/*
- * Copyright 2012 Facebook, Inc.
+ * Copyright 2014 Facebook, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* limitations under the License.
*/
-//
// @author Mark Rabkin (mrabkin@fb.com)
// @author Andrei Alexandrescu (andrei.alexandrescu@fb.com)
-//
-#include "folly/Range.h"
+#include <folly/Range.h>
+
+#if FOLLY_HAVE_EMMINTRIN_H
+#include <emmintrin.h> // __v16qi
+#endif
+#include <iostream>
namespace folly {
/**
-Predicates that can be used with qfind and startsWith
+ * Predicates that can be used with qfind and startsWith
*/
const AsciiCaseSensitive asciiCaseSensitive = AsciiCaseSensitive();
const AsciiCaseInsensitive asciiCaseInsensitive = AsciiCaseInsensitive();
-std::ostream& operator<<(std::ostream& os, const StringPiece& piece) {
+std::ostream& operator<<(std::ostream& os, const StringPiece piece) {
+ os.write(piece.start(), piece.size());
+ return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const MutableStringPiece piece) {
os.write(piece.start(), piece.size());
return os;
}
-} // namespace folly
+namespace detail {
+
+size_t qfind_first_byte_of_memchr(const StringPiece haystack,
+ const StringPiece needles) {
+ size_t best = haystack.size();
+ for (char needle: needles) {
+ const void* ptr = memchr(haystack.data(), needle, best);
+ if (ptr) {
+ auto found = static_cast<const char*>(ptr) - haystack.data();
+ best = std::min<size_t>(best, found);
+ }
+ }
+ if (best == haystack.size()) {
+ return StringPiece::npos;
+ }
+ return best;
+}
+
+} // namespace detail
+
+namespace {
+
+// It's okay if pages are bigger than this (as powers of two), but they should
+// not be smaller.
+constexpr size_t kMinPageSize = 4096;
+static_assert(kMinPageSize >= 16,
+ "kMinPageSize must be at least SSE register size");
+#define PAGE_FOR(addr) \
+ (reinterpret_cast<uintptr_t>(addr) / kMinPageSize)
+
+
+// Earlier versions of GCC (for example, Clang on Mac OS X, which is based on
+// GCC 4.2) do not have a full compliment of SSE builtins.
+#if FOLLY_HAVE_EMMINTRIN_H && __GNUC_PREREQ(4, 6)
+inline size_t nextAlignedIndex(const char* arr) {
+ auto firstPossible = reinterpret_cast<uintptr_t>(arr) + 1;
+ return 1 + // add 1 because the index starts at 'arr'
+ ((firstPossible + 15) & ~0xF) // round up to next multiple of 16
+ - firstPossible;
+}
+
+// build sse4.2-optimized version even if -msse4.2 is not passed to GCC
+size_t qfind_first_byte_of_needles16(const StringPiece haystack,
+ const StringPiece needles)
+ __attribute__ ((__target__("sse4.2"), noinline))
+ FOLLY_DISABLE_ADDRESS_SANITIZER;
+
+// helper method for case where needles.size() <= 16
+size_t qfind_first_byte_of_needles16(const StringPiece haystack,
+ const StringPiece needles) {
+ DCHECK(!haystack.empty());
+ DCHECK(!needles.empty());
+ DCHECK_LE(needles.size(), 16);
+ // benchmarking shows that memchr beats out SSE for small needle-sets
+ // with large haystacks.
