greg/sparsemap
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity

cleanup

Browse source
This commit is contained in:
Gregory Burd 2024-04-15 10:57:22 -04:00
parent 5d5c7f1584
commit 6d2e355e19
12 changed files with 1409 additions and 317 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
Makefile
View file

@ -12,7 +12,7 @@ CFLAGS = -DSPARSEMAP_DIAGNOSTIC -DSPARSEMAP_ASSERT -Wall -Wextra -Wpedantic -Og
TEST_FLAGS = -Wall -Wextra -Wpedantic -Og -g -std=c11 -Iinclude/ -Itests/ -fPIC TEST_FLAGS = -Wall -Wextra -Wpedantic -Og -g -std=c11 -Iinclude/ -Itests/ -fPIC
TESTS = tests/test TESTS = tests/test
TEST_OBJS = tests/test.o tests/munit.o tests/common.o TEST_OBJS = tests/test.o tests/munit.o tests/tdigest.o tests/common.o
EXAMPLES = examples/ex_1 examples/ex_2 examples/ex_3 examples/ex_4 EXAMPLES = examples/ex_1 examples/ex_2 examples/ex_3 examples/ex_4
.PHONY: all shared static clean test examples mls .PHONY: all shared static clean test examples mls
@ -39,7 +39,7 @@ check: test
env ASAN_OPTIONS=detect_leaks=1 LSAN_OPTIONS=verbosity=1:log_threads=1 ./tests/test env ASAN_OPTIONS=detect_leaks=1 LSAN_OPTIONS=verbosity=1:log_threads=1 ./tests/test
tests/test: $(TEST_OBJS) $(STATIC_LIB) tests/test: $(TEST_OBJS) $(STATIC_LIB)
$(CC) $^ -o $@ $(TEST_FLAGS) $(CC) $^ -lm -o $@ $(TEST_FLAGS)
clean: clean:
rm -f $(OBJS) rm -f $(OBJS)
@ -76,4 +76,7 @@ examples/ex_3: examples/common.o examples/ex_3.o $(STATIC_LIB)
examples/ex_4: examples/common.o examples/ex_4.o $(STATIC_LIB) examples/ex_4: examples/common.o examples/ex_4.o $(STATIC_LIB)
$(CC) $^ -o $@ $(CFLAGS) $(TEST_FLAGS) $(CC) $^ -o $@ $(CFLAGS) $(TEST_FLAGS)
todo:
rg -i 'todo|gsb'
# cp src/sparsemap.c /tmp && clang-tidy src/sparsemap.c -fix -fix-errors -checks="readability-braces-around-statements" -- -DDEBUG -DSPARSEMAP_DIAGNOSTIC -DSPARSEMAP_ASSERT -Wall -Wextra -Wpedantic -Og -g -std=c11 -Iinclude/ -fPIC # cp src/sparsemap.c /tmp && clang-tidy src/sparsemap.c -fix -fix-errors -checks="readability-braces-around-statements" -- -DDEBUG -DSPARSEMAP_DIAGNOSTIC -DSPARSEMAP_ASSERT -Wall -Wextra -Wpedantic -Og -g -std=c11 -Iinclude/ -fPIC

13
examples/ex_1.c
View file

@ -1,4 +1,7 @@
#include <assert.h> #include <assert.h>
#include <stdbool.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h> #include <stdio.h>
#include "../include/sparsemap.h" #include "../include/sparsemap.h"
@ -23,7 +26,9 @@ int
main() main()
{ {
size_t size = 4; size_t size = 4;
setbuf(stderr, 0); // disable buffering setvbuf(stdout, NULL, _IONBF, 0); // Disable buffering for stdout
setvbuf(stderr, NULL, _IONBF, 0); // Disable buffering for stdout
__diag("Please wait a moment..."); __diag("Please wait a moment...");
sparsemap_t mmap, *map = &mmap; sparsemap_t mmap, *map = &mmap;
uint8_t buffer[1024]; uint8_t buffer[1024];
@ -135,7 +140,7 @@ main()
sparsemap_set(map, i, true); sparsemap_set(map, i, true);
} }
for (int i = 0; i < 100000; i++) { for (int i = 0; i < 100000; i++) {
assert(sparsemap_select(map, i) == (unsigned)i); assert(sparsemap_select(map, i, true) == (unsigned)i);
} }
sparsemap_clear(map); sparsemap_clear(map);
@ -145,7 +150,7 @@ main()
sparsemap_set(map, i, true); sparsemap_set(map, i, true);
} }
for (int i = 1; i < 513; i++) { for (int i = 1; i < 513; i++) {
assert(sparsemap_select(map, i - 1) == (unsigned)i); assert(sparsemap_select(map, i - 1, true) == (unsigned)i);
} }
sparsemap_clear(map); sparsemap_clear(map);
@ -155,7 +160,7 @@ main()
sparsemap_set(map, i * 10, true); sparsemap_set(map, i * 10, true);
} }
for (size_t i = 0; i < 8; i++) { for (size_t i = 0; i < 8; i++) {
assert(sparsemap_select(map, i) == i * 10); assert(sparsemap_select(map, i, true) == (sm_loc_t)i * 10);
} }
// split and move, aligned to MiniMap capacity // split and move, aligned to MiniMap capacity

21
examples/ex_2.c
View file

@ -1,9 +1,7 @@
#include <assert.h> #include <assert.h>
#include <stdarg.h> #include <stdbool.h>
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <time.h>
#include <unistd.h>
#include "../include/sparsemap.h" #include "../include/sparsemap.h"
@ -16,29 +14,28 @@
} while (0) } while (0)
#pragma GCC diagnostic pop #pragma GCC diagnostic pop
#define SEED
int int
main(void) main(void)
{ {
int i = 0; int i;
// disable buffering // disable buffering
setbuf(stderr, 0); setvbuf(stdout, NULL, _IONBF, 0); // Disable buffering for stdout
setvbuf(stderr, NULL, _IONBF, 0); // Disable buffering for stdout
// start with a 1KiB buffer, 1024 bits // start with a 1KiB buffer, 1024 bits
uint8_t *buf = calloc(1024, sizeof(uint8_t)); uint8_t *buf = calloc(1024, sizeof(uint8_t));
// create the sparse bitmap // create the sparse bitmap
sparsemap_t *map = sparsemap(buf, sizeof(uint8_t) * 1024); sparsemap_t *map = sparsemap_wrap(buf, sizeof(uint8_t) * 1024);
// Set every other bit (pathologically worst case) to see what happens // Set every other bit (pathologically worst case) to see what happens
// when the map is full. // when the map is full.
for (i = 0; i < 7744; i++) { for (i = 0; i < 7744; i++) {
if (i % 2) if (!i % 2) {
continue; sparsemap_set(map, i, true);
sparsemap_set(map, i, true); assert(sparsemap_is_set(map, i) == true);
assert(sparsemap_is_set(map, i) == true); }
} }
// On 1024 KiB of buffer with every other bit set the map holds 7744 bits // On 1024 KiB of buffer with every other bit set the map holds 7744 bits
// and then runs out of space. This next _set() call will fail/abort. // and then runs out of space. This next _set() call will fail/abort.

8
examples/ex_3.c
View file

@ -1,9 +1,7 @@
#include <assert.h> #include <assert.h>
#include <stdarg.h> #include <stdbool.h>
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <time.h>
#include <unistd.h>
#include "../include/sparsemap.h" #include "../include/sparsemap.h"
#include "../tests/common.h" #include "../tests/common.h"
@ -11,7 +9,7 @@
int int
main(void) main(void)
{ {
int i = 0; int i;
int array[1024] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, int array[1024] = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37,
38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76,
77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112,
@ -60,7 +58,7 @@ main(void)
uint8_t *buf = calloc(1024, sizeof(uint8_t)); uint8_t *buf = calloc(1024, sizeof(uint8_t));
// create the sparse bitmap // create the sparse bitmap
sparsemap_t *map = sparsemap(buf, sizeof(uint8_t) * 1024); sparsemap_t *map = sparsemap_wrap(buf, sizeof(uint8_t) * 1024);
// set all the bits on in a random order // set all the bits on in a random order
for (i = 0; i < 1024; i++) { for (i = 0; i < 1024; i++) {

4
examples/ex_4.c
View file

@ -24,7 +24,7 @@ main(void)
uint8_t *buf = calloc((size_t)3 * 1024, sizeof(uint8_t)); uint8_t *buf = calloc((size_t)3 * 1024, sizeof(uint8_t));
// create the sparse bitmap // create the sparse bitmap
sparsemap_t *map = sparsemap(buf, sizeof(uint8_t) * 3 * 1024); sparsemap_t *map = sparsemap_wrap(buf, sizeof(uint8_t) * 3 * 1024);
// create an array of ints // create an array of ints
setup_test_array(array, TEST_ARRAY_SIZE, 1024 * 3); setup_test_array(array, TEST_ARRAY_SIZE, 1024 * 3);
@ -60,7 +60,7 @@ main(void)
assert(sparsemap_is_set(map, array[i]) == true); assert(sparsemap_is_set(map, array[i]) == true);
} }
has_span(map, array, TEST_ARRAY_SIZE, (int)len); has_span(map, array, TEST_ARRAY_SIZE, (int)len);
size_t l = sparsemap_span(map, 0, len); size_t l = sparsemap_span(map, 0, len, true);
if (l != (size_t)-1) { if (l != (size_t)-1) {
__diag("Found span in map starting at %lu of length %lu\n", l, len); __diag("Found span in map starting at %lu of length %lu\n", l, len);
__diag("is_span(%lu, %lu) == %s\n", l, len, is_span(array, TEST_ARRAY_SIZE, l, len) ? "yes" : "no"); __diag("is_span(%lu, %lu) == %s\n", l, len, is_span(array, TEST_ARRAY_SIZE, l, len) ? "yes" : "no");

