Cycles: OpenCL image texture support, fix an attribute node issue and refactor

feature enabling #defines a bit.

8 files changed, 129 insertions, 15 deletions

diff --git a/intern/cycles/kernel/kernel_compat_cpu.h b/intern/cycles/kernel/kernel_compat_cpu.h
index 79f894bfdac..2bd0b61b4fa 100644
--- a/intern/cycles/kernel/kernel_compat_cpu.h
+++ b/intern/cycles/kernel/kernel_compat_cpu.h
@@ -146,6 +146,7 @@ typedef texture<float> texture_float;
 typedef texture<uint> texture_uint;
 typedef texture<int> texture_int;
 typedef texture<uint4> texture_uint4;
+typedef texture<uchar4> texture_uchar4;
 typedef texture_image<float4> texture_image_float4;
 typedef texture_image<uchar4> texture_image_uchar4;
 
diff --git a/intern/cycles/kernel/kernel_compat_cuda.h b/intern/cycles/kernel/kernel_compat_cuda.h
index cc719bfadbc..2f9f2c45e88 100644
--- a/intern/cycles/kernel/kernel_compat_cuda.h
+++ b/intern/cycles/kernel/kernel_compat_cuda.h
@@ -50,6 +50,7 @@ typedef texture<float, 1> texture_float;
 typedef texture<uint, 1> texture_uint;
 typedef texture<int, 1> texture_int;
 typedef texture<uint4, 1> texture_uint4;
+typedef texture<uchar4, 1> texture_uchar4;
 typedef texture<float4, 2> texture_image_float4;
 typedef texture<uchar4, 2, cudaReadModeNormalizedFloat> texture_image_uchar4;
 
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index cd9557bd0bf..e9e7fbd4ca1 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -59,6 +59,7 @@ __device float3 area_light_sample(float3 axisu, float3 axisv, float randu, float
 	return axisu*randu + axisv*randv;
 }
 
+#ifdef __BACKGROUND_MIS__
 __device float3 background_light_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
 {
 	/* for the following, the CDF values are actually a pair of floats, with the
@@ -165,6 +166,7 @@ __device float background_light_pdf(KernelGlobals *kg, float3 direction)
 
 	return pdf * kernel_data.integrator.pdf_lights;
 }
+#endif
 
 __device void regular_light_sample(KernelGlobals *kg, int point,
 	float randu, float randv, float3 P, LightSample *ls, float *pdf)
diff --git a/intern/cycles/kernel/kernel_textures.h b/intern/cycles/kernel/kernel_textures.h
index f4de4c100c4..4ab2574c8ad 100644
--- a/intern/cycles/kernel/kernel_textures.h
+++ b/intern/cycles/kernel/kernel_textures.h
@@ -7,7 +7,6 @@
 #define KERNEL_IMAGE_TEX(type, ttype, name)
 #endif
 
-
 /* bvh */
 KERNEL_TEX(float4, texture_float4, __bvh_nodes)
 KERNEL_TEX(float4, texture_float4, __tri_woop)
@@ -151,6 +150,10 @@ KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_097)
 KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_098)
 KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_099)
 
+/* packed image (opencl) */
+KERNEL_TEX(uchar4, texture_uchar4, __tex_image_packed)
+KERNEL_TEX(uint4, texture_uint4, __tex_image_packed_info)
+
 #undef KERNEL_TEX
 #undef KERNEL_IMAGE_TEX
 
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index edca9f8d34d..a64c850d35a 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -49,8 +49,30 @@ CCL_NAMESPACE_BEGIN
 #endif
 
 #ifdef __KERNEL_OPENCL__
-//#define __KERNEL_SHADING__
-//#define __KERNEL_ADV_SHADING__
+
+#ifdef __KERNEL_OPENCL_NVIDIA__
+#define __KERNEL_SHADING__
+#define __MULTI_CLOSURE__
+#endif
+
+#ifdef __KERNEL_OPENCL_APPLE__
+//#define __SVM__
+//#define __EMISSION__
+//#define __IMAGE_TEXTURES__
+//#define __HOLDOUT__
+//#define __PROCEDURAL_TEXTURES__
+//#define __EXTRA_NODES__
+#endif
+
+#ifdef __KERNEL_OPENCL_AMD__
+#define __SVM__
+#define __EMISSION__
+#define __IMAGE_TEXTURES__
+#define __HOLDOUT__
+#define __PROCEDURAL_TEXTURES__
+#define __EXTRA_NODES__
+#endif
+
 #endif
 
