From e9a069ef92f2477371f2956d95f8432a3bb7cd5c Mon Sep 17 00:00:00 2001
From: Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
Date: Fri, 20 Oct 2023 11:28:25 -0700
Subject: [PATCH 1/5] Replace long double with double

Use lround instead of lroundl, for perf improvements on aarch64 machines.
---
 src/h3lib/lib/coordijk.c | 44 ++++++++++++++++++++--------------------
 1 file changed, 22 insertions(+), 22 deletions(-)

diff --git a/src/h3lib/lib/coordijk.c b/src/h3lib/lib/coordijk.c
index 137084c75..9db3f163e 100644
--- a/src/h3lib/lib/coordijk.c
+++ b/src/h3lib/lib/coordijk.c
@@ -1,5 +1,5 @@
 /*
- * Copyright 2016-2018, 2020-2022 Uber Technologies, Inc.
+ * Copyright 2016-2018, 2020-2023 Uber Technologies, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -67,7 +67,7 @@ void _hex2dToCoordIJK(const Vec2d *v, CoordIJK *h) {
 
     // first do a reverse conversion
     x2 = a2 / M_SIN60;
-    x1 = a1 + x2 / 2.0L;
+    x1 = a1 + x2 / 2.0;
 
     // check if we have the center of a hex
     m1 = x1;
@@ -77,9 +77,9 @@ void _hex2dToCoordIJK(const Vec2d *v, CoordIJK *h) {
     r1 = x1 - m1;
     r2 = x2 - m2;
 
-    if (r1 < 0.5L) {
-        if (r1 < 1.0L / 3.0L) {
-            if (r2 < (1.0L + r1) / 2.0L) {
+    if (r1 < 0.5) {
+        if (r1 < 1.0 / 3.0) {
+            if (r2 < (1.0 + r1) / 2.0) {
                 h->i = m1;
                 h->j = m2;
             } else {
@@ -87,33 +87,33 @@ void _hex2dToCoordIJK(const Vec2d *v, CoordIJK *h) {
                 h->j = m2 + 1;
             }
         } else {
-            if (r2 < (1.0L - r1)) {
+            if (r2 < (1.0 - r1)) {
                 h->j = m2;
             } else {
                 h->j = m2 + 1;
             }
 
-            if ((1.0L - r1) <= r2 && r2 < (2.0 * r1)) {
+            if ((1.0 - r1) <= r2 && r2 < (2.0 * r1)) {
                 h->i = m1 + 1;
             } else {
                 h->i = m1;
             }
         }
     } else {
-        if (r1 < 2.0L / 3.0L) {
-            if (r2 < (1.0L - r1)) {
+        if (r1 < 2.0 / 3.0) {
+            if (r2 < (1.0 - r1)) {
                 h->j = m2;
             } else {
                 h->j = m2 + 1;
             }
 
-            if ((2.0L * r1 - 1.0L) < r2 && r2 < (1.0L - r1)) {
+            if ((2.0 * r1 - 1.0) < r2 && r2 < (1.0 - r1)) {
                 h->i = m1;
             } else {
                 h->i = m1 + 1;
             }
         } else {
-            if (r2 < (r1 / 2.0L)) {
+            if (r2 < (r1 / 2.0)) {
                 h->i = m1 + 1;
                 h->j = m2;
             } else {
@@ -125,7 +125,7 @@ void _hex2dToCoordIJK(const Vec2d *v, CoordIJK *h) {
 
     // now fold across the axes if necessary
 
-    if (v->x < 0.0L) {
+    if (v->x < 0.0) {
         if ((h->j % 2) == 0)  // even
         {
             long long int axisi = h->j / 2;
@@ -138,7 +138,7 @@ void _hex2dToCoordIJK(const Vec2d *v, CoordIJK *h) {
         }
     }
 
-    if (v->y < 0.0L) {
+    if (v->y < 0.0) {
         h->i = h->i - (2 * h->j + 1) / 2;
         h->j = -1 * h->j;
     }
@@ -156,7 +156,7 @@ void _ijkToHex2d(const CoordIJK *h, Vec2d *v) {
     int i = h->i - h->k;
     int j = h->j - h->k;
 
-    v->x = i - 0.5L * j;
+    v->x = i - 0.5 * j;
     v->y = j * M_SQRT3_2;
 }
 
@@ -346,8 +346,8 @@ H3Error _upAp7Checked(CoordIJK *ijk) {
     }
 
     // TODO: Do the int math parts here in long double?
-    ijk->i = (int)lroundl(((i * 3) - j) / 7.0L);
-    ijk->j = (int)lroundl((i + (j * 2)) / 7.0L);
+    ijk->i = (int)lround(((i * 3) - j) / 7.0);
+    ijk->j = (int)lround((i + (j * 2)) / 7.0);
     ijk->k = 0;
 
     // Expected not to be reachable, because max + min or max - min would need
@@ -395,8 +395,8 @@ H3Error _upAp7rChecked(CoordIJK *ijk) {
     }
 
     // TODO: Do the int math parts here in long double?
-    ijk->i = (int)lroundl(((i * 2) + j) / 7.0L);
-    ijk->j = (int)lroundl(((j * 3) - i) / 7.0L);
+    ijk->i = (int)lround(((i * 2) + j) / 7.0);
+    ijk->j = (int)lround(((j * 3) - i) / 7.0);
     ijk->k = 0;
 
