Fix #35684: cycles unable to use full 6GB of memory on NVidia Titan GPU. We now

use arrays instead of textures for general storage on this card (image textures
are still stored as texture). Textures were found to be faster on older cards,
but the limits on 1D texture size have not increased along with the memory size,
which meant that the full 6 GB could not be used.

The performance actually seems to be slightly better with arrays in some tests
on Titan. For older cards there seems to be a bit of a mix, some are better and
others not. We may change those to use arrays too, but more testing is needed,
only Titan and Tesla K20 (sm_35) is changed for now.

The fact that arrays are faster is a bit surprising, as others found textures
to be faster on Kepler. However even if they were, the memory limitation is
more important to solve anyway.
https://research.nvidia.com/publication/understanding-efficiency-ray-traversal-gpus-kepler-and-fermi-addendum

intern/cycles/kernel/kernel_compat_cuda.h

@@ -57,7 +57,18 @@ typedef texture<uchar4, 2, cudaReadModeNormalizedFloat> texture_image_uchar4;
 
 /* Macros to handle different memory storage on different devices */
 
+/* In order to use full 6GB of memory on Titan cards, use arrays instead
+ * of textures. On earlier cards this seems slower, but on Titan it is
+ * actually slightly faster in tests. */
+#if __CUDA_ARCH__ < 350
+#define __KERNEL_CUDA_TEX_STORAGE__
+#endif
+
+#ifdef __KERNEL_CUDA_TEX_STORAGE__
 #define kernel_tex_fetch(t, index) tex1Dfetch(t, index)
+#else
+#define kernel_tex_fetch(t, index) t[(index)]
+#endif
 #define kernel_tex_image_interp(t, x, y) tex2D(t, x, y)
 
 #define kernel_data __data