123
include/sparsemap.h
View file

@ -69,14 +69,14 @@
#ifndef SPARSEMAP_H #ifndef SPARSEMAP_H
#define SPARSEMAP_H #define SPARSEMAP_H
#include <sys/types.h>
#include <assert.h>
#include <limits.h> #include <limits.h>
#include <stdbool.h> #include <stdbool.h>
#include <stddef.h>
#include <stdint.h> #include <stdint.h>
#include <stdio.h>
#include <string.h> #if defined(__cplusplus)
extern "C" {
#endif
/* /*
* The public interface for a sparse bit-mapped index, a "sparse map". * The public interface for a sparse bit-mapped index, a "sparse map".
@ -88,55 +88,114 @@
*/ */
typedef struct sparsemap sparsemap_t; typedef struct sparsemap sparsemap_t;
typedef long sm_loc_t;
#define SM_LOC_MAX LONG_MAX
#define SM_LOC_MIN LONG_MIN
typedef uint32_t sm_idx_t; typedef uint32_t sm_idx_t;
typedef uint64_t sm_bitvec_t; typedef uint64_t sm_bitvec_t;
/* Allocate on a sparsemap_t on the heap and initialize it. */ /**
sparsemap_t *sparsemap(uint8_t *data, size_t size); * Create a new, empty sparsemap_t with a buffer of |size|.
* Default when set to 0 is 1024.
*/
sparsemap_t *sparsemap(size_t size);
/* Initialize sparsemap_t with data. */ /**
* Allocate on a sparsemap_t on the heap to wrap the provided fixed-size
* buffer (heap or stack allocated).
*/
sparsemap_t *sparsemap_wrap(uint8_t *data, size_t size);
/**
* Initialize a (possibly stack allocated) sparsemap_t with data (potentially
* also on the stack).
*/
void sparsemap_init(sparsemap_t *map, uint8_t *data, size_t size); void sparsemap_init(sparsemap_t *map, uint8_t *data, size_t size);
/* Clears the whole buffer. */ /**
void sparsemap_clear(sparsemap_t *map); * Opens an existing sparsemap contained within the specified buffer.
*/
/* Opens an existing sparsemap at the specified buffer. */
void sparsemap_open(sparsemap_t *, uint8_t *data, size_t data_size); void sparsemap_open(sparsemap_t *, uint8_t *data, size_t data_size);
/* Resizes the data range. */ /**
* Resets values and empties the buffer making it ready to accept new data.
*/
void sparsemap_clear(sparsemap_t *map);
/**
* Resizes the data range within the limits of the provided buffer.
*/
void sparsemap_set_data_size(sparsemap_t *map, size_t data_size); void sparsemap_set_data_size(sparsemap_t *map, size_t data_size);
/* Calculate remaining capacity, full when 0. */ /**
* Calculate remaining capacity, approaches 0 when full.
*/
double sparsemap_capacity_remaining(sparsemap_t *map); double sparsemap_capacity_remaining(sparsemap_t *map);
/* Returns the size of the underlying byte array. */ /**
* Returns the capacity of the underlying byte array.
*/
size_t sparsemap_get_capacity(sparsemap_t *map); size_t sparsemap_get_capacity(sparsemap_t *map);
/* Returns the value of a bit at index |idx|. */ /**
bool sparsemap_is_set(sparsemap_t *map, size_t idx); * Returns the value of a bit at index |idx|, either on/true/1 or off/false/0.
* When |idx| is negative it is an error.
*/
bool sparsemap_is_set(sparsemap_t *map, sm_loc_t idx);
/* Sets the bit at index |idx| to true or false, depending on |value|. */ /**
void sparsemap_set(sparsemap_t *map, size_t idx, bool value); * Sets the bit at index |idx| to true or false, depending on |value|.
* When |idx| is negative is it an error.
*/
void sparsemap_set(sparsemap_t *map, sm_loc_t idx, bool value);
/* Returns the offset of the very first bit. */ /**
sm_idx_t sparsemap_get_start_offset(sparsemap_t *map); * Returns the offset of the very first/last bit in the map.
*/
sm_idx_t sparsemap_get_starting_offset(sparsemap_t *map);
/* Returns the used size in the data buffer. */ /**
* Returns the used size in the data buffer in bytes.
*/
size_t sparsemap_get_size(sparsemap_t *map); size_t sparsemap_get_size(sparsemap_t *map);
/* Decompresses the whole bitmap; calls scanner for all bits. */ /**
void sparsemap_scan(sparsemap_t *map, void (*scanner)(sm_idx_t[], size_t), size_t skip); * Decompresses the whole bitmap; calls scanner for all bits with a set of
* |n| vectors |vec| each a sm_bitmap_t which can be masked and read using
* bit operators to read the values for each position in the bitmap index.
* Setting |skip| will start the scan after "skip" bits.
*/
void sparsemap_scan(sparsemap_t *map, void (*scanner)(sm_idx_t vec[], size_t n), size_t skip);
/* Appends all chunk maps from |map| starting at |sstart| to |other|, then /**
reduces the chunk map-count appropriately. */ * Appends all chunk maps from |map| starting at |offset| to |other|, then
void sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other); * reduces the chunk map-count appropriately.
*/
void sparsemap_split(sparsemap_t *map, sm_loc_t offset, sparsemap_t *other);
/* Returns the index of the n'th set bit; uses a 0-based index. */ /**
size_t sparsemap_select(sparsemap_t *map, size_t n); * Finds the offset of the n'th bit either set (|value| is true) or unset
* (|value| is false) from the start (positive |n|), or end (negative |n|),
* of the bitmap and returns that (uses a 0-based index). Returns -inf or +inf
* if not found (where "inf" is SM_LOC_MAX and "-inf" is SM_LOC_MIN).
*/
sm_loc_t sparsemap_select(sparsemap_t *map, sm_loc_t n, bool value);
/* Counts the set bits in the range [offset, idx]. */ /**
size_t sparsemap_rank(sparsemap_t *map, size_t offset, size_t idx); * Counts the set (|value| is true) or unset (|value| is false) bits starting
* at |x| bits (0-based) in the range [x, y] (inclusive on either end).
*/
size_t sparsemap_rank(sparsemap_t *map, size_t x, size_t y, bool value);
size_t sparsemap_span(sparsemap_t *map, size_t loc, size_t len); /**
* Finds the first span (i.e. a contiguous set of bits), in the bitmap that
* are set (|value| is true) or unset (|value| is false) and returns the
* starting offset for the span (0-based).
*/
size_t sparsemap_span(sparsemap_t *map, sm_loc_t idx, size_t len, bool value);
#if defined(__cplusplus)
}
#endif #endif
#endif /* !defined(SPARSEMAP_H) */

