[libc][NFC] Split memset implementations per platform

This is a follow up on D154800 and D154770 to make the code structure more principled and avoid too many nested #ifdef/#endif.

Reviewed By: courbet

Differential Revision: https://reviews.llvm.org/D155174

5 files changed, 180 insertions, 109 deletions

diff --git a/libc/src/string/memory_utils/CMakeLists.txt b/libc/src/string/memory_utils/CMakeLists.txt
index 60a2482..ecccd2d 100644
--- a/libc/src/string/memory_utils/CMakeLists.txt
+++ b/libc/src/string/memory_utils/CMakeLists.txt
@@ -5,6 +5,7 @@ add_header_library(
     aarch64/bcmp_implementations.h
     aarch64/memcmp_implementations.h
     aarch64/memcpy_implementations.h
+    aarch64/memset_implementations.h
     bcmp_implementations.h
     bzero_implementations.h
     generic/aligned_access.h
@@ -19,10 +20,12 @@ add_header_library(
     op_x86.h
     riscv/bcmp_implementations.h
     riscv/memcpy_implementations.h
+    riscv/memset_implementations.h
     utils.h
     x86_64/bcmp_implementations.h
     x86_64/memcmp_implementations.h
     x86_64/memcpy_implementations.h
+    x86_64/memset_implementations.h
   DEPS
     libc.src.__support.common
     libc.src.__support.CPP.bit
diff --git a/libc/src/string/memory_utils/aarch64/memset_implementations.h b/libc/src/string/memory_utils/aarch64/memset_implementations.h
new file mode 100644
index 0000000..a457897
--- /dev/null
+++ b/libc/src/string/memory_utils/aarch64/memset_implementations.h
@@ -0,0 +1,61 @@
+//===-- Memset implementation for aarch64 -----------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef LIBC_SRC_STRING_MEMORY_UTILS_AARCH64_MEMSET_IMPLEMENTATIONS_H
+#define LIBC_SRC_STRING_MEMORY_UTILS_AARCH64_MEMSET_IMPLEMENTATIONS_H
+
+#include "src/__support/macros/attributes.h" // LIBC_INLINE
+#include "src/string/memory_utils/op_aarch64.h"
+#include "src/string/memory_utils/op_generic.h"
+#include "src/string/memory_utils/utils.h" // Ptr, CPtr
+
+#include <stddef.h> // size_t
+
+namespace __llvm_libc {
+
+[[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;
+  if (count == 0)
+    return;
+  if (count <= 3) {
+    generic::Memset<uint8_t>::block(dst, value);
+    if (count > 1)
+      generic::Memset<uint16_t>::tail(dst, value, count);
+    return;
+  }
+  if (count <= 8)
+    return generic::Memset<uint32_t>::head_tail(dst, value, count);
+  if (count <= 16)
+    return generic::Memset<uint64_t>::head_tail(dst, value, count);
+  if (count <= 32)
+    return generic::Memset<uint128_t>::head_tail(dst, value, count);
+  if (count <= (32 + 64)) {
+    generic::Memset<uint256_t>::block(dst, value);
+    if (count <= 64)
+      return generic::Memset<uint256_t>::tail(dst, value, count);
+    generic::Memset<uint256_t>::block(dst + 32, value);
+    generic::Memset<uint256_t>::tail(dst, value, count);
+    return;
+  }
+  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);
+  } else {
+    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);
+  }
+}
+
+} // namespace __llvm_libc
+
+#endif // LIBC_SRC_STRING_MEMORY_UTILS_AARCH64_MEMSET_IMPLEMENTATIONS_H
diff --git a/libc/src/string/memory_utils/memset_implementations.h b/libc/src/string/memory_utils/memset_implementations.h
index 691d7f2..37ceff9 100644
--- a/libc/src/string/memory_utils/memset_implementations.h
+++ b/libc/src/string/memory_utils/memset_implementations.h
@@ -9,126 +9,35 @@
 #ifndef LLVM_LIBC_SRC_STRING_MEMORY_UTILS_MEMSET_IMPLEMENTATIONS_H
 #define LLVM_LIBC_SRC_STRING_MEMORY_UTILS_MEMSET_IMPLEMENTATIONS_H
 
-#include "src/__support/common.h"
-#include "src/__support/macros/optimization.h"
-#include "src/__support/macros/properties/architectures.h"
-#include "src/string/memory_utils/generic/aligned_access.h"
-#include "src/string/memory_utils/generic/byte_per_byte.h"
-#include "src/string/memory_utils/op_aarch64.h"
-#include "src/string/memory_utils/op_builtin.h"
-#include "src/string/memory_utils/op_generic.h"
-#include "src/string/memory_utils/op_x86.h"
-#include "src/string/memory_utils/utils.h"
+#include "src/__support/macros/config.h"                   // LIBC_INLINE
+#include "src/__support/macros/properties/architectures.h" // LIBC_TARGET_ARCH_IS_
+#include "src/string/memory_utils/utils.h"                 // Ptr, CPtr
 
