Revert "HADOOP-11505. Various native parts use bswap incorrectly and unportably (Alan Burlison via aw)"
This reverts commit 6725e7f1be
.
This commit is contained in:
parent
2577e5b2c1
commit
c37c3f41b3
|
@ -537,9 +537,6 @@ Trunk (Unreleased)
|
|||
|
||||
HADOOP-12553. [JDK8] Fix javadoc error caused by illegal tag. (aajisaka)
|
||||
|
||||
HADOOP-11505. Various native parts use bswap incorrectly and unportably
|
||||
(Alan Burlison via aw)
|
||||
|
||||
OPTIMIZATIONS
|
||||
|
||||
HADOOP-7761. Improve the performance of raw comparisons. (todd)
|
||||
|
|
|
@ -41,131 +41,6 @@ endif()
|
|||
# Configure JNI.
|
||||
include(HadoopJNI)
|
||||
|
||||
#
|
||||
# Endian configuration, as per http://austingroupbugs.net/view.php?id=162#c665
|
||||
#
|
||||
|
||||
# Work out the endianness, set header macro values.
|
||||
include(TestBigEndian)
|
||||
include(CheckIncludeFile)
|
||||
include(CheckSymbolExists)
|
||||
test_big_endian(_bigendian)
|
||||
if(_bigendian)
|
||||
set(HADOOP_BYTE_ORDER "HADOOP_BIG_ENDIAN")
|
||||
else()
|
||||
set(HADOOP_BYTE_ORDER "HADOOP_LITTLE_ENDIAN")
|
||||
endif()
|
||||
|
||||
# Linux, NetBSD, FreeBSD and OpenBSD all provide htoXXX definitions in endian.h or sys/endian.h.
|
||||
check_include_file("endian.h" _endian_h)
|
||||
if (_endian_h)
|
||||
set(HADOOP_ENDIAN_H "endian.h")
|
||||
else()
|
||||
check_include_file("sys/endian.h" _sys_endian_h)
|
||||
if (_sys_endian_h)
|
||||
set(HADOOP_ENDIAN_H "sys/endian.h")
|
||||
endif()
|
||||
endif()
|
||||
if(DEFINED HADOOP_ENDIAN_H)
|
||||
check_symbol_exists("be64toh" ${HADOOP_ENDIAN_H} _be64toh)
|
||||
if( _be64toh)
|
||||
set(HADOOP_HTOBE16 "htobe16")
|
||||
set(HADOOP_HTOLE16 "htole16")
|
||||
set(HADOOP_BE16TOH "be16toh")
|
||||
set(HADOOP_LE16TOH "le16toh")
|
||||
set(HADOOP_HTOBE32 "htobe32")
|
||||
set(HADOOP_HTOLE32 "htole32")
|
||||
set(HADOOP_BE32TOH "be32toh")
|
||||
set(HADOOP_LE32TOH "le32toh")
|
||||
set(HADOOP_HTOBE64 "htobe64")
|
||||
set(HADOOP_HTOLE64 "htole64")
|
||||
set(HADOOP_BE64TOH "be64toh")
|
||||
set(HADOOP_LE64TOH "le64toh")
|
||||
set(_have_endian TRUE)
|
||||
unset(_be64toh)
|
||||
else()
|
||||
message(FATAL_ERROR "endian.h located but doesn't contain be64toh")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
# Solaris doesn't provide htoXXX, we have to provide alternatives.
|
||||
if(NOT _have_endian)
|
||||
check_include_file("sys/byteorder.h" _sys_byteorder_h)
|
||||
if(_sys_byteorder_h)
|
||||
set(HADOOP_ENDIAN_H "sys/byteorder.h")
|
||||
check_symbol_exists("BSWAP_64" ${HADOOP_ENDIAN_H} _bswap_64)
|
||||
endif()
|
||||
if(_sys_byteorder_h AND _bswap_64)
|
||||
if(_bigendian)
|
||||
set(HADOOP_HTOBE16 "")
|
||||
set(HADOOP_HTOLE16 "BSWAP_16")
|
||||
set(HADOOP_BE16TOH "")
|
||||
set(HADOOP_LE16TOH "BSWAP_16")
|
||||
set(HADOOP_HTOBE32 "")
|
||||
set(HADOOP_HTOLE32 "BSWAP_32")
|
||||
set(HADOOP_BE32TOH "")
|
||||
set(HADOOP_LE32TOH "BSWAP_32")
|
||||
set(HADOOP_HTOBE64 "")
|
||||
set(HADOOP_HTOLE64 "BSWAP_64")
|
||||
set(HADOOP_BE64TOH "")
|
||||
set(HADOOP_LE64TOH "BSWAP_64")
|
||||
else()
|
||||
set(HADOOP_HTOBE16 "BSWAP_16")
|
||||
set(HADOOP_HTOLE16 "")
|
||||
set(HADOOP_BE16TOH "BSWAP_16")
|
||||
set(HADOOP_LE16TOH "")
|
||||
set(HADOOP_HTOBE32 "BSWAP_32")
|
||||
set(HADOOP_HTOLE32 "")
|
||||
set(HADOOP_BE32TOH "BSWAP_32")
|
||||
set(HADOOP_LE32TOH "")
|
||||
set(HADOOP_HTOBE64 "BSWAP_64")
|
||||
set(HADOOP_HTOLE64 "")
|
||||
set(HADOOP_BE64TOH "BSWAP_64")
|
||||
set(HADOOP_LE64TOH "")
|
||||
endif()
|
||||
set(_have_endian TRUE)
|
||||
unset(_sys_byteorder_h)
