dry@pcrat.UUCP (Dhrystone) (07/07/88)
07/07/88 Here comes the Dhrystone 2.1 benchmark sources, in three parts: dry2shar.1 (Stuff I added) dry2shar.2 (Reinhold's C version) dry2shar.3 (Reinhold's Pascal version) Look in newsgroup "comp.arch" for results to follow. #! /bin/sh # This is a shell archive, meaning: # 1. Remove everything above the #! /bin/sh line. # 2. Save the resulting text in a file. # 3. Execute the file with /bin/sh (not csh) to create: # README.RER # clarify.doc # Makefile # submit.frm # This archive created: Wed Jul 6 16:51:15 1988 export PATH; PATH=/bin:/usr/bin:$PATH if test -f 'README.RER' then echo shar: "will not over-write existing file 'README.RER'" else sed 's/^X//' << \SHAR_EOF > 'README.RER' XHere is Reinhold Weicker's DHRYSTONE 2.1 benchmark, slightly modified Xby me for instrumentation purposes only. This is an integer CPU Xbenchmark. The differences between 2.0 and 2.1 are minor, and XI believe that 2.0 and 2.1 results are comparable. X XI've enclosed a new submission form (note new address for mailings). XPlease deluge this mailbox (..!uunet!pcrat!dry2) with your results. XI'll summarize and repost when the dust clears. Please do not Xassume that I will pull posted results off of the net (I won't, its Xtoo much work). X XI've attempted to include a Makefile for UNIX and Microsoft C (with ndmake). XPay particular attention to the HZ parameter, even though your power may Xbe 50 or 60 hertz, your computer may not be. You may have to ask someone, Xread the manual, or check: X /usr/include/sys/param.h X /usr/include/limits.h (CLK_TCK==HZ) Xfor this information. X XThere are two versions to run, one with register variables, and one Xwithout. Please let the benchmark run for 30,000 loops on sixteen Xbit machines, and for much longer (a minute or two) on faster machines. XPlease note that "time(2)" has a resolution of 1 second, and may give Xvariable results. No matter how time is measured, a sanity check with Xa stopwatch is prudent. We've run into systems that lie about time, Xand there is always the configuration error problem. When it comes Xto time measurement on UNIX, there is no widely adhered to standard. X XFor segmented architectures, it is appropriate to submit results for Xall memory models, as shown below. X XThe CODESIZE information is new. On UNIX, the size command may Xbe used to get this info: X size dhry_1.o dhry_2.o Xand then add the first number on each line together. MS-DOS provides Xno such utility. I think you have to rummage around the object Xcode listing. We are only concerned with the actual object module code Xsize, and not the size of the loaded program and libraries, or their file Xsizes. I've only included one size parameter; use the NOREG or the REG Xsize, whichever is smaller. X XHere's a sample submission of results: X XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: IBM XMODEL: PC/AT XPROC: 80286 XCLOCK: 8 XOS: Venix XOVERSION: SVr2.3 XCOMPILER: AT&T cc XCVERSION: 11/8/84 XOPTIONS: -O XNOREG: 1450 XREG: 1450 XNOTES: Small Model XDATE: 03/04/88 XSUBMITTER: pcrat!rick (Rick Richardson) XCODESIZE: 1901 XMAILTO: uunet!pcrat!dry2 X XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: IBM XMODEL: PC/AT XPROC: 80286 XCLOCK: 8 XOS: Venix XOVERSION: SVr2.3 XCOMPILER: AT&T cc XCVERSION: 11/8/84 XOPTIONS: -O -Ml XNOREG: 1043 XREG: 1043 XNOTES: Large Model XDATE: 03/04/88 XSUBMITTER: pcrat!rick (Rick Richardson) XCODESIZE: 2403 XMAILTO: uunet!pcrat!dry2 X XThe program that processes submission forms is rather dumb. Please Xdo not change the order, add or removes lines in the form. If your XNOTES are longer than the space provided, then they are too long for Xthe summary. Keep it terse, please. X XA form consists of all lines between: X DHRYSTONE 2.1 BENCHMARK REPORTING FORM Xand X MAILTO: uunet!pcrat!dry2 Xboth lines must be present for the form to be processed. If Xa field does not apply or is not known, leave it blank. The fields Xare: X MANUF: Computer manufacturer, e.g. AT&T, IBM X MODEL: Model number of computer X PROC: If a microprocessor CPU, the part number, e.g. 68030 X CLOCK: Clock in Mhz, if known. Numeric only, e.g. 16.67 X OS: Operating system, e.g. UNIX X OVERSION: OS version, e.g. SVR3 X COMPILER: Compiler name, e.g. cc, Microsoft, or Green Hills X CVERSION: Compiler version, e.g. 5.10 X OPTIONS: Relevant compiler options, e.g. -O3 X NOREG: Dhrystones/second, no register attribute X REG: Dhrystones/second, with register attribute X NOTES: Additional, terse comments on one line X DATE: Date of test, US format MM/DD/YY X CODESIZE: ".text" size of dhry_1.o plus