Patch to add L3 and L4 in GEM5

main
Atharva 5 months ago committed by GitHub
parent fe3e8ca97c
commit 909750cb9d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

@ -0,0 +1,232 @@
From c6965615b1fad9c3e60ef286ce8e980874a86dc9 Mon Sep 17 00:00:00 2001
From: Atharva Gondhalekar <atharva1@hazelnut.rlogin>
Date: Sat, 10 Feb 2024 15:05:46 -0500
Subject: [PATCH] Patch for L3 and L4

---
configs/common/CacheConfig.py | 44 ++++++++++++++++++++++++++++++++---
configs/common/Caches.py | 17 ++++++++++++++
configs/common/Options.py | 7 +++++-
src/cpu/BaseCPU.py | 32 +++++++++++++++++++++++++
src/mem/XBar.py | 18 ++++++++++++++
5 files changed, 114 insertions(+), 4 deletions(-)

diff --git a/configs/common/CacheConfig.py b/configs/common/CacheConfig.py
index 4f21a43924..00015817b1 100644
--- a/configs/common/CacheConfig.py
+++ b/configs/common/CacheConfig.py
@@ -90,7 +90,7 @@ def config_cache(options, system):
print("O3_ARM_v7a_3 is unavailable. Did you compile the O3 model?")
sys.exit(1)
- dcache_class, icache_class, l2_cache_class, walk_cache_class = (
+ dcache_class, icache_class, l2_cache_class, l3_cache_class, walk_cache_class = (
core.O3_ARM_v7a_DCache,
core.O3_ARM_v7a_ICache,
core.O3_ARM_v7aL2,
@@ -103,17 +103,19 @@ def config_cache(options, system):
print("HPI is unavailable.")
sys.exit(1)
- dcache_class, icache_class, l2_cache_class, walk_cache_class = (
+ dcache_class, icache_class, l2_cache_class, l3_cache_class, l4_cache_class, walk_cache_class = (
core.HPI_DCache,
core.HPI_ICache,
core.HPI_L2,
None,
)
else:
- dcache_class, icache_class, l2_cache_class, walk_cache_class = (
+ dcache_class, icache_class, l2_cache_class, l3_cache_class, l4_cache_class, walk_cache_class = (
L1_DCache,
L1_ICache,
L2Cache,
+ L3Cache,
+ L4Cache,
None,
)
@@ -138,6 +140,42 @@ def config_cache(options, system):
system.tol2bus = L2XBar(clk_domain=system.cpu_clk_domain)
system.l2.cpu_side = system.tol2bus.mem_side_ports
system.l2.mem_side = system.membus.cpu_side_ports
+
+ #Adding L3 cache implementation below
+ if options.l2cache and options.l3cache:
+ system.l2 = l2_cache_class(clk_domain=system.cpu_clk_domain, size=options.l2_size, assoc=options.l2_assoc)
+
+ system.l3 = l3_cache_class(clk_domain=system.cpu_clk_domain, size=options.l3_size, assoc=options.l3_assoc)
+
+ system.tol2bus = L2XBar(clk_domain = system.cpu_clk_domain)
+ system.tol3bus = L3XBar(clk_domain = system.cpu_clk_domain)
+
+ system.l2.cpu_side = system.tol2bus.master
+ system.l2.mem_side = system.tol3bus.slave
+
+ system.l3.cpu_side = system.tol3bus.master
+ system.l3.mem_side = system.membus.slave
+
+ #Adding L4 cache implementation here
+ if options.l2cache and options.l3cache and options.l4cache:
+ system.l2 = l2_cache_class(clk_domain=system.cpu_clk_domain, size=options.l2_size, assoc=options.l2_assoc)
+
+ system.l3 = l3_cache_class(clk_domain=system.cpu_clk_domain, size=options.l3_size, assoc=options.l3.assoc)
+
+ system.l4 = l4_cache_class(clk_domain=system.cpu_clk_domain, size=options.l4_size, assoc=options.l4.assoc)
+
+ system.tol2bus = L2XBar(clk_domain = system.cpu_clk_domain)
+ system.tol3bus = L3XBar(clk_domain = system.cpu_clk_domain)
+ system.tol4bus = L4XBar(clk_domain = system.cpu_clk_domain)
+
+ system.l2.cpu_side = system.tol2bus.master
+ system.l2.mem_side = system.tol3bus.slave
+
+ system.l3.cpu_sude = system.tol3bus.master
+ system.l3.mem_side = system.membus.slave
+
+ system.l4.cpu_sude = system.tol4bus.master
+ system.l4.mem_side = system.membus.slavem
if options.memchecker:
system.memchecker = MemChecker()
diff --git a/configs/common/Caches.py b/configs/common/Caches.py
index fed9ac7d19..ee8d28453f 100644
--- a/configs/common/Caches.py
+++ b/configs/common/Caches.py
@@ -77,6 +77,23 @@ class L2Cache(Cache):
tgts_per_mshr = 12
write_buffers = 8
+class L3Cache(Cache):
+ assoc = 64
+ tag_latency = 32
+ data_latency = 32
+ response_latency = 32
+ mshrs = 32
+ tgts_per_mshr = 24
+ write_buffers = 16
+
+class L4Cache(Cache):
+ assoc = 64
+ tag_latency = 32
+ data_latency = 32
+ response_latency = 32
+ mshrs = 32
+ tgts_per_mshr = 24
+ write_buffers = 16
class IOCache(Cache):
assoc = 8
