URL
https://opencores.org/ocsvn/openrisc_2011-10-31/openrisc_2011-10-31/trunk
Subversion Repositories openrisc_2011-10-31
[/] [openrisc/] [trunk/] [or1ksim/] [cpu/] [common/] [stats.c] - Rev 30
Go to most recent revision | Compare with Previous | Blame | View Log
/* stats.c -- Various statistics about instruction scheduling etc. Copyright (C) 1999 Damjan Lampret, lampret@opencores.org Copyright (C) 2008 Embecosm Limited Contributor Jeremy Bennett <jeremy.bennett@embecosm.com> This file is part of Or1ksim, the OpenRISC 1000 Architectural Simulator. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see <http://www.gnu.org/licenses/>. */ /* This program is commented throughout in a fashion suitable for processing with Doxygen. */ /* Autoconf and/or portability configuration */ #include "config.h" #include "port.h" /* Package includes */ #include "stats.h" #include "sim-config.h" #include "icache-model.h" #include "spr-defs.h" #include "execute.h" #define DSTATS_LEN 3000 #define SSTATS_LEN 300 #define FSTATS_LEN 200 /* Used by safe division - increment divisor by one if it is zero */ #define SD(X) (X != 0 ? X : 1) struct branchstat { int taken; int nottaken; int forward; int backward; }; /*! @see also enum insn_type in abstract.h */ static const char func_unit_str[30][30] = { "unknown", "exception", "arith", "shift", "compare", "branch", "jump", "load", "store", "movimm", "move", "extend", "nop", "mac" }; struct dstats_entry { int insn1; int insn2; int cnt_dynamic; int depend; }; /*!< double stats */ struct sstats_entry { int insn; int cnt_dynamic; }; /*!< single stats */ struct fstats_entry { enum insn_type insn1; enum insn_type insn2; int cnt_dynamic; int depend; }; /*!< functional units stats */ /* Globally visible statistics data. Renamed mstats to or1k_mstats because Mac OS X has a lib function called mstats */ struct mstats_entry or1k_mstats = { 0 }; /*!< misc units stats */ struct cachestats_entry ic_stats = { 0 }; /*!< instruction cache stats */ struct cachestats_entry dc_stats = { 0 }; /*!< data cache stats */ struct immustats_entry immu_stats = { 0 }; /*!< insn MMU stats */ struct dmmustats_entry dmmu_stats = { 0 }; /*!< data MMU stats */ struct raw_stats raw_stats; /*!< RAW hazard stats */ /* Statistics data strutures used just here */ static struct dstats_entry dstats[DSTATS_LEN]; /*!< dependency stats */ static struct sstats_entry sstats[SSTATS_LEN]; /*!< single stats */ static struct fstats_entry fstats[FSTATS_LEN]; /*!< func units stats */ void addsstats (int item, int cnt_dynamic) { int i = 0; while (sstats[i].insn != item && sstats[i].insn >= 0 && i < SSTATS_LEN) i++; if (i >= SSTATS_LEN - 1) return; if (sstats[i].insn >= 0) { sstats[i].cnt_dynamic += cnt_dynamic; } else { sstats[i].insn = item; sstats[i].cnt_dynamic = cnt_dynamic; } } void adddstats (int item1, int item2, int cnt_dynamic, int depend) { int i = 0; while ((dstats[i].insn1 != item1 || dstats[i].insn2 != item2) && (i < DSTATS_LEN) && dstats[i].insn1 >= 0) i++; if (i >= DSTATS_LEN - 1) return; if (dstats[i].insn1 >= 0) { dstats[i].cnt_dynamic += cnt_dynamic; dstats[i].depend += depend; } else { dstats[i].insn1 = item1; dstats[i].insn2 = item2; dstats[i].cnt_dynamic = cnt_dynamic; dstats[i].depend = depend; } } void addfstats (enum insn_type item1, enum insn_type item2, int cnt_dynamic, int depend) { int i = 0; while (((fstats[i].insn1 != item1) || (fstats[i].insn2 != item2)) && (fstats[i].insn1 != it_unknown) && (i < FSTATS_LEN)) i++; if (i >= FSTATS_LEN - 1) return; if ((fstats[i].insn1 == item1) && (fstats[i].insn2 == item2)) { fstats[i].cnt_dynamic += cnt_dynamic; fstats[i].depend += depend; } else { fstats[i].insn1 = item1; fstats[i].insn2 = item2; fstats[i].cnt_dynamic = cnt_dynamic; fstats[i].depend = depend; } } void initstats () { int i; memset (sstats, 0, sizeof (sstats)); for (i = 0; i < SSTATS_LEN; i++) sstats[i].insn = -1; memset (dstats, 0, sizeof (dstats)); for (i = 0; i < DSTATS_LEN; i++) dstats[i].insn1 = dstats[i].insn2 = -1; memset (fstats, 0, sizeof (fstats)); memset (&or1k_mstats, 0, sizeof (or1k_mstats)); memset (&ic_stats, 0, sizeof (ic_stats)); memset (&dc_stats, 0, sizeof (dc_stats)); memset (&raw_stats, 0, sizeof (raw_stats)); } static void printotherstats (int which) { PRINTF ("\n"); if (config.bpb.enabled) { struct branchstat bf; struct branchstat bnf; long bf_all, bnf_all; bf.taken = or1k_mstats.bf[1][0] + or1k_mstats.bf[1][1]; bf.nottaken = or1k_mstats.bf[0][0] + or1k_mstats.bf[0][1]; bf.forward = or1k_mstats.bf[0][1] + or1k_mstats.bf[1][1]; bf.backward = or1k_mstats.bf[0][0] + or1k_mstats.bf[1][0]; bf_all = bf.forward + bf.backward; bnf.taken = or1k_mstats.bnf[1][0] + or1k_mstats.bf[1][1]; bnf.nottaken = or1k_mstats.bnf[0][0] + or1k_mstats.bf[0][1]; bnf.forward = or1k_mstats.bnf[0][1] + or1k_mstats.bf[1][1]; bnf.backward = or1k_mstats.bnf[0][0] + or1k_mstats.bf[1][0]; bnf_all = bnf.forward + bnf.backward; PRINTF ("bnf: %d (%ld%%) taken,", bf.taken, (bf.taken * 100) / SD (bf_all)); PRINTF (" %d (%ld%%) not taken,", bf.nottaken, (bf.nottaken * 100) / SD (bf_all)); PRINTF (" %d (%ld%%) forward,", bf.forward, (bf.forward * 100) / SD (bf_all)); PRINTF (" %d (%ld%%) backward\n", bf.backward, (bf.backward * 100) / SD (bf_all)); PRINTF ("bf: %d (%ld%%) taken,", bnf.taken, (bnf.taken * 100) / SD (bnf_all)); PRINTF (" %d (%ld%%) not taken,", bnf.nottaken, (bnf.nottaken * 100) / SD (bnf_all)); PRINTF (" %d (%ld%%) forward,", bnf.forward, (bnf.forward * 100) / SD (bnf_all)); PRINTF (" %d (%ld%%) backward\n", bnf.backward, (bnf.backward * 100) / SD (bnf_all)); PRINTF ("StaticBP bnf(%s): correct %ld%%\n", config.bpb.sbp_bnf_fwd ? "forward" : "backward", (or1k_mstats.bnf[0][config.bpb.sbp_bnf_fwd] * 100) / SD (bnf_all)); PRINTF ("StaticBP bf(%s): correct %ld%%\n", config.bpb.sbp_bf_fwd ? "forward" : "backward", (or1k_mstats.bnf[1][config.bpb.sbp_bf_fwd] * 100) / SD (bf_all)); PRINTF ("BPB: hit %d (correct %d%%), miss %d\n", or1k_mstats.bpb.hit, (or1k_mstats.bpb.correct * 100) / SD (or1k_mstats.bpb.hit), or1k_mstats.bpb.miss); } else PRINTF ("BPB simulation disabled. Enable it to see BPB analysis\n"); if (config.bpb.btic) { PRINTF ("BTIC: hit %d(%d%%), miss %d\n", or1k_mstats.btic.hit, (or1k_mstats.btic.hit * 100) / SD (or1k_mstats.btic.hit + or1k_mstats.btic.miss), or1k_mstats.btic.miss); } else PRINTF ("BTIC simulation disabled. Enabled it to see BTIC analysis\n"); if ((NULL != ic_state) && ic_state->enabled) { PRINTF ("IC read: hit %d(%d%%), miss %d\n", ic_stats.readhit, (ic_stats.readhit * 100) / SD (ic_stats.readhit + ic_stats.readmiss), ic_stats.readmiss); } else PRINTF ("No ICache. Enable it to see IC results.\n"); if (config.dc.enabled) { PRINTF ("DC read: hit %d(%d%%), miss %d\n", dc_stats.readhit, (dc_stats.readhit * 100) / SD (dc_stats.readhit + dc_stats.readmiss), dc_stats.readmiss); PRINTF ("DC write: hit %d(%d%%), miss %d\n", dc_stats.writehit, (dc_stats.writehit * 100) / SD (dc_stats.writehit + dc_stats.writemiss), dc_stats.writemiss); } else PRINTF ("No DCache. Enable it to see DC results.