FlaxEngine/Source/ThirdParty/NvCloth/ps/windows/PsWindowsIntrinsics.h

// This code contains NVIDIA Confidential Information and is disclosed to you
// under a form of NVIDIA software license agreement provided separately to you.
//
// Notice
// NVIDIA Corporation and its licensors retain all intellectual property and
// proprietary rights in and to this software and related documentation and
// any modifications thereto. Any use, reproduction, disclosure, or
// distribution of this software and related documentation without an express
// license agreement from NVIDIA Corporation is strictly prohibited.
//
// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
//
// Information and code furnished is believed to be accurate and reliable.
// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
// information or for any infringement of patents or other rights of third parties that may
// result from its use. No license is granted by implication or otherwise under any patent
// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
// This code supersedes and replaces all information previously supplied.
// NVIDIA Corporation products are not authorized for use as critical
// components in life support devices or systems without express written approval of
// NVIDIA Corporation.
//
// Copyright (c) 2008-2020 NVIDIA Corporation. All rights reserved.
// Copyright (c) 2004-2008 AGEIA Technologies, Inc. All rights reserved.
// Copyright (c) 2001-2004 NovodeX AG. All rights reserved.

#ifndef PSFOUNDATION_PSWINDOWSINTRINSICS_H
#define PSFOUNDATION_PSWINDOWSINTRINSICS_H

#include "Ps.h"

// this file is for internal intrinsics - that is, intrinsics that are used in
// cross platform code but do not appear in the API

#if !PX_WINDOWS_FAMILY
#error "This file should only be included by Windows builds!!"
#endif

#pragma warning(push)
//'symbol' is not defined as a preprocessor macro, replacing with '0' for 'directives'
#pragma warning(disable : 4668)
#if PX_VC == 10
#pragma warning(disable : 4987) // nonstandard extension used: 'throw (...)'
#endif
#include <intrin.h>
#pragma warning(pop)

#pragma warning(push)
#pragma warning(disable : 4985) // 'symbol name': attributes not present on previous declaration
#include <math.h>
#pragma warning(pop)

#include <float.h>
#include <mmintrin.h>

#pragma intrinsic(_BitScanForward)
#pragma intrinsic(_BitScanReverse)

/** \brief NVidia namespace */
namespace nv
{
/** \brief nvcloth namespace */
namespace cloth
{
namespace ps
{

/*
* Implements a memory barrier
*/
PX_FORCE_INLINE void PxMemoryBarrier()
{
	_ReadWriteBarrier();
	/* long Barrier;
	__asm {
	    xchg Barrier, eax
	}*/
}

/*!
Returns the index of the highest set bit. Not valid for zero arg.
*/
PX_FORCE_INLINE uint32_t PxHighestSetBitUnsafe(uint32_t v)
{
	unsigned long retval;
	_BitScanReverse(&retval, v);
	return retval;
}

/*!
Returns the index of the highest set bit. Undefined for zero arg.
*/
PX_FORCE_INLINE uint32_t PxLowestSetBitUnsafe(uint32_t v)
{
	unsigned long retval;
	_BitScanForward(&retval, v);
	return retval;
}

/*!
Returns the number of leading zeros in v. Returns 32 for v=0.
*/
PX_FORCE_INLINE uint32_t PxCountLeadingZeros(uint32_t v)
{
	if(v)
	{
		unsigned long bsr = (unsigned long)-1;
		_BitScanReverse(&bsr, v);
		return 31 - bsr;
	}
	else
		return 32;
}

/*!
Prefetch aligned cache size around \c ptr+offset.
*/
#if !PX_ARM && !PX_A64
PX_FORCE_INLINE void PxPrefetchLine(const void* ptr, uint32_t offset = 0)
{
	// cache line on X86/X64 is 64-bytes so a 128-byte prefetch would require 2 prefetches.
	// However, we can only dispatch a limited number of prefetch instructions so we opt to prefetch just 1 cache line
	/*_mm_prefetch(((const char*)ptr + offset), _MM_HINT_T0);*/
	// We get slightly better performance prefetching to non-temporal addresses instead of all cache levels
	_mm_prefetch(((const char*)ptr + offset), _MM_HINT_NTA);
}
#else
PX_FORCE_INLINE void PxPrefetchLine(const void* ptr, uint32_t offset = 0)
{
	// arm does have 32b cache line size
	__prefetch(((const char*)ptr + offset));
}
#endif

/*!
Prefetch \c count bytes starting at \c ptr.
*/
#if !PX_ARM
PX_FORCE_INLINE void PxPrefetch(const void* ptr, uint32_t count = 1)
{
	const char* cp = (char*)ptr;
	uint64_t p = size_t(ptr);
	uint64_t startLine = p >> 6, endLine = (p + count - 1) >> 6;
	uint64_t lines = endLine - startLine + 1;
	do
	{
		PxPrefetchLine(cp);
		cp += 64;
	} while(--lines);
}
#else
PX_FORCE_INLINE void PxPrefetch(const void* ptr, uint32_t count = 1)
{
	const char* cp = (char*)ptr;
	uint32_t p = size_t(ptr);
	uint32_t startLine = p >> 5, endLine = (p + count - 1) >> 5;
	uint32_t lines = endLine - startLine + 1;
	do
	{
		PxPrefetchLine(cp);
		cp += 32;
	} while(--lines);
}
#endif

} // namespace ps
} // namespace cloth
} // namespace nv

#endif // #ifndef PSFOUNDATION_PSWINDOWSINTRINSICS_H