- adding placeholders for new builtin registers

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 6848]

4 files changed, 86 insertions, 26 deletions

diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index f3aa24a..5decd2b 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -2977,7 +2977,7 @@ void xor_impl( const ptx_instruction *pI, ptx_thread_info *thread )
 
 void inst_not_implemented( const ptx_instruction * pI ) 
 {
-   printf("Execution error (%s:%u): instruction \"%s\" not (yet) implemented\n",
+   printf("GPGPU-Sim PTX: ERROR (%s:%u) instruction \"%s\" not (yet) implemented\n",
           pI->source_file(), 
           pI->source_line(), 
           pI->get_opcode_cstr() );
diff --git a/src/cuda-sim/opcodes.h b/src/cuda-sim/opcodes.h
index 221bf2e..1a06415 100644
--- a/src/cuda-sim/opcodes.h
+++ b/src/cuda-sim/opcodes.h
@@ -71,12 +71,26 @@ enum opcode_t {
 };
 
 enum special_regs {
-   CLOCK_ID,
-   CTA_ID,
-   NTID_ID,
-   GRIDID_ID,
-   NCTAID_ID,
-   TID_ID
+   CLOCK_REG,
+   CLOCK64_REG,
+   CTAID_REG,
+   ENVREG_REG,
+   GRIDID_REG,
+   LANEID_REG,
+   LANEMASK_EQ_REG,
+   LANEMASK_LE_REG,
+   LANEMASK_LT_REG,
+   LANEMASK_GE_REG,
+   LANEMASK_GT_REG,
+   NCTAID_REG,
+   NTID_REG,
+   NSMID_REG,
+   NWARPID_REG,
+   PM_REG,
+   SMID_REG,
+   TID_REG,
+   WARPID_REG,
+   WARPSZ_REG
 };
 
 #endif
diff --git a/src/cuda-sim/ptx.l b/src/cuda-sim/ptx.l
index e4c9fef..b7dd56f 100644
--- a/src/cuda-sim/ptx.l
+++ b/src/cuda-sim/ptx.l
@@ -71,6 +71,11 @@
 char linebuf[1024];
 unsigned col = 0;
 #define TC col+=strlen(ptx_text); 
+#define CHECK_UNSIGNED \
+	if( yytext[strlen(yytext)-1]=='U' ) { \
+		printf("GPGPU-Sim: ERROR ** U modifier not implemented\n"); \
+		abort(); \
+	}
 int ptx_error( const char *s );
 %}
 
@@ -165,7 +170,7 @@ xor     TC; ptx_lval.int_value = XOR_OP; return OPCODE;
 \.byte	TC; return BYTE_DIRECTIVE; /* not in PTX 2.1 */
 \.callprototype TC; return CALLPROTOTYPE_DIRECTIVE;
 \.calltargets TC; return CALLTARGETS_DIRECTIVE;
-\.const\[[0-9]+\] TC; return CONST_DIRECTIVE;
+\.const\[[0-9]+\] TC; /*fixme: should use const offset value*/ abort(); return CONST_DIRECTIVE;
 \.const TC; return CONST_DIRECTIVE;
 \.entry TC; return ENTRY_DIRECTIVE;
 \.extern TC; return EXTERN_DIRECTIVE;
@@ -194,18 +199,37 @@ xor     TC; ptx_lval.int_value = XOR_OP; return OPCODE;
 \.visible TC; return VISIBLE_DIRECTIVE;
 
