From df54ffb86fd309a0a70ad0e76e815aca14893dfd Mon Sep 17 00:00:00 2001
From: Ilya Grebnov <Ilya.Grebnov@gmail.com>
Date: Sat, 2 Dec 2023 17:39:55 -0800
Subject: [PATCH] Small performance optimization for esa_matchfinder_advance
 API.

---
 CHANGES           |  3 ++
 README.md         |  2 ++
 VERSION           |  2 +-
 esa_matchfinder.c | 81 +++++++++++++++++++++++++++++++++++++++++++++++
 esa_matchfinder.h |  4 +--
 5 files changed, 89 insertions(+), 3 deletions(-)

diff --git a/CHANGES b/CHANGES
index 0bb6703..a43b1a1 100644
--- a/CHANGES
+++ b/CHANGES
@@ -1,3 +1,6 @@
+Changes in 1.2.0 (December 2, 2023)
+- Small performance optimization for esa_matchfinder_advance API.
+
 Changes in 1.1.0 (November 30, 2023)
 - New API to find matches within specified sliding window.
 
diff --git a/README.md b/README.md
index 4a96723..65ed225 100644
--- a/README.md
+++ b/README.md
@@ -29,6 +29,8 @@ The esa-matchfinder finds all distance optimal matches (between min_match_length
 The esa-matchfinder released under the [Apache License Version 2.0](LICENSE "Apache license") and is considered suitable for production use. However, no warranty or fitness for a particular purpose is expressed or implied.
 
 ## Changes
+* December 2, 2023 (1.2.0)
+  * Small performance optimization for esa_matchfinder_advance API.
 * November 30, 2023 (1.1.0)
   * New API to find matches within specified sliding window.
 * June 19, 2022 (1.0.1)
diff --git a/VERSION b/VERSION
index 1cc5f65..867e524 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.1.0
\ No newline at end of file
+1.2.0
\ No newline at end of file
diff --git a/esa_matchfinder.c b/esa_matchfinder.c
index 12656d0..5fcacb5 100644
--- a/esa_matchfinder.c
+++ b/esa_matchfinder.c
@@ -914,8 +914,89 @@ ESA_MATCHFINDER_MATCH esa_matchfinder_find_best_match_in_window(void * mf, uint6
     }
 }
 
