Create README.md

README.md

@@ -0,0 +1,65 @@
+To get full use of the repo, you need a modern processor which has AVX512 instructions. 
+If your processor only has  AVX2, you need to change target instruction sets in the projects to AVX2, and don’t generate AVX512
+In the projects because your machine wont run them.
+
+The projects build with clang, IC2022 and VS2019.
+Select x64 and solution configuration  for IC2022, release, debug and clang
+
+Getting started shows some example use cases for vectors filters and views, together with an experimental
+vectorised forward AAD.
+
+Accumulate example shows some of the use cases given in the cppCon2022 talk.
+Additionally it gives an example of error correction in Khan accumulation
+
+The example build an run with VS2019, clang and intel compilers. The target instruction set 
+generated by the framework can be changed by changing the namespace. These are double and float
+types  VecDb is pair of doubles. Uncomment the namespace ande build the example.
+
+//using namespace DRC::VecDb;
+//using namespace DRC::VecD2D;  //sse2   double
+using namespace DRC::VecD4D;	//avx2   double
+//using namespace DRC::VecF8F;	// avx2  float
+//using namespace DRC::VecD8D;  //avx512 double
+//using namespace DRC::VecF16F; //avx512   float
+
+For a machine supporting AVX512 ensure all the VC projects are set to use  enhanced instruction set must be set 
+ConfigurationProperties C++/Instruction Set /Enable Enhanced Instruction Set to ARCh:AVX512
+If your machine doesnt support this, reduce to AVX2 or SSE2, and dont select a namespace in the code requiring more advanced instruction
+sets.
+
+Uncomment one of the Using namespace lines  select the instruction set that you wish to run  
+Those ending in F have float type as underlying, those ending with D have a double.
+
+The project is set to compile using the AVX512  enhanced instruction set. The namespace selection 
+choses the type of the intrinsics that are used to instantiate lambdas.
+
+If your hardware does not support AVX512 chose the next level down AVX2 and avoid using namespaces 
+DRC::VecD8D or DRC::VecF16F which will cause generation of code with instructions that your computer doesn't support. 
+
+check device manager/processor to determine what processor you have and check against web site 
+https://ark.intel.com/content/www/us/en/ark/products/123550/intel-xeon-silver-4114-processor-13-75m-cache-2-20-ghz.html
+or
+https://www.intel.com/content/www/us/en/products/details/processors/xeon/scalable.html
+
+
+The getting started project shows the useage of vectors lambdas and filters
+
+The accumulateExample builds performance examples covered in the cppCon2022 talk. 
+They give the user the chance to change between ICC,clang and VS2019 builds but changing the
+instruction set used via the using declaration.
+
+The inverseCumNormalExample  gives the performance example shown in cppCon2022, although there might be some slight
+perfrormance regression on one or two of the examples.   Its instructive to run the examples after building with the
+different compilers and  chosing different instruction sets for the Lambdas (via namespace).
+
+The AVX512Dance function runs a routine which finds the max value in in array, using AVX2 amnd AVX512. By monitoring the
+power useage using something like openhardware monitor its possible to see that using the AVX512 instructions, use less
+energy to do the compute than teh AVX2 ( on this silver4114 xeon).
+
+VectorTest is a selection of tests using googletest.  
+The main library is  Vectorisation.  This refrence a local copy of the VCL2 library.  It has a slight change to enable
+VCL2 to be used with the intel IC2022 compiler.
+
+
+
+