Implemented permuted longest common prefix array (PLCP) construction …

…of an integer array.

CHANGES

@@ -1,4 +1,5 @@
-Changes in 2.7.6 (March 3, 2024)
+Changes in 2.8.0 (March 3, 2024)
+- Implemented permuted longest common prefix array (PLCP) construction of an integer array.
 - Fixed compilation error when compiling the library with OpenMP enabled.
 
 Changes in 2.7.5 (February 26, 2024)

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.10)
 
-project(libsais VERSION 2.7.6 LANGUAGES C DESCRIPTION "libsais is a library for linear time suffix array, longest common prefix array and burrows wheeler transform construction based on induced sorting algorithm.")
+project(libsais VERSION 2.8.0 LANGUAGES C DESCRIPTION "libsais is a library for linear time suffix array, longest common prefix array and burrows wheeler transform construction based on induced sorting algorithm.")
 
 set(CMAKE_C_STANDARD 99)
 set(CMAKE_C_STANDARD_REQUIRED ON)

README.md

@@ -23,7 +23,8 @@ The libsais provides simple C99 API to construct suffix array and Burrows-Wheele
 The libsais is released under the [Apache License Version 2.0](LICENSE "Apache license")
 
 ## Changes
-* March 3, 2024 (2.7.6)
+* March 3, 2024 (2.8.0)
+  * Implemented permuted longest common prefix array (PLCP) construction of an integer array.
   * Fixed compilation error when compiling the library with OpenMP enabled.
 * February 26, 2024 (2.7.5)
   * Improved performance of suffix array and burrows wheeler transform construction on degenerate inputs.

VERSION

@@ -1 +1 @@
-2.7.6
+2.8.0

include/libsais.h

@@ -25,9 +25,9 @@ Please see the file LICENSE for full copyright information.
 #define LIBSAIS_H 1
 
 #define LIBSAIS_VERSION_MAJOR   2
-#define LIBSAIS_VERSION_MINOR   7
-#define LIBSAIS_VERSION_PATCH   5
-#define LIBSAIS_VERSION_STRING  "2.7.5"
+#define LIBSAIS_VERSION_MINOR   8
+#define LIBSAIS_VERSION_PATCH   0
+#define LIBSAIS_VERSION_STRING  "2.8.0"
 
 #ifdef _WIN32
     #ifdef LIBSAIS_SHARED
@@ -332,6 +332,16 @@ extern "C" {
     */
     LIBSAIS_API int32_t libsais_plcp(const uint8_t * T, const int32_t * SA, int32_t * PLCP, int32_t n);
 
+    /**
+    * Constructs the permuted longest common prefix array (PLCP) of a integer array and a suffix array.
+    * @param T [0..n-1] The input integer array.
+    * @param SA [0..n-1] The input suffix array.
+    * @param PLCP [0..n-1] The output permuted longest common prefix array.
+    * @param n The length of the integer array and the suffix array.
+    * @return 0 if no error occurred, -1 otherwise.
+    */
+    LIBSAIS_API int32_t libsais_plcp_int(const int32_t * T, const int32_t * SA, int32_t * PLCP, int32_t n);
+
     /**
     * Constructs the longest common prefix array (LCP) of a given permuted longest common prefix array (PLCP) and a suffix array.
     * @param PLCP [0..n-1] The input permuted longest common prefix array.
@@ -354,6 +364,17 @@ extern "C" {
     */
     LIBSAIS_API int32_t libsais_plcp_omp(const uint8_t * T, const int32_t * SA, int32_t * PLCP, int32_t n, int32_t threads);
 
+    /**
+    * Constructs the permuted longest common prefix array (PLCP) of a given integer array and a suffix array in parallel using OpenMP.
+    * @param T [0..n-1] The input integer array.
+    * @param SA [0..n-1] The input suffix array.
+    * @param PLCP [0..n-1] The output permuted longest common prefix array.
+    * @param n The length of the integer array and the suffix array.
+    * @param threads The number of OpenMP threads to use (can be 0 for OpenMP default).
+    * @return 0 if no error occurred, -1 otherwise.
+    */
+    LIBSAIS_API int32_t libsais_plcp_int_omp(const int32_t * T, const int32_t * SA, int32_t * PLCP, int32_t n, int32_t threads);
+
     /**
     * Constructs the longest common prefix array (LCP) of a given permuted longest common prefix array (PLCP) and a suffix array in parallel using OpenMP.
     * @param PLCP [0..n-1] The input permuted longest common prefix array.

include/libsais16.h

@@ -25,9 +25,9 @@ Please see the file LICENSE for full copyright information.
 #define LIBSAIS16_H 1
 
 #define LIBSAIS16_VERSION_MAJOR   2
-#define LIBSAIS16_VERSION_MINOR   7
-#define LIBSAIS16_VERSION_PATCH   5
-#define LIBSAIS16_VERSION_STRING  "2.7.5"
+#define LIBSAIS16_VERSION_MINOR   8
+#define LIBSAIS16_VERSION_PATCH   0
+#define LIBSAIS16_VERSION_STRING  "2.8.0"
 
 #ifdef _WIN32
     #ifdef LIBSAIS_SHARED

include/libsais64.h

@@ -25,9 +25,9 @@ Please see the file LICENSE for full copyright information.
 #define LIBSAIS64_H 1
 
