From 5dfce9310190fc9c8ae653208ed8ce84d7bb02e6 Mon Sep 17 00:00:00 2001
From: guoshengCS <guosheng@baidu.com>
Date: Thu, 31 Jan 2019 01:44:09 +0800
Subject: [PATCH] To make CUDA_LAUNCH_KERNEL_HELPER support large size.
 test=develop

---
 paddle/fluid/platform/cuda_device_function.h | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/paddle/fluid/platform/cuda_device_function.h b/paddle/fluid/platform/cuda_device_function.h
index 2ce8f141d..31b6c38d6 100644
--- a/paddle/fluid/platform/cuda_device_function.h
+++ b/paddle/fluid/platform/cuda_device_function.h
@@ -53,10 +53,12 @@ inline static int RoundToPowerOfTwo(int dim) {
     __VA_ARGS__;                           \
   } break
 
-#define CUDA_LAUNCH_KERNEL_HELPER(...)         \
-  CUDA_LAUNCH_KERNEL_BASE(256, ##__VA_ARGS__); \
-  CUDA_LAUNCH_KERNEL_BASE(128, ##__VA_ARGS__); \
-  CUDA_LAUNCH_KERNEL_BASE(64, ##__VA_ARGS__);  \
+#define CUDA_LAUNCH_KERNEL_HELPER(...)          \
+  CUDA_LAUNCH_KERNEL_BASE(1024, ##__VA_ARGS__); \
+  CUDA_LAUNCH_KERNEL_BASE(512, ##__VA_ARGS__);  \
+  CUDA_LAUNCH_KERNEL_BASE(256, ##__VA_ARGS__);  \
+  CUDA_LAUNCH_KERNEL_BASE(128, ##__VA_ARGS__);  \
+  CUDA_LAUNCH_KERNEL_BASE(64, ##__VA_ARGS__);   \
   CUDA_LAUNCH_KERNEL_BASE(32, ##__VA_ARGS__);
 
 template <typename T>
-- 
GitLab