MyFavMirrors/minio
1
0
Fork 0
mirror of https://github.com/minio/minio.git synced 2025-12-01 22:02:33 -05:00
Code Issues Packages Projects Releases Wiki Activity

Add erasure coding and decoding using Intel Storage Acceleration library

Browse Source 
- move contrib/erasure --> contrib/isal
 - bring in low level 'isal' package for Go for exposing C functions
 - Implement Erasure 'encoding'
   Supports - Reed Solomon Codes, Cauchy Codes
 - Implement Erasure 'decoding'
   Supports - Reed Solomon Codes, Cauchy Codes
 - Renames Minios -> Minio at all the references
This commit is contained in:
Harshavardhana
2014-11-05 03:09:40 -08:00
parent cbce7fd66f
commit 1e7515a7df
85 changed files with 839 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6680
erasure/isal/include/ec-base.h Normal file
View File

File diff suppressed because it is too large Load Diff

659
erasure/isal/include/erasure-code.h Normal file
View File

@@ -0,0 +1,659 @@
/**********************************************************************
Copyright(c) 2011-2014 Intel Corporation All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**********************************************************************/
#ifndef _ERASURE_CODE_H_
#define _ERASURE_CODE_H_
/**
* @file erasure_code.h
* @brief Interface to functions supporting erasure code encode and decode.
*
* This file defines the interface to optimized functions used in erasure
* codes. Encode and decode of erasures in GF(2^8) are made by calculating the
* dot product of the symbols (bytes in GF(2^8)) across a set of buffers and a
* set of coefficients. Values for the coefficients are determined by the type
* of erasure code. Using a general dot product means that any sequence of
* coefficients may be used including erasure codes based on random
* coefficients.
* Multiple versions of dot product are supplied to calculate 1-6 output
* vectors in one pass.
* Base GF multiply and divide functions can be sped up by defining
* GF_LARGE_TABLES at the expense of memory size.
*
*/
#include "gf-vect-mul.h"
#ifdef __cplusplus
extern "C" {
#endif
/**
* @brief Initialize tables for fast Erasure Code encode and decode.
*
* Generates the expanded tables needed for fast encode or decode for erasure
* codes on blocks of data. 32bytes is generated for each input coefficient.
*
* @param k The number of vector sources or rows in the generator matrix
* for coding.
* @param rows The number of output vectors to concurrently encode/decode.
* @param a Pointer to sets of arrays of input coefficients used to encode
* or decode data.
* @param gftbls Pointer to start of space for concatenated output tables
* generated from input coefficients. Must be of size 32*k*rows.
* @returns none
*/
void ec_init_tables(int k, int rows, unsigned char* a, unsigned char* gftbls);
/**
* @brief Generate or decode erasure codes on blocks of data.
*
* Given a list of source data blocks, generate one or multiple blocks of
* encoded data as specified by a matrix of GF(2^8) coefficients. When given a
* suitable set of coefficients, this function will perform the fast generation
* or decoding of Reed-Solomon type erasure codes.
*
* @requires SSE4.1
* @param len Length of each block of data (vector) of source or dest data.
* @param k The number of vector sources or rows in the generator matrix
* for coding.
* @param rows The number of output vectors to concurrently encode/decode.
* @param gftbls Pointer to array of input tables generated from coding
* coefficients in ec_init_tables(). Must be of size 32*k*rows
* @param data Array of pointers to source input buffers.
* @param coding Array of pointers to coded output buffers.
* @returns none
*/
void ec_encode_data_sse(int len, int k, int rows, unsigned char *gftbls, unsigned char **data, unsigned char **coding);
/**
* @brief Generate or decode erasure codes on blocks of data, runs appropriate version.
*
* Given a list of source data blocks, generate one or multiple blocks of
* encoded data as specified by a matrix of GF(2^8) coefficients. When given a
* suitable set of coefficients, this function will perform the fast generation
* or decoding of Reed-Solomon type erasure codes.
*
* This function determines what instruction sets are enabled and
* selects the appropriate version at runtime.
*
* @param len Length of each block of data (vector) of source or dest data.
* @param k The number of vector sources or rows in the generator matrix
* for coding.
* @param rows The number of output vectors to concurrently encode/decode.
* @param gftbls Pointer to array of input tables generated from coding
* coefficients in ec_init_tables(). Must be of size 32*k*rows
* @param data Array of pointers to source input buffers.
* @param coding Array of pointers to coded output buffers.
* @returns none
*/
void ec_encode_data(int len, int k, int rows, unsigned char *gftbls, unsigned char **data, unsigned char **coding);
/**
* @brief Generate or decode erasure codes on blocks of data, runs baseline version.
*
* Given a list of source data blocks, generate one or multiple blocks of
* encoded data as specified by a matrix of GF(2^8) coefficients. When given a
* suitable set of coefficients, this function will perform the fast generation
* or decoding of Reed-Solomon type erasure codes.
*
* @param len Length of each block of data (vector) of source or dest data.
* @param srcs The number of vector sources or rows in the generator matrix
* for coding.
* @param dests The number of output vectors to concurrently encode/decode.
* @param v Pointer to array of input tables generated from coding
* coefficients in ec_init_tables(). Must be of size 32*k*rows
* @param src Array of pointers to source input buffers.
* @param dest Array of pointers to coded output buffers.
* @returns none
*/
void ec_encode_data_base(int len, int srcs, int dests, unsigned char *v, unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product.
*
* Does a GF(2^8) dot product across each byte of the input array and a constant
* set of coefficients to produce each byte of the output. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 32*vlen byte constant array based on the input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 32*vlen byte array of pre-calculated constants based
