kannonier8/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity
064caae7b2
blender/intern/cycles/kernel/bvh/qbvh_traversal.h
Sergey Sharybin 064caae7b2 Cycles: BVH-related SSE optimization 
Several ideas here:

- Optimize calculation of near_{x,y,z} in a way that does not require
  3 if() statements per update, which avoids negative effect of wrong
  branch prediction.

- Optimization of direction clamping for BVH.

- Optimization of point/direction transform.

Brings ~1.5% speedup again depending on a scene (unfortunately, this
speedup can't be sum across all previous commits because speedup of
each of the changes varies from scene to scene, but it still seems to
be nice solid speedup of few percent on Linux and bigger speedup was
reported on Windows).

Once again ,thanks Maxym for inspiration!

Still TODO: We have multiple places where we need to calculate near
x,y,z indices in BVH, for now it's only done for main BVH traversal.
Will try to move this calculation to an utility function and see if
that can be easily re-used across all the BVH flavors.
2016-10-25 14:47:34 +02:00

549 lines
17 KiB
C
Raw Blame History

/*
* Copyright 2011-2013 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/* This is a template BVH traversal function, where various features can be
* enabled/disabled. This way we can compile optimized versions for each case
* without new features slowing things down.
*
* BVH_INSTANCING: object instancing
* BVH_HAIR: hair curve rendering
* BVH_HAIR_MINIMUM_WIDTH: hair curve rendering with minimum width
* BVH_MOTION: motion blur rendering
*
*/
#if BVH_FEATURE(BVH_HAIR)
# define NODE_INTERSECT qbvh_node_intersect
# define NODE_INTERSECT_ROBUST qbvh_node_intersect_robust
#else
# define NODE_INTERSECT qbvh_aligned_node_intersect
# define NODE_INTERSECT_ROBUST qbvh_aligned_node_intersect_robust
#endif
ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
const Ray *ray,
Intersection *isect,
const uint visibility
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
,uint *lcg_state,
float difl,
float extmax
#endif
)
{
/* TODO(sergey):
* - Test if pushing distance on the stack helps (for non shadow rays).
* - Separate version for shadow rays.
* - Likely and unlikely for if() statements.
* - Test restrict attribute for pointers.
*/
/* Traversal stack in CUDA thread-local memory. */
QBVHStackItem traversal_stack[BVH_QSTACK_SIZE];
traversal_stack[0].addr = ENTRYPOINT_SENTINEL;
traversal_stack[0].dist = -FLT_MAX;
/* Traversal variables in registers. */
int stack_ptr = 0;
int node_addr = kernel_data.bvh.root;
float node_dist = -FLT_MAX;
/* Ray parameters in registers. */
float3 P = ray->P;
float3 dir = bvh_clamp_direction(ray->D);
float3 idir = bvh_inverse_direction(dir);
int object = OBJECT_NONE;
#if BVH_FEATURE(BVH_MOTION)
Transform ob_itfm;
#endif
#ifndef __KERNEL_SSE41__
if(!isfinite(P.x)) {
return false;
}
#endif
isect->t = ray->t;
isect->u = 0.0f;
isect->v = 0.0f;
isect->prim = PRIM_NONE;
isect->object = OBJECT_NONE;
BVH_DEBUG_INIT();
ssef tnear(0.0f), tfar(ray->t);
#if BVH_FEATURE(BVH_HAIR)
sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z));
#endif
sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
float3 P_idir = P*idir;
sse3f P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#endif
#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
sse3f org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
/* Offsets to select the side that becomes the lower or upper bound. */
#ifdef __KERNEL_SSE__
int near_x = 0, near_y = 2, near_z = 4;
int far_x = 1, far_y = 3, far_z = 5;
const size_t mask = movemask(ssef(idir.m128));
const int mask_x = mask & 1;
const int mask_y = (mask & 2) >> 1;
const int mask_z = (mask & 4) >> 2;
near_x += mask_x; far_x -= mask_x;
near_y += mask_y; far_y -= mask_y;
near_z += mask_z; far_z -= mask_z;
#else
int near_x, near_y, near_z;
int far_x, far_y, far_z;
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
#endif
IsectPrecalc isect_precalc;
triangle_intersect_precalc(dir, &isect_precalc);
/* Traversal loop. */
do {
do {
/* Traverse internal nodes. */
while(node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) {
