[Bf-blender-cvs] [803337f3f64] master: \0; 115; 0cCycles: Cleanup, use ccl_restrict instead of ccl_restrict_ptr
Sergey Sharybin
noreply at git.blender.org
Fri May 19 12:41:08 CEST 2017
Commit: 803337f3f64fed240e9adc6f286d5f9d13a5026a
Author: Sergey Sharybin
Date: Fri May 19 12:33:28 2017 +0200
Branches: master
https://developer.blender.org/rB803337f3f64fed240e9adc6f286d5f9d13a5026a
\0;115;0cCycles: Cleanup, use ccl_restrict instead of ccl_restrict_ptr
There were following issues with ccl_restrict_ptr:
- We already had ccl_restrict for all platforms.
- It was secretly adding `const` qualifier to the declaration,
which is quite weird since non-const pointer can also be
declared as restricted.
- We never in Blender are using foo_ptr or FooPtr type definitions,
so not sure why we should introduce such a thing here.
- It is absolutely wrong from semantic point of view to put pointer
into the restrict macro -- const is a part of type, not part of
hint for compiler that some pointer is never aliased.
===================================================================
M intern/cycles/kernel/filter/filter_features.h
M intern/cycles/kernel/filter/filter_features_sse.h
M intern/cycles/kernel/filter/filter_nlm_cpu.h
M intern/cycles/kernel/filter/filter_nlm_gpu.h
M intern/cycles/kernel/filter/filter_prefilter.h
M intern/cycles/kernel/filter/filter_reconstruction.h
M intern/cycles/kernel/filter/filter_transform.h
M intern/cycles/kernel/filter/filter_transform_gpu.h
M intern/cycles/kernel/filter/filter_transform_sse.h
M intern/cycles/kernel/kernel_compat_cpu.h
M intern/cycles/kernel/kernel_compat_cuda.h
M intern/cycles/kernel/kernel_compat_opencl.h
M intern/cycles/kernel/kernels/cuda/filter.cu
M intern/cycles/kernel/kernels/opencl/filter.cl
M intern/cycles/util/util_math_matrix.h
===================================================================
diff --git a/intern/cycles/kernel/filter/filter_features.h b/intern/cycles/kernel/filter/filter_features.h
index 41998c792b6..53d703de143 100644
--- a/intern/cycles/kernel/filter/filter_features.h
+++ b/intern/cycles/kernel/filter/filter_features.h
@@ -28,7 +28,11 @@
pixel_buffer += buffer_w - (high.x - low.x); \
}
-ccl_device_inline void filter_get_features(int2 pixel, ccl_global float ccl_restrict_ptr buffer, float *features, float ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_features(int2 pixel,
+ const ccl_global float *ccl_restrict buffer,
+ float *features,
+ const float *ccl_restrict mean,
+ int pass_stride)
{
features[0] = pixel.x;
features[1] = pixel.y;
@@ -46,7 +50,11 @@ ccl_device_inline void filter_get_features(int2 pixel, ccl_global float ccl_rest
}
}
-ccl_device_inline void filter_get_feature_scales(int2 pixel, ccl_global float ccl_restrict_ptr buffer, float *scales, float ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_feature_scales(int2 pixel,
+ const ccl_global float *ccl_restrict buffer,
+ float *scales,
+ const float *ccl_restrict mean,
+ int pass_stride)
{
scales[0] = fabsf(pixel.x - mean[0]);
scales[1] = fabsf(pixel.y - mean[1]);
@@ -70,19 +78,21 @@ ccl_device_inline void filter_calculate_scale(float *scale)
scale[3] = scale[4] = scale[5] = 1.0f/max(sqrtf(scale[3]), 0.01f);
}
-ccl_device_inline float3 filter_get_pixel_color(ccl_global float ccl_restrict_ptr buffer, int pass_stride)
+ccl_device_inline float3 filter_get_pixel_color(const ccl_global float *ccl_restrict buffer,
+ int pass_stride)
{
return make_float3(ccl_get_feature(buffer, 0), ccl_get_feature(buffer, 1), ccl_get_feature(buffer, 2));
}
-ccl_device_inline float filter_get_pixel_variance(ccl_global float ccl_restrict_ptr buffer, int pass_stride)
+ccl_device_inline float filter_get_pixel_variance(const ccl_global float *ccl_restrict buffer,
+ int pass_stride)
{
return average(make_float3(ccl_get_feature(buffer, 0), ccl_get_feature(buffer, 1), ccl_get_feature(buffer, 2)));
}
ccl_device_inline void design_row_add(float *design_row,
int rank,
- ccl_global float ccl_restrict_ptr transform,
+ const ccl_global float *ccl_restrict transform,
int stride,
int row,
float feature)
@@ -94,13 +104,13 @@ ccl_device_inline void design_row_add(float *design_row,
/* Fill the design row. */
ccl_device_inline void filter_get_design_row_transform(int2 p_pixel,
- ccl_global float ccl_restrict_ptr p_buffer,
