Optimize perf of EventBase with new option.

[folly.git] / folly / GroupVarint.h

diff --git a/folly/GroupVarint.h b/folly/GroupVarint.h
index e3b452ae8f5ddfd52eef771602ef706ba7a395ee..ff1c33c385da09eb4d5b61c9f617899a4cb4ea30 100644 (file)
--- a/folly/GroupVarint.h
+++ b/folly/GroupVarint.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2013 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.
@@ -21,15 +21,16 @@
 #error GroupVarint.h requires GCC
 #endif
 
-#if !defined(__x86_64__) && !defined(__i386__)
-#error GroupVarint.h requires x86_64 or i386
-#endif
+#include <folly/Portability.h>
+
+#if FOLLY_X64 || defined(__i386__)
+#define HAVE_GROUP_VARINT 1
 
 #include <cstdint>
 #include <limits>
-#include "folly/detail/GroupVarintDetail.h"
-#include "folly/Bits.h"
-#include "folly/Range.h"
+#include <folly/detail/GroupVarintDetail.h>
+#include <folly/Bits.h>
+#include <folly/Range.h>
 #include <glog/logging.h>
 
 #ifdef __SSSE3__
@@ -175,7 +176,7 @@ class GroupVarint<uint32_t> : public detail::GroupVarintBase<uint32_t> {
     p += k2+1;
     size_t k3 = b3key(k);
     *d = loadUnaligned<uint32_t>(p) & kMask[k3];
-    p += k3+1;
+    // p += k3+1;
     return end;
   }
 
@@ -188,6 +189,10 @@ class GroupVarint<uint32_t> : public detail::GroupVarintBase<uint32_t> {
   }
 
 #ifdef __SSSE3__
+  /**
+   * Just like the non-SSSE3 decode below, but with the additional constraint
+   * that we must be able to read at least 17 bytes from the input pointer, p.
+   */
   static const char* decode(const char* p, uint32_t* dest) {
     uint8_t key = p[0];
     __m128i val = _mm_loadu_si128((const __m128i*)(p+1));
@@ -197,6 +202,10 @@ class GroupVarint<uint32_t> : public detail::GroupVarintBase<uint32_t> {
     return p + detail::groupVarintLengths[key];
   }
 
+  /**
+   * Just like decode_simple, but with the additional constraint that
+   * we must be able to read at least 17 bytes from the input pointer, p.
+   */
   static const char* decode(const char* p, uint32_t* a, uint32_t* b,
                             uint32_t* c, uint32_t* d) {
     uint8_t key = p[0];
@@ -510,15 +519,17 @@ class GroupVarintDecoder {
     : rrest_(data.end()),
       p_(data.data()),
       end_(data.end()),
+      limit_(end_),
       pos_(0),
       count_(0),
       remaining_(maxCount) {
   }
 
-  void reset(StringPiece data, size_t maxCount=(size_t)-1) {
+  void reset(StringPiece data, size_t maxCount = (size_t)-1) {
     rrest_ = data.end();
     p_ = data.data();
     end_ = data.end();
+    limit_ = end_;
     pos_ = 0;
     count_ = 0;
     remaining_ = maxCount;
@@ -541,10 +552,11 @@ class GroupVarintDecoder {
       // The best way to ensure this is to ensure that data has at least
       // Base::kMaxSize - 1 bytes readable *after* the end, otherwise we'll copy
       // into a temporary buffer.
-      if (rem < Base::kMaxSize) {
+      if (limit_ - p_ < Base::kMaxSize) {
         memcpy(tmp_, p_, rem);
         p_ = tmp_;
         end_ = p_ + rem;
+        limit_ = tmp_ + sizeof(tmp_);
       }
       pos_ = 0;
       const char* n = Base::decode(p_, buf_);
@@ -592,7 +604,8 @@ class GroupVarintDecoder {
   const char* rrest_;
   const char* p_;
   const char* end_;
-  char tmp_[Base::kMaxSize];
+  const char* limit_;
+  char tmp_[2 * Base::kMaxSize];
   type buf_[Base::kGroupSize];
   size_t pos_;
   size_t count_;
@@ -604,5 +617,5 @@ typedef GroupVarintDecoder<uint64_t> GroupVarint64Decoder;
 
 }  // namespace folly
 
+#endif /* FOLLY_X64 || defined(__i386__) */
 #endif /* FOLLY_GROUPVARINT_H_ */
-