From liujun at ict.ac.cn Thu Nov 6 06:59:18 2008 From: liujun at ict.ac.cn (=?gb2312?B?wfW+/A==?=) Date: Thu, 6 Nov 2008 22:59:18 +0800 Subject: [Hotspot] about hotspot-4.1 Message-ID: <200811062259180311622@ict.ac.cn> hi, When I use the function void read_power_block (block_model_t *model, double *power, char *file), however,it couldn't read gcc.ptrace file, it shows "invalid power file format". I find the function is not used in the file of hotspot.c. What's wrong with it? Jun Liu -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20081106/ba75d28f/attachment.html From ks4kk at cs.virginia.edu Thu Nov 6 09:05:32 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Thu, 6 Nov 2008 12:05:32 -0500 (EST) Subject: [Hotspot] about hotspot-4.1 In-Reply-To: <200811062259180311622@ict.ac.cn> References: <200811062259180311622@ict.ac.cn> Message-ID: The read_power_* functions are for reading power numbers output by the corresponding dump_power_* functions, which essentially output *one* set of power numbers in a two-column format. The power trace file contains *many* sets of power numbers with one set per row. For reading the power trace file, refer to the code in hotspot.c that does it. The relevant snippet from hotspot.c is given below: /* read the instantaneous power trace */ vals = dvector(MAX_UNITS); while ((num=read_vals(pin, vals)) != 0) { ... } Thanks, -karthik On Thu, 6 Nov 2008, ???? wrote: > hi, > When I use the function > void read_power_block (block_model_t *model, double *power, char *file), however,it couldn't read gcc.ptrace file, it shows > "invalid power file format". I find the function is not used in the file of hotspot.c. What's wrong with it? > > > > Jun Liu > From augusto.vega at bsc.es Sat Nov 8 02:09:19 2008 From: augusto.vega at bsc.es (Augusto Vega) Date: Sat, 08 Nov 2008 11:09:19 +0100 Subject: [Hotspot] Time step in power trace Message-ID: <4915654F.1050505@bsc.es> Hi, I've a simulator that generates a power trace file to be used with HotSpot. Unlike Wattch power traces, my power traces also have information of time (every row has the time when the power was calculated) because power is *not* computed every cycle in my simulator. So, because my time step is not constant (I've seen that HotSpot considere a constant sampling interval of 3.333e-06 sec) I need to modify HotSpot: first, to read the time values from the trace and second, to compute temperatures using not constant intervals. In order to do that I'll change all calls to `compute_temp()' to pass the time step read from power trace instead of the constant one. What I would like to know is if this change is correct and enough and if there could be "side effects" by doing that. Thank you in advance. -- Augusto Vega BSC-CNS Barcelona Supercomputing Center Centro Nacional de Supercomputaci?n Telf.: +34 93 413 79 46 e-mail: augusto.vega at bsc.es From liujun at ict.ac.cn Sat Nov 8 00:32:24 2008 From: liujun at ict.ac.cn (=?gb2312?B?wfW+/A==?=) Date: Sat, 8 Nov 2008 16:32:24 +0800 Subject: [Hotspot] about hotspot-4.1 read_power() Message-ID: <200811081632245783109@ict.ac.cn> Hi,every one. I'm using the the function void read_power_grid(block_model_t *model, double *power, char *file) for simulate 3D chip which has multi-layers, but I do not know what format is needed for the function workinging properply. It always inform "invalid power format". Would you give me a example for the format of the power file format? Thanks, looking forward to your reply. Jun Liu 2008-11-08 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.cs.virginia.edu/pipermail/hotspot/attachments/20081108/3b16cc59/attachment.html From ks4kk at cs.virginia.edu Sun Nov 9 07:47:02 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Sun, 9 Nov 2008 10:47:02 -0500 (EST) Subject: [Hotspot] Time step in power trace In-Reply-To: <4915654F.1050505@bsc.es> References: <4915654F.1050505@bsc.es> Message-ID: Hello Augusto, Your approach is correct. You can pass any reasonable value as the time interval for compute_temp. There are no side effects. Thanks, -karthik On Sat, 8 Nov 2008, Augusto Vega wrote: > Hi, > > I've a simulator that generates a power trace file to be used with > HotSpot. Unlike Wattch power traces, my power traces also have > information of time (every row has the time when the power was > calculated) because power is *not* computed every cycle in my simulator. > > So, because my time step is not constant (I've seen that HotSpot > considere a constant sampling interval of 3.333e-06 sec) I need to > modify HotSpot: first, to read the time values from the trace and > second, to compute temperatures using not constant intervals. In order > to do that I'll change all calls to `compute_temp()' to pass the time > step read from power trace instead of the constant one. > > What I would like to know is if this change is correct and enough and if > there could be "side effects" by doing that. > > Thank you in advance. > > -- > Augusto Vega > > BSC-CNS > Barcelona Supercomputing Center > Centro Nacional de Supercomputaci?n > Telf.: +34 93 413 79 46 > e-mail: augusto.vega at bsc.es > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > From ks4kk at cs.virginia.edu Sun Nov 9 08:28:25 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Sun, 9 Nov 2008 11:28:25 -0500 (EST) Subject: [Hotspot] about hotspot-4.1 read_power() In-Reply-To: <200811081632245783109@ict.ac.cn> References: <200811081632245783109@ict.ac.cn> Message-ID: Hello Jun, read_power_grid reads files that are in a two-column format. If you are using a 3D floorplan, for each power dissipating layer "" (this precludes heatsink and heat spreader layers which do not dissipate power), the line is of the form "layer__\t\n" where is the name of the functional unit in the floorplan and is its power. Here is a sample corresponding to the "example.lcf" file distributed with HotSpot. Let us assume that each unit dissipates 1W of power. Then the format is: layer_0_Unit1 1.0 layer_0_Unit2 1.0 layer_0_Unit3 1.0 layer_2_L2_left 1.0 layer_2_L2 1.0 layer_2_L2_right 1.0 layer_2_Icache 1.0 layer_2_Dcache 1.0 layer_2_Bpred 1.0 layer_2_DTB 1.0 layer_2_FPAdd 1.0 layer_2_FPReg 1.0 layer_2_FPMul 1.0 layer_2_FPMap 1.0 layer_2_IntMap 1.0 layer_2_IntQ 1.0 layer_2_IntReg 1.0 layer_2_IntExec 1.0 layer_2_FPQ 1.0 layer_2_LdStQ 1.0 layer_2_ITB 1.0 Note that layer_1 and layer_3 are missing because they do not dissipate power. If this explanation is not clear, as mentioned in our previous e-mail, we encourage you to use the "dump_power_grid" function first with a default (uninitialized) power vector allocated using "hotspot_vector". The "dump_power_grid" function will output values to the speficied file in the format above. You can then edit that file as needed to use it with "read_power_grid". Hope this helps. Thanks, -karthik On Sat, 8 Nov 2008, ???? wrote: > Hi,every one. I'm using the the function void > read_power_grid(block_model_t *model, double *power, char *file) for > simulate 3D chip which has multi-layers, but I do not know what format > is needed for the function workinging properply. It always inform > "invalid power format". Would you give me a example for the format of > the power file format? Thanks, looking forward to your reply. > > > > > Jun Liu > 2008-11-08 > The read_power_* functions are for reading power numbers output by the corresponding dump_power_* functions, which essentially output *one* set of power numbers in a two-column format. The power trace file contains *many* sets of power numbers with one set per row. For reading the power trace file, refer to the code in hotspot.c that does it. The relevant snippet from hotspot.c is given below: /* read the instantaneous power trace */ vals = dvector(MAX_UNITS); while ((num=read_vals(pin, vals)) != 0) { ... } Thanks, -karthik