* on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Pointer to destination data array.
* @returns none
*/
void gf_vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char *dest);
/**
* @brief GF(2^8) vector dot product.
*
* Does a GF(2^8) dot product across each byte of the input array and a constant
* set of coefficients to produce each byte of the output. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 32*vlen byte constant array based on the input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 32*vlen byte array of pre-calculated constants based
* on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Pointer to destination data array.
* @returns none
*/
void gf_vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char *dest);
/**
* @brief GF(2^8) vector dot product.
*
* Does a GF(2^8) dot product across each byte of the input array and a constant
* set of coefficients to produce each byte of the output. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 32*vlen byte constant array based on the input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 32*vlen byte array of pre-calculated constants based
* on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Pointer to destination data array.
* @returns none
*/
void gf_vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char *dest);
/**
* @brief GF(2^8) vector dot product with two outputs.
*
* Vector dot product optimized to calculate two ouputs at a time. Does two
* GF(2^8) dot products across each byte of the input array and two constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 2*32*vlen byte constant array based on the two sets of input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 2*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_2vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with two outputs.
*
* Vector dot product optimized to calculate two ouputs at a time. Does two
* GF(2^8) dot products across each byte of the input array and two constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 2*32*vlen byte constant array based on the two sets of input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 2*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_2vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with two outputs.
*
* Vector dot product optimized to calculate two ouputs at a time. Does two
* GF(2^8) dot products across each byte of the input array and two constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 2*32*vlen byte constant array based on the two sets of input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 2*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_2vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with three outputs.
*
* Vector dot product optimized to calculate three ouputs at a time. Does three
* GF(2^8) dot products across each byte of the input array and three constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 3*32*vlen byte constant array based on the three sets of input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 3*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_3vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with three outputs.
*
* Vector dot product optimized to calculate three ouputs at a time. Does three
* GF(2^8) dot products across each byte of the input array and three constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 3*32*vlen byte constant array based on the three sets of input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 3*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_3vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with three outputs.
*
* Vector dot product optimized to calculate three ouputs at a time. Does three
* GF(2^8) dot products across each byte of the input array and three constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 3*32*vlen byte constant array based on the three sets of input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 3*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_3vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with four outputs.
*
* Vector dot product optimized to calculate four ouputs at a time. Does four
* GF(2^8) dot products across each byte of the input array and four constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 4*32*vlen byte constant array based on the four sets of input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 4*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_4vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with four outputs.
*
* Vector dot product optimized to calculate four ouputs at a time. Does four
* GF(2^8) dot products across each byte of the input array and four constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 4*32*vlen byte constant array based on the four sets of input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 4*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_4vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with four outputs.
*
* Vector dot product optimized to calculate four ouputs at a time. Does four
* GF(2^8) dot products across each byte of the input array and four constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 4*32*vlen byte constant array based on the four sets of input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 4*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_4vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with five outputs.
*
* Vector dot product optimized to calculate five ouputs at a time. Does five
* GF(2^8) dot products across each byte of the input array and five constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 5*32*vlen byte constant array based on the five sets of input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 5*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_5vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with five outputs.
*
* Vector dot product optimized to calculate five ouputs at a time. Does five
* GF(2^8) dot products across each byte of the input array and five constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 5*32*vlen byte constant array based on the five sets of input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 5*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_5vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with five outputs.
