[SystemZ/ZOS] Add support for getHostNumPhysicalCores()

[Kai]

Summary

The information about the CPs online in an LPAR is stored in a control block.
The code navigates to the right control block and returns the number.

[abhina.sreeskantharajan]

llvm/unittests/Support/Host.cpp

39–40

Minor nit: grammar

[hubert.reinterpretcast]

llvm/unittests/Support/Host.cpp

39–40

The x86_64 part of the sentence is meant to be distributed across "Linux" and "Darwin" and not to z/OS. With the new formulation, x86_64 needs to be added to before "Darwin" to retain the intended meaning.

43

Minor nit: Aside for Windows, where the expression here is not sensitive to the platform architecture, can we structure this to consistently group by architecture? Alternatively, consistently group by OS.

[MaskRay]

llvm/lib/Support/Host.cpp

1334

If z/OS supports something similar to taskset -c 0-3 $program, make sure the function returns a number less than or equal to 4.

[Kai]

llvm/lib/Support/Host.cpp

1334

Thanks for the hint. This kind of command is not supported on z/OS.
The reason is the fault tolerance available in z/OS: if a defect in a CP is detected, then this CP can go offline and a spare CP can continue to run the code exactly at the point where the former CP went offline. This does not work if the process is tied to a certain set of CPs, and therefore it's not supported.

[Kai]

• Sorted condition according to OS
• Updated/fixed comment

[Kai]

llvm/unittests/Support/Host.cpp

39–40

I hope the new order is better.
As alternative I also thought about removing this part of the comment, as it only rephrases the condition without giving new information.

[hubert.reinterpretcast]

LGTM with minor comment.

llvm/unittests/Support/Host.cpp

39–40

I think the comment doesn't need to indicate the list (note that the duplication led to Windows being in one list and not ther other). The comment does help in explaining what the function does though.

[hubert.reinterpretcast]

llvm/lib/Support/Host.cpp

1334

@MaskRay, I think the test you propose imposes an additional restriction that I am not sure the existing implementations of the function adhere to. Namely, I believe the test that you proposed indicates that the intent is to return a number that is one greater than the highest processor index number on the system. I think this might differ from the count of available processors. My understanding is that the function is meant to return the latter.

[aganea]

llvm/lib/Support/Host.cpp

1334

@hubert.reinterpretcast Actually both the Linux & Windows implementations return the restricted set of CPUs/processors that are usable within the process (ie. CPUs represented by the affinity mask).
https://github.com/llvm/llvm-project/blob/62a933b72c5b060bcb2c7332d05082f002d6c65a/llvm/lib/Support/Host.cpp#L1234

[hubert.reinterpretcast]

llvm/lib/Support/Host.cpp

1334

The restricted set is what I meant by "available". In other words, I understand that the current implementations returns what you describe.

My comment was on the relationship between the count and the index. If the first usable index is higher than the count, then the command MaskRay proposed would fail.

[aganea]

llvm/lib/Support/Host.cpp

1334

I think @MaskRay was only suggesting to take the affinity mask into account. On Linux you could say taskset -c 6-8,10,15 and in that case computeHostNumPhysicalCores() would return 5. Or maybe I'm misunderstanding to which count and index are you referring to?

[hubert.reinterpretcast]

llvm/lib/Support/Host.cpp

1334

Your example is more clear (not fixed to 0 on one end). If that's what @MaskRay meant, then I agree.

[aganea]

llvm/lib/Support/Host.cpp

1334

Sorry wrong example, that should return 4, as CPU 6 & 7 should be on the same core on a SMT system, but you got the idea.

[hubert.reinterpretcast]

llvm/lib/Support/Host.cpp

1334

@aganea, thanks. This last was something I missed. The "physical" part of the function name is actually significant.

[Kai]

• Remove list of systems from comment.

llvm/lib/Support/Host.cpp

if (count < 1) {
     nm[0] = CTL_HW;
     nm[1] = HW_AVAILCPU;
     sysctl(nm, 2, &count, &len, NULL, 0);
     if (count < 1)
       return -1;
   }
   return count;
 }
+#elif defined(__MVS__)
+int computeHostNumPhysicalCores() {
+  enum {
+    // Byte offset of the pointer to the Communications Vector Table (CVT) in
+    // the Prefixed Save Area (PSA). The table entry is a 31-bit pointer and
+    // will be zero-extended to uintptr_t.
+    FLCCVT = 16,
+    // Byte offset of the pointer to the Common System Data Area (CSD) in the
+    // CVT. The table entry is a 31-bit pointer and will be zero-extended to
+    // uintptr_t.
+    CVTCSD = 660,
+    // Byte offset to the number of live CPs in the LPAR, stored as a signed
+    // 32-bit value in the table.
+    CSD_NUMBER_ONLINE_STANDARD_CPS = 264,
+  };
+  char *PSA = 0;
+  char *CVT = reinterpret_cast<char *>(
+      static_cast<uintptr_t>(reinterpret_cast<unsigned int &>(PSA[FLCCVT])));
+  char *CSD = reinterpret_cast<char *>(
+      static_cast<uintptr_t>(reinterpret_cast<unsigned int &>(CVT[CVTCSD])));
+  return reinterpret_cast<int &>(CSD[CSD_NUMBER_ONLINE_STANDARD_CPS]);
+}
 #elif defined(_WIN32) && LLVM_ENABLE_THREADS != 0
 // Defined in llvm/lib/Support/Windows/Threading.inc
 int computeHostNumPhysicalCores();
 #else
 // On other systems, return -1 to indicate unknown.
 static int computeHostNumPhysicalCores() { return -1; }
 #endif
 

llvm/unittests/Support/Host.cpp

 using namespace llvm;
 
 class HostTest : public testing::Test {
   Triple Host;
 
 protected:
   bool isSupportedArchAndOS() {
     // Initially this is only testing detection of the number of
-    // physical cores, which is currently only supported/tested for
-    // x86_64 Linux and Darwin.
+    // physical cores, which is currently only supported/tested on
+    // some systems.
     return (Host.isOSWindows() && llvm_is_multithreaded()) ||
-           (Host.isX86() &&
-            (Host.isOSDarwin() || Host.getOS() == Triple::Linux)) ||
-           (Host.getOS() == Triple::Linux &&
-            (Host.isPPC64() || Host.isSystemZ()));
+           (Host.isX86() && (Host.isOSDarwin() || Host.isOSLinux())) ||
+           (Host.isPPC64() && Host.isOSLinux()) ||
+           (Host.isSystemZ() && (Host.isOSLinux() || Host.isOSzOS()));
   }
 
   HostTest() : Host(Triple::normalize(sys::getProcessTriple())) {}
 };
 
 TEST_F(HostTest, NumPhysicalCores) {
   int Num = sys::getHostNumPhysicalCores();