dhry_2.o. Do not X indicate the "a.out" or ".exe" size. One number only, X in bytes. Leave blank if unknown. File sizes are X inappropriate. X SUBMITTER: uucp or domain address (full name) X X-- X Rick Richardson, President, PC Research, Inc. X X(201) 542-3734 (voice, nights) OR (201) 389-8963 (voice, days) Xuunet!pcrat!rick (UUCP) rick%pcrat.uucp@uunet.uu.net (INTERNET) X uunet!pcrat!dry2 (Dhrystone submission forms only) SHAR_EOF fi if test -f 'clarify.doc' then echo shar: "will not over-write existing file 'clarify.doc'" else sed 's/^X//' << \SHAR_EOF > 'clarify.doc' XCLARIFICATION XThere seems to have been a great deal of confusion over what this Xbenchmark measures, and how to use these results. Let me try to clarify Xthis: X X 1) DHRYSTONE is a measure of processor+compiler efficiency in X executing a 'typical' program. The 'typical' program was X designed by measuring statistics on a great number of X 'real' programs. The 'typical' program was then written X by Reinhold P. Weicker using these statistics. The X program is balanced according to statement type, as well X as data type. X X 2) DHRYSTONE does not use floating point. Typical programs don't. X X 3) DHRYSTONE does not do I/O. Typical programs do, but then X we'd have a whole can of worms opened up. X X 4) DHRYSTONE does not contain much code that can be optimized X by vector processors. That is why a CRAY doesn't look real X fast, they weren't built to do this sort of computing. X X 5) DHRYSTONE does not measure OS performance, as it avoids X calling the O.S. The O.S. is indicated in the results only X to help in identifying the compiler technology. X X 6) DHRYSTONE is not perfect, but is a hell of a lot better than X the "sieve", or "SI". X X 7) DHRYSTONE gives results in dhrystones/second. Bigger X numbers are better. As a baseline, the original IBM PC X gives around 300-400 dhrystones/second with a good compiler. X The fastest machines today are approaching 100,000. X XIf somebody asked me to pick out the best machine for the money, I Xwouldn't look at just the results of DHRYSTONE. I'd probably: X X 1) Run DHRYSTONE to get a feel for the compiler+processor X speed. X 2) Run any number of benchmarks to check disk I/O bandwidth, X using both sequential and random read/writes. X 3) Run a multitasking benchmark to check multi-user response X time. Typically, these benchmarks run several types of X programs such as editors, shell scripts, sorts, compiles, X and plot the results against the number of simulated users. X 4) If appropriate for the intended use, run something like X WHETSTONE, to determine floating point performance. X 5) If appropriate for intended use, run some programs which do X vector and matrix computations. X 6) Figure out what the box will: X - cost to buy X - cost to operate and maintain X - be worth when it is sold X - be worth if the manufacturer goes out of business X 7) Having done the above, I probably have a hand-full of X machines which meet my price/performance requirements. X Now, I find out if the applications programs I'd like X to use will run on any of these machines. I also find X out how much interest people have in writing new software X for the machine, and look carefully at the migration path X I will have to take when I reach the (inevitable) limits X of the machine. X XTo summarize, DHRYSTONES by themselves are not anything more than Xa way to win free beers when arguing 'Box-A versus Box-B' religion. XThey do provide insight into Box-A/Compiler-A versus Box-A/Compiler-B Xcomparisons. X XA SPECIAL THANKS XI didn't write the DHRYSTONE benchmark. Rheinhold Weicker did. He has Xcertainly provided us with a useful tool for benchmarking, and is Xto be congratulated. X X Rick Richardson X PC Research, Inc. X (201) 389-8963 (9-17 EST) X (201) 542-3734 (7-9,17-24 EST) X ...!uunet!pcrat!rick (normal mail) X ...!uunet!pcrat!dry2 (results only) X X X SHAR_EOF fi if test -f 'Makefile' then echo shar: "will not over-write existing file 'Makefile'" else sed 's/^X//' << \SHAR_EOF > 'Makefile' X# X# Adjust for your system! X# X# Common options for generic UNIX and Microsoft C (under DOS) X# are listed here. You can change them by switching the order, X# placing the ones you want last. Pay particular attention to X# the HZ parameter, which may or may not be listed in some X# header file on your system, such as <sys/param.h> or <limits.h> X# (as CLK_TCK). Even if it is listed, it may be incorrect. X# Also, some operating systems (notably some (all?) versions X# of Microport UNIX) lie about the time. Sanity check with a X# stopwatch. X# X# For Microsoft