Revert "[libc] New version of the mem* framework"

This reverts commit d55f2d8.

libc/src/stdio/printf_core/string_writer.cpp

@@ -33,7 +33,7 @@ void StringWriter::write(char new_char, size_t len) {
     len = available_capacity;
 
   if (len > 0) {
-    inline_memset(cur_buffer, static_cast<uint8_t>(new_char), len);
+    inline_memset(cur_buffer, new_char, len);
     cur_buffer += len;
     available_capacity -= len;
   }

libc/src/string/bcmp.cpp

@@ -14,8 +14,8 @@ namespace __llvm_libc {
 
 LLVM_LIBC_FUNCTION(int, bcmp,
                    (const void *lhs, const void *rhs, size_t count)) {
-  return static_cast<int>(inline_bcmp(static_cast<const char *>(lhs),
-                                      static_cast<const char *>(rhs), count));
+  return inline_bcmp(static_cast<const char *>(lhs),
+                     static_cast<const char *>(rhs), count);
 }
 
 } // namespace __llvm_libc

libc/src/string/memcmp.cpp

@@ -15,8 +15,8 @@ namespace __llvm_libc {
 
 LLVM_LIBC_FUNCTION(int, memcmp,
                    (const void *lhs, const void *rhs, size_t count)) {
-  return static_cast<int>(inline_memcmp(static_cast<const char *>(lhs),
-                                        static_cast<const char *>(rhs), count));
+  return inline_memcmp(static_cast<const char *>(lhs),
+                       static_cast<const char *>(rhs), count);
 }
 
 } // namespace __llvm_libc

libc/src/string/memmove.cpp

@@ -9,61 +9,36 @@
 #include "src/string/memmove.h"
 
 #include "src/__support/common.h"
-#include "src/string/memory_utils/op_aarch64.h"
-#include "src/string/memory_utils/op_generic.h"
-#include "src/string/memory_utils/op_x86.h"
+#include "src/__support/integer_operations.h"
+#include "src/string/memory_utils/elements.h"
 #include <stddef.h> // size_t, ptrdiff_t
 
-#include <stdio.h>
-
 namespace __llvm_libc {
 
 static inline void inline_memmove(char *dst, const char *src, size_t count) {
-#if defined(LLVM_LIBC_ARCH_X86) || defined(LLVM_LIBC_ARCH_AARCH64)
-#if defined(LLVM_LIBC_ARCH_X86)
-  static constexpr size_t kMaxSize = x86::kAvx512F ? 64
-                                     : x86::kAvx   ? 32
-                                     : x86::kSse2  ? 16
-                                                   : 8;
-#elif defined(LLVM_LIBC_ARCH_AARCH64)
-  static constexpr size_t kMaxSize = aarch64::kNeon ? 16 : 8;
-#endif
+  using namespace __llvm_libc::scalar;
   if (count == 0)
     return;
   if (count == 1)
-    return generic::Memmove<1, kMaxSize>::block(dst, src);
+    return move<_1>(dst, src);
   if (count <= 4)
-    return generic::Memmove<2, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_2>>(dst, src, count);
   if (count <= 8)
-    return generic::Memmove<4, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_4>>(dst, src, count);
   if (count <= 16)
-    return generic::Memmove<8, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_8>>(dst, src, count);
   if (count <= 32)
-    return generic::Memmove<16, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_16>>(dst, src, count);
   if (count <= 64)
-    return generic::Memmove<32, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_32>>(dst, src, count);
   if (count <= 128)
-    return generic::Memmove<64, kMaxSize>::head_tail(dst, src, count);
+    return move<HeadTail<_64>>(dst, src, count);
 
-  if (dst < src) {
-    generic::Memmove<32, kMaxSize>::align_forward<Arg::Src>(dst, src, count);
-    return generic::Memmove<64, kMaxSize>::loop_and_tail_forward(dst, src,
-                                                                 count);
-  } else {
-    generic::Memmove<32, kMaxSize>::align_backward<Arg::Src>(dst, src, count);
-    return generic::Memmove<64, kMaxSize>::loop_and_tail_backward(dst, src,
-                                                                  count);
-  }
-#elif defined(LLVM_LIBC_ARCH_ARM)
-  if (count == 0)
-    return;
+  using AlignedMoveLoop = Align<_16, Arg::Src>::Then<Loop<_64>>;
   if (dst < src)
-    return generic::Memmove<1, 1>::loop_forward(dst, src, count);
-  else
-    return generic::Memmove<1, 1>::loop_backward(dst, src, count);
-#else
-#error "Unsupported platform"
-#endif
+    return move<AlignedMoveLoop>(dst, src, count);
+  else if (dst > src)
+    return move_backward<AlignedMoveLoop>(dst, src, count);
 }
 
