OmniSciDB  fe05a0c208
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
RuntimeFunctions.h File Reference
#include "Shared/funcannotations.h"
#include <cassert>
#include <cstdint>
#include <ctime>
#include <limits>
#include <type_traits>
+ Include dependency graph for RuntimeFunctions.h:
+ This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define EMPTY_KEY_64   std::numeric_limits<int64_t>::max()
 
#define EMPTY_KEY_32   std::numeric_limits<int32_t>::max()
 
#define EMPTY_KEY_16   std::numeric_limits<int16_t>::max()
 
#define EMPTY_KEY_8   std::numeric_limits<int8_t>::max()
 

Enumerations

enum  RuntimeInterruptFlags { INT_CHECK = 0, INT_ABORT = -1, INT_RESET = -2 }
 

Functions

int64_t agg_sum (int64_t *agg, const int64_t val)
 
void agg_max (int64_t *agg, const int64_t val)
 
void agg_min (int64_t *agg, const int64_t val)
 
void agg_sum_double (int64_t *agg, const double val)
 
void agg_max_double (int64_t *agg, const double val)
 
void agg_min_double (int64_t *agg, const double val)
 
int32_t agg_sum_int32_skip_val (int32_t *agg, const int32_t val, const int32_t skip_val)
 
int64_t agg_sum_skip_val (int64_t *agg, const int64_t val, const int64_t skip_val)
 
void agg_max_skip_val (int64_t *agg, const int64_t val, const int64_t skip_val)
 
void agg_min_skip_val (int64_t *agg, const int64_t val, const int64_t skip_val)
 
void agg_sum_float_skip_val (int32_t *agg, const float val, const float skip_val)
 
void agg_sum_double_skip_val (int64_t *agg, const double val, const double skip_val)
 
void agg_max_double_skip_val (int64_t *agg, const double val, const double skip_val)
 
void agg_min_double_skip_val (int64_t *agg, const double val, const double skip_val)
 
int32_t agg_sum_int32 (int32_t *agg, const int32_t val)
 
void agg_max_int32 (int32_t *agg, const int32_t val)
 
void agg_max_int16 (int16_t *agg, const int16_t val)
 
void agg_max_int8 (int8_t *agg, const int8_t val)
 
void agg_min_int32 (int32_t *agg, const int32_t val)
 
void agg_min_int16 (int16_t *agg, const int16_t val)
 
void agg_min_int8 (int8_t *agg, const int8_t val)
 
void agg_sum_float (int32_t *agg, const float val)
 
void agg_max_float (int32_t *agg, const float val)
 
void agg_min_float (int32_t *agg, const float val)
 
void agg_max_int32_skip_val (int32_t *agg, const int32_t val, const int32_t skip_val)
 
void agg_max_int16_skip_val (int16_t *agg, const int16_t val, const int16_t skip_val)
 
void agg_max_int8_skip_val (int8_t *agg, const int8_t val, const int8_t skip_val)
 
void agg_min_int32_skip_val (int32_t *agg, const int32_t val, const int32_t skip_val)
 
void agg_min_int16_skip_val (int16_t *agg, const int16_t val, const int16_t skip_val)
 
void agg_min_int8_skip_val (int8_t *agg, const int8_t val, const int8_t skip_val)
 
void agg_max_float_skip_val (int32_t *agg, const float val, const float skip_val)
 
void agg_min_float_skip_val (int32_t *agg, const float val, const float skip_val)
 
void agg_count_distinct_bitmap (int64_t *agg, const int64_t val, const int64_t min_val)
 
RUNTIME_EXPORT uint32_t key_hash (const int64_t *key, const uint32_t key_qw_count, const uint32_t key_byte_width)
 
RUNTIME_EXPORT int64_t * get_group_value (int64_t *groups_buffer, const uint32_t groups_buffer_entry_count, const int64_t *key, const uint32_t key_count, const uint32_t key_width, const uint32_t row_size_quad)
 
bool check_interrupt ()
 
bool check_interrupt_init (unsigned command)
 
RUNTIME_EXPORT int64_t * get_group_value_with_watchdog (int64_t *groups_buffer, const uint32_t groups_buffer_entry_count, const int64_t *key, const uint32_t key_count, const uint32_t key_width, const uint32_t row_size_quad)
 
RUNTIME_EXPORT int64_t * get_group_value_columnar (int64_t *groups_buffer, const uint32_t groups_buffer_entry_count, const int64_t *key, const uint32_t key_qw_count)
 
RUNTIME_EXPORT int64_t * get_group_value_columnar_with_watchdog (int64_t *groups_buffer, const uint32_t groups_buffer_entry_count, const int64_t *key, const uint32_t key_qw_count)
 
RUNTIME_EXPORT int64_t * get_group_value_fast (int64_t *groups_buffer, const int64_t key, const int64_t min_key, const int64_t bucket, const uint32_t row_size_quad)
 
