Integrate mov.pred fix from TM branch.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 15538]

2 files changed, 10 insertions, 0 deletions

diff --git a/CHANGES b/CHANGES
index 700fb1d..5630583 100644
--- a/CHANGES
+++ b/CHANGES
@@ -67,6 +67,8 @@ Version 3.2.0+edits (development branch) versus 3.2.0
       GPGPUSIM_ROOT was not the first entry in the list;.
     - Fixed a GPUWattch bug where the performance counters were not being correctly
       reset when a kernel completes. 
+    - Fixed implementation for mov.pred with literal input (need to explicitly
+      treat zero as false and non-zero as true). 
 
 Version 3.2.0 versus 3.1.2
 - Added GPUWattch GPGPU power model based on McPAT 0.8beta. 
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 49b27d2..89212f2 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -2574,6 +2574,14 @@ void mov_impl( const ptx_instruction *pI, ptx_thread_info *thread )
       } else {
          thread->set_operand_value(dst,tmp_bits, i_type, thread, pI);
       }
+   } else if (i_type == PRED_TYPE and src1.is_literal() == true) {
+      // in ptx, literal input translate to predicate as 0 = false and 1 = true 
+      // we have adopted the opposite to simplify implementation of zero flags in ptxplus 
+      data = thread->get_operand_value(src1, dst, i_type, thread, 1);
+
+      ptx_reg_t finaldata; 
+      finaldata.pred = (data.u32 == 0)? 1 : 0;  // setting zero-flag in predicate 
+      thread->set_operand_value(dst, finaldata, i_type, thread, pI);
    } else {
 
       data = thread->get_operand_value(src1, dst, i_type, thread, 1);