mirror of
https://github.com/horizon-rust/horizon.git
synced 2026-11-07 11:06:11 -10:00
Ludovic MarieandClaude Opus 5.5
3380705af9
Since #4160 (10-29) the game's allocator is mimalloc v3, from libmimalloc-sys 0.1.49. Built by MSVC, its mi_popcount is `__popcnt64`, which MSVC turns into the POPCNT instruction whatever the target. mimalloc runs before main, so on a Core 2 (no POPCNT) the Windows game died at once with 0xc000001d and no log (#7264). v0.2.97 had no POPCNT at all; nightlies 2026-10-31 and 11-01 have 9. third_party/libmimalloc-sys is the published crate with one change: on MSVC x86/x64, mi_popcount takes mimalloc's portable count, as the GCC builds without -mpopcnt already do. Only arena bitmaps count bits, never malloc's fast path. [patch.crates-io] points at it until a release carries microsoft/mimalloc#1291. tools/check_cpu_baseline.py disassembles the Windows and Linux programs in the game build and fails on any POPCNT or LZCNT: it flags 2026-11-01's horizon-client.exe and passes v0.2.97's and the Linux builds. Fixes #7264 Co-Authored-By: Claude Opus 5.5 <[email protected]>
95 lines
3.6 KiB
Python
95 lines
3.6 KiB
Python
#!/usr/bin/env python3
|
|
"""Fail when an x86_64 release binary uses an instruction that baseline x86-64 CPUs lack.
|
|
|
|
The game must start on players' oldest PCs: x86-64 with SSE2 only (Core 2, AMD K8). A
|
|
single unguarded POPCNT killed it with 0xc000001d before `main` on a Core 2 (#7264):
|
|
mimalloc's C code, built by MSVC, whose `__popcnt` intrinsic is the instruction whatever
|
|
the target.
|
|
|
|
Only POPCNT and LZCNT are checked. They are what C compilers emit for count intrinsics, and
|
|
no runtime-dispatched code in the game uses them today. AVX, BMI2, SSSE3 and SSE4 do appear,
|
|
always behind a CPUID check (ring's assembly, memchr, half's F16C, compiler_builtins' FMA,
|
|
the MSVC C runtime), and the Windows programs have no symbols to tell those functions from
|
|
unguarded ones. `tzcnt` is not checked either: LLVM writes `rep bsf` for trailing_zeros,
|
|
which disassembles as tzcnt and runs as bsf on every CPU.
|
|
|
|
Needs llvm-objdump: $LLVM_OBJDUMP, rustup's llvm-tools component, or one on PATH.
|
|
"""
|
|
|
|
import argparse
|
|
import glob
|
|
import os
|
|
import re
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
|
|
FORBIDDEN = ("popcnt", "lzcnt")
|
|
LABEL = re.compile(r"^([0-9a-f]+) <(.*)>:$")
|
|
INSTRUCTION = re.compile(r"^\s*([0-9a-f]+):\s+(\S+)")
|
|
|
|
|
|
def find_objdump():
|
|
if os.environ.get("LLVM_OBJDUMP"):
|
|
return os.environ["LLVM_OBJDUMP"]
|
|
try:
|
|
sysroot = subprocess.run(["rustc", "--print", "sysroot"], capture_output=True, text=True,
|
|
check=True).stdout.strip()
|
|
found = glob.glob(os.path.join(sysroot, "lib", "rustlib", "*", "bin", "llvm-objdump*"))
|
|
if found:
|
|
return found[0]
|
|
except (OSError, subprocess.CalledProcessError):
|
|
pass
|
|
return shutil.which("llvm-objdump")
|
|
|
|
|
|
def forbidden_uses(lines):
|
|
"""(address, enclosing symbol, mnemonic) of each forbidden instruction in llvm-objdump -d output."""
|
|
symbol = "?"
|
|
for line in lines:
|
|
line = line.rstrip("\n")
|
|
label = LABEL.match(line)
|
|
if label:
|
|
symbol = label.group(2)
|
|
continue
|
|
insn = INSTRUCTION.match(line)
|
|
if not insn:
|
|
continue
|
|
mnemonic = insn.group(2)
|
|
# AT&T syntax adds the operand size: popcntq, lzcntl.
|
|
base = mnemonic[:-1] if mnemonic[-1:] in "wlq" and mnemonic[:-1] in FORBIDDEN else mnemonic
|
|
if base in FORBIDDEN:
|
|
yield insn.group(1), symbol, base
|
|
|
|
|
|
def check(objdump, path):
|
|
proc = subprocess.Popen([objdump, "-d", "--no-show-raw-insn", path], stdout=subprocess.PIPE,
|
|
text=True, errors="replace")
|
|
uses = list(forbidden_uses(proc.stdout))
|
|
if proc.wait() != 0:
|
|
sys.exit(f"llvm-objdump failed on {path}")
|
|
for address, symbol, mnemonic in uses[:20]:
|
|
print(f"{path}: {mnemonic} at 0x{address} in {symbol}")
|
|
if len(uses) > 20:
|
|
print(f"{path}: ... {len(uses) - 20} more")
|
|
return not uses
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
|
|
parser.add_argument("binaries", nargs="+")
|
|
args = parser.parse_args()
|
|
objdump = find_objdump()
|
|
if not objdump:
|
|
sys.exit("llvm-objdump not found: rustup component add llvm-tools, or set LLVM_OBJDUMP")
|
|
failed = [path for path in args.binaries if not check(objdump, path)]
|
|
if failed:
|
|
print(f"::error::{', '.join(failed)}: {'/'.join(FORBIDDEN)} without a CPU check crash CPUs "
|
|
"older than Nehalem (#7264); find the C library or target_feature that emits them")
|
|
sys.exit(1)
|
|
print(f"no {'/'.join(FORBIDDEN)} in {len(args.binaries)} binaries")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|