diff options
author | Hristo Gueorguiev <> | 2016-10-23 00:38:42 +0300 |
---|---|---|
committer | Brecht Van Lommel <brechtvanlommel@gmail.com> | 2016-10-23 00:49:29 +0300 |
commit | 8905c5c87495ad8fd876de83a968d79ac54f0e17 (patch) | |
tree | 81d626a6e472c138578fd3994b89c3ccd62ae47e /intern/cycles/kernel/svm | |
parent | 371d3570e0a5fd04c2181a286034841919eabd34 (diff) |
Cycles: OpenCL 3d textures support.
Note that volume rendering is not supported yet, this is a step towards that.
Reviewed By: brecht
Differential Revision: https://developer.blender.org/D2299
Diffstat (limited to 'intern/cycles/kernel/svm')
-rw-r--r-- | intern/cycles/kernel/svm/svm_image.h | 145 | ||||
-rw-r--r-- | intern/cycles/kernel/svm/svm_voxel.h | 8 |
2 files changed, 7 insertions, 146 deletions
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h index 378ce650129..9606064492e 100644 --- a/intern/cycles/kernel/svm/svm_image.h +++ b/intern/cycles/kernel/svm/svm_image.h @@ -29,147 +29,6 @@ CCL_NAMESPACE_BEGIN # define TEX_NUM_FLOAT4_IMAGES TEX_NUM_FLOAT4_OPENCL #endif -#ifdef __KERNEL_OPENCL__ - -/* For OpenCL all images are packed in a single array, and we do manual lookup - * and interpolation. */ - -ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int offset) -{ - /* Float4 */ - if(id < TEX_START_BYTE4_OPENCL) { - return kernel_tex_fetch(__tex_image_float4_packed, offset); - } - /* Byte4 */ - else if(id < TEX_START_FLOAT_OPENCL) { - uchar4 r = kernel_tex_fetch(__tex_image_byte4_packed, offset); - float f = 1.0f/255.0f; - return make_float4(r.x*f, r.y*f, r.z*f, r.w*f); - } - /* Float */ - else if(id < TEX_START_BYTE_OPENCL) { - float f = kernel_tex_fetch(__tex_image_float_packed, offset); - return make_float4(f, f, f, 1.0f); - } - /* Byte */ - else { - uchar r = kernel_tex_fetch(__tex_image_byte_packed, offset); - float f = r * (1.0f/255.0f); - return make_float4(f, f, f, 1.0f); - } -} - -ccl_device_inline int svm_image_texture_wrap_periodic(int x, int width) -{ - x %= width; - if(x < 0) - x += width; - return x; -} - -ccl_device_inline int svm_image_texture_wrap_clamp(int x, int width) -{ - return clamp(x, 0, width-1); -} - -ccl_device_inline float svm_image_texture_frac(float x, int *ix) -{ - int i = float_to_int(x) - ((x < 0.0f)? 1: 0); - *ix = i; - return x - (float)i; -} - -ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb, uint use_alpha) -{ - uint4 info = kernel_tex_fetch(__tex_image_packed_info, id); - uint width = info.x; - uint height = info.y; - uint offset = info.z; - - /* Image Options */ - uint interpolation = (info.w & (1 << 0)) ? INTERPOLATION_CLOSEST : INTERPOLATION_LINEAR; - uint extension; - if(info.w & (1 << 1)) - extension = EXTENSION_REPEAT; - else if(info.w & (1 << 2)) - extension = EXTENSION_EXTEND; - else - extension = EXTENSION_CLIP; - - float4 r; - int ix, iy, nix, niy; - if(interpolation == INTERPOLATION_CLOSEST) { - svm_image_texture_frac(x*width, &ix); - svm_image_texture_frac(y*height, &iy); - - if(extension == EXTENSION_REPEAT) { - ix = svm_image_texture_wrap_periodic(ix, width); - iy = svm_image_texture_wrap_periodic(iy, height); - } - else if(extension == EXTENSION_CLIP) { - if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) - return make_float4(0.0f, 0.0f, 