add listing function for ptx debugging

print out compilation errors encountered during OpenCL to PTX conversion (still a bit cryptic)

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 7162]

4 files changed, 68 insertions, 13 deletions

diff --git a/libopencl/nvopencl_wrapper.cc b/libopencl/nvopencl_wrapper.cc
index 7c12075..6d446a4 100644
--- a/libopencl/nvopencl_wrapper.cc
+++ b/libopencl/nvopencl_wrapper.cc
@@ -65,10 +65,12 @@
 #include <string.h>
 #include <stdarg.h>
 
+#define PREAMBLE "GPGPU-Sim nvopencl_wrapper"
+
 void vmyexit(int code, const char *str,va_list ap)
 {
    char buffer[1024];
-   snprintf(buffer,1024,"GPGPU-Sim API: nvopencl_wrapper ERROR ** %s\n", str);
+   snprintf(buffer,1024,"%s: ERROR ** %s\n", PREAMBLE, str);
    vprintf(buffer,ap);
    fflush(stdout);
    if( code ) 
@@ -123,9 +125,12 @@ int main(int argc, const char **argv)
    if ( errcode != CL_SUCCESS ) myexit(3,"clGetPlatformIDs returned %d",errcode);
    errcode = clGetPlatformInfo(platforms[0], CL_PLATFORM_NAME, 1024, &buffer, NULL);
    if ( errcode != CL_SUCCESS ) myexit(3,"clGetPlatformInfo returned %d",errcode);
-   printf("GPGPU-Sim OpenCL API: Generating PTX using OpenCL platform \'%s\'\n",buffer);
+   printf("%s: Generating PTX using OpenCL platform \'%s\'\n",PREAMBLE,buffer);
+
    errcode = clGetDeviceIDs(platforms[0], CL_DEVICE_TYPE_GPU, 0, NULL, &num_devices);
    if ( errcode != CL_SUCCESS ) myexit(4,"clGetDeviceIDs returned %d",errcode);
+   printf("%s: found %u native OpenCL devices\n",PREAMBLE,num_devices);
+
    cl_device_id *devices = (cl_device_id *)malloc(num_devices * sizeof(cl_device_id) );
    errcode = clGetDeviceIDs(platforms[0], CL_DEVICE_TYPE_GPU, num_devices, devices, NULL);
    if ( errcode != CL_SUCCESS ) myexit(5,"clGetDeviceIDs returned %d",errcode);
@@ -143,7 +148,19 @@ int main(int argc, const char **argv)
       n+= 2;
    }
    errcode = clBuildProgram(pgm, 0, NULL, options, NULL, NULL);
-   if ( errcode != CL_SUCCESS ) myexit(8,"clBuildProgram returned %d",errcode);
+   if ( errcode != CL_SUCCESS ) {
+      printf("%s: clBuildProgram returned %d (error) -- build log:\n\n",PREAMBLE,errcode);
+      size_t build_log_length=0;
+      errcode = clGetProgramBuildInfo(pgm,devices[0],CL_PROGRAM_BUILD_LOG,0,NULL,&build_log_length);
+      if( errcode != CL_SUCCESS ) myexit(8,"clGetProgramBuildInfo returned %d",errcode);
+      char *build_log = (char*)calloc(1,build_log_length);
+      errcode = clGetProgramBuildInfo(pgm,devices[0],CL_PROGRAM_BUILD_LOG,build_log_length,
+                                      build_log,&build_log_length);
+      printf("%s",build_log);
+      printf("\n\n%s: end of build log\n", PREAMBLE);
+      printf("%s: exiting early because the OpenCL code had errors (see above).\n", PREAMBLE);
+      exit(8);
+   }
 
    size_t nbytes1=0;
    errcode = clGetProgramInfo(pgm,CL_PROGRAM_NUM_DEVICES,sizeof(cl_uint),&num_devices,&nbytes1);
diff --git a/src/cuda-sim/ptx_sim.cc b/src/cuda-sim/ptx_sim.cc
index d7fa510..9a2acf6 100644
--- a/src/cuda-sim/ptx_sim.cc
+++ b/src/cuda-sim/ptx_sim.cc
@@ -434,6 +434,16 @@ std::string ptx_thread_info::get_location() const
    return std::string(buf);
 }
 
+const ptx_instruction *ptx_thread_info::get_inst() const
+{
+   return m_func_info->get_instruction(m_PC);
+}
+
+const ptx_instruction *ptx_thread_info::get_inst( addr_t pc ) const
+{
+   return m_func_info->get_instruction(pc);
+}
+
 void ptx_thread_info::dump_regs()
 {
    printf("Register File Contents:\n");
diff --git a/src/cuda-sim/ptx_sim.h b/src/cuda-sim/ptx_sim.h
index 8de4127..c2bb533 100644
--- a/src/cuda-sim/ptx_sim.h
+++ b/src/cuda-sim/ptx_sim.h
@@ -400,6 +400,8 @@ public:
    bool callstack_pop();
    void dump_callstack() const;
    std::string get_location() const;
+   const ptx_instruction *get_inst() const;
+   const ptx_instruction *get_inst( addr_t pc ) const;
    bool rpc_updated() const { return m_RPC_updated; }
    bool last_was_call() const { return m_last_was_call; }
    unsigned get_rpc() const { return m_RPC; }
diff --git a/src/debug.cc b/src/debug.cc
index 3e71209..0691d78 100644
--- a/src/debug.cc
+++ b/src/debug.cc
@@ -51,6 +51,8 @@ void gpgpu_debug()
       done=false;
 
