Commit: 4e94a6f576de76d670b5ed3dd69500fd8c539e67
Author: Geraldine Chua
Date: Mon Jun 11 22:57:59 2018 +0800
Branches: soc-2018-cycles-volumes
https://developer.blender.org/rB4e94a6f576de76d670b5ed3dd69500fd8c539e67
Minor sparse tile fixes.
Fixed some really obvious errors with CUDA sampling, and moved the
coordinates to index calculation back into kernel, saving a small
perecentage of time.
===================================================================
M intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
M intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
M intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
M intern/cycles/render/image.cpp
M intern/cycles/render/mesh_volume.cpp
M intern/cycles/util/util_sparse_grid.h
===================================================================
diff --git a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
index c43b94db7e0..a0d6b54245b 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
+++ b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
@@ -30,6 +30,11 @@ template<typename T> struct TextureInterpolator {
u[3] = (1.0f / 6.0f) * t * t * t; \
} (void)0
+ static ccl_always_inline int flatten(int x, int y, int z, int width,
int height)
+ {
+ return x + width * (y + z * height);
+ }
+
static ccl_always_inline float4 read(float4 r)
{
return r;
@@ -82,7 +87,7 @@ template<typename T> struct TextureInterpolator {
int tix = x / TILE_SIZE, itix = x % TILE_SIZE,
tiy = y / TILE_SIZE, itiy = y % TILE_SIZE,
tiz = z / TILE_SIZE, itiz = z % TILE_SIZE;
- int dense_index = compute_index_fast(tix, tiy, tiz, tiw, tih) *
2;
+ int dense_index = flatten(tix, tiy, tiz, tiw, tih) * 2;
int sparse_index = grid_info[dense_index];
int dims = grid_info[dense_index + 1];
if(sparse_index < 0) {
@@ -90,7 +95,7 @@ template<typename T> struct TextureInterpolator {
}
int itiw = dims & (1 << ST_SHIFT_TRUNCATE_WIDTH) ? ltw :
TILE_SIZE;
int itih = dims & (1 << ST_SHIFT_TRUNCATE_HEIGHT) ? lth :
TILE_SIZE;
- int in_tile_index = compute_index_fast(itix, itiy, itiz, itiw,
itih);
+ int in_tile_index = flatten(itix, itiy, itiz, itiw, itih);
return read(data[sparse_index + in_tile_index]);
}
@@ -318,7 +323,7 @@ template<typename T> struct TextureInterpolator {
info.tiled_width, info.tiled_height,
info.last_tile_width,
info.last_tile_height);
}
- return read(data[compute_index_fast(ix, iy, iz, width,
height)]);
+ return read(data[flatten(ix, iy, iz, width, height)]);
}
static ccl_always_inline float4 interp_3d_linear(const TextureInfo&
info,
@@ -384,14 +389,14 @@ template<typename T> struct TextureInterpolator {
r += tz*ty*tx * read(data,
gi, nix, niy, niz, tiw, tih, ltw, lth);
}
else {
- r = (1.0f - tz)*(1.0f - ty)*(1.0f - tx) *
read(data[compute_index_fast(ix, iy, iz, width, height)]);
- r += (1.0f - tz)*(1.0f - ty)*tx *
read(data[compute_index_fast(nix, iy, iz, width, height)]);
- r += (1.0f - tz)*ty*(1.0f - tx) *
read(data[compute_index_fast(ix, niy, iz, width, height)]);
- r += (1.0f - tz)*ty*tx
* read(data[compute_index_fast(nix, niy, iz, width, height)]);
- r += tz*(1.0f - ty)*(1.0f - tx) *
read(data[compute_index_fast(ix, iy, niz, width, height)]);
- r += tz*(1.0f - ty)*tx
* read(data[compute_index_fast(nix, iy, niz, width, height)]);
- r += tz*ty*(1.0f - tx)
* read(data[compute_index_fast(ix, niy, niz, width, height)]);
- r += tz*ty*tx
* read(data[compute_index_fast(nix, niy, niz, width, height)]);
+ r = (1.0f - tz)*(1.0f - ty)*(1.0f - tx) *
read(data[flatten(ix, iy, iz, width, height)]);
+ r += (1.0f - tz)*(1.0f - ty)*tx *
read(data[flatten(nix, iy, iz, width, height)]);
+ r += (1.0f - tz)*ty*(1.0f - tx) *
read(data[flatten(ix, niy, iz, width, height)]);
+ r += (1.0f - tz)*ty*tx
* read(data[flatten(nix, niy, iz, width, height)]);
+ r += tz*(1.0f - ty)*(1.0f - tx) *
read(data[flatten(ix, iy, niz, width, height)]);
+ r += tz*(1.0f - ty)*tx
* read(data[flatten(nix, iy, niz, width, height)]);
+ r += tz*ty*(1.0f - tx)
* read(data[flatten(ix, niy, niz, width, height)]);
+ r += tz*ty*tx
* read(data[flatten(nix, niy, niz, width, height)]);
}
return r;
diff --git a/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
b/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
index dcd1b33722a..f73f5cdbb3a 100644
--- a/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
+++ b/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
@@ -58,7 +58,8 @@ ccl_device float cubic_h1(float a)
return 1.0f + cubic_w3(a) / (cubic_w2(a) + cubic_w3(a)) + 0.5f;
}
-ccl_device bool compute_sparse_coordinates(const TextureInfo *info, float fx,
float fy, float fz)
+/* Converts coordinates from normal volume textures dense to sparse ones. */
+ccl_device bool compute_sparse_coordinates(const TextureInfo *info, float &fx,