RUNTIME_EXPORT int64_t * get_group_value_fast_with_original_key (int64_t *groups_buffer, const int64_t key, const int64_t orig_key, const int64_t min_key, const int64_t bucket, const uint32_t row_size_quad)
 
RUNTIME_EXPORT uint32_t get_columnar_group_bin_offset (int64_t *key_base_ptr, const int64_t key, const int64_t min_key, const int64_t bucket)
 
int64_t * get_matching_group_value_perfect_hash (int64_t *groups_buffer, const uint32_t h, const int64_t *key, const uint32_t key_qw_count, const uint32_t row_size_quad)
 
int64_t * get_matching_group_value_perfect_hash_keyless (int64_t *groups_buffer, const uint32_t hashed_index, const uint32_t row_size_quad)
 
int32_t * get_bucketized_hash_slot (int32_t *buff, const int64_t key, const int64_t min_key, const int64_t bucket_normalization=1)
 
int32_t * get_hash_slot (int32_t *buff, const int64_t key, const int64_t min_key)
 
int32_t * get_hash_slot_sharded (int32_t *buff, const int64_t key, const int64_t min_key, const uint32_t entry_count_per_shard, const uint32_t num_shards, const uint32_t device_count)
 
int32_t * get_bucketized_hash_slot_sharded (int32_t *buff, const int64_t key, const int64_t min_key, const uint32_t entry_count_per_shard, const uint32_t num_shards, const uint32_t device_count, const int64_t bucket_normalization)
 
int32_t * get_hash_slot_sharded_opt (int32_t *buff, const int64_t key, const int64_t min_key, const uint32_t entry_count_per_shard, const uint32_t shard, const uint32_t num_shards, const uint32_t device_count)
 
int32_t * get_bucketized_hash_slot_sharded_opt (int32_t *buff, const int64_t key, const int64_t min_key, const uint32_t entry_count_per_shard, const uint32_t shard, const uint32_t num_shards, const uint32_t device_count, const int64_t bucket_normalization)
 
void linear_probabilistic_count (uint8_t *bitmap, const uint32_t bitmap_bytes, const uint8_t *key_bytes, const uint32_t key_len)
 
int64_t fixed_width_int_decode_noinline (const int8_t *byte_stream, const int32_t byte_width, const int64_t pos)
 
int64_t fixed_width_unsigned_decode_noinline (const int8_t *byte_stream, const int32_t byte_width, const int64_t pos)
 
float fixed_width_float_decode_noinline (const int8_t *byte_stream, const int64_t pos)
 
double fixed_width_double_decode_noinline (const int8_t *byte_stream, const int64_t pos)
 
int64_t fixed_width_small_date_decode_noinline (const int8_t *byte_stream, const int32_t byte_width, const int32_t null_val, const int64_t ret_null_val, const int64_t pos)
 
int8_t * extract_str_ptr_noinline (const uint64_t str_and_len)
 
int32_t extract_str_len_noinline (const uint64_t str_and_len)
 
template<typename T = int64_t>
get_empty_key ()
 
template<>
int32_t get_empty_key ()
 

Macro Definition Documentation

#define EMPTY_KEY_16   std::numeric_limits<int16_t>::max()

Definition at line 118 of file RuntimeFunctions.h.

#define EMPTY_KEY_32   std::numeric_limits<int32_t>::max()

Definition at line 117 of file RuntimeFunctions.h.

Referenced by get_empty_key().

#define EMPTY_KEY_64   std::numeric_limits<int64_t>::max()

Definition at line 116 of file RuntimeFunctions.h.

Referenced by get_empty_key().

#define EMPTY_KEY_8   std::numeric_limits<int8_t>::max()

Definition at line 119 of file RuntimeFunctions.h.

Enumeration Type Documentation

Enumerator
INT_CHECK 
INT_ABORT 
INT_RESET 

Definition at line 133 of file RuntimeFunctions.h.

Function Documentation

void agg_count_distinct_bitmap ( int64_t *  agg,
const int64_t  val,
const int64_t  min_val 
)

Definition at line 302 of file RuntimeFunctions.cpp.

Referenced by agg_count_distinct_bitmap_skip_val(), WindowFunctionContext::fillPartitionEnd(), WindowFunctionContext::fillPartitionStart(), anonymous_namespace{WindowContext.cpp}::index_to_partition_end(), and InValuesBitmap::InValuesBitmap().

304  {
305  const uint64_t bitmap_idx = val - min_val;
306  reinterpret_cast<int8_t*>(*agg)[bitmap_idx >> 3] |= (1 << (bitmap_idx & 7));
307 }

+ Here is the caller graph for this function:

void agg_max ( int64_t *  agg,
const int64_t  val 
)

Definition at line 367 of file RuntimeFunctions.cpp.