269
src/sparsemap.c
View file

@ -20,12 +20,18 @@
* SOFTWARE. * SOFTWARE.
*/ */
#include <sys/types.h>
#include <assert.h> #include <assert.h>
#include <popcount.h> #include <popcount.h>
#include <sparsemap.h> #include <sparsemap.h>
#include <stdbool.h> #include <stdbool.h>
#include <stdint.h> #include <stdint.h>
#include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <stdarg.h>
#include <stddef.h>
#include <string.h>
#ifdef SPARSEMAP_DIAGNOSTIC #ifdef SPARSEMAP_DIAGNOSTIC
#pragma GCC diagnostic push #pragma GCC diagnostic push
@ -83,7 +89,7 @@ enum __SM_CHUNK_INFO {
/* sm_bitvec_t is not used (2#01) */ /* sm_bitvec_t is not used (2#01) */
SM_PAYLOAD_NONE = 1, SM_PAYLOAD_NONE = 1,
/* a mask for checking flags (2 bits) */ /* a mask for checking flags (2 bits, 2#11) */
SM_FLAG_MASK = 3, SM_FLAG_MASK = 3,
/* return code for set(): ok, no further action required */ /* return code for set(): ok, no further action required */
@ -388,10 +394,11 @@ __sm_chunk_map_set(__sm_chunk_t *map, size_t idx, bool value, size_t *pos, sm_bi
* value of |n|. * value of |n|.
*/ */
static size_t static size_t
__sm_chunk_map_select(__sm_chunk_t *map, size_t n, ssize_t *pnew_n) __sm_chunk_map_select(__sm_chunk_t *map, size_t n, ssize_t *pnew_n, bool value)
{ {
size_t ret = 0; size_t ret = 0;
register uint8_t *p; register uint8_t *p;
(void)value; // TODO
p = (uint8_t *)map->m_data; p = (uint8_t *)map->m_data;
for (size_t i = 0; i < sizeof(sm_bitvec_t); i++, p++) { for (size_t i = 0; i < sizeof(sm_bitvec_t); i++, p++) {
@ -446,7 +453,7 @@ __sm_chunk_map_select(__sm_chunk_t *map, size_t n, ssize_t *pnew_n)
* '*offset' has been reached 0. * '*offset' has been reached 0.
*/ */
static size_t static size_t
__sm_chunk_map_rank(__sm_chunk_t *map, size_t *offset, size_t idx) __sm_chunk_map_rank(__sm_chunk_t *map, size_t *offset, size_t idx, sm_bitvec_t *vec)
{ {
size_t ret = 0; size_t ret = 0;
@ -466,6 +473,7 @@ __sm_chunk_map_rank(__sm_chunk_t *map, size_t *offset, size_t idx)
*offset = 0; *offset = 0;
} }
} else { } else {
*vec = 0;
return (ret); return (ret);
} }
} else if (flags == SM_PAYLOAD_ONES) { } else if (flags == SM_PAYLOAD_ONES) {
@ -480,6 +488,7 @@ __sm_chunk_map_rank(__sm_chunk_t *map, size_t *offset, size_t idx)
*offset = 0; *offset = 0;
} }
} else { } else {
*vec = UINT64_MAX;
return (ret + idx); return (ret + idx);
} }
} else if (flags == SM_PAYLOAD_MIXED) { } else if (flags == SM_PAYLOAD_MIXED) {
@ -493,8 +502,11 @@ __sm_chunk_map_rank(__sm_chunk_t *map, size_t *offset, size_t idx)
/* Create a mask for the range between offset and idx inclusive [*offset, idx]. */ /* Create a mask for the range between offset and idx inclusive [*offset, idx]. */
uint64_t offset_mask = (((uint64_t)1 << *offset) - 1); uint64_t offset_mask = (((uint64_t)1 << *offset) - 1);
uint64_t idx_mask = idx >= 63 ? UINT64_MAX : ((uint64_t)1 << (idx + 1)) - 1; uint64_t idx_mask = idx >= 63 ? UINT64_MAX : ((uint64_t)1 << (idx + 1)) - 1;
ret += popcountll(w & (idx_mask - offset_mask)); sm_bitvec_t mw = w & (idx_mask - offset_mask);
ret += popcountll(mw);
*offset = *offset > idx ? *offset - idx : 0; *offset = *offset > idx ? *offset - idx : 0;
*vec = mw;
(*vec) <<= *offset;
return (ret); return (ret);
} }
} }
@ -593,7 +605,7 @@ __sm_get_chunk_map_count(sparsemap_t *map)
static inline uint8_t * static inline uint8_t *
__sm_get_chunk_map_data(sparsemap_t *map, size_t offset) __sm_get_chunk_map_data(sparsemap_t *map, size_t offset)
{ {
return (uint8_t *)(&map->m_data[SM_SIZEOF_OVERHEAD + offset]); return (&map->m_data[SM_SIZEOF_OVERHEAD + offset]);
} }
/** /**
@ -644,33 +656,51 @@ __sm_get_aligned_offset(size_t idx)
} }
/** /**
* Returns the byte offset of a __sm_chunk_t in m_data * Returns the byte offset of a __sm_chunk_t in m_data.
*/ */
static ssize_t static ssize_t
__sm_get_chunk_map_offset(sparsemap_t *map, size_t idx) __sm_get_chunk_map_offset(sparsemap_t *map, sm_loc_t idx)
{ {
size_t count; int count;
count = __sm_get_chunk_map_count(map); count = __sm_get_chunk_map_count(map);
if (count == 0) { if (count == 0) {
return (-1); return (-1);
} }
uint8_t *start = __sm_get_chunk_map_data(map, 0); if (idx > 0 || idx == 0) {
uint8_t *p = start; uint8_t *start = __sm_get_chunk_map_data(map, 0);
uint8_t *p = start;
for (size_t i = 0; i < count - 1; i++) { for (sm_loc_t i = 0; i < count - 1; i++) {
sm_idx_t start = *(sm_idx_t *)p; sm_idx_t s = *(sm_idx_t *)p;
__sm_assert(start == __sm_get_aligned_offset(start)); __sm_assert(s == __sm_get_aligned_offset(start));
__sm_chunk_t chunk; __sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t)); __sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t));
if (start >= idx || idx < start + __sm_chunk_map_get_capacity(&chunk)) { if (s >= idx || (unsigned long)idx < s + __sm_chunk_map_get_capacity(&chunk)) {
break; break;
}
p += sizeof(sm_idx_t) + __sm_chunk_map_get_size(&chunk);
} }
p += sizeof(sm_idx_t) + __sm_chunk_map_get_size(&chunk);
}
return ((ssize_t)(p - start)); return ((ssize_t)(p - start));
} else {
uint8_t *end = __sm_get_chunk_map_data(map, count - 1);
uint8_t *p = end;
for (sm_loc_t i = count - 1; i >= 0; i--) {
sm_idx_t e = *(sm_idx_t *)p;
__sm_assert(e == __sm_get_aligned_offset(end));
__sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t));
if (e >= idx || (unsigned long)idx < e + __sm_chunk_map_get_capacity(&chunk)) {
break;
}
p += sizeof(sm_idx_t) + __sm_chunk_map_get_size(&chunk);
}
return ((ssize_t)(p - end));
}
} }
/** /**
@ -743,11 +773,21 @@ sparsemap_clear(sparsemap_t *map)
__sm_set_chunk_map_count(map, 0); __sm_set_chunk_map_count(map, 0);
} }
/**
* Allocate on a sparsemap_t on the heap and initialize it.
*/
sparsemap_t * sparsemap_t *
sparsemap(uint8_t *data, size_t size) sparsemap(size_t size)
{
if (size == 0) {
size = 1024;
}
sparsemap_t *map = (sparsemap_t *)calloc(1, sizeof(sparsemap_t) + (size * sizeof(uint8_t)));
if (map) {
sparsemap_init(map, (uint8_t *)map + sizeof(sparsemap_t), size);
}
return map;
}
sparsemap_t *
sparsemap_wrap(uint8_t *data, size_t size)
{ {
sparsemap_t *map = (sparsemap_t *)calloc(1, sizeof(sparsemap_t)); sparsemap_t *map = (sparsemap_t *)calloc(1, sizeof(sparsemap_t));
if (map) { if (map) {
@ -756,9 +796,6 @@ sparsemap(uint8_t *data, size_t size)
return map; return map;
} }
/**
* Initialize sparsemap_t with data.
*/
void void
sparsemap_init(sparsemap_t *map, uint8_t *data, size_t size) sparsemap_init(sparsemap_t *map, uint8_t *data, size_t size)
{ {
@ -768,9 +805,6 @@ sparsemap_init(sparsemap_t *map, uint8_t *data, size_t size)
sparsemap_clear(map); sparsemap_clear(map);
} }
/**
* Opens an existing sparsemap at the specified buffer.
*/
void void
sparsemap_open(sparsemap_t *map, uint8_t *data, size_t data_size) sparsemap_open(sparsemap_t *map, uint8_t *data, size_t data_size)
{ {
@ -779,9 +813,7 @@ sparsemap_open(sparsemap_t *map, uint8_t *data, size_t data_size)
map->m_capacity = data_size; map->m_capacity = data_size;
} }
/** /*
* Resizes the data range.
*
* TODO/NOTE: This is a dangerous operation because we cannot verify that * TODO/NOTE: This is a dangerous operation because we cannot verify that
* data_size is not exceeding the size of the underlying buffer. * data_size is not exceeding the size of the underlying buffer.
*/ */
@ -791,10 +823,6 @@ sparsemap_set_data_size(sparsemap_t *map, size_t data_size)
map->m_capacity = data_size; map->m_capacity = data_size;
} }
/**
* Calculates the remaining capacity as an integer that approaches 0 to
* indicate full.
*/
double double
sparsemap_capacity_remaining(sparsemap_t *map) sparsemap_capacity_remaining(sparsemap_t *map)
{ {
@ -807,23 +835,21 @@ sparsemap_capacity_remaining(sparsemap_t *map)
return 100 - (((double)map->m_data_used / (double)map->m_capacity) * 100); return 100 - (((double)map->m_data_used / (double)map->m_capacity) * 100);
} }
/**
* Returns the size of the underlying byte array.
*/
size_t size_t
sparsemap_get_capacity(sparsemap_t *map) sparsemap_get_capacity(sparsemap_t *map)
{ {
return (map->m_capacity); return (map->m_capacity);
} }
/**
* Returns the value of a bit at index |idx|.
*/
bool bool
sparsemap_is_set(sparsemap_t *map, size_t idx) sparsemap_is_set(sparsemap_t *map, sm_loc_t idx)
{ {
__sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD); __sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD);
if (idx < 0) {
return (false);
}
/* Get the __sm_chunk_t which manages this index */ /* Get the __sm_chunk_t which manages this index */
ssize_t offset = __sm_get_chunk_map_offset(map, idx); ssize_t offset = __sm_get_chunk_map_offset(map, idx);
@ -839,8 +865,8 @@ sparsemap_is_set(sparsemap_t *map, size_t idx)
__sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t)); __sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t));
/* Determine if the bit is out of bounds of the __sm_chunk_t; if yes then /* Determine if the bit is out of bounds of the __sm_chunk_t; if yes then
the bit is not set. */ the bit is not set. */
if (idx < start || idx - start >= __sm_chunk_map_get_capacity(&chunk)) { if (idx < start || (unsigned long)idx - start >= __sm_chunk_map_get_capacity(&chunk)) {
return (false); return (false);
} }
@ -848,14 +874,15 @@ sparsemap_is_set(sparsemap_t *map, size_t idx)
return (__sm_chunk_map_is_set(&chunk, idx - start)); return (__sm_chunk_map_is_set(&chunk, idx - start));
} }
/**
* Sets the bit at index |idx| to true or false, depending on |value|.
*/
void void
sparsemap_set(sparsemap_t *map, size_t idx, bool value) sparsemap_set(sparsemap_t *map, sm_loc_t idx, bool value)
{ {
__sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD); __sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD);
if (idx < 0) {
return;
}
/* Get the __sm_chunk_t which manages this index */ /* Get the __sm_chunk_t which manages this index */
ssize_t offset = __sm_get_chunk_map_offset(map, idx); ssize_t offset = __sm_get_chunk_map_offset(map, idx);
bool dont_grow = false; bool dont_grow = false;
