Add more print statements to debug api calls

Fixed bug with 32bit rem
Fixed bug with launching kernels in ptxjitplus
Removed extraneous files
Cleaned up more pointers
Added documentation for ptxjitplus

2 files changed, 18 insertions, 2 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index df8d806..d2f096f 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1309,7 +1309,9 @@ void function_info::ptx_jit_config(std::map<unsigned long long, size_t> mallocPt
     do{
         fprintf(fout, "%s", buff);
         fgets(buff, 1024, fin);
-        assert(!feof(fin));
+        if(feof(fin)){
+            break;
+        }
     } while(strstr(buff, "entry")==NULL);
 
     fclose(fin);
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 034a7b9..9e18772 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -3190,7 +3190,21 @@ void rem_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    src1_data = thread->get_operand_value(src1, dst, i_type, thread, 1);
    src2_data = thread->get_operand_value(src2, dst, i_type, thread, 1);
 
-   data.u64 = src1_data.u64 % src2_data.u64;
+   switch ( i_type ) {
+   case S32_TYPE:
+      data.s32 = src1_data.s32 % src2_data.s32;
+      break;
+   case S64_TYPE:
+      data.s64 = src1_data.s64 % src2_data.s64;
+      break;
+   case U32_TYPE:
+      data.u32 = src1_data.u32 % src2_data.u32;
+      break;
+   case U64_TYPE:
+      data.u64 = src1_data.u64 % src2_data.u64;
+      break;
+   default: assert(0); break;
+   }
 
    thread->set_operand_value(dst,data, i_type, thread, pI);
 }