[Bf-blender-cvs] [a9da879efa] soc-2016-cycles_denoising: Cycles Denoising: Remove hidden NLM-only option
Lukas Stockner
noreply at git.blender.org
Thu Jan 12 05:14:08 CET 2017
Commit: a9da879efa690be65ed3ea45a598be45525e2ff6
Author: Lukas Stockner
Date: Wed Jan 11 06:26:19 2017 +0100
Branches: soc-2016-cycles_denoising
https://developer.blender.org/rBa9da879efa690be65ed3ea45a598be45525e2ff6
Cycles Denoising: Remove hidden NLM-only option
===================================================================
M intern/cycles/device/device_cpu.cpp
M intern/cycles/kernel/filter/filter_nlm.h
M intern/cycles/kernel/kernels/cpu/kernel_cpu.h
M intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h
===================================================================
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 4e713e8469..c8203fe802 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -142,7 +142,6 @@ public:
KernelFunctions<void(*)(KernelGlobals*, int, float*, int, int, int, int, float*, void*, float*, int*, int*)> filter_final_pass_wlr_kernel;
KernelFunctions<void(*)(KernelGlobals*, int, float*, int, int, int, int, float*, void*, float*, int*, int*)> filter_final_pass_nlm_kernel;
KernelFunctions<void(*)(KernelGlobals*, int, int, int, float*, int, int)> filter_divide_combined_kernel;
- KernelFunctions<void(*)(int, int, float**, float**, float**, float**, int*, int, int, float, float)> filter_non_local_means_3_kernel;
#define KERNEL_FUNCTIONS(name) \
KERNEL_NAME_EVAL(cpu, name), \
@@ -166,8 +165,7 @@ public:
filter_estimate_wlr_params_kernel(KERNEL_FUNCTIONS(filter_estimate_wlr_params)),
filter_final_pass_wlr_kernel(KERNEL_FUNCTIONS(filter_final_pass_wlr)),
filter_final_pass_nlm_kernel(KERNEL_FUNCTIONS(filter_final_pass_nlm)),
- filter_divide_combined_kernel(KERNEL_FUNCTIONS(filter_divide_combined)),
- filter_non_local_means_3_kernel(KERNEL_FUNCTIONS(filter_non_local_means_3))
+ filter_divide_combined_kernel(KERNEL_FUNCTIONS(filter_divide_combined))
{
#ifdef WITH_OSL
kernel_globals.osl = &osl_globals;
@@ -455,7 +453,6 @@ public:
void denoise_run(KernelGlobals *kg, int sample, float *filter_buffer, int4 filter_area, int4 rect, int offset, int stride, float *buffers)
{
- bool only_nlm_filter = getenv("ONLY_NLM_FILTER");
bool use_gradients = kg->__data.integrator.use_gradients;
bool nlm_weights = kg->__data.integrator.use_nlm_weights;
@@ -466,28 +463,7 @@ public:
int w = align_up(rect.z - rect.x, 4), h = (rect.w - rect.y);
int pass_stride = w*h;
- if(only_nlm_filter) {
- float *img[3] = {filter_buffer + 16*pass_stride, filter_buffer + 18*pass_stride, filter_buffer + 20*pass_stride};
- float *var[3] = {filter_buffer + 17*pass_stride, filter_buffer + 19*pass_stride, filter_buffer + 21*pass_stride};
- float *out[3] = {filter_buffer + 0*pass_stride, filter_buffer + 1*pass_stride, filter_buffer + 2*pass_stride};
- for(int y = rect.y; y < rect.w; y++) {
- for(int x = rect.x; x < rect.z; x++) {
- filter_non_local_means_3_kernel()(x, y, img, img, var, out, &rect.x, 10, 4, 1, 0.04f);
- }
- }
- for(int y = 0; y < filter_area.w; y++) {
- int py = y + filter_area.y;
