Integration change. Bug fixes from AMD-CMU trace gen branch.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 8389]

5 files changed, 32 insertions, 30 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index e33181f..9555010 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -313,7 +313,7 @@ memory_space_t whichspace( addr_t addr )
 {
    if( (addr >= GLOBAL_HEAP_START) || (addr < STATIC_ALLOC_LIMIT) ) {
       return global_space;
-   } else if( addr > SHARED_GENERIC_START ) {
+   } else if( addr >= SHARED_GENERIC_START ) {
       return shared_space;
    } else {
       return local_space;
@@ -851,6 +851,19 @@ void init_inst_classification_stat()
    g_inst_op_classification_stat[g_ptx_kernel_count] = StatCreate(kernelname,1,100);
 }
 
+static unsigned get_tex_datasize( const ptx_instruction *pI, ptx_thread_info *thread )
+{
+   const operand_info &src1 = pI->src1(); //the name of the texture
+   std::string texname = src1.name();
+
+   gpgpu_t *gpu = thread->get_gpu();
+   const struct textureReference* texref = gpu->get_texref(texname);
+   const struct textureInfo* texInfo = gpu->get_texinfo(texref);
+
+   unsigned data_size = texInfo->texel_size;
+   return data_size; 
+}
+
 void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id )
 {
    bool skip = false;
@@ -960,31 +973,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id )
    if (pI->get_opcode() == TEX_OP) {
       inst.set_addr(lane_id, last_eaddr() );
       assert( inst.space == last_space() );
-
-      unsigned to_type = pI->get_type();
-      switch ( to_type ) {
-      case B8_TYPE:
-      case S8_TYPE:
-      case U8_TYPE: 
-         inst.data_size = 1; break;
-      case B16_TYPE:
-      case S16_TYPE:
-      case U16_TYPE:
-      case F16_TYPE: 
-         inst.data_size = 2; break;
-      case B32_TYPE:
-      case S32_TYPE:
-      case U32_TYPE:
-      case F32_TYPE: 
-         inst.data_size = 4; break;
-      case B64_TYPE:
-      case S64_TYPE:
-      case U64_TYPE:
-      case F64_TYPE: 
-      case FF64_TYPE:
-         inst.data_size = 8; break;
-      default: assert(0); break;
-      }
+      insn_data_size = get_tex_datasize(pI, this); // texture obtain its data granularity from the texture info 
    }
 
    // Output register information to file and stdout
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 05a9b55..1526ef3 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -711,6 +711,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel )
         m_thread[i].m_cta_id = free_cta_hw_id;
         unsigned warp_id = i/m_config->warp_size;
         nthreads_in_block += ptx_sim_init_thread(kernel,&m_thread[i].m_functional_model_thread_state,m_sid,i,cta_size-(i-start_thread),m_config->n_thread_per_shader,this,free_cta_hw_id,warp_id,m_cluster->get_gpu());
+        m_thread[i].m_active = true; 
         warps.set( warp_id );
     }
     assert( nthreads_in_block > 0 && nthreads_in_block <= m_config->n_thread_per_shader); // should be at least one, but less than max
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 2380bcf..2233566 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -135,6 +135,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu,
    for (unsigned i = 0; i<config->n_thread_per_shader; i++) {
       m_thread[i].m_functional_model_thread_state = NULL;
       m_thread[i].m_cta_id = -1;
+      m_thread[i].m_active = false;
    }
    
    m_icnt = new shader_memory_interface(this,cluster);
@@ -519,12 +520,13 @@ void shader_core_ctx::fetch()
                 bool did_exit=false;
                 for( unsigned t=0; t<m_config->warp_size;t++) {
                     unsigned tid=warp_id*m_config->warp_size+t;
-                    if( m_thread[tid].m_functional_model_thread_state ) {
+                    if( m_thread[tid].m_active == true ) {
+                        m_thread[tid].m_active = false; 
                         unsigned cta_id = m_warp[warp_id].get_cta_id();
                         register_cta_thread_exit(cta_id);
                         m_not_completed -= 1;
                         m_active_threads.reset(tid);
-                        m_thread[tid].m_functional_model_thread_state=NULL;
+                        assert( m_thread[tid].m_functional_model_thread_state != NULL );
                         did_exit=true;
                     }
                 }
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 2b834a6..ee0269f 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -120,6 +120,8 @@ public:
    unsigned n_l1_mis_ac;
    unsigned n_l1_mrghit_ac;
    unsigned n_l1_access_ac; 
+
+   bool m_active; 
 };
 
 class shd_warp_t {
diff --git a/src/stream_manager.cc b/src/stream_manager.cc
index 54e2ee7..b9a83d8 100644
--- a/src/stream_manager.cc
+++ b/src/stream_manager.cc
@@ -370,8 +370,16 @@ void stream_manager::push( stream_operation op )
     print_impl(stdout);
     pthread_mutex_unlock(&m_lock);
     if( m_cuda_launch_blocking || stream == NULL ) {
-        while( !empty() )
-            ; 
+        unsigned int wait_amount = 100; 
+        unsigned int wait_cap = 100000; // 100ms 
+        while( !empty() ) {
+            // sleep to prevent CPU hog by empty spin
+            // sleep time increased exponentially ensure fast response when needed 
+            usleep(wait_amount); 
+            wait_amount *= 2; 
+            if (wait_amount > wait_cap) 
+               wait_amount = wait_cap; 
+        }
     }
 }