+ const ccl_global float *ccl_restrict p_buffer,
int2 q_pixel,
- ccl_global float ccl_restrict_ptr q_buffer,
+ const ccl_global float *ccl_restrict q_buffer,
int pass_stride,
int rank,
float *design_row,
- ccl_global float ccl_restrict_ptr transform,
+ const ccl_global float *ccl_restrict transform,
int stride)
{
design_row[0] = 1.0f;
diff --git a/intern/cycles/kernel/filter/filter_features_sse.h b/intern/cycles/kernel/filter/filter_features_sse.h
index a242a8ed0a1..ad0978d0c1d 100644
--- a/intern/cycles/kernel/filter/filter_features_sse.h
+++ b/intern/cycles/kernel/filter/filter_features_sse.h
@@ -33,7 +33,12 @@ CCL_NAMESPACE_BEGIN
pixel_buffer += buffer_w - (pixel.x - low.x); \
}
-ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y, __m128 active_pixels, float ccl_restrict_ptr buffer, __m128 *features, __m128 ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y,
+ __m128 active_pixels,
+ const float *ccl_restrict buffer,
+ __m128 *features,
+ const __m128 ccl_restrict *mean,
+ int pass_stride)
{
features[0] = x;
features[1] = y;
@@ -53,7 +58,12 @@ ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y, __m128 active
features[i] = _mm_mask_ps(features[i], active_pixels);
}
-ccl_device_inline void filter_get_feature_scales_sse(__m128 x, __m128 y, __m128 active_pixels, float ccl_restrict_ptr buffer, __m128 *scales, __m128 ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_feature_scales_sse(__m128 x, __m128 y,
+ __m128 active_pixels,
+ const float *ccl_restrict buffer,
+ __m128 *scales,
+ const __m128 *ccl_restrict mean,
+ int pass_stride)
{
scales[0] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(x, mean[0])), active_pixels);
scales[1] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(y, mean[1])), active_pixels);
diff --git a/intern/cycles/kernel/filter/filter_nlm_cpu.h b/intern/cycles/kernel/filter/filter_nlm_cpu.h
index 1a314b100be..57222811992 100644
--- a/intern/cycles/kernel/filter/filter_nlm_cpu.h
+++ b/intern/cycles/kernel/filter/filter_nlm_cpu.h
@@ -16,7 +16,15 @@
CCL_NAMESPACE_BEGIN
-ccl_device_inline void kernel_filter_nlm_calc_difference(int dx, int dy, float ccl_restrict_ptr weightImage, float ccl_restrict_ptr varianceImage, float *differenceImage, int4 rect, int w, int channel_offset, float a, float k_2)
+ccl_device_inline void kernel_filter_nlm_calc_difference(int dx, int dy,
+ const float *ccl_restrict weightImage,
+ const float *ccl_restrict varianceImage,
+ float *differenceImage,
+ int4 rect,
+ int w,
+ int channel_offset,
+ float a,
+ float k_2)
{
for(int y = rect.y; y < rect.w; y++) {
for(int x = rect.x; x < rect.z; x++) {
@@ -36,7 +44,11 @@ ccl_device_inline void kernel_filter_nlm_calc_difference(int dx, int dy, float c
}
}
-ccl_device_inline void kernel_filter_nlm_blur(float ccl_restrict_ptr differenceImage, float *outImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_blur(const float *ccl_restrict differenceImage,
+ float *outImage,
+ int4 rect,
+ int w,
+ int f)
{
#ifdef __KERNEL_SSE3__
int aligned_lowx = (rect.x & ~(3));
@@ -65,7 +77,11 @@ ccl_device_inline void kernel_filter_nlm_blur(float ccl_restrict_ptr differenceI
}
}
-ccl_device_inline void kernel_filter_nlm_calc_weight(float ccl_restrict_ptr differenceImage, float *outImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_calc_weight(const float *ccl_restrict differenceImage,
+ float *outImage,
+ int4 rect,
+ int w,
+ int f)
{
for(int y = rect.y; y < rect.w; y++) {
for(int x = rect.x; x < rect.z; x++) {
@@ -90,7 +106,14 @@ ccl_device_inline void kernel_filter_nlm_calc_weight(float ccl_restrict_ptr diff
}
}
-ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy, float ccl_restrict_ptr differenceImage, float ccl_restrict_ptr image, float *outImage, float *accumImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy,
+ const float *ccl_restrict differenceImage,
+ const float *ccl_restrict image,
+ float *outImage,
+ float *accumImage,
+ int4 rect,
+ int w,
+ int f)
{
for(int y = rect.y; y < rect.w; y++) {
for(int x = rect.x; x < rect.z; x++) {
@@ -108,8 +131,8 @@ ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy, float ccl
}
ccl_device_inline void kernel_filter_nlm_construct_gramian(int dx, int dy,
- float ccl_restrict_ptr differenceImage,
- float ccl_restric
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list