diff --git a/configs/common/Options.py b/configs/common/Options.py
index 97335f13b9..e7dd79d864 100644
--- a/configs/common/Options.py
+++ b/configs/common/Options.py
@@ -185,19 +185,24 @@ def addNoISAOptions(parser):
)
parser.add_argument("--caches", action="store_true")
parser.add_argument("--l2cache", action="store_true")
+ parser.add_argument("--l3cache", action="store_true")
+ parser.add_argument("--l4cache", action="store_true")
parser.add_argument("--num-dirs", type=int, default=1)
parser.add_argument("--num-l2caches", type=int, default=1)
parser.add_argument("--num-l3caches", type=int, default=1)
+ parser.add_argument("--num-l4caches", type=int, default=1)
parser.add_argument("--l1d_size", type=str, default="64kB")
parser.add_argument("--l1i_size", type=str, default="32kB")
parser.add_argument("--l2_size", type=str, default="2MB")
parser.add_argument("--l3_size", type=str, default="16MB")
+ parser.add_argument("--l4_size", type=str, default="48MB")
parser.add_argument("--l1d_assoc", type=int, default=2)
parser.add_argument("--l1i_assoc", type=int, default=2)
parser.add_argument("--l2_assoc", type=int, default=8)
parser.add_argument("--l3_assoc", type=int, default=16)
+ parser.add_argument("--l4_assoc", type=int, default=32)
parser.add_argument("--cacheline_size", type=int, default=64)
-
+
# Enable Ruby
parser.add_argument("--ruby", action="store_true")
diff --git a/src/cpu/BaseCPU.py b/src/cpu/BaseCPU.py
index 9ba60ef1b8..5ff3f456a1 100644
--- a/src/cpu/BaseCPU.py
+++ b/src/cpu/BaseCPU.py
@@ -49,6 +49,8 @@ from m5.objects.Platform import Platform
from m5.objects.ResetPort import ResetResponsePort
from m5.objects.SubSystem import SubSystem
from m5.objects.XBar import L2XBar
+from m5.objects.XBar import L3XBar
+from m5.objects.XBar import L4XBar
from m5.params import *
from m5.proxy import *
from m5.SimObject import *
@@ -222,6 +224,36 @@ class BaseCPU(ClockedObject):
self.toL2Bus.mem_side_ports = self.l2cache.cpu_side
self._cached_ports = ["l2cache.mem_side"]
+
+ #Adding L3 cache implementation below
+ def addThreeLevelCacheHierarchy(self, ic, dc, l2c, iwc = None, dwc = None, xbar=None):
+ self.addPrivateSplitL2Caches(ic, dc, iwc, dwc)
+ self.toL3Bus = L3XBar()
+ self.connectCachedPorts(self.toL3Bus)
+ self.l3cache = l3c
+ self.toL3Bus.master = self.l3cache.cpu_side
+ self._cached_ports = ['l3cache.mem_side']
+
+ #Adding 4 level cache heirarchy
+ def addFourLevelCacheHierarchy(self, ic, dc, l2c, l3c, l4c, iwc = None, dwc = None, xbar=None):
+ self.addPrivateSplitL1Caches(ic, dc, iwc, dwc)
+ self.toL2Bus = xbar if xbar else L4XBar()
+ self.connectCachedPorts(self.toL2Bus.cpu_side_ports)
+ self.l2cache = l2c
+ self.toL2Bus.mem_side_ports = self.l2cache.cpu_side
+ self._cached_ports = ['l2cache.mem_side']
+ self.toL3Bus = L3XBar()
+ self.connectCachedPorts(self.toL3Bus)
+ self.l3Cache = l3c
+ self.toL3Bus.master = self.l3cache.cpu_side
+ self._cached_ports = ['l3cache.mem_side']
+ self.toL4Bus = L4Xbar()
+ self.connectCachedPorts(self.toL4Bus)
+ self.l4cache = l4c
+ self.toL4Bus.master = self.l4cache.cpu_side
+ self._cached_ports = ['l4cache.mem_side']
+
+
def createThreads(self):
# If no ISAs have been created, assume that the user wants the
# default ISA.
diff --git a/src/mem/XBar.py b/src/mem/XBar.py
index 927d3bbe36..4a0db7a794 100644
--- a/src/mem/XBar.py
+++ b/src/mem/XBar.py
@@ -178,7 +178,25 @@ class L2XBar(CoherentXBar):
# to the first level of unified cache.
point_of_unification = True
+#Adding L3 cache implementation below
+class L3XBar(CoherentXBar):
+ #256-bit crossbar by default
+ width = 32
+ frontend_latency = 1
+ forward_latency = 0
+ response_latency = 1
+ snoop_response_latency = 1
+ snoop_filter = SnoopFilter(lookup_latency = 0)
+class L4XBar(CoherentXBar):
+ width = 32
+ frontend_latency = 1
+ forward_latency = 0
+ response_latency = 1
+ snoop_response_latency = 1
+ snoop_filter = SnoopFilter(lookup_latency = 0)
+
+
# One of the key coherent crossbar instances is the system
# interconnect, tying together the CPU clusters, GPUs, and any I/O
# coherent requestors, and DRAM controllers.
--
2.43.0

Loading…
Cancel
Save