From d4b23da91f4b6ed58dde36505c5fa04672d2bea3 Mon Sep 17 00:00:00 2001
From: Ben Sander <ben.sander@amd.com>
Date: Tue, 18 Oct 2016 22:25:26 -0500
Subject: [PATCH] Add HIP_WAIT_MODE env var.

Also weaken cases where hipSetDeviceFlags returns hipErrorInvalidValue.

Change-Id: I7f113338be6fe498eaf1ab40fd0fd6b23849bb5e
---
 hipamd/src/hip_device.cpp |  5 +++--
 hipamd/src/hip_hcc.cpp    | 12 +++++++++---
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/hipamd/src/hip_device.cpp b/hipamd/src/hip_device.cpp
index 7168ccfbe6..9d577f5313 100644
--- a/hipamd/src/hip_device.cpp
+++ b/hipamd/src/hip_device.cpp
@@ -285,14 +285,15 @@ hipError_t hipSetDeviceFlags( unsigned int flags)
                    e = hipSuccess;
                    break;
                default:
-                   e = hipErrorInvalidValue;
+                   e = hipSuccess; // TODO - should this be error?  Map to Auto?
+                   //e = hipErrorInvalidValue;
                    break;
            }
        }
 
        unsigned supportedFlags = hipDeviceScheduleMask | hipDeviceMapHost | hipDeviceLmemResizeToMax; 
 
-       if (flags & ~supportedFlags) {
+       if (flags & (~supportedFlags)) {
           e = hipErrorInvalidValue;
        }
     } else {
diff --git a/hipamd/src/hip_hcc.cpp b/hipamd/src/hip_hcc.cpp
index 2d46c1ead8..fd3292a6f4 100644
--- a/hipamd/src/hip_hcc.cpp
+++ b/hipamd/src/hip_hcc.cpp
@@ -66,7 +66,7 @@ int HIP_ATP_MARKER= 0;
 int HIP_DB= 0;
 int HIP_VISIBLE_DEVICES = 0; /* Contains a comma-separated sequence of GPU identifiers */
 int HIP_NUM_KERNELS_INFLIGHT = 128;
-int HIP_FORCE_BLOCKING_SYNC = 0;
+int HIP_WAIT_MODE = 0;
 
 #define HIP_USE_PRODUCT_NAME 0
 //#define DISABLE_COPY_EXT 1
@@ -285,8 +285,14 @@ void ihipStream_t::wait(LockedAccessor_StreamCrit_t &crit, bool assertQueueEmpty
         } else {
             assert(0); // bad wait mode.
         }
+
+        if (HIP_WAIT_MODE == 1) {
+            waitMode = hc::hcWaitModeBlocked;
+        } else if (HIP_WAIT_MODE == 2) {
+            waitMode = hc::hcWaitModeActive;
+        }
             
-        crit->_av.wait(HIP_FORCE_BLOCKING_SYNC ? hc::hcWaitModeBlocked : waitMode);
+        crit->_av.wait(waitMode);
     }
 
     crit->_kernelCnt = 0;
@@ -1120,7 +1126,7 @@ void ihipInit()
     READ_ENV_I(release, HIP_VISIBLE_DEVICES, CUDA_VISIBLE_DEVICES, "Only devices whose index is present in the secquence are visible to HIP applications and they are enumerated in the order of secquence" );
 
 
-    READ_ENV_I(release, HIP_FORCE_BLOCKING_SYNC, 0, "Force blocking synchronization for stream waits.  This may increase latency but is friendlier to other processes. If 0, used .");
+    READ_ENV_I(release, HIP_WAIT_MODE, 0, "Force synchronization mode. 1= force yield, 2=force spin, 0=defaults specified in application");
 
     READ_ENV_I(release, HIP_NUM_KERNELS_INFLIGHT, 128, "Max number of inflight kernels per stream before active synchronization is forced.");