From e03e9f74b667acdcfccd7a05abe568b95a12ac53 Mon Sep 17 00:00:00 2001
From: Lurch <Lurch>
Date: Fri, 13 Jan 2017 20:06:44 +0100
Subject: [PATCH] Changed demo_cdpSimplePrint.py to use DynamicSourceModule.
 Using Context.set_limit() now. Needed a missing limit enum
 CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH (in CUDA8/v8.0/include/cuda.h[975], available
 since 3.5), added CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH and
 CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT (same) to pycuda.driver.limit.

---
 examples/demo_cdpSimplePrint.py | 14 +++-----------
 src/wrapper/wrap_cudadrv.cpp    |  4 ++++
 2 files changed, 7 insertions(+), 11 deletions(-)

diff --git a/examples/demo_cdpSimplePrint.py b/examples/demo_cdpSimplePrint.py
index d00ba750..d5435987 100644
--- a/examples/demo_cdpSimplePrint.py
+++ b/examples/demo_cdpSimplePrint.py
@@ -20,8 +20,7 @@
 import sys, os
 import pycuda.autoinit
 import pycuda.driver as cuda
-from pycuda.compiler import JitLinkModule
-from pycuda.driver import jit_input_type
+from pycuda.compiler import DynamicSourceModule
 
 cdpSimplePrint_cu = '''
 #include <cstdio>
@@ -96,11 +95,7 @@ def main(argv):
 
     print("starting Simple Print (CUDA Dynamic Parallelism)")
 
-    mod = JitLinkModule()
-    mod.add_source(cdpSimplePrint_cu, nvcc_options=['-O3', '-rdc=true', '-lcudadevrt'])
-    mod.add_stdlib('cudadevrt')
-    mod.link()
-
+    mod = DynamicSourceModule(cdpSimplePrint_cu)
     cdp_kernel = mod.get_function('cdp_kernel').prepare('iiii').prepared_call
 
     print("***************************************************************************")
@@ -113,14 +108,11 @@ def main(argv):
         num_blocks *= 4
         print("+%d" % num_blocks)
         sum += num_blocks
-
     print("=%d blocks are launched!!! (%d from the GPU)" % (sum, sum-2))
     print("***************************************************************************\n")
 
-    # TODO: cudaDeviceSetLimit() is not available on PyCuda, works anyway on my GeForce GTX 980; maybe add that function?
-    #cudaDeviceSetLimit( cudaLimitDevRuntimeSyncDepth, max_depth )
+    pycuda.autoinit.context.set_limit(cuda.limit.DEV_RUNTIME_SYNC_DEPTH, max_depth)
 
-    # Launch the kernel from the CPU.
     print("Launching cdp_kernel() with CUDA Dynamic Parallelism:\n")
     cdp_kernel((2,1), (2,1,1), max_depth, 0, 0, -1)
 
diff --git a/src/wrapper/wrap_cudadrv.cpp b/src/wrapper/wrap_cudadrv.cpp
index 4f126065..76f30df4 100644
--- a/src/wrapper/wrap_cudadrv.cpp
+++ b/src/wrapper/wrap_cudadrv.cpp
@@ -1010,6 +1010,10 @@ BOOST_PYTHON_MODULE(_driver)
     .value("PRINTF_FIFO_SIZE", CU_LIMIT_PRINTF_FIFO_SIZE)
 #if CUDAPP_CUDA_VERSION >= 3020
     .value("MALLOC_HEAP_SIZE", CU_LIMIT_MALLOC_HEAP_SIZE)
+#endif
+#if CUDAPP_CUDA_VERSION >= 3050
+    .value("DEV_RUNTIME_SYNC_DEPTH", CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH)
+    .value("DEV_RUNTIME_PENDING_LAUNCH_COUNT", CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT)
 #endif
     ;
 #endif