Unlimited number of textures for Cycles

This patch allows for an unlimited number of textures in Cycles where the hardware allows. It replaces a number static arrays with dynamic arrays and changes the way the flat_slot indices are calculated. Eventually, I'd like to get to a point where there are only flat slots left and textures off all kinds are stored in a single array.

Note that the arrays in DeviceScene are changed from containing device_vector<T> objects to device_vector<T>* pointers. Ideally, I'd like to store objects, but dynamic resizing of a std:vector in pre-C++11 calls the copy constructor, which for a good reason is not implemented for device_vector. Once we require C++11 for Cycles builds, we can implement a move constructor for device_vector and store objects again.

The limits for CUDA Fermi hardware still apply.

Reviewers: tod_baudais, InsigMathK, dingto, #cycles

Reviewed By: dingto, #cycles

Subscribers: dingto, smellslikedonkey

Differential Revision: https://developer.blender.org/D2650

6 files changed, 104 insertions, 68 deletions

diff --git a/intern/cycles/kernel/kernel_globals.h b/intern/cycles/kernel/kernel_globals.h
index c9c97ea977e..dd0971e4a06 100644
--- a/intern/cycles/kernel/kernel_globals.h
+++ b/intern/cycles/kernel/kernel_globals.h
@@ -19,6 +19,10 @@
 #ifndef __KERNEL_GLOBALS_H__
 #define __KERNEL_GLOBALS_H__
 
+#ifdef __KERNEL_CPU__
+#include "util/util_vector.h"
+#endif
+
 CCL_NAMESPACE_BEGIN
 
 /* On the CPU, we pass along the struct KernelGlobals to nearly everywhere in
@@ -38,12 +42,12 @@ struct Intersection;
 struct VolumeStep;
 
 typedef struct KernelGlobals {
-	texture_image_uchar4 texture_byte4_images[TEX_NUM_BYTE4_CPU];
-	texture_image_float4 texture_float4_images[TEX_NUM_FLOAT4_CPU];
-	texture_image_half4 texture_half4_images[TEX_NUM_HALF4_CPU];
-	texture_image_float texture_float_images[TEX_NUM_FLOAT_CPU];
-	texture_image_uchar texture_byte_images[TEX_NUM_BYTE_CPU];
-	texture_image_half texture_half_images[TEX_NUM_HALF_CPU];
+	vector<texture_image_uchar4> texture_byte4_images;
+	vector<texture_image_float4> texture_float4_images;
+	vector<texture_image_half4> texture_half4_images;
+	vector<texture_image_float> texture_float_images;
+	vector<texture_image_uchar> texture_byte_images;
+	vector<texture_image_half> texture_half_images;
 
 #  define KERNEL_TEX(type, ttype, name) ttype name;
 #  define KERNEL_IMAGE_TEX(type, ttype, name)
diff --git a/intern/cycles/kernel/kernel_image_opencl.h b/intern/cycles/kernel/kernel_image_opencl.h
index 0352c58037d..15579f55a41 100644
--- a/intern/cycles/kernel/kernel_image_opencl.h
+++ b/intern/cycles/kernel/kernel_image_opencl.h
@@ -20,18 +20,19 @@
 
 ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int offset)
 {
+	const texture_type = kernel_tex_type(id);
 	/* Float4 */
-	if(id < TEX_START_BYTE4_OPENCL) {
+	if(texture_type == IMAGE_DATA_TYPE_FLOAT4) {
 		return kernel_tex_fetch(__tex_image_float4_packed, offset);
 	}
 	/* Byte4 */
-	else if(id < TEX_START_FLOAT_OPENCL) {
+	else if(texture_type == IMAGE_DATA_TYPE_BYTE4) {
 		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) {
+	else if(texture_type == IMAGE_DATA_TYPE_FLOAT) {
 		float f = kernel_tex_fetch(__tex_image_float_packed, offset);
 		return make_float4(f, f, f, 1.0f);
 	}
diff --git a/intern/cycles/kernel/kernels/cpu/kernel.cpp b/intern/cycles/kernel/kernels/cpu/kernel.cpp
index 16992c681e6..db2de6836d3 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel.cpp
+++ b/intern/cycles/kernel/kernels/cpu/kernel.cpp
@@ -95,9 +95,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_float4")) {
 		texture_image_float4 *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_float4_"));
-		int array_index = id;
+		int array_index = kernel_tex_index(id);
 
