Cycles: Fix/workaround for toggling world MIS causing CUDA to fail

Seems it's just another issue with the compiler, worked around by explicitly
telling not to inline some function.

In theory we can unify this with CPU, but we're quite close to the release
so better be safe than sorry.

1 files changed, 19 insertions, 2 deletions

diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 846ec801f77..76fa754b5fa 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -37,7 +37,16 @@ typedef struct LightSample {
 
 #ifdef __BACKGROUND_MIS__
 
-ccl_device float3 background_light_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
+/* TODO(sergey): In theory it should be all fine to use noinline for all
+ * devices, but we're so close to the release so better not screw things
+ * up for CPU at least.
+ */
+#ifdef __KERNEL_GPU__
+ccl_device_noinline
+#else
+ccl_device
+#endif
+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
 	 * function value as X and the actual CDF as Y.  The last entry's function
@@ -113,7 +122,15 @@ ccl_device float3 background_light_sample(KernelGlobals *kg, float randu, float
 	return -equirectangular_to_direction(u, v);
 }
 
-ccl_device float background_light_pdf(KernelGlobals *kg, float3 direction)
+/* TODO(sergey): Same as above, after the release we should consider using
+ * 'noinline' for all devices.
+ */
+#ifdef __KERNEL_GPU__
+ccl_device_noinline
+#else
+ccl_device
+#endif
+float background_light_pdf(KernelGlobals *kg, float3 direction)
 {
 	float2 uv = direction_to_equirectangular(direction);
 	int res = kernel_data.integrator.pdf_background_res;