|
||||
unset(_bswap_64)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
# OSX uses libkern/OSByteOrder.h and OSSwapXtoY.
|
||||
if(NOT _have_endian)
|
||||
check_include_file("libkern/OSByteOrder.h" _libkern_osbyteorder_h)
|
||||
if(_libkern_osbyteorder_h)
|
||||
set(HADOOP_ENDIAN_H "libkern/OSByteOrder.h")
|
||||
check_symbol_exists("OSSwapHostToLittleInt64" ${HADOOP_ENDIAN_H} _osswaphosttolittleint64)
|
||||
endif()
|
||||
if(_libkern_osbyteorder_h AND _osswaphosttolittleint64)
|
||||
set(HADOOP_HTOBE16 "OSSwapHostToBigInt16")
|
||||
set(HADOOP_HTOLE16 "OSSwapHostToLittleInt16")
|
||||
set(HADOOP_BE16TOH "OSSwapBigToHostInt16")
|
||||
set(HADOOP_LE16TOH "OSSwapLittleToHostInt16")
|
||||
set(HADOOP_HTOBE32 "OSSwapHostToBigInt32")
|
||||
set(HADOOP_HTOLE32 "OSSwapHostToLittleInt32")
|
||||
set(HADOOP_BE32TOH "OSSwapBigToHostInt32")
|
||||
set(HADOOP_LE32TOH "OSSwapLittleToHostInt32")
|
||||
set(HADOOP_HTOBE64 "OSSwapHostToBigInt64")
|
||||
set(HADOOP_HTOLE64 "OSSwapHostToLittleInt64")
|
||||
set(HADOOP_BE64TOH "OSSwapBigToHostInt64")
|
||||
set(HADOOP_LE64TOH "OSSwapLittleToHostInt64")
|
||||
set(_have_endian TRUE)
|
||||
unset(_libkern_osbyteorder_h)
|
||||
unset(_osswaphosttolittleint64)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
# Bail if we don't know the endian definitions for this platform.
|
||||
if(NOT _have_endian)
|
||||
message(FATAL_ERROR "Can't provide endianness definitions for this platform")
|
||||
endif()
|
||||
|
||||
# Configure the hadoop_endian.h header file.
|
||||
configure_file(${CMAKE_SOURCE_DIR}/hadoop_endian.h.cmake ${CMAKE_BINARY_DIR}/hadoop_endian.h)
|
||||
unset(_bigendian)
|
||||
unset(_have_endian)
|
||||
unset(HADOOP_ENDIAN_H)
|
||||
|
||||
# Require zlib.
|
||||
set(STORED_CMAKE_FIND_LIBRARY_SUFFIXES ${CMAKE_FIND_LIBRARY_SUFFIXES})
|
||||
hadoop_set_find_shared_library_version("1")
|
||||
|
|
|
@ -37,7 +37,6 @@
|
|||
#include "crc32c_tables.h"
|
||||
#include "bulk_crc32.h"
|
||||
#include "gcc_optimizations.h"
|
||||
#include "hadoop_endian.h"
|
||||
|
||||
#define CRC_INITIAL_VAL 0xffffffff
|
||||
|
||||
|
@ -164,7 +163,7 @@ static uint32_t crc32c_sb8(uint32_t crc, const uint8_t *buf, size_t length) {
|
|||
for (li=0; li < running_length/8; li++) {
|
||||
uint32_t term1;
|
||||
uint32_t term2;
|
||||
crc ^= hadoop_htole32(*(uint32_t *)buf);
|
||||
crc ^= *(uint32_t *)buf;
|
||||
buf += 4;
|
||||
term1 = CRC32C_T8_7[crc & 0x000000FF] ^
|
||||
CRC32C_T8_6[(crc >> 8) & 0x000000FF];
|
||||
|
@ -172,10 +171,10 @@ static uint32_t crc32c_sb8(uint32_t crc, const uint8_t *buf, size_t length) {
|
|||
crc = term1 ^
|
||||
CRC32C_T8_5[term2 & 0x000000FF] ^
|
||||
CRC32C_T8_4[(term2 >> 8) & 0x000000FF];
|
||||
term1 = CRC32C_T8_3[hadoop_htole32(*(uint32_t *)buf) & 0x000000FF] ^
|
||||
CRC32C_T8_2[(hadoop_htole32(*(uint32_t *)buf) >> 8) & 0x000000FF];
|
||||
term1 = CRC32C_T8_3[(*(uint32_t *)buf) & 0x000000FF] ^
|
||||
CRC32C_T8_2[((*(uint32_t *)buf) >> 8) & 0x000000FF];
|
||||
|
||||
term2 = hadoop_htole32((*(uint32_t *)buf)) >> 16;
|
||||
term2 = (*(uint32_t *)buf) >> 16;
|
||||
crc = crc ^
|
||||
term1 ^
|
||||
CRC32C_T8_1[term2 & 0x000000FF] ^
|
||||
|
@ -210,7 +209,7 @@ static uint32_t crc32_zlib_sb8(
|
|||
for (li=0; li < running_length/8; li++) {
|
||||
uint32_t term1;
|
||||
uint32_t term2;
|
||||
crc ^= hadoop_htole32(*(uint32_t *)buf);
|
||||
crc ^= *(uint32_t *)buf;
|
||||
buf += 4;
|
||||
term1 = CRC32_T8_7[crc & 0x000000FF] ^
|
||||
CRC32_T8_6[(crc >> 8) & 0x000000FF];
|
||||
|
@ -218,10 +217,10 @@ static uint32_t crc32_zlib_sb8(
|
|||
crc = term1 ^
|
||||
CRC32_T8_5[term2 & 0x000000FF] ^
|
||||
CRC32_T8_4[(term2 >> 8) & 0x000000FF];