*
* Vector dot product optimized to calculate five ouputs at a time. Does five
* GF(2^8) dot products across each byte of the input array and five constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 5*32*vlen byte constant array based on the five sets of input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 5*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_5vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with six outputs.
*
* Vector dot product optimized to calculate six ouputs at a time. Does six
* GF(2^8) dot products across each byte of the input array and six constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 6*32*vlen byte constant array based on the six sets of input coefficients.
*
* @requires SSE4.1
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 6*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_6vect_dot_prod_sse(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with six outputs.
*
* Vector dot product optimized to calculate six ouputs at a time. Does six
* GF(2^8) dot products across each byte of the input array and six constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 6*32*vlen byte constant array based on the six sets of input coefficients.
*
* @requires AVX
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 6*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_6vect_dot_prod_avx(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product with six outputs.
*
* Vector dot product optimized to calculate six ouputs at a time. Does six
* GF(2^8) dot products across each byte of the input array and six constant
* sets of coefficients to produce each byte of the outputs. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 6*32*vlen byte constant array based on the six sets of input coefficients.
*
* @requires AVX2
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 6*32*vlen byte array of pre-calculated constants
* based on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Array of pointers to destination data buffers.
* @returns none
*/
void gf_6vect_dot_prod_avx2(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char **dest);
/**
* @brief GF(2^8) vector dot product, runs baseline version.
*
* Does a GF(2^8) dot product across each byte of the input array and a constant
* set of coefficients to produce each byte of the output. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 32*vlen byte constant array based on the input coefficients.
*
* @param len Length of each vector in bytes. Must be >= 16.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 32*vlen byte array of pre-calculated constants based
* on the array of input coefficients. Only elements 32*CONST*j + 1
* of this array are used, where j = (0, 1, 2...) and CONST is the
* number of elements in the array of input coefficients. The
* elements used correspond to the original input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Pointer to destination data array.
* @returns none
*/
void gf_vect_dot_prod_base(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char *dest);
/**
* @brief GF(2^8) vector dot product, runs appropriate version.
*
* Does a GF(2^8) dot product across each byte of the input array and a constant
* set of coefficients to produce each byte of the output. Can be used for
* erasure coding encode and decode. Function requires pre-calculation of a
* 32*vlen byte constant array based on the input coefficients.
*
* This function determines what instruction sets are enabled and
* selects the appropriate version at runtime.
*
* @param len Length of each vector in bytes. Must be >= 32.
* @param vlen Number of vector sources.
* @param gftbls Pointer to 32*vlen byte array of pre-calculated constants based
* on the array of input coefficients.
* @param src Array of pointers to source inputs.
* @param dest Pointer to destination data array.
* @returns none
*/
void gf_vect_dot_prod(int len, int vlen, unsigned char *gftbls,
unsigned char **src, unsigned char *dest);
/**********************************************************************
* The remaining are lib support functions used in GF(2^8) operations.
*/
/**
* @brief Single element GF(2^8) multiply.
*
* @param a Multiplicand a
* @param b Multiplicand b
* @returns Product of a and b in GF(2^8)
*/
unsigned char gf_mul(unsigned char a, unsigned char b);
/**
* @brief Single element GF(2^8) inverse.
*
* @param a Input element
* @returns Field element b such that a x b = {1}
*/
unsigned char gf_inv(unsigned char a);
/**
* @brief Generate a matrix of coefficients to be used for encoding.
*
* Vandermonde matrix example of encoding coefficients where high portion of
* matrix is identity matrix I and lower portion is constructed as 2^{i*(j-k+1)}
* i:{0,k-1} j:{k,m-1}. Commonly used method for choosing coefficients in
* erasure encoding but does not guarantee invertable for every sub matrix. For
* large k it is possible to find cases where the decode matrix chosen from
* sources and parity not in erasure are not invertable. Users may want to
* adjust for k > 5.
*
* @param a [mxk] array to hold coefficients
* @param m number of rows in matrix corresponding to srcs + parity.
* @param k number of columns in matrix corresponding to srcs.
* @returns none
*/
void gf_gen_rs_matrix(unsigned char *a, int m, int k);
/**
* @brief Generate a Cauchy matrix of coefficients to be used for encoding.
*
* Cauchy matrix example of encoding coefficients where high portion of matrix
* is identity matrix I and lower portion is constructed as 1/(i + j) | i != j,
* i:{0,k-1} j:{k,m-1}. Any sub-matrix of a Cauchy matrix should be invertable.
*
* @param a [mxk] array to hold coefficients
* @param m number of rows in matrix corresponding to srcs + parity.
* @param k number of columns in matrix corresponding to srcs.
* @returns none
*/
void gf_gen_cauchy1_matrix(unsigned char *a, int m, int k);
/**
* @brief Invert a matrix in GF(2^8)
*
* @param in input matrix
* @param out output matrix such that [in] x [out] = [I] - identity matrix
* @param n size of matrix [nxn]
* @returns 0 successful, other fail on singular input matrix
*/
int gf_invert_matrix(unsigned char *in, unsigned char *out, const int n);
/*************************************************************/
#ifdef __cplusplus
}
#endif
#endif //_ERASURE_CODE_H_

