[libc-commits] [libc] [libc] Add support for string/memory_utils functions for AArch64 without HW FP/SIMD (PR #137592)

[Guillaume Chatelet via libc-commits]

================
@@ -18,12 +18,34 @@
 
 namespace LIBC_NAMESPACE_DECL {
 
+using uint128_t = generic_v128;
+using uint256_t = generic_v256;
+using uint512_t = generic_v512;
+
 [[maybe_unused]] LIBC_INLINE static void
-inline_memset_aarch64(Ptr dst, uint8_t value, size_t count) {
-  static_assert(aarch64::kNeon, "aarch64 supports vector types");
-  using uint128_t = generic_v128;
-  using uint256_t = generic_v256;
-  using uint512_t = generic_v512;
+inline_memset_aarch64_no_fp(Ptr dst, uint8_t value, size_t count) {
+  generic::Memset<uint128_t>::block(dst, value);
+  align_to_next_boundary<16>(dst, count);
+  return generic::Memset<uint512_t>::loop_and_tail(dst, value, count);
+}
+
+#if defined(__ARM_NEON)
+[[maybe_unused]] LIBC_INLINE static void
+inline_memset_aarch64_with_fp(Ptr dst, uint8_t value, size_t count) {
+  if (count >= 448 && value == 0 && aarch64::neon::hasZva()) {
+    generic::Memset<uint512_t>::block(dst, 0);
+    align_to_next_boundary<64>(dst, count);
+    return aarch64::neon::BzeroCacheLine::loop_and_tail(dst, 0, count);
+  }
+
+  generic::Memset<uint128_t>::block(dst, value);
+  align_to_next_boundary<16>(dst, count);
+  return generic::Memset<uint512_t>::loop_and_tail(dst, value, count);
+}
+#endif
+
+[[gnu::flatten]] [[maybe_unused]] LIBC_INLINE static void
----------------
gchatelet wrote:

ditto

https://github.com/llvm/llvm-project/pull/137592