- add support for ld/st with generic addresses

- add support for parsing cache operators (still need to pass to timing model)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 6868]

1 files changed, 25 insertions, 0 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index f8a8db0..705a80a 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -294,6 +294,7 @@ ptx_instruction::ptx_instruction( int opcode,
    m_lo = false;
    m_uni = false;
    m_to_option = false;
+   m_cache_option = 0;
    m_rounding_mode = RN_OPTION;
    m_compare_op = -1;
    m_saturation_mode = 0;
@@ -405,6 +406,9 @@ ptx_instruction::ptx_instruction( int opcode,
       case TO_OPTION:
          m_to_option = true;
          break;
+      case CA_OPTION: case CG_OPTION: case CS_OPTION: case LU_OPTION: case CV_OPTION:
+         m_cache_option = last_ptx_inst_option;
+         break;
 
       default:
          assert(0);
@@ -655,6 +659,11 @@ addr_t shared_to_generic( unsigned smid, addr_t addr )
    return SHARED_GENERIC_START + smid*SHARED_MEM_SIZE_MAX + addr;
 }
 
+addr_t global_to_generic( addr_t addr )
+{
+   return addr;
+}
+
 bool isspace_shared( unsigned smid, addr_t addr )
 {
    addr_t start = SHARED_GENERIC_START + smid*SHARED_MEM_SIZE_MAX;
@@ -669,6 +678,17 @@ bool isspace_global( addr_t addr )
    return (addr > GLOBAL_HEAP_START) || (addr < STATIC_ALLOC_LIMIT);
 }
 
+unsigned whichspace( addr_t addr )
+{
+   if( (addr > GLOBAL_HEAP_START) || (addr < STATIC_ALLOC_LIMIT) ) {
+      return GLOBAL_DIRECTIVE;
+   } else if( addr > SHARED_GENERIC_START ) {
+      return SHARED_DIRECTIVE;
+   } else {
+      return LOCAL_DIRECTIVE;
+   }
+}
+
 addr_t generic_to_shared( unsigned smid, addr_t addr )
 {
    assert(isspace_shared(smid,addr));
@@ -696,6 +716,11 @@ addr_t generic_to_local( unsigned smid, unsigned hwtid, addr_t addr )
    return addr - (LOCAL_GENERIC_START + (TOTAL_LOCAL_MEM_PER_SM * smid) + (LOCAL_MEM_SIZE_MAX * hwtid));
 }
 
+addr_t generic_to_global( addr_t addr )
+{
+   return addr;
+}
+
 
 unsigned long long g_dev_malloc=GLOBAL_HEAP_START;