# @(#)M		1.4  src/bos/usr/sbin/perf/pmapi/libpmapi/POWER3.dms, pmapi, bos720 3/30/07 03:05:42
# IBM_PROLOG_BEGIN_TAG 
# This is an automatically generated prolog. 
#  
# bos720 src/bos/usr/sbin/perf/pmapi/libpmapi/POWER3.dms 1.4 
#  
# Licensed Materials - Property of IBM 
#  
# COPYRIGHT International Business Machines Corp. 2005,2007 
# All Rights Reserved 
#  
# US Government Users Restricted Rights - Use, duplication or 
# disclosure restricted by GSA ADP Schedule Contract with IBM Corp. 
#  
# IBM_PROLOG_END_TAG 

# Global variables

CACHE_LINE_SIZE = 128;
MEM_LINE_SIZE   = 128;

# Derived metric descriptions

@PMD_UTI_RATE;Utilization rate;%
@@(PM_CYC * 100 / proc_freq) / total_time
@@@group=performance

@PMD_PRC_TLB_MISS;% TLB misses per cycle;%
@@100 * (PM_TLB_MISS / PM_CYC)
@@@group=performance

@PMD_LD_PER_TLB;Number of loads per TLB miss
@@PM_LD_CMPL / PM_TLB_MISS
@@@group=performance

@PMD_LD_ST_PER_TLB;Number of load/store per TLB miss
@@(PM_LD_CMPL + PM_ST_CMPL) / PM_TLB_MISS
@@@group=performance

@PMD_LD_ST;Total load and store operations;M
@@(PM_ST_CMPL + PM_LD_CMPL) * 0.000001
@@@group=performance

@PMD_INST_PER_LD_ST;Instructions per load/store
@@PM_INST_CMPL /  (PM_ST_CMPL + PM_LD_CMPL)
@@@group=performance

@PMD_MIPS;MIPS;MIPS
@@(0.000001 * PM_INST_CMPL) / total_time
@@@group=performance

@PMD_INST_PER_CYC;Instructions per cycle 
@@PM_INST_CMPL / PM_CYC
@@@group=performance

@PMD_HW_FP_PER_CYC;HW floating point instructions per Cycle
@@(PM_FPU0_CMPL + PM_FPU1_CMPL) / PM_CYC
@@@group=performance

@PMD_HW_FP_PER_UTIME;HW floating point instructions / user time;M HWflops/s
@@(PM_FPU0_CMPL + PM_FPU1_CMPL) / (PM_CYC / proc_freq)
@@@group=performance

@PMD_HW_FP_RATE;HW floating point rate (HW Flops / WCT);M HWflops/s
@@0.000001 * (PM_FPU0_CMPL + PM_FPU1_CMPL) / total_time
@@@group=performance

@PMD_FLIPS;Total floating point instructions + FMAs (flips);Mflip
@@(PM_FPU0_CMPL + PM_FPU1_CMPL + PM_FPU_FMA) * 0.000001
@@@group=performance

@PMD_FLIPS_WCT;Flip rate (flips / WCT);Mflip/s
@@((PM_FPU0_CMPL + PM_FPU1_CMPL + PM_FPU_FMA) * 0.000001) / total_time
@@@group=performance

@PMD_FLIPS_UTIME;Flips / user time;Mflip/s
@@((PM_FPU0_CMPL + PM_FPU1_CMPL + PM_FPU_FMA) * 0.000001) / (PM_CYC / proc_freq)
@@@group=performance

@PMD_COMP;Computation intensity
@@(PM_FPU0_CMPL + PM_FPU1_CMPL + PM_FPU_FMA) / (PM_LD_CMPL + PM_ST_CMPL)
@@@group=performance

@PMD_FMA_PRC;FMA percentage;%
@@(2 * 100 * PM_FPU_FMA) / (PM_FPU0_CMPL + PM_FPU1_CMPL + PM_FPU_FMA)
@@@group=performance

@PMD_TLB_EST_LAT;Estimated latency from TLB miss;s
@@HPM_TLB_LATENCY * PM_TLB_MISS / proc_freq
@@@group=performance

@PMD_LD_DISP_PER_TLB;Number of loads dispatched per TLB miss
@@PM_LD_DISP / PM_TLB_MISS
@@@group=performance

@PMD_LD_ST_DISP_PER_TLB;Number of load/store dispatched per TLB miss
@@(PM_LD_DISP + PM_ST_DISP) / PM_TLB_MISS
@@@group=performance

@PMD_DATA_L2;Total L2 data cache accesses;M
@@(PM_LD_MISS_L1 + PM_ST_MISS_L1) * 0.000001
@@@group=performance

@PMD_PRC_L2_ACCESS;% accesses from L2 per cycle;%
@@100 * (PM_LD_MISS_L1 + PM_ST_MISS_L1) / PM_CYC
@@@group=performance

