Pierre-Terdiman

See https://github.com/jbikker/tinybvh
PhysX is a general purpose library not focused on raycasts and the best setup for raycast performance may not be obvious.
Here is the code I used in the tiny BVH benchmark. Results: https://x.com/PierreTerdiman/status/1854629430246748455
Adapted from https://github.com/NVIDIA-Omniverse/PhysX/tree/main/physx/snippets/snippetpathtracing

For building:

#ifdef PHYSX_BUILD
	const bool quantized = false; // quantized trees use less memory but are usually slower for raycasts
	const bool useSAH = false;    // SAH build is much slower but can give slightly better perf. Not the PhysX default.

Pierre-Terdiman

#include <ctype.h>
#include <chrono>
#include <iostream>
#include "PxPhysicsAPI.h"
#include "../snippetcommon/SnippetPrint.h"
#include "../snippetcommon/SnippetPVD.h"
#include "../snippetutils/SnippetUtils.h"
#ifdef RENDER_SNIPPET
	#include "../snippetrender/SnippetRender.h"
#endif

Pierre-Terdiman

#ifdef SAFE_VERSION
	#define SIMD_OVERLAP_TEST	(_mm_movemask_ps(_mm_cmpngt_ps(b, _mm_load_ps(box)))==15)
#else
	#define SIMD_OVERLAP_TEST	(_mm_movemask_ps(_mm_cmpnle_ps(_mm_load_ps(box), b))==12)
#endif


bool Meshmerizer::CompleteBoxPruning(udword nb, const AABB* list, Container& pairs)
{
	// Checkings