 #define LIBSAIS64_VERSION_MAJOR   2
-#define LIBSAIS64_VERSION_MINOR   7
-#define LIBSAIS64_VERSION_PATCH   5
-#define LIBSAIS64_VERSION_STRING  "2.7.5"
+#define LIBSAIS64_VERSION_MINOR   8
+#define LIBSAIS64_VERSION_PATCH   0
+#define LIBSAIS64_VERSION_STRING  "2.8.0"
 
 #ifdef _WIN32
     #ifdef LIBSAIS_SHARED

src/libsais.c

@@ -7761,6 +7761,54 @@ static void libsais_compute_plcp_omp(const uint8_t * RESTRICT T, sa_sint_t * RES
     }
 }
 
+static void libsais_compute_plcp_int(const int32_t * RESTRICT T, sa_sint_t * RESTRICT PLCP, fast_sint_t n, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
+{
+    const fast_sint_t prefetch_distance = 32;
+
+    fast_sint_t i, j, l = 0;
+    for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance; i < j; i += 1)
+    {
+        libsais_prefetchw(&PLCP[i + 2 * prefetch_distance]);
+        libsais_prefetchr(&T[PLCP[i + prefetch_distance] + l]);
+
+        fast_sint_t k = PLCP[i], m = n - (i > k ? i : k);
+        while (l < m && T[i + l] == T[k + l]) { l++; }
+
+        PLCP[i] = (sa_sint_t)l; l -= (l != 0);
+    }
+
+    for (j += prefetch_distance; i < j; i += 1)
+    {
+        fast_sint_t k = PLCP[i], m = n - (i > k ? i : k);
+        while (l < m && T[i + l] == T[k + l]) { l++; }
+
+        PLCP[i] = (sa_sint_t)l; l -= (l != 0);
+    }
+}
+
+static void libsais_compute_plcp_int_omp(const int32_t * RESTRICT T, sa_sint_t * RESTRICT PLCP, sa_sint_t n, sa_sint_t threads)
+{
+#if defined(LIBSAIS_OPENMP)
+    #pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
+#endif
+    {
+#if defined(LIBSAIS_OPENMP)
+        fast_sint_t omp_thread_num    = omp_get_thread_num();
+        fast_sint_t omp_num_threads   = omp_get_num_threads();
+#else
+        UNUSED(threads);
+
+        fast_sint_t omp_thread_num    = 0;
+        fast_sint_t omp_num_threads   = 1;
+#endif
+        fast_sint_t omp_block_stride  = (n / omp_num_threads) & (-16);
+        fast_sint_t omp_block_start   = omp_thread_num * omp_block_stride;
+        fast_sint_t omp_block_size    = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
+
+        libsais_compute_plcp_int(T, PLCP, n, omp_block_start, omp_block_size);
+    }
+}
+
 static void libsais_compute_lcp(const sa_sint_t * RESTRICT PLCP, const sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT LCP, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
 {
     const fast_sint_t prefetch_distance = 32;
@@ -7831,6 +7879,24 @@ int32_t libsais_plcp(const uint8_t * T, const int32_t * SA, int32_t * PLCP, int3
     return 0;
 }
 
+int32_t libsais_plcp_int(const int32_t * T, const int32_t * SA, int32_t * PLCP, int32_t n)
+{
+    if ((T == NULL) || (SA == NULL) || (PLCP == NULL) || (n < 0))
+    {
+        return -1;
+    }
+    else if (n <= 1)
+    {
+        if (n == 1) { PLCP[0] = 0; }
+        return 0;
+    }
+
+    libsais_compute_phi_omp(SA, PLCP, n, 1);
+    libsais_compute_plcp_int_omp(T, PLCP, n, 1);
+
+    return 0;
+}
+
 int32_t libsais_lcp(const int32_t * PLCP, const int32_t * SA, int32_t * LCP, int32_t n)
 {
     if ((PLCP == NULL) || (SA == NULL) || (LCP == NULL) || (n < 0))
@@ -7870,6 +7936,26 @@ int32_t libsais_plcp_omp(const uint8_t * T, const int32_t * SA, int32_t * PLCP,
     return 0;
 }
 
+int32_t libsais_plcp_int_omp(const int32_t * T, const int32_t * SA, int32_t * PLCP, int32_t n, int32_t threads)
+{
+    if ((T == NULL) || (SA == NULL) || (PLCP == NULL) || (n < 0) || (threads < 0))
+    {
+        return -1;
+    }
+    else if (n <= 1)
+    {
+        if (n == 1) { PLCP[0] = 0; }
+        return 0;
+    }
+
+    threads = threads > 0 ? threads : omp_get_max_threads();
+
+    libsais_compute_phi_omp(SA, PLCP, n, threads);
+    libsais_compute_plcp_int_omp(T, PLCP, n, threads);
+
+    return 0;
+}
+
 int32_t libsais_lcp_omp(const int32_t * PLCP, const int32_t * SA, int32_t * LCP, int32_t n, int32_t threads)
 {
     if ((PLCP == NULL) || (SA == NULL) || (LCP == NULL) || (n < 0) || (threads < 0))

src/libsais64.c

@@ -6288,7 +6288,7 @@ static void libsais64_convert_inplace_32u_to_64u_omp(uint32_t * V, sa_sint_t n,
             fast_sint_t omp_block_start     = omp_thread_num * omp_block_stride;
             fast_sint_t omp_block_size      = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
 
-            libsais64_convert_32u_to_64u((uint32_t *)V, (uint64_t *)V, n + omp_block_start, omp_block_size);
+            libsais64_convert_32u_to_64u(((uint32_t *)(void *)V) + n, ((uint64_t *)(void *)V) + n, omp_block_start, omp_block_size);
         }
     }