C under DOS, you need a real make, not MSC make, X# to run this Makefile. The public domain "ndmake" will suffice. X# XCC= cl # C compiler name goes here (MSC) XCC= cc # C compiler name goes here (UNIX) X XPROGS= msc # Programs to build (MSC) XPROGS= unix # Programs to build (UNIX) X XTIME_FUNC= -DMSC_CLOCK # Use Microsoft clock() for measurement XTIME_FUNC= -DTIME # Use time(2) for measurement XTIME_FUNC= -DTIMES # Use times(2) for measurement XHZ= 50 # Frequency of times(2) clock ticks XHZ= 60 # Frequency of times(2) clock ticks XHZ= 100 # Frequency of times(2) clock ticks XHZ= 1 # Give bogus result unless changed! X XSTRUCTASSIGN= -DNOSTRUCTASSIGN # Compiler cannot assign structs XSTRUCTASSIGN= # Compiler can assign structs X XENUMS= -DNOENUMS # Compiler doesn't have enum type XENUMS= # Compiler does have enum type X XOPTIMIZE= -Ox -G2 # Optimization Level (MSC, 80286) XOPTIMIZE= -O # Optimization Level (generic UNIX) X XLFLAGS= #Loader Flags X XCFLAGS= $(OPTIMIZE) $(TIME_FUNC) -DHZ=$(HZ) $(ENUMS) $(STRUCTASSIGN) $(CFL) X X# X# You shouldn't need to touch the rest X# XSRC= dhry_1.c dhry_2.c XHDR= dhry.h X XUNIX_PROGS= dry2 dry2reg XMSC_PROGS= sdry2.exe sdry2reg.exe mdry2.exe mdry2reg.exe \ X ldry2.exe ldry2reg.exe cdry2.exe cdry2reg.exe \ X hdry2.exe hdry2reg.exe X X# Files added by rer: XFILES1= README.RER clarify.doc Makefile submit.frm X# Reinhold's files: XFILES2= Rationale $(HDR) $(SRC) XFILES3= dhry.p X Xall: $(PROGS) X Xunix: $(UNIX_PROGS) X Xmsc: $(MSC_PROGS) X Xdry2: $(SRC) $(HDR) X $(CC) $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xdry2reg: $(SRC) $(HDR) X $(CC) $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xsdry2.exe: $(SRC) $(HDR) X $(CC) -AS $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xsdry2reg.exe: $(SRC) $(HDR) X $(CC) -AS $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xmdry2.exe: $(SRC) $(HDR) X $(CC) -AM $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xmdry2reg.exe: $(SRC) $(HDR) X $(CC) -AM $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xldry2.exe: $(SRC) $(HDR) X $(CC) -AL $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xldry2reg.exe: $(SRC) $(HDR) X $(CC) -AL $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xcdry2.exe: $(SRC) $(HDR) X $(CC) -AC $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xcdry2reg.exe: $(SRC) $(HDR) X $(CC) -AC $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xhdry2.exe: $(SRC) $(HDR) X $(CC) -AH $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xhdry2reg.exe: $(SRC) $(HDR) X $(CC) -AH $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xshar: dry2shar.1 dry2shar.2 dry2shar.3 X Xarc: $(FILES1) $(FILES2) $(FILES3) X arc a dry2.arc $(FILES1) $(FILES2) $(FILES3) X Xdry2shar.1: $(FILES1) X shar -p X $(FILES1) >$@ X Xdry2shar.2: $(FILES2) X shar -p X $(FILES2) >$@ X Xdry2shar.3: $(FILES3) X shar -p X $(FILES3) >$@ X Xclean: X -rm -f *.o *.obj X Xclobber: clean X -rm -f $(UNIX_PROGS) $(MSC_PROGS) dry2shar.* dry2.arc X Xpost: dry2shar.1 dry2shar.2 dry2shar.3 X for i in 1 2 3;\ X do\ X cat HEADERS BOILER.$$i dry2shar.$$i |\ X inews -h -t "Dhrystone 2.1 ($$i of 3)" -n comp.arch;\ X done X Xmail: dry2shar.1 dry2shar.2 dry2shar.3 X for i in 1 2 3;\ X do\ X cat BOILER.$$i dry2shar.$$i |\ X mailx -s "Dhrystone 2.1 ($$i of 3)" $(ADDR);\ X done SHAR_EOF fi if test -f 'submit.frm' then echo shar: "will not over-write existing file 'submit.frm'" else sed 's/^X//' << \SHAR_EOF > 'submit.frm' XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: XMODEL: XPROC: XCLOCK: XOS: XOVERSION: XCOMPILER: XCVERSION: XOPTIONS: XNOREG: XREG: XNOTES: XDATE: XSUBMITTER: XCODESIZE: XMAILTO: uunet!pcrat!dry2 SHAR_EOF fi exit 0 # End of shell archive
dry@pcrat.UUCP (Dhrystone) (07/07/88)
#! /bin/sh # This is a shell archive, meaning: # 1. Remove everything above the #! /bin/sh line. # 2. Save the resulting text in a file. # 3. Execute the file with /bin/sh (not csh) to create: # Rationale # dhry.h # dhry_1.c # dhry_2.c # This archive created: Wed Jul 6 16:50:06 1988 export PATH; PATH=/bin:/usr/bin:$PATH if test -f 'Rationale' then echo shar: "will not over-write existing file 'Rationale'" else sed 's/^X//' << \SHAR_EOF > 'Rationale' XDhrystone Benchmark: Rationale for Version 2 and Measurement Rules X X Reinhold P. Weicker X Siemens AG, E STE 35 X Postfach 3240 X D-8520 Erlangen X Germany (West) X X X X XThe Dhrystone benchmark program [1] has become a popular benchmark for XCPU/compiler performance measurement, in particular in the area of Xminicomputers, workstations, PC's and microprocesors. It apparently Xsatisfies a need for an easy-to-use