 #include <stddef.h> // size_t
 
-namespace __llvm_libc {
-
-#if defined(LIBC_TARGET_ARCH_IS_X86)
-[[maybe_unused]] LIBC_INLINE static void
-inline_memset_x86(Ptr dst, uint8_t value, size_t count) {
-#if defined(__AVX512F__)
-  using uint128_t = generic_v128;
-  using uint256_t = generic_v256;
-  using uint512_t = generic_v512;
-#elif defined(__AVX__)
-  using uint128_t = generic_v128;
-  using uint256_t = generic_v256;
-  using uint512_t = cpp::array<generic_v256, 2>;
-#elif defined(__SSE2__)
-  using uint128_t = generic_v128;
-  using uint256_t = cpp::array<generic_v128, 2>;
-  using uint512_t = cpp::array<generic_v128, 4>;
-#else
-  using uint128_t = cpp::array<uint64_t, 2>;
-  using uint256_t = cpp::array<uint64_t, 4>;
-  using uint512_t = cpp::array<uint64_t, 8>;
-#endif
-
-  if (count == 0)
-    return;
-  if (count == 1)
-    return generic::Memset<uint8_t>::block(dst, value);
-  if (count == 2)
-    return generic::Memset<uint16_t>::block(dst, value);
-  if (count == 3)
-    return generic::MemsetSequence<uint16_t, uint8_t>::block(dst, value);
-  if (count <= 8)
-    return generic::Memset<uint32_t>::head_tail(dst, value, count);
-  if (count <= 16)
-    return generic::Memset<uint64_t>::head_tail(dst, value, count);
-  if (count <= 32)
-    return generic::Memset<uint128_t>::head_tail(dst, value, count);
-  if (count <= 64)
-    return generic::Memset<uint256_t>::head_tail(dst, value, count);
-  if (count <= 128)
-    return generic::Memset<uint512_t>::head_tail(dst, value, count);
-  // Aligned loop
-  generic::Memset<uint256_t>::block(dst, value);
-  align_to_next_boundary<32>(dst, count);
-  return generic::Memset<uint256_t>::loop_and_tail(dst, value, count);
-}
-#endif // defined(LIBC_TARGET_ARCH_IS_X86)
-
-#if defined(LIBC_TARGET_ARCH_IS_AARCH64)
-[[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;
-  if (count == 0)
-    return;
-  if (count <= 3) {
-    generic::Memset<uint8_t>::block(dst, value);
-    if (count > 1)
-      generic::Memset<uint16_t>::tail(dst, value, count);
-    return;
-  }
-  if (count <= 8)
-    return generic::Memset<uint32_t>::head_tail(dst, value, count);
-  if (count <= 16)
-    return generic::Memset<uint64_t>::head_tail(dst, value, count);
-  if (count <= 32)
-    return generic::Memset<uint128_t>::head_tail(dst, value, count);
-  if (count <= (32 + 64)) {
-    generic::Memset<uint256_t>::block(dst, value);
-    if (count <= 64)
-      return generic::Memset<uint256_t>::tail(dst, value, count);
-    generic::Memset<uint256_t>::block(dst + 32, value);
-    generic::Memset<uint256_t>::tail(dst, value, count);
-    return;
-  }
-  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);
-  } else {
-    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 // defined(LIBC_TARGET_ARCH_IS_AARCH64)
-
-LIBC_INLINE static void inline_memset(Ptr dst, uint8_t value, size_t count) {
 #if defined(LIBC_TARGET_ARCH_IS_X86)
-  return inline_memset_x86(dst, value, count);
+#include "src/string/memory_utils/x86_64/memset_implementations.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_x86
 #elif defined(LIBC_TARGET_ARCH_IS_AARCH64)
-  return inline_memset_aarch64(dst, value, count);
-#elif defined(LIBC_TARGET_ARCH_IS_RISCV64)
-  return inline_memset_aligned_access_64bit(dst, value, count);
-#elif defined(LIBC_TARGET_ARCH_IS_RISCV32)
-  return inline_memset_aligned_access_32bit(dst, value, count);
+#include "src/string/memory_utils/aarch64/memset_implementations.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_aarch64
+#elif defined(LIBC_TARGET_ARCH_IS_ANY_RISCV)
+#include "src/string/memory_utils/riscv/memset_implementations.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_riscv
 #else
-  return inline_memset_byte_per_byte(dst, value, count);
+// We may want to error instead of defaulting to suboptimal implementation.
+#include "src/string/memory_utils/generic/byte_per_byte.h"
+#define LIBC_SRC_STRING_MEMORY_UTILS_MEMSET inline_memset_byte_per_byte
 #endif
-}
+
+namespace __llvm_libc {
 
 LIBC_INLINE static void inline_memset(void *dst, uint8_t value, size_t count) {
-  inline_memset(reinterpret_cast<Ptr>(dst), value, count);
+  LIBC_SRC_STRING_MEMORY_UTILS_MEMSET(reinterpret_cast<Ptr>(dst), value, count);
 }
 