 
-"%clock" TC; ptx_lval.int_value = CLOCK_ID; return SPECIAL_REGISTER;
-"%ctaid" TC; ptx_lval.int_value = CTA_ID; return SPECIAL_REGISTER;
-"%ntid"  TC; ptx_lval.int_value = NTID_ID; return SPECIAL_REGISTER;
-"%gridid" TC; ptx_lval.int_value = GRIDID_ID; return SPECIAL_REGISTER;
-"%nctaid" TC; ptx_lval.int_value = NCTAID_ID; return SPECIAL_REGISTER;
-"%tid"  TC; ptx_lval.int_value = TID_ID; return SPECIAL_REGISTER;
+"%clock" TC; ptx_lval.int_value = CLOCK_REG; return SPECIAL_REGISTER;
+"%clock64" TC; ptx_lval.int_value = CLOCK64_REG; return SPECIAL_REGISTER;
+"%ctaid" TC; ptx_lval.int_value = CTAID_REG; return SPECIAL_REGISTER;
+"%envreg"[0-9]+ TC; sscanf(yytext+7,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += ENVREG_REG; return SPECIAL_REGISTER;
+"%gridid" TC; ptx_lval.int_value = GRIDID_REG; return SPECIAL_REGISTER;
+"%laneid"  TC; ptx_lval.int_value = LANEID_REG; return SPECIAL_REGISTER;
+"%lanemask_eq"  TC; ptx_lval.int_value = LANEMASK_EQ_REG; return SPECIAL_REGISTER;
+"%lanemask_le"  TC; ptx_lval.int_value = LANEMASK_LE_REG; return SPECIAL_REGISTER;
+"%lanemask_lt"  TC; ptx_lval.int_value = LANEMASK_LT_REG; return SPECIAL_REGISTER;
+"%lanemask_ge"  TC; ptx_lval.int_value = LANEMASK_GE_REG; return SPECIAL_REGISTER;
+"%lanemask_gt"  TC; ptx_lval.int_value = LANEMASK_GT_REG; return SPECIAL_REGISTER;
+"%nctaid" TC; ptx_lval.int_value = NCTAID_REG; return SPECIAL_REGISTER;
+"%ntid"  TC; ptx_lval.int_value = NTID_REG; return SPECIAL_REGISTER;
+"%nsmid"  TC; ptx_lval.int_value = NSMID_REG; return SPECIAL_REGISTER;
+"%nwarpid"  TC; ptx_lval.int_value = NWARPID_REG; return SPECIAL_REGISTER;
+"%pm"[0-3]  TC; sscanf(yytext+3,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += PM_REG; return SPECIAL_REGISTER;
+"%smid"  TC; ptx_lval.int_value = SMID_REG; return SPECIAL_REGISTER;
+"%tid"  TC; ptx_lval.int_value = TID_REG; return SPECIAL_REGISTER;
+"%warpid"  TC; ptx_lval.int_value = WARPID_REG; return SPECIAL_REGISTER;
+"WARP_SZ"  TC; ptx_lval.int_value = WARPSZ_REG; return SPECIAL_REGISTER;
 
-[_A-Za-z$%][_0-9A-Za-z$]*  TC; ptx_lval.string_value = strdup(yytext); /*printf("\n<<identifier=\"%s\">>\n", yytext); fflush(stdout);*/ return IDENTIFIER;
+[a-zA-Z][a-zA-Z0-9_$]*  TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER;
+[_$%][a-zA-Z0-9_$]+  TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER;
 
 [0-9]+\.[0-9]+ 	 TC; sscanf(yytext,"%lf", &ptx_lval.double_value); return DOUBLE_OPERAND;
-[-]{0,1}[0-9]+		 TC; ptx_lval.int_value =  atoi(yytext); return INT_OPERAND;
-0[xX][0-9]+	 TC;  sscanf(yytext,"%x", &ptx_lval.int_value); return INT_OPERAND;
+	
+0[xX][0-9]+U?   TC; CHECK_UNSIGNED; sscanf(yytext,"%x", &ptx_lval.int_value); return INT_OPERAND;
+0[0-7]+U?   	TC; printf("GPGPU-Sim: ERROR ** parsing octal not (yet) implemented\n"); abort(); return INT_OPERAND;
+0[bB][01]+U?  	TC; printf("GPGPU-Sim: ERROR ** parsing binary not (yet) implemented\n"); abort(); return INT_OPERAND;
+[-]?[0-9]+U?    TC; CHECK_UNSIGNED; ptx_lval.int_value =  atoi(yytext); return INT_OPERAND;
+
 0[fF][0-9a-fA-F]{8}  TC; sscanf(yytext+2,"%x", (unsigned*)(void*)&ptx_lval.float_value); return FLOAT_OPERAND;
 0[dD][0-9a-fA-F]{16}  TC; sscanf(yytext+2,"%Lx", (unsigned long long*)(void*)&ptx_lval.double_value); return DOUBLE_OPERAND;
 
