diff options
| author | Lukas Stockner <lukas.stockner@freenet.de> | 2018-10-06 21:39:01 +0300 |
|---|---|---|
| committer | Lukas Stockner <lukas.stockner@freenet.de> | 2018-10-06 22:49:54 +0300 |
| commit | a0cc7bd961ef2cd501468dce08998992a88b3bed (patch) | |
| tree | 90a26871e42e7aafc7d17c0172b4513a501896f8 /intern/cycles/device/device_cpu.cpp | |
| parent | 6e08b01bd9ab87f63d8176bf2ec0b4f073a92ee3 (diff) | |
Cycles: Implement vectorized NLM kernels for faster CPU denoising
Diffstat (limited to 'intern/cycles/device/device_cpu.cpp')
| -rw-r--r-- | intern/cycles/device/device_cpu.cpp | 12 |
1 files changed, 7 insertions, 5 deletions
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp index 7c72ab1a009..4986bb809fb 100644 --- a/intern/cycles/device/device_cpu.cpp +++ b/intern/cycles/device/device_cpu.cpp @@ -184,11 +184,11 @@ public: KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int)> filter_detect_outliers_kernel; KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int)> filter_combine_halves_kernel; - KernelFunctions<void(*)(int, int, float*, float*, float*, int*, int, int, float, float)> filter_nlm_calc_difference_kernel; - KernelFunctions<void(*)(float*, float*, int*, int, int)> filter_nlm_blur_kernel; - KernelFunctions<void(*)(float*, float*, int*, int, int)> filter_nlm_calc_weight_kernel; - KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int, int)> filter_nlm_update_output_kernel; - KernelFunctions<void(*)(float*, float*, int*, int)> filter_nlm_normalize_kernel; + KernelFunctions<void(*)(int, int, float*, float*, float*, int*, int, int, float, float)> filter_nlm_calc_difference_kernel; + KernelFunctions<void(*)(float*, float*, int*, int, int)> filter_nlm_blur_kernel; + KernelFunctions<void(*)(float*, float*, int*, int, int)> filter_nlm_calc_weight_kernel; + KernelFunctions<void(*)(int, int, float*, float*, float*, float*, float*, int*, int, int)> filter_nlm_update_output_kernel; + KernelFunctions<void(*)(float*, float*, int*, int)> filter_nlm_normalize_kernel; KernelFunctions<void(*)(float*, int, int, int, float*, int*, int*, int, int, float)> filter_construct_transform_kernel; KernelFunctions<void(*)(int, int, float*, float*, float*, int*, float*, float3*, int*, int*, int, int, int)> filter_nlm_construct_gramian_kernel; @@ -475,6 +475,7 @@ public: float *blurDifference = temporary_mem; float *difference = temporary_mem + task->buffer.pass_stride; float *weightAccum = temporary_mem + 2*task->buffer.pass_stride; + float *temp_image = temporary_mem + 3*task->buffer.pass_stride; memset(weightAccum, 0, sizeof(float)*w*h); memset((float*) out_ptr, 0, sizeof(float)*w*h); @@ -499,6 +500,7 @@ public: filter_nlm_update_output_kernel()(dx, dy, blurDifference, (float*) image_ptr, + temp_image, (float*) out_ptr, weightAccum, local_rect, |