Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
Brad, this patch to affects the number of ticks required for performing a particular number of loads. I don't expect such a thing to happen. Do you? -- Nilay On Wed, 9 Feb 2011, Brad Beckmann wrote: --- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/#review839 --- Overall, this looks great. A pretty simple change that offers significant speedup. I just have one question about the parameters to storeEventInfo. src/mem/ruby/buffers/MessageBuffer.cc http://reviews.m5sim.org/r/328/#comment1196 Is there a reason why you want to pass the message pointer instead of just the vnet id? src/mem/ruby/network/simple/PerfectSwitch.cc http://reviews.m5sim.org/r/328/#comment1197 It seems that you could remove the safe_cast and message pointer dereference if you passed in the vnet id as the first parameter. Am I missing something? - Brad ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
I realized the reason for this. Actually, the wakeup function performs a round-robin scheduling between the incoming links. This code had moved inside the a new if condition that has been introduced in the code. I have moved that code before the if condition so that the timing remains same as before. -- Nilay On Mon, 14 Feb 2011, Nilay Vaish wrote: Brad, this patch to affects the number of ticks required for performing a particular number of loads. I don't expect such a thing to happen. Do you? -- Nilay On Wed, 9 Feb 2011, Brad Beckmann wrote: --- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/#review839 --- Overall, this looks great. A pretty simple change that offers significant speedup. I just have one question about the parameters to storeEventInfo. src/mem/ruby/buffers/MessageBuffer.cc http://reviews.m5sim.org/r/328/#comment1196 Is there a reason why you want to pass the message pointer instead of just the vnet id? src/mem/ruby/network/simple/PerfectSwitch.cc http://reviews.m5sim.org/r/328/#comment1197 It seems that you could remove the safe_cast and message pointer dereference if you passed in the vnet id as the first parameter. Am I missing something? - Brad ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
--- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/#review839 --- Overall, this looks great. A pretty simple change that offers significant speedup. I just have one question about the parameters to storeEventInfo. src/mem/ruby/buffers/MessageBuffer.cc http://reviews.m5sim.org/r/328/#comment1196 Is there a reason why you want to pass the message pointer instead of just the vnet id? src/mem/ruby/network/simple/PerfectSwitch.cc http://reviews.m5sim.org/r/328/#comment1197 It seems that you could remove the safe_cast and message pointer dereference if you passed in the vnet id as the first parameter. Am I missing something? - Brad On 2011-02-05 12:47:34, Nilay Vaish wrote: --- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/ --- (Updated 2011-02-05 12:47:34) Review request for Default. Summary --- Currently the wakeup function for the PerfectSwitch contains three loops - loop on number of virtual networks loop on number of incoming links loop till all messages for this (link, network) have been routed With an 8 processor mesh network and Hammer protocol, about 11-12% of the was observed to have been spent in this function, which is the highest amongst all the functions. It was found that the innermost loop is executed about 45 times per invocation of the wakeup function, when each invocation of the wakeup function processes just about one message. The patch tries to do away with the redundant executions of the innermost loop. Counters have been added for each virtual network that record the number of messages that need to be routed for that virtual network. The inner loops are only executed when the number of messages for that particular virtual network 0. This does away with almost 80% of the executions of the innermost loop. The function now consumes about 5-6% of the total execution time. Diffs - src/mem/ruby/buffers/MessageBuffer.hh UNKNOWN src/mem/ruby/buffers/MessageBuffer.cc UNKNOWN src/mem/ruby/common/Consumer.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.cc UNKNOWN src/mem/ruby/slicc_interface/Message.hh UNKNOWN src/mem/ruby/slicc_interface/NetworkMessage.hh UNKNOWN Diff: http://reviews.m5sim.org/r/328/diff Testing --- Thanks, Nilay ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
On 2011-02-09 09:44:20, Brad Beckmann wrote: src/mem/ruby/buffers/MessageBuffer.cc, line 234 http://reviews.m5sim.org/r/328/diff/2/?file=10257#file10257line234 Is there a reason why you want to pass the message pointer instead of just the vnet id? I would change that. It is a left over from the earlier approach that I was thinking of taking, the one in which all messages were queued in the Perfect Switch as well. - Nilay --- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/#review839 --- On 2011-02-05 12:47:34, Nilay Vaish wrote: --- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/ --- (Updated 2011-02-05 12:47:34) Review request for Default. Summary --- Currently the wakeup function for the PerfectSwitch contains three loops - loop on number of virtual networks loop on number of incoming links loop till all messages for this (link, network) have been routed With an 8 processor mesh network and Hammer