81
erasure/isal/include/erasure/tests.h Normal file
View File

@@ -0,0 +1,81 @@
/**********************************************************************
Copyright(c) 2011-2014 Intel Corporation All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**********************************************************************/
#ifndef __ERASURE_TESTS_H
#define __ERASURE_TESTS_H
#ifdef __cplusplus
extern "C" {
#endif
// Use sys/time.h functions for time
#include <sys/time.h>
struct perf{
struct timeval tv;
};
inline int perf_start(struct perf *p)
{
return gettimeofday(&(p->tv), 0);
}
inline int perf_stop(struct perf *p)
{
return gettimeofday(&(p->tv), 0);
}
inline void perf_print(struct perf stop, struct perf start, long long dsize)
{
long long secs = stop.tv.tv_sec - start.tv.tv_sec;
long long usecs = secs * 1000000 + stop.tv.tv_usec - start.tv.tv_usec;
printf("runtime = %10lld usecs", usecs);
if (dsize != 0) {
#if 1 // not bug in printf for 32-bit
printf(", bandwidth %lld MB in %.4f sec = %.2f MB/s\n", dsize/(1024*1024),
((double) usecs)/1000000, ((double) dsize) / (double)usecs);
#else
printf(", bandwidth %lld MB ", dsize/(1024*1024));
printf("in %.4f sec ",(double)usecs/1000000);
printf("= %.2f MB/s\n", (double)dsize/usecs);
#endif
}
else
printf("\n");
}
#ifdef __cplusplus
}
#endif
#endif // __ERASURE_TESTS_H

80
erasure/isal/include/erasure/types.h Normal file
View File

@@ -0,0 +1,80 @@
/**********************************************************************
Copyright(c) 2011-2014 Intel Corporation All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**********************************************************************/
/**
* @file types.h
* @brief Defines standard width types.
*
*/
#ifndef __ERASURE_TYPES_H
#define __ERASURE_TYPES_H
#ifdef __cplusplus
extern "C" {
#endif
#ifndef __unix__
#ifdef __MINGW32__
# include <_mingw.h>
#endif
typedef unsigned __int64 UINT64;
typedef __int64 INT64;
typedef unsigned __int32 UINT32;
typedef unsigned __int16 UINT16;
typedef unsigned char UINT8;
#else
typedef unsigned long int UINT64;
typedef long int INT64;
typedef unsigned int UINT32;
typedef unsigned short int UINT16;
typedef unsigned char UINT8;
#endif
#ifdef __unix__
# define DECLARE_ALIGNED(decl, alignval) decl __attribute__((aligned(alignval)))
# define __forceinline static inline
#else
# define DECLARE_ALIGNED(decl, alignval) __declspec(align(alignval)) decl
# define posix_memalign(p, algn, len) (NULL == (*((char**)(p)) = (void*) _aligned_malloc(len, algn)))
#endif
#ifdef DEBUG
# define DEBUG_PRINT(x) printf x
#else
# define DEBUG_PRINT(x) do {} while (0)
#endif
#ifdef __cplusplus
}
#endif
#endif //__ERASURE_TYPES_H

