From yix at colostate.edu Mon Nov 1 12:11:25 2010 From: yix at colostate.edu (Yi Xiang) Date: Mon, 1 Nov 2010 14:11:25 -0600 Subject: [Hotspot] About read_power() in Hotfloorplan Message-ID: Hello Hotspot, There seems to be a small bug in Hotfloorplanner. For read_power_block() : ////////////////////////////////////////////////////////////////////////// 1 while(!feof(fp)) { 2 fgets(str1, LINE_SIZE, fp); 3 *if (feof(fp))* 4 *break;* 5 strcpy(str2, str1); 6 7 /* ignore comments and empty lines */ 8 ptr = strtok(str1, " \r\t\n"); 9 if (!ptr || ptr[0] == '#') 10 continue; 11 12 if (sscanf(str2, "%s%lf", name, &val) != 2) 13 fatal("invalid power file format\n"); 14 idx = get_blk_index(flp, name); 15 if (idx >= 0) 16 power[idx] = val; 17 else /* since get_blk_index calls fatal, the line below cannot be reached */ 18 fatal ("unit in power file not found in floorplan\n"); 19 } ////////////////////////////////////////////////////////////////////////// In line 3, program will check if *fp points to the EOF. However, when the information of last block in avp.p is just read in, *fp will point to next line, which is EOF. As a result, the while loop will break before the power value of last block can be assigned to power[] in line 16. In debugging, I could see that the value of last block is ignored and remained to be 0. On the other hand, I modified avg.p to give last block a very high value. As expected, the cost result did not change. So, the average power value of last block does not count in. Do I misunderstand or make some mistakes on this? Sorry to take your time. Best, Yi Xiang -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101101/f6966703/attachment.html From yix at colostate.edu Mon Nov 1 13:23:56 2010 From: yix at colostate.edu (Yi Xiang) Date: Mon, 1 Nov 2010 15:23:56 -0600 Subject: [Hotspot] About read_power() in Hotfloorplan In-Reply-To: References: Message-ID: Sorry, I executed hotfloorplan with original makefile. This part became correct. It seems that the mechanism of fgets() changed in my modified version. I am not clear why this happens at present. Sorry to take your time. Best, Yi On Mon, Nov 1, 2010 at 2:11 PM, Yi Xiang wrote: > Hello Hotspot, > > There seems to be a small bug in Hotfloorplanner. > For read_power_block() : > > ////////////////////////////////////////////////////////////////////////// > 1 while(!feof(fp)) { > 2 fgets(str1, LINE_SIZE, fp); > 3 *if (feof(fp))* > 4 *break;* > 5 strcpy(str2, str1); > 6 > 7 /* ignore comments and empty lines */ > 8 ptr = strtok(str1, " \r\t\n"); > 9 if (!ptr || ptr[0] == '#') > 10 continue; > 11 > 12 if (sscanf(str2, "%s%lf", name, &val) != 2) > 13 fatal("invalid power file format\n"); > 14 idx = get_blk_index(flp, name); > 15 if (idx >= 0) > 16 power[idx] = val; > 17 else /* since get_blk_index calls fatal, the line below cannot > be reached */ > 18 fatal ("unit in power file not found in floorplan\n"); > 19 } > ////////////////////////////////////////////////////////////////////////// > > > In line 3, program will check if *fp points to the EOF. However, when the > information of last block in avp.p is just read in, *fp will point to next > line, which is EOF. > As a result, the while loop will break before the power value of last block > can be assigned to power[] in line 16. > > In debugging, I could see that the value of last block is ignored and > remained to be 0. On the other hand, I modified avg.p to give last block a > very high value. As expected, the cost result did not change. > So, the average power value of last block does not count in. > > Do I misunderstand or make some mistakes on this? > Sorry to take your time. > > Best, > Yi Xiang > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101101/87a18036/attachment.html From Runjie at virginia.edu Wed Nov 3 19:01:04 2010 From: Runjie at virginia.edu (Runjie Zhang) Date: Wed, 3 Nov 2010 23:01:04 -0400 Subject: [Hotspot] How to change McPAT to generate power trace for HotSpot Message-ID: Hi, Xiwei I am also forwarding this email thread to the entire mailing list since many people might be interested in this. In McPAT (I assume version 0.8_r274), all the power info were stored in the class "Processor", which was defined in processor.h. For example, if you want to get the dynamic power for a core, you can use "core.power.readOp.dynamic" in a member function of class Processor. The steps I would suggest to do modify McPAT are: 1. Specify the output file where McPAT writes traces for HotSpot. You can do that in main.cc and easily make the output file a cmd line parameter. 2. Build a function to write the first line in your power trace file. The quantity of each function unit is stored in class ParseXML and based on that, you can print out names and ID numbers for all units according to your granularity requirements. 