summaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
authorAndreas Sandberg <andreas.sandberg@arm.com>2017-02-27 13:17:51 +0000
committerAndreas Sandberg <andreas.sandberg@arm.com>2017-03-07 11:50:35 +0000
commitb043dcf58ad766582aeab162fb855cc3fc95f2cf (patch)
tree1da517346bef667cf20f80a54c934a3931129e4f /src
parentc07a2d68f34ab4cef8b3ed4d079316d2b4306244 (diff)
downloadgem5-b043dcf58ad766582aeab162fb855cc3fc95f2cf.tar.xz
gpu-compute: Fix Python/C++ object hierarchy discrepancies
The GPUCoalescer and the Shader classes have different base classes in C++ and Python. This causes subtle bugs in SWIG and compilation errors for PyBind. Change-Id: I1ddd2a8ea43f083470538ddfea891347b21d14d8 Reviewed-by: Andreas Hansson <andreas.hansson@arm.com> Reviewed-on: https://gem5-review.googlesource.com/2228 Maintainer: Andreas Sandberg <andreas.sandberg@arm.com> Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com> Reviewed-by: Pierre-Yves PĂ©neau <pierre-yves.peneau@lirmm.fr> Reviewed-by: Bradford Beckmann <brad.beckmann@amd.com>
Diffstat (limited to 'src')
-rw-r--r--src/gpu-compute/shader.cc2
-rw-r--r--src/gpu-compute/shader.hh2
-rw-r--r--src/mem/ruby/system/GPUCoalescer.hh11
-rw-r--r--src/mem/ruby/system/GPUCoalescer.py10
4 files changed, 17 insertions, 8 deletions
diff --git a/src/gpu-compute/shader.cc b/src/gpu-compute/shader.cc
index 6deaaab94..6c328a7fd 100644
--- a/src/gpu-compute/shader.cc
+++ b/src/gpu-compute/shader.cc
@@ -50,7 +50,7 @@
#include "mem/ruby/system/RubySystem.hh"
#include "sim/sim_exit.hh"
-Shader::Shader(const Params *p) : SimObject(p),
+Shader::Shader(const Params *p) : ClockedObject(p),
clock(p->clk_domain->clockPeriod()), cpuThread(nullptr), gpuTc(nullptr),
cpuPointer(p->cpu_pointer), tickEvent(this), timingSim(p->timing),
hsail_mode(SIMT), impl_kern_boundary_sync(p->impl_kern_boundary_sync),
diff --git a/src/gpu-compute/shader.hh b/src/gpu-compute/shader.hh
index 13afab977..55c3feef9 100644
--- a/src/gpu-compute/shader.hh
+++ b/src/gpu-compute/shader.hh
@@ -73,7 +73,7 @@ static const int LDS_SIZE = 65536;
// Class Shader: This describes a single shader instance. Most
// configurations will only have a single shader.
-class Shader : public SimObject
+class Shader : public ClockedObject
{
protected:
// Shader's clock period in terms of number of ticks of curTime,
diff --git a/src/mem/ruby/system/GPUCoalescer.hh b/src/mem/ruby/system/GPUCoalescer.hh
index 557d39235..2b42e1933 100644
--- a/src/mem/ruby/system/GPUCoalescer.hh
+++ b/src/mem/ruby/system/GPUCoalescer.hh
@@ -49,7 +49,7 @@
#include "mem/request.hh"
#include "mem/ruby/common/Address.hh"
#include "mem/ruby/common/Consumer.hh"
-#include "mem/ruby/system/RubyPort.hh"
+#include "mem/ruby/system/Sequencer.hh"
class DataBlock;
class CacheMsg;
@@ -255,10 +255,6 @@ class GPUCoalescer : public RubyPort
bool handleLlsc(Addr address, GPUCoalescerRequest* request);
- // Private copy constructor and assignment operator
- GPUCoalescer(const GPUCoalescer& obj);
- GPUCoalescer& operator=(const GPUCoalescer& obj);
-
class IssueEvent : public Event
{
private:
@@ -370,6 +366,11 @@ class GPUCoalescer : public RubyPort
std::vector<Stats::Histogram *> m_InitialToForwardDelayHist;
std::vector<Stats::Histogram *> m_ForwardToFirstResponseDelayHist;
std::vector<Stats::Histogram *> m_FirstResponseToCompletionDelayHist;
+
+private:
+ // Private copy constructor and assignment operator
+ GPUCoalescer(const GPUCoalescer& obj);
+ GPUCoalescer& operator=(const GPUCoalescer& obj);
};
inline std::ostream&
diff --git a/src/mem/ruby/system/GPUCoalescer.py b/src/mem/ruby/system/GPUCoalescer.py
index 0c19f875d..101a5fe30 100644
--- a/src/mem/ruby/system/GPUCoalescer.py
+++ b/src/mem/ruby/system/GPUCoalescer.py
@@ -36,7 +36,7 @@ from m5.params import *
from m5.proxy import *
from Sequencer import *
-class RubyGPUCoalescer(RubySequencer):
+class RubyGPUCoalescer(RubyPort):
type = 'RubyGPUCoalescer'
cxx_class = 'GPUCoalescer'
cxx_header = "mem/ruby/system/GPUCoalescer.hh"
@@ -46,3 +46,11 @@ class RubyGPUCoalescer(RubySequencer):
"max requests (incl. prefetches) outstanding")
assume_rfo = Param.Bool(True, "assume protocol implementes Read for "
"Ownership coherence");
+
+ icache = Param.RubyCache("")
+ dcache = Param.RubyCache("")
+ deadlock_threshold = Param.Cycles(500000,
+ "max outstanding cycles for a request before " \
+ "deadlock/livelock declared")
+ garnet_standalone = Param.Bool(False, "")
+ dcache_hit_latency = Param.Cycles(1, "Data cache hit latency")