 } // namespace __llvm_libc
 
+#undef LIBC_SRC_STRING_MEMORY_UTILS_MEMSET
+
 #endif // LLVM_LIBC_SRC_STRING_MEMORY_UTILS_MEMSET_IMPLEMENTATIONS_H
diff --git a/libc/src/string/memory_utils/riscv/memset_implementations.h b/libc/src/string/memory_utils/riscv/memset_implementations.h
new file mode 100644
index 0000000..80be228
--- /dev/null
+++ b/libc/src/string/memory_utils/riscv/memset_implementations.h
@@ -0,0 +1,33 @@
+//===-- Memset implementation for riscv -------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef LIBC_SRC_STRING_MEMORY_UTILS_RISCV_MEMSET_IMPLEMENTATIONS_H
+#define LIBC_SRC_STRING_MEMORY_UTILS_RISCV_MEMSET_IMPLEMENTATIONS_H
+
+#include "src/__support/macros/attributes.h"               // LIBC_INLINE
+#include "src/__support/macros/properties/architectures.h" // LIBC_TARGET_ARCH_IS_RISCV64
+#include "src/string/memory_utils/generic/aligned_access.h"
+#include "src/string/memory_utils/utils.h" // Ptr, CPtr
+
+#include <stddef.h> // size_t
+
+namespace __llvm_libc {
+
+LIBC_INLINE static void inline_memset_riscv(Ptr dst, uint8_t value,
+                                            size_t count) {
+#if defined(LIBC_TARGET_ARCH_IS_RISCV64)
+  return inline_memset_aligned_access_64bit(dst, value, count);
+#elif defined(LIBC_TARGET_ARCH_IS_RISCV32)
+  return inline_memset_aligned_access_32bit(dst, value, count);
+#else
+#error "Unimplemented"
+#endif
+}
+
+} // namespace __llvm_libc
+
+#endif // LIBC_SRC_STRING_MEMORY_UTILS_RISCV_MEMSET_IMPLEMENTATIONS_H
diff --git a/libc/src/string/memory_utils/x86_64/memset_implementations.h b/libc/src/string/memory_utils/x86_64/memset_implementations.h
new file mode 100644
index 0000000..cb26854
--- /dev/null
+++ b/libc/src/string/memory_utils/x86_64/memset_implementations.h
@@ -0,0 +1,65 @@
+//===-- Memset implementation for x86_64 ------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef LIBC_SRC_STRING_MEMORY_UTILS_X86_64_MEMSET_IMPLEMENTATIONS_H
+#define LIBC_SRC_STRING_MEMORY_UTILS_X86_64_MEMSET_IMPLEMENTATIONS_H
+
+#include "src/__support/macros/attributes.h" // LIBC_INLINE
+#include "src/string/memory_utils/op_generic.h"
+#include "src/string/memory_utils/op_x86.h"
+#include "src/string/memory_utils/utils.h" // Ptr, CPtr
+
+#include <stddef.h> // size_t
+
+namespace __llvm_libc {
+
+[[maybe_unused]] LIBC_INLINE static void
+inline_memset_x86(Ptr dst, uint8_t value, size_t count) {
+#if defined(__AVX512F__)
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = generic_v512;
+#elif defined(__AVX__)
+  using uint128_t = generic_v128;
+  using uint256_t = generic_v256;
+  using uint512_t = cpp::array<generic_v256, 2>;
+#elif defined(__SSE2__)
+  using uint128_t = generic_v128;
+  using uint256_t = cpp::array<generic_v128, 2>;
+  using uint512_t = cpp::array<generic_v128, 4>;
+#else
+  using uint128_t = cpp::array<uint64_t, 2>;
+  using uint256_t = cpp::array<uint64_t, 4>;
+  using uint512_t = cpp::array<uint64_t, 8>;
+#endif
+
+  if (count == 0)
+    return;
+  if (count == 1)
+    return generic::Memset<uint8_t>::block(dst, value);
+  if (count == 2)
+    return generic::Memset<uint16_t>::block(dst, value);
+  if (count == 3)
+    return generic::MemsetSequence<uint16_t, uint8_t>::block(dst, value);
+  if (count <= 8)
+    return generic::Memset<uint32_t>::head_tail(dst, value, count);
+  if (count <= 16)
+    return generic::Memset<uint64_t>::head_tail(dst, value, count);
+  if (count <= 32)
+    return generic::Memset<uint128_t>::head_tail(dst, value, count);
+  if (count <= 64)
+    return generic::Memset<uint256_t>::head_tail(dst, value, count);
+  if (count <= 128)
+    return generic::Memset<uint512_t>::head_tail(dst, value, count);
+  // Aligned loop
+  generic::Memset<uint256_t>::block(dst, value);
+  align_to_next_boundary<32>(dst, count);
+  return generic::Memset<uint256_t>::loop_and_tail(dst, value, count);
+}
+} // namespace __llvm_libc
+
+#endif // LIBC_SRC_STRING_MEMORY_UTILS_X86_64_MEMSET_IMPLEMENTATIONS_H