From 6005bfac9e14c07a602afc2eb967ca97a8b04af5 Mon Sep 17 00:00:00 2001
From: Malcolm Roberts <malcolm.roberts@amd.com>
Date: Mon, 21 Nov 2022 15:51:13 -0700
Subject: [PATCH] Use hipGetDevice and hipSetDevice to ensure RTC threads are
 on the correct device.

---
 library/src/rtc_kernel.cpp | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/library/src/rtc_kernel.cpp b/library/src/rtc_kernel.cpp
index 05388f13..21b050e3 100644
--- a/library/src/rtc_kernel.cpp
+++ b/library/src/rtc_kernel.cpp
@@ -114,6 +114,13 @@ std::shared_future<std::unique_ptr<RTCKernel>> RTCKernel::runtime_compile(
 {
 
 #ifdef ROCFFT_RUNTIME_COMPILE
+
+    int deviceId = 0;
+    if(hipGetDevice(&deviceId) != hipSuccess)
+    {
+        throw std::runtime_error("failed to get device");
+    }
+
     RTCGenerator generator;
     // try each type of generator until one is valid
     generator = RTCKernelStockham::generate_from_node(node, gpu_arch, enable_callbacks);
@@ -122,6 +129,10 @@ std::shared_future<std::unique_ptr<RTCKernel>> RTCKernel::runtime_compile(
         std::string kernel_name = generator.generate_name();
 
         auto compile = [=]() {
+            if(hipSetDevice(deviceId) != hipSuccess)
+            {
+                throw std::runtime_error("failed to set device");
+            }
             try
             {
                 std::vector<char> code = cached_compile(