-		if(array_index >= 0 && array_index < TEX_NUM_FLOAT4_CPU) {
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_float4_images.size()) {
+				kg->texture_float4_images.resize(array_index+1);
+			}
 			tex = &kg->texture_float4_images[array_index];
 		}
 
@@ -111,9 +114,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_float")) {
 		texture_image_float *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_float_"));
-		int array_index = id - TEX_START_FLOAT_CPU;
-
-		if(array_index >= 0 && array_index < TEX_NUM_FLOAT_CPU) {
+		int array_index = kernel_tex_index(id);
+		
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_float_images.size()) {
+				kg->texture_float_images.resize(array_index+1);
+			}
 			tex = &kg->texture_float_images[array_index];
 		}
 
@@ -127,9 +133,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_byte4")) {
 		texture_image_uchar4 *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_byte4_"));
-		int array_index = id - TEX_START_BYTE4_CPU;
-
-		if(array_index >= 0 && array_index < TEX_NUM_BYTE4_CPU) {
+		int array_index = kernel_tex_index(id);
+		
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_byte4_images.size()) {
+				kg->texture_byte4_images.resize(array_index+1);
+			}
 			tex = &kg->texture_byte4_images[array_index];
 		}
 
@@ -143,9 +152,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_byte")) {
 		texture_image_uchar *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_byte_"));
-		int array_index = id - TEX_START_BYTE_CPU;
-
-		if(array_index >= 0 && array_index < TEX_NUM_BYTE_CPU) {
+		int array_index = kernel_tex_index(id);
+		
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_byte_images.size()) {
+				kg->texture_byte_images.resize(array_index+1);
+			}
 			tex = &kg->texture_byte_images[array_index];
 		}
 
@@ -159,9 +171,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_half4")) {
 		texture_image_half4 *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_half4_"));
-		int array_index = id - TEX_START_HALF4_CPU;
-
-		if(array_index >= 0 && array_index < TEX_NUM_HALF4_CPU) {
+		int array_index = kernel_tex_index(id);
+		
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_half4_images.size()) {
+				kg->texture_half4_images.resize(array_index+1);
+			}
 			tex = &kg->texture_half4_images[array_index];
 		}
 
@@ -175,9 +190,12 @@ void kernel_tex_copy(KernelGlobals *kg,
 	else if(strstr(name, "__tex_image_half")) {
 		texture_image_half *tex = NULL;
 		int id = atoi(name + strlen("__tex_image_half_"));
-		int array_index = id - TEX_START_HALF_CPU;
-
-		if(array_index >= 0 && array_index < TEX_NUM_HALF_CPU) {
+		int array_index = kernel_tex_index(id);
+		
+		if(array_index >= 0) {
+			if(array_index >= kg->texture_half_images.size()) {
+				kg->texture_half_images.resize(array_index+1);
+			}
 			tex = &kg->texture_half_images[array_index];
 		}
 
diff --git a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
index af68907a5c2..f6bb4c25012 100644
--- a/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
+++ b/intern/cycles/kernel/kernels/cpu/kernel_cpu_image.h
@@ -23,51 +23,59 @@ CCL_NAMESPACE_BEGIN
 