 LLVM_LIBC_FUNCTION(void *, memmove,

libc/src/string/memory_utils/CMakeLists.txt

@@ -2,17 +2,13 @@
 add_header_library(
   memory_utils
   HDRS
+    utils.h
+    elements.h
     bcmp_implementations.h
     bzero_implementations.h
     memcmp_implementations.h
     memcpy_implementations.h
     memset_implementations.h
-    op_aarch64.h
-    op_higher_order.h
-    op_builtin.h
-    op_generic.h
-    op_x86.h
-    utils.h
   DEPS
     libc.src.__support.CPP.bit
 )

libc/src/string/memory_utils/bcmp_implementations.h

@@ -11,136 +11,49 @@
 
 #include "src/__support/architectures.h"
 #include "src/__support/common.h"
-#include "src/string/memory_utils/op_aarch64.h"
-#include "src/string/memory_utils/op_generic.h"
-#include "src/string/memory_utils/op_x86.h"
+#include "src/string/memory_utils/elements.h"
 
 #include <stddef.h> // size_t
 
 namespace __llvm_libc {
 
-#if defined(LLVM_LIBC_ARCH_X86) || defined(LLVM_LIBC_ARCH_AARCH64)
-static inline BcmpReturnType inline_bcmp_generic_gt16(CPtr p1, CPtr p2,
-                                                      size_t count) {
-  if (count < 256)
-    return generic::Bcmp<16>::loop_and_tail(p1, p2, count);
-  if (auto value = generic::Bcmp<64>::block(p1, p2))
-    return value;
-  align_to_next_boundary<64, Arg::P1>(p1, p2, count);
-  return generic::Bcmp<64>::loop_and_tail(p1, p2, count);
+// Fixed-size difference between 'lhs' and 'rhs'.
+template <typename Element> bool differs(const char *lhs, const char *rhs) {
+  return !Element::equals(lhs, rhs);
 }
-#endif // defined(LLVM_LIBC_ARCH_X86) || defined(LLVM_LIBC_ARCH_AARCH64)
-
-#if defined(LLVM_LIBC_ARCH_X86)
-static inline BcmpReturnType inline_bcmp_x86_sse2_gt16(CPtr p1, CPtr p2,
-                                                       size_t count) {
-  if (count <= 32)
-    return x86::sse2::Bcmp<16>::head_tail(p1, p2, count);
-  if (count < 256)
-    return x86::sse2::Bcmp<16>::loop_and_tail(p1, p2, count);
-  if (auto value = x86::sse2::Bcmp<16>::block(p1, p2))
-    return value;
-  align_to_next_boundary<16, Arg::P1>(p1, p2, count);
-  return x86::sse2::Bcmp<64>::loop_and_tail(p1, p2, count);
-}
-
-static inline BcmpReturnType inline_bcmp_x86_avx2_gt16(CPtr p1, CPtr p2,
-                                                       size_t count) {
-  if (count <= 32)
-    return x86::sse2::Bcmp<16>::head_tail(p1, p2, count);
-  if (count <= 64)
-    return x86::avx2::Bcmp<32>::head_tail(p1, p2, count);
-  if (count <= 128)
-    return x86::avx2::Bcmp<64>::head_tail(p1, p2, count);
-  if (unlikely(count >= 256)) {
-    if (auto value = x86::avx2::Bcmp<64>::block(p1, p2))
-      return value;
-    align_to_next_boundary<64, Arg::P1>(p1, p2, count);
-  }
-  return x86::avx2::Bcmp<64>::loop_and_tail(p1, p2, count);
-}
-
-static inline BcmpReturnType inline_bcmp_x86_avx512bw_gt16(CPtr p1, CPtr p2,
-                                                           size_t count) {
-  if (count <= 32)
-    return x86::sse2::Bcmp<16>::head_tail(p1, p2, count);
-  if (count <= 64)
-    return x86::avx2::Bcmp<32>::head_tail(p1, p2, count);
-  if (count <= 128)
-    return x86::avx512bw::Bcmp<64>::head_tail(p1, p2, count);
-  if (unlikely(count >= 256)) {
-    if (auto value = x86::avx512bw::Bcmp<64>::block(p1, p2))
-      return value;
-    align_to_next_boundary<64, Arg::P1>(p1, p2, count);
-  }
-  return x86::avx512bw::Bcmp<64>::loop_and_tail(p1, p2, count);
+// Runtime-size difference between 'lhs' and 'rhs'.
+template <typename Element>
+bool differs(const char *lhs, const char *rhs, size_t size) {
+  return !Element::equals(lhs, rhs, size);
 }
-#endif // defined(LLVM_LIBC_ARCH_X86)
 