float &fy, float &fz)
{
float *ix, *iy, *iz;
modff(fx, *ix);
@@ -76,9 +77,9 @@ ccl_device bool compute_sparse_coordinates(const TextureInfo
*info, float fx, fl
int tile_y = info->grid_info[dense_index + 1];
int tile_z = info->grid_info[dense_index + 2];
int dims = info->grid_info[dense_index + 3];
- fx += tile_x + itix + (dims & (1 << ST_SHIFT_X_LHS_PAD));
- fy += tile_y + itiy + (dims & (1 << ST_SHIFT_Y_LHS_PAD));
- fz += tile_z + itiz + (dims & (1 << ST_SHIFT_Z_LHS_PAD));
+ fx += tile_x + itix + (dims & (1 << ST_SHIFT_X_PAD));
+ fy += tile_y + itiy + (dims & (1 << ST_SHIFT_Y_PAD));
+ fz += tile_z + itiz + (dims & (1 << ST_SHIFT_Z_PAD));
return true;
}
diff --git a/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
b/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
index 2e0db6609f7..be1996af3a0 100644
--- a/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
+++ b/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
@@ -67,6 +67,7 @@ ccl_device_inline float4 svm_image_texture_read(KernelGlobals
*kg, const ccl_glo
}
}
+/* Calculates the index for sparse volume textures. */
ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg,
const ccl_global TextureInfo
*info,
int id, int x, int y, int z)
diff --git a/intern/cycles/render/image.cpp b/intern/cycles/render/image.cpp
index 3789821e7d1..e0337bfab42 100644
--- a/intern/cycles/render/image.cpp
+++ b/intern/cycles/render/image.cpp
@@ -739,14 +739,26 @@ void ImageManager::file_make_image_sparse(Device *device,
int real_depth = tex_img->data_depth;
vector<DeviceType> sparse_grid;
vector<int> grid_info;
+ int voxel_count;
- int voxel_count = create_sparse_grid<DeviceType>(tex_img->data(),
+ if(device->info.type != DEVICE_CUDA) {
+ voxel_count = create_sparse_grid<DeviceType>(tex_img->data(),
real_width,
real_height,
real_depth,
img->isovalue,
&sparse_grid,
&grid_info);
+ }
+ else {
+ voxel_count =
create_sparse_grid_cuda<DeviceType>(tex_img->data(),
+ real_width,
+ real_height,
+ real_depth,
+ img->isovalue,
+ &sparse_grid,
+ &grid_info);
+ }
if(voxel_count < 1) {
VLOG(1) << "Could not make sparse grid for "
diff --git a/intern/cycles/render/mesh_volume.cpp
b/intern/cycles/render/mesh_volume.cpp
index 8e91105afdf..5ce8198b9d6 100644
--- a/intern/cycles/render/mesh_volume.cpp
+++ b/intern/cycles/render/mesh_volume.cpp
@@ -467,6 +467,7 @@ void MeshManager::create_volume_mesh(Scene *scene,
const int3 last_tile_res = make_int3(resolution.x % TILE_SIZE,
resolution.y % TILE_SIZE,
resolution.z % TILE_SIZE);
+ const bool using_cuda = (scene->device->info.type == DEVICE_CUDA);
if(attr) {
const Transform *tfm = attr->data_transform();
@@ -493,12 +494,24 @@ void MeshManager::create_volume_mesh(Scene *scene,
int voxel_index;
if(grid_info) {
- voxel_index =
compute_index(grid_info, x, y, z,
-
tiled_res.x,
-
tiled_res.y,
-
tiled_res.z,
-
last_tile_res.x,
-
last_tile_res.y);
+ if(!using_cuda) {
+ voxel_index =
compute_index(grid_info, x, y, z,
+
tiled_res.x,
+
tiled_res.y,
+
tiled_res.z,
+
last_tile_res.x,
+
last_tile_res.y);
+ }
+ else {
+ voxel_index =
compute_index_cuda(grid_info,
+
x, y, z,
+
resolution.x,
+
resolution.y,
+
resolution.z,
+
tiled_res.x,
+
tiled_res.y,
+
tiled_res.z);
+ }
if(voxel_index < 0) {
continue;
}
diff --git a/intern/cycles/util/util_sparse_grid.h
b/intern/cycles/util/util_sparse_grid.h
index dff9c3949ed..d63cee25fda 100644
--- a/intern/cycles/util/util_sparse_grid.h
+++ b/intern/cycles/util/util_sparse_grid.h
@@ -69,8 +69,8 @@ const inline int compute_index(const size_t x, const size_t
y, const size_t z,
return x + width * (y + z * height);
}
-const inline int compute_index_fast(const size_t x, const size_t y, const
size_t z,
- const size_t width, const size_t height)
+const inline int compute_index(const size_t x, const size_t y, const size_t z,
+ const size_t width, const size_t height)
{
return x + width * (y + z * height);
}
@@ -114,11 +114,10 @@ const inline int compute_index(const int *grid_info,
int itih = dims & (1 << ST_SHIFT_TRUNCATE_HEIGHT) ? lth : TILE_SIZE;
/* Look up voxel in the tile.
* Need to check whether or not a tile is padded on any of its 6 faces.
*/
- int in_tile_index = compute_index_fast(itix, itiy, itiz, itiw, itih);
+ int in_tile_index = compute_index(itix, itiy, itiz, itiw, itih);
return sparse_index + in_tile_index;
}
-
/* Do not call this function in the kernel. */
const inline int compute_index_cuda(const int *grid_info,
int x, int y, int z,
@@ -176,7 +175,7
@@ Diff output truncated at 10240 characters. @@
_______________________________________________
Bf-blender-cvs mailing list
[email protected]
https://lists.blender.org/mailman/listinfo/bf-blender-cvs