    /// check if we've reached a breakpoint
+   const ptx_thread_info *brk_thd = NULL;
+   const ptx_instruction *brk_inst = NULL;
 
    for( std::map<unsigned,brk_pt>::iterator i=breakpoints.begin(); i!=breakpoints.end(); i++) {
       unsigned num=i->first;
@@ -59,19 +61,19 @@ void gpgpu_debug()
          unsigned addr = b.get_addr();
          unsigned new_value = read_location(addr);
          if( new_value != b.get_value() ) {
-            printf( "GPGPU-Sim DBG: watch point %u triggered (old value=%x, new value=%x)\n",
+            printf( "GPGPU-Sim PTX DBG: watch point %u triggered (old value=%x, new value=%x)\n",
                      num,b.get_value(),new_value );
             std::map<unsigned,watchpoint_event>::iterator w=g_watchpoint_hits.find(num);
             if( w==g_watchpoint_hits.end() ) 
-               printf( "GPGPU-Sim DBG: memory transfer modified value\n");
+               printf( "GPGPU-Sim PTX DBG: memory transfer modified value\n");
             else {
                watchpoint_event wa = w->second;
-               const ptx_thread_info *thd = wa.thread();
-               const ptx_instruction *pI = wa.inst();
-               printf( "GPGPU-Sim DBG: modified by thread uid=%u, sid=%u, hwtid=%u\n",
-                       thd->get_uid(),thd->get_hw_sid(), thd->get_hw_tid() );
-               printf( "GPGPU-Sim DBG: ");
-               pI->print_insn(stdout);
+               brk_thd = wa.thread();
+               brk_inst = wa.inst();
+               printf( "GPGPU-Sim PTX DBG: modified by thread uid=%u, sid=%u, hwtid=%u\n",
+                       brk_thd->get_uid(),brk_thd->get_hw_sid(), brk_thd->get_hw_tid() );
+               printf( "GPGPU-Sim PTX DBG: ");
+               brk_inst->print_insn(stdout);
                printf( "\n" );
                g_watchpoint_hits.erase(w);
             }
@@ -84,8 +86,15 @@ void gpgpu_debug()
             if( !fvi ) continue;
             if( thread_at_brkpt(fvi->ptx_thd_info, b) ) {
                done = false;
-               printf("GPGPU-Sim DBG: reached breakpoint %u at %s (sm=%u, hwtid=%u)\n", 
+               printf("GPGPU-Sim PTX DBG: reached breakpoint %u at %s (sm=%u, hwtid=%u)\n", 
                       num, b.location().c_str(), sid, fvi->hw_thread_id );
+               brk_thd = (ptx_thread_info*)fvi->ptx_thd_info;
+               brk_inst = brk_thd->get_inst();
+               printf( "GPGPU-Sim PTX DBG: reached by thread uid=%u, sid=%u, hwtid=%u\n",
+                       brk_thd->get_uid(),brk_thd->get_hw_sid(), brk_thd->get_hw_tid() );
+               printf( "GPGPU-Sim PTX DBG: ");
+               brk_inst->print_insn(stdout);
+               printf( "\n" );
             }
          }
       }
@@ -97,7 +106,7 @@ void gpgpu_debug()
    /// enter interactive debugger loop
 
    while (!done) {
-      printf("(gpgpu-sim dbg) ");
+      printf("(ptx debugger) ");
       fflush(stdout);
       
       char line[1024];
@@ -145,6 +154,22 @@ void gpgpu_debug()
          unsigned value = read_location(addr);
          g_global_mem->set_watch(addr,next_brkpt); 
          breakpoints[next_brkpt++] = brk_pt(addr,value);
+      } else if( !strcmp(tok,"l") ) {
+         if( brk_thd == NULL  ) 
+            printf("no thread selected");
+         addr_t pc = brk_thd->get_pc();
+         addr_t start_pc = (pc<5)?0:(pc-5);
+         for( addr_t p=start_pc;  p <= pc+5; p++ ) {
+            const ptx_instruction *i = brk_thd->get_inst(p);
+            if( i ) {
+               if( p != pc )
+                  printf( "    " );
+               else
+                  printf( "==> " );
+                   i->print_insn(stdout);
+               printf( "\n" );
+            }
+         }
       } else if( !strcmp(tok,"h") ) {
          printf("commands:\n");
          printf("  q                           - quit GPGPU-Sim\n");
@@ -153,6 +178,7 @@ void gpgpu_debug()
          printf("  del <n>                     - delete breakpoint\n");
          printf("  s                           - single step one shader cycle (all cores)\n");
          printf("  c                           - continue simulation without single stepping\n");
+         printf("  l                           - list PTX around current breakpoint\n");
          printf("  dp <n>                      - display pipeline contents on SM <n>\n");
          printf("  h                           - print this message\n");
       } else {