0.0f, 0.0f); - } - else { /* EXTENSION_EXTEND */ - ix = svm_image_texture_wrap_clamp(ix, width); - iy = svm_image_texture_wrap_clamp(iy, height); - } - - r = svm_image_texture_read(kg, id, offset + ix + iy*width); - } - else { /* INTERPOLATION_LINEAR */ - float tx = svm_image_texture_frac(x*width - 0.5f, &ix); - float ty = svm_image_texture_frac(y*height - 0.5f, &iy); - - if(extension == EXTENSION_REPEAT) { - ix = svm_image_texture_wrap_periodic(ix, width); - iy = svm_image_texture_wrap_periodic(iy, height); - - nix = svm_image_texture_wrap_periodic(ix+1, width); - niy = svm_image_texture_wrap_periodic(iy+1, height); - } - else { - if(extension == EXTENSION_CLIP) { - if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) { - return make_float4(0.0f, 0.0f, 0.0f, 0.0f); - } - } - nix = svm_image_texture_wrap_clamp(ix+1, width); - niy = svm_image_texture_wrap_clamp(iy+1, height); - ix = svm_image_texture_wrap_clamp(ix, width); - iy = svm_image_texture_wrap_clamp(iy, height); - } - - r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, id, offset + ix + iy*width); - r += (1.0f - ty)*tx*svm_image_texture_read(kg, id, offset + nix + iy*width); - r += ty*(1.0f - tx)*svm_image_texture_read(kg, id, offset + ix + niy*width); - r += ty*tx*svm_image_texture_read(kg, id, offset + nix + niy*width); - } - - if(use_alpha && r.w != 1.0f && r.w != 0.0f) { - float invw = 1.0f/r.w; - r.x *= invw; - r.y *= invw; - r.z *= invw; - - if(id >= TEX_NUM_FLOAT4_IMAGES) { - r.x = min(r.x, 1.0f); - r.y = min(r.y, 1.0f); - r.z = min(r.z, 1.0f); - } - } - - if(srgb) { - r.x = color_srgb_to_scene_linear(r.x); - r.y = color_srgb_to_scene_linear(r.y); - r.z = color_srgb_to_scene_linear(r.z); - } - - return r; -} - -#else - ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, uint srgb, uint use_alpha) { #ifdef __KERNEL_CPU__ @@ -180,6 +39,8 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, # else float4 r = kernel_tex_image_interp(id, x, y); # endif +#elif defined(__KERNEL_OPENCL__) + float4 r = kernel_tex_image_interp(kg, id, x, y); #else float4 r; @@ -339,8 +200,6 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y, return r; } -#endif - /* Remap coordnate from 0..1 box to -1..-1 */ ccl_device_inline float3 texco_remap_square(float3 co) { diff --git a/intern/cycles/kernel/svm/svm_voxel.h b/intern/cycles/kernel/svm/svm_voxel.h index f54f4e8e888..a8b3604a8a7 100644 --- a/intern/cycles/kernel/svm/svm_voxel.h +++ b/intern/cycles/kernel/svm/svm_voxel.h @@ -43,7 +43,7 @@ ccl_device void svm_node_tex_voxel(KernelGlobals *kg, co = transform_point(&tfm, co); } float4 r; -# if defined(__KERNEL_GPU__) +# if defined(__KERNEL_CUDA__) # if __CUDA_ARCH__ >= 300 CUtexObject tex = kernel_tex_fetch(__bindless_mapping, id); if(id < 2048) /* TODO(dingto): Make this a variable */ @@ -55,9 +55,11 @@ ccl_device void svm_node_tex_voxel(KernelGlobals *kg, # else /* __CUDA_ARCH__ >= 300 */ r = volume_image_texture_3d(id, co.x, co.y, co.z); # endif -# else /* __KERNEL_GPU__ */ +# elif defined(__KERNEL_OPENCL__) + r = kernel_tex_image_interp_3d(kg, id, co.x, co.y, co.z); +# else r = kernel_tex_image_interp_3d(id, co.x, co.y, co.z); -# endif +# endif /* __KERNEL_CUDA__ */ #else float4 r = make_float4(0.0f, 0.0f, 0.0f, 0.0f); #endif |