On Thu, 6 Nov 2008, ???? wrote: > hi, > When I use the function > void read_power_block (block_model_t *model, double *power, char *file), > however,it couldn't read gcc.ptrace file, it shows > "invalid power file format". I find the function is not used in the file > of hotspot.c. What's wrong with it? > > > > Jun Liu From yao1 at wisc.edu Mon Nov 17 12:18:05 2008 From: yao1 at wisc.edu (CHUNHUA YAO) Date: Mon, 17 Nov 2008 14:18:05 -0600 Subject: [Hotspot] Units in HotSpot 4.1 In-Reply-To: References: Message-ID: Hi, I am new to HotSpot 4.1. I wonder what's the units of the data inside the example floorplan and power trance files. In the floorplan file "ev6.flp": L2_left 0.004900 0.006200 0.000000 0.009800 (I suppose it means the width of the L2_left is 0.0049 meter, which is 4.9mm. Am I right?) In the power trace file "gcc.ptrace": L2_left L2_right 1.44 7.37 (Is the power consumption of L2_left 1.44 watts or 1.44 milliwatts?) Thanks very much. Chunhua From ks4kk at cs.virginia.edu Mon Nov 17 13:25:31 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Mon, 17 Nov 2008 16:25:31 -0500 (EST) Subject: [Hotspot] Units in HotSpot 4.1 In-Reply-To: References: Message-ID: Hi Chunhua, You are correct - the units of length are in meters. The power numbers are in watts. -karthik On Mon, 17 Nov 2008, CHUNHUA YAO wrote: > Hi, > > I am new to HotSpot 4.1. I wonder what's the units of the data inside the example floorplan and power trance files. > > In the floorplan file "ev6.flp": > L2_left 0.004900 0.006200 0.000000 0.009800 > (I suppose it means the width of the L2_left is 0.0049 meter, which is 4.9mm. Am I right?) > > In the power trace file "gcc.ptrace": > L2_left L2_right > 1.44 7.37 > (Is the power consumption of L2_left 1.44 watts or 1.44 milliwatts?) > > Thanks very much. > > Chunhua > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > From augusto.vega at bsc.es Tue Nov 25 02:16:03 2008 From: augusto.vega at bsc.es (Augusto Vega) Date: Tue, 25 Nov 2008 11:16:03 +0100 Subject: [Hotspot] Modeling granularity in HotSpot Message-ID: <492BD063.6070901@bsc.es> Hi, I'm executing HotSpot for very long power traces and it takes a long time. In "Temperature-Aware Microarchitecture: Modeling and Implementation" (Skadron et al.) I found the following comment: "Although it is feasible to recompute temperatures every cycle, this is wasteful, since even at the fine granularity of architectural units, temperatures take at least 100K cycles to rise by 0.1?C." I would like to find the proof for this statement (some paper to cite) because I'll try to run HotSpot with a bigger step (e.g., 1M cycles). Where can I find it? Thank you in advance. -- Augusto Vega BSC-CNS Barcelona Supercomputing Center Centro Nacional de Supercomputaci?n Telf.: +34 93 413 79 46 e-mail: augusto.vega at bsc.es From wh6p at cms.mail.virginia.edu Tue Nov 25 07:38:54 2008 From: wh6p at cms.mail.virginia.edu (Wei Huang) Date: Tue, 25 Nov 2008 10:38:54 -0500 Subject: [Hotspot] Modeling granularity in HotSpot In-Reply-To: <492BD063.6070901@bsc.es> References: <492BD063.6070901@bsc.es> Message-ID: Hi Augusto, The claim in our ISCA paper was empirical, based on the observations of our hottest benchmark. However, you can find a more direct experiment for typical thermal packages in out recent Trans on Computers paper: Fig.9 in the paper shows transient response for different time scales for rather high power density pulses (2W/mm^2). If you want a more rigorous analysis, here is one: we found that the short-term (~ms or less) transient response in silicon is mainly determined by the thermal time constant of the die, which is about a few milliseconds. Assuming the hotspot's steady-state temperature is 50 degrees above ambient (ie. if your ambient is 40C, then hotspot temperature is around 90C), the transient temperature as a function of time can be approximately expressed as