     // Expected not to be reachable, because max + min or max - min would need
@@ -419,8 +419,8 @@ void _upAp7(CoordIJK *ijk) {
     int i = ijk->i - ijk->k;
     int j = ijk->j - ijk->k;
 
-    ijk->i = (int)lroundl((3 * i - j) / 7.0L);
-    ijk->j = (int)lroundl((i + 2 * j) / 7.0L);
+    ijk->i = (int)lround((3 * i - j) / 7.0);
+    ijk->j = (int)lround((i + 2 * j) / 7.0);
     ijk->k = 0;
     _ijkNormalize(ijk);
 }
@@ -436,8 +436,8 @@ void _upAp7r(CoordIJK *ijk) {
     int i = ijk->i - ijk->k;
     int j = ijk->j - ijk->k;
 
-    ijk->i = (int)lroundl((2 * i + j) / 7.0L);
-    ijk->j = (int)lroundl((3 * j - i) / 7.0L);
+    ijk->i = (int)lround((2 * i + j) / 7.0);
+    ijk->j = (int)lround((3 * j - i) / 7.0);
     ijk->k = 0;
     _ijkNormalize(ijk);
 }

From f2f229dbedfb4f92dfdfdc308766db9271920f8e Mon Sep 17 00:00:00 2001
From: Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
Date: Fri, 20 Oct 2023 11:30:40 -0700
Subject: [PATCH 2/5] Replace long double with double

Use lround instead of lroundl, for perf improvements on aarch64 machines.
---
 src/h3lib/lib/faceijk.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/h3lib/lib/faceijk.c b/src/h3lib/lib/faceijk.c
index b46f38e04..aeb29bfd9 100644
--- a/src/h3lib/lib/faceijk.c
+++ b/src/h3lib/lib/faceijk.c
@@ -33,7 +33,7 @@
 #include "vec3d.h"
 
 /** square root of 7 */
-#define M_SQRT7 2.6457513110645905905016157536392604257102L
+#define M_SQRT7 2.6457513110645905905016157536392604257102
 
 /** @brief icosahedron face centers in lat/lng radians */
 const LatLng faceCenterGeo[NUM_ICOSA_FACES] = {
@@ -395,7 +395,7 @@ void _geoToHex2d(const LatLng *g, int res, int *face, Vec2d *v) {
     double r = acos(1 - sqd / 2);
 
     if (r < EPSILON) {
-        v->x = v->y = 0.0L;
+        v->x = v->y = 0.0;
         return;
     }
 

From f9f487df7b6840405392f3d0e112b5f8ab87a8d3 Mon Sep 17 00:00:00 2001
From: Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
Date: Fri, 20 Oct 2023 11:31:57 -0700
Subject: [PATCH 3/5] Replace long double with double

Use lround instead of lroundl, for perf improvements on aarch64 machines.
---
 src/h3lib/lib/latLng.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/h3lib/lib/latLng.c b/src/h3lib/lib/latLng.c
index fbaed5e6c..5c2c098e0 100644
--- a/src/h3lib/lib/latLng.c
+++ b/src/h3lib/lib/latLng.c
@@ -1,5 +1,5 @@
 /*
- * Copyright 2016-2021 Uber Technologies, Inc.
+ * Copyright 2016-2023 Uber Technologies, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -34,7 +34,7 @@
  * @return The normalized radians value.
  */
 double _posAngleRads(double rads) {
-    double tmp = ((rads < 0.0L) ? rads + M_2PI : rads);
+    double tmp = ((rads < 0.0) ? rads + M_2PI : rads);
     if (rads >= M_2PI) tmp -= M_2PI;
     return tmp;
 }

From b04acb0ef889c3f8147fb06ea8cc05d75e4217a4 Mon Sep 17 00:00:00 2001
From: Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
Date: Fri, 20 Oct 2023 11:33:05 -0700
Subject: [PATCH 4/5] Update faceijk.c

Update copyright year.
---
 src/h3lib/lib/faceijk.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/h3lib/lib/faceijk.c b/src/h3lib/lib/faceijk.c
index aeb29bfd9..aee1d92a7 100644
--- a/src/h3lib/lib/faceijk.c
+++ b/src/h3lib/lib/faceijk.c
@@ -1,5 +1,5 @@
 /*
- * Copyright 2016-2021 Uber Technologies, Inc.
+ * Copyright 2016-2023 Uber Technologies, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

From ef0a87ac3c8ff38f301bc8aa0cafdb2b35d154f8 Mon Sep 17 00:00:00 2001
From: Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
Date: Fri, 20 Oct 2023 12:18:50 -0700
Subject: [PATCH 5/5] Update CHANGELOG.md

---
 CHANGELOG.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 45343b075..38f6b36c7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,7 @@ The public API of this library consists of the functions declared in file
 ## [Unreleased]
 ### Changed
 - Reorganize tests into public / internal. (#762)
+- Performance enhancement for aarch64, should not affect other platforms
 
 ## [4.1.0] - 2023-01-18
 ### Added