From 5fc7bf5872aad126a09cad4b385054c4b3a094aa Mon Sep 17 00:00:00 2001
From: Jin Wang <wangjin.jane@gmail.com>
Date: Mon, 20 Oct 2014 17:19:32 -0400
Subject: BUG: do not handle cudaGetParameterBufferV2 and cudaLaunchDeviceV2 as
 call.uni in reconvergence

---
 src/abstract_hardware_model.h | 2 ++
 src/cuda-sim/cuda-sim.cc      | 6 ++++--
 src/cuda-sim/ptx_ir.cc        | 3 +++
 3 files changed, 9 insertions(+), 2 deletions(-)

(limited to 'src')

diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 16f4b31..11fee10 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -864,6 +864,7 @@ public:
         m_mem_accesses_created=false;
         m_cache_hit=false;
         m_is_printf=false;
+        m_is_cdp = false;
     }
     virtual ~warp_inst_t(){
     }
@@ -1016,6 +1017,7 @@ protected:
     unsigned cycles; // used for implementing initiation interval delay
     bool m_isatomic;
     bool m_is_printf;
+    bool m_is_cdp;
     unsigned m_warp_id;
     unsigned m_dynamic_warp_id; 
     const core_config *m_config; 
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 4281913..58bd4e0 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -638,7 +638,7 @@ void ptx_instruction::set_opcode_and_latency()
    case MEMBAR_OP: op = MEMORY_BARRIER_OP; break;
    case CALL_OP:
    {
-       if(m_is_printf)
+       if(m_is_printf || m_is_cdp)
            op = ALU_OP;
        else
            op = CALL_OPS;
@@ -646,7 +646,7 @@ void ptx_instruction::set_opcode_and_latency()
    }
    case CALLP_OP:
    {
-       if(m_is_printf)
+       if(m_is_printf || m_is_cdp)
                op = ALU_OP;
            else
                op = CALL_OPS;
@@ -1219,6 +1219,8 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
    bool skip = false;
    int op_classification = 0;
    addr_t pc = next_instr();
+   if(pc == 440)
+    pc = 440;
    assert( pc == inst.pc ); // make sure timing model and functional model are in sync
    const ptx_instruction *pI = m_func_info->get_instruction(pc);
    set_npc( pc + pI->inst_size() );
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 915c623..861f0dc 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -1241,6 +1241,9 @@ ptx_instruction::ptx_instruction( int opcode,
        if (fname =="vprintf"){
            m_is_printf = true;
        }
+       if (fname == "cudaGetParameterBufferV2"
+        || fname == "cudaLaunchDeviceV2")
+           m_is_cdp = true;
 
    }
 }
-- 
cgit v1.3