MaskedOcclusionCullingPrivate::SetResolution() uses the intrinsic _mm256_set1_ps() which in the Debug|x64 build in Visual Studio 2017 (cl.exe version 19.10.25019) compiles to the register to register version of vbroadcastss which is an AVX2 instruction (according to https://en.wikipedia.org/wiki/Advanced_Vector_Extensions). Locally adding a lambda to replace this with the memory to register version appears to fix this:

	auto _mm256_set1_ps = [](float f)
	{
		return _mm256_broadcast_ss(&f);
	};

This doesn't appear required for the Release|x64 build which heavily flattens SetResolution().

Thanks for the great sample code!

From b15fdcf11b4390a606a6a91178358237279f4321 Mon Sep 17 00:00:00 2001
From: Daniel Silver [email protected]
Date: Sat, 24 Mar 2018 18:58:56 -0500
Subject: [PATCH] Fixed a buffer over-run bug in the CPUID code on certain
processors.

MaskedOcclusionCulling/MaskedOcclusionCulling.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/MaskedOcclusionCulling/MaskedOcclusionCulling.cpp b/MaskedOcclusionCulling/MaskedOcclusionCulling.cpp
index d5d0a78..15afac0 100644
--- a/MaskedOcclusionCulling/MaskedOcclusionCulling.cpp
+++ b/MaskedOcclusionCulling/MaskedOcclusionCulling.cpp
@@ -54,7 +54,7 @@ static MaskedOcclusionCulling::Implementation DetectCPUFeatures(MaskedOcclusionC

 //cpuIdEx.resize(regs[0] - 0x80000000);
 size_t cpuIdExCount = regs[0] - 0x80000000;

• CpuInfo * cpuIdEx = (CpuInfo*)alignedAlloc( 64, sizeof( CpuInfo ) * cpuIdCount );

• CpuInfo * cpuIdEx = (CpuInfo*)alignedAlloc( 64, sizeof( CpuInfo ) * cpuIdExCount);

  for (size_t i = 0; i < cpuIdExCount; ++i)
  __cpuidex(cpuIdEx[i].regs, 0x80000000 + (int)i, 0);
  --
  2.16.1.windows.4

This line should read like below.

	float total = 0.0f;
	for(int i = 0; i < AVG_FRAMES; i++)
	{
		total += mFPSAvg[i];
	}

Also, consider adding a \t in the output string to avoid the FPS counter jumping around.

swprintf_s(&wcstring[0], CPUT_MAX_STRING_LENGTH, _L("FPS:%.1f \t(AVG:%.1f)"), fps, total);