-static inline BcmpReturnType inline_bcmp(CPtr p1, CPtr p2, size_t count) {
+static inline int inline_bcmp(const char *lhs, const char *rhs, size_t count) {
 #if defined(LLVM_LIBC_ARCH_X86)
+  using namespace ::__llvm_libc::x86;
+#elif defined(LLVM_LIBC_ARCH_AARCH64)
+  using namespace ::__llvm_libc::aarch64;
+#else
+  using namespace ::__llvm_libc::scalar;
+#endif
   if (count == 0)
-    return BcmpReturnType::ZERO();
+    return 0;
   if (count == 1)
-    return generic::Bcmp<1>::block(p1, p2);
+    return differs<_1>(lhs, rhs);
   if (count == 2)
-    return generic::Bcmp<2>::block(p1, p2);
-  if (count <= 4)
-    return generic::Bcmp<2>::head_tail(p1, p2, count);
+    return differs<_2>(lhs, rhs);
+  if (count == 3)
+    return differs<_3>(lhs, rhs);
   if (count <= 8)
-    return generic::Bcmp<4>::head_tail(p1, p2, count);
+    return differs<HeadTail<_4>>(lhs, rhs, count);
   if (count <= 16)
-    return generic::Bcmp<8>::head_tail(p1, p2, count);
-  if constexpr (x86::kAvx512BW)
-    return inline_bcmp_x86_avx512bw_gt16(p1, p2, count);
-  else if constexpr (x86::kAvx2)
-    return inline_bcmp_x86_avx2_gt16(p1, p2, count);
-  else if constexpr (x86::kSse2)
-    return inline_bcmp_x86_sse2_gt16(p1, p2, count);
-  else
-    return inline_bcmp_generic_gt16(p1, p2, count);
-#elif defined(LLVM_LIBC_ARCH_AARCH64)
-  if (likely(count <= 32)) {
-    if (unlikely(count >= 16)) {
-      return generic::Bcmp<16>::head_tail(p1, p2, count);
-    }
-    switch (count) {
-    case 0:
-      return BcmpReturnType::ZERO();
-    case 1:
-      return generic::Bcmp<1>::block(p1, p2);
-    case 2:
-      return generic::Bcmp<2>::block(p1, p2);
-    case 3:
-      return generic::Bcmp<2>::head_tail(p1, p2, count);
-    case 4:
-      return generic::Bcmp<4>::block(p1, p2);
-    case 5 ... 7:
-      return generic::Bcmp<4>::head_tail(p1, p2, count);
-    case 8:
-      return generic::Bcmp<8>::block(p1, p2);
-    case 9 ... 15:
-      return generic::Bcmp<8>::head_tail(p1, p2, count);
-    }
-  }
-
+    return differs<HeadTail<_8>>(lhs, rhs, count);
+  if (count <= 32)
+    return differs<HeadTail<_16>>(lhs, rhs, count);
   if (count <= 64)
-    return generic::Bcmp<32>::head_tail(p1, p2, count);
-
-  // Aligned loop if > 256, otherwise normal loop
-  if (count > 256) {
-    if (auto value = generic::Bcmp<32>::block(p1, p2))
-      return value;
-    align_to_next_boundary<16, Arg::P1>(p1, p2, count);
-  }
-  return generic::Bcmp<32>::loop_and_tail(p1, p2, count);
-#elif defined(LLVM_LIBC_ARCH_ARM)
-  if (count == 0)
-    return BcmpReturnType::ZERO();
-  return generic::Bcmp<1>::loop(p1, p2, count);
-#else
-#error "Unsupported platform"
-#endif
+    return differs<HeadTail<_32>>(lhs, rhs, count);
+  if (count <= 128)
+    return differs<HeadTail<_64>>(lhs, rhs, count);
+  return differs<Align<_32>::Then<Loop<_32>>>(lhs, rhs, count);
 }
 
 } // namespace __llvm_libc