AMDGPU: Don't avoid clamp of bit shift in BFE pattern

changpeng · changpeng · commit aeb611ad0d1e · 2024-11-07T14:10:42.000-08:00
Use "countMaxTrailingOnes() &lt;= 5" instead of "countMinLeadingZeros() &gt;= 27"
in uint5Bits PatLeaf definition.
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -3554,7 +3554,7 @@ def : AMDGPUPat <
 >;
 
 def uint5Bits : PatLeaf<(i32 VGPR_32:$width), [{
-  return CurDAG->computeKnownBits(SDValue(N, 0)).countMinLeadingZeros() >= 27;
+  return CurDAG->computeKnownBits(SDValue(N, 0)).countMaxTrailingOnes() <= 5;
 }]>;
 
 // x << (bitwidth - y) >> (bitwidth - y)
