Attempted fix for issue with latest CUDA kernel for 32 bit windows/linux/mac,

there seems to be some sort of compiler bug in CUDA toolkit 4.2, uninlining a
few functions seems to avoid it.

5 files changed, 11 insertions, 10 deletions

diff --git a/intern/cycles/kernel/kernel_emission.h b/intern/cycles/kernel/kernel_emission.h
index 4048bbd9dfc..3b27a1f49c9 100644
--- a/intern/cycles/kernel/kernel_emission.h
+++ b/intern/cycles/kernel/kernel_emission.h
@@ -20,7 +20,7 @@ CCL_NAMESPACE_BEGIN
 
 /* Direction Emission */
 
-__device float3 direct_emissive_eval(KernelGlobals *kg, float rando,
+__device_noinline float3 direct_emissive_eval(KernelGlobals *kg, float rando,
 	LightSample *ls, float u, float v, float3 I, float t, float time)
 {
 	/* setup shading at emitter */
@@ -74,7 +74,7 @@ __device float3 direct_emissive_eval(KernelGlobals *kg, float rando,
 	return eval;
 }
 
-__device bool direct_emission(KernelGlobals *kg, ShaderData *sd, int lindex,
+__device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd, int lindex,
 	float randt, float rando, float randu, float randv, Ray *ray, BsdfEval *eval,
 	bool *is_lamp)
 {
@@ -148,7 +148,7 @@ __device bool direct_emission(KernelGlobals *kg, ShaderData *sd, int lindex,
 
 /* Indirect Primitive Emission */
 
-__device float3 indirect_primitive_emission(KernelGlobals *kg, ShaderData *sd, float t, int path_flag, float bsdf_pdf)
+__device_noinline float3 indirect_primitive_emission(KernelGlobals *kg, ShaderData *sd, float t, int path_flag, float bsdf_pdf)
 {
 	/* evaluate emissive closure */
 	float3 L = shader_emissive_eval(kg, sd);
@@ -171,7 +171,7 @@ __device float3 indirect_primitive_emission(KernelGlobals *kg, ShaderData *sd, f
 
 /* Indirect Lamp Emission */
 
-__device bool indirect_lamp_emission(KernelGlobals *kg, Ray *ray, int path_flag, float bsdf_pdf, float randt, float3 *emission)
+__device_noinline bool indirect_lamp_emission(KernelGlobals *kg, Ray *ray, int path_flag, float bsdf_pdf, float randt, float3 *emission)
 {
 	LightSample ls;
 	int lamp = lamp_light_eval_sample(kg, randt);
@@ -200,7 +200,7 @@ __device bool indirect_lamp_emission(KernelGlobals *kg, Ray *ray, int path_flag,
 
 /* Indirect Background */
 
-__device float3 indirect_background(KernelGlobals *kg, Ray *ray, int path_flag, float bsdf_pdf)
+__device_noinline float3 indirect_background(KernelGlobals *kg, Ray *ray, int path_flag, float bsdf_pdf)
 {
 #ifdef __BACKGROUND__
 	/* evaluate background closure */
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 10a32226f17..f6fbd3599ad 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -454,7 +454,7 @@ __device void triangle_light_sample(KernelGlobals *kg, int prim, int object,
 	ls->lamp = ~0;
 	ls->shader |= SHADER_USE_MIS;
 	ls->t = 0.0f;
-	ls->type = LIGHT_AREA;
+	ls->type = LIGHT_TRIANGLE;
 	ls->eval_fac = 1.0f;
 
 	object_transform_light_sample(kg, ls, object, time);
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index b7d87427bd3..df86b352697 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -53,7 +53,7 @@ __device_noinline void shader_setup_object_transforms(KernelGlobals *kg, ShaderD
 }
 #endif
 
-__device void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
+__device_noinline void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
 	const Intersection *isect, const Ray *ray)
 {
 #ifdef __INSTANCING__
@@ -160,7 +160,7 @@ __device void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
 
 /* ShaderData setup from position sampled on mesh */
 
-__device void shader_setup_from_sample(KernelGlobals *kg, ShaderData *sd,
+__device_noinline void shader_setup_from_sample(KernelGlobals *kg, ShaderData *sd,
 	const float3 P, const float3 Ng, const float3 I,
 	int shader, int object, int prim, float u, float v, float t, float time, int segment = ~0)
 {
@@ -283,7 +283,7 @@ __device void shader_setup_from_sample(KernelGlobals *kg, ShaderData *sd,
 
 /* ShaderData setup for displacement */
 
-__device void shader_setup_from_displace(KernelGlobals *kg, ShaderData *sd,
+__device_noinline void shader_setup_from_displace(KernelGlobals *kg, ShaderData *sd,
 	int object, int prim, float u, float v)
 {
 	/* Note: no OSLShader::init call here, this is done in shader_setup_from_sample! */
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 52bdddc72fb..773ea0c43fb 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -299,6 +299,7 @@ typedef enum LightType {
 	LIGHT_AREA,
 	LIGHT_AO,
 	LIGHT_SPOT,
+	LIGHT_TRIANGLE,
 	LIGHT_STRAND
 } LightType;
 
diff --git a/intern/cycles/render/osl.h b/intern/cycles/render/osl.h
index 3c599caa893..d17f4c81362 100644
--- a/intern/cycles/render/osl.h
+++ b/intern/cycles/render/osl.h
@@ -36,7 +36,7 @@ class Device;
 class DeviceScene;
 class ImageManager;
 class OSLRenderServices;
-class OSLGlobals;
+struct OSLGlobals;
 class Scene;
 class ShaderGraph;
 class ShaderNode;