@PMD_L2_TRAF;L2 traffic;MBytes
@@(PM_LD_MISS_L1 + PM_ST_MISS_L1) * CACHE_LINE_SIZE / (1024 * 1024)
@@@group=performance

@PMD_L2_BDW;L2 bandwidth per processor;MBytes/s
@@(PM_LD_MISS_L1 + PM_ST_MISS_L1) * CACHE_LINE_SIZE / (1024 * 1024) / total_time
@@@group=performance

@PMD_LD_ST_DISP;Total load and store dispatched operations;M
@@(PM_ST_DISP + PM_LD_DISP) * 0.000001
@@@group=performance

@PMD_LD_DISP_PER_LD_MISS;Number of loads dispatched per load miss
@@PM_LD_DISP / PM_LD_MISS_L1
@@@group=performance

@PMD_ST_DISP_PER_ST_MISS;Number of stores dispatched per store miss
@@PM_ST_DISP / PM_ST_MISS_L1
@@@group=performance

@PMD_LD_ST_DISP_PER_L1_MISS;Number of load/stores dispatched per L1 miss
@@(PM_ST_DISP + PM_LD_DISP) / (PM_ST_MISS_L1 + PM_LD_MISS_L1)
@@@group=performance

@PMD_L1_HIT_RATE_LDST_DISP;L1 cache hit rate (versus load/store dispatched);%
@@100 * (1 - ((PM_LD_MISS_L1 + PM_ST_MISS_L1) / (PM_LD_DISP + PM_ST_DISP)))
@@@group=performance

@PMD_PRC_LSU_IDLE;% Cycles LSU is idle ;%
@@(PM_LSU_IDLE * 100) / PM_CYC
@@@group=performance

@PMD_LD_DISP_PER_LD_MISS_L2HIT;Number of loads dispatched per load miss (L2 hit)
@@PM_LD_DISP / PM_LD_MISS_L2HIT
@@@group=performance

@PMD_ST_DISP_PER_L2_ST_MISS;Number of stores dispatched per L2 store miss
@@PM_ST_DISP / PM_ST_MISS_L2
@@@group=performance

@PMD_LD_DISP_PER_L2_LD_MISS;Number of loads dispatched per L2 load miss
@@PM_LD_DISP / PM_BIU_LD_NORTRY
@@@group=performance

@PMD_LD_ST_DISP_PER_L2_MISS;Number of load/stores dispatched per L2 miss
@@(PM_LD_DISP + PM_ST_DISP) / PM_BIU_LD_NORTRY
@@@group=performance

@PMD_MEM_TRAF;Memory traffic;MBytes
@@(PM_BIU_LD_NORTRY + PM_BIU_ST_NORTRY) * CACHE_LINE_SIZE / (1024 * 1024)
@@@group=performance

@PMD_MEM_BDW;Memory bandwidth per processor ;MBytes/s
@@(PM_BIU_LD_NORTRY + PM_BIU_ST_NORTRY) * CACHE_LINE_SIZE / (1024 * 1024) / total_time
@@@group=performance

@PMD_PRC_0INST_CMPL;% of cycles with 0 instructions completed;%
@@(PM_0INST_CMPL * 100) / PM_CYC
@@@group=performance

@PMD_FX;Total Fixed point operations;M
@@(PM_FXU0_PROD_RESULT + PM_FXU1_PROD_RESULT + PM_FXU2_PROD_RESULT) * 0.000001
@@@group=performance

@PMD_FX_PER_CYC;Fixed point operations per Cycle
@@(PM_FXU0_PROD_RESULT + PM_FXU1_PROD_RESULT + PM_FXU2_PROD_RESULT)  / PM_CYC
@@@group=performance

@PMD_FX_PER_LDST;Fixed point operations per load/stores
@@(PM_FXU0_PROD_RESULT + PM_FXU1_PROD_RESULT + PM_FXU2_PROD_RESULT)  / (PM_LD_CMPL + PM_ST_CMPL)
@@@group=performance

@PMD_MBR_PRC;Branches mispredicated percentage;%
@@(PM_BR_MPRED_GC * 100) / PM_BR_PRED
@@@group=performance

@PMD_INST_PER_IC_MISS;Instructions per I Cache Miss
@@PM_INST_CMPL / PM_IC_MISS
@@@group=performance

@PMD_PRC_INST_DISP_CMPL;% Instructions dispatched that completed;%
@@PM_INST_CMPL * 100 / PM_INST_DISP
@@@group=performance

@PMD_ST_L1_MISS_PER_L2_ST_MISS;L1 store miss per L2 store miss;%
@@PM_ST_MISS_L1 / PM_ST_MISS_L2
@@@group=performance

@PMD_SNOOP_HIT_PER_SNOOP;Snoop hit ratio
@@(PM_SNOOP_L2HIT * 100) / PM_SNOOP
@@@group=performance