@ -917,7 +944,7 @@ sparsemap_set(sparsemap_t *map, size_t idx, bool value)
else { else {
__sm_chunk_t chunk; __sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t)); __sm_chunk_map_init(&chunk, p + sizeof(sm_idx_t));
if (idx - start >= __sm_chunk_map_get_capacity(&chunk)) { if (idx - (unsigned long)start >= __sm_chunk_map_get_capacity(&chunk)) {
if (value == false) { if (value == false) {
/* nothing to do */ /* nothing to do */
return; return;
@ -931,7 +958,7 @@ sparsemap_set(sparsemap_t *map, size_t idx, bool value)
__sm_insert_data(map, offset, &buf[0], sizeof(buf)); __sm_insert_data(map, offset, &buf[0], sizeof(buf));
start += __sm_chunk_map_get_capacity(&chunk); start += __sm_chunk_map_get_capacity(&chunk);
if ((size_t)start + SM_CHUNK_MAX_CAPACITY < idx) { if ((size_t)start + SM_CHUNK_MAX_CAPACITY < (unsigned long)idx) {
start = __sm_get_fully_aligned_offset(idx); start = __sm_get_fully_aligned_offset(idx);
} }
*(sm_idx_t *)p = start; *(sm_idx_t *)p = start;
@ -984,16 +1011,15 @@ sparsemap_set(sparsemap_t *map, size_t idx, bool value)
__sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD); __sm_assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD);
} }
/**
* Returns the offset of the very first bit.
*/
sm_idx_t sm_idx_t
sparsemap_get_start_offset(sparsemap_t *map) sparsemap_get_starting_offset(sparsemap_t *map)
{ {
if (__sm_get_chunk_map_count(map) == 0) { size_t count = __sm_get_chunk_map_count(map);
if (count == 0) {
return (0); return (0);
} }
return (*(sm_idx_t *)__sm_get_chunk_map_data(map, 0)); sm_idx_t *chunk = (sm_idx_t *)__sm_get_chunk_map_data(map, 0);
return *chunk;
} }
/** /**
@ -1032,14 +1058,13 @@ sparsemap_scan(sparsemap_t *map, void (*scanner)(sm_idx_t[], size_t), size_t ski
} }
} }
/**
* Appends all chunk maps from |sstart| to |other|, then reduces the chunk
* map-count appropriately. |sstart| must be BitVector-aligned!
*/
void void
sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other) sparsemap_split(sparsemap_t *map, sm_loc_t offset, sparsemap_t *other)
{ {
assert(sstart % SM_BITS_PER_VECTOR == 0); assert(offset % SM_BITS_PER_VECTOR == 0);
if (offset < 0)
return;
/* |dst| points to the destination buffer */ /* |dst| points to the destination buffer */
uint8_t *dst = __sm_get_chunk_map_end(other); uint8_t *dst = __sm_get_chunk_map_end(other);
@ -1047,9 +1072,9 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
/* |src| points to the source-chunk map */ /* |src| points to the source-chunk map */
uint8_t *src = __sm_get_chunk_map_data(map, 0); uint8_t *src = __sm_get_chunk_map_data(map, 0);
/* |sstart| is relative to the beginning of this sparsemap_t; best /* |offset| is relative to the beginning of this sparsemap_t; best
make it absolute. */ make it absolute. */
sstart += *(sm_idx_t *)src; offset += *(sm_idx_t *)src;
bool in_middle = false; bool in_middle = false;
uint8_t *prev = src; uint8_t *prev = src;
@ -1058,14 +1083,14 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
sm_idx_t start = *(sm_idx_t *)src; sm_idx_t start = *(sm_idx_t *)src;
__sm_chunk_t chunk; __sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, src + sizeof(sm_idx_t)); __sm_chunk_map_init(&chunk, src + sizeof(sm_idx_t));
if (start == sstart) { if (start == offset) {
break; break;
} }
if (start + __sm_chunk_map_get_capacity(&chunk) > sstart) { if (start + __sm_chunk_map_get_capacity(&chunk) > (unsigned long)offset) {
in_middle = true; in_middle = true;
break; break;
} }
if (start > sstart) { if (start > offset) {
src = prev; src = prev;
i--; i--;
break; break;
@ -1083,12 +1108,12 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
/* Now copy all the remaining chunks. */ /* Now copy all the remaining chunks. */
int moved = 0; int moved = 0;
/* If |sstart| is in the middle of a chunk then this chunk has to be split */ /* If |offset| is in the middle of a chunk then this chunk has to be split */
if (in_middle) { if (in_middle) {
uint8_t buf[sizeof(sm_idx_t) + sizeof(sm_bitvec_t) * 2] = { 0 }; uint8_t buf[sizeof(sm_idx_t) + sizeof(sm_bitvec_t) * 2] = { 0 };
memcpy(dst, &buf[0], sizeof(buf)); memcpy(dst, &buf[0], sizeof(buf));
*(sm_idx_t *)dst = sstart; *(sm_idx_t *)dst = offset;
dst += sizeof(sm_idx_t); dst += sizeof(sm_idx_t);
/* the |other| sparsemap_t now has one additional chunk */ /* the |other| sparsemap_t now has one additional chunk */
@ -1104,11 +1129,11 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
__sm_chunk_t d_chunk; __sm_chunk_t d_chunk;
__sm_chunk_map_init(&d_chunk, dst); __sm_chunk_map_init(&d_chunk, dst);
__sm_chunk_map_set_capacity(&d_chunk, capacity - (sstart % capacity)); __sm_chunk_map_set_capacity(&d_chunk, capacity - (offset % capacity));
/* Now copy the bits. */ /* Now copy the bits. */
size_t d = sstart; sm_loc_t d = offset;
for (size_t j = sstart % capacity; j < capacity; j++, d++) { for (size_t j = offset % capacity; j < capacity; j++, d++) {
if (__sm_chunk_map_is_set(&s_chunk, j)) { if (__sm_chunk_map_is_set(&s_chunk, j)) {
sparsemap_set(other, d, true); sparsemap_set(other, d, true);
} }
@ -1120,7 +1145,7 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
i++; i++;
/* Reduce the capacity of the source-chunk map. */ /* Reduce the capacity of the source-chunk map. */
__sm_chunk_map_set_capacity(&s_chunk, sstart % capacity); __sm_chunk_map_set_capacity(&s_chunk, offset % capacity);
} }
/* Now continue with all remaining minimaps. */ /* Now continue with all remaining minimaps. */
@ -1152,93 +1177,99 @@ sparsemap_split(sparsemap_t *map, size_t sstart, sparsemap_t *other)
assert(sparsemap_get_size(other) > SM_SIZEOF_OVERHEAD); assert(sparsemap_get_size(other) > SM_SIZEOF_OVERHEAD);
} }
/** sm_loc_t
* Returns the index of the n'th set bit; uses a 0-based index, sparsemap_select(sparsemap_t *map, sm_loc_t n, bool value)
* i.e. n == 0 for the first bit which is set, n == 1 for the second bit etc.
*/
size_t
sparsemap_select(sparsemap_t *map, size_t n)
{ {
assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD); assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD);
size_t result; size_t result;
size_t count = __sm_get_chunk_map_count(map); size_t count = __sm_get_chunk_map_count(map);
uint8_t *p = __sm_get_chunk_map_data(map, 0); if (n >= 0) {
uint8_t *p = __sm_get_chunk_map_data(map, 0);
for (size_t i = 0; i < count; i++) { for (size_t i = 0; i < count; i++) {
result = *(sm_idx_t *)p; result = *(sm_idx_t *)p;
p += sizeof(sm_idx_t); p += sizeof(sm_idx_t);
__sm_chunk_t chunk; __sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, p); __sm_chunk_map_init(&chunk, p);
ssize_t new_n = (ssize_t)n; ssize_t new_n = (ssize_t)n;
size_t index = __sm_chunk_map_select(&chunk, n, &new_n); size_t index = __sm_chunk_map_select(&chunk, n, &new_n, value);
if (new_n == -1) { if (new_n == -1) {
return (result + index); return (result + index);
}
n = new_n;
p += __sm_chunk_map_get_size(&chunk);
} }
n = new_n;
p += __sm_chunk_map_get_size(&chunk);
}
#ifdef DEBUG #ifdef DEBUG
assert(!"shouldn't be here"); assert(!"shouldn't be here");
#endif #endif
return (size_t)-1; return SM_LOC_MAX;
} else {
return SM_LOC_MIN; // TODO... sparsemap_select(map, -n, value);
}
} }
/**
* Counts the set bits starting at 'offset' until and including 'idx', meaning
* [offset, idx] inclusive.
*/
size_t size_t
sparsemap_rank(sparsemap_t *map, size_t offset, size_t idx) sparsemap_rank_vec(sparsemap_t *map, size_t x, size_t y, bool value, sm_bitvec_t *vec)
{ {
(void)value; //TODO
assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD); assert(sparsemap_get_size(map) >= SM_SIZEOF_OVERHEAD);
size_t result = 0, prev = 0, count = __sm_get_chunk_map_count(map); size_t result = 0, prev = 0, count = __sm_get_chunk_map_count(map);
uint8_t *p = __sm_get_chunk_map_data(map, 0); uint8_t *p = __sm_get_chunk_map_data(map, 0);
for (size_t i = 0; i < count; i++) { for (size_t i = 0; i < count; i++) {
sm_idx_t start = *(sm_idx_t *)p; sm_idx_t start = *(sm_idx_t *)p;
if (start > idx) { if (start > y) {
return (result); return (result);
} }
offset -= start - prev; x -= start - prev;
prev = start; prev = start;
p += sizeof(sm_idx_t); p += sizeof(sm_idx_t);
__sm_chunk_t chunk; __sm_chunk_t chunk;
__sm_chunk_map_init(&chunk, p); __sm_chunk_map_init(&chunk, p);
result += __sm_chunk_map_rank(&chunk, &offset, idx - start); result += __sm_chunk_map_rank(&chunk, &x, y - start, vec);
p += __sm_chunk_map_get_size(&chunk); p += __sm_chunk_map_get_size(&chunk);
} }
return (result); return (result);
// TODO: sparsemap_rank(map, x, y, false)
} }
/**
* Finds a span of set bits of at least |len| after |loc|. Returns the index of
* the n'th set bit that starts a span of at least |len| bits set to true.
*/
size_t size_t
sparsemap_span(sparsemap_t *map, size_t loc, size_t len) sparsemap_rank(sparsemap_t *map, size_t x, size_t y, bool value)
{ {
size_t offset, nth = 0, count; sm_bitvec_t vec;
(void)loc; // TODO return sparsemap_rank_vec(map, x, y, value, &vec);
}
offset = sparsemap_select(map, 0); size_t
sparsemap_span(sparsemap_t *map, sm_loc_t idx, size_t len, bool value)
{
size_t count, nth = 0;
sm_bitvec_t vec = 0;
sm_loc_t offset;
offset = sparsemap_select(map, nth++, value);
if (len == 1) { if (len == 1) {
return offset; return offset;
} }
do { do {
count = sparsemap_rank(map, offset, offset + len); count = sparsemap_rank_vec(map, offset, offset + len, value, &vec);
if (count == len) { if (count == len) {
return offset; return offset;
} else { } else {
count = len; // TODO: what is nth when len > SM_BITS_PER_VECTOR?
while (--count && sparsemap_is_set(map, offset)) { int c = len > SM_BITS_PER_VECTOR ? SM_BITS_PER_VECTOR : len;
for (int b = 0; b < c && (vec & 1 << b); b++) {
nth++; nth++;
} }
} }
offset = sparsemap_select(map, nth); if (count)
} while (offset != ((size_t)-1)); nth++;
/* Use select to potentially jump very far forward in the map. */
offset = sparsemap_select(map, nth, value);
} while (offset != SM_LOC_MAX);
return offset; return idx > 0 ? SM_LOC_MAX : SM_LOC_MIN;
} }

