Fix implicit conversion loses integer precision warning

The type of m is long in 64 bits build, and results implicit conversion
loses integer precision, which was found by improved clang warning
(-Wshorten-64-to-32)

Bug: chromium:1124085
Change-Id: Ic9f22508bd817a06d5c90162b1ac3554a7171529
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2391323
Commit-Queue: Zequan Wu <zequanwu@google.com>
Auto-Submit: Zequan Wu <zequanwu@google.com>
Reviewed-by: Nico Weber <thakis@chromium.org>
Reviewed-by: Igor Sheludko <ishell@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69686}

src/base/macros.h

@@ -336,14 +336,14 @@ inline T RoundDown(T x, intptr_t m) {
   STATIC_ASSERT(std::is_integral<T>::value);
   // m must be a power of two.
   DCHECK(m != 0 && ((m & (m - 1)) == 0));
-  return x & -m;
+  return x & static_cast<T>(-m);
 }
 template <intptr_t m, typename T>
 constexpr inline T RoundDown(T x) {
   STATIC_ASSERT(std::is_integral<T>::value);
   // m must be a power of two.
   STATIC_ASSERT(m != 0 && ((m & (m - 1)) == 0));
-  return x & -m;
+  return x & static_cast<T>(-m);
 }
 
 // Return the smallest multiple of m which is >= x.