T(t)=50*(1-exp(-t/1ms)). For a 0.1C increase in T(t), if you solve the equation, the time needed is about 2us, which is in the ballpark of the claim in our ISCA paper. Of course, if you use a drastically different package, or your ambient temperature is different, or your applied power density is very high, you may want to re-do this quick and dirty analysis to find a proper range of the transient simulation granularity. Hope this is helpful, Regards, -Wei --On Tuesday, November 25, 2008 11:16 AM +0100 Augusto Vega wrote: > Hi, > > I'm executing HotSpot for very long power traces and it > takes a long time. In "Temperature-Aware > Microarchitecture: Modeling and Implementation" (Skadron > et al.) I found the following comment: > > "Although it is feasible to recompute temperatures every > cycle, this is wasteful, since even at the fine > granularity of architectural units, temperatures take at > least 100K cycles to rise by 0.1?C." > > I would like to find the proof for this statement (some > paper to cite) because I'll try to run HotSpot with a > bigger step (e.g., 1M cycles). Where can I find it? > > Thank you in advance. > > -- > Augusto Vega > > BSC-CNS > Barcelona Supercomputing Center > Centro Nacional de Supercomputaci?n > Telf.: +34 93 413 79 46 > e-mail: augusto.vega at bsc.es > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot From augusto.vega at bsc.es Tue Nov 25 10:44:28 2008 From: augusto.vega at bsc.es (Augusto Vega) Date: Tue, 25 Nov 2008 19:44:28 +0100 Subject: [Hotspot] HotSpot on PowerPC? Message-ID: <492C478C.4080700@bsc.es> Hi, I'm trying to compile HotSpot in a PowerPC 970 machine. It compiles ok, but it only executes fine if optimization level is "O0". For optimization levels greater than 0, I've a segmentation fault at the very end of the main() function. This is the output of Valgrind: ... inode_10 322.65 inode_11 321.50 ==29655== ==29655== Invalid read of size 4 ==29655== at 0x100018CC: main (hotspot.c:579) ==29655== Address 0x43300004 is not stack'd, malloc'd or (recently) free'd ==29655== ==29655== Process terminating with default action of signal 11 (SIGSEGV) ==29655== Access not within mapped region at address 0x43300004 ==29655== at 0x100018CC: main (hotspot.c:579) ==29655== ==29655== ERROR SUMMARY: 3 errors from 3 contexts (suppressed: 8 from 3) ==29655== malloc/free: in use at exit: 0 bytes in 0 blocks. ==29655== malloc/free: 25,151 allocs, 25,151 frees, 27,626,720 bytes allocated. ==29655== For counts of detected errors, rerun with: -v ==29655== All heap blocks were freed -- no leaks are possible. Segmentation fault This is the way I compiled it: gcc -O3 -DVERBOSE=1 -DMATHACCEL=0 -g -c xxx.c Has HotSpot been tested in PowerPC? Thank you in advance. -- Augusto Vega BSC-CNS Barcelona Supercomputing Center Centro Nacional de Supercomputaci?n Telf.: +34 93 413 79 46 e-mail: augusto.vega at bsc.es From ks4kk at cs.virginia.edu Tue Nov 25 11:46:33 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Tue, 25 Nov 2008 14:46:33 -0500 (EST) Subject: [Hotspot] HotSpot on PowerPC? In-Reply-To: <492C478C.4080700@bsc.es> References: <492C478C.4080700@bsc.es> Message-ID: Hello Augusto, Thank you for letting us know. We did test an earlier version of HotSpot on Power PC (during the addition of the MATHACCEL engine) and found no problems. Can you provide us with the following information: 1) The version of HotSpot you are using 2) The exact command line/input files used for the run 3) The Makefile parameters specific to your platform We will try to see if we can reproduce this error. Thanks, -karthik On Tue, 25 Nov 2008, Augusto Vega wrote: > Hi, > > I'm trying to compile HotSpot in a PowerPC 970 machine. > > It compiles ok, but it only executes fine if optimization level is "O0". > For optimization levels greater than 0, I've a segmentation fault at the > very end of the main() function. > > This is the output of Valgrind: > > ... > inode_10 322.65 > inode_11 321.50 > ==29655== > ==29655== Invalid read of size 4 > ==29655== at 0x100018CC: main (hotspot.c:579) > ==29655== Address 0x43300004 is not stack'd, malloc'd or (recently) > free'd > ==29655== > ==29655== Process terminating with default action of signal 11 > (SIGSEGV) > ==29655== Access not within mapped region at address 0x43300004 > ==29655== at 0x100018CC: main (hotspot.c:579) > ==29655== > ==29655== ERROR SUMMARY: 3 errors from 3 contexts (suppressed: 8 from > 3) > ==29655== malloc/free: in use at exit: 0 bytes in 0 blocks. > ==29655== malloc/free: 25,151 allocs, 25,151 frees, 27,626,720 bytes > allocated. > ==29655== For counts of detected errors, rerun with: -v > ==29655== All heap blocks were freed -- no leaks are possible. > Segmentation fault > > > This is the way I compiled it: > > gcc -O3 -DVERBOSE=1 -DMATHACCEL=0 -g -c xxx.c > > > Has HotSpot been tested in PowerPC? > > Thank you in advance. > > > -- > Augusto Vega > > BSC-CNS > Barcelona Supercomputing Center > Centro Nacional de Supercomputaci?n > Telf.: +34 93 413 79 46 > e-mail: augusto.vega at bsc.es > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot > > From fazal256 at yahoo.com Thu Nov 27 09:35:55 2008 From: fazal256 at yahoo.com (fazal hameed) Date: Thu, 27 Nov 2008 09:35:55 -0800 (PST) Subject: [Hotspot] HotSpot Intel math kernel library In-Reply-To: Message-ID: <327047.96204.qm@web54203.mail.re2.yahoo.com> Dear All, I am trying to compile Hotspot on Intel IBM thinkpad (Intel (R) Core (TM)2 T7400) with SUSE Linux 11.0 operating system and i am getting the error below. I think there is some incompatiblity with the Intel Library?. undefined reference to `mkl_serv_load_fun' undefined reference to `mkl_serv_load_dll' collect2: ld returned 1 exit status make: *** [hotspot] Error 1 I am new to Linux. Can anybody help me in this regard? gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c hotspot.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c RCutil.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_block.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_grid.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp_desc.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c npe.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c shape.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c util.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c wire.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -o hotspot hotspot.o temperature.o RCutil.o temperature_block.o temperature_grid.o flp.o flp_desc.o npe.o shape.o util.o wire.o -lmkl_lapack -lmkl_ia32 -lguide -lpthread -lm /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_fun' /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_dll' collect2: ld returned 1 exit status make: *** [hotspot] Error 1 regards, Fazal Hameed From fazal256 at yahoo.com Thu Nov 27 09:35:55 2008 From: fazal256 at yahoo.com (fazal hameed) Date: Thu, 27 Nov 2008 09:35:55 -0800 (PST) Subject: [Hotspot] HotSpot Intel math kernel library In-Reply-To: Message-ID: <327047.96204.qm@web54203.mail.re2.yahoo.com> Dear All, I am trying to compile Hotspot on Intel IBM thinkpad (Intel (R) Core (TM)2 T7400) with SUSE Linux 11.0 operating system and i am getting the error below. I think there is some incompatiblity with the Intel Library?. undefined reference to `mkl_serv_load_fun' undefined reference to `mkl_serv_load_dll' collect2: ld returned 1 exit status make: *** [hotspot] Error 1 I am new to Linux. Can anybody help me in this regard? gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c hotspot.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c RCutil.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_block.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_grid.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp_desc.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c npe.