Revert "Remove all custom CopyCharsUnsigned implementations"

This reverts commit 5d8c489000.

Reason for revert: Fails on UBSan bot:
https://ci.chromium.org/p/v8/builders/ci/V8%20Linux64%20UBSan/7946

Original change's description:
> Remove all custom CopyCharsUnsigned implementations
> 
> It's unclear whether the custom implementation have any advantage over
> the standard library one's.
> Since we update our toolchain and standard library regularly, it might
> well be the case that the custom implementations are slower by now.
> 
> Thus this CL removes all {CopyCharsUnsigned} implementations and
> implements {CopyChars} generically using {std::copy_n}.
> 
> Note that this does not touch the {MemMove} and {MemCopy} functions
> yet, as we have seen regressions when trying to remove them before
> (https://crbug.com/v8/8675#c5).
> 
> R=​leszeks@chromium.org
> 
> Bug: v8:9396
> Change-Id: I97a183afebcccd2fbb567bdba02e827331475608
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1800577
> Commit-Queue: Clemens Hammacher <clemensh@chromium.org>
> Reviewed-by: Leszek Swirski <leszeks@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#63808}

TBR=leszeks@chromium.org,clemensh@chromium.org

Change-Id: Ia16da942c7c28ba71076d1e3b0b8a6388a4ba359
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: v8:9396
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1806103
Reviewed-by: Adam Klein <adamk@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#63810}

src/builtins/arm/builtins-arm.cc

@@ -3167,6 +3167,51 @@ void Builtins::Generate_MemCopyUint8Uint8(MacroAssembler* masm) {
   __ Ret();
 }
 
+void Builtins::Generate_MemCopyUint16Uint8(MacroAssembler* masm) {
+  Register dest = r0;
+  Register src = r1;
+  Register chars = r2;
+
+  {
+    UseScratchRegisterScope temps(masm);
+
+    Register temp1 = r3;
+    Register temp2 = temps.Acquire();
+    Register temp3 = lr;
+    Register temp4 = r4;
+    Label loop;
+    Label not_two;
+
+    __ Push(lr, r4);
+    __ bic(temp2, chars, Operand(0x3));
+    __ add(temp2, dest, Operand(temp2, LSL, 1));
+
+    __ bind(&loop);
+    __ ldr(temp1, MemOperand(src, 4, PostIndex));
+    __ uxtb16(temp3, temp1);
+    __ uxtb16(temp4, temp1, 8);
+    __ pkhbt(temp1, temp3, Operand(temp4, LSL, 16));
+    __ str(temp1, MemOperand(dest));
+    __ pkhtb(temp1, temp4, Operand(temp3, ASR, 16));
+    __ str(temp1, MemOperand(dest, 4));
+    __ add(dest, dest, Operand(8));
+    __ cmp(dest, temp2);
+    __ b(&loop, ne);
+
+    __ mov(chars, Operand(chars, LSL, 31), SetCC);  // bit0 => ne, bit1 => cs
+    __ b(&not_two, cc);
+    __ ldrh(temp1, MemOperand(src, 2, PostIndex));
+    __ uxtb(temp3, temp1, 8);
+    __ mov(temp3, Operand(temp3, LSL, 16));
+    __ uxtab(temp3, temp3, temp1);
+    __ str(temp3, MemOperand(dest, 4, PostIndex));
+    __ bind(&not_two);
+    __ ldrb(temp1, MemOperand(src), ne);
+    __ strh(temp1, MemOperand(dest), ne);
+    __ Pop(pc, r4);
+  }
+}
+
 #undef __
 
 }  // namespace internal

src/builtins/builtins-definitions.h

@@ -1113,6 +1113,7 @@ namespace internal {
   TFS(SetProperty, kReceiver, kKey, kValue)                                    \
   TFS(SetPropertyInLiteral, kReceiver, kKey, kValue)                           \
   ASM(MemCopyUint8Uint8, CCall)                                                \
+  ASM(MemCopyUint16Uint8, CCall)                                               \
   ASM(MemMove, CCall)                                                          \
                                                                                \
   /* Trace */                                                                  \

src/builtins/builtins-internal-gen.cc

@@ -933,6 +933,12 @@ void Builtins::Generate_MemCopyUint8Uint8(MacroAssembler* masm) {
 }
 #endif  // !defined(V8_TARGET_ARCH_ARM) && !defined(V8_TARGET_ARCH_MIPS)
 
+#ifndef V8_TARGET_ARCH_ARM
+void Builtins::Generate_MemCopyUint16Uint8(MacroAssembler* masm) {
+  masm->Call(BUILTIN_CODE(masm->isolate(), Illegal), RelocInfo::CODE_TARGET);
+}
+#endif  // V8_TARGET_ARCH_ARM
+
 #ifndef V8_TARGET_ARCH_IA32
 void Builtins::Generate_MemMove(MacroAssembler* masm) {
   masm->Call(BUILTIN_CODE(masm->isolate(), Illegal), RelocInfo::CODE_TARGET);

src/parsing/scanner-character-streams.cc

@@ -265,7 +265,7 @@ class BufferedCharacterStream : public Utf16CharacterStream {
     }
 
     size_t length = Min(kBufferSize, range.length());
-    i::CopyChars(buffer_, range.start, length);
+    i::CopyCharsUnsigned(buffer_, range.start, length);
     buffer_end_ = &buffer_[length];
     return true;
   }

src/utils/memcopy.cc

@@ -25,8 +25,18 @@ V8_EXPORT_PRIVATE void MemMove(void* dest, const void* src, size_t size) {
   (*memmove_function)(dest, src, size);
 }
 #elif V8_OS_POSIX && V8_HOST_ARCH_ARM
+void MemCopyUint16Uint8Wrapper(uint16_t* dest, const uint8_t* src,
+                               size_t chars) {
+  uint16_t* limit = dest + chars;
+  while (dest < limit) {
+    *dest++ = static_cast<uint16_t>(*src++);
+  }
+}
+
 V8_EXPORT_PRIVATE MemCopyUint8Function memcopy_uint8_function =
     &MemCopyUint8Wrapper;
+MemCopyUint16Uint8Function memcopy_uint16_uint8_function =
+    &MemCopyUint16Uint8Wrapper;
 #elif V8_OS_POSIX && V8_HOST_ARCH_MIPS
 V8_EXPORT_PRIVATE MemCopyUint8Function memcopy_uint8_function =
     &MemCopyUint8Wrapper;
@@ -44,6 +54,9 @@ void init_memcopy_functions() {
     EmbeddedData d = EmbeddedData::FromBlob();
     memcopy_uint8_function = reinterpret_cast<MemCopyUint8Function>(
         d.InstructionStartOfBuiltin(Builtins::kMemCopyUint8Uint8));
+    memcopy_uint16_uint8_function =
+        reinterpret_cast<MemCopyUint16Uint8Function>(
+            d.InstructionStartOfBuiltin(Builtins::kMemCopyUint16Uint8));
   }
 #elif V8_OS_POSIX && V8_HOST_ARCH_MIPS
   if (Isolate::CurrentEmbeddedBlobIsBinaryEmbedded()) {

src/utils/memcopy.h

@@ -8,7 +8,6 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
-#include <algorithm>
 
 #include "src/base/logging.h"
 #include "src/base/macros.h"
@@ -56,8 +55,17 @@ V8_EXPORT_PRIVATE V8_INLINE void MemMove(void* dest, const void* src,
   memmove(dest, src, size);
 }
 
+using MemCopyUint16Uint8Function = void (*)(uint16_t* dest, const uint8_t* src,
+                                            size_t size);
+extern MemCopyUint16Uint8Function memcopy_uint16_uint8_function;
+void MemCopyUint16Uint8Wrapper(uint16_t* dest, const uint8_t* src,
+                               size_t chars);
 // For values < 12, the assembler function is slower than the inlined C code.
 const int kMinComplexConvertMemCopy = 12;
+V8_INLINE void MemCopyUint16Uint8(uint16_t* dest, const uint8_t* src,
+                                  size_t size) {
+  (*memcopy_uint16_uint8_function)(dest, src, size);
+}
 #elif defined(V8_HOST_ARCH_MIPS)
 using MemCopyUint8Function = void (*)(uint8_t* dest, const uint8_t* src,
                                       size_t size);
@@ -196,39 +204,309 @@ inline void MemsetPointer(T** dest, U* value, size_t counter) {
                 reinterpret_cast<Address>(value), counter);
 }
 
-// Copy from 8bit/16bit chars to 8bit/16bit chars. Values are zero-extended if
+template <typename sourcechar, typename sinkchar>
-// needed. Ranges are not allowed to overlap unless the type sizes match (hence
+V8_INLINE static void CopyCharsUnsigned(sinkchar* dest, const sourcechar* src,
-// {memmove} is used internally).
+                                        size_t chars);
-template <typename SrcType, typename DstType>
+#if defined(V8_HOST_ARCH_ARM)
-void CopyChars(DstType* dst, const SrcType* src, size_t count) {
+V8_INLINE void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src,
-  STATIC_ASSERT(std::is_integral<SrcType>::value);
+                                 size_t chars);
-  STATIC_ASSERT(std::is_integral<DstType>::value);
+V8_INLINE void CopyCharsUnsigned(uint16_t* dest, const uint8_t* src,
-
+                                 size_t chars);
-  // If the size of {SrcType} and {DstType} matches, we switch to the more
+V8_INLINE void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src,
-  // general and potentially faster {memmove}. Note that this decision is made
+                                 size_t chars);
-  // statically at compile time.
+#elif defined(V8_HOST_ARCH_MIPS)
-  // This {memmove} also allows to pass overlapping ranges if the sizes match.
+V8_INLINE void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src,
-  // We probably should not call {CopyChars} with overlapping ranges, but
+                                 size_t chars);
-  // unfortunately we sometimes do.
+V8_INLINE void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src,
-  if (sizeof(SrcType) == sizeof(DstType)) {
+                                 size_t chars);
-    memmove(dst, src, count * sizeof(SrcType));
+#elif defined(V8_HOST_ARCH_PPC) || defined(V8_HOST_ARCH_S390)
-    return;
+V8_INLINE void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src,
-  }
+                                 size_t chars);
-
+V8_INLINE void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src,
-  using SrcTypeUnsigned = typename std::make_unsigned<SrcType>::type;
+                                 size_t chars);
-  using DstTypeUnsigned = typename std::make_unsigned<DstType>::type;
-#ifdef DEBUG
-  // Check for no overlap, otherwise {std::copy_n} cannot be used.
-  Address src_start = reinterpret_cast<Address>(src);
-  Address src_end = src_start + count * sizeof(SrcType);
-  Address dst_start = reinterpret_cast<Address>(dst);
-  Address dst_end = dst_start + count * sizeof(DstType);
-  DCHECK(src_end < dst_start || dst_end < src_start);
 #endif
-  std::copy_n(reinterpret_cast<const SrcTypeUnsigned*>(src), count,
+
-              reinterpret_cast<DstTypeUnsigned*>(dst));
+// Copy from 8bit/16bit chars to 8bit/16bit chars.
+template <typename sourcechar, typename sinkchar>
+V8_INLINE void CopyChars(sinkchar* dest, const sourcechar* src, size_t chars);
+
+template <typename sourcechar, typename sinkchar>
+void CopyChars(sinkchar* dest, const sourcechar* src, size_t chars) {
+  DCHECK_LE(sizeof(sourcechar), 2);
+  DCHECK_LE(sizeof(sinkchar), 2);
+  if (sizeof(sinkchar) == 1) {
+    if (sizeof(sourcechar) == 1) {
+      CopyCharsUnsigned(reinterpret_cast<uint8_t*>(dest),
+                        reinterpret_cast<const uint8_t*>(src), chars);
+    } else {
+      CopyCharsUnsigned(reinterpret_cast<uint8_t*>(dest),
+                        reinterpret_cast<const uint16_t*>(src), chars);
+    }
+  } else {
+    if (sizeof(sourcechar) == 1) {
+      CopyCharsUnsigned(reinterpret_cast<uint16_t*>(dest),
+                        reinterpret_cast<const uint8_t*>(src), chars);
+    } else {
+      CopyCharsUnsigned(reinterpret_cast<uint16_t*>(dest),
+                        reinterpret_cast<const uint16_t*>(src), chars);
+    }
+  }
 }
 
+template <typename sourcechar, typename sinkchar>
+void CopyCharsUnsigned(sinkchar* dest, const sourcechar* src, size_t chars) {
+  sinkchar* limit = dest + chars;
+  if ((sizeof(*dest) == sizeof(*src)) &&
+      (chars >= kMinComplexMemCopy / sizeof(*dest))) {
+    MemCopy(dest, src, chars * sizeof(*dest));
+  } else {
+    while (dest < limit) *dest++ = static_cast<sinkchar>(*src++);
+  }
+}
+
+#if defined(V8_HOST_ARCH_ARM)
+void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src, size_t chars) {
+  switch (static_cast<unsigned>(chars)) {
+    case 0:
+      break;
+    case 1:
+      *dest = *src;
+      break;
+    case 2:
+      memcpy(dest, src, 2);
+      break;
+    case 3:
+      memcpy(dest, src, 3);
+      break;
+    case 4:
+      memcpy(dest, src, 4);
+      break;
+    case 5:
+      memcpy(dest, src, 5);
+      break;
+    case 6:
+      memcpy(dest, src, 6);
+      break;
+    case 7:
+      memcpy(dest, src, 7);
+      break;
+    case 8:
+      memcpy(dest, src, 8);
+      break;
+    case 9:
+      memcpy(dest, src, 9);
+      break;
+    case 10:
+      memcpy(dest, src, 10);
+      break;
+    case 11:
+      memcpy(dest, src, 11);
+      break;
+    case 12:
+      memcpy(dest, src, 12);
+      break;
+    case 13:
+      memcpy(dest, src, 13);
+      break;
+    case 14:
+      memcpy(dest, src, 14);
+      break;
+    case 15:
+      memcpy(dest, src, 15);
+      break;
+    default:
+      MemCopy(dest, src, chars);
+      break;
+  }
+}
+
+void CopyCharsUnsigned(uint16_t* dest, const uint8_t* src, size_t chars) {
+  if (chars >= static_cast<size_t>(kMinComplexConvertMemCopy)) {
+    MemCopyUint16Uint8(dest, src, chars);
+  } else {
+    MemCopyUint16Uint8Wrapper(dest, src, chars);
+  }
+}
+
+void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src, size_t chars) {
+  switch (static_cast<unsigned>(chars)) {
+    case 0:
+      break;
+    case 1:
+      *dest = *src;
+      break;
+    case 2:
+      memcpy(dest, src, 4);
+      break;
+    case 3:
+      memcpy(dest, src, 6);
+      break;
+    case 4:
+      memcpy(dest, src, 8);
+      break;
+    case 5:
+      memcpy(dest, src, 10);
+      break;
+    case 6:
+      memcpy(dest, src, 12);
+      break;
+    case 7:
+      memcpy(dest, src, 14);
+      break;
+    default:
+      MemCopy(dest, src, chars * sizeof(*dest));
+      break;
+  }
+}
+
+#elif defined(V8_HOST_ARCH_MIPS)
+void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src, size_t chars) {
+  if (chars < kMinComplexMemCopy) {
+    memcpy(dest, src, chars);
+  } else {
+    MemCopy(dest, src, chars);
+  }
+}
+
+void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src, size_t chars) {
+  if (chars < kMinComplexMemCopy) {
+    memcpy(dest, src, chars * sizeof(*dest));
+  } else {
+    MemCopy(dest, src, chars * sizeof(*dest));
+  }
+}
+#elif defined(V8_HOST_ARCH_PPC) || defined(V8_HOST_ARCH_S390)
+#define CASE(n)           \
+  case n:                 \
+    memcpy(dest, src, n); \
+    break
+void CopyCharsUnsigned(uint8_t* dest, const uint8_t* src, size_t chars) {
+  switch (static_cast<unsigned>(chars)) {
+    case 0:
+      break;
+    case 1:
+      *dest = *src;
+      break;
+      CASE(2);
+      CASE(3);
+      CASE(4);
+      CASE(5);
+      CASE(6);
+      CASE(7);
+      CASE(8);
+      CASE(9);
+      CASE(10);
+      CASE(11);
+      CASE(12);
+      CASE(13);
+      CASE(14);
+      CASE(15);
+      CASE(16);
+      CASE(17);
+      CASE(18);
+      CASE(19);
+      CASE(20);
+      CASE(21);
+      CASE(22);
+      CASE(23);
+      CASE(24);
+      CASE(25);
+      CASE(26);
+      CASE(27);
+      CASE(28);
+      CASE(29);
+      CASE(30);
+      CASE(31);
+      CASE(32);
+      CASE(33);
+      CASE(34);
+      CASE(35);
+      CASE(36);
+      CASE(37);
+      CASE(38);
+      CASE(39);
+      CASE(40);
+      CASE(41);
+      CASE(42);
+      CASE(43);
+      CASE(44);
+      CASE(45);
+      CASE(46);
+      CASE(47);
+      CASE(48);
+      CASE(49);
+      CASE(50);
+      CASE(51);
+      CASE(52);
+      CASE(53);
+      CASE(54);
+      CASE(55);
+      CASE(56);
+      CASE(57);
+      CASE(58);
+      CASE(59);
+      CASE(60);
+      CASE(61);
+      CASE(62);
+      CASE(63);
+      CASE(64);
+    default:
+      memcpy(dest, src, chars);
+      break;
+  }
+}
+#undef CASE
+
+#define CASE(n)               \
+  case n:                     \
+    memcpy(dest, src, n * 2); \
+    break
+void CopyCharsUnsigned(uint16_t* dest, const uint16_t* src, size_t chars) {
+  switch (static_cast<unsigned>(chars)) {
+    case 0:
+      break;
+    case 1:
+      *dest = *src;
+      break;
+      CASE(2);
+      CASE(3);
+      CASE(4);
+      CASE(5);
+      CASE(6);
+      CASE(7);
+      CASE(8);
+      CASE(9);
+      CASE(10);
+      CASE(11);
+      CASE(12);
+      CASE(13);
+      CASE(14);
+      CASE(15);
+      CASE(16);
+      CASE(17);
+      CASE(18);
+      CASE(19);
+      CASE(20);
+      CASE(21);
+      CASE(22);
+      CASE(23);
+      CASE(24);
+      CASE(25);
+      CASE(26);
+      CASE(27);
+      CASE(28);
+      CASE(29);
+      CASE(30);
+      CASE(31);
+      CASE(32);
+    default:
+      memcpy(dest, src, chars * 2);
+      break;
+  }
+}
+#undef CASE
+#endif
+
 }  // namespace internal
 }  // namespace v8