integer benchmark; it gives a first Xperformance indication which is more meaningful than MIPS numbers Xwhich, in their literal meaning (million instructions per second), Xcannot be used across different instruction sets (e.g. RISC vs. CISC). XWith the increasing use of the benchmark, it seems necessary to Xreconsider the benchmark and to check whether it can still fulfill this Xfunction. Version 2 of Dhrystone is the result of such a re- Xevaluation, it has been made for two reasons: X Xo Dhrystone has been published in Ada [1], and Versions in Ada, Pascal X and C have been distributed by Reinhold Weicker via floppy disk. X However, the version that was used most often for benchmarking has X been the version made by Rick Richardson by another translation from X the Ada version into the C programming language, this has been the X version distributed via the UNIX network Usenet [2]. X X There is an obvious need for a common C version of Dhrystone, since C X is at present the most popular system programming language for the X class of systems (microcomputers, minicomputers, workstations) where X Dhrystone is used most. There should be, as far as possible, only X one C version of Dhrystone such that results can be compared without X restrictions. In the past, the C versions distributed by Rick X Richardson (Version 1.1) and by Reinhold Weicker had small (though X not significant) differences. X X Together with the new C version, the Ada and Pascal versions have X been updated as well. X Xo As far as it is possible without changes to the Dhrystone statistics, X optimizing compilers should be prevented from removing significant X statements. It has turned out in the past that optimizing compilers X suppressed code generation for too many statements (by "dead code X removal" or "dead variable elimination"). This has lead to the X danger that benchmarking results obtained by a naive application of X Dhrystone - without inspection of the code that was generated - could X become meaningless. X XThe overall policiy for version 2 has been that the distribution of Xstatements, operand types and operand locality described in [1] should Xremain unchanged as much as possible. (Very few changes were Xnecessary; their impact should be negligible.) Also, the order of Xstatements should remain unchanged. Although I am aware of some Xcritical remarks on the benchmark - I agree with several of them - and Xknow some suggestions for improvement, I didn't want to change the Xbenchmark into something different from what has become known as X"Dhrystone"; the confusion generated by such a change would probably Xoutweight the benefits. If I were to write a new benchmark program, I Xwouldn't give it the name "Dhrystone" since this denotes the program Xpublished in [1]. However, I do recognize the need for a larger number Xof representative programs that can be used as benchmarks; users should Xalways be encouraged to use more than just one benchmark. X XThe new versions (version 2.1 for C, Pascal and Ada) will be Xdistributed as widely as possible. (Version 2.1 differs from version X2.0 distributed via the UNIX Network Usenet in March 1988 only in a few Xcorrections for minor deficiencies found by users of version 2.0.) XReaders who want to use the benchmark for their own measurements can Xobtain a copy in machine-readable form on floppy disk (MS-DOS or XENIX Xformat) from the author. X X XIn general, version 2 follows - in the parts that are significant for Xperformance measurement, i.e. within the measurement loop - the Xpublished (Ada) version and the C versions previously distributed. XWhere the versions distributed by Rick Richardson [2] and Reinhold XWeicker have been different, it follows the version distributed by XReinhold Weicker. (However, the differences have been so small that Xtheir impact on execution time in all likelihood has been negligible.) XThe initialization and UNIX instrumentation part - which had been Xomitted in [1] - follows mostly the ideas of Rick Richardson [2]. XHowever, any changes in the initialization part and in the printing of Xthe result have no impact on performance measurement since they are Xoutside the measaurement loop. As a concession to older compilers, Xnames have been made unique within the first 8 characters for the C Xversion. X XThe original publication of Dhrystone did not contain any statements Xfor time measurement since they are necessarily system-dependent. XHowever, it turned out that it is not enough just to inclose the main Xprocedure of Dhrystone in a loop and to measure the execution time. If Xthe variables