protocol, about 11-12% of the was observed to have been spent in this function, which is the highest amongst all the functions. It was found that the innermost loop is executed about 45 times per invocation of the wakeup function, when each invocation of the wakeup function processes just about one message. The patch tries to do away with the redundant executions of the innermost loop. Counters have been added for each virtual network that record the number of messages that need to be routed for that virtual network. The inner loops are only executed when the number of messages for that particular virtual network 0. This does away with almost 80% of the executions of the innermost loop. The function now consumes about 5-6% of the total execution time. Diffs - src/mem/ruby/buffers/MessageBuffer.hh UNKNOWN src/mem/ruby/buffers/MessageBuffer.cc UNKNOWN src/mem/ruby/common/Consumer.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.cc UNKNOWN src/mem/ruby/slicc_interface/Message.hh UNKNOWN src/mem/ruby/slicc_interface/NetworkMessage.hh UNKNOWN Diff: http://reviews.m5sim.org/r/328/diff Testing --- Thanks, Nilay ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
One simple nitpick before you commit is to fix the commit message so you put a proper summary line. Nate On Sat, Feb 5, 2011 at 12:47 PM, Nilay Vaish ni...@cs.wisc.edu wrote: This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/ Review request for Default. By Nilay Vaish. Description Currently the wakeup function for the PerfectSwitch contains three loops - loop on number of virtual networks loop on number of incoming links loop till all messages for this (link, network) have been routed With an 8 processor mesh network and Hammer protocol, about 11-12% of the was observed to have been spent in this function, which is the highest amongst all the functions. It was found that the innermost loop is executed about 45 times per invocation of the wakeup function, when each invocation of the wakeup function processes just about one message. The patch tries to do away with the redundant executions of the innermost loop. Counters have been added for each virtual network that record the number of messages that need to be routed for that virtual network. The inner loops are only executed when the number of messages for that particular virtual network 0. This does away with almost 80% of the executions of the innermost loop. The function now consumes about 5-6% of the total execution time. Diffs - src/mem/ruby/buffers/MessageBuffer.hh (UNKNOWN) - src/mem/ruby/buffers/MessageBuffer.cc (UNKNOWN) - src/mem/ruby/common/Consumer.hh (UNKNOWN) - src/mem/ruby/network/simple/PerfectSwitch.hh (UNKNOWN) - src/mem/ruby/network/simple/PerfectSwitch.cc (UNKNOWN) - src/mem/ruby/slicc_interface/Message.hh (UNKNOWN) - src/mem/ruby/slicc_interface/NetworkMessage.hh (UNKNOWN) View Diff http://reviews.m5sim.org/r/328/diff/ ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
On Wed, 9 Feb 2011, nathan binkert wrote: One simple nitpick before you commit is to fix the commit message so you put a proper summary line. Nate What would you like it to be? ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
Re: [m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
One simple nitpick before you commit is to fix the commit message so you put a proper summary line. Nate What would you like it to be? Whatever you want. Remember that it's supposed to be a single descriptive word followed by a single line summary describing the commit. It's described in the commit access guide. http://www.m5sim.org/wiki/index.php/Commit_Access Nate ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
[m5-dev] Review Request: Ruby: Change PerfectSwitch's wakeup function
--- This is an automatically generated e-mail. To reply, visit: http://reviews.m5sim.org/r/328/ --- Review request for Default. Summary --- Currently the wakeup function for the PerfectSwitch contains three loops - loop on number of virtual networks loop on number of incoming links loop till all messages for this (link, network) have been routed With an 8 processor mesh network and Hammer protocol, about 11-12% of the was observed to have been spent in this function, which is the highest amongst all the functions. It was found that the innermost loop is executed about 45 times per invocation of the wakeup function, when each invocation of the wakeup function processes just about one message. The patch tries to do away with the redundant executions of the innermost loop. Counters have been added for each virtual network that record the number of messages that need to be routed for that virtual network. The inner loops are only executed when the number of messages for that particular virtual network 0. This does away with almost 80% of the executions of the innermost loop. The function now consumes about 5-6% of the total execution time. Diffs - src/mem/ruby/buffers/MessageBuffer.hh UNKNOWN src/mem/ruby/buffers/MessageBuffer.cc UNKNOWN src/mem/ruby/common/Consumer.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.hh UNKNOWN src/mem/ruby/network/simple/PerfectSwitch.cc UNKNOWN src/mem/ruby/slicc_interface/Message.hh UNKNOWN src/mem/ruby/slicc_interface/NetworkMessage.hh UNKNOWN Diff: http://reviews.m5sim.org/r/328/diff Testing --- Thanks, Nilay ___ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev