sandey/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity

Fix T43027: OpenCL kernel compilation broken after QBVH

Browse Source 
OpenCL apparently does not support templates, so the idea of generic
function for swapping is a bit of a failure. Now it is either inlined
into the code (in triangle intersection) or has specific implementation
for QBVH.

This is probably even better, because we can't create QBVH-specific
function in util_math anyway.
This commit is contained in:
Sergey Sharybin 2015-01-02 14:58:01 +05:00
parent da66a2c871
commit 4f2583ee13
3 changed files with 22 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
intern/cycles/kernel/geom/geom_qbvh.h
View File

@ -19,14 +19,24 @@ struct QBVHStackItem {
float dist;
};
/* TOOD(sergey): Investigate if using instrinsics helps here. */
/* TOOD(sergey): Investigate if using instrinsics helps for both
* stack item swap and float comparison.
*/
ccl_device_inline void qbvh_item_swap(QBVHStackItem *__restrict a,
QBVHStackItem *__restrict b)
{
QBVHStackItem tmp = *a;
*a = *b;
*b = tmp;
}
ccl_device_inline void qbvh_stack_sort(QBVHStackItem *__restrict s1,
QBVHStackItem *__restrict s2,
QBVHStackItem *__restrict s3)
{
if(s2->dist < s1->dist) { util_swap(s2, s1); }
if(s3->dist < s2->dist) { util_swap(s3, s2); }
if(s2->dist < s1->dist) { util_swap(s2, s1); }
if(s2->dist < s1->dist) { qbvh_item_swap(s2, s1); }
if(s3->dist < s2->dist) { qbvh_item_swap(s3, s2); }
if(s2->dist < s1->dist) { qbvh_item_swap(s2, s1); }
}
ccl_device_inline void qbvh_stack_sort(QBVHStackItem *__restrict s1,
@ -34,11 +44,11 @@ ccl_device_inline void qbvh_stack_sort(QBVHStackItem *__restrict s1,
QBVHStackItem *__restrict s3,
QBVHStackItem *__restrict s4)
{
if(s2->dist < s1->dist) { util_swap(s2, s1); }
if(s4->dist < s3->dist) { util_swap(s4, s3); }
if(s3->dist < s1->dist) { util_swap(s3, s1); }
if(s4->dist < s2->dist) { util_swap(s4, s2); }
if(s3->dist < s2->dist) { util_swap(s3, s2); }
if(s2->dist < s1->dist) { qbvh_item_swap(s2, s1); }
if(s4->dist < s3->dist) { qbvh_item_swap(s4, s3); }
if(s3->dist < s1->dist) { qbvh_item_swap(s3, s1); }
if(s4->dist < s2->dist) { qbvh_item_swap(s4, s2); }
if(s3->dist < s2->dist) { qbvh_item_swap(s3, s2); }
}
ccl_device_inline int qbvh_node_intersect(KernelGlobals *__restrict kg,

4
intern/cycles/kernel/geom/geom_triangle_intersect.h
View File

@ -67,7 +67,9 @@ void triangle_intersect_precalc(float3 dir,
/* Swap kx and ky dimensions to preserve winding direction of triangles. */
if(IDX(dir, kz) < 0.0f) {
util_swap(&kx, &ky);
float tmp = kx;
kx = ky;
kx = tmp;
}
/* Calculate the shear constants. */

19
intern/cycles/util/util_math.h
View File

@ -1486,25 +1486,6 @@ ccl_device_inline int util_max_axis(float3 vec)
}
}
/* NOTE: We don't use std::swap here because of number of reasons:
*
* - We don't want current context to be polluted with all the templated
* functions from stl which might cause some interference about which
* function is used.
*
* - Different devices in theory might want to use intrinsics to optimize
* this function for specific type.
*
* - We don't want ot use references because of OpenCL state at this moment.
*/
template <typename T>
ccl_device_inline void util_swap(T *__restrict a, T *__restrict b)
{
T c = *a;
*a = *b;
*b = c;
}
CCL_NAMESPACE_END
#endif /* __UTIL_MATH_H__ */