367  {
368  *agg = std::max(*agg, val);
369 }
void agg_max_double ( int64_t *  agg,
const double  val 
)

Definition at line 591 of file RuntimeFunctions.cpp.

References test_fsi::r.

591  {
592  const auto r = std::max(*reinterpret_cast<const double*>(agg), val);
593  *agg = *(reinterpret_cast<const int64_t*>(may_alias_ptr(&r)));
594 }
tuple r
Definition: test_fsi.py:16
void agg_max_double_skip_val ( int64_t *  agg,
const double  val,
const double  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_max_float ( int32_t *  agg,
const float  val 
)

Definition at line 632 of file RuntimeFunctions.cpp.

References test_fsi::r.

632  {
633  const auto r = std::max(*reinterpret_cast<const float*>(agg), val);
634  *agg = *(reinterpret_cast<const int32_t*>(may_alias_ptr(&r)));
635 }
tuple r
Definition: test_fsi.py:16
void agg_max_float_skip_val ( int32_t *  agg,
const float  val,
const float  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_max_int16 ( int16_t *  agg,
const int16_t  val 
)
void agg_max_int16_skip_val ( int16_t *  agg,
const int16_t  val,
const int16_t  skip_val 
)
void agg_max_int32 ( int32_t *  agg,
const int32_t  val 
)
void agg_max_int32_skip_val ( int32_t *  agg,
const int32_t  val,
const int32_t  skip_val 
)
void agg_max_int8 ( int8_t *  agg,
const int8_t  val 
)
void agg_max_int8_skip_val ( int8_t *  agg,
const int8_t  val,
const int8_t  skip_val 
)
void agg_max_skip_val ( int64_t *  agg,
const int64_t  val,
const int64_t  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_min ( int64_t *  agg,
const int64_t  val 
)

Definition at line 371 of file RuntimeFunctions.cpp.

371  {
372  *agg = std::min(*agg, val);
373 }
void agg_min_double ( int64_t *  agg,
const double  val 
)

Definition at line 596 of file RuntimeFunctions.cpp.

References test_fsi::r.

596  {
597  const auto r = std::min(*reinterpret_cast<const double*>(agg), val);
598  *agg = *(reinterpret_cast<const int64_t*>(may_alias_ptr(&r)));
599 }
tuple r
Definition: test_fsi.py:16
void agg_min_double_skip_val ( int64_t *  agg,
const double  val,
const double  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_min_float ( int32_t *  agg,
const float  val 
)

Definition at line 637 of file RuntimeFunctions.cpp.

References test_fsi::r.

637  {
638  const auto r = std::min(*reinterpret_cast<const float*>(agg), val);
639  *agg = *(reinterpret_cast<const int32_t*>(may_alias_ptr(&r)));
640 }
tuple r
Definition: test_fsi.py:16
void agg_min_float_skip_val ( int32_t *  agg,
const float  val,
const float  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_min_int16 ( int16_t *  agg,
const int16_t  val 
)
void agg_min_int16_skip_val ( int16_t *  agg,
const int16_t  val,
const int16_t  skip_val 
)
void agg_min_int32 ( int32_t *  agg,
const int32_t  val 
)
void agg_min_int32_skip_val ( int32_t *  agg,
const int32_t  val,
const int32_t  skip_val 
)
void agg_min_int8 ( int8_t *  agg,
const int8_t  val 
)
void agg_min_int8_skip_val ( int8_t *  agg,
const int8_t  val,
const int8_t  skip_val 
)
void agg_min_skip_val ( int64_t *  agg,
const int64_t  val,
const int64_t  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

int64_t agg_sum ( int64_t *  agg,
const int64_t  val 
)

Definition at line 361 of file RuntimeFunctions.cpp.

Referenced by agg_sum_skip_val().

361  {
362  const auto old = *agg;
363  *agg += val;
364  return old;
365 }

+ Here is the caller graph for this function:

void agg_sum_double ( int64_t *  agg,
const double  val 
)

Definition at line 586 of file RuntimeFunctions.cpp.

References test_fsi::r.

586  {
587  const auto r = *reinterpret_cast<const double*>(agg) + val;
588  *agg = *reinterpret_cast<const int64_t*>(may_alias_ptr(&r));
589 }
tuple r
Definition: test_fsi.py:16
void agg_sum_double_skip_val ( int64_t *  agg,
const double  val,
const double  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

void agg_sum_float ( int32_t *  agg,
const float  val 
)

Definition at line 627 of file RuntimeFunctions.cpp.

References test_fsi::r.

627  {
628  const auto r = *reinterpret_cast<const float*>(agg) + val;
629  *agg = *reinterpret_cast<const int32_t*>(may_alias_ptr(&r));
630 }
tuple r
Definition: test_fsi.py:16
void agg_sum_float_skip_val ( int32_t *  agg,
const float  val,
const float  skip_val 
)

Referenced by Executor::reduceResults().

