[llvm] [AMDGPU] Narrow 64 bit math to 32 bit if profitable (PR #130577)

[Shilei Tian via llvm-commits]

================
@@ -1224,6 +1224,69 @@ static bool foldLibCalls(Instruction &I, TargetTransformInfo &TTI,
   return false;
 }
 
+static bool isSaveToNarrow(unsigned opc, uint64_t num1, uint64_t num2) {
+  if (num1 > 0xffffffff || num2 > 0xffffffff) {
+    // if `num > 0xffffffff`, then `%and = and i64 %a, num` may or may not have
+    // higher 32bit set. Which cause truncate possibly lose infomation
+    return false;
+  }
+  switch (opc) {
+  // If `%and = and i64 %a, num` where num <= 0xffffffff, then `%and` must be
+  // positive.
+  // Since add and mul both increasing function on positive integer domain and
+  // `%ai <= numi`, then if `(num1 op num2) <= 0xffffffff` we have `%a1 + %a2 <=
+  // 0xffffffff`
+  case Instruction::Add:
+    return (num1 + num2) <= 0xffffffff;
+  case Instruction::Mul:
+    return (num1 * num2) <= 0xffffffff;
+    break;
+  }
+
+  return false;
+}
+
+static bool tryNarrowMathIfNoOverflow(Instruction &I,
+                                      TargetTransformInfo &TTI) {
+  unsigned opc = I.getOpcode();
+  if (opc != Instruction::Add && opc != Instruction::Mul) {
+    return false;
+  }
+  LLVMContext &ctx = I.getContext();
+  Type *i64type = Type::getInt64Ty(ctx);
+  Type *i32type = Type::getInt32Ty(ctx);
+
+  if (I.getType() != i64type || !TTI.isTruncateFree(i64type, i32type)) {
+    return false;
+  }
+  InstructionCost costOp64 =
----------------
shiltian wrote:

Variable names in this functions don't follow LLVM standard.

https://github.com/llvm/llvm-project/pull/130577