|
||||
term1 = CRC32_T8_3[hadoop_htole32(*(uint32_t *)buf) & 0x000000FF] ^
|
||||
CRC32_T8_2[(hadoop_htole32(*(uint32_t *)buf) >> 8) & 0x000000FF];
|
||||
term1 = CRC32_T8_3[(*(uint32_t *)buf) & 0x000000FF] ^
|
||||
CRC32_T8_2[((*(uint32_t *)buf) >> 8) & 0x000000FF];
|
||||
|
||||
term2 = hadoop_htole32(*(uint32_t *)buf) >> 16;
|
||||
term2 = (*(uint32_t *)buf) >> 16;
|
||||
crc = crc ^
|
||||
term1 ^
|
||||
CRC32_T8_1[term2 & 0x000000FF] ^
|
||||
|
|
|
@ -90,7 +90,6 @@ include_directories(
|
|||
${SRC}/src/util
|
||||
${SRC}/src/lib
|
||||
${SRC}/test
|
||||
../../../../hadoop-common-project/hadoop-common/target/native
|
||||
${CMAKE_CURRENT_SOURCE_DIR}
|
||||
${CMAKE_BINARY_DIR}
|
||||
${JNI_INCLUDE_DIRS}
|
||||
|
|
|
@ -42,8 +42,8 @@ import org.apache.hadoop.classification.InterfaceStability;
|
|||
* <code>
|
||||
* int HivePlatform::HiveKeyComparator(const char * src, uint32_t srcLength,
|
||||
* const char * dest, uint32_t destLength) {
|
||||
* uint32_t sl = hadoop_be32toh(*(uint32_t*)src);
|
||||
* uint32_t dl = hadoop_be32toh(*(uint32_t*)dest);
|
||||
* uint32_t sl = bswap(*(uint32_t*)src);
|
||||
* uint32_t dl = bswap(*(uint32_t*)dest);
|
||||
* return NativeObjectFactory::BytesComparator(src + 4, sl, dest + 4, dl);
|
||||
* }
|
||||
* </code>
|
||||
|
|
|
@ -40,6 +40,15 @@ enum NativeObjectType {
|
|||
BatchHandlerType = 1,
|
||||
};
|
||||
|
||||
/**
|
||||
* Enduim setting
|
||||
*
|
||||
*/
|
||||
enum Endium {
|
||||
LITTLE_ENDIUM = 0,
|
||||
LARGE_ENDIUM = 1
|
||||
};
|
||||
|
||||
#define NATIVE_COMBINER "native.combiner.class"
|
||||
#define NATIVE_PARTITIONER "native.partitioner.class"
|
||||
#define NATIVE_MAPPER "native.mapper.class"
|
||||
|
|
|
@ -104,8 +104,8 @@ int32_t BlockDecompressStream::read(void * buff, uint32_t length) {
|
|||
THROW_EXCEPTION(IOException, "readFully get incomplete data");
|
||||
}
|
||||
_compressedBytesRead += rd;
|
||||
sizes[0] = hadoop_be32toh(sizes[0]);
|
||||
sizes[1] = hadoop_be32toh(sizes[1]);
|
||||
sizes[0] = bswap(sizes[0]);
|
||||
sizes[1] = bswap(sizes[1]);
|
||||
if (sizes[0] <= length) {
|
||||
uint32_t len = decompressOneBlock(sizes[1], buff, sizes[0]);
|
||||
if (len != sizes[0]) {
|
||||
|
|
|
@ -38,8 +38,8 @@ void Lz4CompressStream::compressOneBlock(const void * buff, uint32_t length) {
|
|||
int ret = LZ4_compress((char*)buff, _tempBuffer + 8, length);
|
||||
if (ret > 0) {
|
||||
compressedLength = ret;
|
||||
((uint32_t*)_tempBuffer)[0] = hadoop_be32toh(length);
|
||||
((uint32_t*)_tempBuffer)[1] = hadoop_be32toh((uint32_t)compressedLength);
|
||||
((uint32_t*)_tempBuffer)[0] = bswap(length);
|
||||
((uint32_t*)_tempBuffer)[1] = bswap((uint32_t)compressedLength);
|
||||
_stream->write(_tempBuffer, compressedLength + 8);
|
||||
_compressedBytesWritten += (compressedLength + 8);
|
||||
} else {
|
||||
|
|
|
@ -37,8 +37,8 @@ void SnappyCompressStream::compressOneBlock(const void * buff, uint32_t length)
|
|||
snappy_status ret = snappy_compress((const char*)buff, length, _tempBuffer + 8,
|
||||
&compressedLength);
|
||||
if (ret == SNAPPY_OK) {
|
||||
((uint32_t*)_tempBuffer)[0] = hadoop_be32toh(length);
|
||||
((uint32_t*)_tempBuffer)[1] = hadoop_be32toh((uint32_t)compressedLength);
|
||||
((uint32_t*)_tempBuffer)[0] = bswap(length);
|
||||
((uint32_t*)_tempBuffer)[1] = bswap((uint32_t)compressedLength);
|
||||
_stream->write(_tempBuffer, compressedLength + 8);
|
||||
_compressedBytesWritten += (compressedLength + 8);
|
||||
} else if (ret == SNAPPY_INVALID_INPUT) {
|
||||
|
|
|
@ -15,7 +15,6 @@
|
|||
* See the License for the specific language governing permissions and
|
||||
* limitations under the License.
|
||||
*/
|
||||
|
||||
#include "CombineHandler.h"
|
||||
|
||||
namespace NativeTask {
|
||||
|
@ -49,8 +48,8 @@ uint32_t CombineHandler::feedDataToJavaInWritableSerialization() {
|
|||
|
||||
if (_kvCached) {
|
||||
uint32_t kvLength = _key.outerLength + _value.outerLength + KVBuffer::headerLength();
|
||||
outputInt(hadoop_be32toh(_key.outerLength));
|
||||
outputInt(hadoop_be32toh(_value.outerLength));
|
||||
outputInt(bswap(_key.outerLength));
|
||||
outputInt(bswap(_value.outerLength));
|
||||
outputKeyOrValue(_key, _kType);
|
||||
outputKeyOrValue(_value, _vType);
|
||||
|
||||
|
@ -74,8 +73,8 @@ uint32_t CombineHandler::feedDataToJavaInWritableSerialization() {
|
|||
} else {
|
||||
firstKV = false;
|
||||
//write final key length and final value length
|
||||
outputInt(hadoop_be32toh(_key.outerLength));
|
||||
outputInt(hadoop_be32toh(_value.outerLength));
|
||||
outputInt(bswap(_key.outerLength));
|
||||
outputInt(bswap(_value.outerLength));
|
||||
outputKeyOrValue(_key, _kType);
|
||||
outputKeyOrValue(_value, _vType);
|
||||
|
||||
|
@ -102,7 +101,7 @@ void CombineHandler::outputKeyOrValue(SerializeInfo & KV, KeyValueType type) {
|
|||
output(KV.buffer.data(), KV.buffer.length());
|
||||
break;
|
||||
case BytesType:
|
||||
outputInt(hadoop_be32toh(KV.buffer.length()));
|
||||
outputInt(bswap(KV.buffer.length()));
|
||||
output(KV.buffer.data(), KV.buffer.length());
|
||||
break;
|
||||
default:
|
||||
|
@ -203,8 +202,8 @@ void CombineHandler::write(char * buf, uint32_t length) {
|
|||
uint32_t outputRecordCount = 0;
|
||||
while (remain > 0) {
|
||||
kv = (KVBuffer *)pos;
|
||||
kv->keyLength = hadoop_be32toh(kv->keyLength);
|
||||
kv->valueLength = hadoop_be32toh(kv->valueLength);
|
||||
kv->keyLength = bswap(kv->keyLength);
|
||||
kv->valueLength = bswap(kv->valueLength);
|
||||
_writer->write(kv->getKey(), kv->keyLength, kv->getValue(), kv->valueLength);
|
||||
outputRecordCount++;
|
||||
remain -= kv->length();
|
||||
|
|
|
@ -30,7 +30,7 @@ using std::vector;
|
|||
namespace NativeTask {
|
||||
|
||||
MCollectorOutputHandler::MCollectorOutputHandler()
|
||||
: _collector(NULL), _dest(NULL) {
|
||||
: _collector(NULL), _dest(NULL), _endium(LARGE_ENDIUM) {
|
||||
}
|
||||
|
||||
MCollectorOutputHandler::~MCollectorOutputHandler() {
|
||||
|
@ -73,9 +73,11 @@ void MCollectorOutputHandler::handleInput(ByteBuffer & in) {
|
|||
THROW_EXCEPTION(IOException, "k/v meta information incomplete");
|
||||
}
|
||||
|
||||
kvBuffer->partitionId = hadoop_be32toh(kvBuffer->partitionId);
|
||||
kvBuffer->buffer.keyLength = hadoop_be32toh(kvBuffer->buffer.keyLength);
|
||||
kvBuffer->buffer.valueLength = hadoop_be32toh(kvBuffer->buffer.valueLength);
|
||||
if (_endium == LARGE_ENDIUM) {
|
||||
kvBuffer->partitionId = bswap(kvBuffer->partitionId);
|
||||
kvBuffer->buffer.keyLength = bswap(kvBuffer->buffer.keyLength);
|
||||
kvBuffer->buffer.valueLength = bswap(kvBuffer->buffer.valueLength);
|
||||
}
|
||||
|
||||
uint32_t kvLength = kvBuffer->buffer.length();
|
||||
|
||||
|
|
|
@ -35,6 +35,8 @@ private:
|
|||
// state info for large KV pairs
|
||||
char * _dest;
|
||||
|
||||
Endium _endium;
|
||||
|
||||
public:
|
||||
MCollectorOutputHandler();
|
||||
virtual ~MCollectorOutputHandler();
|
||||
|
|
|
@ -115,7 +115,7 @@ public:
|
|||
* read uint32_t big endian
|
||||
*/
|
||||
inline uint32_t read_uint32_be() {
|
||||
return hadoop_be32toh(read_uint32_le());
|
||||
return bswap(read_uint32_le());
|
||||
}
|
||||
};
|
||||
|
||||
|
@ -198,7 +198,7 @@ public:
|
|||
}
|
||||
|
||||
inline void write_uint32_be(uint32_t v) {
|
||||
write_uint32_le(hadoop_be32toh(v));
|
||||
write_uint32_le(bswap(v));
|
||||
}
|
||||
|
||||
inline void write_uint64_le(uint64_t v) {
|
||||
|
@ -211,7 +211,7 @@ public:
|
|||
}
|
||||
|
||||
inline void write_uint64_be(uint64_t v) {
|
||||
write_uint64_le(hadoop_be64toh(v));
|
||||
write_uint64_le(bswap64(v));
|
||||
}
|
||||
|
||||
inline void write_vlong(int64_t v) {
|
||||
|
@ -278,11 +278,12 @@ struct KVBuffer {
|
|||
}
|
||||
|
||||
uint32_t length() {
|
||||
return keyLength + valueLength + SIZE_OF_KV_LENGTH;
|
||||
return keyLength + valueLength + SIZE_OF_KEY_LENGTH + SIZE_OF_VALUE_LENGTH;
|
||||
}
|
||||
|
||||
uint32_t lengthConvertEndium() {
|
||||
return hadoop_be32toh(keyLength) + hadoop_be32toh(valueLength) + SIZE_OF_KV_LENGTH;
|
||||
long value = bswap64(*((long *)this));
|
||||
return (value >> 32) + value + SIZE_OF_KEY_LENGTH + SIZE_OF_VALUE_LENGTH;
|
||||
}
|
||||
|
||||
void fill(const void * key, uint32_t keylen, const void * value, uint32_t vallen) {
|
||||
|
