- print out instructions at branch divergence/reconvergence points (more informative)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 13399]

4 files changed, 18 insertions, 4 deletions

diff --git a/CHANGES b/CHANGES
index 2c3deb6..2da61e0 100644
--- a/CHANGES
+++ b/CHANGES
@@ -10,6 +10,7 @@ Version 3.1.0+edits (development branch) versus 3.1.0
 - Added TeslaC2050 configuration.
 - Added a two level warp scheduler (as presented at ISCA 2012 tutorial).
 - Cleaner makedepend that doesn't interfere with Makefiles
+- Print out instructions at branch divergence/reconvergence points
 - Bug Fixes:
     - Fixed a bug in executing call_imp and callp_imp with the pure functional simulation mode.
     - Fixed a couple of memory errors in cuobjdump_to_ptxplus code.
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index e08e521..252b230 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -267,12 +267,12 @@ void function_info::ptx_assemble()
       print_ipostdominators();
    }
 
-   printf("GPGPU-Sim PTX: pre-decoding instructions for \'%s\'... ", m_name.c_str() );
+   printf("GPGPU-Sim PTX: pre-decoding instructions for \'%s\'...\n", m_name.c_str() );
    for ( unsigned ii=0; ii < n; ii += m_instr_mem[ii]->inst_size() ) { // handle branch instructions
       ptx_instruction *pI = m_instr_mem[ii];
       pI->pre_decode();
    }
-   printf("  done.\n");
+   printf("GPGPU-Sim PTX: ... done pre-decoding instructions for \'%s\'.\n", m_name.c_str() );
    fflush(stdout);
 
    m_assembled = true;
@@ -1785,8 +1785,16 @@ struct rec_pts find_reconvergence_points( function_info *finfo )
       gpgpu_recon_t *kernel_recon_points = (struct gpgpu_recon_t*) calloc(num_recon, sizeof(struct gpgpu_recon_t));
       finfo->get_reconvergence_pairs(kernel_recon_points);
       printf("GPGPU-Sim PTX: Reconvergence Pairs for %s\n", finfo->get_name().c_str() );
-      for (int i=0;i<num_recon;i++) 
-         printf("GPGPU-Sim PTX:   branch pc = %d\ttarget pc = %d\n", kernel_recon_points[i].source_pc, kernel_recon_points[i].target_pc); 
+      for (int i=0;i<num_recon;i++) {
+         printf("GPGPU-Sim PTX:   branch pc = %4d : ", kernel_recon_points[i].source_pc );
+         kernel_recon_points[i].source_inst->print_insn();
+         printf("\n");
+         printf("GPGPU-Sim PTX:   target pc = %4d : ", kernel_recon_points[i].target_pc ); 
+         if( kernel_recon_points[i].target_inst )
+            kernel_recon_points[i].target_inst->print_insn();
+         printf("\n");
+      }
+
       tmp.s_kernel_recon_points = kernel_recon_points;
       tmp.s_num_recon = num_recon;
       g_rpts[finfo] = tmp;
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 50669b6..102ae40 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -783,14 +783,17 @@ void function_info::get_reconvergence_pairs(gpgpu_recon_t* recon_points)
          fflush(stdout);
 #endif
          recon_points[idx].source_pc = m_basic_blocks[i]->ptx_end->get_PC();
+         recon_points[idx].source_inst = m_basic_blocks[i]->ptx_end;
 #ifdef DEBUG_GET_RECONVERG_PAIRS
          printf("\trecon_points[idx].source_pc=%d\n", recon_points[idx].source_pc);
 #endif
          if( m_basic_blocks[m_basic_blocks[i]->immediatepostdominator_id]->ptx_begin ) {
             recon_points[idx].target_pc = m_basic_blocks[m_basic_blocks[i]->immediatepostdominator_id]->ptx_begin->get_PC();
+            recon_points[idx].target_inst = m_basic_blocks[m_basic_blocks[i]->immediatepostdominator_id]->ptx_begin;
          } else {
             // reconverge after function return
             recon_points[idx].target_pc = -2;
+            recon_points[idx].target_inst = NULL;
          }
 #ifdef DEBUG_GET_RECONVERG_PAIRS
          m_basic_blocks[m_basic_blocks[i]->immediatepostdominator_id]->ptx_begin->print_insn();
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 1c945d7..737a610 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -747,6 +747,8 @@ struct basic_block_t {
 struct gpgpu_recon_t {
    address_type source_pc;
    address_type target_pc;
+   class ptx_instruction* source_inst;
+   class ptx_instruction* target_inst;
 };
 
 class ptx_instruction : public warp_inst_t {