+ Here is the caller graph for this function:

int32_t agg_sum_int32 ( int32_t *  agg,
const int32_t  val 
)

Definition at line 419 of file RuntimeFunctions.cpp.

Referenced by agg_sum_int32_skip_val().

419  {
420  const auto old = *agg;
421  *agg += val;
422  return old;
423 }

+ Here is the caller graph for this function:

int32_t agg_sum_int32_skip_val ( int32_t *  agg,
const int32_t  val,
const int32_t  skip_val 
)

Definition at line 504 of file RuntimeFunctions.cpp.

References agg_sum_int32().

506  {
507  const auto old = *agg;
508  if (val != skip_val) {
509  if (old != skip_val) {
510  return agg_sum_int32(agg, val);
511  } else {
512  *agg = val;
513  }
514  }
515  return old;
516 }
ALWAYS_INLINE int32_t agg_sum_int32(int32_t *agg, const int32_t val)

+ Here is the call graph for this function:

int64_t agg_sum_skip_val ( int64_t *  agg,
const int64_t  val,
const int64_t  skip_val 
)

Definition at line 490 of file RuntimeFunctions.cpp.

References agg_sum().

Referenced by Executor::reduceResults().

492  {
493  const auto old = *agg;
494  if (val != skip_val) {
495  if (old != skip_val) {
496  return agg_sum(agg, val);
497  } else {
498  *agg = val;
499  }
500  }
501  return old;
502 }
ALWAYS_INLINE int64_t agg_sum(int64_t *agg, const int64_t val)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool check_interrupt ( )

Definition at line 160 of file cuda_mapd_rt.cu.

References check_interrupt_init(), INT_CHECK, and runtime_interrupt_flag.

Referenced by check_interrupt_rt(), ColumnarResults::compactAndCopyEntriesWithoutTargetSkipping(), ColumnarResults::compactAndCopyEntriesWithTargetSkipping(), ColumnFetcher::getAllTableColumnFragments(), QueryExecutionContext::groupBufferToDeinterleavedResults(), ColumnFetcher::linearizeColumnFragments(), ColumnarResults::locateAndCountEntries(), ColumnFetcher::makeJoinColumn(), ColumnarResults::materializeAllColumnsThroughIteration(), ColumnarResults::materializeAllLazyColumns(), and ResultSetStorage::reduceEntriesNoCollisionsColWise().

160  {
161  return (runtime_interrupt_flag == 1) ? true : false;
162 }
__device__ int32_t runtime_interrupt_flag
Definition: cuda_mapd_rt.cu:96

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool check_interrupt_init ( unsigned  command)

Definition at line 1413 of file RuntimeFunctions.cpp.

References INT_ABORT, INT_CHECK, INT_RESET, and runtime_interrupt_flag.

Referenced by check_interrupt(), Executor::interrupt(), and Executor::resetInterrupt().

1413  {
1414  static std::atomic_bool runtime_interrupt_flag{false};
1415 
1416  if (command == static_cast<unsigned>(INT_CHECK)) {
1417  if (runtime_interrupt_flag.load()) {
1418  return true;
1419  }
1420  return false;
1421  }
1422  if (command == static_cast<unsigned>(INT_ABORT)) {
1423  runtime_interrupt_flag.store(true);
1424  return false;
1425  }
1426  if (command == static_cast<unsigned>(INT_RESET)) {
1427  runtime_interrupt_flag.store(false);
1428  return false;
1429  }
1430  return false;
1431 }
__device__ int32_t runtime_interrupt_flag
Definition: cuda_mapd_rt.cu:96

+ Here is the caller graph for this function:

int32_t extract_str_len_noinline ( const uint64_t  str_and_len)

Definition at line 1224 of file RuntimeFunctions.cpp.

References extract_str_len().

Referenced by string_compress().

1224  {
1225  return extract_str_len(str_and_len);
1226 }
ALWAYS_INLINE int32_t extract_str_len(const uint64_t str_and_len)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int8_t* extract_str_ptr_noinline ( const uint64_t  str_and_len)

Definition at line 1220 of file RuntimeFunctions.cpp.

References extract_str_ptr().

Referenced by string_compress().

1220  {
1221  return extract_str_ptr(str_and_len);
1222 }
ALWAYS_INLINE int8_t * extract_str_ptr(const uint64_t str_and_len)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

double fixed_width_double_decode_noinline ( const int8_t *  byte_stream,
const int64_t  pos 
)

Definition at line 134 of file DecodersImpl.h.

References fixed_width_double_decode(), and SUFFIX.

Referenced by compute_bucket_sizes_impl(), JoinColumnIterator::getElementSwitch(), result_set::lazy_decode(), and OverlapsKeyHandler::operator()().

134  {
135  return SUFFIX(fixed_width_double_decode)(byte_stream, pos);
136 }
#define SUFFIX(name)
DEVICE ALWAYS_INLINE double SUFFIX() fixed_width_double_decode(const int8_t *byte_stream, const int64_t pos)
Definition: DecodersImpl.h:126

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

float fixed_width_float_decode_noinline ( const int8_t *  byte_stream,
const int64_t  pos 
)

Definition at line 121 of file DecodersImpl.h.

References fixed_width_float_decode(), and SUFFIX.

Referenced by result_set::lazy_decode().

121  {
122  return SUFFIX(fixed_width_float_decode)(byte_stream, pos);
123 }
#define SUFFIX(name)
DEVICE ALWAYS_INLINE float SUFFIX() fixed_width_float_decode(const int8_t *byte_stream, const int64_t pos)
Definition: DecodersImpl.h:113

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int64_t fixed_width_int_decode_noinline ( const int8_t *  byte_stream,
const int32_t  byte_width,
const int64_t  pos 
)

Definition at line 91 of file DecodersImpl.h.

References fixed_width_int_decode(), and SUFFIX.

Referenced by JoinColumnIterator::getElementSwitch(), and result_set::lazy_decode().

93  {
94  return SUFFIX(fixed_width_int_decode)(byte_stream, byte_width, pos);
95 }
DEVICE ALWAYS_INLINE int64_t SUFFIX() fixed_width_int_decode(const int8_t *byte_stream, const int32_t byte_width, const int64_t pos)
Definition: DecodersImpl.h:31
#define SUFFIX(name)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int64_t fixed_width_small_date_decode_noinline ( const int8_t *  byte_stream,
const int32_t  byte_width,
const int32_t  null_val,
const int64_t  ret_null_val,
const int64_t  pos 
)

Definition at line 149 of file DecodersImpl.h.

References fixed_width_small_date_decode(), and SUFFIX.

Referenced by JoinColumnIterator::getElementSwitch(), and result_set::lazy_decode().

153  {
155  byte_stream, byte_width, null_val, ret_null_val, pos);
156 }
#define SUFFIX(name)
DEVICE ALWAYS_INLINE int64_t SUFFIX() fixed_width_small_date_decode(const int8_t *byte_stream, const int32_t byte_width, const int32_t null_val, const int64_t ret_null_val, const int64_t pos)
Definition: DecodersImpl.h:139

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int64_t fixed_width_unsigned_decode_noinline ( const int8_t *  byte_stream,
const int32_t  byte_width,
const int64_t  pos 
)

Definition at line 98 of file DecodersImpl.h.

References fixed_width_unsigned_decode(), and SUFFIX.

Referenced by JoinColumnIterator::getElementSwitch(), and result_set::lazy_decode().

100  {
101  return SUFFIX(fixed_width_unsigned_decode)(byte_stream, byte_width, pos);
102 }
#define SUFFIX(name)
DEVICE ALWAYS_INLINE int64_t SUFFIX() fixed_width_unsigned_decode(const int8_t *byte_stream, const int32_t byte_width, const int64_t pos)
Definition: DecodersImpl.h:61

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int32_t* get_bucketized_hash_slot ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key,
const int64_t  bucket_normalization = 1 
)

Definition at line 31 of file JoinHashImpl.h.

Referenced by bucketized_hash_join_idx(), count_matches_bucketized(), fill_hash_join_buff_bucketized(), and fill_row_ids_bucketized().

35  {
36  return buff + (key - min_key) / bucket_normalization;
37 }

+ Here is the caller graph for this function:

int32_t* get_bucketized_hash_slot_sharded ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key,
const uint32_t  entry_count_per_shard,
const uint32_t  num_shards,
const uint32_t  device_count,
const int64_t  bucket_normalization 
)

Definition at line 45 of file JoinHashImpl.h.

References SHARD_FOR_KEY.

Referenced by fill_row_ids_sharded_bucketized().

52  {
53  const uint32_t shard = SHARD_FOR_KEY(key, num_shards);
54  const uint32_t shard_buffer_index =
55  shard / device_count; // shard sub-buffer index within `buff`
56  int32_t* shard_buffer = buff + shard_buffer_index * entry_count_per_shard;
57  return shard_buffer + (key - min_key) / bucket_normalization / num_shards;
58 }
#define SHARD_FOR_KEY(key, num_shards)
Definition: shard_key.h:20

+ Here is the caller graph for this function:

int32_t* get_bucketized_hash_slot_sharded_opt ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key,
const uint32_t  entry_count_per_shard,
const uint32_t  shard,
const uint32_t  num_shards,
const uint32_t  device_count,
const int64_t  bucket_normalization 
)

Definition at line 74 of file JoinHashImpl.h.

Referenced by fill_hash_join_buff_sharded_bucketized().

