Mercurial > public > mercurial-scm > hg
diff contrib/python-zstandard/zstd/dictBuilder/zdict.h @ 42937:69de49c4e39c
zstandard: vendor python-zstandard 0.12
The upstream source distribution from PyPI was extracted. Unwanted
files were removed.
The clang-format ignore list was updated to reflect the new source
of files.
test-repo-compengines.t was updated to reflect a change in behavior
of the zstd library.
The project contains a vendored copy of zstandard 1.4.3. The old
version was 1.3.8. This should result in some minor performance wins.
# no-check-commit because 3rd party code has different style guidelines
Differential Revision: https://phab.mercurial-scm.org/D6858
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Sun, 15 Sep 2019 20:04:00 -0700 |
parents | 73fef626dae3 |
children |
line wrap: on
line diff
--- a/contrib/python-zstandard/zstd/dictBuilder/zdict.h Sun Sep 15 00:07:30 2019 -0400 +++ b/contrib/python-zstandard/zstd/dictBuilder/zdict.h Sun Sep 15 20:04:00 2019 -0700 @@ -46,7 +46,12 @@ * The resulting dictionary will be saved into `dictBuffer`. * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) * or an error code, which can be tested with ZDICT_isError(). - * Note: ZDICT_trainFromBuffer() requires about 9 bytes of memory for each input byte. + * Note: Dictionary training will fail if there are not enough samples to construct a + * dictionary, or if most of the samples are too small (< 8 bytes being the lower limit). + * If dictionary training fails, you should use zstd without a dictionary, as the dictionary + * would've been ineffective anyways. If you believe your samples would benefit from a dictionary + * please open an issue with details, and we can look into it. + * Note: ZDICT_trainFromBuffer()'s memory usage is about 6 MB. * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. * In general, it's recommended to provide a few thousands samples, though this can vary a lot. @@ -89,6 +94,8 @@ unsigned steps; /* Number of steps : Only used for optimization : 0 means default (40) : Higher means more parameters checked */ unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */ double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */ + unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */ + unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */ ZDICT_params_t zParams; } ZDICT_cover_params_t; @@ -100,6 +107,9 @@ unsigned nbThreads; /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */ double splitPoint; /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */ unsigned accel; /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */ + unsigned shrinkDict; /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking */ + unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */ + ZDICT_params_t zParams; } ZDICT_fastCover_params_t; @@ -110,6 +120,7 @@ * The resulting dictionary will be saved into `dictBuffer`. * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) * or an error code, which can be tested with ZDICT_isError(). + * See ZDICT_trainFromBuffer() for details on failure modes. * Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte. * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. @@ -133,8 +144,9 @@ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [50, 2000]. * * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) - * or an error code, which can be tested with ZDICT_isError(). - * On success `*parameters` contains the parameters selected. + * or an error code, which can be tested with ZDICT_isError(). + * On success `*parameters` contains the parameters selected. + * See ZDICT_trainFromBuffer() for details on failure modes. * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread. */ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover( @@ -151,7 +163,8 @@ * The resulting dictionary will be saved into `dictBuffer`. * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) * or an error code, which can be tested with ZDICT_isError(). - * Note: ZDICT_trainFromBuffer_fastCover() requires about 1 bytes of memory for each input byte and additionally another 6 * 2^f bytes of memory . + * See ZDICT_trainFromBuffer() for details on failure modes. + * Note: ZDICT_trainFromBuffer_fastCover() requires 6 * 2^f bytes of memory. * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. * In general, it's recommended to provide a few thousands samples, though this can vary a lot. @@ -175,9 +188,10 @@ * If accel is zero, default value of 1 is used. * * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) - * or an error code, which can be tested with ZDICT_isError(). - * On success `*parameters` contains the parameters selected. - * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 1 byte of memory for each input byte and additionally another 6 * 2^f bytes of memory for each thread. + * or an error code, which can be tested with ZDICT_isError(). + * On success `*parameters` contains the parameters selected. + * See ZDICT_trainFromBuffer() for details on failure modes. + * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread. */ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity, const void* samplesBuffer, @@ -195,7 +209,7 @@ * maxDictSize must be >= dictContentSize, and must be >= ZDICT_DICTSIZE_MIN bytes. * * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`), - * or an error code, which can be tested by ZDICT_isError(). + * or an error code, which can be tested by ZDICT_isError(). * Note: ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0. * Note 2: dictBuffer and dictContent can overlap */ @@ -219,6 +233,7 @@ * `parameters` is optional and can be provided with values set to 0 to mean "default". * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`) * or an error code, which can be tested with ZDICT_isError(). + * See ZDICT_trainFromBuffer() for details on failure modes. * Tips: In general, a reasonable dictionary has a size of ~ 100 KB. * It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`. * In general, it's recommended to provide a few thousands samples, though this can vary a lot.