From 95121fefda110e5984e36ae630186d0c73887202 Mon Sep 17 00:00:00 2001
From: J <jonathanslew@outlook.com>
Date: Tue, 7 Aug 2018 17:02:39 -0700
Subject: fix for deadlock that should serialize bank accesses

---
 src/gpgpu-sim/shader.cc | 27 +++++++++++----------------
 1 file changed, 11 insertions(+), 16 deletions(-)

(limited to 'src/gpgpu-sim/shader.cc')

diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index da85bae..0a5e11f 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -3025,25 +3025,20 @@ int register_bank(int regnum, int wid, unsigned num_banks, unsigned bank_warp_sh
    return bank % num_banks;
 }
 
-bool opndcoll_rfu_t::writeback( const warp_inst_t &inst )
+bool opndcoll_rfu_t::writeback( warp_inst_t &inst )
 {
    assert( !inst.empty() );
    std::list<unsigned> regs = m_shader->get_regs_written(inst);
-   std::list<unsigned>::iterator r;
-   unsigned n=0;
-   for( r=regs.begin(); r!=regs.end();r++,n++ ) {
-      unsigned reg = *r;
-      unsigned bank = register_bank(reg,inst.warp_id(),m_num_banks,m_bank_warp_shift);
-      unsigned count = 0;
-      while( !m_arbiter.bank_idle(bank) ) {
-          assert((++count)<m_num_banks);
-          bank++;
-          bank %= m_num_banks;
-      }
-      if( m_arbiter.bank_idle(bank) ) {
-          m_arbiter.allocate_bank_for_write(bank,op_t(&inst,reg,m_num_banks,m_bank_warp_shift));
-      } else {
-          return false;
+   for( unsigned op=0; op < MAX_REG_OPERANDS; op++ ) {
+      int reg_num = inst.arch_reg.dst[op]; // this math needs to match that used in function_info::ptx_decode_inst
+      if( reg_num >= 0 ){ // valid register
+         unsigned bank = register_bank(reg_num,inst.warp_id(),m_num_banks,m_bank_warp_shift);
+         if( m_arbiter.bank_idle(bank) ) {
+             m_arbiter.allocate_bank_for_write(bank,op_t(&inst,reg_num,m_num_banks,m_bank_warp_shift));
+             inst.arch_reg.dst[op] = -1;
+         } else {
+             return false;
+         }
       }
    }
    for(unsigned i=0;i<(unsigned)regs.size();i++){
-- 
cgit v1.3