148
erasure/isal/include/gf-vect-mul.h Normal file
View File

@@ -0,0 +1,148 @@
/**********************************************************************
Copyright(c) 2011-2014 Intel Corporation All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**********************************************************************/
#ifndef _GF_VECT_MUL_H
#define _GF_VECT_MUL_H
/**
* @file gf-vect-mul.h
* @brief Interface to functions for vector (block) multiplication in GF(2^8).
*
* This file defines the interface to routines used in fast RAID rebuild and
* erasure codes.
*/
#ifdef __cplusplus
extern "C" {
#endif
/**
* @brief GF(2^8) vector multiply by constant.
*
* Does a GF(2^8) vector multiply b = Ca where a and b are arrays and C
* is a single field element in GF(2^8). Can be used for RAID6 rebuild
* and partial write functions. Function requires pre-calculation of a
* 32-element constant array based on constant C. gftbl(C) = {C{00},
* C{01}, C{02}, ... , C{0f} }, {C{00}, C{10}, C{20}, ... , C{f0} }. Len
* and src must be aligned to 32B.
* @requires SSE4.1
* @param len Length of vector in bytes. Must be aligned to 32B.
* @param gftbl Pointer to 32-byte array of pre-calculated constants based on C.
* @param src Pointer to src data array. Must be aligned to 32B.
* @param dest Pointer to destination data array. Must be aligned to 32B.
* @returns 0 pass, other fail
*/
int gf_vect_mul_sse(int len, unsigned char *gftbl, void *src, void *dest);
/**
* @brief GF(2^8) vector multiply by constant.
*
* Does a GF(2^8) vector multiply b = Ca where a and b are arrays and C
* is a single field element in GF(2^8). Can be used for RAID6 rebuild
* and partial write functions. Function requires pre-calculation of a
* 32-element constant array based on constant C. gftbl(C) = {C{00},
* C{01}, C{02}, ... , C{0f} }, {C{00}, C{10}, C{20}, ... , C{f0} }. Len
* and src must be aligned to 32B.
* @requires AVX
* @param len Length of vector in bytes. Must be aligned to 32B.
* @param gftbl Pointer to 32-byte array of pre-calculated constants based on C.
* @param src Pointer to src data array. Must be aligned to 32B.
* @param dest Pointer to destination data array. Must be aligned to 32B.
* @returns 0 pass, other fail
*/
int gf_vect_mul_avx(int len, unsigned char *gftbl, void *src, void *dest);
/**
* @brief GF(2^8) vector multiply by constant, runs appropriate version.
*
* Does a GF(2^8) vector multiply b = Ca where a and b are arrays and C
* is a single field element in GF(2^8). Can be used for RAID6 rebuild
* and partial write functions. Function requires pre-calculation of a
* 32-element constant array based on constant C. gftbl(C) = {C{00},
* C{01}, C{02}, ... , C{0f} }, {C{00}, C{10}, C{20}, ... , C{f0} }.
* Len and src must be aligned to 32B.
*
* This function determines what instruction sets are enabled
* and selects the appropriate version at runtime.
*
* @param len Length of vector in bytes. Must be aligned to 32B.
* @param gftbl Pointer to 32-byte array of pre-calculated constants based on C.
* @param src Pointer to src data array. Must be aligned to 32B.
* @param dest Pointer to destination data array. Must be aligned to 32B.
* @returns 0 pass, other fail
*/
int gf_vect_mul(int len, unsigned char *gftbl, void *src, void *dest);
/**
* @brief Initialize 32-byte constant array for GF(2^8) vector multiply
*
* Calculates array {C{00}, C{01}, C{02}, ... , C{0f} }, {C{00}, C{10},
* C{20}, ... , C{f0} } as required by other fast vector multiply
* functions.
* @param c Constant input.
* @param gftbl Table output.
*/
void gf_vect_mul_init(unsigned char c, unsigned char* gftbl);
/**
* @brief GF(2^8) vector multiply by constant, runs baseline version.
*
* Does a GF(2^8) vector multiply b = Ca where a and b are arrays and C
* is a single field element in GF(2^8). Can be used for RAID6 rebuild
* and partial write functions. Function requires pre-calculation of a
* 32-element constant array based on constant C. gftbl(C) = {C{00},
* C{01}, C{02}, ... , C{0f} }, {C{00}, C{10}, C{20}, ... , C{f0} }. Len
* and src must be aligned to 32B.
*
* @param len Length of vector in bytes. Must be aligned to 32B.
* @param a Pointer to 32-byte array of pre-calculated constants based on C.
* only use 2nd element is used.
* @param src Pointer to src data array. Must be aligned to 32B.
* @param dest Pointer to destination data array. Must be aligned to 32B.
*/
void gf_vect_mul_base(int len, unsigned char *a, unsigned char *src,
unsigned char *dest);
#ifdef __cplusplus
}
#endif
#endif //_GF_VECT_MUL_H

