Merge pull request #21 from abdallm/dev-purdue-integration

A few little things and some sizeable changes: Added memory partition indexing, some configuration file updates, fixed the texture cache so apps that use tex memory will no longer crash.

2 files changed, 27 insertions, 11 deletions

diff --git a/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt b/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt
index 7820e4e..c399db9 100644
--- a/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt
@@ -1,6 +1,6 @@
 //21*1 fly with 32 flits per packet under gpgpusim injection mode
 use_map = 0;
-flit_size = 32; 
+flit_size = 40; 
 
 // currently we do not use this, see subnets below
 network_count = 2;
@@ -17,7 +17,7 @@ routing_function = dest_tag;
 // Flow control
 
 num_vcs     = 1;
-vc_buf_size = 8;
+vc_buf_size = 64;
 
 wait_for_tail_credit = 0;
 
@@ -32,9 +32,9 @@ routing_delay  = 0;
 vc_alloc_delay = 1;
 sw_alloc_delay = 1;
 
-input_speedup     = 2;
+input_speedup     = 1;
 output_speedup    = 1;
-internal_speedup  = 1.0;
+internal_speedup  = 2.0;
 
 // Traffic, GPGPU-Sim does not use this
 
diff --git a/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config b/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config
index 03fcda1..c96432b 100644
--- a/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config
+++ b/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config
@@ -50,20 +50,28 @@
 # <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
 # ** Optional parameter - Required when mshr_type==Texture Fifo
 # Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
--gpgpu_cache:dl1  N:32:128:4,L:L:m:N:H,A:32:8,8
+-gpgpu_cache:dl1  N:32:128:4,L:L:m:N:H,S:64:8,8
 -gpgpu_shmem_size 49152
+-icnt_flit_size 40
+-gmem_skip_L1D 0
+-gpgpu_n_cluster_ejection_buffer_size 32
+-l1_latency 35
+-smem_latency 26
 
 # The alternative configuration for fermi in case cudaFuncCachePreferL1 is selected
-#-gpgpu_cache:dl1  N:64:128:6,L:L:m:N:H,A:32:8,8
+#-gpgpu_cache:dl1  N:64:128:6,L:L:m:N:H,S:64:8,8
 #-gpgpu_shmem_size 16384
 
 # 64 sets, each 128 bytes 8-way for each memory sub partition. This gives 786KB L2 cache
--gpgpu_cache:dl2 N:64:128:8,L:B:m:W:L,A:32:4,4:0,32
+-gpgpu_cache:dl2 S:64:128:8,L:B:m:L:L,A:256:4,4:0,32
 -gpgpu_cache:dl2_texture_only 0 
+-gpgpu_dram_partition_queues 64:64:64:64
+-perf_sim_memcpy 0
+-memory_partition_indexing 0
 
--gpgpu_cache:il1 N:4:128:4,L:R:f:N:L,A:2:32,4
--gpgpu_tex_cache:l1 N:4:128:24,L:R:m:N:L,F:128:4,128:2
--gpgpu_const_cache:l1 N:64:64:2,L:R:f:N:L,A:2:32,4
+-gpgpu_cache:il1 N:4:128:4,L:R:f:N:L,S:2:32,4
+-gpgpu_tex_cache:l1 N:4:128:24,L:R:m:N:L,T:128:4,128:2
+-gpgpu_const_cache:l1 N:64:64:2,L:R:f:N:L,S:2:32,4
 
 # enable operand collector 
 -gpgpu_operand_collector_num_units_sp 6
@@ -76,6 +84,7 @@
 -gpgpu_shmem_num_banks 32
 -gpgpu_shmem_limited_broadcast 0
 -gpgpu_shmem_warp_parts 1
+-gpgpu_coalesce_arch 20
 
 -gpgpu_max_insn_issue_per_warp 1
 
@@ -94,7 +103,7 @@
 # To allow 100% DRAM utility, there should at least be enough buffer to sustain
 # the minimum DRAM latency (100 core cycles).  I.e. 
 #   Total buffer space required = 100 x 924MHz / 700MHz = 132
--gpgpu_frfcfs_dram_sched_queue_size 16
+-gpgpu_frfcfs_dram_sched_queue_size 64
 -gpgpu_dram_return_queue_size 116
 
 # for Fermi, bus width is 384bits, this is 8 bytes (4 bytes at each DRAM chip) per memory partition
@@ -110,6 +119,13 @@
 -gpgpu_dram_timing_opt "nbk=16:CCD=2:RRD=6:RCD=12:RAS=28:RP=12:RC=40:
                         CL=12:WL=4:CDLR=5:WR=12:nbkgrp=4:CCDL=3:RTPL=2"
 
+# select lower bits for bnkgrp to increase bnkgrp parallelism
+-dram_bnk_indexing_policy 0
+-dram_bnkgrp_indexing_policy 1
+
+#-Seperate_Write_Queue_Enable 1
+#-Write_Queue_Size 64:56:32
+
 # Fermi has two schedulers per core
 -gpgpu_num_sched_per_core 2
 # Two Level Scheduler with active and pending pools