diff options
| author | Inderpreet Singh <[email protected]> | 2011-05-25 16:01:16 -0800 |
|---|---|---|
| committer | Inderpreet Singh <[email protected]> | 2011-05-25 16:01:16 -0800 |
| commit | e1cffaf08733754ab55052b6a2729dae57665d29 (patch) | |
| tree | 6bd1fda8aedad981d8295fdf765c85df1e1c8f3d /src/gpgpu-sim/shader.h | |
| parent | 07376c3d981fbfea6bfad99657c2055b9bdb0e4f (diff) | |
Fix bug #100: local memory address translation returns multiple addresses
Fix bug #101: Coalescing allows multiple accesses per thread for local memory access
This will break atomics which assume at most one thread per mem_fetch. It did not break scoreboard as that logic tracks mem_fetches at warp level, not thread level.
[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 9303]
Diffstat (limited to 'src/gpgpu-sim/shader.h')
| -rw-r--r-- | src/gpgpu-sim/shader.h | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index 8d6c290..7cb3af7 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1242,7 +1242,9 @@ private: friend class scheduler_unit; //this is needed to use private issue warp. void issue_warp( warp_inst_t *&warp, const warp_inst_t *pI, const active_mask_t &active_mask, unsigned warp_id ); void func_exec_inst( warp_inst_t &inst ); - address_type translate_local_memaddr(address_type localaddr, unsigned tid, unsigned num_shader ); + + // Returns numbers of addresses in translated_addrs + unsigned translate_local_memaddr( address_type localaddr, unsigned tid, unsigned num_shader, unsigned datasize, new_addr_type* translated_addrs ); void read_operands(); |