[Support][X86] Change getHostNumPhsicalCores() to return number of physical cores enabled by affinity

[MaskRay]

Summary

Fixes https://bugs.llvm.org/show_bug.cgi?id=45556

While here, make the x86-64 code available for x86-32.

The output has been available and stable since
https://git.kernel.org/torvalds/c/3dd9d514846cdca1dcef2e4fce666d85e199e844 (2005)

processor:
...
physical id:
siblings:
core id:

Don't check HAVE_SCHED_GETAFFINITY/HAVE_CPU_COUNT. The interface is
simply available in every libc which can build LLVM.

[aganea]

Check if this fixes PR45556; also if it works with -DLLVM_ENABLE_THREADS=OFF (it should, but who knows).

llvm/lib/Support/Host.cpp

1293

cpu_set_t Affinity; ?

1306–1307

Since you're here, it seems really overkill to use a Set to represent a bitfield (also, this will allocate if processors > 32). Can we have cpu_set_t Enabled here instead, and use CPU_COUNT(Enabled) at the end?

1321

CPU_SET(CurProcessor, Enabled); as suggested above?

[MaskRay]

Rename Set to Affinity
Add comments.
Mention https://bugs.llvm.org/show_bug.cgi?id=45556

llvm/lib/Support/Host.cpp

1306–1307

The idea is that different processor ids may share the same (physical id, core id) pair, so we need to count the number of unique (physical id, core id) pair.

We arbitrarily define it as: if one processor id is allowed by the CPU affinity mask, consider its (physical id, core id) available.

[aganea]

LGTM.

llvm/lib/Support/Host.cpp

1300

One more (almost mandatory) allocation here. :-(

6-core:
$ cat /proc/cpuinfo | wc -l
312

18-core:
$ cat /proc/cpuinfo | wc -l
936

I know this function is only called once, but all these allocations pile up on the startup time. Can't help seeing this. You don't have to fix it.

1306–1307

Ok. I guess we could read "siblings" and do CPU_SET(CurPhysicalId * SiblingsId + CurCoreId, &Enabled); since it's a flat bitfield anyway. Unless we expect asymmetric configurations, with different core count on 2+ CPU sockets, I'm not even sure if that's possible.
Let's leave this for later then if you think it's too much trouble.

[MaskRay]

Address comments

[aganea]

Thanks for doing this, still looking good.

So to be complete, only Darwin remains, if ever. @dexonsmith

llvm/lib/Support/Host.cpp

switch (hostInfo.cpu_subtype) {
     }
 
   return "generic";
 }
 #else
 StringRef sys::getHostCPUName() { return "generic"; }
 #endif
 
-#if defined(__linux__) && defined(__x86_64__)
+#if defined(__linux__) && (defined(__i386__) || defined(__x86_64__))
 // On Linux, the number of physical cores can be computed from /proc/cpuinfo,
 // using the number of unique physical/core id pairs. The following
 // implementation reads the /proc/cpuinfo format on an x86_64 system.
 int computeHostNumPhysicalCores() {
+  // Count physical id/core id pairs with at least one processor id enabled by
+  // the CPU affinity mask.
+  cpu_set_t Affinity;
+  if (sched_getaffinity(0, sizeof(Affinity), &Affinity) != 0)
+    return -1;
+
   // Read /proc/cpuinfo as a stream (until EOF reached). It cannot be
   // mmapped because it appears to have 0 size.
   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> Text =
       llvm::MemoryBuffer::getFileAsStream("/proc/cpuinfo");
   if (std::error_code EC = Text.getError()) {
     llvm::errs() << "Can't read "
                  << "/proc/cpuinfo: " << EC.message() << "\n";
     return -1;
   }
   SmallVector<StringRef, 8> strs;
   (*Text)->getBuffer().split(strs, "\n", /*MaxSplit=*/-1,
                              /*KeepEmpty=*/false);
+  int CurProcessor = -1;
   int CurPhysicalId = -1;
   int CurCoreId = -1;
   SmallSet<std::pair<int, int>, 32> UniqueItems;
-  for (auto &Line : strs) {
-    Line = Line.trim();
-    if (!Line.startswith("physical id") && !Line.startswith("core id"))
-      continue;
+  for (StringRef Line : strs) {
     std::pair<StringRef, StringRef> Data = Line.split(':');
     auto Name = Data.first.trim();
     auto Val = Data.second.trim();
-    if (Name == "physical id") {
-      assert(CurPhysicalId == -1 &&
-             "Expected a core id before seeing another physical id");
+    if (Name == "processor")
+      Val.getAsInteger(10, CurProcessor);
+    else if (Name == "physical id")
       Val.getAsInteger(10, CurPhysicalId);
-    }
-    if (Name == "core id") {
-      assert(CurCoreId == -1 &&
+    else if (Name == "core id") {
+      assert((CurProcessor != -1 && CurPhysicalId != -1 && CurCoreId == -1) &&
              "Expected a physical id before seeing another core id");
       Val.getAsInteger(10, CurCoreId);
-    }
-    if (CurPhysicalId != -1 && CurCoreId != -1) {
+      // The processor id corresponds to an index into cpu_set_t.
+      if (CPU_ISSET(CurProcessor, &Affinity))
         UniqueItems.insert(std::make_pair(CurPhysicalId, CurCoreId));
+      CurProcessor = -1;
       CurPhysicalId = -1;
       CurCoreId = -1;
     }
   }
   return UniqueItems.size();
 }
 #elif defined(__APPLE__) && defined(__x86_64__)
 #include <sys/param.h>