82  {
83  const uint32_t shard_buffer_index =
84  shard / device_count; // shard sub-buffer index within `buff`
85  int32_t* shard_buffer = buff + shard_buffer_index * entry_count_per_shard;
86  return shard_buffer + (key - min_key) / bucket_normalization / num_shards;
87 }

+ Here is the caller graph for this function:

RUNTIME_EXPORT uint32_t get_columnar_group_bin_offset ( int64_t *  key_base_ptr,
const int64_t  key,
const int64_t  min_key,
const int64_t  bucket 
)

Definition at line 228 of file GroupByRuntime.cpp.

References EMPTY_KEY_64.

231  {
232  int64_t off = key - min_key;
233  if (bucket) {
234  off /= bucket;
235  }
236  if (key_base_ptr[off] == EMPTY_KEY_64) {
237  key_base_ptr[off] = key;
238  }
239  return off;
240 }
#define EMPTY_KEY_64
template<typename T = int64_t>
T get_empty_key ( )
inline

Definition at line 264 of file RuntimeFunctions.h.

References EMPTY_KEY_64.

264  {
265  static_assert(std::is_same<T, int64_t>::value,
266  "Unsupported template parameter other than int64_t for now");
267  return EMPTY_KEY_64;
268 }
#define EMPTY_KEY_64
template<>
int32_t get_empty_key ( )
inline

Definition at line 271 of file RuntimeFunctions.h.

References EMPTY_KEY_32.

271  {
272  return EMPTY_KEY_32;
273 }
#define EMPTY_KEY_32
RUNTIME_EXPORT int64_t* get_group_value ( int64_t *  groups_buffer,
const uint32_t  groups_buffer_entry_count,
const int64_t *  key,
const uint32_t  key_count,
const uint32_t  key_width,
const uint32_t  row_size_quad 
)

Definition at line 25 of file GroupByRuntime.cpp.

References get_matching_group_value(), and key_hash().

Referenced by ResultSetStorage::moveOneEntryToBuffer().

31  {
32  uint32_t h = key_hash(key, key_count, key_width) % groups_buffer_entry_count;
33  int64_t* matching_group = get_matching_group_value(
34  groups_buffer, h, key, key_count, key_width, row_size_quad);
35  if (matching_group) {
36  return matching_group;
37  }
38  uint32_t h_probe = (h + 1) % groups_buffer_entry_count;
39  while (h_probe != h) {
40  matching_group = get_matching_group_value(
41  groups_buffer, h_probe, key, key_count, key_width, row_size_quad);
42  if (matching_group) {
43  return matching_group;
44  }
45  h_probe = (h_probe + 1) % groups_buffer_entry_count;
46  }
47  return NULL;
48 }
__device__ int64_t * get_matching_group_value(int64_t *groups_buffer, const uint32_t h, const T *key, const uint32_t key_count, const uint32_t row_size_quad)
RUNTIME_EXPORT ALWAYS_INLINE DEVICE uint32_t key_hash(const int64_t *key, const uint32_t key_count, const uint32_t key_byte_width)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

RUNTIME_EXPORT int64_t* get_group_value_columnar ( int64_t *  groups_buffer,
const uint32_t  groups_buffer_entry_count,
const int64_t *  key,
const uint32_t  key_qw_count 
)

Definition at line 139 of file GroupByRuntime.cpp.

References get_matching_group_value_columnar(), and key_hash().

Referenced by ResultSetStorage::moveOneEntryToBuffer().

143  {
144  uint32_t h = key_hash(key, key_qw_count, sizeof(int64_t)) % groups_buffer_entry_count;
145  int64_t* matching_group = get_matching_group_value_columnar(
146  groups_buffer, h, key, key_qw_count, groups_buffer_entry_count);
147  if (matching_group) {
148  return matching_group;
149  }
150  uint32_t h_probe = (h + 1) % groups_buffer_entry_count;
151  while (h_probe != h) {
152  matching_group = get_matching_group_value_columnar(
153  groups_buffer, h_probe, key, key_qw_count, groups_buffer_entry_count);
154  if (matching_group) {
155  return matching_group;
156  }
157  h_probe = (h_probe + 1) % groups_buffer_entry_count;
158  }
159  return NULL;
160 }
__device__ int64_t * get_matching_group_value_columnar(int64_t *groups_buffer, const uint32_t h, const int64_t *key, const uint32_t key_qw_count, const size_t entry_count)
RUNTIME_EXPORT ALWAYS_INLINE DEVICE uint32_t key_hash(const int64_t *key, const uint32_t key_count, const uint32_t key_byte_width)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

RUNTIME_EXPORT int64_t* get_group_value_columnar_with_watchdog ( int64_t *  groups_buffer,
const uint32_t  groups_buffer_entry_count,
const int64_t *  key,
const uint32_t  key_qw_count 
)

Definition at line 163 of file GroupByRuntime.cpp.