\n"); if (cpu_state.sprs[SPR_UPR] & SPR_UPR_IMP) { PRINTF ("IMMU read: hit %d(%d%%), miss %d\n", immu_stats.fetch_tlbhit, (immu_stats.fetch_tlbhit * 100) / SD (immu_stats.fetch_tlbhit + immu_stats.fetch_tlbmiss), immu_stats.fetch_tlbmiss); } else PRINTF ("No IMMU. Set UPR[IMP]\n"); if (cpu_state.sprs[SPR_UPR] & SPR_UPR_DMP) { PRINTF ("DMMU read: hit %d(%d%%), miss %d\n", dmmu_stats.loads_tlbhit, (dmmu_stats.loads_tlbhit * 100) / SD (dmmu_stats.loads_tlbhit + dmmu_stats.loads_tlbmiss), dmmu_stats.loads_tlbmiss); } else PRINTF ("No DMMU. Set UPR[DMP]\n"); PRINTF ("Additional LOAD CYCLES: %u STORE CYCLES: %u\n", runtime.sim.loadcycles, runtime.sim.storecycles); } void printstats (int which) { int i, all = 0, dependall = 0; if (which > 1 && which <= 5 && !config.cpu.dependstats) { PRINTF ("Hazard analysis disabled. Enable it to see analysis results.\n"); return; } switch (which) { case 1: PRINTF ("stats 1: Misc stats\n"); printotherstats (which); break; case 2: PRINTF ("stats 2: Instruction usage\n"); for (i = 0; i < SSTATS_LEN; i++) all += sstats[i].cnt_dynamic; for (i = 0; i < SSTATS_LEN; i++) if (sstats[i].cnt_dynamic) PRINTF (" %-15s used %6dx (%5.1f%%)\n", insn_name (sstats[i].insn), sstats[i].cnt_dynamic, (sstats[i].cnt_dynamic * 100.) / SD (all)); PRINTF ("%d instructions (dynamic, single stats)\n", all); break; case 3: PRINTF ("stats 3: Instruction dependencies\n"); for (i = 0; i < DSTATS_LEN; i++) { all += dstats[i].cnt_dynamic; dependall += dstats[i].depend; } for (i = 0; i < DSTATS_LEN; i++) if (dstats[i].cnt_dynamic) { char temp[100]; sprintf (temp, "%s, %s ", insn_name (dstats[i].insn1), insn_name (dstats[i].insn2)); PRINTF (" %-30s %6dx (%5.1f%%)", temp, dstats[i].cnt_dynamic, (dstats[i].cnt_dynamic * 100.) / SD (all)); PRINTF (" depend: %5.1f%%\n", (dstats[i].depend * 100.) / dstats[i].cnt_dynamic); } PRINTF ("%d instructions (dynamic, dependency stats) depend: %d%%\n", all, (dependall * 100) / SD (all)); break; case 4: PRINTF ("stats 4: Functional units dependencies\n"); for (i = 0; i < FSTATS_LEN; i++) { all += fstats[i].cnt_dynamic; dependall += fstats[i].depend; } for (i = 0; i < FSTATS_LEN; i++) if (fstats[i].cnt_dynamic) { char temp[100]; sprintf (temp, "%s, %s", func_unit_str[fstats[i].insn1], func_unit_str[fstats[i].insn2]); PRINTF (" %-30s %6dx (%5.1f%%)", temp, fstats[i].cnt_dynamic, (fstats[i].cnt_dynamic * 100.) / SD (all)); PRINTF (" depend: %5.1f%%\n", (fstats[i].depend * 100.) / fstats[i].cnt_dynamic); } PRINTF ("%d instructions (dynamic, functional units stats) depend: %d%%\n\n", all, (dependall * 100) / SD (all)); break; case 5: PRINTF ("stats 5: Raw register usage over time\n"); #if RAW_RANGE_STATS for (i = 0; (i < RAW_RANGE); i++) PRINTF (" Register set and reused in %d. cycle: %d cases\n", i, raw_stats.range[i]); #endif break; case 6: if (config.cpu.sbuf_len) { extern int sbuf_total_cyc, sbuf_wait_cyc; PRINTF ("stats 6: Store buffer analysis\n"); PRINTF ("Using store buffer of length %i.\n", config.cpu.sbuf_len); PRINTF ("Number of total memory store cycles: %i/%lli\n", sbuf_total_cyc, runtime.sim.cycles + sbuf_total_cyc - sbuf_wait_cyc); PRINTF ("Number of cycles waiting for memory stores: %i\n", sbuf_wait_cyc); PRINTF ("Number of memory cycles spared: %i\n", sbuf_total_cyc - sbuf_wait_cyc); PRINTF ("Store speedup %3.2f%%, total speedup %3.2f%%\n", 100. * (sbuf_total_cyc - sbuf_wait_cyc) / sbuf_total_cyc, 100. * (sbuf_total_cyc - sbuf_wait_cyc) / (runtime.sim.cycles + sbuf_total_cyc - sbuf_wait_cyc)); } else PRINTF ("Store buffer analysis disabled. Enable it to see analysis results.\n"); break; default: PRINTF ("Please specify a stats group (1-6).\n"); break; } }
Go to most recent revision | Compare with Previous | Blame | View Log