91
tests/common.c
View file

@ -1,13 +1,19 @@
#include <sys/types.h> #define _POSIX_C_SOURCE 199309L
#define X86_INTRIN
#include <assert.h> #include <assert.h>
#include <pthread.h> #include <pthread.h> // If using threads
#include <sparsemap.h>
#include <stdbool.h> #include <stdbool.h>
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h> #include <unistd.h>
#ifdef X86_INTRIN
#include <x86intrin.h>
#endif
#include "../include/sparsemap.h"
#include "common.h" #include "common.h"
#pragma GCC diagnostic push #pragma GCC diagnostic push
@ -22,84 +28,25 @@
uint64_t uint64_t
tsc(void) tsc(void)
{ {
#ifdef X86_INTRIN
return __rdtsc();
#else
uint32_t low, high; uint32_t low, high;
__asm__ volatile("rdtsc" : "=a"(low), "=d"(high)); __asm__ volatile("rdtsc" : "=a"(low), "=d"(high));
return ((uint64_t)high << 32) | low; return ((uint64_t)high << 32) | low;
} #endif
static uint64_t
get_tsc_frequency()
{
uint32_t high, low;
__asm__ volatile("rdtsc" : "=a"(low), "=d"(high));
__asm__ volatile("rdtsc");
return ((uint64_t)high << 32) | low;
} }
double double
tsc_ticks_to_ns(uint64_t tsc_ticks) nsts()
{ {
static uint64_t tsc_freq = 0; struct timespec ts;
if (tsc_freq == 0) {
tsc_freq = get_tsc_frequency();
}
return (double)tsc_ticks / (double)tsc_freq * 1e9;
}
void if (clock_gettime(CLOCK_REALTIME, &ts) == -1) {
est_sift_up(uint64_t *heap, int child_index) perror("clock_gettime");
{ return -1.0; // Return -1.0 on error
while (child_index > 0) {
int parent_index = (child_index - 1) / 2;
if (heap[parent_index] > heap[child_index]) {
// Swap parent and child
uint64_t temp = heap[parent_index];
heap[parent_index] = heap[child_index];
heap[child_index] = temp;
child_index = parent_index;
} else {
break; // Heap property satisfied
}
}
}
void
est_sift_down(uint64_t *heap, int heap_size, int parent_index)
{
int child_index = 2 * parent_index + 1; // Left child
while (child_index < heap_size) {
// Right child exists and is smaller than left child
if (child_index + 1 < heap_size && heap[child_index + 1] < heap[child_index]) {
child_index++;
}
// If the smallest child is smaller than the parent, swap them
if (heap[child_index] < heap[parent_index]) {
uint64_t temp = heap[child_index];
heap[child_index] = heap[parent_index];
heap[parent_index] = temp;
parent_index = child_index;
child_index = 2 * parent_index + 1;
} else {
break; // Heap property satisfied
}
}
}
void
est_insert_value(uint64_t *heap, int heap_max_size, int *heap_size, uint64_t value)
{
if (*heap_size < heap_max_size) { // Heap not full, insert value
heap[*heap_size] = value;
est_sift_up(heap, *heap_size);
(*heap_size)++;
} else {
// Heap is full, replace root with new value with a certain probability
// This is a very naive approach to maintain a sample of the input
if (rand() % 2) {
heap[0] = value;
est_sift_down(heap, heap_max_size, 0);
}
} }
return ts.tv_sec + ts.tv_nsec / 1e9;
} }
int __xorshift32_state = 0; int __xorshift32_state = 0;

13
tests/common.h
View file

@ -23,20 +23,9 @@
#define XORSHIFT_SEED_VALUE ((unsigned int)time(NULL) ^ getpid()) #define XORSHIFT_SEED_VALUE ((unsigned int)time(NULL) ^ getpid())
#endif #endif
#define EST_MEDIAN_DECL(decl, size) \
uint64_t heap_##decl[size] = { 0 }; \
int heap_##decl##_max_size = size; \
int heap_##decl##_size = 0;
#define EST_MEDIAN_ADD(decl, value) est_insert_value(heap_##decl, heap_##decl##_max_size, &heap_##decl##_size, (value));
#define EST_MEDIAN_GET(decl) heap_##decl[0]
uint64_t tsc(void); uint64_t tsc(void);
double tsc_ticks_to_ns(uint64_t tsc_ticks); double tsc_ticks_to_ns(uint64_t tsc_ticks);
void est_sift_up(uint64_t *heap, int child_index); double nsts();
void est_sift_down(uint64_t *heap, int heap_size, int parent_index);
void est_insert_value(uint64_t *heap, int heap_max_size, int *heap_size, uint64_t value);
void xorshift32_seed(); void xorshift32_seed();
uint32_t xorshift32(); uint32_t xorshift32();