@ -298,7 +299,7 @@ struct KVBuffer {
|
|||
}
|
||||
|
||||
static uint32_t headerLength() {
|
||||
return SIZE_OF_KV_LENGTH;
|
||||
return SIZE_OF_KEY_LENGTH + SIZE_OF_VALUE_LENGTH;
|
||||
}
|
||||
};
|
||||
|
||||
|
|
|
@ -60,7 +60,7 @@ bool IFileReader::nextPartition() {
|
|||
if (4 != _stream->readFully(&chsum, 4)) {
|
||||
THROW_EXCEPTION(IOException, "read ifile checksum failed");
|
||||
}
|
||||
uint32_t actual = hadoop_be32toh(chsum);
|
||||
uint32_t actual = bswap(chsum);
|
||||
uint32_t expect = _source->getChecksum();
|
||||
if (actual != expect) {
|
||||
THROW_EXCEPTION_EX(IOException, "read ifile checksum not match, actual %x expect %x", actual,
|
||||
|
@ -130,7 +130,7 @@ void IFileWriter::endPartition() {
|
|||
}
|
||||
|
||||
uint32_t chsum = _dest->getChecksum();
|
||||
chsum = hadoop_be32toh(chsum);
|
||||
chsum = bswap(chsum);
|
||||
_stream->write(&chsum, sizeof(chsum));
|
||||
_stream->flush();
|
||||
IFileSegment * info = &(_spillFileSegments[_spillFileSegments.size() - 1]);
|
||||
|
|
|
@ -74,7 +74,7 @@ public:
|
|||
keyLen = WritableUtils::ReadVInt(kvbuff, len);
|
||||
break;
|
||||
case BytesType:
|
||||
keyLen = hadoop_be32toh(*(uint32_t*)kvbuff);
|
||||
keyLen = bswap(*(uint32_t*)kvbuff);
|
||||
len = 4;
|
||||
break;
|
||||
default:
|
||||
|
@ -89,7 +89,7 @@ public:
|
|||
_valuePos = vbuff + len;
|
||||
break;
|
||||
case BytesType:
|
||||
_valueLen = hadoop_be32toh(*(uint32_t*)vbuff);
|
||||
_valueLen = bswap(*(uint32_t*)vbuff);
|
||||
_valuePos = vbuff + 4;
|
||||
break;
|
||||
default:
|
||||
|
|
|
@ -317,8 +317,8 @@ int NativeObjectFactory::IntComparator(const char * src, uint32_t srcLength, con
|
|||
uint32_t destLength) {
|
||||
int result = (*src) - (*dest);
|
||||
if (result == 0) {
|
||||
uint32_t from = hadoop_be32toh(*(uint32_t*)src);
|
||||
uint32_t to = hadoop_be32toh(*(uint32_t*)dest);
|
||||
uint32_t from = bswap(*(uint32_t*)src);
|
||||
uint32_t to = bswap(*(uint32_t*)dest);
|
||||
if (from > to) {
|
||||
return 1;
|
||||
} else if (from == to) {
|
||||
|
@ -335,8 +335,8 @@ int NativeObjectFactory::LongComparator(const char * src, uint32_t srcLength, co
|
|||
int result = (int)(*src) - (int)(*dest);
|
||||
if (result == 0) {
|
||||
|
||||
uint64_t from = hadoop_be64toh(*(uint64_t*)src);
|
||||
uint64_t to = hadoop_be64toh(*(uint64_t*)dest);
|
||||
uint64_t from = bswap64(*(uint64_t*)src);
|
||||
uint64_t to = bswap64(*(uint64_t*)dest);
|
||||
if (from > to) {
|
||||
return 1;
|
||||
} else if (from == to) {
|
||||
|
@ -380,8 +380,8 @@ int NativeObjectFactory::FloatComparator(const char * src, uint32_t srcLength, c
|
|||
THROW_EXCEPTION_EX(IOException, "float comparator, while src/dest lengt is not 4");
|
||||
}
|
||||
|
||||
uint32_t from = hadoop_be32toh(*(uint32_t*)src);
|
||||
uint32_t to = hadoop_be32toh(*(uint32_t*)dest);
|
||||
uint32_t from = bswap(*(uint32_t*)src);
|
||||
uint32_t to = bswap(*(uint32_t*)dest);
|
||||
|
||||
float * srcValue = (float *)(&from);
|
||||
float * destValue = (float *)(&to);
|
||||
|
@ -401,8 +401,8 @@ int NativeObjectFactory::DoubleComparator(const char * src, uint32_t srcLength,
|
|||
THROW_EXCEPTION_EX(IOException, "double comparator, while src/dest lengt is not 4");
|
||||
}
|
||||
|
||||
uint64_t from = hadoop_be64toh(*(uint64_t*)src);
|
||||
uint64_t to = hadoop_be64toh(*(uint64_t*)dest);
|
||||
uint64_t from = bswap64(*(uint64_t*)src);
|
||||
uint64_t to = bswap64(*(uint64_t*)dest);
|
||||
|
||||
double * srcValue = (double *)(&from);
|
||||
double * destValue = (double *)(&to);
|
||||
|
|
|
@ -58,10 +58,10 @@ void SingleSpillInfo::writeSpillInfo(const std::string & filepath) {
|
|||
appendBuffer.flush();
|
||||
uint32_t chsum = dest.getChecksum();
|
||||
#ifdef SPILLRECORD_CHECKSUM_UINT
|
||||
chsum = hadoop_be32toh(chsum);
|
||||
chsum = bswap(chsum);
|
||||
fout->write(&chsum, sizeof(uint32_t));
|
||||
#else
|
||||
uint64_t wtchsum = hadoop_be64toh((uint64_t)chsum);
|
||||
uint64_t wtchsum = bswap64((uint64_t)chsum);
|
||||
fout->write(&wtchsum, sizeof(uint64_t));
|
||||
#endif
|
||||
}
|
||||
|
|
|
@ -41,7 +41,6 @@
|
|||
#include <map>
|
||||
#include <algorithm>
|
||||
|
||||
#include "hadoop_endian.h"
|
||||
#include "lib/primitives.h"
|
||||
#include "lib/Log.h"
|
||||
#include "NativeTask.h"
|
||||
|
|
|
@ -28,7 +28,6 @@
|
|||
#include <stdint.h>
|
||||
#include <assert.h>
|
||||
#include <string>
|
||||
#include "hadoop_endian.h"
|
||||
|
||||
#ifdef __GNUC__
|
||||
#define likely(x) __builtin_expect((x),1)
|
||||
|
@ -94,6 +93,39 @@ inline void simple_memcpy(void * dest, const void * src, size_t len) {
|
|||
|
||||
#endif
