From e79d63ed5681187d15ea5fc048da4687cae8ac49 Mon Sep 17 00:00:00 2001
From: Riley Murray <rjmurr@sandia.gov>
Date: Wed, 11 Sep 2024 22:14:53 -0400
Subject: [PATCH] turn off compiler optimizations in box-muller. Change default
 error threshold for weights_to_cdf. Add missing include in sparse_skops.hh

---
 RandBLAS/random_gen.hh   | 2 ++
 RandBLAS/sparse_skops.hh | 1 +
 RandBLAS/util.hh         | 6 +++++-
 3 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/RandBLAS/random_gen.hh b/RandBLAS/random_gen.hh
index f502e981..2247ea52 100644
--- a/RandBLAS/random_gen.hh
+++ b/RandBLAS/random_gen.hh
@@ -74,7 +74,9 @@ static inline void sincospi(double x, double *s, double *c) {
 #include <Random123/philox.h>
 #include <Random123/threefry.h>
 // NOTE: we do not support Random123's AES or ARS generators.
+#pragma clang optimize off
 #include <Random123/boxmuller.hpp>
+#pragma clang optimize on
 #include <Random123/uniform.hpp>
 
 /// our extensions to random123
diff --git a/RandBLAS/sparse_skops.hh b/RandBLAS/sparse_skops.hh
index a946ec08..1bd320cb 100644
--- a/RandBLAS/sparse_skops.hh
+++ b/RandBLAS/sparse_skops.hh
@@ -40,6 +40,7 @@
 #include <cstdio>
 #include <cmath>
 #include <algorithm>
+#include <unordered_map>
 
 #define MAX(a, b) (((a) < (b)) ? (b) : (a))
 #define MIN(a, b) (((a) < (b)) ? (a) : (b))
diff --git a/RandBLAS/util.hh b/RandBLAS/util.hh
index 3cd6b7a1..6cfd02ea 100644
--- a/RandBLAS/util.hh
+++ b/RandBLAS/util.hh
@@ -328,6 +328,10 @@ void transpose_square(T* A, int64_t n, int64_t lda) {
     return;
 }
 
+template <typename T>
+T sqrt_epsilon() {
+    return std::sqrt(std::numeric_limits<T>::epsilon());
+}
 
 // =============================================================================
 /// \fn weights_to_cdf(int64_t n, T* w, T error_if_below = -std::numeric_limits<T>::epsilon())
@@ -344,7 +348,7 @@ void transpose_square(T* A, int64_t n, int64_t lda) {
 /// On exit, \math{w} is a CDF suitable for use with sample_indices_iid.
 ///
 template <typename T>
-void weights_to_cdf(int64_t n, T* w, T error_if_below = -std::numeric_limits<T>::epsilon()) {
+void weights_to_cdf(int64_t n, T* w, T error_if_below = -sqrt_epsilon<T>()) {
     T sum = 0.0;
     for (int64_t i = 0; i < n; ++i) {
         T val = w[i];