float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr+0);
if(UNLIKELY(node_dist > isect->t)
#ifdef __VISIBILITY_FLAG__
|| (__float_as_uint(inodes.x) & visibility) == 0)
#endif
{
/* Pop. */
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
continue;
}
int child_mask;
ssef dist;
BVH_DEBUG_NEXT_STEP();
#if BVH_FEATURE(BVH_HAIR_MINIMUM_WIDTH)
if(difl != 0.0f) {
/* NOTE: We extend all the child BB instead of fetching
* and checking visibility flags for each of the,
*
* Need to test if doing opposite would be any faster.
*/
child_mask = NODE_INTERSECT_ROBUST(kg,
tnear,
tfar,
# ifdef __KERNEL_AVX2__
P_idir4,
# endif
# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4,
# endif
# if BVH_FEATURE(BVH_HAIR)
dir4,
# endif
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
node_addr,
difl,
&dist);
}
else
#endif /* BVH_HAIR_MINIMUM_WIDTH */
{
child_mask = NODE_INTERSECT(kg,
tnear,
tfar,
#ifdef __KERNEL_AVX2__
P_idir4,
#endif
#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4,
#endif
#if BVH_FEATURE(BVH_HAIR)
dir4,
#endif
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
node_addr,
&dist);
}
if(child_mask != 0) {
float4 cnodes;
/* TODO(sergey): Investigate whether moving cnodes upwards
* gives a speedup (will be different cache pattern but will
* avoid extra check here),
*/
#if BVH_FEATURE(BVH_HAIR)
if(__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) {
cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+13);
}
else
#endif
{
cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+7);
}
/* One child is hit, continue with that child. */
int r = __bscf(child_mask);
float d0 = ((float*)&dist)[r];
if(child_mask == 0) {
node_addr = __float_as_int(cnodes[r]);
node_dist = d0;
continue;
}
/* Two children are hit, push far child, and continue with
* closer child.
*/
int c0 = __float_as_int(cnodes[r]);
r = __bscf(child_mask);
int c1 = __float_as_int(cnodes[r]);
float d1 = ((float*)&dist)[r];
if(child_mask == 0) {
if(d1 < d0) {
node_addr = c1;
node_dist = d1;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c0;
traversal_stack[stack_ptr].dist = d0;
continue;
}
else {
node_addr = c0;
node_dist = d0;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c1;
traversal_stack[stack_ptr].dist = d1;
continue;
}
}
/* Here starts the slow path for 3 or 4 hit children. We push
* all nodes onto the stack to sort them there.
*/
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c1;
traversal_stack[stack_ptr].dist = d1;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c0;
traversal_stack[stack_ptr].dist = d0;
/* Three children are hit, push all onto stack and sort 3
* stack items, continue with closest child.
*/
r = __bscf(child_mask);
int c2 = __float_as_int(cnodes[r]);
float d2 = ((float*)&dist)[r];
if(child_mask == 0) {
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c2;
traversal_stack[stack_ptr].dist = d2;
qbvh_stack_sort(&traversal_stack[stack_ptr],
&traversal_stack[stack_ptr - 1],
&traversal_stack[stack_ptr - 2]);
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
continue;
}
/* Four children are hit, push all onto stack and sort 4
* stack items, continue with closest child.
*/
r = __bscf(child_mask);
int c3 = __float_as_int(cnodes[r]);
float d3 = ((float*)&dist)[r];
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c3;
traversal_stack[stack_ptr].dist = d3;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c2;
traversal_stack[stack_ptr].dist = d2;
qbvh_stack_sort(&traversal_stack[stack_ptr],
&traversal_stack[stack_ptr - 1],
&traversal_stack[stack_ptr - 2],
&traversal_stack[stack_ptr - 3]);
}
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
}
/* If node is leaf, fetch triangle list. */
if(node_addr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr-1));
#ifdef __VISIBILITY_FLAG__
if(UNLIKELY((node_dist > isect->t) ||
((__float_as_uint(leaf.z) & visibility) == 0)))
#else
if(UNLIKELY((node_dist > isect->t)))
#endif
{
/* Pop. */
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
continue;
}
int prim_addr = __float_as_int(leaf.x);
#if BVH_FEATURE(BVH_INSTANCING)
if(prim_addr >= 0) {
#endif
int prim_addr2 = __float_as_int(leaf.y);