|
||||
|
||||
/**
|
||||
* little-endian to big-endian or vice versa
|
||||
*/
|
||||
inline uint32_t bswap(uint32_t val) {
|
||||
#ifdef __aarch64__
|
||||
__asm__("rev %w[dst], %w[src]" : [dst]"=r"(val) : [src]"r"(val));
|
||||
#else
|
||||
__asm__("bswap %0" : "=r" (val) : "0" (val));
|
||||
#endif
|
||||
return val;
|
||||
}
|
||||
|
||||
inline uint64_t bswap64(uint64_t val) {
|
||||
#ifdef __aarch64__
|
||||
__asm__("rev %[dst], %[src]" : [dst]"=r"(val) : [src]"r"(val));
|
||||
#else
|
||||
#ifdef __X64
|
||||
__asm__("bswapq %0" : "=r" (val) : "0" (val));
|
||||
#else
|
||||
|
||||
uint64_t lower = val & 0xffffffffU;
|
||||
uint32_t higher = (val >> 32) & 0xffffffffU;
|
||||
|
||||
lower = bswap(lower);
|
||||
higher = bswap(higher);
|
||||
|
||||
return (lower << 32) + higher;
|
||||
|
||||
#endif
|
||||
#endif
|
||||
return val;
|
||||
}
|
||||
|
||||
/**
|
||||
* Fast memcmp
|
||||
*/
|
||||
|
@ -126,16 +158,16 @@ inline int64_t fmemcmp(const char * src, const char * dest, uint32_t len) {
|
|||
return ((int64_t)src8[2] - (int64_t)dest8[2]);
|
||||
}
|
||||
case 4: {
|
||||
return (int64_t)hadoop_be32toh(*(uint32_t*)src) - (int64_t)hadoop_be32toh(*(uint32_t*)dest);
|
||||
return (int64_t)bswap(*(uint32_t*)src) - (int64_t)bswap(*(uint32_t*)dest);
|
||||
}
|
||||
}
|
||||
if (len < 8) {
|
||||
int64_t ret = ((int64_t)hadoop_be32toh(*(uint32_t*)src) - (int64_t)hadoop_be32toh(*(uint32_t*)dest));
|
||||
int64_t ret = ((int64_t)bswap(*(uint32_t*)src) - (int64_t)bswap(*(uint32_t*)dest));
|
||||
if (ret) {
|
||||
return ret;
|
||||
}
|
||||
return ((int64_t)hadoop_be32toh(*(uint32_t*)(src + len - 4))
|
||||
- (int64_t)hadoop_be32toh(*(uint32_t*)(dest + len - 4)));
|
||||
return ((int64_t)bswap(*(uint32_t*)(src + len - 4))
|
||||
- (int64_t)bswap(*(uint32_t*)(dest + len - 4)));
|
||||
}
|
||||
uint32_t cur = 0;
|
||||
uint32_t end = len & (0xffffffffU << 3);
|
||||
|
@ -143,8 +175,8 @@ inline int64_t fmemcmp(const char * src, const char * dest, uint32_t len) {
|
|||
uint64_t l = *(uint64_t*)(src8 + cur);
|
||||
uint64_t r = *(uint64_t*)(dest8 + cur);
|
||||
if (l != r) {
|
||||
l = hadoop_be64toh(l);
|
||||
r = hadoop_be64toh(r);
|
||||
l = bswap64(l);
|
||||
r = bswap64(r);
|
||||
return l > r ? 1 : -1;
|
||||
}
|
||||
cur += 8;
|
||||
|
@ -152,8 +184,8 @@ inline int64_t fmemcmp(const char * src, const char * dest, uint32_t len) {
|
|||
uint64_t l = *(uint64_t*)(src8 + len - 8);
|
||||
uint64_t r = *(uint64_t*)(dest8 + len - 8);
|
||||
if (l != r) {
|
||||
l = hadoop_be64toh(l);
|
||||
r = hadoop_be64toh(r);
|
||||
l = bswap64(l);
|
||||
r = bswap64(r);
|
||||
return l > r ? 1 : -1;
|
||||
}
|
||||
return 0;
|
||||
|
|
|
@ -120,29 +120,29 @@ void WritableUtils::WriteVLongInner(int64_t v, char * pos, uint32_t & len) {
|
|||
len = 4;
|
||||
} else if (value < (1ULL << 32)) {
|
||||
*(pos++) = base - 3;
|
||||
*(uint32_t*)(pos) = hadoop_be32toh((uint32_t)value);
|
||||
*(uint32_t*)(pos) = bswap((uint32_t)value);
|
||||
len = 5;
|
||||
} else if (value < (1ULL << 40)) {
|
||||
*(pos++) = base - 4;
|
||||
*(uint32_t*)(pos) = hadoop_be32toh((uint32_t)(value >> 8));
|
||||
*(uint32_t*)(pos) = bswap((uint32_t)(value >> 8));
|
||||
*(uint8_t*)(pos + 4) = value;
|
||||
len = 6;
|
||||
} else if (value < (1ULL << 48)) {
|
||||
*(pos++) = base - 5;
|
||||
*(uint32_t*)(pos) = hadoop_be32toh((uint32_t)(value >> 16));
|
||||
*(uint32_t*)(pos) = bswap((uint32_t)(value >> 16));
|
||||
*(uint8_t*)(pos + 4) = value >> 8;
|
||||
*(uint8_t*)(pos + 5) = value;
|
||||
len = 7;
|
||||
} else if (value < (1ULL << 56)) {
|
||||
*(pos++) = base - 6;
|
||||
*(uint32_t*)(pos) = hadoop_be32toh((uint32_t)(value >> 24));
|
||||
*(uint32_t*)(pos) = bswap((uint32_t)(value >> 24));
|
||||
*(uint8_t*)(pos + 4) = value >> 16;
|
||||
*(uint8_t*)(pos + 5) = value >> 8;
|
||||
*(uint8_t*)(pos + 6) = value;
|
||||
len = 8;
|
||||
} else {
|
||||
*(pos++) = base - 7;
|
||||
*(uint64_t*)pos = hadoop_be64toh(value);
|
||||
*(uint64_t*)pos = bswap64(value);
|
||||
len = 9;
|
||||
}
|
||||
}
|
||||
|
@ -168,7 +168,7 @@ int64_t WritableUtils::ReadLong(InputStream * stream) {
|
|||
if (stream->readFully(&ret, 8) != 8) {
|
||||
THROW_EXCEPTION(IOException, "ReadLong reach EOF");
|
||||
}
|
||||
return (int64_t)hadoop_be64toh(ret);
|
||||
return (int64_t)bswap64(ret);
|
||||
}
|
||||
|
||||
int32_t WritableUtils::ReadInt(InputStream * stream) {
|
||||
|
@ -176,7 +176,7 @@ int32_t WritableUtils::ReadInt(InputStream * stream) {
|
|||
if (stream->readFully(&ret, 4) != 4) {
|
||||
THROW_EXCEPTION(IOException, "ReadInt reach EOF");
|
||||
}
|
||||
