[bigint] Move String-to-BigInt parsing to src/bigint/

No changes to the algorithm, approximately 4x performance
improvement thanks to reduced overhead.

Bug: v8:11515
Change-Id: Id3f6c91bd650f6ae47ac8f169dc780420091998e
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3046185
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Maya Lekova <mslekova@chromium.org>
Cr-Commit-Position: refs/heads/master@{#76022}

BUILD.bazel

@@ -2685,6 +2685,7 @@ filegroup(
         "src/bigint/div-helpers.cc",
         "src/bigint/div-helpers.h",
         "src/bigint/div-schoolbook.cc",
+        "src/bigint/fromstring.cc",
         "src/bigint/mul-fft.cc",
         "src/bigint/mul-karatsuba.cc",
         "src/bigint/mul-schoolbook.cc",

BUILD.gn

@@ -5005,6 +5005,7 @@ v8_source_set("v8_bigint") {
     "src/bigint/div-helpers.cc",
     "src/bigint/div-helpers.h",
     "src/bigint/div-schoolbook.cc",
+    "src/bigint/fromstring.cc",
     "src/bigint/mul-karatsuba.cc",
     "src/bigint/mul-schoolbook.cc",
     "src/bigint/tostring.cc",

src/bigint/bigint-internal.h

@@ -67,6 +67,9 @@ class ProcessorImpl : public Processor {
   void ToStringImpl(char* out, int* out_length, Digits X, int radix, bool sign,
                     bool use_fast_algorithm);
 
+  void FromString(RWDigits Z, FromStringAccumulator* accumulator);
+  void FromStringClassic(RWDigits Z, FromStringAccumulator* accumulator);
+
   bool should_terminate() { return status_ == Status::kInterrupted; }
 
   // Each unit is supposed to represent approximately one CPU {mul} instruction.

src/bigint/bigint.h

@@ -10,6 +10,7 @@
 #include <algorithm>
 #include <cstring>
 #include <iostream>
+#include <vector>
 
 namespace v8 {
 namespace bigint {
@@ -235,6 +236,8 @@ bool SubtractSigned(RWDigits Z, Digits X, bool x_negative, Digits Y,
 
 enum class Status { kOk, kInterrupted };
 
+class FromStringAccumulator;
+
 class Processor {
  public:
   // Takes ownership of {platform}.
@@ -258,6 +261,8 @@ class Processor {
   // {out_length} initially contains the allocated capacity of {out}, and
   // upon return will be set to the actual length of the result string.
   Status ToString(char* out, int* out_length, Digits X, int radix, bool sign);
+
+  Status FromString(RWDigits Z, FromStringAccumulator* accumulator);
 };
 
 inline int AddResultLength(int x_length, int y_length) {
@@ -296,9 +301,130 @@ int ToStringResultLength(Digits X, int radix, bool sign);
 // In DEBUG builds, the result of {ToString} will be initialized to this value.
 constexpr char kStringZapValue = '?';
 
