[Support] Refactor getN1Bits so it does not work around any g++ bug

[AtariDreams]

This also folds better than the previous version as well.

Proof:
https://alive2.llvm.org/ce/z/6uwy95

[llvmbot]

@llvm/pr-subscribers-llvm-support

Author: AtariDreams (AtariDreams)

Changes

This also folds better than the previous version as well.

---

Full diff: https://github.com/llvm/llvm-project/pull/78933.diff

1 Files Affected:

• (modified) llvm/include/llvm/Support/Discriminator.h (+2-5)

diff --git a/llvm/include/llvm/Support/Discriminator.h b/llvm/include/llvm/Support/Discriminator.h
index fa78cf3045de35..e2d2a5bceee4b9 100644
--- a/llvm/include/llvm/Support/Discriminator.h
+++ b/llvm/include/llvm/Support/Discriminator.h
@@ -121,12 +121,9 @@ static inline unsigned getBaseFSBitEnd() {
 }
 
 // Set bits in range of [0 .. n] to 1. Used in FS Discriminators.
-static inline unsigned getN1Bits(int N) {
-  // Work around the g++ bug that folding "(1U << (N + 1)) - 1" to 0.
-  if (N == 31)
-    return 0xFFFFFFFF;
+static inline unsigned getN1Bits(unsigned N) {
   assert((N < 32) && "N is invalid");
-  return (1U << (N + 1)) - 1;
+  return 0xFFFFFFFFU >> (31 - N);
 }
 
 } // namespace llvm

[boomanaiden154]

@AtariDreams It doesn't appear like this PR (or #87560, #87313, #85162, or #91386) have been approved by a reviewer? The LLVM review policy technically does allow post-commit review, but that is typically de-facto reserved for code owners working in their specific area. These PRs should probably be reverted until they are adequately reviewed.

[nikic]

I've reverted the changes.

[preames]

@AtariDreams It doesn't appear like this PR (or #87560, #87313, #85162, or #91386) have been approved by a reviewer? The LLVM review policy technically does allow post-commit review, but that is typically de-facto reserved for code owners working in their specific area. These PRs should probably be reverted until they are adequately reviewed.

I want to state this far more strongly. Landing PRs without appropriate review is a significant violation of the review policy and our developer practices. This is wildly inappropriate, and if continued would be a reason for revoking commit rights entirely.

In your case, please err strongly towards waiting on review. Only commit without pre-review if you have established trust with contributors in a given area. As an observer, it seems clear you have lost any trust you may have built, so please do not land a change without pre review at all for near future.