Fixed x2APIC discovery for 256-processor architectures

[AndreyChurbanov]

Summary

CPUID returns values in bits 0-15 of ebx those were mistakenly masked by 0xFF reading only bits 0-7. Thus on 256-processor architecture we got value 0 that is wrong. Masking by 0xFFFF fixes the problem.

[tlwilmar]

LGTM

openmp/trunk/runtime/src/kmp_affinity.cpp

for (level = 0;; level++) {
         int kind = (buf.ecx >> 8) & 0xff;
         if (kind == 1) {
             //
             // SMT level
             //
             threadLevel = level;
             coreLevel = -1;
             pkgLevel = -1;
-            __kmp_nThreadsPerCore = buf.ebx & 0xff;
+            __kmp_nThreadsPerCore = buf.ebx & 0xffff;
             if (__kmp_nThreadsPerCore == 0) {
                 *msg_id = kmp_i18n_str_InvalidCpuidInfo;
                 return -1;
             }
         }
         else if (kind == 2) {
             //
             // core level
             //
             coreLevel = level;
             pkgLevel = -1;
-            nCoresPerPkg = buf.ebx & 0xff;
+            nCoresPerPkg = buf.ebx & 0xffff;
             if (nCoresPerPkg == 0) {
                 *msg_id = kmp_i18n_str_InvalidCpuidInfo;
                 return -1;
             }
         }
         else {
             if (level <= 0) {
                 *msg_id = kmp_i18n_str_InvalidCpuidInfo;
                 return -1;
             }
             if (pkgLevel >= 0) {
                 continue;
             }
             pkgLevel = level;
-            nPackages = buf.ebx & 0xff;
+            nPackages = buf.ebx & 0xffff;
             if (nPackages == 0) {
                 *msg_id = kmp_i18n_str_InvalidCpuidInfo;
                 return -1;
             }
         }
     }
     int depth = level;