[llvm] [DAG] Improve known bits of Zext/Sext loads with range metadata (PR #80829)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-selectiondag

Author: David Green (davemgreen)

<details>
<summary>Changes</summary>

This extends the known bits for extending loads which have range metadata, handling the range metadata on the original memory type, extending that to the correct BitWidth.

---
Full diff: https://github.com/llvm/llvm-project/pull/80829.diff


2 Files Affected:

- (modified) llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (+29-19) 
- (modified) llvm/test/CodeGen/AArch64/setcc_knownbits.ll (+1-3) 


``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 3c1343836187a..eec09219ad01a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3612,32 +3612,42 @@ KnownBits SelectionDAG::computeKnownBits(SDValue Op, const APInt &DemandedElts,
           }
         }
       }
-    } else if (ISD::isZEXTLoad(Op.getNode()) && Op.getResNo() == 0) {
-      // If this is a ZEXTLoad and we are looking at the loaded value.
-      EVT VT = LD->getMemoryVT();
-      unsigned MemBits = VT.getScalarSizeInBits();
-      Known.Zero.setBitsFrom(MemBits);
-    } else if (const MDNode *Ranges = LD->getRanges()) {
-      EVT VT = LD->getValueType(0);
-
-      // TODO: Handle for extending loads
-      if (LD->getExtensionType() == ISD::NON_EXTLOAD) {
+    } else if (Op.getResNo() == 0) {
+      KnownBits Known0(!LD->getMemoryVT().isScalableVT()
+                           ? LD->getMemoryVT().getSizeInBits()
+                           : BitWidth);
+      EVT VT = Op.getValueType();
+      // Fill in any known bits from range information. There are 3 types being
+      // used. The results VT (same vector elt size as BitWidth), the loaded
+      // MemoryVT (which may or may not be vector) and the range VTs original
+      // type. The range matadata needs the full range (i.e
+      // MemoryVT().getSizeInBits()), which is truncated to the correct elt size
+      // if it is know. These are then extended to the original VT sizes below.
+      if (const MDNode *MD = LD->getRanges()) {
+        computeKnownBitsFromRangeMetadata(*MD, Known0);
         if (VT.isVector()) {
           // Handle truncation to the first demanded element.
           // TODO: Figure out which demanded elements are covered
           if (DemandedElts != 1 || !getDataLayout().isLittleEndian())
             break;
+          Known0 = Known0.trunc(BitWidth);
+        }
+      }
 
-          // Handle the case where a load has a vector type, but scalar memory
-          // with an attached range.
-          EVT MemVT = LD->getMemoryVT();
-          KnownBits KnownFull(MemVT.getSizeInBits());
+      if (LD->getMemoryVT().isVector())
+        Known0 = Known0.trunc(LD->getMemoryVT().getScalarSizeInBits());
 
-          computeKnownBitsFromRangeMetadata(*Ranges, KnownFull);
-          Known = KnownFull.trunc(BitWidth);
-        } else
-          computeKnownBitsFromRangeMetadata(*Ranges, Known);
-      }
+      // Extend the Known bits from memory to the size of the result.
+      if (ISD::isZEXTLoad(Op.getNode()))
+        Known = Known0.zext(BitWidth);
+      else if (ISD::isSEXTLoad(Op.getNode()))
+        Known = Known0.sext(BitWidth);
+      else if (ISD::isEXTLoad(Op.getNode()))
+        Known = Known0.anyext(BitWidth);
+      else
+        Known = Known0;
+      assert(Known.getBitWidth() == BitWidth);
+      return Known;
     }
     break;
   }
diff --git a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
index 9e9c814be0266..60d6187e3423d 100644
--- a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
+++ b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
@@ -21,9 +21,7 @@ define noundef i1 @logger(i32 noundef %logLevel, ptr %ea, ptr %pll) {
 ; CHECK-NEXT:    ret
 ; CHECK-NEXT:  .LBB1_2: // %land.rhs
 ; CHECK-NEXT:    ldr x8, [x1]
-; CHECK-NEXT:    ldrb w8, [x8]
-; CHECK-NEXT:    cmp w8, #0
-; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ldrb w0, [x8]
 ; CHECK-NEXT:    ret
 entry:
   %0 = load i32, ptr %pll, align 4

``````````

</details>


https://github.com/llvm/llvm-project/pull/80829