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c shape.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c util.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c wire.c gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -o hotspot hotspot.o temperature.o RCutil.o temperature_block.o temperature_grid.o flp.o flp_desc.o npe.o shape.o util.o wire.o -lmkl_lapack -lmkl_ia32 -lguide -lpthread -lm /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_fun' /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_dll' collect2: ld returned 1 exit status make: *** [hotspot] Error 1 regards, Fazal Hameed From wh6p at cms.mail.virginia.edu Thu Nov 27 10:54:00 2008 From: wh6p at cms.mail.virginia.edu (Wei Huang) Date: Thu, 27 Nov 2008 13:54:00 -0500 Subject: [Hotspot] HotSpot Intel math kernel library Message-ID: Hi, I am forwarding an earlier HotSpot mailing list email containing suggestions on solving a similar issue (thanks to my colleague Karthik). See below... Regards, -Wei ---------- Forwarded Message ---------- Date: Monday, September 29, 2008 4:16 PM -0400 From: Karthik Sankaranarayanan To: Wei Huang Subject: Re: Auotmatic reject (fwd) Hi Wei, Not sure what version of MKL he is using but this is an MKL problem. The LIBS flags need to be set from the MKL installation manual. The default flags in HotSpot are 32-bit flags from MKL 8.0.2. He needs the 64-bit flags. Following pages give information on what flags to use: http://www.intel.com/support/performancetools/libraries/mkl /linux/sb/CS-028423.htm http://www.intel.com/support/performancetools/sb/CS-028699. htm http://www.intel.com/software/products/mkl/docs/linux/WebHe lp/mkl_ug_linking_your_application/selecting_libraries_to_l ink.htm It appears that in the layered default model, following might work although I haven't tried it myself: -lmkl_lapack -lmkl -lguide -lm -lpthread -karthik On Mon, 29 Sep 2008, Wei Huang wrote: > Karthik, > > I've got this user subscribed to the mailing list. > > Would you mind have a look at his question during > compiling HotSpot? It seems there are some compatibility > issues between his OS and the library he used. > > Thanks, > -Wei > > ---------- Forwarded Message ---------- > Date: Monday, September 29, 2008 1:44 PM -0400 > From: Yan Zhang > To: hotspot-owner at mail.cs.virginia.edu > Subject: Auotmatic reject > > Hi there, > > I have subscribe the list but still being rejected. Just > want to post my question during installation. > > --------------------------------------------------------- > -- > --------------------------------------------------------- > -- ------------------------------------------------ > > After install INTEL's MKL, I have set up the configuration > like this: > --------------------------------------------------------- > -- -- > MATHACCEL = intel > INCDIR = /Program/mkl_intel/include > LIBDIR = /Program/mkl_intel/lib/32 > LIBS = -lm -lmkl_lapack -lmkl_ia32 -lguide > -lpthread EXTRAFLAGS = > --------------------------------------------------------- > -- --- > My Linux distribution is Ubuntu 7.04, running on Intel > Core2 Duo. After makefile, I got error: > --------------------------------------------------------- > -- --------------------------------- > make > gcc -O3 -I/Program/mkl_intel/include > -L/Program/mkl_intel/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -o > hotspot hotspot.o temperature.o RCutil.o > temperature_block.o temperature_grid.o flp.o flp_desc.o > npe.o shape.o util.o wire.o -lmkl_lapack -lmkl_ia32 > -lguide -lpthread -lm > /Program/mkl_intel/lib/32/libmkl_lapack.so: undefined > reference to `mkl_serv_load_fun' > /Program/mkl_intel/lib/32/libmkl_lapack.so: undefined > reference to `mkl_serv_load_dll' > collect2: ld returned 1 exit status > make: *** [hotspot] Error 1 > > --------------------------------------------------------- > -- -------------------------------- > I think the default