[llvm] [InstCombine] Fold shift+cttz with power of 2 operands (PR #127055)

Matthew Devereau via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 13 08:56:46 PST 2025


================
@@ -103,4 +103,38 @@ entry:
   ret i32 %res
 }
 
+define i64 @fold_cttz_64() vscale_range(1,16) {
+; CHECK-LABEL: define i64 @fold_cttz_64(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    ret i64 4
+;
+entry:
+  %0 = tail call i64 @llvm.vscale.i64()
+  %1 = shl nuw nsw i64 %0, 4
+  %2 = shl nuw nsw i64 %0, 2
+  %3 = tail call range(i64 2, 65) i64 @llvm.cttz.i64(i64 %2, i1 true)
+  %div1 = lshr i64 %1, %3
+  ret i64 %div1
+}
+
+define i32 @fold_cttz_32() vscale_range(1,16) {
+; CHECK-LABEL: define i32 @fold_cttz_32(
+; CHECK-SAME: ) #[[ATTR0]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    ret i32 4
+;
+entry:
+  %0 = tail call i32 @llvm.vscale.i32()
+  %1 = shl nuw nsw i32 %0, 4
+  %2 = shl nuw nsw i32 %0, 2
+  %3 = tail call range(i32 2, 65) i32 @llvm.cttz.i32(i32 %2, i1 true)
+  %div1 = lshr i32 %1, %3
+  ret i32 %div1
+}
+
+declare i64 @llvm.vscale.i64()
+declare i64 @llvm.cttz.i64(i64, i1 immarg)
+declare i32 @llvm.vscale.i32()
+declare i32 @llvm.cttz.i32(i32, i1 immarg)
----------------
MDevereau wrote:

Removed

https://github.com/llvm/llvm-project/pull/127055


More information about the llvm-commits mailing list