+ if ((needles.size() <= 2 && haystack.size() >= 256) ||
+ // must bail if we can't even SSE-load a single segment of haystack
+ (haystack.size() < 16 &&
+ PAGE_FOR(haystack.end() - 1) != PAGE_FOR(haystack.data() + 15)) ||
+ // can't load needles into SSE register if it could cross page boundary
+ PAGE_FOR(needles.end() - 1) != PAGE_FOR(needles.data() + 15)) {
+ return detail::qfind_first_byte_of_memchr(haystack, needles);
+ }
+
+ auto arr2 = __builtin_ia32_loaddqu(needles.data());
+ // do an unaligned load for first block of haystack
+ auto arr1 = __builtin_ia32_loaddqu(haystack.data());
+ auto index = __builtin_ia32_pcmpestri128(arr2, needles.size(),
+ arr1, haystack.size(), 0);
+ if (index < 16) {
+ return index;
+ }
+
+ // Now, we can do aligned loads hereafter...
+ size_t i = nextAlignedIndex(haystack.data());
+ for (; i < haystack.size(); i+= 16) {
+ void* ptr1 = __builtin_assume_aligned(haystack.data() + i, 16);
+ auto arr1 = *reinterpret_cast<const __v16qi*>(ptr1);
+ auto index = __builtin_ia32_pcmpestri128(arr2, needles.size(),
+ arr1, haystack.size() - i, 0);
+ if (index < 16) {
+ return i + index;
+ }
+ }
+ return StringPiece::npos;
+}
+#endif // FOLLY_HAVE_EMMINTRIN_H && GCC 4.6+
+
+// Aho, Hopcroft, and Ullman refer to this trick in "The Design and Analysis
+// of Computer Algorithms" (1974), but the best description is here:
+// http://research.swtch.com/sparse
+class FastByteSet {
+ public:
+ FastByteSet() : size_(0) { } // no init of arrays required!
+
+ inline void add(uint8_t i) {
+ if (!contains(i)) {
+ dense_[size_] = i;
+ sparse_[i] = size_;
+ size_++;
+ }
+ }
+ inline bool contains(uint8_t i) const {
+ DCHECK_LE(size_, 256);
+ return sparse_[i] < size_ && dense_[sparse_[i]] == i;
+ }
+
+ private:
+ uint16_t size_; // can't use uint8_t because it would overflow if all
+ // possible values were inserted.
+ uint8_t sparse_[256];
+ uint8_t dense_[256];
+};
+
+} // namespace
+
+namespace detail {
+
+size_t qfind_first_byte_of_byteset(const StringPiece haystack,
+ const StringPiece needles) {
+ FastByteSet s;
+ for (auto needle: needles) {
+ s.add(needle);
+ }
+ for (size_t index = 0; index < haystack.size(); ++index) {
+ if (s.contains(haystack[index])) {
+ return index;
+ }
+ }
+ return StringPiece::npos;
+}
+
+#if FOLLY_HAVE_EMMINTRIN_H && __GNUC_PREREQ(4, 6)
+
+template <bool HAYSTACK_ALIGNED>
+size_t scanHaystackBlock(const StringPiece haystack,
+ const StringPiece needles,
+ int64_t idx)
+// inline is okay because it's only called from other sse4.2 functions
+ __attribute__ ((__target__("sse4.2")))
+// Turn off ASAN because the "arr2 = ..." assignment in the loop below reads
+// up to 15 bytes beyond end of the buffer in #needles#. That is ok because
+// ptr2 is always 16-byte aligned, so the read can never span a page boundary.
+// Also, the extra data that may be read is never actually used.
+ FOLLY_DISABLE_ADDRESS_SANITIZER;
+
+// Scans a 16-byte block of haystack (starting at blockStartIdx) to find first
+// needle. If HAYSTACK_ALIGNED, then haystack must be 16byte aligned.
+// If !HAYSTACK_ALIGNED, then caller must ensure that it is safe to load the
+// block.
+template <bool HAYSTACK_ALIGNED>
+size_t scanHaystackBlock(const StringPiece haystack,
+ const StringPiece needles,
+ int64_t blockStartIdx) {
+ DCHECK_GT(needles.size(), 16); // should handled by *needles16() method
+ DCHECK(blockStartIdx + 16 <= haystack.size() ||
+ (PAGE_FOR(haystack.data() + blockStartIdx) ==
+ PAGE_FOR(haystack.data() + blockStartIdx + 15)));
+
+ __v16qi arr1;
+ if (HAYSTACK_ALIGNED) {
+ void* ptr1 = __builtin_assume_aligned(haystack.data() + blockStartIdx, 16);
+ arr1 = *reinterpret_cast<const __v16qi*>(ptr1);
+ } else {
+ arr1 = __builtin_ia32_loaddqu(haystack.data() + blockStartIdx);
+ }
+
+ // This load is safe because needles.size() >= 16
+ auto arr2 = __builtin_ia32_loaddqu(needles.data());
+ size_t b = __builtin_ia32_pcmpestri128(
+ arr2, 16, arr1, haystack.size() - blockStartIdx, 0);
+
+ size_t j = nextAlignedIndex(needles.data());
+ for (; j < needles.size(); j += 16) {
+ void* ptr2 = __builtin_assume_aligned(needles.data() + j, 16);
+ arr2 = *reinterpret_cast<const __v16qi*>(ptr2);
+
+ auto index = __builtin_ia32_pcmpestri128(
+ arr2, needles.size() - j, arr1, haystack.size() - blockStartIdx, 0);
+ b = std::min<size_t>(index, b);
+ }
+
+ if (b < 16) {
+ return blockStartIdx + b;
+ }
+ return StringPiece::npos;
+}
+
+size_t qfind_first_byte_of_sse42(const StringPiece haystack,
+ const StringPiece needles)
+ __attribute__ ((__target__("sse4.2"), noinline));
+
+size_t qfind_first_byte_of_sse42(const StringPiece haystack,
+ const StringPiece needles) {
+ if (UNLIKELY(needles.empty() || haystack.empty())) {
+ return StringPiece::npos;
+ } else if (needles.size() <= 16) {
+ // we can save some unnecessary load instructions by optimizing for
+ // the common case of needles.size() <= 16
+ return qfind_first_byte_of_needles16(haystack, needles);
+ }
+
+ if (haystack.size() < 16 &&
+ PAGE_FOR(haystack.end() - 1) != PAGE_FOR(haystack.data() + 16)) {
+ // We can't safely SSE-load haystack. Use a different approach.
+ if (haystack.size() <= 2) {
+ return qfind_first_of(haystack, needles, asciiCaseSensitive);
+ }
+ return qfind_first_byte_of_byteset(haystack, needles);
+ }
+
+ auto ret = scanHaystackBlock<false>(haystack, needles, 0);
+ if (ret != StringPiece::npos) {
+ return ret;
+ }
+
+ size_t i = nextAlignedIndex(haystack.data());
+ for (; i < haystack.size(); i += 16) {
+ auto ret = scanHaystackBlock<true>(haystack, needles, i);
+ if (ret != StringPiece::npos) {
+ return ret;
+ }
+ }
+
+ return StringPiece::npos;
+}
+#endif // FOLLY_HAVE_EMMINTRIN_H && GCC 4.6+
+
+size_t qfind_first_byte_of_nosse(const StringPiece haystack,
+ const StringPiece needles) {
+ if (UNLIKELY(needles.empty() || haystack.empty())) {
+ return StringPiece::npos;
+ }
+ // The thresholds below were empirically determined by benchmarking.
+ // This is not an exact science since it depends on the CPU, the size of
+ // needles, and the size of haystack.
+ if (haystack.size() == 1 ||
+ (haystack.size() < 4 && needles.size() <= 16)) {
+ return qfind_first_of(haystack, needles, asciiCaseSensitive);
+ } else if ((needles.size() >= 4 && haystack.size() <= 10) ||
+ (needles.size() >= 16 && haystack.size() <= 64) ||
+ needles.size() >= 32) {
+ return qfind_first_byte_of_byteset(haystack, needles);
+ }
+
+ return qfind_first_byte_of_memchr(haystack, needles);
+}
+
+} // namespace detail
+} // namespace folly