configuration is for 32 bit. How > should I change the configuration for 64-bit? Many thanks. > Yan Zhang > young.chang at hotmail.com > ------------------------------------- > --------------------------------------------------------- > -- > --------------------------------------------------------- > -- ----------------------------------------------------- > > > > > __________________________________________________ > Want to do more with Windows Live? Learn ?0 hidden > secrets?from Jamie. Learn Now > > ---------- End Forwarded Message ---------- > > > > > > ---------- End Forwarded Message ---------- --On Wednesday, October 08, 2008 11:04 AM -0700 Sohaib Majzoub wrote: > Hi Everyone, > I am trying to compile Hotspot on Debian 64bit and i am > getting the error below. > It seems something incompatible with the Intel Library? > Appreciate any help, > Thanks > Sohaib > > gcc -O3 -I/intel/mkl/9.1.023/include > -L/intel/mkl/9.1.023/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -o > hotspot hotspot.o temperature.o RCutil.o > temperature_block.o temperature_grid.o flp.o flp_desc.o > npe.o shape.o util.o wire.o -lmkl_lapack -lmkl_ia32 > -lguide -lpthread -lm > /usr/bin/ld: skipping incompatible > /intel/mkl/9.1.023/lib/32/libmkl_lapack.so when searching > for -lmkl_lapack > /usr/bin/ld: skipping incompatible > /intel/mkl/9.1.023/lib/32/libmkl_lapack.a when searching > for -lmkl_lapack > /usr/bin/ld: cannot find -lmkl_lapack > collect2: ld returned 1 exit status > make: *** [hotspot] Error 1 > > > ---------- End Forwarded Message ---------- From ks4kk at cs.virginia.edu Fri Nov 28 08:24:09 2008 From: ks4kk at cs.virginia.edu (Karthik Sankaranarayanan) Date: Fri, 28 Nov 2008 11:24:09 -0500 (EST) Subject: [Hotspot] HotSpot Intel math kernel library In-Reply-To: <327047.96204.qm@web54203.mail.re2.yahoo.com> References: <327047.96204.qm@web54203.mail.re2.yahoo.com> Message-ID: Hi, This seems to be an Intel MKL issue. Core 2 is a 64-bit processor but the library being linked with here is a 32-bit one. Please refer to the MKL documentation to find the flags to link with the appropriate libraries. Thanks, -karthik On Thu, 27 Nov 2008, fazal hameed wrote: > Dear All, > I am trying to compile Hotspot on Intel IBM thinkpad (Intel (R) Core (TM)2 T7400) with SUSE Linux 11.0 operating system and i am getting the error below. I think there is some incompatiblity with the Intel Library?. > > undefined reference to `mkl_serv_load_fun' > undefined reference to `mkl_serv_load_dll' > collect2: ld returned 1 exit status > make: *** [hotspot] Error 1 > > I am new to Linux. Can anybody help me in this regard? > > > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c hotspot.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c RCutil.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_block.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c temperature_grid.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c flp_desc.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c npe.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c shape.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c util.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -c wire.c > gcc -O3 -I/opt/intel/mkl/10.1.0.015/include -L/opt/intel/mkl/10.1.0.015/lib/32 -DVERBOSE=1 -DMATHACCEL=1 -o hotspot hotspot.o temperature.o RCutil.o temperature_block.o temperature_grid.o flp.o flp_desc.o npe.o shape.o util.o wire.o -lmkl_lapack -lmkl_ia32 -lguide -lpthread -lm > /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_fun' > /opt/intel/mkl/10.1.0.015/lib/32/libmkl_lapack.so: undefined reference to `mkl_serv_load_dll' > collect2: ld returned 1 exit status > make: *** [hotspot] Error 1 > > > regards, > Fazal Hameed > > > > > _______________________________________________ > HotSpot mailing list > HotSpot at mail.cs.virginia.edu > http://www.cs.virginia.edu/mailman/listinfo/hotspot >