3. Print power values of each unit. As mentioned above, power values are stored in class Processor It it worth mentioning that besides the power trace, HotSpot requires another important input, floorplan (.flp). For each power trace you get from McPAT, a floorplan is also required to drive HotSpot. You may want to make sure that your expectation for power trace granularity won't exceed the floorplan generating methods' limits. Thanks, Runjie -------- Original Message -------- Hi, I am using HotSpot and need to generate the power trace file. Besides the common choice Wattch, I found that McPAT could also be used. Can you give me some hints on how to change McPAT to generate the power trace? Thanks & Regards, Xiwei -- Runjie Zhang Computer Engineering University of Virginia -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101103/44bc6c73/attachment-0001.html From fsamie at gmail.com Wed Nov 3 22:02:02 2010 From: fsamie at gmail.com (farzad samie) Date: Thu, 4 Nov 2010 09:32:02 +0330 Subject: [Hotspot] PTscalar (Solid Repel) Message-ID: Hi Kamesh, PTscalar is very similar to simplescalar and you can use many of tutorial which belong to simplescalar to install and use PTScalar. to install PTScalar, in command line: 1- make config-pisa OR make config-alpha 2- make after instalation, you can run PTScalar by: ./sim-outorder ... Use simplescalar README and HOWTO files for more help on switches and arguments. > Message: 1 > Date: Tue, 19 Oct 2010 15:46:48 +0300 > From: Solid Repel > Subject: [Hotspot] PTscalar > To: hotspot at mail.cs.virginia.edu > Message-ID: > > Content-Type: text/plain; charset="iso-8859-1" > > Hello Hotspot, > > I am new to hotspot and other simulators. Can any one let me know how to > run > PTscalar (basically the command to be run on linux). As far as I can see > there is no "Readme" corresponding to PTscalar in the archive describing > what it can accomplish and how it is to be run. All the readme files talk > about simplescalar. I understand that this is not the forum to talk about > PTscalar but I assume lot many people would be using it in here and would > be > able to help. > > Thanks and Regards, > Kamesh. > > -- Farzad Samie -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101104/843672d0/attachment.html From swayamsiddha at gmail.com Mon Nov 15 10:34:49 2010 From: swayamsiddha at gmail.com (Swayamsiddha Dandekar) Date: Mon, 15 Nov 2010 10:34:49 -0800 Subject: [Hotspot] Library issue Message-ID: Hello Hotspot, I am new to hotspot and other simulators. I have an issue regarding changing the library path. I am using HotSpot to do Voltage and Temperature modeling of the OpenSparcT2 processor. I am trying to follow the HOWTO steps for installation and building HotSpot, but I having trouble when it says "For Linux/UNIX systems with BLAS/LAPACK libraries installed,uncomment the lines corresponding to your installation of the BLAS/LAPACK vendor libraries (see http://www.netlib.org/blas/faq.html#5) and set the path and compiler options corresponding to your library" , but it doesn't specify where am I supposed to go change this path and this link is not helpful either. Can someone please explain where can I change its path to my library (of the processor) which I want to do the modeling for. Thanks a lot -- Regards Swayams -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101115/95914a5b/attachment.html From ks4kk at virginia.edu Mon Nov 15 12:20:19 2010 From: ks4kk at virginia.edu (Karthik Sankaranarayanan) Date: Mon, 15 Nov 2010 12:20:19 -0800 Subject: [Hotspot] Library issue In-Reply-To: References: Message-ID: Hi, Please uncomment the lines in Makefile corresponding to your installation. Thanks, -karthik On Mon, Nov 15, 2010 at 10:34 AM, Swayamsiddha Dandekar < swayamsiddha at gmail.com> wrote: > Hello Hotspot, > > I am new to hotspot and other simulators. I have an issue regarding > changing the library path. I am using HotSpot to do Voltage and Temperature > modeling of the OpenSparcT2 processor. I am trying to follow the HOWTO steps > for installation and building HotSpot, but I having trouble when it says "For > Linux/UNIX systems with BLAS/LAPACK libraries installed,uncomment the > lines corresponding to your installation of the BLAS/LAPACK vendor > libraries (see http://www.netlib.org/blas/faq.html#5) and set the path and > compiler options corresponding to your library" , but it doesn't specify > where am I supposed to go change this path and this link is not helpful > either. > > Can someone please explain where can I change its path to my library (of > the processor) which I want to do the modeling for. > > Thanks a lot > > -- > Regards > Swayams > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20101115/b16b1d92/attachment.html