - t)*(1.0f - t); - - decomp.y.x = P.x; - decomp.y.y = P.y; - decomp.y.z = P.z; -#endif - - /* linear interpolation for rotation and scale */ - if(t < 0.5f) { - t *= 2.0f; - - decomp.x = quat_interpolate(motion->pre.x, motion->mid.x, t); -#ifdef CURVED_MOTION_INTERPOLATE - decomp.y.w = (1.0f - t)*motion->pre.y.w + t*motion->mid.y.w; -#else - decomp.y = (1.0f - t)*motion->pre.y + t*motion->mid.y; -#endif - decomp.z = (1.0f - t)*motion->pre.z + t*motion->mid.z; - decomp.w = (1.0f - t)*motion->pre.w + t*motion->mid.w; - } - else { - t = (t - 0.5f)*2.0f; - - decomp.x = quat_interpolate(motion->mid.x, motion->post.x, t); -#ifdef CURVED_MOTION_INTERPOLATE - decomp.y.w = (1.0f - t)*motion->mid.y.w + t*motion->post.y.w; -#else - decomp.y = (1.0f - t)*motion->mid.y + t*motion->post.y; -#endif - decomp.z = (1.0f - t)*motion->mid.z + t*motion->post.z; - decomp.w = (1.0f - t)*motion->mid.w + t*motion->post.w; - } - - /* compose rotation, translation, scale into matrix */ + /* Figure out which steps we need to interpolate. */ + int maxstep = numsteps-1; + int step = min((int)(time*maxstep), maxstep-1); + float t = time*maxstep - step; + + const ccl_global DecomposedTransform *a = motion + step; + const ccl_global DecomposedTransform *b = motion + step + 1; + + /* Interpolate rotation, translation and scale. */ + DecomposedTransform decomp; + decomp.x = quat_interpolate(a->x, b->x, t); + decomp.y = (1.0f - t)*a->y + t*b->y; + decomp.z = (1.0f - t)*a->z + t*b->z; + decomp.w = (1.0f - t)*a->w + t*b->w; + + /* Compose rotation, translation, scale into matrix. */ transform_compose(tfm, &decomp); } @@ -513,13 +426,13 @@ ccl_device void transform_motion_interpolate(Transform *tfm, const MotionTransfo class BoundBox2D; -ccl_device_inline bool operator==(const MotionTransform& A, const MotionTransform& B) +ccl_device_inline bool operator==(const DecomposedTransform& A, const DecomposedTransform& B) { - return (A.pre == B.pre && A.post == B.post); + return memcmp(&A, &B, sizeof(DecomposedTransform)) == 0; } float4 transform_to_quat(const Transform& tfm); -void transform_motion_decompose(MotionTransform *decomp, const MotionTransform *motion, const Transform *mid); +void transform_motion_decompose(DecomposedTransform *decomp, const Transform *motion, size_t size); Transform transform_from_viewplane(BoundBox2D& viewplane); #endif |