Hi,
Number of L2 requests should be equal to the number of L1 misses. So using L1 instruction and data misses you should be able to calculate L2 miss rate.
Thanks.
On Mon, Sep 7, 2009 at 2:40 PM, Edward Lee <edwl202@xxxxxxxxx> wrote:
Hi all,
I have a little confusion about the ruby stats regarding the L2 Cache accesses and correspondingly calculating the L2 Cache Miss rate. For example, below is a copy of the stats file from an FFT benchmark (based on MOESI_SMP_directory protocol).
As I checked earlier posts, total number of memory accesses is the sum of Insn requests and Data requests: (356241870 + 76057261) = 432299131. I am interested in calculating the L2 cache miss rate. So, by just dividing the L2_cache_total_misses: 138053 to total memory accesses gives me the miss rate?
I suspect this would be incorrect as inst. requests dominate this total number of memory accesses but thats rarely used in L2 cache requests. So, how can I isolate the L2 Cache accesses here? Can anybody help me on this?
Thanks,
Ed
----------------------- Selected parts of the stats file -------------------------------------
L1D_cache cache stats: L1D_cache_total_misses: 2117964 L1D_cache_total_demand_misses: 2117964
L1D_cache_total_prefetches: 0 L1D_cache_total_sw_prefetches: 0 L1D_cache_total_hw_prefetches: 0 L1D_cache_misses_per_transaction: 2.11796e+06 L1D_cache_misses_per_instruction: 0.00594529 L1D_cache_instructions_per_misses: 168.2
L1D_cache_request_type_LD: 97.3424% L1D_cache_request_type_ST: 2.46463% L1D_cache_request_type_ATOMIC: 0.192921%
L1D_cache_access_mode_type_SupervisorMode: 1987116 93.822% L1D_cache_access_mode_type_UserMode: 130848 6.17801%
L1D_cache_request_size: [binsize: log2 max: 8 count: 2117964 average: 2.99419 | standard deviation: 2.2293 | 0 9343 1749447 10013 349161 ]
L1I_cache cache stats: L1I_cache_total_misses: 485 L1I_cache_total_demand_misses: 485
L1I_cache_total_prefetches: 0 L1I_cache_total_sw_prefetches: 0 L1I_cache_total_hw_prefetches: 0 L1I_cache_misses_per_transaction: 485 L1I_cache_misses_per_instruction: 1.36143e-06 L1I_cache_instructions_per_misses: 734520
L1I_cache_request_type_IFETCH: 100%
L1I_cache_access_mode_type_UserMode: 485 100% L1I_cache_request_size: [binsize: log2 max: 4 count: 485 average: 4 | standard deviation: 0 | 0 0 0 485 ]
L2_cache cache stats: L2_cache_total_misses: 138053 L2_cache_total_demand_misses: 138053 L2_cache_total_prefetches: 0 L2_cache_total_sw_prefetches: 0 L2_cache_total_hw_prefetches: 0 L2_cache_misses_per_transaction: 138053
L2_cache_misses_per_instruction: 0.000387525 L2_cache_instructions_per_misses: 2580.48
L2_cache_request_type_LD: 69.8203% L2_cache_request_type_ST: 27.8929% L2_cache_request_type_ATOMIC: 2.06587%
L2_cache_request_type_IFETCH: 0.22093%
L2_cache_access_mode_type_SupervisorMode: 44311 32.0971% L2_cache_access_mode_type_UserMode: 93742 67.9029% L2_cache_request_size: [binsize: log2 max: 8 count: 138053 average: 7.00595 | standard deviation: 2.23386 | 0 6675 11351 5600 114427 ]
Simics Driver Transaction Stats ---------------------------------- Insn requests: 356241870 Data requests: 76057261
Simics Driver Transaction Results Stats ------------------------------------------
Fast path: 73296358 Request missed: 2118449
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.
-- Regards,
Sharad Bade Graduate Student, Computer Engineering North Carolina State University, Raleigh, NC, USA.
|