traverse.cl

// gpu-side code for ray traversal

struct Ray
{
	// data is defined here as 16-byte values to encourage the compilers
	// to fetch 16 bytes at a time: 12 (so, 8 + 4) will be slower.
	float4 O, D, rD; // 48 byte
	float4 hit; // 16 byte
};

// #define CWBVH_COMPRESSED_TRIS // sync with tiny_bvh.h
// #define BVH4_GPU_COMPRESSED_TRIS // sync with tiny_bvh.h

// BVH traversal stack size 
#define STACK_SIZE 32

// Low-level optimizations for specific platforms
#ifdef ISINTEL // Iris Xe, Arc, ..
// #define USE_VLOAD_VSTORE
#define SIMD_AABBTEST
#elif defined ISNVIDIA // 2080, 3080, 4080, ..
#define USE_VLOAD_VSTORE
// #define SIMD_AABBTEST
#elif defined ISAMD
#define USE_VLOAD_VSTORE
// #define SIMD_AABBTEST
#else // unkown GPU
// #define USE_VLOAD_VSTORE
#define SIMD_AABBTEST
#endif

// Includes for traversal kernel implementations:
#include "traverse_bvh2.cl"
#include "traverse_bvh4.cl"
#include "traverse_cwbvh.cl"