From 8c564e9b57dc0fb913ce31de99c6ad556ae04fa7 Mon Sep 17 00:00:00 2001
From: Christian Sigg <chsigg@users.noreply.github.com>
Date: Fri, 6 Dec 2019 18:14:55 +0100
Subject: [PATCH 1/2] Fix clang build (#271)

Clang doesn't understand `optimize("O0")`. It has `noopt`, which GCC doesn't understand. Wrap the difference in a macro.
---
 src/init.cc | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/src/init.cc b/src/init.cc
index 1ee2a73..15a9e7c 100644
--- a/src/init.cc
+++ b/src/init.cc
@@ -124,10 +124,18 @@ ncclResult_t ncclGetUniqueId(ncclUniqueId* out) {
 }
 
 // Prevent compiler from optimizing out these operations
-void __attribute__((optimize("O0"))) commPoison(ncclComm_t comm) {
+#ifdef __clang__
+#define NCCL_NO_OPTIMIZE __attribute__((noopt)))
+#else
+#define NCCL_NO_OPTIMIZE __attribute__((optimize("O0")))
+#endif
+
+void NCCL_NO_OPTIMIZE commPoison(ncclComm_t comm) {
   comm->rank = comm->cudaDev = comm->busId = comm->nRanks = -1;
 }
 
+#undef NCCL_NO_OPTIMIZE
+
 static ncclResult_t commFree(ncclComm_t comm) {
   if (comm == NULL)
     return ncclSuccess;

From aa15dfb29c05adc4f2c895aa72f8873e125560fb Mon Sep 17 00:00:00 2001
From: Sylvain Jeaugey <sjeaugey@nvidia.com>
Date: Fri, 6 Dec 2019 09:55:54 -0800
Subject: [PATCH 2/2] Fix clang compilation

---
 src/graph/tuning.cc | 2 +-
 src/init.cc         | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/graph/tuning.cc b/src/graph/tuning.cc
index 89a97a3..87afb2f 100644
--- a/src/graph/tuning.cc
+++ b/src/graph/tuning.cc
@@ -192,7 +192,7 @@ ncclResult_t ncclSetThresholds(struct ncclComm* comm, int minCompCap, int maxCom
   // Override defaults with user env
   char* str = getenv("NCCL_THREAD_THRESHOLDS");
   if (str) {
-    ssize_t t[NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS] = { -2 };
+    ssize_t t[NCCL_NUM_ALGORITHMS][NCCL_NUM_PROTOCOLS] = {{ -2, -2, -2 }, { -2, -2, -2}};
     sscanf(str, "%ld %ld %ld %ld %ld %ld", t[0], t[0]+1, t[0]+2, t[1], t[1]+1, t[1]+2);
     for (int a=0; a<NCCL_NUM_ALGORITHMS; a++) {
       for (int p=0; p<NCCL_NUM_PROTOCOLS; p++) {
diff --git a/src/init.cc b/src/init.cc
index 15a9e7c..9da676a 100644
--- a/src/init.cc
+++ b/src/init.cc
@@ -125,7 +125,7 @@ ncclResult_t ncclGetUniqueId(ncclUniqueId* out) {
 
 // Prevent compiler from optimizing out these operations
 #ifdef __clang__
-#define NCCL_NO_OPTIMIZE __attribute__((noopt)))
+#define NCCL_NO_OPTIMIZE __attribute__((noopt))
 #else
 #define NCCL_NO_OPTIMIZE __attribute__((optimize("O0")))
 #endif