References dynamic_watchdog(), get_matching_group_value_columnar(), and key_hash().

166  {
167  uint32_t h = key_hash(key, key_qw_count, sizeof(int64_t)) % groups_buffer_entry_count;
168  int64_t* matching_group = get_matching_group_value_columnar(
169  groups_buffer, h, key, key_qw_count, groups_buffer_entry_count);
170  if (matching_group) {
171  return matching_group;
172  }
173  uint32_t watchdog_countdown = 100;
174  uint32_t h_probe = (h + 1) % groups_buffer_entry_count;
175  while (h_probe != h) {
176  matching_group = get_matching_group_value_columnar(
177  groups_buffer, h_probe, key, key_qw_count, groups_buffer_entry_count);
178  if (matching_group) {
179  return matching_group;
180  }
181  h_probe = (h_probe + 1) % groups_buffer_entry_count;
182  if (--watchdog_countdown == 0) {
183  if (dynamic_watchdog()) {
184  return NULL;
185  }
186  watchdog_countdown = 100;
187  }
188  }
189  return NULL;
190 }
__device__ bool dynamic_watchdog()
__device__ int64_t * get_matching_group_value_columnar(int64_t *groups_buffer, const uint32_t h, const int64_t *key, const uint32_t key_qw_count, const size_t entry_count)
RUNTIME_EXPORT ALWAYS_INLINE DEVICE uint32_t key_hash(const int64_t *key, const uint32_t key_count, const uint32_t key_byte_width)

+ Here is the call graph for this function:

RUNTIME_EXPORT int64_t* get_group_value_fast ( int64_t *  groups_buffer,
const int64_t  key,
const int64_t  min_key,
const int64_t  bucket,
const uint32_t  row_size_quad 
)

Definition at line 192 of file GroupByRuntime.cpp.

References EMPTY_KEY_64.

197  {
198  int64_t key_diff = key - min_key;
199  if (bucket) {
200  key_diff /= bucket;
201  }
202  int64_t off = key_diff * row_size_quad;
203  if (groups_buffer[off] == EMPTY_KEY_64) {
204  groups_buffer[off] = key;
205  }
206  return groups_buffer + off + 1;
207 }
#define EMPTY_KEY_64
RUNTIME_EXPORT int64_t* get_group_value_fast_with_original_key ( int64_t *  groups_buffer,
const int64_t  key,
const int64_t  orig_key,
const int64_t  min_key,
const int64_t  bucket,
const uint32_t  row_size_quad 
)

Definition at line 210 of file GroupByRuntime.cpp.

References EMPTY_KEY_64.

215  {
216  int64_t key_diff = key - min_key;
217  if (bucket) {
218  key_diff /= bucket;
219  }
220  int64_t off = key_diff * row_size_quad;
221  if (groups_buffer[off] == EMPTY_KEY_64) {
222  groups_buffer[off] = orig_key;
223  }
224  return groups_buffer + off + 1;
225 }
#define EMPTY_KEY_64
RUNTIME_EXPORT int64_t* get_group_value_with_watchdog ( int64_t *  groups_buffer,
const uint32_t  groups_buffer_entry_count,
const int64_t *  key,
const uint32_t  key_count,
const uint32_t  key_width,
const uint32_t  row_size_quad 
)

Definition at line 52 of file GroupByRuntime.cpp.

References dynamic_watchdog(), get_matching_group_value(), and key_hash().

58  {
59  uint32_t h = key_hash(key, key_count, key_width) % groups_buffer_entry_count;
60  int64_t* matching_group = get_matching_group_value(
61  groups_buffer, h, key, key_count, key_width, row_size_quad);
62  if (matching_group) {
63  return matching_group;
64  }
65  uint32_t watchdog_countdown = 100;
66  uint32_t h_probe = (h + 1) % groups_buffer_entry_count;
67  while (h_probe != h) {
68  matching_group = get_matching_group_value(
69  groups_buffer, h_probe, key, key_count, key_width, row_size_quad);
70  if (matching_group) {
71  return matching_group;
72  }
73  h_probe = (h_probe + 1) % groups_buffer_entry_count;
74  if (--watchdog_countdown == 0) {
75  if (dynamic_watchdog()) {
76  return NULL;
77  }
78  watchdog_countdown = 100;
79  }
80  }
81  return NULL;
82 }
__device__ bool dynamic_watchdog()
__device__ int64_t * get_matching_group_value(int64_t *groups_buffer, const uint32_t h, const T *key, const uint32_t key_count, const uint32_t row_size_quad)
RUNTIME_EXPORT ALWAYS_INLINE DEVICE uint32_t key_hash(const int64_t *key, const uint32_t key_count, const uint32_t key_byte_width)

+ Here is the call graph for this function:

int32_t* get_hash_slot ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key 
)

Definition at line 39 of file JoinHashImpl.h.