- for(int x = 0; x < filter_area.z; x++) {
- int px = x + filter_area.x;
- int i = (py - rect.y)*w + (px - rect.x);
- float *loc_buf = buffers + (offset + py*stride + px)*kg->__data.film.pass_stride;
- loc_buf[0] = sample*filter_buffer[0*pass_stride + i];
- loc_buf[1] = sample*filter_buffer[1*pass_stride + i];
- loc_buf[2] = sample*filter_buffer[2*pass_stride + i];
- }
- }
- }
- else if(nlm_weights) {
+ if(nlm_weights) {
for(int y = 0; y < filter_area.w; y++) {
for(int x = 0; x < filter_area.z; x++) {
filter_construct_transform_kernel()(kg, sample, filter_buffer, x + filter_area.x, y + filter_area.y, storage + y*filter_area.z + x, &rect.x);
diff --git a/intern/cycles/kernel/filter/filter_nlm.h b/intern/cycles/kernel/filter/filter_nlm.h
index 59e9fed2f2..8888660370 100644
--- a/intern/cycles/kernel/filter/filter_nlm.h
+++ b/intern/cycles/kernel/filter/filter_nlm.h
@@ -66,74 +66,4 @@ ccl_device float nlm_weight(int px, int py, int qx, int qy, float ccl_readonly_p
return fast_expf(-max(0.0f, dI));
}
-ccl_device void kernel_filter_non_local_means_3(int x, int y, float ccl_readonly_ptr noisyImage[3], float ccl_readonly_ptr weightImage[3], float ccl_readonly_ptr variance[3], float *filteredImage[3], int4 rect, int r, int f, float a, float k_2)
-{
- int2 low = make_int2(max(rect.x, x - r),
- max(rect.y, y - r));
- int2 high = make_int2(min(rect.z, x + r + 1),
- min(rect.w, y + r + 1));
-
- float sum_image[3] = {0.0f}, sum_weight = 0.0f;
-
- int w = align_up(rect.z - rect.x, 4);
- int p_idx = (y-rect.y)*w + (x - rect.x);
- int q_idx = (low.y-rect.y)*w + (low.x-rect.x);
-#ifdef __KERNEL_SSE41__
- __m128 a_sse = _mm_set1_ps(a), k_2_sse = _mm_set1_ps(k_2);
-#endif
- /* Loop over the q's, center pixels of all relevant patches. */
- for(int qy = low.y; qy < high.y; qy++) {
- for(int qx = low.x; qx < high.x; qx++, q_idx++) {
- int2 low_dPatch = make_int2(max(max(rect.x - qx, rect.x - x), -f), max(max(rect.y - qy, rect.y - y), -f));
- int2 high_dPatch = make_int2(min(min(rect.z - qx, rect.z - x), f+1), min(min(rect.w - qy, rect.w - y), f+1));
- /* Loop over the pixels in the patch.
- * Note that the patch must be small enough to be fully inside the rect, both at p and q.
- * Do avoid doing all the coordinate calculations twice, the code here computes both weights at once. */
-#ifdef __KERNEL_SSE41__
- __m128 dI_sse = _mm_setzero_ps();
- __m128 highX_sse = _mm_set1_ps(high_dPatch.x);
- for(int k = 0; k < 3; k++) {
- int dIdx = low_dPatch.x + low_dPatch.y*w;
- for(int dy = low_dPatch.y; dy < high_dPatch.y; dy++) {
- int dx;
- for(dx = low_dPatch.x; dx < high_dPatch.x; dx+=4, dIdx+=4) {
- __m128 diff = _mm_sub_ps(_mm_loadu_ps(weightImage[k] + p_idx + dIdx), _mm_loadu_ps(weightImage[k] + q_idx + dIdx));
- __m128 pvar = _mm_loadu_ps(variance[k] + p_idx + dIdx);
- __m128 qvar = _mm_loadu_ps(variance[k] + q_idx + dIdx);
- __m128 d = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(diff, diff), _mm_mul_ps(a_sse, _mm_add_ps(pvar, _mm_min_ps(pvar, qvar)))), _mm_rcp_ps(_mm_add_ps(_mm_set1_ps(1e-7f), _mm_mul_ps(k_2_sse, _mm_add_ps(pvar, qvar)))));
