ruby: new atomics implementation

This patch changes the way that Ruby handles atomic RMW instructions. This implementation, unlike the prior one, is protocol independent. It works by locking an address from the sequencer immediately after the read portion of an RMW completes. When that address is locked, the coherence controller will only satisfy requests coming from one port (e.g., the mandatory queue) and will ignore all others. After the write portion completed, the line is unlocked. This should also work with multi-line atomics, as long as the blocks are always acquired in the same order.

3 files changed, 10 insertions, 10 deletions

diff --git a/src/mem/ruby/config/MI_example-homogeneous.rb b/src/mem/ruby/config/MI_example-homogeneous.rb
index 71e20c318..d409e6782 100644
--- a/src/mem/ruby/config/MI_example-homogeneous.rb
+++ b/src/mem/ruby/config/MI_example-homogeneous.rb
@@ -13,7 +13,7 @@ RubySystem.reset
 # default values
 
 num_cores = 2
-l1_cache_size_kb = 32768
+l1_cache_size_bytes = 32768
 l1_cache_assoc = 8
 l1_cache_latency = 1
 num_memories = 2
diff --git a/src/mem/ruby/config/TwoLevel_SplitL1UnifiedL2.rb b/src/mem/ruby/config/TwoLevel_SplitL1UnifiedL2.rb
index a8ef1eceb..ee22df656 100644
--- a/src/mem/ruby/config/TwoLevel_SplitL1UnifiedL2.rb
+++ b/src/mem/ruby/config/TwoLevel_SplitL1UnifiedL2.rb
@@ -68,8 +68,8 @@ assert((protocol == "MESI_CMP_directory" or protocol == "MOESI_CMP_directory"),
 require protocol+".rb"
 
 num_cores.times { |n|
-  icache = SetAssociativeCache.new("l1i_"+n.to_s, l1_icache_size_kb, l1_icache_latency, l1_icache_assoc, "PSEUDO_LRU")
-  dcache = SetAssociativeCache.new("l1d_"+n.to_s, l1_dcache_size_kb, l1_dcache_latency, l1_dcache_assoc, "PSEUDO_LRU")
+  icache = SetAssociativeCache.new("l1i_"+n.to_s, l1_icache_size_kb*1024, l1_icache_latency, l1_icache_assoc, "PSEUDO_LRU")
+  dcache = SetAssociativeCache.new("l1d_"+n.to_s, l1_dcache_size_kb*1024, l1_dcache_latency, l1_dcache_assoc, "PSEUDO_LRU")
   sequencer = Sequencer.new("Sequencer_"+n.to_s, icache, dcache)
   iface_ports << sequencer
   if protocol == "MOESI_CMP_directory"
@@ -87,7 +87,7 @@ num_cores.times { |n|
   end
 }
 num_l2_banks.times { |n|
-  cache = SetAssociativeCache.new("l2u_"+n.to_s, l2_cache_size_kb/num_l2_banks, l2_cache_latency, l2_cache_assoc, "PSEUDO_LRU")
+  cache = SetAssociativeCache.new("l2u_"+n.to_s, (l2_cache_size_kb*1024)/num_l2_banks, l2_cache_latency, l2_cache_assoc, "PSEUDO_LRU")
   if protocol == "MOESI_CMP_directory"
     net_ports << MOESI_CMP_directory_L2CacheController.new("L2CacheController_"+n.to_s,
                                                            "L2Cache",
diff --git a/src/mem/ruby/config/cfg.rb b/src/mem/ruby/config/cfg.rb
index d57862420..a20562243 100644
--- a/src/mem/ruby/config/cfg.rb
+++ b/src/mem/ruby/config/cfg.rb
@@ -385,12 +385,12 @@ class DMAController < NetPort
 end
 
 class Cache < LibRubyObject
-  param :size_kb, Integer
+  param :size, Integer
   param :latency, Integer
   param :controller, NetPort
-  def initialize(obj_name, size_kb, latency)
+  def initialize(obj_name, size, latency)
     super(obj_name)
-    self.size_kb = size_kb
+    self.size = size
     self.latency = latency
     # controller must be set manually by the configuration script
     # because there is a cyclic dependence
@@ -406,8 +406,8 @@ class SetAssociativeCache < Cache
   #  when an integer, it represents the number of cycles for a hit
   #  when a float, it represents the cache access time in ns
   #  when set to "auto", libruby will attempt to find a realistic latency by running CACTI
-  def initialize(obj_name, size_kb, latency, assoc, replacement_policy)
-    super(obj_name, size_kb, latency)
+  def initialize(obj_name, size, latency, assoc, replacement_policy)
+    super(obj_name, size, latency)
     self.assoc = assoc
     self.replacement_policy = replacement_policy
   end
@@ -415,7 +415,7 @@ class SetAssociativeCache < Cache
   def calculateLatency()
     if self.latency == "auto"
       cacti_args = Array.new()
-      cacti_args << (self.size_kb*1024) <<  RubySystem.block_size_bytes << self.assoc
+      cacti_args << (self.size*1024) <<  RubySystem.block_size_bytes << self.assoc
       cacti_args << 1 << 0 << 0 << 0 << 1
       cacti_args << RubySystem.tech_nm << RubySystem.block_size_bytes*8
       cacti_args << 0 << 0 << 0 << 1 << 0 << 0 << 0 << 0 << 1