680
tests/tdigest.c Normal file
View file

@ -0,0 +1,680 @@
#include <stdlib.h>
#include <stdbool.h>
#include <string.h>
#include <math.h>
#include "tdigest.h"
#include <errno.h>
#include <stdint.h>
#ifndef TD_MALLOC_INCLUDE
#define TD_MALLOC_INCLUDE "td_malloc.h"
#endif
#ifndef TD_ALLOC_H
#define TD_ALLOC_H
#define __td_malloc malloc
#define __td_calloc calloc
#define __td_realloc realloc
#define __td_free free
#endif
#define __td_max(x, y) (((x) > (y)) ? (x) : (y))
#define __td_min(x, y) (((x) < (y)) ? (x) : (y))
static inline double weighted_average_sorted(double x1, double w1, double x2, double w2) {
const double x = (x1 * w1 + x2 * w2) / (w1 + w2);
return __td_max(x1, __td_min(x, x2));
}
static inline bool _tdigest_long_long_add_safe(long long a, long long b) {
if (b < 0) {
return (a >= __LONG_LONG_MAX__ - b);
} else {
return (a <= __LONG_LONG_MAX__ - b);
}
}
static inline double weighted_average(double x1, double w1, double x2, double w2) {
if (x1 <= x2) {
return weighted_average_sorted(x1, w1, x2, w2);
} else {
return weighted_average_sorted(x2, w2, x1, w1);
}
}
static inline void swap(double *arr, int i, int j) {
const double temp = arr[i];
arr[i] = arr[j];
arr[j] = temp;
}
static inline void swap_l(long long *arr, int i, int j) {
const long long temp = arr[i];
arr[i] = arr[j];
arr[j] = temp;
}
static unsigned int partition(double *means, long long *weights, unsigned int start,
unsigned int end, unsigned int pivot_idx) {
const double pivotMean = means[pivot_idx];
swap(means, pivot_idx, end);
swap_l(weights, pivot_idx, end);
int i = start - 1;
for (unsigned int j = start; j < end; j++) {
// If current element is smaller than the pivot
if (means[j] < pivotMean) {
// increment index of smaller element
i++;
swap(means, i, j);
swap_l(weights, i, j);
}
}
swap(means, i + 1, end);
swap_l(weights, i + 1, end);
return i + 1;
}
/**
* Standard quick sort except that sorting rearranges parallel arrays
*
* @param means Values to sort on
* @param weights The auxillary values to sort.
* @param start The beginning of the values to sort
* @param end The value after the last value to sort
*/
static void td_qsort(double *means, long long *weights, unsigned int start, unsigned int end) {
if (start < end) {
// two elements can be directly compared
if ((end - start) == 1) {
if (means[start] > means[end]) {
swap(means, start, end);
swap_l(weights, start, end);
}
return;
}
// generating a random number as a pivot was very expensive vs the array size
// const unsigned int pivot_idx = start + rand()%(end - start + 1);
const unsigned int pivot_idx = (end + start) / 2; // central pivot
const unsigned int new_pivot_idx = partition(means, weights, start, end, pivot_idx);
if (new_pivot_idx > start) {
td_qsort(means, weights, start, new_pivot_idx - 1);
}
td_qsort(means, weights, new_pivot_idx + 1, end);
}
}
static inline size_t cap_from_compression(double compression) {
if ((size_t)compression > ((SIZE_MAX / sizeof(double) / 6) - 10)) {
return 0;
}
return (6 * (size_t)(compression)) + 10;
}
static inline bool should_td_compress(td_histogram_t *h) {
return ((h->merged_nodes + h->unmerged_nodes) >= (h->cap - 1));
}
static inline int next_node(td_histogram_t *h) { return h->merged_nodes + h->unmerged_nodes; }
int td_compress(td_histogram_t *h);
static inline int _check_overflow(const double v) {
// double-precision overflow detected on h->unmerged_weight
if (v == INFINITY) {
return EDOM;
}
return 0;
}
static inline int _check_td_overflow(const double new_unmerged_weight,
const double new_total_weight) {
// double-precision overflow detected on h->unmerged_weight
if (new_unmerged_weight == INFINITY) {
return EDOM;
}
if (new_total_weight == INFINITY) {
return EDOM;
}
const double denom = 2 * MM_PI * new_total_weight * log(new_total_weight);
if (denom == INFINITY) {
return EDOM;
}
return 0;
}
int td_centroid_count(td_histogram_t *h) { return next_node(h); }
void td_reset(td_histogram_t *h) {
if (!h) {
return;
}
h->min = __DBL_MAX__;
h->max = -h->min;
h->merged_nodes = 0;
h->merged_weight = 0;
h->unmerged_nodes = 0;
h->unmerged_weight = 0;
h->total_compressions = 0;
}
int td_init(double compression, td_histogram_t **result) {
const size_t capacity = cap_from_compression(compression);
if (capacity < 1) {
return 1;
}
td_histogram_t *histogram;
histogram = (td_histogram_t *)__td_malloc(sizeof(td_histogram_t));
if (!histogram) {
return 1;
}
histogram->cap = capacity;
histogram->compression = (double)compression;
td_reset(histogram);
histogram->nodes_mean = (double *)__td_calloc(capacity, sizeof(double));
if (!histogram->nodes_mean) {
td_free(histogram);
return 1;
}
histogram->nodes_weight = (long long *)__td_calloc(capacity, sizeof(long long));
if (!histogram->nodes_weight) {
td_free(histogram);
return 1;
}
*result = histogram;
return 0;
}
td_histogram_t *td_new(double compression) {
td_histogram_t *mdigest = NULL;
td_init(compression, &mdigest);
return mdigest;
}
void td_free(td_histogram_t *histogram) {
if (histogram->nodes_mean) {
__td_free((void *)(histogram->nodes_mean));
}
if (histogram->nodes_weight) {
__td_free((void *)(histogram->nodes_weight));
}
__td_free((void *)(histogram));
}
int td_merge(td_histogram_t *into, td_histogram_t *from) {
if (td_compress(into) != 0)
return EDOM;
if (td_compress(from) != 0)
return EDOM;
const int pos = from->merged_nodes + from->unmerged_nodes;
for (int i = 0; i < pos; i++) {
const double mean = from->nodes_mean[i];
const long long weight = from->nodes_weight[i];
if (td_add(into, mean, weight) != 0) {
return EDOM;
}
}
return 0;
}
long long td_size(td_histogram_t *h) { return h->merged_weight + h->unmerged_weight; }
double td_cdf(td_histogram_t *h, double val) {
td_compress(h);
// no data to examine
if (h->merged_nodes == 0) {
return NAN;
}
// bellow lower bound
if (val < h->min) {
return 0;
}
// above upper bound
if (val > h->max) {
return 1;
}
if (h->merged_nodes == 1) {
// exactly one centroid, should have max==min
const double width = h->max - h->min;
if (val - h->min <= width) {
// min and max are too close together to do any viable interpolation
return 0.5;
} else {
// interpolate if somehow we have weight > 0 and max != min
return (val - h->min) / width;
}
}
const int n = h->merged_nodes;
// check for the left tail
const double left_centroid_mean = h->nodes_mean[0];
const double left_centroid_weight = (double)h->nodes_weight[0];
const double merged_weight_d = (double)h->merged_weight;
if (val < left_centroid_mean) {
// note that this is different than h->nodes_mean[0] > min
// ... this guarantees we divide by non-zero number and interpolation works
const double width = left_centroid_mean - h->min;
if (width > 0) {
// must be a sample exactly at min
if (val == h->min) {
return 0.5 / merged_weight_d;
} else {
return (1 + (val - h->min) / width * (left_centroid_weight / 2 - 1)) /
merged_weight_d;
}
} else {
// this should be redundant with the check val < h->min
return 0;
}
}
// and the right tail
const double right_centroid_mean = h->nodes_mean[n - 1];
const double right_centroid_weight = (double)h->nodes_weight[n - 1];
if (val > right_centroid_mean) {
const double width = h->max - right_centroid_mean;
if (width > 0) {
if (val == h->max) {
return 1 - 0.5 / merged_weight_d;
} else {
// there has to be a single sample exactly at max
const double dq = (1 + (h->max - val) / width * (right_centroid_weight / 2 - 1)) /
merged_weight_d;
return 1 - dq;
}
} else {
return 1;
}
}
// we know that there are at least two centroids and mean[0] < x < mean[n-1]
// that means that there are either one or more consecutive centroids all at exactly x
// or there are consecutive centroids, c0 < x < c1
double weightSoFar = 0;
for (int it = 0; it < n - 1; it++) {
// weightSoFar does not include weight[it] yet
if (h->nodes_mean[it] == val) {
// we have one or more centroids == x, treat them as one
// dw will accumulate the weight of all of the centroids at x
double dw = 0;
while (it < n && h->nodes_mean[it] == val) {
dw += (double)h->nodes_weight[it];
it++;
}
return (weightSoFar + dw / 2) / (double)h->merged_weight;
} else if (h->nodes_mean[it] <= val && val < h->nodes_mean[it + 1]) {
const double node_weight = (double)h->nodes_weight[it];
const double node_weight_next = (double)h->nodes_weight[it + 1];
const double node_mean = h->nodes_mean[it];
const double node_mean_next = h->nodes_mean[it + 1];
// landed between centroids ... check for floating point madness
if (node_mean_next - node_mean > 0) {
// note how we handle singleton centroids here
// the point is that for singleton centroids, we know that their entire
// weight is exactly at the centroid and thus shouldn't be involved in
// interpolation
double leftExcludedW = 0;
double rightExcludedW = 0;
if (node_weight == 1) {
if (node_weight_next == 1) {
// two singletons means no interpolation
// left singleton is in, right is out
return (weightSoFar + 1) / merged_weight_d;
} else {
leftExcludedW = 0.5;
}
} else if (node_weight_next == 1) {
rightExcludedW = 0.5;
}
double dw = (node_weight + node_weight_next) / 2;
// adjust endpoints for any singleton
double dwNoSingleton = dw - leftExcludedW - rightExcludedW;
double base = weightSoFar + node_weight / 2 + leftExcludedW;
return (base + dwNoSingleton * (val - node_mean) / (node_mean_next - node_mean)) /
merged_weight_d;
} else {
// this is simply caution against floating point madness
// it is conceivable that the centroids will be different
// but too near to allow safe interpolation
double dw = (node_weight + node_weight_next) / 2;
return (weightSoFar + dw) / merged_weight_d;
}
} else {
weightSoFar += (double)h->nodes_weight[it];
}
}
return 1 - 0.5 / merged_weight_d;
}
static double td_internal_iterate_centroids_to_index(const td_histogram_t *h, const double index,
const double left_centroid_weight,
const int total_centroids, double *weightSoFar,
int *node_pos) {
if (left_centroid_weight > 1 && index < left_centroid_weight / 2) {
// there is a single sample at min so we interpolate with less weight
return h->min + (index - 1) / (left_centroid_weight / 2 - 1) * (h->nodes_mean[0] - h->min);
}
// usually the last centroid will have unit weight so this test will make it moot
if (index > h->merged_weight - 1) {
return h->max;
}
// if the right-most centroid has more than one sample, we still know
// that one sample occurred at max so we can do some interpolation
const double right_centroid_weight = (double)h->nodes_weight[total_centroids - 1];
const double right_centroid_mean = h->nodes_mean[total_centroids - 1];
if (right_centroid_weight > 1 &&
(double)h->merged_weight - index <= right_centroid_weight / 2) {
return h->max - ((double)h->merged_weight - index - 1) / (right_centroid_weight / 2 - 1) *
(h->max - right_centroid_mean);
}
for (; *node_pos < total_centroids - 1; (*node_pos)++) {
const int i = *node_pos;
const double node_weight = (double)h->nodes_weight[i];
const double node_weight_next = (double)h->nodes_weight[i + 1];
const double node_mean = h->nodes_mean[i];
const double node_mean_next = h->nodes_mean[i + 1];
const double dw = (node_weight + node_weight_next) / 2;
if (*weightSoFar + dw > index) {
// centroids i and i+1 bracket our current point
// check for unit weight
double leftUnit = 0;
if (node_weight == 1) {
if (index - *weightSoFar < 0.5) {
// within the singleton's sphere
return node_mean;
} else {
leftUnit = 0.5;
}
}
double rightUnit = 0;
if (node_weight_next == 1) {
if (*weightSoFar + dw - index <= 0.5) {
// no interpolation needed near singleton
return node_mean_next;
}
rightUnit = 0.5;
}
const double z1 = index - *weightSoFar - leftUnit;
const double z2 = *weightSoFar + dw - index - rightUnit;
return weighted_average(node_mean, z2, node_mean_next, z1);
}
*weightSoFar += dw;
}
// weightSoFar = totalWeight - weight[total_centroids-1]/2 (very nearly)
// so we interpolate out to max value ever seen
const double z1 = index - h->merged_weight - right_centroid_weight / 2.0;
const double z2 = right_centroid_weight / 2 - z1;
return weighted_average(right_centroid_mean, z1, h->max, z2);
}
double td_quantile(td_histogram_t *h, double q) {
td_compress(h);
// q should be in [0,1]
if (q < 0.0 || q > 1.0 || h->merged_nodes == 0) {
return NAN;
}
// with one data point, all quantiles lead to Rome
if (h->merged_nodes == 1) {
return h->nodes_mean[0];
}
// if values were stored in a sorted array, index would be the offset we are interested in
const double index = q * (double)h->merged_weight;
// beyond the boundaries, we return min or max
// usually, the first centroid will have unit weight so this will make it moot
if (index < 1) {
return h->min;
}
// we know that there are at least two centroids now
const int n = h->merged_nodes;
// if the left centroid has more than one sample, we still know
// that one sample occurred at min so we can do some interpolation
const double left_centroid_weight = (double)h->nodes_weight[0];
// in between extremes we interpolate between centroids
double weightSoFar = left_centroid_weight / 2;
int i = 0;
return td_internal_iterate_centroids_to_index(h, index, left_centroid_weight, n, &weightSoFar,
&i);
}
int td_quantiles(td_histogram_t *h, const double *quantiles, double *values, size_t length) {
td_compress(h);
if (NULL == quantiles || NULL == values) {
return EINVAL;
}
const int n = h->merged_nodes;
if (n == 0) {
for (size_t i = 0; i < length; i++) {
values[i] = NAN;
}
return 0;
}
if (n == 1) {
for (size_t i = 0; i < length; i++) {
const double requested_quantile = quantiles[i];
// q should be in [0,1]
if (requested_quantile < 0.0 || requested_quantile > 1.0) {
values[i] = NAN;
} else {
// with one data point, all quantiles lead to Rome
values[i] = h->nodes_mean[0];
}
}
return 0;
}
// we know that there are at least two centroids now
// if the left centroid has more than one sample, we still know
// that one sample occurred at min so we can do some interpolation
const double left_centroid_weight = (double)h->nodes_weight[0];
// in between extremes we interpolate between centroids
double weightSoFar = left_centroid_weight / 2;
int node_pos = 0;
// to avoid allocations we use the values array for intermediate computation
// i.e. to store the expected cumulative count at each percentile
for (size_t qpos = 0; qpos < length; qpos++) {
const double index = quantiles[qpos] * (double)h->merged_weight;
values[qpos] = td_internal_iterate_centroids_to_index(h, index, left_centroid_weight, n,
&weightSoFar, &node_pos);
}
return 0;
}
static double td_internal_trimmed_mean(const td_histogram_t *h, const double leftmost_weight,
const double rightmost_weight) {
double count_done = 0;
double trimmed_sum = 0;
double trimmed_count = 0;
for (int i = 0; i < h->merged_nodes; i++) {
const double n_weight = (double)h->nodes_weight[i];
// Assume the whole centroid falls into the range
double count_add = n_weight;
// If we haven't reached the low threshold yet, skip appropriate part of the centroid.
count_add -= __td_min(__td_max(0, leftmost_weight - count_done), count_add);
// If we have reached the upper threshold, ignore the overflowing part of the centroid.
count_add = __td_min(__td_max(0, rightmost_weight - count_done), count_add);
// consider the whole centroid processed
count_done += n_weight;
// increment the sum / count
trimmed_sum += h->nodes_mean[i] * count_add;
trimmed_count += count_add;
// break once we cross the high threshold
if (count_done >= rightmost_weight)
break;
}
return trimmed_sum / trimmed_count;
}
double td_trimmed_mean_symmetric(td_histogram_t *h, double proportion_to_cut) {
td_compress(h);
// proportion_to_cut should be in [0,1]
if (h->merged_nodes == 0 || proportion_to_cut < 0.0 || proportion_to_cut > 1.0) {
return NAN;
}
// with one data point, all values lead to Rome
if (h->merged_nodes == 1) {
return h->nodes_mean[0];
}
/* translate the percentiles to counts */
const double leftmost_weight = floor((double)h->merged_weight * proportion_to_cut);
const double rightmost_weight = ceil((double)h->merged_weight * (1.0 - proportion_to_cut));
return td_internal_trimmed_mean(h, leftmost_weight, rightmost_weight);
}
double td_trimmed_mean(td_histogram_t *h, double leftmost_cut, double rightmost_cut) {
td_compress(h);
// leftmost_cut and rightmost_cut should be in [0,1]
if (h->merged_nodes == 0 || leftmost_cut < 0.0 || leftmost_cut > 1.0 || rightmost_cut < 0.0 ||
rightmost_cut > 1.0) {
return NAN;
}
// with one data point, all values lead to Rome
if (h->merged_nodes == 1) {
return h->nodes_mean[0];
}
/* translate the percentiles to counts */
const double leftmost_weight = floor((double)h->merged_weight * leftmost_cut);
const double rightmost_weight = ceil((double)h->merged_weight * rightmost_cut);
return td_internal_trimmed_mean(h, leftmost_weight, rightmost_weight);
}
int td_add(td_histogram_t *h, double mean, long long weight) {
if (should_td_compress(h)) {
const int overflow_res = td_compress(h);
if (overflow_res != 0)
return overflow_res;
}
const int pos = next_node(h);
if (pos >= h->cap)
return EDOM;
if (_tdigest_long_long_add_safe(h->unmerged_weight, weight) == false)
return EDOM;
const long long new_unmerged_weight = h->unmerged_weight + weight;
if (_tdigest_long_long_add_safe(new_unmerged_weight, h->merged_weight) == false)
return EDOM;
const long long new_total_weight = new_unmerged_weight + h->merged_weight;
// double-precision overflow detected
const int overflow_res =
_check_td_overflow((double)new_unmerged_weight, (double)new_total_weight);
if (overflow_res != 0)
return overflow_res;
if (mean < h->min) {
h->min = mean;
}
if (mean > h->max) {
h->max = mean;
}
h->nodes_mean[pos] = mean;
h->nodes_weight[pos] = weight;
h->unmerged_nodes++;
h->unmerged_weight = new_unmerged_weight;
return 0;
}
int td_compress(td_histogram_t *h) {
if (h->unmerged_nodes == 0) {
return 0;
}
int N = h->merged_nodes + h->unmerged_nodes;
td_qsort(h->nodes_mean, h->nodes_weight, 0, N - 1);
const double total_weight = (double)h->merged_weight + (double)h->unmerged_weight;
// double-precision overflow detected
const int overflow_res = _check_td_overflow((double)h->unmerged_weight, (double)total_weight);
if (overflow_res != 0)
return overflow_res;
if (total_weight <= 1)
return 0;
const double denom = 2 * MM_PI * total_weight * log(total_weight);
if (_check_overflow(denom) != 0)
return EDOM;
// Compute the normalizer given compression and number of points.
const double normalizer = h->compression / denom;
if (_check_overflow(normalizer) != 0)
return EDOM;
int cur = 0;
double weight_so_far = 0;
for (int i = 1; i < N; i++) {
const double proposed_weight = (double)h->nodes_weight[cur] + (double)h->nodes_weight[i];
const double z = proposed_weight * normalizer;
// quantile up to cur
const double q0 = weight_so_far / total_weight;
// quantile up to cur + i
const double q2 = (weight_so_far + proposed_weight) / total_weight;
// Convert a quantile to the k-scale
const bool should_add = (z <= (q0 * (1 - q0))) && (z <= (q2 * (1 - q2)));
// next point will fit
// so merge into existing centroid
if (should_add) {
h->nodes_weight[cur] += h->nodes_weight[i];
const double delta = h->nodes_mean[i] - h->nodes_mean[cur];
const double weighted_delta = (delta * h->nodes_weight[i]) / h->nodes_weight[cur];
h->nodes_mean[cur] += weighted_delta;
} else {
weight_so_far += h->nodes_weight[cur];
cur++;
h->nodes_weight[cur] = h->nodes_weight[i];
h->nodes_mean[cur] = h->nodes_mean[i];
}
if (cur != i) {
h->nodes_weight[i] = 0;
h->nodes_mean[i] = 0.0;
}
}
h->merged_nodes = cur + 1;
h->merged_weight = total_weight;
h->unmerged_nodes = 0;
h->unmerged_weight = 0;
h->total_compressions++;
return 0;
}
double td_min(td_histogram_t *h) { return h->min; }
double td_max(td_histogram_t *h) { return h->max; }
int td_compression(td_histogram_t *h) { return h->compression; }
const long long *td_centroids_weight(td_histogram_t *h) { return h->nodes_weight; }
const double *td_centroids_mean(td_histogram_t *h) { return h->nodes_mean; }
long long td_centroids_weight_at(td_histogram_t *h, int pos) { return h->nodes_weight[pos]; }
double td_centroids_mean_at(td_histogram_t *h, int pos) {
if (pos < 0 || pos > h->merged_nodes) {
return NAN;
}
return h->nodes_mean[pos];
}

258
tests/tdigest.h Normal file
View file

@ -0,0 +1,258 @@
#pragma once
#include <stdlib.h>
/**
* Adaptive histogram based on something like streaming k-means crossed with Q-digest.
* The implementation is a direct descendent of MergingDigest
* https://github.com/tdunning/t-digest/
*
* Copyright (c) 2021 Redis, All rights reserved.
* Copyright (c) 2018 Andrew Werner, All rights reserved.
*
* The special characteristics of this algorithm are:
*
* - smaller summaries than Q-digest
*
* - provides part per million accuracy for extreme quantiles and typically &lt;1000 ppm accuracy
* for middle quantiles
*
* - fast
*
* - simple
*
* - easy to adapt for use with map-reduce
*/
#define MM_PI 3.14159265358979323846
struct td_histogram {
// compression is a setting used to configure the size of centroids when merged.
double compression;
double min;
double max;
// cap is the total size of nodes
int cap;
// merged_nodes is the number of merged nodes at the front of nodes.
int merged_nodes;
// unmerged_nodes is the number of buffered nodes.
int unmerged_nodes;
// we run the merge in reverse every other merge to avoid left-to-right bias in merging
long long total_compressions;
long long merged_weight;
long long unmerged_weight;
double *nodes_mean;
long long *nodes_weight;
};
typedef struct td_histogram td_histogram_t;
#ifdef __cplusplus
extern "C" {
#endif
/**
* Allocate the memory, initialise the t-digest, and return the histogram as output parameter.
* @param compression The compression parameter.
* 100 is a common value for normal uses.
* 1000 is extremely large.
* The number of centroids retained will be a smallish (usually less than 10) multiple of this
* number.
* @return the histogram on success, NULL if allocation failed.
*/
td_histogram_t *td_new(double compression);
/**
* Allocate the memory and initialise the t-digest.
*
* @param compression The compression parameter.
* 100 is a common value for normal uses.
* 1000 is extremely large.
* The number of centroids retained will be a smallish (usually less than 10) multiple of this
* number.
* @param result Output parameter to capture allocated histogram.
* @return 0 on success, 1 if allocation failed.
*/
int td_init(double compression, td_histogram_t **result);
/**
* Frees the memory associated with the t-digest.
*
* @param h The histogram you want to free.
*/
void td_free(td_histogram_t *h);
/**
* Reset a histogram to zero - empty out a histogram and re-initialise it
*
* If you want to re-use an existing histogram, but reset everything back to zero, this
* is the routine to use.
*
* @param h The histogram you want to reset to empty.
*
*/
void td_reset(td_histogram_t *h);
/**
* Adds a sample to a histogram.
*
* @param val The value to add.
* @param weight The weight of this point.
* @return 0 on success, EDOM if overflow was detected as a consequence of adding the provided
* weight.
*
*/
int td_add(td_histogram_t *h, double val, long long weight);
/**
* Re-examines a t-digest to determine whether some centroids are redundant. If your data are
* perversely ordered, this may be a good idea. Even if not, this may save 20% or so in space.
*
* The cost is roughly the same as adding as many data points as there are centroids. This
* is typically &lt; 10 * compression, but could be as high as 100 * compression.
* This is a destructive operation that is not thread-safe.
*
* @param h The histogram you want to compress.
* @return 0 on success, EDOM if overflow was detected as a consequence of adding the provided
* weight. If overflow is detected the histogram is not changed.
*
*/
int td_compress(td_histogram_t *h);
/**
* Merges all of the values from 'from' to 'this' histogram.
*
* @param h "This" pointer
* @param from Histogram to copy values from.
* * @return 0 on success, EDOM if overflow was detected as a consequence of merging the the
* provided histogram. If overflow is detected the original histogram is not detected.
*/
int td_merge(td_histogram_t *h, td_histogram_t *from);
/**
* Returns the fraction of all points added which are &le; x.
*
* @param x The cutoff for the cdf.
* @return The fraction of all data which is less or equal to x.
*/
double td_cdf(td_histogram_t *h, double x);
/**
* Returns an estimate of the cutoff such that a specified fraction of the data
* added to this TDigest would be less than or equal to the cutoff.
*
* @param q The desired fraction
* @return The value x such that cdf(x) == q;
*/
double td_quantile(td_histogram_t *h, double q);
/**
* Returns an estimate of the cutoff such that a specified fraction of the data
* added to this TDigest would be less than or equal to the cutoffs.
*
* @param quantiles The ordered percentiles array to get the values for.
* @param values Destination array containing the values at the given quantiles.
* The values array should be allocated by the caller.
* @return 0 on success, ENOMEM if the provided destination array is null.
*/
int td_quantiles(td_histogram_t *h, const double *quantiles, double *values, size_t length);
/**
* Returns the trimmed mean ignoring values outside given cutoff upper and lower limits.
*
* @param leftmost_cut Fraction to cut off of the left tail of the distribution.
* @param rightmost_cut Fraction to cut off of the right tail of the distribution.
* @return The trimmed mean ignoring values outside given cutoff upper and lower limits;
*/
double td_trimmed_mean(td_histogram_t *h, double leftmost_cut, double rightmost_cut);
/**
* Returns the trimmed mean ignoring values outside given a symmetric cutoff limits.
*
* @param proportion_to_cut Fraction to cut off of the left and right tails of the distribution.
* @return The trimmed mean ignoring values outside given cutoff upper and lower limits;
*/
double td_trimmed_mean_symmetric(td_histogram_t *h, double proportion_to_cut);
/**
* Returns the current compression factor.
*
* @return The compression factor originally used to set up the TDigest.
*/
int td_compression(td_histogram_t *h);
/**
* Returns the number of points that have been added to this TDigest.
*
* @return The sum of the weights on all centroids.
*/
long long td_size(td_histogram_t *h);
/**
* Returns the number of centroids being used by this TDigest.
*
* @return The number of centroids being used.
*/
int td_centroid_count(td_histogram_t *h);
/**
* Get minimum value from the histogram. Will return __DBL_MAX__ if the histogram
* is empty.
*
* @param h "This" pointer
*/
double td_min(td_histogram_t *h);
/**
* Get maximum value from the histogram. Will return - __DBL_MAX__ if the histogram
* is empty.
*
* @param h "This" pointer
*/
double td_max(td_histogram_t *h);
/**
* Get the full centroids weight array for 'this' histogram.
*
* @param h "This" pointer
*
* @return The full centroids weight array.
*/
const long long *td_centroids_weight(td_histogram_t *h);
/**
* Get the full centroids mean array for 'this' histogram.
*
* @param h "This" pointer
*
* @return The full centroids mean array.
*/
const double *td_centroids_mean(td_histogram_t *h);
/**
* Get the centroid weight for 'this' histogram and 'pos'.
*
* @param h "This" pointer
* @param pos centroid position.
*
* @return The centroid weight.
*/
long long td_centroids_weight_at(td_histogram_t *h, int pos);
/**
* Get the centroid mean for 'this' histogram and 'pos'.
*
* @param h "This" pointer
* @param pos centroid position.
*
* @return The centroid mean.
*/
double td_centroids_mean_at(td_histogram_t *h, int pos);
#ifdef __cplusplus
}
#endif

