[Beignet] [PATCH] Change global barrier test temporary to check local barrier whether need fence.

[Yang, Rong R]

I just change the global memory barrier test case to local barrier, and it is random fail. 
If change barrier(CLK_LOCAL_MEM_FENCE) to barrier(CLK_GLOBAL_MEM_FENCE), it always pass. 
The diff between local fence and global fence is whether has memory fence before barrier.

-----Original Message-----
From: Yang, Rong R 
Sent: Tuesday, June 18, 2013 11:24 AM
To: beignet at lists.freedesktop.org
Cc: Yang, Rong R
Subject: [PATCH] Change global barrier test temporary to check local barrier whether need fence.


Signed-off-by: Yang Rong <rong.r.yang at intel.com>
---
 kernels/compiler_global_memory_barrier.cl |    4 ++--
 utests/compiler_global_memory_barrier.cpp |    9 +++++----
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/kernels/compiler_global_memory_barrier.cl b/kernels/compiler_global_memory_barrier.cl
index 99bb940..705517d 100644
--- a/kernels/compiler_global_memory_barrier.cl
+++ b/kernels/compiler_global_memory_barrier.cl
@@ -1,7 +1,7 @@
-__kernel void compiler_global_memory_barrier(__global int *dst, __global int *src) {
+__kernel void compiler_global_memory_barrier(__global int *dst, __local 
+int *src) {
   src[get_local_size(0) * (2 * get_group_id(0)) + get_local_id(0)] = get_local_id(0);
   src[get_local_size(0) * (2 * get_group_id(0) + 1) + get_local_id(0)] = get_local_id(0);
-  barrier(CLK_GLOBAL_MEM_FENCE);
+  barrier(CLK_LOCAL_MEM_FENCE);
   dst[get_local_size(0) * (2 * get_group_id(0)) + get_local_id(0)] = src[get_local_size(0) * 2 * get_group_id(0) + get_local_size(0) - (get_local_id(0) + 1)];
   dst[get_local_size(0) * (2 * get_group_id(0) + 1) + get_local_id(0)] = src[get_local_size(0) * (2 * get_group_id(0) + 1) + get_local_size(0) - (get_local_id(0) + 1)];  } diff --git a/utests/compiler_global_memory_barrier.cpp b/utests/compiler_global_memory_barrier.cpp
index a6496a7..311d093 100644
--- a/utests/compiler_global_memory_barrier.cpp
+++ b/utests/compiler_global_memory_barrier.cpp
@@ -4,16 +4,17 @@ static void compiler_global_memory_barrier(void)
 {
   const size_t n = 16*1024;
 
+  globals[0] = n/2;
+  locals[0] = 32;
+
   // Setup kernel and buffers
   OCL_CREATE_KERNEL("compiler_global_memory_barrier");
   OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(uint32_t), NULL);
-  OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(uint32_t), NULL);
+  //OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(uint32_t), NULL);
   OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
-  OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
+  OCL_SET_ARG(1, locals[0] * 2 * sizeof(uint32_t), NULL);
 
   // Run the kernel
-  globals[0] = n/2;
-  locals[0] = 32;
   OCL_NDRANGE(1);
   OCL_MAP_BUFFER(0);
 
--
1.7.10.4