Fixing a bug exposed by the fix for bug 42.

   The "_" "null" register potentially generated by ptx and intentionally generated by ptxplus was being initialized without a
   type.  This caused the parser to think it was not a register.
   Fix is to allow the parser to think of it as register, but ensure the arch-sim does not by adding a flag indicating
   that it is special.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 15305]

3 files changed, 52 insertions, 6 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 1dbff50..cba5262 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -844,7 +844,8 @@ void ptx_instruction::pre_decode()
    for ( ; opr != op_iter_end(); opr++, n++ ) { //process operands
       const operand_info &o = *opr;
       if ( has_dst && n==0 ) {
-         if ( o.is_reg() ) {
+         // Do not set the null register "_" as an architectural register
+         if ( o.is_reg() && !o.is_non_arch_reg() ) {
             out[0] = o.reg_num();
             arch_reg.dst[0] = o.arch_reg_num();
          } else if ( o.is_vector() ) {
@@ -858,7 +859,7 @@ void ptx_instruction::pre_decode()
                arch_reg.dst[i] = o.arch_reg_num(i);
          }
       } else {
-         if ( o.is_reg() ) {
+         if ( o.is_reg() && !o.is_non_arch_reg() ) {
             int reg_num = o.reg_num();
             arch_reg.src[m] = o.arch_reg_num();
             switch ( m ) {
@@ -898,6 +899,8 @@ void ptx_instruction::pre_decode()
          const operand_info &o = *op;
 
          if(o.is_memory_operand()) {
+             // We do not support the null register as a memory operand
+             assert( !o.is_non_arch_reg() );
 
             // Check PTXPlus-type operand
             // memory operand with addressing (ex. s[0x4] or g[$r1])
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 0dffe70..678febf 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -191,8 +191,23 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
    } else {
       assert( !prior_decl );
       *sym_table = new symbol_table( "", entry_point, this );
-      symbol *null_reg = (*sym_table)->add_variable("_",NULL,0,"",0); 
+      
+      // Initial setup code to support a register represented as "_".
+      // This register is used when an instruction operand is
+      // not read or written.  However, the parser must recognize it
+      // as a legitimate register but we do not want to pass
+      // it to the micro-architectural register to the performance simulator.
+      // For this purpose we add a symbol to the symbol table but
+      // mark it as a non_arch_reg so it does not effect the performance sim.
+      type_info_key null_key( reg_space, 0, 0, 0, 0, 0 );
+      null_key.set_is_non_arch_reg();
+      // First param is null - which is bad.
+      // However, the first parameter is actually unread in the constructor...
+      // TODO - remove the symbol_table* from type_info
+      type_info* null_type_info = new type_info( NULL, null_key );
+      symbol *null_reg = (*sym_table)->add_variable( "_", null_type_info, 0, "", 0 ); 
       null_reg->set_regno(0, 0);
+      
       (*sym_table)->set_name(name);
       (*func_info)->set_symtab(*sym_table);
       m_function_symtab_lookup[key] = *sym_table;
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 9a24e47..26469d5 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -47,10 +47,12 @@ class type_info_key {
 public:
    type_info_key()
    {
+      m_is_non_arch_reg = false;
       m_init = false;
    }
    type_info_key( memory_space_t space_spec, int scalar_type_spec, int vector_spec, int alignment_spec, int extern_spec, int array_dim )
    {
+      m_is_non_arch_reg = false;
       m_init = true;
       m_space_spec = space_spec; 
       m_scalar_type_spec = scalar_type_spec;
@@ -75,7 +77,9 @@ public:
 
    void set_array_dim( int array_dim ) { m_array_dim = array_dim; }
    int get_array_dim() const { assert(m_init); return m_array_dim; }
+   void set_is_non_arch_reg() { m_is_non_arch_reg = true;  }
 
+   bool is_non_arch_reg() const { return m_is_non_arch_reg; }
    bool is_reg() const { return m_space_spec == reg_space;} 
    bool is_param_kernel() const { return m_space_spec == param_space_kernel;}
    bool is_param_local() const { return m_space_spec == param_space_local; }
@@ -99,6 +103,7 @@ private:
    int m_extern_spec;
    int m_array_dim;
    int m_is_function;
+   bool m_is_non_arch_reg;
 
    friend class type_info_key_compare;
 };
@@ -219,11 +224,20 @@ public:
    bool is_param_local() const { return m_is_param_local; }
    bool is_tex() const { return m_is_tex;}
    bool is_func_addr() const { return m_is_func_addr; }
-   bool is_reg() const {
-      if( m_type == NULL ) 
-         return false;
+   bool is_reg() const
+   {
+       if ( m_type == NULL ) {
+           return false;
+       }
       return m_type->get_key().is_reg(); 
    }
+   bool is_non_arch_reg() const
+   {
+       if ( m_type == NULL ) {
+           return false;
+       }
+      return m_type->get_key().is_non_arch_reg(); 
+   }
 
    void add_initializer( const std::list<operand_info> &init );
    bool has_initializer() const 
@@ -335,6 +349,7 @@ class operand_info {
 public:
    operand_info()
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -346,6 +361,7 @@ public:
    }
    operand_info( const symbol *addr )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -374,6 +390,8 @@ public:
       } else {
          m_type = reg_t;
       }
+      
+      m_is_non_arch_reg = addr->is_non_arch_reg();
       m_value.m_symbolic = addr;
       m_addr_offset = 0;
       m_vector = false;
@@ -383,6 +401,7 @@ public:
    }
    operand_info( const symbol *addr1, const symbol *addr2 )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -404,6 +423,7 @@ public:
    }
    operand_info( int builtin_id, int dim_mod )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -421,6 +441,7 @@ public:
    }
    operand_info( const symbol *addr, int offset )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -438,6 +459,7 @@ public:
    }
    operand_info( unsigned x )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -455,6 +477,7 @@ public:
    }
    operand_info( int x )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -472,6 +495,7 @@ public:
    }
    operand_info( float x )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -489,6 +513,7 @@ public:
    }
    operand_info( double x )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -506,6 +531,7 @@ public:
    }
    operand_info( const symbol *s1, const symbol *s2, const symbol *s3, const symbol *s4 )
    {
+      m_is_non_arch_reg = false;
       m_addr_space = undefined_space;
       m_operand_lohi = 0;
       m_double_operand_type = 0;
@@ -688,6 +714,7 @@ public:
    bool get_operand_neg() const { return m_operand_neg; }
    void set_const_mem_offset(addr_t set_value) { m_const_mem_offset = set_value; }
    addr_t get_const_mem_offset() const { return m_const_mem_offset; }
+   bool is_non_arch_reg() const { return m_is_non_arch_reg; }
 
 private:
    unsigned m_uid;
@@ -717,6 +744,7 @@ private:
 
    bool m_neg_pred;
    bool m_is_return_var;
+   bool m_is_non_arch_reg;
 
    static unsigned sm_next_uid;
    unsigned get_uid();