summaryrefslogtreecommitdiff
path: root/src/cuda-sim
diff options
context:
space:
mode:
authorMengchi Zhang <[email protected]>2019-06-22 13:09:02 -0400
committerGitHub <[email protected]>2019-06-22 13:09:02 -0400
commit067ca5fcce6ef4fa51ed3949dbffe62f60f6ad12 (patch)
treeb344dfc77c47d1d789ac503b0420f650c904972a /src/cuda-sim
parent3cac7525e1046285337e4dcf7fd8aa07c3a1f7c9 (diff)
parent2f749f6969cd0582b7ac2bfe33309034185cc77c (diff)
Merge pull request #19 from echoedit/dev
Dev
Diffstat (limited to 'src/cuda-sim')
-rw-r--r--src/cuda-sim/instructions.cc1
-rw-r--r--src/cuda-sim/ptx_sim.h1
2 files changed, 2 insertions, 0 deletions
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 186bb1e..46a262d 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1528,6 +1528,7 @@ void bar_impl( const ptx_instruction *pIin, ptx_thread_info *thread )
op2_data = thread->get_operand_value(op2, op2, PRED_TYPE, thread, 1);
op2_data.u32=!(op2_data.pred & 0x0001);
pI->set_bar_id(op1_data.u32);
+ pI->set_bar_count(thread->get_ntid().x * thread->get_ntid().y * thread->get_ntid().z);
switch(red_op){
case ATOMIC_POPC:
thread->popc_reduction(ctaid,op1_data.u32,op2_data.u32);
diff --git a/src/cuda-sim/ptx_sim.h b/src/cuda-sim/ptx_sim.h
index d226fbe..a045a76 100644
--- a/src/cuda-sim/ptx_sim.h
+++ b/src/cuda-sim/ptx_sim.h
@@ -341,6 +341,7 @@ public:
dim3 get_ctaid() const { return m_ctaid; }
dim3 get_tid() const { return m_tid; }
+ dim3 get_ntid() const { return m_ntid; }
class gpgpu_sim *get_gpu() { return (gpgpu_sim*)m_gpu;}
unsigned get_hw_tid() const { return m_hw_tid;}
unsigned get_hw_ctaid() const { return m_hw_ctaid;}