 ccl_device float4 kernel_tex_image_interp_impl(KernelGlobals *kg, int tex, float x, float y)
 {
-	if(tex >= TEX_START_HALF_CPU)
-		return kg->texture_half_images[tex - TEX_START_HALF_CPU].interp(x, y);
-	else if(tex >= TEX_START_BYTE_CPU)
-		return kg->texture_byte_images[tex - TEX_START_BYTE_CPU].interp(x, y);
-	else if(tex >= TEX_START_FLOAT_CPU)
-		return kg->texture_float_images[tex - TEX_START_FLOAT_CPU].interp(x, y);
-	else if(tex >= TEX_START_HALF4_CPU)
-		return kg->texture_half4_images[tex - TEX_START_HALF4_CPU].interp(x, y);
-	else if(tex >= TEX_START_BYTE4_CPU)
-		return kg->texture_byte4_images[tex - TEX_START_BYTE4_CPU].interp(x, y);
-	else
-		return kg->texture_float4_images[tex].interp(x, y);
+	switch(kernel_tex_type(tex)) {
+		case IMAGE_DATA_TYPE_HALF:
+			return kg->texture_half_images[kernel_tex_index(tex)].interp(x, y);
+		case IMAGE_DATA_TYPE_BYTE:
+			return kg->texture_byte_images[kernel_tex_index(tex)].interp(x, y);
+		case IMAGE_DATA_TYPE_FLOAT:
+			return kg->texture_float_images[kernel_tex_index(tex)].interp(x, y);
+		case IMAGE_DATA_TYPE_HALF4:
+			return kg->texture_half4_images[kernel_tex_index(tex)].interp(x, y);
+		case IMAGE_DATA_TYPE_BYTE4:
+			return kg->texture_byte4_images[kernel_tex_index(tex)].interp(x, y);
+		case IMAGE_DATA_TYPE_FLOAT4:
+		default:
+			return kg->texture_float4_images[kernel_tex_index(tex)].interp(x, y);
+	}
 }
 
 ccl_device float4 kernel_tex_image_interp_3d_impl(KernelGlobals *kg, int tex, float x, float y, float z)
 {
-	if(tex >= TEX_START_HALF_CPU)
-		return kg->texture_half_images[tex - TEX_START_HALF_CPU].interp_3d(x, y, z);
-	else if(tex >= TEX_START_BYTE_CPU)
-		return kg->texture_byte_images[tex - TEX_START_BYTE_CPU].interp_3d(x, y, z);
-	else if(tex >= TEX_START_FLOAT_CPU)
-		return kg->texture_float_images[tex - TEX_START_FLOAT_CPU].interp_3d(x, y, z);
-	else if(tex >= TEX_START_HALF4_CPU)
-		return kg->texture_half4_images[tex - TEX_START_HALF4_CPU].interp_3d(x, y, z);
-	else if(tex >= TEX_START_BYTE4_CPU)
-		return kg->texture_byte4_images[tex - TEX_START_BYTE4_CPU].interp_3d(x, y, z);
-	else
-		return kg->texture_float4_images[tex].interp_3d(x, y, z);
-
+	switch(kernel_tex_type(tex)) {
+		case IMAGE_DATA_TYPE_HALF:
+			return kg->texture_half_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+		case IMAGE_DATA_TYPE_BYTE:
+			return kg->texture_byte_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+		case IMAGE_DATA_TYPE_FLOAT:
+			return kg->texture_float_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+		case IMAGE_DATA_TYPE_HALF4:
+			return kg->texture_half4_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+		case IMAGE_DATA_TYPE_BYTE4:
+			return kg->texture_byte4_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+		case IMAGE_DATA_TYPE_FLOAT4:
+		default:
+			return kg->texture_float4_images[kernel_tex_index(tex)].interp_3d(x, y, z);
+	}
 }
 
