Cycles / CUDA: Increase maximum image textures on GPU.

Instead of 95, we can use 145 images now. This only affects Kepler and above (sm30, sm_35 and sm_50).

This can be increased further if needed, but let's first test if this does not come with a performance impact.

Originally developed during my GSoC 2013.

4 files changed, 21 insertions, 10 deletions

diff --git a/intern/cycles/render/image.cpp b/intern/cycles/render/image.cpp
index 8dde642f70b..86755badc42 100644
--- a/intern/cycles/render/image.cpp
+++ b/intern/cycles/render/image.cpp
@@ -59,11 +59,16 @@ void ImageManager::set_osl_texture_system(void *texture_system)
 	osl_texture_system = texture_system;
 }
 
-void ImageManager::set_extended_image_limits(void)
+void ImageManager::set_extended_image_limits(const DeviceInfo& info)
 {
-	tex_num_images = TEX_EXTENDED_NUM_IMAGES;
-	tex_num_float_images = TEX_EXTENDED_NUM_FLOAT_IMAGES;
-	tex_image_byte_start = TEX_EXTENDED_IMAGE_BYTE_START;
+	if(info.type == DEVICE_CPU) {
+		tex_num_images = TEX_EXTENDED_NUM_IMAGES_CPU;
+		tex_num_float_images = TEX_EXTENDED_NUM_FLOAT_IMAGES;
+		tex_image_byte_start = TEX_EXTENDED_IMAGE_BYTE_START;
+	}
+	else if((info.type == DEVICE_CUDA || info.type == DEVICE_MULTI) && info.extended_images) {
+		tex_num_images = TEX_EXTENDED_NUM_IMAGES_GPU;
+	}
 }
 
 bool ImageManager::set_animation_frame_update(int frame)
diff --git a/intern/cycles/render/image.h b/intern/cycles/render/image.h
index a862ffce5c3..561550fe0d2 100644
--- a/intern/cycles/render/image.h
+++ b/intern/cycles/render/image.h
@@ -17,6 +17,7 @@
 #ifndef __IMAGE_H__
 #define __IMAGE_H__
 
+#include "device.h"
 #include "device_memory.h"
 
 #include "util_string.h"
@@ -27,11 +28,16 @@
 
 CCL_NAMESPACE_BEGIN
 
+/* generic */
 #define TEX_NUM_IMAGES			95
 #define TEX_IMAGE_BYTE_START	TEX_NUM_FLOAT_IMAGES
 
+/* extended gpu */
+#define TEX_EXTENDED_NUM_IMAGES_GPU		145
+
+/* extended cpu */
 #define TEX_EXTENDED_NUM_FLOAT_IMAGES	1024
-#define TEX_EXTENDED_NUM_IMAGES			1024
+#define TEX_EXTENDED_NUM_IMAGES_CPU		1024
 #define TEX_EXTENDED_IMAGE_BYTE_START	TEX_EXTENDED_NUM_FLOAT_IMAGES
 
 /* color to use when textures are not found */
@@ -59,7 +65,7 @@ public:
 
 	void set_osl_texture_system(void *texture_system);
 	void set_pack_images(bool pack_images_);
-	void set_extended_image_limits(void);
+	void set_extended_image_limits(const DeviceInfo& info);
 	bool set_animation_frame_update(int frame);
 
 	bool need_update;
diff --git a/intern/cycles/render/scene.cpp b/intern/cycles/render/scene.cpp
index 4a42b2c9e35..686a1bc406e 100644
--- a/intern/cycles/render/scene.cpp
+++ b/intern/cycles/render/scene.cpp
@@ -63,8 +63,8 @@ Scene::Scene(const SceneParams& params_, const DeviceInfo& device_info_)
 	else
 		shader_manager = ShaderManager::create(this, SceneParams::SVM);
 
-	if (device_info_.type == DEVICE_CPU)
-		image_manager->set_extended_image_limits();
+	/* Extended image limits for CPU and GPUs */
+	image_manager->set_extended_image_limits(device_info_);
 }
 
 Scene::~Scene()
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index c913b4c59da..0f0bb725823 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -105,8 +105,8 @@ public:
 	/* integrator */
 	device_vector<uint> sobol_directions;
 
-	/* images */
-	device_vector<uchar4> tex_image[TEX_EXTENDED_NUM_IMAGES];
+	/* cpu images */
+	device_vector<uchar4> tex_image[TEX_EXTENDED_NUM_IMAGES_CPU];
 	device_vector<float4> tex_float_image[TEX_EXTENDED_NUM_FLOAT_IMAGES];
 
 	/* opencl images */