[llvm-bugs] [Bug 49024] New: Failure to optimize add loop in -O3 as well as in -Oz

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=49024

            Bug ID: 49024
           Summary: Failure to optimize add loop in -O3 as well as in -Oz
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: gabravier at gmail.com
                CC: llvm-bugs at lists.llvm.org

int f(uint8_t max_value) {
    int sum = 0;
    for (int i = 0; i < max_value; i++) {
        sum += i;
    }
    return sum;
}

This can be optimized to `return (((unsigned _ExtInt(33))max_value) * (unsigned
_ExtInt(33))((uint32_t)max_value - 1)) >> 1;`. This transformation is done
under `-Oz`, but `-O3` seems to generate far inferior code, as the IR shows:

define dso_local i32 @_Z1fh(i8 zeroext %0) local_unnamed_addr #0 {
  %2 = zext i8 %0 to i32
  %3 = icmp eq i8 %0, 0
  br i1 %3, label %14, label %4

4: ; preds = %1
  %5 = add nsw i32 %2, -1
  %6 = zext i32 %5 to i33
  %7 = add nsw i32 %2, -2
  %8 = zext i32 %7 to i33
  %9 = mul i33 %6, %8
  %10 = lshr i33 %9, 1
  %11 = trunc i33 %10 to i32
  %12 = add i32 %2, %11
  %13 = add i32 %12, -1
  br label %14

14: ; preds = %4, %1
  %15 = phi i32 [ 0, %1 ], [ %13, %4 ]
  ret i32 %15
}

Compared to the far simpler and faster code generated on -Oz:

define dso_local i32 @_Z1fh(i8 zeroext %0) local_unnamed_addr #0 {
  %2 = zext i8 %0 to i32
  %3 = zext i8 %0 to i33
  %4 = add nsw i32 %2, -1
  %5 = zext i32 %4 to i33
  %6 = mul i33 %3, %5
  %7 = lshr i33 %6, 1
  %8 = trunc i33 %7 to i32
  ret i32 %8
}

Godbolt comparison: https://godbolt.org/z/vW5jKz
alive2 comparison: https://alive2.llvm.org/ce/z/NVhmdm

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210203/5da96e8d/attachment.html>