[RegisterBankInfo] Make addRegBankCoverage more capable to ease

targeting jobs.
Now, addRegBankCoverage also adds the subreg-classes not just the
sub-classes of the given register class.

llvm-svn: 265469

1 files changed, 67 insertions, 4 deletions

diff --git a/llvm/lib/CodeGen/GlobalISel/RegisterBankInfo.cpp b/llvm/lib/CodeGen/GlobalISel/RegisterBankInfo.cpp
index 8c86491ec8a..80c0b0380be 100644
--- a/llvm/lib/CodeGen/GlobalISel/RegisterBankInfo.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/RegisterBankInfo.cpp
@@ -76,15 +76,23 @@ void RegisterBankInfo::addRegBankCoverage(unsigned ID,
     const uint32_t *SubClassMask = CurRC.getSubClassMask();
     // The subclasses mask is broken down into chunks of uint32_t, but it still
     // represents all register classes.
+    bool First = true;
     for (unsigned Base = 0; Base < NbOfRegClasses; Base += 32) {
       unsigned Idx = Base;
       for (uint32_t Mask = *SubClassMask++; Mask; Mask >>= 1, ++Idx) {
         unsigned Offset = countTrailingZeros(Mask);
         unsigned SubRCId = Idx + Offset;
-        if (!Covered.test(SubRCId))
+        if (!Covered.test(SubRCId)) {
+          if (First)
+            DEBUG(dbgs() << "  Enqueue sub-class: ");
+          DEBUG(dbgs() << TRI.getRegClassName(TRI.getRegClass(SubRCId))
+                       << ", ");
           WorkList.push_back(SubRCId);
-        // Remember that we saw the sub class.
-        Covered.set(SubRCId);
+          // Remember that we saw the sub class.
+          Covered.set(SubRCId);
+          First = false;
+        }
+
         // Move the cursor to the next sub class.
         // I.e., eat up the zeros then move to the next bit.
         // This last part is done as part of the loop increment.
@@ -96,6 +104,61 @@ void RegisterBankInfo::addRegBankCoverage(unsigned ID,
         Idx += Offset;
       }
     }
-
+    if (!First)
+      DEBUG(dbgs() << '\n');
+
+    // Push also all the register classes that can be accessed via a
+    // subreg index, i.e., its subreg-class (which is different than
+    // its subclass).
+    //
+    // Note: It would probably be faster to go the other way around
+    // and have this method add only super classes, since this
+    // information is available in a more efficient way. However, it
+    // feels less natural for the client of this APIs plus we will
+    // TableGen the whole bitset at some point, so compile time for
+    // the initialization is not very important.
+    First = true;
+    for (unsigned SubRCId = 0; SubRCId < NbOfRegClasses; ++SubRCId) {
+      if (Covered.test(SubRCId))
+        continue;
+      bool Pushed = false;
+      const TargetRegisterClass *SubRC = TRI.getRegClass(SubRCId);
+      for (SuperRegClassIterator SuperRCIt(SubRC, &TRI); SuperRCIt.isValid();
+           ++SuperRCIt) {
+        if (Pushed)
+          break;
+        const uint32_t *SuperRCMask = SuperRCIt.getMask();
+        for (unsigned Base = 0; Base < NbOfRegClasses; Base += 32) {
+          unsigned Idx = Base;
+          for (uint32_t Mask = *SuperRCMask++; Mask; Mask >>= 1, ++Idx) {
+            unsigned Offset = countTrailingZeros(Mask);
+            unsigned SuperRCId = Idx + Offset;
+            if (SuperRCId == RCId) {
+              if (First)
+                DEBUG(dbgs() << "  Enqueue subreg-class: ");
+              DEBUG(dbgs() << TRI.getRegClassName(SubRC) << ", ");
+              WorkList.push_back(SubRCId);
+              // Remember that we saw the sub class.
+              Covered.set(SubRCId);
+              Pushed = true;
+              First = false;
+              break;
+            }
+
+            // Move the cursor to the next sub class.
+            // I.e., eat up the zeros then move to the next bit.
+            // This last part is done as part of the loop increment.
+
+            // By construction, Offset must be less than 32.
+            // Otherwise, than means Mask was zero. I.e., no UB.
+            Mask >>= Offset;
+            // Remember that we shifted the base offset.
+            Idx += Offset;
+          }
+        }
+      }
+    }
+    if (!First)
+      DEBUG(dbgs() << '\n');
   } while (!WorkList.empty());
 }