 /* kernel features */
@@ -69,7 +91,9 @@ CCL_NAMESPACE_BEGIN
 #ifdef __KERNEL_SHADING__
 #define __SVM__
 #define __EMISSION__
-#define __TEXTURES__
+#define __PROCEDURAL_TEXTURES__
+#define __IMAGE_TEXTURES__
+#define __EXTRA_NODES__
 #define __HOLDOUT__
 #endif
 
@@ -85,7 +109,6 @@ CCL_NAMESPACE_BEGIN
 //#define __MULTI_LIGHT__
 //#define __OSL__
 //#define __SOBOL_FULL_SCREEN__
-//#define __MODIFY_TP__
 //#define __QBVH__
 
 /* Shader Evaluation */
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index 5f4d7bbd0c4..12ed61673d1 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -216,13 +216,15 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_JUMP:
 				offset = node.y;
 				break;
-#ifdef __TEXTURES__
+#ifdef __IMAGE_TEXTURES__
 			case NODE_TEX_IMAGE:
 				svm_node_tex_image(kg, sd, stack, node);
 				break;
 			case NODE_TEX_ENVIRONMENT:
 				svm_node_tex_environment(kg, sd, stack, node);
 				break;
+#endif
+#ifdef __PROCEDURAL_TEXTURES__
 			case NODE_TEX_SKY:
 				svm_node_tex_sky(kg, sd, stack, node.y, node.z);
 				break;
@@ -254,6 +256,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_GEOMETRY:
 				svm_node_geometry(sd, stack, node.y, node.z);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_GEOMETRY_BUMP_DX:
 				svm_node_geometry_bump_dx(sd, stack, node.y, node.z);
 				break;
@@ -263,6 +266,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_PATH:
 				svm_node_light_path(sd, stack, node.y, node.z, path_flag);
 				break;
+#endif
 			case NODE_CONVERT:
 				svm_node_convert(sd, stack, node.y, node.z, node.w);
 				break;
@@ -272,6 +276,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_VALUE_V:
 				svm_node_value_v(kg, sd, stack, node.y, &offset);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_INVERT:
 				svm_node_invert(sd, stack, node.y, node.z, node.w);
 				break;
@@ -293,21 +298,25 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_HSV:
 				svm_node_hsv(kg, sd, stack, node.y, node.z, node.w, &offset);
 				break;
+#endif
 			case NODE_ATTR:
 				svm_node_attr(kg, sd, stack, node);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_ATTR_BUMP_DX:
 				svm_node_attr_bump_dx(kg, sd, stack, node);
 				break;
 			case NODE_ATTR_BUMP_DY:
 				svm_node_attr_bump_dy(kg, sd, stack, node);
 				break;
+#endif
 			case NODE_FRESNEL:
 				svm_node_fresnel(sd, stack, node.y, node.z, node.w);
 				break;
 			case NODE_LAYER_WEIGHT:
 				svm_node_layer_weight(sd, stack, node);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_SET_DISPLACEMENT:
 				svm_node_set_displacement(sd, stack, node.y);
 				break;
@@ -323,6 +332,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_NORMAL:
 				svm_node_normal(kg, sd, stack, node.y, node.z, node.w, &offset);
 				break;
+#endif
 			case NODE_MAPPING:
 				svm_node_mapping(kg, sd, stack, node.y, node.z, &offset);
 				break;
@@ -332,15 +342,18 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_TEX_COORD:
 				svm_node_tex_coord(kg, sd, stack, node.y, node.z);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_TEX_COORD_BUMP_DX:
 				svm_node_tex_coord_bump_dx(kg, sd, stack, node.y, node.z);
 				break;
 			case NODE_TEX_COORD_BUMP_DY:
 				svm_node_tex_coord_bump_dy(kg, sd, stack, node.y, node.z);
 				break;
+#endif
 			case NODE_EMISSION_SET_WEIGHT_TOTAL:
 				svm_node_emission_set_weight_total(kg, sd, node.y, node.z, node.w);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_RGB_RAMP:
 				svm_node_rgb_ramp(kg, sd, stack, node, &offset);
 				break;
@@ -350,6 +363,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_FALLOFF:
 				svm_node_light_falloff(sd, stack, node);
 				break;
+#endif
 			case NODE_END:
 			default:
 #ifndef __MULTI_CLOSURE__
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index 3a94f08d42f..ed70a6dc423 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -22,7 +22,7 @@ CCL_NAMESPACE_BEGIN
 