- dI_sse = _mm_add_ps(dI_sse, _mm_mask_ps(d, _mm_cmplt_ps(_mm_add_ps(_mm_set1_ps(dx), _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f)), highX_sse)));
- }
- dIdx += w-(dx - low_dPatch.x);
- }
- }
- float dI = _mm_hsum_ss(dI_sse);
-#else
- float dI = 0.0f;
- for(int k = 0; k < 3; k++) {
- int dIdx = low_dPatch.x + low_dPatch.y*w;
- for(int dy = low_dPatch.y; dy < high_dPatch.y; dy++) {
- for(int dx = low_dPatch.x; dx < high_dPatch.x; dx++, dIdx++) {
- float diff = weightImage[k][p_idx+dIdx] - weightImage[k][q_idx+dIdx];
- dI += (diff*diff - a*(variance[k][p_idx+dIdx] + min(variance[k][p_idx+dIdx], variance[k][q_idx+dIdx]))) * (1.0f / (1e-7f + k_2*(variance[k][p_idx+dIdx] + variance[k][q_idx+dIdx])));
- }
- dIdx += w-(high_dPatch.x - low_dPatch.x);
- }
- }
-#endif
- dI *= 1.0f / (3.0f * (high_dPatch.x - low_dPatch.x) * (high_dPatch.y - low_dPatch.y));
-
- float wI = fast_expf(-max(0.0f, dI));
- sum_image[0] += wI*noisyImage[0][q_idx];
- sum_image[1] += wI*noisyImage[1][q_idx];
- sum_image[2] += wI*noisyImage[2][q_idx];
- sum_weight += wI;
- }
- q_idx += w-(high.x-low.x);
- }
-
- filteredImage[0][p_idx] = sum_image[0] / sum_weight;
- filteredImage[1][p_idx] = sum_image[1] / sum_weight;
- filteredImage[2][p_idx] = sum_image[2] / sum_weight;
-}
-
CCL_NAMESPACE_END
\ No newline at end of file
diff --git a/intern/cycles/kernel/kernels/cpu/kernel_cpu.h b/intern/cycles/kernel/kernels/cpu/kernel_cpu.h
index 4274ef271b..52a05ddf3b 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel_cpu.h
+++ b/intern/cycles/kernel/kernels/cpu/kernel_cpu.h
@@ -88,15 +88,6 @@ void KERNEL_FUNCTION_FULL_NAME(filter_non_local_means)(int x, int y,
int r, int f,
float a, float k_2);
-void KERNEL_FUNCTION_FULL_NAME(filter_non_local_means_3)(int x, int y,
- float *noisyImage[3],
- float *weightImage[3],
- float *variance[3],
- float *filteredImage[3],
- int* rect,
- int r, int f,
- float a, float k_2);
-
void KERNEL_FUNCTION_FULL_NAME(filter_combine_halves)(int x, int y,
float *mean,
float *variance,
diff --git a/intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h b/intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h
index 30e3bec3a7..56aa7e52f8 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h
+++ b/intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h
@@ -216,26 +216,6 @@ void KERNEL_FUNCTION_FULL_NAME(filter_non_local_means)(int x, int y,
#endif
}
-void KERNEL_FUNCTION_FULL_NAME(filter_non_local_means_3)(int x, int y,
- float *noisyImage[3],
- float *weightImage[3],
- float *variance[3],
- float *filteredImage[3],
- int* filter_rect,
- int r, int f,
- float a, float k_2)
-{
-#ifdef KERNEL_STUB
- STUB_ASSERT(KERNEL_ARCH, filter_non_local_means_3);
-#else
- int4 rect = make_int4(filter_rect[0], filter_rect[1], filter_rect[2], filter_rect[3]);
- kernel_filter_non_local_means_3(x, y,
- (float ccl_readonly_ptr*) noisyImage,
- (float ccl_readonly_ptr*) weightImage,
- (float ccl_readonly_ptr*) variance,
- filteredImage, rect, r, f, a, k_2);
-#endif
-}
void KERNEL_FUNCTION_FULL_NAME(filter_combine_halves)(int x, int y,
float *mean,
More information about the Bf-blender-cvs
mailing list