Referenced by count_matches(), fill_hash_join_buff(), fill_row_ids(), and hash_join_idx().

41  {
42  return buff + (key - min_key);
43 }

+ Here is the caller graph for this function:

int32_t* get_hash_slot_sharded ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key,
const uint32_t  entry_count_per_shard,
const uint32_t  num_shards,
const uint32_t  device_count 
)

Definition at line 60 of file JoinHashImpl.h.

References SHARD_FOR_KEY.

Referenced by count_matches_sharded(), fill_row_ids_sharded(), and hash_join_idx_sharded().

66  {
67  const uint32_t shard = SHARD_FOR_KEY(key, num_shards);
68  const uint32_t shard_buffer_index =
69  shard / device_count; // shard sub-buffer index within `buff`
70  int32_t* shard_buffer = buff + shard_buffer_index * entry_count_per_shard;
71  return shard_buffer + (key - min_key) / num_shards;
72 }
#define SHARD_FOR_KEY(key, num_shards)
Definition: shard_key.h:20

+ Here is the caller graph for this function:

int32_t* get_hash_slot_sharded_opt ( int32_t *  buff,
const int64_t  key,
const int64_t  min_key,
const uint32_t  entry_count_per_shard,
const uint32_t  shard,
const uint32_t  num_shards,
const uint32_t  device_count 
)

Definition at line 89 of file JoinHashImpl.h.

Referenced by fill_hash_join_buff_sharded().

96  {
97  const uint32_t shard_buffer_index =
98  shard / device_count; // shard sub-buffer index within `buff`
99  int32_t* shard_buffer = buff + shard_buffer_index * entry_count_per_shard;
100  return shard_buffer + (key - min_key) / num_shards;
101 }

+ Here is the caller graph for this function:

int64_t* get_matching_group_value_perfect_hash ( int64_t *  groups_buffer,
const uint32_t  h,
const int64_t *  key,
const uint32_t  key_qw_count,
const uint32_t  row_size_quad 
)

Definition at line 1144 of file RuntimeFunctions.cpp.

References EMPTY_KEY_64, and i.

1149  {
1150  uint32_t off = hashed_index * row_size_quad;
1151  if (groups_buffer[off] == EMPTY_KEY_64) {
1152  for (uint32_t i = 0; i < key_count; ++i) {
1153  groups_buffer[off + i] = key[i];
1154  }
1155  }
1156  return groups_buffer + off + key_count;
1157 }
#define EMPTY_KEY_64
int64_t* get_matching_group_value_perfect_hash_keyless ( int64_t *  groups_buffer,
const uint32_t  hashed_index,
const uint32_t  row_size_quad 
)

For a particular hashed index (only used with multi-column perfect hash group by) it returns the row-wise offset of the group in the output buffer. Since it is intended for keyless hash use, it assumes there is no group columns prepending the output buffer.

Definition at line 1165 of file RuntimeFunctions.cpp.

1168  {
1169  return groups_buffer + row_size_quad * hashed_index;
1170 }
RUNTIME_EXPORT uint32_t key_hash ( const int64_t *  key,
const uint32_t  key_qw_count,
const uint32_t  key_byte_width 
)

Definition at line 21 of file GroupByRuntime.cpp.

References MurmurHash1().

Referenced by get_group_value(), get_group_value_columnar(), anonymous_namespace{ResultSetReduction.cpp}::get_group_value_columnar_reduction(), get_group_value_columnar_slot(), get_group_value_columnar_slot_with_watchdog(), get_group_value_columnar_with_watchdog(), result_set::get_group_value_reduction(), and get_group_value_with_watchdog().

21  {
22  return MurmurHash1(key, key_byte_width * key_count, 0);
23 }
RUNTIME_EXPORT NEVER_INLINE DEVICE uint32_t MurmurHash1(const void *key, int len, const uint32_t seed)
Definition: MurmurHash.cpp:20

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

void linear_probabilistic_count ( uint8_t *  bitmap,
const uint32_t  bitmap_bytes,
const uint8_t *  key_bytes,
const uint32_t  key_len 
)

Definition at line 1205 of file cuda_mapd_rt.cu.

References MurmurHash1().

1208  {
1209  const uint32_t bit_pos = MurmurHash1(key_bytes, key_len, 0) % (bitmap_bytes * 8);
1210  const uint32_t word_idx = bit_pos / 32;
1211  const uint32_t bit_idx = bit_pos % 32;
1212  atomicOr(((uint32_t*)bitmap) + word_idx, 1 << bit_idx);
1213 }
RUNTIME_EXPORT NEVER_INLINE DEVICE uint32_t MurmurHash1(const void *key, int len, const uint32_t seed)
Definition: MurmurHash.cpp:20

+ Here is the call graph for this function: