[llvm] [AMDGPU] Rework getDivNumBits API (PR #119768)

[Jay Foad via llvm-commits]

================
@@ -1189,27 +1188,32 @@ static Value* getMulHu(IRBuilder<> &Builder, Value *LHS, Value *RHS) {
   return getMul64(Builder, LHS, RHS).second;
 }
 
-/// Figure out how many bits are really needed for this division. \p AtLeast is
-/// an optimization hint to bypass the second ComputeNumSignBits call if we the
-/// first one is insufficient. Returns -1 on failure.
-int AMDGPUCodeGenPrepareImpl::getDivNumBits(BinaryOperator &I, Value *Num,
-                                            Value *Den, unsigned AtLeast,
-                                            bool IsSigned) const {
+/// Figure out how many bits are really needed for this division.
+/// \p MaxDivBits is an optimization hint to bypass the second
+/// ComputeNumSignBits/computeKnownBits call if we the first one is
+/// insufficient.
+unsigned AMDGPUCodeGenPrepareImpl::getDivNumBits(BinaryOperator &I, Value *Num,
+                                                 Value *Den,
+                                                 unsigned MaxDivBits,
+                                                 bool IsSigned) const {
   assert(Num->getType()->getScalarSizeInBits() ==
          Den->getType()->getScalarSizeInBits());
   unsigned SSBits = Num->getType()->getScalarSizeInBits();
   if (IsSigned) {
     unsigned RHSSignBits = ComputeNumSignBits(Den, DL, 0, AC, &I);
-    if (RHSSignBits < AtLeast)
-      return -1;
+    // a SignBit needs to be reserved for shrinking
+    unsigned DivBits = SSBits - RHSSignBits + 1;
+    if (DivBits > MaxDivBits)
+      return DivBits;
 
     unsigned LHSSignBits = ComputeNumSignBits(Num, DL, 0, AC, &I);
-    if (LHSSignBits < AtLeast)
-      return -1;
+    DivBits = SSBits - LHSSignBits + 1;
----------------
jayfoad wrote:

There is no need for this early-out, since it does not save any expensive calls to ComputeNumSignBits. Same comments apply to the unsigned cases below.

https://github.com/llvm/llvm-project/pull/119768