+static void esa_matchfinder_advance_backwards(void * mf, int32_t count)
+{
+    ESA_MF_CONTEXT * ESA_MF_RESTRICT const          matchfinder_ctx     = (ESA_MF_CONTEXT *)mf;
+
+    const ptrdiff_t                                 prefetch_distance   = 4;
+    const uint64_t                                  current_position    = matchfinder_ctx->position;
+    const uint64_t                                  target_position     = matchfinder_ctx->position += (uint64_t)count;
+
+    uint64_t * ESA_MF_RESTRICT const                sa_parent_link      = matchfinder_ctx->sa_parent_link;
+    uint32_t * ESA_MF_RESTRICT const                plcp_leaf_link      = matchfinder_ctx->plcp_leaf_link;
+
+    memset(matchfinder_ctx->prefetch, 0, sizeof(matchfinder_ctx->prefetch));
+
+    for (uint64_t position = target_position + prefetch_distance * 8; position-- != target_position; )
+    {
+        uint64_t * ESA_MF_RESTRICT const prefetch = &matchfinder_ctx->prefetch[position & (prefetch_distance - 1)][0];
+
+        esa_matchfinder_prefetchw(&sa_parent_link[              (sa_parent_link[prefetch[0]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[0] = (sa_parent_link[prefetch[1]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[1] = (sa_parent_link[prefetch[2]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[2] = (sa_parent_link[prefetch[3]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[3] = (sa_parent_link[prefetch[4]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[4] = (sa_parent_link[prefetch[5]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[5] = (sa_parent_link[prefetch[6]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[6] = (plcp_leaf_link[position - 8 * prefetch_distance])]);
+        esa_matchfinder_prefetchr(&plcp_leaf_link[position - 9 * prefetch_distance]);
+    }
+
+    for (uint64_t position = target_position; position-- != current_position; )
+    {
+        if (position >= 8 * prefetch_distance)
+        {
+            uint64_t * ESA_MF_RESTRICT const prefetch = &matchfinder_ctx->prefetch[position & (prefetch_distance - 1)][0];
+
+            esa_matchfinder_prefetchw(&sa_parent_link[              (sa_parent_link[prefetch[0]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[0] = (sa_parent_link[prefetch[1]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[1] = (sa_parent_link[prefetch[2]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[2] = (sa_parent_link[prefetch[3]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[3] = (sa_parent_link[prefetch[4]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[4] = (sa_parent_link[prefetch[5]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[5] = (sa_parent_link[prefetch[6]] & ESA_MF_PARENT_MASK)]);
+            esa_matchfinder_prefetchw(&sa_parent_link[prefetch[6] = (plcp_leaf_link[position - 8 * prefetch_distance])]);
+            esa_matchfinder_prefetchr(&plcp_leaf_link[position - 9 * prefetch_distance]);
+        }
+
+        const uint64_t new_offset       = (uint64_t)position << ESA_MF_OFFSET_SHIFT;
+        uint64_t reference              = plcp_leaf_link[position];
+        uint64_t interval               = sa_parent_link[reference];
+
+        while ((interval & ESA_MF_OFFSET_MASK) < new_offset)
+        {
+            sa_parent_link[reference]   = (interval & (~ESA_MF_OFFSET_MASK)) + new_offset;
+            reference                   = interval & ESA_MF_PARENT_MASK;
+            interval                    = sa_parent_link[reference];
+        }
+    }
+
+    memset(matchfinder_ctx->prefetch, 0, sizeof(matchfinder_ctx->prefetch));
+
+    for (uint64_t position = target_position - prefetch_distance * 8; position != target_position; position += 1)
+    {
+        uint64_t * ESA_MF_RESTRICT const prefetch = &matchfinder_ctx->prefetch[position & (prefetch_distance - 1)][0];
+
+        esa_matchfinder_prefetchw(&sa_parent_link[              (sa_parent_link[prefetch[0]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[0] = (sa_parent_link[prefetch[1]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[1] = (sa_parent_link[prefetch[2]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[2] = (sa_parent_link[prefetch[3]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[3] = (sa_parent_link[prefetch[4]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[4] = (sa_parent_link[prefetch[5]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[5] = (sa_parent_link[prefetch[6]] & ESA_MF_PARENT_MASK)]);
+        esa_matchfinder_prefetchw(&sa_parent_link[prefetch[6] = (plcp_leaf_link[position + 8 * prefetch_distance])]);
+        esa_matchfinder_prefetchr(&plcp_leaf_link[position + 9 * prefetch_distance]);
+    }
+}
+
 void esa_matchfinder_advance(void * mf, int32_t count)
 {
+    if (count >= /*ESA_MF_ADVANCE_BACKWARDS_THRESHOLD*/ 64)
+    {
+        esa_matchfinder_advance_backwards(mf, count);
+        return;
+    }
+
     ESA_MF_CONTEXT * ESA_MF_RESTRICT const          matchfinder_ctx     = (ESA_MF_CONTEXT *)mf;
 
     const ptrdiff_t                                 prefetch_distance   = 4;
diff --git a/esa_matchfinder.h b/esa_matchfinder.h
index 4e6a8d1..c69f780 100644
--- a/esa_matchfinder.h
+++ b/esa_matchfinder.h
@@ -33,9 +33,9 @@ Please see the file LICENSE for full copyright and license details.
 #define ESA_MATCHFINDER_BAD_PARAMETER       (-1)
 
 #define ESA_MATCHFINDER_VERSION_MAJOR       1
-#define ESA_MATCHFINDER_VERSION_MINOR       1
+#define ESA_MATCHFINDER_VERSION_MINOR       2
 #define ESA_MATCHFINDER_VERSION_PATCH       0
-#define ESA_MATCHFINDER_VERSION_STRING      "1.1.0"
+#define ESA_MATCHFINDER_VERSION_STRING      "1.2.0"
 
 #ifdef __cplusplus
 extern "C" {