Several changes to ptxplus towards getting the sphyraena benchmark to work. Add .false modifier to cuobjdump_to_ptxplus. It isn't supported in gpgpusim yet since we don't know what it does. Fixed most of the issues with global (const14) variable not being declared properly. Added "BRX" instruction to cuobjdump_to_ptxplus and some support for it in gpgpusim. There are other issues with it that still need to be worked out. Added support for translating IADD.CARRY sass instruction into ptxplus addp instructions. Fixed a bug with generating ".half" modifiers on instructions so the ptxplus instruction addresses should match sass addresses now. Separated ssy instructions from nop. However, they are still simulated as nop in gpgpusim. Sometime cuobjdump generate a blank link that takes up an address slot. This was changed to a nop.

[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 14001]

4 files changed, 20 insertions, 0 deletions

diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index ab2d9f3..fe2d0ef 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1214,6 +1214,15 @@ void bra_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    thread->set_npc(target_pc);
 }
 
+void brx_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
+{
+   const operand_info &target  = pI->dst();
+   ptx_reg_t target_pc = thread->get_operand_value(target, target, U32_TYPE, thread, 1);
+
+   thread->m_branch_taken = true;
+   thread->set_npc(target_pc);
+}
+
 void break_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
 {
    const operand_info &target  = thread->pop_breakaddr();
@@ -3485,6 +3494,12 @@ void sqrt_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    thread->set_operand_value(dst,d, i_type, thread, pI);
 }
 
+void ssy_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
+{
+   //printf("Execution Warning: unimplemented ssy instruction is treated as a nop\n");
+   // TODO: add implementation
+}
+
 void st_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
 {
    const operand_info &dst = pI->dst();
diff --git a/src/cuda-sim/opcodes.def b/src/cuda-sim/opcodes.def
index 0087d1e..b4912c0 100644
--- a/src/cuda-sim/opcodes.def
+++ b/src/cuda-sim/opcodes.def
@@ -49,6 +49,7 @@ OP_DEF(BFE_OP,bfe_impl,"bfe",1,1)
 OP_DEF(BFI_OP,bfi_impl,"bfi",1,1)
 OP_DEF(BFIND_OP,bfind_impl,"bfind",1,1)
 OP_DEF(BRA_OP,bra_impl,"bra",0,3)
+OP_DEF(BRX_OP,brx_impl,"brx",0,3)
 OP_DEF(BREV_OP,brev_impl,"brev",1,1)
 OP_DEF(BRKPT_OP,brkpt_impl,"brkpt",1,9)
 OP_DEF(CALL_OP,call_impl,"call",1,3)
@@ -101,6 +102,7 @@ OP_DEF(SHR_OP,shr_impl,"shr",1,1)
 OP_DEF(SIN_OP,sin_impl,"sin",1,4)
 OP_DEF(SLCT_OP,slct_impl,"slct",1,1)
 OP_DEF(SQRT_OP,sqrt_impl,"sqrt",1,4)
+OP_DEF(SSY_OP,ssy_impl,"ssy",0,3)
 OP_DEF(ST_OP,st_impl,"st",0,5)
 OP_DEF(SUB_OP,sub_impl,"sub",1,1)
 OP_DEF(SUBC_OP,subc_impl,"subc",1,1)
diff --git a/src/cuda-sim/ptx.l b/src/cuda-sim/ptx.l
index 04100c8..de041fc 100644
--- a/src/cuda-sim/ptx.l
+++ b/src/cuda-sim/ptx.l
@@ -65,6 +65,7 @@ bfe     TC; ptx_lval.int_value = BFE_OP; return OPCODE;
 bfi     TC; ptx_lval.int_value = BFI_OP; return OPCODE;
 bfind   TC; ptx_lval.int_value = BFIND_OP; return OPCODE;
 bra     TC; ptx_lval.int_value = BRA_OP; return OPCODE;
+brx     TC; ptx_lval.int_value = BRX_OP; return OPCODE;
 brev    TC; ptx_lval.int_value = BREV_OP; return OPCODE;
 brkpt   TC; ptx_lval.int_value = BRKPT_OP; return OPCODE;
 call	TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = CALL_OP; return OPCODE; // blocking opcode token in case the callee has the same name as an opcode
@@ -118,6 +119,7 @@ shr     TC; ptx_lval.int_value = SHR_OP; return OPCODE;
 sin	TC; ptx_lval.int_value = SIN_OP; return OPCODE;
 slct	TC; ptx_lval.int_value = SLCT_OP; return OPCODE;
 sqrt	TC; ptx_lval.int_value = SQRT_OP; return OPCODE;
+ssy     TC; ptx_lval.int_value = SSY_OP; return OPCODE;
 st      TC; ptx_lval.int_value = ST_OP; return OPCODE;
 st.volatile TC; ptx_lval.int_value = ST_OP; return OPCODE;
 sub	TC; ptx_lval.int_value = SUB_OP; return OPCODE;
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index 6e39e50..360be51 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -376,6 +376,7 @@ opcode_spec: OPCODE { add_opcode($1); } option_list
 
 pred_spec: PRED IDENTIFIER  { add_pred($2,0, -1); }
 	| PRED EXCLAMATION IDENTIFIER { add_pred($3,1, -1); } 
+	| PRED IDENTIFIER LT_OPTION  { add_pred($2,0,1); }
 	| PRED IDENTIFIER EQ_OPTION  { add_pred($2,0,2); }
 	| PRED IDENTIFIER LE_OPTION  { add_pred($2,0,3); }
 	| PRED IDENTIFIER NE_OPTION  { add_pred($2,0,5); }