+// Support for parsing BigInts from Strings, using an Accumulator object
+// for intermediate state.
+
+class ProcessorImpl;
+
+#if defined(__GNUC__) || defined(__clang__)
+// Clang supports this since 3.9, GCC since 5.x.
+#define HAVE_BUILTIN_MUL_OVERFLOW 1
+#else
+#define HAVE_BUILTIN_MUL_OVERFLOW 0
+#endif
+
+// A container object for all metadata required for parsing a BigInt from
+// a string.
+// Aggressively optimized not to waste instructions for small cases, while
+// also scaling transparently to huge cases.
+// Defined here in the header so that {ConsumeChar} can be inlined.
+class FromStringAccumulator {
+ public:
+  // {max_digits} is only used for refusing to grow beyond a given size
+  // (see "Step 1" below). Does not cause pre-allocation, so feel free to
+  // specify a large maximum.
+  // TODO(jkummerow): The limit applies to the number of intermediate chunks,
+  // whereas the final result will be slightly smaller (depending on {radix}).
+  // So setting max_digits=N here will, for sufficiently large N, not actually
+  // allow parsing BigInts with N digits. We can fix that if/when anyone cares.
+  FromStringAccumulator(int radix, int max_digits)
+      : radix_(radix),
+#if !HAVE_BUILTIN_MUL_OVERFLOW
+        max_multiplier_((~digit_t{0}) / radix),
+#endif
+        max_digits_(max_digits),
+        limit_digit_(radix < 10 ? radix : 10),
+        limit_alpha_(radix > 10 ? radix - 10 : 0) {
+  }
+
+  ~FromStringAccumulator() {
+    delete parts_;
+    delete multipliers_;
+  }
+
+  // Step 1: Call this method repeatedly to read all characters.
+  // This method will return quickly; it does not perform heavy processing.
+  enum class Result { kOk, kInvalidChar, kMaxSizeExceeded };
+  Result ConsumeChar(uint32_t c) {
+    digit_t d;
+    if (c - '0' < limit_digit_) {
+      d = c - '0';
+    } else if ((c | 32u) - 'a' < limit_alpha_) {
+      d = (c | 32u) - 'a' + 10;
+    } else {
+      return Result::kInvalidChar;
+    }
+#if HAVE_BUILTIN_MUL_OVERFLOW
+    digit_t m;
+    if (!__builtin_mul_overflow(multiplier_, radix_, &m)) {
+      multiplier_ = m;
+      part_ = part_ * radix_ + d;
+    }
+#else
+    if (multiplier_ <= max_multiplier_) {
+      multiplier_ *= radix_;
+      part_ = part_ * radix_ + d;
+    }
+#endif
+    else {  // NOLINT(readability/braces)
+      if (!AddPart(multiplier_, part_)) return Result::kMaxSizeExceeded;
+      multiplier_ = radix_;
+      part_ = d;
+    }
+    return Result::kOk;
+  }
+
+  // Step 2: Call this method to determine the required size for the result.
+  int ResultLength() {
+    if (!parts_) return part_ > 0 ? 1 : 0;
+    if (multiplier_ > 1) {
+      multipliers_->push_back(multiplier_);
+      parts_->push_back(part_);
+      // {ResultLength} should be idempotent.
+      multiplier_ = 1;
+      part_ = 0;
+    }
+    return parts_size();
+  }
+
+  // Step 3: Use BigIntProcessor::FromString() to retrieve the result into an
+  // {RWDigits} struct allocated for the size returned by step 2.
+
+ private:
+  friend class ProcessorImpl;
+  int parts_size() { return static_cast<int>(parts_->size()); }
+
+  bool AddPart(digit_t multiplier, digit_t part) {
+    if (!parts_) {
+      parts_ = new std::vector<digit_t>;
+      multipliers_ = new std::vector<digit_t>;
+    } else if (parts_size() == max_digits_) {
+      return false;
+    }
+    multipliers_->push_back(multiplier);
+    parts_->push_back(part);
+    return true;
+  }
+
+  const digit_t radix_;
+#if !HAVE_BUILTIN_MUL_OVERFLOW
+  const digit_t max_multiplier_;
+#endif
+  // The next part to be added to {parts_}, or the only part when sufficient.
+  digit_t part_{0};
+  digit_t multiplier_{1};
+  const int max_digits_;
+  const uint32_t limit_digit_;
+  const uint32_t limit_alpha_;
+  // Avoid allocating these unless we actually need them.
+  std::vector<digit_t>* parts_{nullptr};
+  std::vector<digit_t>* multipliers_{nullptr};
+};
+
 }  // namespace bigint
 }  // namespace v8
 