return (int32_t)hadoop_be32toh(ret);
|
||||
return (int32_t)bswap(ret);
|
||||
}
|
||||
|
||||
int16_t WritableUtils::ReadShort(InputStream * stream) {
|
||||
|
@ -192,7 +192,7 @@ float WritableUtils::ReadFloat(InputStream * stream) {
|
|||
if (stream->readFully(&ret, 4) != 4) {
|
||||
THROW_EXCEPTION(IOException, "ReadFloat reach EOF");
|
||||
}
|
||||
ret = hadoop_be32toh(ret);
|
||||
ret = bswap(ret);
|
||||
return *(float*)&ret;
|
||||
}
|
||||
|
||||
|
@ -232,12 +232,12 @@ void WritableUtils::WriteVLong(OutputStream * stream, int64_t v) {
|
|||
}
|
||||
|
||||
void WritableUtils::WriteLong(OutputStream * stream, int64_t v) {
|
||||
uint64_t be = hadoop_be64toh((uint64_t)v);
|
||||
uint64_t be = bswap64((uint64_t)v);
|
||||
stream->write(&be, 8);
|
||||
}
|
||||
|
||||
void WritableUtils::WriteInt(OutputStream * stream, int32_t v) {
|
||||
uint32_t be = hadoop_be32toh((uint32_t)v);
|
||||
uint32_t be = bswap((uint32_t)v);
|
||||
stream->write(&be, 4);
|
||||
}
|
||||
|
||||
|
@ -249,7 +249,7 @@ void WritableUtils::WriteShort(OutputStream * stream, int16_t v) {
|
|||
|
||||
void WritableUtils::WriteFloat(OutputStream * stream, float v) {
|
||||
uint32_t intv = *(uint32_t*)&v;
|
||||
intv = hadoop_be32toh(intv);
|
||||
intv = bswap(intv);
|
||||
stream->write(&intv, 4);
|
||||
}
|
||||
|
||||
|
@ -286,10 +286,10 @@ void WritableUtils::toString(string & dest, KeyValueType type, const void * data
|
|||
dest.append(*(uint8_t*)data ? "true" : "false");
|
||||
break;
|
||||
case IntType:
|
||||
dest.append(StringUtil::ToString((int32_t)hadoop_be32toh(*(uint32_t*)data)));
|
||||
dest.append(StringUtil::ToString((int32_t)bswap(*(uint32_t*)data)));
|
||||
break;
|
||||
case LongType:
|
||||
dest.append(StringUtil::ToString((int64_t)hadoop_be64toh(*(uint64_t*)data)));
|
||||
dest.append(StringUtil::ToString((int64_t)bswap64(*(uint64_t*)data)));
|
||||
break;
|
||||
case FloatType:
|
||||
dest.append(StringUtil::ToString(*(float*)data));
|
||||
|
|
|
@ -190,7 +190,7 @@ TEST(IFile, TestGlibCBug) {
|
|||
reader->nextPartition();
|
||||
uint32_t index = 0;
|
||||
while (NULL != (key = reader->nextKey(length))) {
|
||||
int32_t realKey = (int32_t)hadoop_be32toh(*(uint32_t *)(key));
|
||||
int32_t realKey = (int32_t)bswap(*(uint32_t *)(key));
|
||||
ASSERT_LT(index, 5);
|
||||
ASSERT_EQ(expect[index], realKey);
|
||||
index++;
|
||||
|
|
|
@ -38,8 +38,8 @@ inline int fmemcmporig(const char * src, const char * dest, uint32_t len) {
|
|||
uint64_t l = *src8;
|
||||
uint64_t r = *dest8;
|
||||
if (l != r) {
|
||||
l = hadoop_be64toh(l);
|
||||
r = hadoop_be64toh(r);
|
||||
l = bswap64(l);
|
||||
r = bswap64(r);
|
||||
return l > r ? 1 : -1;
|
||||
}
|
||||
++src8;
|
||||
|
@ -59,8 +59,8 @@ inline int fmemcmporig(const char * src, const char * dest, uint32_t len) {
|
|||
if (l == r) {
|
||||
return 0;
|
||||
}
|
||||
l = hadoop_be64toh(l);
|
||||
r = hadoop_be64toh(r);
|
||||
l = bswap64(l);
|
||||
r = bswap64(r);
|
||||
return l > r ? 1 : -1;
|
||||
}
|
||||
|
||||
|
|
|
@ -43,8 +43,8 @@ TEST(KVBuffer, test) {
|
|||
ASSERT_EQ(8, kv1->getKey() - buff);
|
||||
ASSERT_EQ(strlen(KEY) + 8, kv1->getValue() - buff);
|
||||
|
||||
kv1->keyLength = hadoop_be32toh(kv1->keyLength);
|
||||
kv1->valueLength = hadoop_be32toh(kv1->valueLength);
|
||||
kv1->keyLength = bswap(kv1->keyLength);
|
||||
kv1->valueLength = bswap(kv1->valueLength);
|
||||
|
||||
ASSERT_EQ(8, kv1->headerLength());
|
||||
ASSERT_EQ(strlen(KEY) + strlen(VALUE) + 8, kv1->lengthConvertEndium());
|
||||
|
|
|
@ -59,7 +59,7 @@ class MemoryBlockFactory {
|
|||
kv->keyLength = 4;
|
||||
kv->valueLength = 4;
|
||||
uint32_t * key = (uint32_t *)kv->getKey();
|
||||
*key = hadoop_be32toh(index);
|
||||
*key = bswap(index);
|
||||
}
|
||||
return block1;
|
||||
}
|
||||
|
|
|
@ -85,17 +85,17 @@ TEST(MemoryBlock, sort) {
|
|||
medium->keyLength = 4;
|
||||
medium->valueLength = 4;
|
||||
uint32_t * mediumKey = (uint32_t *)medium->getKey();
|
||||
*mediumKey = hadoop_be32toh(MEDIUM);
|
||||
*mediumKey = bswap(MEDIUM);
|
||||
|
||||
small->keyLength = 4;
|
||||
small->valueLength = 4;
|
||||
uint32_t * smallKey = (uint32_t *)small->getKey();
|
||||
*smallKey = hadoop_be32toh(SMALL);
|
||||
*smallKey = bswap(SMALL);
|
||||
|
||||
big->keyLength = 4;
|
||||
big->valueLength = 4;
|
||||
uint32_t * bigKey = (uint32_t *)big->getKey();
|
||||
*bigKey = hadoop_be32toh(BIG);
|
||||
*bigKey = bswap(BIG);
|
||||
|
||||
ComparatorPtr bytesComparator = NativeTask::get_comparator(BytesType, NULL);
|
||||
block.sort(CPPSORT, bytesComparator);
|
||||
|
|
|
@ -16,7 +16,7 @@
|
|||
* limitations under the License.
