Optional support for faster CRC32C checksum through Intel IPP

Grid/util/Sha.h

@@ -28,17 +28,31 @@
 extern "C" {
 #include <openssl/sha.h>
 }
+#ifdef USE_IPP
+#include "ipp.h"
+#endif
 
 #pragma once
 
 class GridChecksum
 {
 public:
-  static inline uint32_t crc32(const void *data,size_t bytes)
+  static inline uint32_t crc32(const void *data, size_t bytes)
   {
     return ::crc32(0L,(unsigned char *)data,bytes);
   }
 
+#ifdef USE_IPP
+  static inline uint32_t crc32c(const void* data, size_t bytes)
+  {
+      uint32_t crc32c = ~(uint32_t)0;
+      ippsCRC32C_8u(reinterpret_cast<const unsigned char *>(data), bytes, &crc32c);
+      ippsSwapBytes_32u_I(&crc32c, 1);
+  
+      return ~crc32c;
+  }
+#endif
+
   template <typename T>
   static inline std::string sha256_string(const std::vector<T> &hash)
   {

Hadrons/DiskVector.hpp

@@ -176,7 +176,11 @@ private:
         f.read(reinterpret_cast<char *>(obj.data()), matSize);
         tRead += usecond();
         tHash  = -usecond();
+#ifdef USE_IPP
+        check  = GridChecksum::crc32c(obj.data(), matSize);
+#else
         check  = GridChecksum::crc32(obj.data(), matSize);
+#endif
         tHash += usecond();
         DV_DEBUG_MSG(this, "Eigen read " << tRead/1.0e6 << " sec " << matSize/tRead*1.0e6/1024/1024 << " MB/s");
         DV_DEBUG_MSG(this, "Eigen crc32 " << std::hex << check << std::dec 
@@ -199,7 +203,11 @@ private:
         nCol    = obj.cols();
         matSize = nRow*nCol*sizeof(T);
         tHash   = -usecond();
+#ifdef USE_IPP
+        crc     = GridChecksum::crc32c(obj.data(), matSize);
+#else
         crc     = GridChecksum::crc32(obj.data(), matSize);
+#endif
         tHash  += usecond();
         f.write(reinterpret_cast<char *>(&crc), sizeof(crc));
         f.write(reinterpret_cast<char *>(&nRow), sizeof(nRow));

configure.ac

@@ -123,10 +123,13 @@ case ${ac_SFW_FP16} in
       AC_MSG_ERROR(["SFW FP16 option not supported ${ac_SFW_FP16}"]);;
 esac
 
-############### MKL
+############### Intel libraries
 AC_ARG_ENABLE([mkl],
     [AC_HELP_STRING([--enable-mkl=yes|no|prefix], [enable Intel MKL for LAPACK & FFTW])],
     [ac_MKL=${enable_mkl}], [ac_MKL=no])
+AC_ARG_ENABLE([ipp],
+    [AC_HELP_STRING([--enable-ipp=yes|no|prefix], [enable Intel IPP for fast CRC32C])],
+    [ac_IPP=${enable_mkl}], [ac_IPP=no])
 
 case ${ac_MKL} in
     no)
@@ -139,6 +142,17 @@ case ${ac_MKL} in
         AC_DEFINE([USE_MKL], [1], [Define to 1 if you use the Intel MKL]);;
 esac
 
+case ${ac_IPP} in
+    no)
+        ;;
+    yes)
+        AC_DEFINE([USE_IPP], [1], [Define to 1 if you use the Intel IPP]);;
+    *)
+        AM_CXXFLAGS="-I$ac_IPP/include $AM_CXXFLAGS"
+        AM_LDFLAGS="-L$ac_IPP/lib $AM_LDFLAGS"
+        AC_DEFINE([USE_IPP], [1], [Define to 1 if you use the Intel IPP]);;
+esac
+
 ############### HDF5
 AC_ARG_WITH([hdf5],
     [AS_HELP_STRING([--with-hdf5=prefix],
@@ -170,7 +184,13 @@ AC_CHECK_FUNCS([gettimeofday])
 
 if test "${ac_MKL}x" != "nox"; then
     AC_SEARCH_LIBS([mkl_set_interface_layer], [mkl_rt], [],
-                   [AC_MSG_ERROR("MKL enabled but library not found")])
+                   [AC_MSG_ERROR("Intel MKL enabled but library not found")])
+fi
+
+if test "${ac_IPP}x" != "nox"; then
+     AC_SEARCH_LIBS([ippsCRC32C_8u], [ippdc], 
+                   [LIBS="${LIBS} -lippdc -lippvm -lipps -lippcore"],
+                   [AC_MSG_ERROR("Intel IPP enabled but library not found")])
 fi
 
 AC_SEARCH_LIBS([__gmpf_init], [gmp],