96
erasure/isal/include/reg-sizes.asm Normal file
View File

@@ -0,0 +1,96 @@
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Copyright(c) 2011-2014 Intel Corporation All rights reserved.
;
; Redistribution and use in source and binary forms, with or without
; modification, are permitted provided that the following conditions
; are met:
; * Redistributions of source code must retain the above copyright
; notice, this list of conditions and the following disclaimer.
; * Redistributions in binary form must reproduce the above copyright
; notice, this list of conditions and the following disclaimer in
; the documentation and/or other materials provided with the
; distribution.
; * Neither the name of Intel Corporation nor the names of its
; contributors may be used to endorse or promote products derived
; from this software without specific prior written permission.
;
; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
%define EFLAGS_HAS_CPUID (1<<21)
%define FLAG_CPUID1_ECX_CLMUL (1<<1)
%define FLAG_CPUID1_EDX_SSE2 (1<<26)
%define FLAG_CPUID1_ECX_SSE3 (1)
%define FLAG_CPUID1_ECX_SSE4_1 (1<<19)
%define FLAG_CPUID1_ECX_SSE4_2 (1<<20)
%define FLAG_CPUID1_ECX_POPCNT (1<<23)
%define FLAG_CPUID1_ECX_AESNI (1<<25)
%define FLAG_CPUID1_ECX_OSXSAVE (1<<27)
%define FLAG_CPUID1_ECX_AVX (1<<28)
%define FLAG_CPUID1_EBX_AVX2 (1<<5)
%define FLAG_XGETBV_EAX_XMM_YMM 0x6
%define FLAG_CPUID1_EAX_AVOTON 0x000406d0
; define d and w variants for registers
%define raxd eax
%define raxw ax
%define raxb al
%define rbxd ebx
%define rbxw bx
%define rbxb bl
%define rcxd ecx
%define rcxw cx
%define rcxb cl
%define rdxd edx
%define rdxw dx
%define rdxb dl
%define rsid esi
%define rsiw si
%define rsib sil
%define rdid edi
%define rdiw di
%define rdib dil
%define rbpd ebp
%define rbpw bp
%define rbpb bpl
%define ymm0x xmm0
%define ymm1x xmm1
%define ymm2x xmm2
%define ymm3x xmm3
%define ymm4x xmm4
%define ymm5x xmm5
%define ymm6x xmm6
%define ymm7x xmm7
%define ymm8x xmm8
%define ymm9x xmm9
%define ymm10x xmm10
%define ymm11x xmm11
%define ymm12x xmm12
%define ymm13x xmm13
%define ymm14x xmm14
%define ymm15x xmm15
%define DWORD(reg) reg %+ d
%define WORD(reg) reg %+ w
%define BYTE(reg) reg %+ b
%define XWORD(reg) reg %+ x