|
||||
*/
|
||||
|
||||
#include "hadoop_endian.h"
|
||||
#include "lib/commons.h"
|
||||
#include "test_commons.h"
|
||||
#include "lib/PartitionBucket.h"
|
||||
#include "lib/PartitionBucketIterator.h"
|
||||
|
@ -129,15 +129,15 @@ TEST(PartitionBucket, sort) {
|
|||
const uint32_t BIG = 1000;
|
||||
|
||||
kv1->keyLength = 4;
|
||||
*((uint32_t *)kv1->getKey()) = hadoop_be32toh(BIG);
|
||||
*((uint32_t *)kv1->getKey()) = bswap(BIG);
|
||||
kv1->valueLength = KV_SIZE - kv1->headerLength() - kv1->keyLength;
|
||||
|
||||
kv2->keyLength = 4;
|
||||
*((uint32_t *)kv2->getKey()) = hadoop_be32toh(SMALL);
|
||||
*((uint32_t *)kv2->getKey()) = bswap(SMALL);
|
||||
kv2->valueLength = KV_SIZE - kv2->headerLength() - kv2->keyLength;
|
||||
|
||||
kv3->keyLength = 4;
|
||||
*((uint32_t *)kv3->getKey()) = hadoop_be32toh(MEDIUM);
|
||||
*((uint32_t *)kv3->getKey()) = bswap(MEDIUM);
|
||||
kv3->valueLength = KV_SIZE - kv3->headerLength() - kv3->keyLength;
|
||||
|
||||
bucket->sort(DUALPIVOTSORT);
|
||||
|
@ -148,13 +148,13 @@ TEST(PartitionBucket, sort) {
|
|||
Buffer value;
|
||||
iter->next(key, value);
|
||||
|
||||
ASSERT_EQ(SMALL, hadoop_be32toh(*(uint32_t * )key.data()));
|
||||
ASSERT_EQ(SMALL, bswap(*(uint32_t * )key.data()));
|
||||
|
||||
iter->next(key, value);
|
||||
ASSERT_EQ(MEDIUM, hadoop_be32toh(*(uint32_t * )key.data()));
|
||||
ASSERT_EQ(MEDIUM, bswap(*(uint32_t * )key.data()));
|
||||
|
||||
iter->next(key, value);
|
||||
ASSERT_EQ(BIG, hadoop_be32toh(*(uint32_t * )key.data()));
|
||||
ASSERT_EQ(BIG, bswap(*(uint32_t * )key.data()));
|
||||
|
||||
delete iter;
|
||||
delete bucket;
|
||||
|
@ -181,15 +181,15 @@ TEST(PartitionBucket, spill) {
|
|||
const uint32_t BIG = 1000;
|
||||
|
||||
kv1->keyLength = 4;
|
||||
*((uint32_t *)kv1->getKey()) = hadoop_be32toh(BIG);
|
||||
*((uint32_t *)kv1->getKey()) = bswap(BIG);
|
||||
kv1->valueLength = KV_SIZE - KVBuffer::headerLength() - kv1->keyLength;
|
||||
|
||||
kv2->keyLength = 4;
|
||||
*((uint32_t *)kv2->getKey()) = hadoop_be32toh(SMALL);
|
||||
*((uint32_t *)kv2->getKey()) = bswap(SMALL);
|
||||
kv2->valueLength = KV_SIZE - KVBuffer::headerLength() - kv2->keyLength;
|
||||
|
||||
kv3->keyLength = 4;
|
||||
*((uint32_t *)kv3->getKey()) = hadoop_be32toh(MEDIUM);
|
||||
*((uint32_t *)kv3->getKey()) = bswap(MEDIUM);
|
||||
kv3->valueLength = KV_SIZE - KVBuffer::headerLength() - kv3->keyLength;
|
||||
|
||||
bucket->sort(DUALPIVOTSORT);
|
||||
|
@ -203,17 +203,17 @@ TEST(PartitionBucket, spill) {
|
|||
KVBuffer * first = (KVBuffer *)writer.buff();
|
||||
ASSERT_EQ(4, first->keyLength);
|
||||
ASSERT_EQ(KV_SIZE - KVBuffer::headerLength() - 4, first->valueLength);
|
||||
ASSERT_EQ(hadoop_be32toh(SMALL), (*(uint32_t * )(first->getKey())));
|
||||
ASSERT_EQ(bswap(SMALL), (*(uint32_t * )(first->getKey())));
|
||||
|
||||
KVBuffer * second = first->next();
|
||||
ASSERT_EQ(4, second->keyLength);
|
||||
ASSERT_EQ(KV_SIZE - KVBuffer::headerLength() - 4, second->valueLength);
|
||||
ASSERT_EQ(hadoop_be32toh(MEDIUM), (*(uint32_t * )(second->getKey())));
|
||||
ASSERT_EQ(bswap(MEDIUM), (*(uint32_t * )(second->getKey())));
|
||||
|
||||
KVBuffer * third = second->next();
|
||||
ASSERT_EQ(4, third->keyLength);
|
||||
ASSERT_EQ(KV_SIZE - KVBuffer::headerLength() - 4, third->valueLength);
|
||||
ASSERT_EQ(hadoop_be32toh(BIG), (*(uint32_t * )(third->getKey())));
|
||||
ASSERT_EQ(bswap(BIG), (*(uint32_t * )(third->getKey())));
|
||||
|
||||
delete [] buff;
|
||||
delete bucket;
|
||||
|
|
Loading…
Reference in New Issue