[llvm] PreISelIntrinsicLowering: Lower llvm.exp to a loop if scalable vec arg (PR #117568)

[Matt Arsenault via llvm-commits]

================
@@ -453,6 +454,11 @@ bool PreISelIntrinsicLowering::lowerIntrinsics(Module &M) const {
     case Intrinsic::objc_sync_exit:
       Changed |= lowerObjCCall(F, "objc_sync_exit");
       break;
+    case Intrinsic::exp:
+      Changed |= forEachCall(F, [&](CallInst *CI) {
+        return lowerUnaryMathIntrinsicWithScalableVecArgAsLoop(M, CI);
----------------
arsenm wrote:

For IR passes that implement lowerings, we have a mixed bag of inconsistent strategies. Various APIs are implemented in TargetLowering, or TargetTransformInfo (some of which call into TargetLowering). 

At the most base level, you would want to check the legalization action for the corresponding ISD:: node to the intrinsic. At least in the fixed vector case, it probably depends on more context.

For a more brute-force approach, shouldExpandReduction used by ExpandReductions is similar. I don't really like this style, it's too specific and duplicates logic that should be present in the legalize rules 

https://github.com/llvm/llvm-project/pull/117568