diff --git a/src/cuda-sim/ptx_sim.cc b/src/cuda-sim/ptx_sim.cc
index 8ec60df..a1a77fd 100644
--- a/src/cuda-sim/ptx_sim.cc
+++ b/src/cuda-sim/ptx_sim.cc
@@ -65,6 +65,8 @@
 #include <string>
 #include "ptx_ir.h"
 
+void feature_not_implemented( const char *f );
+
 std::set<unsigned long long> g_ptx_cta_info_sm_idx_used;
 unsigned long long g_ptx_cta_info_uid = 1;
 extern int gpgpu_option_spread_blocks_across_cores;
@@ -256,23 +258,37 @@ extern signed long long gpu_tot_sim_cycle;
 unsigned ptx_thread_info::get_builtin( int builtin_id, unsigned dim_mod ) 
 {
    assert( m_valid );
-   switch (builtin_id) {
-   case NTID_ID:
-      assert( dim_mod < 3 );
-      return m_ntid[dim_mod];
-   case CLOCK_ID:
+   switch ((builtin_id&0xFFFF)) {
+   case CLOCK_REG:
+      return (unsigned)(gpu_sim_cycle + gpu_tot_sim_cycle);
+   case CLOCK64_REG:
+      abort(); // change return value to unsigned long long?
       return gpu_sim_cycle + gpu_tot_sim_cycle;
-   case CTA_ID:
+   case CTAID_REG:
       assert( dim_mod < 3 );
       return m_ctaid[dim_mod];
-   case GRIDID_ID:
+   case ENVREG_REG: feature_not_implemented( "%envreg" ); return 0;
+   case GRIDID_REG:
       return m_gridid;
-   case NCTAID_ID:
+   case LANEID_REG: feature_not_implemented( "%laneid" ); return 0;
+   case LANEMASK_EQ_REG: feature_not_implemented( "%lanemask_eq" ); return 0;
+   case LANEMASK_LE_REG: feature_not_implemented( "%lanemask_le" ); return 0;
+   case LANEMASK_LT_REG: feature_not_implemented( "%lanemask_lt" ); return 0;
+   case LANEMASK_GE_REG: feature_not_implemented( "%lanemask_ge" ); return 0;
+   case LANEMASK_GT_REG: feature_not_implemented( "%lanemask_gt" ); return 0;
+   case NCTAID_REG:
       assert( dim_mod < 3 );
       return m_nctaid[dim_mod];
-   case TID_ID:
+   case NTID_REG:
+      assert( dim_mod < 3 );
+      return m_ntid[dim_mod];
+   case NWARPID_REG: feature_not_implemented( "%nwarpid" ); return 0;
+   case PM_REG: feature_not_implemented( "%pm" ); return 0;
+   case SMID_REG: feature_not_implemented( "%smid" ); return 0;
+   case TID_REG:
       assert( dim_mod < 3 );
       return m_tid[dim_mod];
+   case WARPSZ_REG: feature_not_implemented( "WARP_SZ" ); return 0;
    default:
       assert(0);
    }
@@ -430,3 +446,9 @@ void ptx_thread_info::set_npc( const function_info *f )
    m_func_info = const_cast<function_info*>( f );
    m_symbol_table = m_func_info->get_symtab();
 }
+
+void feature_not_implemented( const char *f ) 
+{
+   printf("GPGPU-Sim: feature '%s' not supported\n", f );
+   abort();
+}