 #undef BIGINT_H_DCHECK
+#undef HAVE_BUILTIN_MUL_OVERFLOW
 
 #endif  // V8_BIGINT_BIGINT_H_

src/bigint/fromstring.cc

+// Copyright 2021 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/bigint/bigint-internal.h"
+#include "src/bigint/vector-arithmetic.h"
+
+namespace v8 {
+namespace bigint {
+
+// The classic algorithm: for every part, multiply the accumulator with
+// the appropriate multiplier, and add the part. O(n²) overall.
+void ProcessorImpl::FromStringClassic(RWDigits Z,
+                                      FromStringAccumulator* accumulator) {
+  Z[0] = (*accumulator->parts_)[0];
+  RWDigits already_set(Z, 0, 1);
+  for (int i = 1; i < Z.len(); i++) Z[i] = 0;
+  for (int i = 1; i < accumulator->parts_size(); i++) {
+    MultiplySingle(Z, already_set, (*accumulator->multipliers_)[i]);
+    if (should_terminate()) return;
+    Add(Z, (*accumulator->parts_)[i]);
+    already_set.set_len(already_set.len() + 1);
+  }
+}
+
+void ProcessorImpl::FromString(RWDigits Z, FromStringAccumulator* accumulator) {
+  if (!accumulator->parts_) {
+    if (Z.len() > 0) Z[0] = accumulator->part_;
+    for (int i = 1; i < Z.len(); i++) Z[i] = 0;
+  } else {
+    FromStringClassic(Z, accumulator);
+  }
+}
+
+Status Processor::FromString(RWDigits Z, FromStringAccumulator* accumulator) {
+  ProcessorImpl* impl = static_cast<ProcessorImpl*>(this);
+  impl->FromString(Z, accumulator);
+  return impl->get_and_clear_status();
+}
+
+}  // namespace bigint
+}  // namespace v8

src/execution/local-isolate.cc

@@ -4,6 +4,7 @@
 
 #include "src/execution/local-isolate.h"
 
+#include "src/bigint/bigint.h"
 #include "src/execution/isolate.h"
 #include "src/execution/thread-id.h"
 #include "src/handles/handles-inl.h"
@@ -24,7 +25,9 @@ LocalIsolate::LocalIsolate(Isolate* isolate, ThreadKind kind,
                        : GetCurrentStackPosition() - FLAG_stack_size * KB),
       runtime_call_stats_(runtime_call_stats) {}
 
-LocalIsolate::~LocalIsolate() = default;
+LocalIsolate::~LocalIsolate() {
+  if (bigint_processor_) bigint_processor_->Destroy();
+}
 
 void LocalIsolate::RegisterDeserializerStarted() {
   return isolate_->RegisterDeserializerStarted();
@@ -46,6 +49,12 @@ bool LocalIsolate::is_collecting_type_profile() const {
   return isolate_->is_collecting_type_profile();
 }
 
+// Used for lazy initialization, based on an assumption that most
+// LocalIsolates won't be used to parse any BigInt literals.
+void LocalIsolate::InitializeBigIntProcessor() {
+  bigint_processor_ = bigint::Processor::New(new bigint::Platform());
+}
+
 // static
 bool StackLimitCheck::HasOverflowed(LocalIsolate* local_isolate) {
   return GetCurrentStackPosition() < local_isolate->stack_limit();

src/execution/local-isolate.h

@@ -15,6 +15,11 @@
 #include "src/heap/local-heap.h"
 
 namespace v8 {
+
+namespace bigint {
+class Processor;
+}
+
 namespace internal {
 
 class Isolate;
@@ -92,6 +97,10 @@ class V8_EXPORT_PRIVATE LocalIsolate final : private HiddenLocalFactory {
   ThreadId thread_id() const { return thread_id_; }
   Address stack_limit() const { return stack_limit_; }
   RuntimeCallStats* runtime_call_stats() const { return runtime_call_stats_; }
+  bigint::Processor* bigint_processor() {
+    if (!bigint_processor_) InitializeBigIntProcessor();
+    return bigint_processor_;
+  }
 
   bool is_main_thread() const { return heap_.is_main_thread(); }
 
@@ -106,6 +115,8 @@ class V8_EXPORT_PRIVATE LocalIsolate final : private HiddenLocalFactory {
  private:
   friend class v8::internal::LocalFactory;
 
+  void InitializeBigIntProcessor();
+
   LocalHeap heap_;
 
   // TODO(leszeks): Extract out the fields of the Isolate we want and store
@@ -117,6 +128,7 @@ class V8_EXPORT_PRIVATE LocalIsolate final : private HiddenLocalFactory {
   Address const stack_limit_;
 
   RuntimeCallStats* runtime_call_stats_;
+  bigint::Processor* bigint_processor_{nullptr};
 };
 
 template <base::MutexSharedType kIsShared>

src/objects/bigint.cc

@@ -126,10 +126,6 @@ class MutableBigInt : public FreshlyAllocatedBigInt {
       Isolate* isolate, Handle<BigIntBase> x, Handle<BigIntBase> y,
       MutableBigInt result_storage = MutableBigInt());
 