 __device void svm_node_attr_init(KernelGlobals *kg, ShaderData *sd,
 	uint4 node, NodeAttributeType *type,
-	NodeAttributeType *mesh_type, AttributeElement *elem, uint *offset, uint *out_offset)
+	NodeAttributeType *mesh_type, AttributeElement *elem, int *offset, uint *out_offset)
 {
 	if(sd->object != ~0) {
 		/* find attribute by unique id */
@@ -35,7 +35,7 @@ __device void svm_node_attr_init(KernelGlobals *kg, ShaderData *sd,
 
 		/* return result */
 		*elem = (AttributeElement)attr_map.y;
-		*offset = attr_map.z;
+		*offset = as_int(attr_map.z);
 		*mesh_type = (NodeAttributeType)attr_map.w;
 	}
 	else {
@@ -53,7 +53,8 @@ __device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, uin
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
@@ -84,7 +85,8 @@ __device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *st
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
@@ -119,7 +121,8 @@ __device void svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *st
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index eddd0f7034a..388f006c40f 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -18,6 +18,75 @@
 
 CCL_NAMESPACE_BEGIN
 
+#ifdef __KERNEL_OPENCL__
+
+/* For OpenCL all images are packed in a single array, and we do manual lookup
+ * and interpolation. */
+
+__device_inline float4 svm_image_texture_read(KernelGlobals *kg, int offset)
+{
+	uchar4 r = kernel_tex_fetch(__tex_image_packed, offset);
+	float f = 1.0f/255.0f;
+	return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
+}
+
+__device_inline int svm_image_texture_wrap_periodic(int x, int width)
+{
+	x %= width;
+	if(x < 0)
+		x += width;
+	return x;
+}
+
+__device_inline int svm_image_texture_wrap_clamp(int x, int width)
+{
+	return clamp(x, 0, width-1);
+}
+
+__device_inline float svm_image_texture_frac(float x, int *ix)
+{
+	int i = (int)x - ((x < 0.0f)? 1: 0);
+	*ix = i;
+	return x - (float)i;
+}
+
+__device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
+{
+	uint4 info = kernel_tex_fetch(__tex_image_packed_info, id);
+	uint width = info.x;
+	uint height = info.y;
+	uint offset = info.z;
+	uint periodic = info.w;
+
+	int ix, iy, nix, niy;
+	float tx = svm_image_texture_frac(x*width, &ix);
+	float ty = svm_image_texture_frac(y*height, &iy);
+
+	if(periodic) {
+		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 {
+		ix = svm_image_texture_wrap_clamp(ix, width);
+		iy = svm_image_texture_wrap_clamp(iy, height);
+
+		nix = svm_image_texture_wrap_clamp(ix+1, width);
+		niy = svm_image_texture_wrap_clamp(iy+1, height);
+	}
+
+	float4 r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + iy*width);
+	r += (1.0f - ty)*tx*svm_image_texture_read(kg, offset + nix + iy*width);
+	r += ty*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + niy*width);
+	r += ty*tx*svm_image_texture_read(kg, offset + nix + niy*width);
+
+	return r;
+}
+
+#else
+
 __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 {
 	float4 r;
@@ -31,9 +100,6 @@ __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 	   also note that cuda has 128 textures limit, we use 100 now, since
 	   we still need some for other storage */
 
-#ifdef __KERNEL_OPENCL__
-	r = make_float4(0.0f, 0.0f, 0.0f, 0.0f); /* todo */
-#else
 	switch(id) {
 		case 0: r = kernel_tex_image_interp(__tex_image_000, x, y); break;
 		case 1: r = kernel_tex_image_interp(__tex_image_001, x, y); break;
@@ -139,11 +205,12 @@ __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 			kernel_assert(0);
 			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 	}
-#endif
 
 	return r;
 }
 
+#endif
+
 __device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
 	uint id = node.y;