239
tests/test.c
View file

@ -10,6 +10,7 @@
#define MUNIT_NO_FORK (1) #define MUNIT_NO_FORK (1)
#define MUNIT_ENABLE_ASSERT_ALIASES (1) #define MUNIT_ENABLE_ASSERT_ALIASES (1)
#include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <unistd.h> #include <unistd.h>
@ -32,6 +33,8 @@ struct user_data {
int foo; int foo;
}; };
/* -------------------------- Supporting Functions for Testing */
void void
populate_map(sparsemap_t *map, int size, int max_value) populate_map(sparsemap_t *map, int size, int max_value)
{ {
@ -42,7 +45,8 @@ populate_map(sparsemap_t *map, int size, int max_value)
shuffle(array, size); shuffle(array, size);
for (int i = 0; i < size; i++) { for (int i = 0; i < size; i++) {
sparsemap_set(map, array[i], true); sparsemap_set(map, array[i], true);
munit_assert_true(sparsemap_is_set(map, array[i])); bool set = sparsemap_is_set(map, array[i]);
munit_assert_true(set);
} }
} }
@ -65,6 +69,8 @@ test_api_tear_down(void *fixture)
free(map); free(map);
} }
/* -------------------------- API Tests */
static MunitResult static MunitResult
test_api_static_init(const MunitParameter params[], void *data) test_api_static_init(const MunitParameter params[], void *data)
{ {
@ -232,7 +238,7 @@ test_api_remaining_capacity(const MunitParameter params[], void *data)
} }
static void * static void *
test_api_get_range_size_setup(const MunitParameter params[], void *user_data) test_api_get_capacity_setup(const MunitParameter params[], void *user_data)
{ {
uint8_t *buf = munit_calloc(1024, sizeof(uint8_t)); uint8_t *buf = munit_calloc(1024, sizeof(uint8_t));
sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data); sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data);
@ -243,14 +249,14 @@ test_api_get_range_size_setup(const MunitParameter params[], void *user_data)
return (void *)map; return (void *)map;
} }
static void static void
test_api_get_range_size_tear_down(void *fixture) test_api_get_capacity_tear_down(void *fixture)
{ {
sparsemap_t *map = (sparsemap_t *)fixture; sparsemap_t *map = (sparsemap_t *)fixture;
free(map->m_data); free(map->m_data);
test_api_tear_down(fixture); test_api_tear_down(fixture);
} }
static MunitResult static MunitResult
test_api_get_range_size(const MunitParameter params[], void *data) test_api_get_capacity(const MunitParameter params[], void *data)
{ {
sparsemap_t *map = (sparsemap_t *)data; sparsemap_t *map = (sparsemap_t *)data;
(void)params; (void)params;
@ -259,8 +265,7 @@ test_api_get_range_size(const MunitParameter params[], void *data)
sparsemap_set(map, 42, true); sparsemap_set(map, 42, true);
assert_true(sparsemap_is_set(map, 42)); assert_true(sparsemap_is_set(map, 42));
size_t size = sparsemap_get_capacity(map); assert_true(sparsemap_get_capacity(map) == 1024);
assert_true(size == 1024);
return MUNIT_OK; return MUNIT_OK;
} }
@ -337,7 +342,7 @@ test_api_set(const MunitParameter params[], void *data)
} }
static void * static void *
test_api_get_start_offset_setup(const MunitParameter params[], void *user_data) test_api_get_starting_offset_setup(const MunitParameter params[], void *user_data)
{ {
uint8_t *buf = munit_calloc(1024, sizeof(uint8_t)); uint8_t *buf = munit_calloc(1024, sizeof(uint8_t));
sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data); sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data);
@ -348,14 +353,14 @@ test_api_get_start_offset_setup(const MunitParameter params[], void *user_data)
return (void *)map; return (void *)map;
} }
static void static void
test_api_get_start_offset_tear_down(void *fixture) test_api_get_starting_offset_tear_down(void *fixture)
{ {
sparsemap_t *map = (sparsemap_t *)fixture; sparsemap_t *map = (sparsemap_t *)fixture;
free(map->m_data); free(map->m_data);
test_api_tear_down(fixture); test_api_tear_down(fixture);
} }
static MunitResult static MunitResult
test_api_get_start_offset(const MunitParameter params[], void *data) test_api_get_starting_offset(const MunitParameter params[], void *data)
{ {
sparsemap_t *map = (sparsemap_t *)data; sparsemap_t *map = (sparsemap_t *)data;
(void)params; (void)params;
@ -364,7 +369,7 @@ test_api_get_start_offset(const MunitParameter params[], void *data)
sparsemap_set(map, 42, true); sparsemap_set(map, 42, true);
assert_true(sparsemap_is_set(map, 42)); assert_true(sparsemap_is_set(map, 42));
size_t offset = sparsemap_get_start_offset(map); size_t offset = sparsemap_get_starting_offset(map);
assert_true(offset == 0); assert_true(offset == 0);
return MUNIT_OK; return MUNIT_OK;
@ -513,10 +518,41 @@ test_api_select(const MunitParameter params[], void *data)
/* NOTE: select() is 0-based, to get the bit position of the 1st logical bit set /* NOTE: select() is 0-based, to get the bit position of the 1st logical bit set
call select(map, 0), to get the 18th, select(map, 17), etc. */ call select(map, 0), to get the 18th, select(map, 17), etc. */
assert_true(sparsemap_select(map, 0) == 1); assert_true(sparsemap_select(map, 0, true) == 1);
assert_true(sparsemap_select(map, 4) == 6); assert_true(sparsemap_select(map, 4, true) == 6);
assert_true(sparsemap_select(map, 17) == 26); assert_true(sparsemap_select(map, 17, true) == 26);
#if 0 // TODO
size_t f = sparsemap_select(map, 0, false);
for (int i = 0; i <= f; i++) {
assert_false(sparsemap_is_set(map, i % 2));
}
sparsemap_clear(map);
for (int i = 0; i < 1000; i++) {
sparsemap_set(map, i, i % 2 ? true : false);
}
f = sparsemap_select(map, 0, false);
assert_true(f == 1000);
sparsemap_clear(map);
sparsemap_set(map, 42, true);
sparsemap_set(map, 420, true);
sparsemap_set(map, 4200, true);
f = sparsemap_select(map, 0, false);
assert_true(f == 0);
f = sparsemap_select(map, -1, true);
assert_true(f == 4200);
f = sparsemap_select(map, -2, true);
assert_true(f == 420);
f = sparsemap_select(map, -3, true);
assert_true(f == 42);
#endif
return MUNIT_OK; return MUNIT_OK;
} }
@ -559,15 +595,15 @@ test_api_rank(const MunitParameter params[], void *data)
range as [start, end] of [0, 9] counts the bits set in the first 10 range as [start, end] of [0, 9] counts the bits set in the first 10
positions (starting from the LSB) in the index. */ positions (starting from the LSB) in the index. */
r1 = rank_uint64((uint64_t)-1, 0, 9); r1 = rank_uint64((uint64_t)-1, 0, 9);
r2 = sparsemap_rank(map, 0, 9); r2 = sparsemap_rank(map, 0, 9, true);
assert_true(r1 == r2); assert_true(r1 == r2);
assert_true(sparsemap_rank(map, 0, 9) == 10); assert_true(sparsemap_rank(map, 0, 9, true) == 10);
assert_true(sparsemap_rank(map, 1000, 1050) == 0); assert_true(sparsemap_rank(map, 1000, 1050, true) == 0);
for (int i = 0; i < 10; i++) { for (int i = 0; i < 10; i++) {
for (int j = i; j < 10; j++) { for (int j = i; j < 10; j++) {
r1 = rank_uint64((uint64_t)-1, i, j); r1 = rank_uint64((uint64_t)-1, i, j);
r2 = sparsemap_rank(map, i, j); r2 = sparsemap_rank(map, i, j, true);
assert_true(r1 == r2); assert_true(r1 == r2);
} }
} }
@ -603,26 +639,26 @@ test_api_span(const MunitParameter params[], void *data)
int located_at, placed_at, amt = 10000; int located_at, placed_at, amt = 10000;
placed_at = create_sequential_set_in_empty_map(map, amt, 1); placed_at = create_sequential_set_in_empty_map(map, amt, 1);
located_at = sparsemap_span(map, 0, 1); located_at = sparsemap_span(map, 0, 1, true);
assert_true(located_at == placed_at); assert_true(located_at == placed_at);
sparsemap_clear(map); sparsemap_clear(map);
placed_at = create_sequential_set_in_empty_map(map, amt, 50); placed_at = create_sequential_set_in_empty_map(map, amt, 50);
located_at = sparsemap_span(map, 0, 50); located_at = sparsemap_span(map, 0, 50, true);
assert_true(located_at == placed_at); assert_true(located_at == placed_at);
sparsemap_clear(map); sparsemap_clear(map);
placed_at = create_sequential_set_in_empty_map(map, amt, 50); placed_at = create_sequential_set_in_empty_map(map, amt, 50);
located_at = sparsemap_span(map, placed_at / 2, 50); located_at = sparsemap_span(map, placed_at / 2, 50, true);
assert_true(located_at == placed_at); assert_true(located_at == placed_at);
/* TODO /* TODO
sparsemap_clear(map); sparsemap_clear(map);
placed_at = create_sequential_set_in_empty_map(map, amt, amt - 1); placed_at = create_sequential_set_in_empty_map(map, amt, amt - 1);
located_at = sparsemap_span(map, 0, amt - 1); located_at = sparsemap_span(map, 0, amt - 1, true);
assert_true(located_at == placed_at); assert_true(located_at == placed_at);
*/ */
@ -637,11 +673,11 @@ static MunitTest api_test_suite[] = {
{ (char *)"/set_data_size", test_api_set_data_size, test_api_set_data_size_setup, test_api_set_data_size_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/set_data_size", test_api_set_data_size, test_api_set_data_size_setup, test_api_set_data_size_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/remaining_capacity", test_api_remaining_capacity, test_api_remaining_capacity_setup, test_api_remaining_capacity_tear_down, { (char *)"/remaining_capacity", test_api_remaining_capacity, test_api_remaining_capacity_setup, test_api_remaining_capacity_tear_down,
MUNIT_TEST_OPTION_NONE, NULL }, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/get_range_size", test_api_get_range_size, test_api_get_range_size_setup, test_api_get_range_size_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/get_capacity", test_api_get_capacity, test_api_get_capacity_setup, test_api_get_capacity_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/is_set", test_api_is_set, test_api_is_set_setup, test_api_is_set_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/is_set", test_api_is_set, test_api_is_set_setup, test_api_is_set_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/set", test_api_set, test_api_set_setup, test_api_set_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/set", test_api_set, test_api_set_setup, test_api_set_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/get_start_offset", test_api_get_start_offset, test_api_get_start_offset_setup, test_api_get_start_offset_tear_down, MUNIT_TEST_OPTION_NONE, { (char *)"/get_starting_offset", test_api_get_starting_offset, test_api_get_starting_offset_setup, test_api_get_starting_offset_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
NULL }, //TODO { (char *)"/get_ending_offset", test_api_get_ending_offset, test_api_get_ending_offset_setup, test_api_get_ending_offset_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/get_size", test_api_get_size, test_api_get_size_setup, test_api_get_size_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/get_size", test_api_get_size, test_api_get_size_setup, test_api_get_size_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/scan", test_api_scan, test_api_scan_setup, test_api_scan_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/scan", test_api_scan, test_api_scan_setup, test_api_scan_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/split", test_api_split, test_api_split_setup, test_api_split_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/split", test_api_split, test_api_split_setup, test_api_split_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
@ -652,6 +688,98 @@ static MunitTest api_test_suite[] = {
}; };
// clang-format on // clang-format on
/* -------------------------- Scale Tests */
static void *
test_scale_best_case_setup(const MunitParameter params[], void *user_data)
{
uint8_t *buf = munit_calloc(1024, sizeof(uint8_t));
sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data);
sparsemap_init(map, buf, 1024);
return (void *)map;
}
static void
test_scale_best_case_tear_down(void *fixture)
{
sparsemap_t *map = (sparsemap_t *)fixture;
free(map->m_data);
test_api_tear_down(fixture);
}
static MunitResult
test_scale_best_case(const MunitParameter params[], void *data)
{
sparsemap_t *map = (sparsemap_t *)data;
(void)params;
assert_ptr_not_null(map);
/* Best case a map can contain 2048 bits in 8 bytes.
So, in a 1KiB buffer you have:
(1024 KiB / 8 bytes) * 2048 = 268,435,456 bits
or 1.09 TiB of 4KiB pages. Let's investigate, and find out if that's the case.
TODO: Actually, 172032 are stored before SEGV, or 706 MiB of 4KiB pages.
*/
/* Set every bit on, that should be the best case. */
for (int i = 0; i < 268435456; i++) {
/* ANSI esc code to clear line, carrage return, then print on the same line */
// printf("\033[2K\r%d", i);
// fflush(stdout);
sparsemap_set(map, i, true);
}
return MUNIT_OK;
}
static void *
test_scale_worst_case_setup(const MunitParameter params[], void *user_data)
{
uint8_t *buf = munit_calloc(1024, sizeof(uint8_t));
sparsemap_t *map = (sparsemap_t *)test_api_setup(params, user_data);
sparsemap_init(map, buf, 1024);
return (void *)map;
}
static void
test_scale_worst_case_tear_down(void *fixture)
{
sparsemap_t *map = (sparsemap_t *)fixture;
free(map->m_data);
test_api_tear_down(fixture);
}
static MunitResult
test_scale_worst_case(const MunitParameter params[], void *data)
{
sparsemap_t *map = (sparsemap_t *)data;
(void)params;
assert_ptr_not_null(map);
/* Worst case a map can contain 2048 bits in 265 + 8 = 264 bytes.
So, in a 1KiB buffer you have:
(1024 KiB / 264 bytes) * 2048 = 8,134,407.75758 bits
or 33.3 GiB of 4KiB pages. Let's investigate, and find out if that's the case.
TODO: actually 7744 are stored before SEGV, or 31MiB of 4KiB pages.
*/
/* Set every other bit, that has to be the "worst case" for this index. */
for (int i = 0; i < 8134407; i += 2) {
/* ANSI esc code to clear line, carrage return, then print on the same line */
// printf("\033[2K\r%d", i);
// fflush(stdout);
sparsemap_set(map, i, true);
}
return MUNIT_OK;
}
/* -------------------------- Performance Tests */
static void * static void *
test_perf_span_solo_setup(const MunitParameter params[], void *user_data) test_perf_span_solo_setup(const MunitParameter params[], void *user_data)
{ {
@ -669,37 +797,33 @@ test_perf_span_solo_tear_down(void *fixture)
free(map->m_data); free(map->m_data);
test_api_tear_down(fixture); test_api_tear_down(fixture);
} }
EST_MEDIAN_DECL(solo, 10000)
static MunitResult static MunitResult
test_perf_span_solo(const MunitParameter params[], void *data) test_perf_span_solo(const MunitParameter params[], void *data)
{ {
sparsemap_t *map = (sparsemap_t *)data; sparsemap_t *map = (sparsemap_t *)data;
uint64_t stop, start; // double stop, start;
(void)params; (void)params;
int located_at, placed_at, amt = 1000; int located_at, placed_at, amt = 500;
assert_ptr_not_null(map); assert_ptr_not_null(map);
for (int i = 1; i < amt; i++) { for (int i = 1; i < amt; i++) {
for (int j = 1; j < amt / 10; j++) { for (int j = 1; j <= 100; j++) {
sparsemap_clear(map); sparsemap_clear(map);
placed_at = create_sequential_set_in_empty_map(map, amt, j); placed_at = create_sequential_set_in_empty_map(map, amt, j);
// logf("i = %d, j = %d\tplaced_at %d\n", i, j, placed_at); // logf("i = %d, j = %d\tplaced_at %d\n", i, j, placed_at);
// whats_set(map, 5000); // whats_set(map, 5000);
start = tsc(); // start = nsts();
located_at = sparsemap_span(map, 0, j); located_at = sparsemap_span(map, 0, j, true);
stop = tsc(); // stop = nsts();
// fprintf(stdout, "%ll - %ll = %ll\n", stop, start, stop - start); // double amt = (stop - start) * 1e6;
EST_MEDIAN_ADD(solo, stop - start); // if (amt > 0) {
// fprintf(stdout, "%0.8f\n", amt);
// }
if (placed_at != located_at) if (placed_at != located_at)
logf("a: i = %d, j = %d\tplaced_at %d located_at %d\n", i, j, placed_at, located_at); logf("a: i = %d, j = %d\tplaced_at %d located_at %d\n", i, j, placed_at, located_at);
} }
} }
uint64_t est = EST_MEDIAN_GET(solo);
// fprintf(stdout, "median time %zu or %f ns\n", est, tsc_ticks_to_ns(est)); // measured 228
assert_true(est < 500);
fflush(stdout);
return MUNIT_OK; return MUNIT_OK;
} }
@ -720,45 +844,45 @@ test_perf_span_tainted_tear_down(void *fixture)
free(map->m_data); free(map->m_data);
test_api_tear_down(fixture); test_api_tear_down(fixture);
} }
EST_MEDIAN_DECL(tainted, 10000)
static MunitResult static MunitResult
test_perf_span_tainted(const MunitParameter params[], void *data) test_perf_span_tainted(const MunitParameter params[], void *data)
{ {
sparsemap_t *map = (sparsemap_t *)data; sparsemap_t *map = (sparsemap_t *)data;
uint64_t stop, start; // double stop, start;
(void)params; (void)params;
assert_ptr_not_null(map); assert_ptr_not_null(map);
int located_at, placed_at, amt = 1000; int located_at, placed_at, amt = 500;
for (int i = 1; i < amt; i++) { for (int i = 1; i < amt; i++) {
for (int j = 1; j < amt / 10; j++) { for (int j = 100; j <= 10; j++) {
sparsemap_clear(map); sparsemap_clear(map);
populate_map(map, 1024, 1 * 1024); populate_map(map, 1024, 1 * 1024);
placed_at = create_sequential_set_in_empty_map(map, amt, j); placed_at = create_sequential_set_in_empty_map(map, amt, j);
start = tsc(); // start = nsts();
located_at = sparsemap_span(map, 0, j); located_at = sparsemap_span(map, 0, j, true);
stop = tsc(); // stop = nsts();
EST_MEDIAN_ADD(tainted, stop - start); // double amt = (stop - start) * 1e6;
// if (amt > 0) {
// fprintf(stdout, "%0.8f\n", amt);
// }
if (located_at >= placed_at) if (located_at >= placed_at)
logf("b: i = %d, j = %d\tplaced_at %d located_at %d\n", i, j, placed_at, located_at); logf("b: i = %d, j = %d\tplaced_at %d located_at %d\n", i, j, placed_at, located_at);
// assert_true(located_at >= placed_at);
// start = tsc();
// located_at = sparsemap_span(map, (placed_at < j ? 0 : placed_at / 2), i);
// stop = tsc();
// EST_MEDIAN_ADD(solo, stop - start);
// assert_true(placed_at == located_at);
} }
} }
uint64_t est = EST_MEDIAN_GET(tainted);
// fprintf(stdout, "median time %zu or %f ns\n", est, tsc_ticks_to_ns(est)); // measured 228
assert_true(est < 500);
return MUNIT_OK; return MUNIT_OK;
} }
// clang-format off // clang-format off
static MunitTest performance_test_suite[] = { static MunitTest scale_test_suite[] = {
{ (char *)"/best-case", test_scale_best_case, test_scale_best_case_setup, test_scale_best_case_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/worst-case", test_scale_worst_case, test_scale_worst_case_setup, test_scale_worst_case_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ NULL, NULL, NULL, NULL, MUNIT_TEST_OPTION_NONE, NULL } };
// clang-format on
// clang-format off
static MunitTest perf_test_suite[] = {
{ (char *)"/span/solo", test_perf_span_solo, test_perf_span_solo_setup, test_perf_span_solo_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/span/solo", test_perf_span_solo, test_perf_span_solo_setup, test_perf_span_solo_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ (char *)"/span/tainted", test_perf_span_tainted, test_perf_span_tainted_setup, test_perf_span_tainted_tear_down, MUNIT_TEST_OPTION_NONE, NULL }, { (char *)"/span/tainted", test_perf_span_tainted, test_perf_span_tainted_setup, test_perf_span_tainted_tear_down, MUNIT_TEST_OPTION_NONE, NULL },
{ NULL, NULL, NULL, NULL, MUNIT_TEST_OPTION_NONE, NULL } }; { NULL, NULL, NULL, NULL, MUNIT_TEST_OPTION_NONE, NULL } };
@ -766,7 +890,8 @@ static MunitTest performance_test_suite[] = {
// clang-format off // clang-format off
static MunitSuite other_test_suite[] = { static MunitSuite other_test_suite[] = {
{ "/performance", performance_test_suite, NULL, 1, MUNIT_SUITE_OPTION_NONE }, { "/perf", perf_test_suite, NULL, 1, MUNIT_SUITE_OPTION_NONE },
{ "/scale", scale_test_suite, NULL, 1, MUNIT_SUITE_OPTION_NONE },
{ NULL, NULL, NULL, 0, MUNIT_SUITE_OPTION_NONE } }; { NULL, NULL, NULL, 0, MUNIT_SUITE_OPTION_NONE } };
// clang-format on // clang-format on