-  static void InternalMultiplyAdd(BigIntBase source, digit_t factor,
-                                  digit_t summand, int n, MutableBigInt result);
-  void InplaceMultiplyAdd(uintptr_t factor, uintptr_t summand);
-
   // Specialized helpers for shift operations.
   static MaybeHandle<BigInt> LeftShiftByAbsolute(Isolate* isolate,
                                                  Handle<BigIntBase> x,
@@ -152,7 +148,6 @@ class MutableBigInt : public FreshlyAllocatedBigInt {
   // Digit arithmetic helpers.
   static inline digit_t digit_add(digit_t a, digit_t b, digit_t* carry);
   static inline digit_t digit_sub(digit_t a, digit_t b, digit_t* borrow);
-  static inline digit_t digit_mul(digit_t a, digit_t b, digit_t* high);
   static inline bool digit_ismax(digit_t x) {
     return static_cast<digit_t>(~x) == 0;
   }
@@ -1411,50 +1406,6 @@ Handle<MutableBigInt> MutableBigInt::AbsoluteXor(Isolate* isolate,
                            [](digit_t a, digit_t b) { return a ^ b; });
 }
 
-// Multiplies {source} with {factor} and adds {summand} to the result.
-// {result} and {source} may be the same BigInt for inplace modification.
-void MutableBigInt::InternalMultiplyAdd(BigIntBase source, digit_t factor,
-                                        digit_t summand, int n,
-                                        MutableBigInt result) {
-  DCHECK(source.length() >= n);
-  DCHECK(result.length() >= n);
-  digit_t carry = summand;
-  digit_t high = 0;
-  for (int i = 0; i < n; i++) {
-    digit_t current = source.digit(i);
-    digit_t new_carry = 0;
-    // Compute this round's multiplication.
-    digit_t new_high = 0;
-    current = digit_mul(current, factor, &new_high);
-    // Add last round's carryovers.
-    current = digit_add(current, high, &new_carry);
-    current = digit_add(current, carry, &new_carry);
-    // Store result and prepare for next round.
-    result.set_digit(i, current);
-    carry = new_carry;
-    high = new_high;
-  }
-  if (result.length() > n) {
-    result.set_digit(n++, carry + high);
-    // Current callers don't pass in such large results, but let's be robust.
-    while (n < result.length()) {
-      result.set_digit(n++, 0);
-    }
-  } else {
-    CHECK_EQ(carry + high, 0);
-  }
-}
-
-// Multiplies {x} with {factor} and then adds {summand} to it.
-void BigInt::InplaceMultiplyAdd(FreshlyAllocatedBigInt x, uintptr_t factor,
-                                uintptr_t summand) {
-  STATIC_ASSERT(sizeof(factor) == sizeof(digit_t));
-  STATIC_ASSERT(sizeof(summand) == sizeof(digit_t));
-  MutableBigInt bigint = MutableBigInt::cast(x);
-  MutableBigInt::InternalMultiplyAdd(bigint, factor, summand, bigint.length(),
-                                     bigint);
-}
-
 MaybeHandle<BigInt> MutableBigInt::LeftShiftByAbsolute(Isolate* isolate,
                                                        Handle<BigIntBase> x,
                                                        Handle<BigIntBase> y) {
@@ -1591,71 +1542,33 @@ Maybe<BigInt::digit_t> MutableBigInt::ToShiftAmount(Handle<BigIntBase> x) {
   return Just(value);
 }
 
-// Lookup table for the maximum number of bits required per character of a
-// base-N string representation of a number. To increase accuracy, the array
-// value is the actual value multiplied by 32. To generate this table:
-// for (var i = 0; i <= 36; i++) { print(Math.ceil(Math.log2(i) * 32) + ","); }
-constexpr uint8_t kMaxBitsPerChar[] = {
-    0,   0,   32,  51,  64,  75,  83,  90,  96,  // 0..8
-    102, 107, 111, 115, 119, 122, 126, 128,      // 9..16
-    131, 134, 136, 139, 141, 143, 145, 147,      // 17..24
-    149, 151, 153, 154, 156, 158, 159, 160,      // 25..32
-    162, 163, 165, 166,                          // 33..36
-};
-
-static const int kBitsPerCharTableShift = 5;
-static const size_t kBitsPerCharTableMultiplier = 1u << kBitsPerCharTableShift;
+void Terminate(Isolate* isolate) { isolate->TerminateExecution(); }
+// {LocalIsolate} doesn't support interruption or termination.
+void Terminate(LocalIsolate* isolate) { UNREACHABLE(); }
 
 template <typename IsolateT>
-MaybeHandle<FreshlyAllocatedBigInt> BigInt::AllocateFor(
-    IsolateT* isolate, int radix, int charcount, ShouldThrow should_throw,
-    AllocationType allocation) {
-  DCHECK(2 <= radix && radix <= 36);
-  DCHECK_GE(charcount, 0);
-  size_t bits_per_char = kMaxBitsPerChar[radix];
-  uint64_t chars = static_cast<uint64_t>(charcount);
-  const int roundup = kBitsPerCharTableMultiplier - 1;
-  if (chars <=
-      (std::numeric_limits<uint64_t>::max() - roundup) / bits_per_char) {
-    uint64_t bits_min = bits_per_char * chars;
-    // Divide by 32 (see table), rounding up.
-    bits_min = (bits_min + roundup) >> kBitsPerCharTableShift;
-    if (bits_min <= static_cast<uint64_t>(kMaxInt)) {
-      // Divide by kDigitsBits, rounding up.
-      int length = static_cast<int>((bits_min + kDigitBits - 1) / kDigitBits);
-      if (length <= kMaxLength) {
-        Handle<MutableBigInt> result =
-            MutableBigInt::New(isolate, length, allocation).ToHandleChecked();
-        result->InitializeDigits(length);
-        return result;
-      }
-    }
-  }
-  // All the overflow/maximum checks above fall through to here.
-  if (should_throw == kThrowOnError) {
-    return ThrowBigIntTooBig<FreshlyAllocatedBigInt>(isolate);
-  } else {
-    return MaybeHandle<FreshlyAllocatedBigInt>();
+MaybeHandle<BigInt> BigInt::Allocate(IsolateT* isolate,
+                                     bigint::FromStringAccumulator* accumulator,
+                                     bool negative, AllocationType allocation) {
+  int digits = accumulator->ResultLength();
+  DCHECK_LE(digits, kMaxLength);
+  Handle<MutableBigInt> result =
+      MutableBigInt::New(isolate, digits, allocation).ToHandleChecked();
+  bigint::Status status =
+      isolate->bigint_processor()->FromString(GetRWDigits(result), accumulator);
+  if (status == bigint::Status::kInterrupted) {
+    Terminate(isolate);
+    return {};
   }
+  result->set_sign(negative);
+  return MutableBigInt::MakeImmutable(result);
 }
-template MaybeHandle<FreshlyAllocatedBigInt> BigInt::AllocateFor(
-    Isolate* isolate, int radix, int charcount, ShouldThrow should_throw,
-    AllocationType allocation);
-template MaybeHandle<FreshlyAllocatedBigInt> BigInt::AllocateFor(
-    LocalIsolate* isolate, int radix, int charcount, ShouldThrow should_throw,
-    AllocationType allocation);
-
-template <typename IsolateT>
-Handle<BigInt> BigInt::Finalize(Handle<FreshlyAllocatedBigInt> x, bool sign) {
-  Handle<MutableBigInt> bigint = Handle<MutableBigInt>::cast(x);
-  bigint->set_sign(sign);
-  return MutableBigInt::MakeImmutable<Isolate>(bigint);
-}
-
-template Handle<BigInt> BigInt::Finalize<Isolate>(
-    Handle<FreshlyAllocatedBigInt>, bool);
-template Handle<BigInt> BigInt::Finalize<LocalIsolate>(
-    Handle<FreshlyAllocatedBigInt>, bool);
+template MaybeHandle<BigInt> BigInt::Allocate(Isolate*,
+                                              bigint::FromStringAccumulator*,
+                                              bool, AllocationType);
+template MaybeHandle<BigInt> BigInt::Allocate(LocalIsolate*,
+                                              bigint::FromStringAccumulator*,
+                                              bool, AllocationType);
 