 ccl_device float4 kernel_tex_image_interp_3d_ex_impl(KernelGlobals *kg, int tex, float x, float y, float z, int interpolation)
 {
-	if(tex >= TEX_START_HALF_CPU)
-		return kg->texture_half_images[tex - TEX_START_HALF_CPU].interp_3d_ex(x, y, z, interpolation);
-	else if(tex >= TEX_START_BYTE_CPU)
-		return kg->texture_byte_images[tex - TEX_START_BYTE_CPU].interp_3d_ex(x, y, z, interpolation);
-	else if(tex >= TEX_START_FLOAT_CPU)
-		return kg->texture_float_images[tex - TEX_START_FLOAT_CPU].interp_3d_ex(x, y, z, interpolation);
-	else if(tex >= TEX_START_HALF4_CPU)
-		return kg->texture_half4_images[tex - TEX_START_HALF4_CPU].interp_3d_ex(x, y, z, interpolation);
-	else if(tex >= TEX_START_BYTE4_CPU)
-		return kg->texture_byte4_images[tex - TEX_START_BYTE4_CPU].interp_3d_ex(x, y, z, interpolation);
-	else
-		return kg->texture_float4_images[tex].interp_3d_ex(x, y, z, interpolation);
+	switch(kernel_tex_type(tex)) {
+		case IMAGE_DATA_TYPE_HALF:
+			return kg->texture_half_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+		case IMAGE_DATA_TYPE_BYTE:
+			return kg->texture_byte_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+		case IMAGE_DATA_TYPE_FLOAT:
+			return kg->texture_float_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+		case IMAGE_DATA_TYPE_HALF4:
+			return kg->texture_half4_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+		case IMAGE_DATA_TYPE_BYTE4:
+			return kg->texture_byte4_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+		case IMAGE_DATA_TYPE_FLOAT4:
+		default:
+			return kg->texture_float4_images[kernel_tex_index(tex)].interp_3d_ex(x, y, z, interpolation);
+	}
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index 76acc9253a1..4b5e4ebac00 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -151,8 +151,10 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y,
 #  else
 	CUtexObject tex = kernel_tex_fetch(__bindless_mapping, id);
 	/* float4, byte4 and half4 */
-	if(id < TEX_START_FLOAT_CUDA_KEPLER)
+	const int texture_type = kernel_tex_type(id);
+	if(texture_type == IMAGE_DATA_TYPE_FLOAT4 || texture_type == IMAGE_DATA_TYPE_BYTE4 || texture_type == IMAGE_DATA_TYPE_HALF4) {
 		r = kernel_tex_image_interp_float4(tex, x, y);
+	}
 	/* float, byte and half */
 	else {
 		float f = kernel_tex_image_interp_float(tex, x, y);
@@ -166,8 +168,10 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y,
 
 	if(use_alpha && alpha != 1.0f && alpha != 0.0f) {
 		r_ssef = r_ssef / ssef(alpha);
-		if(id >= TEX_NUM_FLOAT4_IMAGES)
+		const int texture_type = kernel_tex_type(id);
+		if(texture_type == IMAGE_DATA_TYPE_BYTE4 || texture_type == IMAGE_DATA_TYPE_BYTE) {
 			r_ssef = min(r_ssef, ssef(1.0f));
+		}
 		r.w = alpha;
 	}
 
@@ -181,8 +185,9 @@ ccl_device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y,
 		r.x *= invw;
 		r.y *= invw;
 		r.z *= invw;
-
-		if(id >= TEX_NUM_FLOAT4_IMAGES) {
+		
+		const int texture_type = kernel_tex_type(id);
+		if(texture_type == IMAGE_DATA_TYPE_BYTE4 || texture_type == IMAGE_DATA_TYPE_BYTE) {
 			r.x = min(r.x, 1.0f);
 			r.y = min(r.y, 1.0f);
 			r.z = min(r.z, 1.0f);
diff --git a/intern/cycles/kernel/svm/svm_voxel.h b/intern/cycles/kernel/svm/svm_voxel.h
index 9e826c8c23f..1d97e8344bd 100644
--- a/intern/cycles/kernel/svm/svm_voxel.h
+++ b/intern/cycles/kernel/svm/svm_voxel.h
@@ -46,7 +46,7 @@ ccl_device void svm_node_tex_voxel(KernelGlobals *kg,
 #  if defined(__KERNEL_CUDA__)
 #    if __CUDA_ARCH__ >= 300
 	CUtexObject tex = kernel_tex_fetch(__bindless_mapping, id);
-	if(id < TEX_START_HALF4_CUDA_KEPLER)
+	if(kernel_tex_type(id) == IMAGE_DATA_TYPE_FLOAT4 || kernel_tex_type(id) == IMAGE_DATA_TYPE_BYTE4 || kernel_tex_type(id) == IMAGE_DATA_TYPE_HALF4)
 		r = kernel_tex_image_interp_3d_float4(tex, co.x, co.y, co.z);
 	else {
 		float f = kernel_tex_image_interp_3d_float(tex, co.x, co.y, co.z);