const uint type = __float_as_int(leaf.w);
/* Pop. */
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
/* Primitive intersection. */
switch(type & PRIMITIVE_ALL) {
case PRIMITIVE_TRIANGLE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
if(triangle_intersect(kg,
&isect_precalc,
isect,
P,
visibility,
object,
prim_addr)) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE) {
return true;
}
}
}
break;
}
#if BVH_FEATURE(BVH_MOTION)
case PRIMITIVE_MOTION_TRIANGLE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
if(motion_triangle_intersect(kg,
isect,
P,
dir,
ray->time,
visibility,
object,
prim_addr)) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE) {
return true;
}
}
}
break;
}
#endif /* BVH_FEATURE(BVH_MOTION) */
#if BVH_FEATURE(BVH_HAIR)
case PRIMITIVE_CURVE:
case PRIMITIVE_MOTION_CURVE: {
for(; prim_addr < prim_addr2; prim_addr++) {
BVH_DEBUG_NEXT_STEP();
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
bool hit;
if(kernel_data.curve.curveflags & CURVE_KN_INTERPOLATE) {
hit = bvh_cardinal_curve_intersect(kg,
isect,
P,
dir,
visibility,
object,
prim_addr,
ray->time,
type,
lcg_state,
difl,
extmax);
}
else {
hit = bvh_curve_intersect(kg,
isect,
P,
dir,
visibility,
object,
prim_addr,
ray->time,
type,
lcg_state,
difl,
extmax);
}
if(hit) {
tfar = ssef(isect->t);
/* Shadow ray early termination. */
if(visibility == PATH_RAY_SHADOW_OPAQUE) {
return true;
}
}
}
break;
}
#endif /* BVH_FEATURE(BVH_HAIR) */
}
}
#if BVH_FEATURE(BVH_INSTANCING)
else {
/* Instance push. */
object = kernel_tex_fetch(__prim_object, -prim_addr-1);
# if BVH_FEATURE(BVH_MOTION)
qbvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist, &ob_itfm);
# else
qbvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist);
# endif
#ifdef __KERNEL_SSE__
near_x = 0; near_y = 2; near_z = 4;
far_x = 1; far_y = 3; far_z = 5;
const size_t mask = movemask(ssef(idir.m128));
const int mask_x = mask & 1;
const int mask_y = (mask & 2) >> 1;
const int mask_z = (mask & 4) >> 2;
near_x += mask_x; far_x -= mask_x;
near_y += mask_y; far_y -= mask_y;
near_z += mask_z; far_z -= mask_z;
#else
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
#endif
tfar = ssef(isect->t);
# if BVH_FEATURE(BVH_HAIR)
dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z));
# endif
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
# ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
# endif
# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
# endif
triangle_intersect_precalc(dir, &isect_precalc);
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL;
traversal_stack[stack_ptr].dist = -FLT_MAX;
node_addr = kernel_tex_fetch(__object_node, object);
BVH_DEBUG_NEXT_INSTANCE();
}
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(node_addr != ENTRYPOINT_SENTINEL);
#if BVH_FEATURE(BVH_INSTANCING)
if(stack_ptr >= 0) {
kernel_assert(object != OBJECT_NONE);
/* Instance pop. */
# if BVH_FEATURE(BVH_MOTION)
bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, &isect->t, &ob_itfm);
# else
bvh_instance_pop(kg, object, ray, &P, &dir, &idir, &isect->t);
# endif
#ifdef __KERNEL_SSE__
near_x = 0; near_y = 2; near_z = 4;
far_x = 1; far_y = 3; far_z = 5;
const size_t mask = movemask(ssef(idir.m128));
const int mask_x = mask & 1;
const int mask_y = (mask & 2) >> 1;
const int mask_z = (mask & 4) >> 2;
near_x += mask_x; far_x -= mask_x;
near_y += mask_y; far_y -= mask_y;
near_z += mask_z; far_z -= mask_z;
#else
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
#endif
tfar = ssef(isect->t);
# if BVH_FEATURE(BVH_HAIR)
dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z));
# endif
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
# ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
# endif
# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
# endif
triangle_intersect_precalc(dir, &isect_precalc);
object = OBJECT_NONE;
node_addr = traversal_stack[stack_ptr].addr;
node_dist = traversal_stack[stack_ptr].dist;
--stack_ptr;
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(node_addr != ENTRYPOINT_SENTINEL);
return (isect->prim != PRIM_NONE);
}
#undef NODE_INTERSECT
#undef NODE_INTERSECT_ROBUST
View Git Blame Copy Permalink