 // The serialization format MUST NOT CHANGE without updating the format
 // version in value-serializer.cc!
@@ -2056,43 +1969,6 @@ inline BigInt::digit_t MutableBigInt::digit_sub(digit_t a, digit_t b,
 #endif
 }
 
-// Returns the low half of the result. High half is in {high}.
-inline BigInt::digit_t MutableBigInt::digit_mul(digit_t a, digit_t b,
-                                                digit_t* high) {
-#if HAVE_TWODIGIT_T
-  twodigit_t result = static_cast<twodigit_t>(a) * static_cast<twodigit_t>(b);
-  *high = result >> kDigitBits;
-  return static_cast<digit_t>(result);
-#else
-  // Multiply in half-pointer-sized chunks.
-  // For inputs [AH AL]*[BH BL], the result is:
-  //
-  //            [AL*BL]  // r_low
-  //    +    [AL*BH]     // r_mid1
-  //    +    [AH*BL]     // r_mid2
-  //    + [AH*BH]        // r_high
-  //    = [R4 R3 R2 R1]  // high = [R4 R3], low = [R2 R1]
-  //
-  // Where of course we must be careful with carries between the columns.
-  digit_t a_low = a & kHalfDigitMask;
-  digit_t a_high = a >> kHalfDigitBits;
-  digit_t b_low = b & kHalfDigitMask;
-  digit_t b_high = b >> kHalfDigitBits;
-
-  digit_t r_low = a_low * b_low;
-  digit_t r_mid1 = a_low * b_high;
-  digit_t r_mid2 = a_high * b_low;
-  digit_t r_high = a_high * b_high;
-
-  digit_t carry = 0;
-  digit_t low = digit_add(r_low, r_mid1 << kHalfDigitBits, &carry);
-  low = digit_add(low, r_mid2 << kHalfDigitBits, &carry);
-  *high =
-      (r_mid1 >> kHalfDigitBits) + (r_mid2 >> kHalfDigitBits) + r_high + carry;
-  return low;
-#endif
-}
-
 #undef HAVE_TWODIGIT_T
 
 void MutableBigInt::set_64_bits(uint64_t bits) {

src/objects/bigint.h

@@ -14,6 +14,11 @@
 #include "src/objects/object-macros.h"
 
 namespace v8 {
+
+namespace bigint {
+class FromStringAccumulator;
+}  // namespace bigint
+
 namespace internal {
 
 void MutableBigInt_AbsoluteAddAndCanonicalize(Address result_addr,
@@ -252,13 +257,9 @@ class BigInt : public BigIntBase {
   static Handle<BigInt> Zero(
       IsolateT* isolate, AllocationType allocation = AllocationType::kYoung);
   template <typename IsolateT>
-  static MaybeHandle<FreshlyAllocatedBigInt> AllocateFor(
-      IsolateT* isolate, int radix, int charcount, ShouldThrow should_throw,
-      AllocationType allocation);
-  static void InplaceMultiplyAdd(FreshlyAllocatedBigInt x, uintptr_t factor,
-                                 uintptr_t summand);
-  template <typename IsolateT>
-  static Handle<BigInt> Finalize(Handle<FreshlyAllocatedBigInt> x, bool sign);
+  static MaybeHandle<BigInt> Allocate(
+      IsolateT* isolate, bigint::FromStringAccumulator* accumulator,
+      bool negative, AllocationType allocation);
 
   // Special functions for ValueSerializer/ValueDeserializer:
   uint32_t GetBitfieldForSerialization() const;

test/cctest/test-thread-termination.cc

@@ -239,6 +239,14 @@ TEST(TerminateBigIntToString) {
       "fail();");
 }
 
+TEST(TerminateBigIntFromString) {
+  TestTerminatingSlowOperation(
+      "var a = '12344567890'.repeat(10000);\n"
+      "terminate();\n"
+      "BigInt(a);\n"
+      "fail();\n");
+}
+
 int call_count = 0;