that are computed are not used somehow, there is the Xdanger that the compiler considers them as "dead variables" and Xsuppresses code generation for a part of the statements. Therefore in Xversion 2 all variables of "main" are printed at the end of the Xprogram. This also permits some plausibility control for correct Xexecution of the benchmark. X XAt several places in the benchmark, code has been added, but only in Xbranches that are not executed. The intention is that optimizing Xcompilers should be prevented from moving code out of the measurement Xloop, or from removing code altogether. Statements that are executed Xhave been changed in very few places only. In these cases, only the Xrole of some operands has been changed, and it was made sure that the Xnumbers defining the "Dhrystone distribution" (distribution of Xstatements, operand types and locality) still hold as much as possible. XExcept for sophisticated optimizing compilers, execution times for Xversion 2.1 should be the same as for previous versions. X XBecause of the self-imposed limitation that the order and distribution Xof the executed statements should not be changed, there are still cases Xwhere optimizing compilers may not generate code for some statements. XTo a certain degree, this is unavoidable for small synthetic Xbenchmarks. Users of the benchmark are advised to check code listings Xwhether code is generated for all statements of Dhrystone. X XContrary to the suggestion in the published paper and its realization Xin the versions previously distributed, no attempt has been made to Xsubtract the time for the measurement loop overhead. (This calculation Xhas proven difficult to implement in a correct way, and its omission Xmakes the program simpler.) However, since the loop check is now part Xof the benchmark, this does have an impact - though a very minor one - Xon the distribution statistics which have been updated for this Xversion. X X XIn this section, all changes are described that affect the measurement Xloop and that are not just renamings of variables. All remarks refer to Xthe C version; the other language versions have been updated similarly. X XIn addition to adding the measurement loop and the printout statements, Xchanges have been made at the following places: X Xo In procedure "main", three statements have been added in the non- X executed "then" part of the statement X if (Enum_Loc == Func_1 (Ch_Index, 'C')) X they are X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 3'RD STRING"); X Int_2_Loc = Run_Index; X Int_Glob = Run_Index; X The string assignment prevents movement of the preceding assignment X to Str_2_Loc (5'th statement of "main") out of the measurement loop X (This probably will not happen for the C version, but it did happen X with another language and compiler.) The assignment to Int_2_Loc X prevents value propagation for Int_2_Loc, and the assignment to X Int_Glob makes the value of Int_Glob possibly dependent from the X value of Run_Index. X Xo In the three arithmetic computations at the end of the measurement X loop in "main ", the role of some variables has been exchanged, to X prevent the division from just cancelling out the multiplication as X it was in [1]. A very smart compiler might have recognized this and X suppressed code generation for the division. X Xo For Proc_2, no code has been changed, but the values of the actual X parameter have changed due to changes in "main". X Xo In Proc_4, the second assignment has been changed from X Bool_Loc = Bool_Loc | Bool_Glob; X to X Bool_Glob = Bool_Loc | Bool_Glob; X It now assigns a value to a global variable instead of a local X variable (Bool_Loc); Bool_Loc would be a "dead variable" which is not X used afterwards. X Xo In Func_1, the statement X Ch_1_Glob = Ch_1_Loc; X was added in the non-executed "else" part of the "if" statement, to X prevent the suppression of code generation for the assignment to X Ch_1_Loc. X Xo In Func_2, the second character comparison statement has been changed X to X if (Ch_Loc == 'R') X ('R' instead of 'X') because a comparison with 'X' is implied in the X preceding "if" statement. X X Also in Func_2, the statement X Int_Glob = Int_Loc; X has been added in the non-executed part of the last "if" statement, X in order to prevent Int_Loc from becoming a dead variable. X Xo In Func_3, a non-executed "else" part has been added to the "if" X statement. While the program would not be incorrect without this X "else" part, it is considered bad programming practice if a function X can be left without a return value. X X To compensate for this change, the (non-executed) "else" part in the X "if" statement of Proc_3 was removed. X XThe distribution statistics have been changed only by the addition of Xthe measurement loop iteration (1 additional statement, 4 additional Xlocal integer operands) and by the change in Proc_4 (one operand Xchanged from local to global). The distribution statistics in the Xcomment headers have been updated accordingly. X X XThe string operations (string assignment and string comparison) have Xnot been changed, to keep the program consistent with the original Xversion. X XThere has been some concern that the string operations are over- Xrepresented in the program, and that execution time is dominated by Xthese operations. This was true in particular when optimizing Xcompilers removed too much code in the main part of the program, this Xshould have been mitigated in version 2. X XIt should be noted that this is a language-dependent issue: Dhrystone Xwas first published in Ada, and with Ada or Pascal semantics, the time Xspent in the string operations is, at least in all implementations Xknown to me, considerably smaller. In Ada and Pascal, assignment and Xcomparison of strings are operators defined in the language, and the Xupper bounds of the strings occuring in Dhrystone are part of the type Xinformation known at compilation time. The compilers can therefore Xgenerate efficient inline code. In C, string assignemt and comparisons Xare not part of the language, so the string operations must be Xexpressed in terms of the C library functions "strcpy" and "strcmp". X(ANSI C allows an implementation to use inline code for these Xfunctions.) In addition to the overhead caused by additional function Xcalls, these functions are defined for null-terminated strings where Xthe length of the strings is not known at compilation time; the Xfunction has to check every byte for the termination condition (the Xnull byte). X XObviously, a C library which includes efficiently coded "strcpy" and X"strcmp" functions helps to obtain good Dhrystone results. However, I Xdon't think that this is unfair since string functions do occur quite Xfrequently in real programs (editors, command interpreters, etc.). If Xthe strings functions are implemented efficiently, this helps real Xprograms as well as benchmark programs. X XI admit that the string comparison in Dhrystone terminates later (after Xscanning 20 characters) than most string comparisons in real programs. XFor consistency with the original benchmark, I didn't change the Xprogram despite this weakness. X X XWhen Dhrystone is used, the following "ground rules" apply: X Xo Separate compilation (Ada and C versions) X X As mentioned in [1], Dhrystone was written to reflect actual X programming practice in systems programming. The division into X several compilation units (5 in the Ada version, 2 in the C version) X is intended, as is the distribution of inter-module and intra-module X subprogram calls. Although on many systems there will be no X difference in execution time to a Dhrystone version where all X compilation units are merged into one file, the rule is that separate X compilation should be used. The intention is that real programming X practice, where programs consist of several independently compiled X units, should be reflected. This also has implies that the compiler, X while compiling one unit, has no information about the use of X variables, register allocation etc. occuring in other compilation X units. Although in real life compilation units will probably be X larger, the intention is that these effects of separate compilation X are modeled in Dhrystone. X X A few language systems have post-linkage optimization available X (e.g., final register allocation is performed after linkage). This X is a borderline case: Post-linkage optimization involves additional X program preparation time (although not as much as compilation in one X unit) which may prevent its general use in practical programming. I X think that since it defeats the intentions given above, it should not X be used for Dhrystone. X X Unfortunately, ISO/ANSI Pascal does not contain language features for X separate compilation. Although most commercial Pascal compilers X provide separate compilation in some way, we cannot use it for X Dhrystone since such a version would not be portable. Therefore, no X attempt has been made to provide a Pascal version with several X compilation units. X Xo No procedure merging X X Although Dhrystone contains some very short procedures where X execution would benefit from procedure merging (inlining, macro X expansion of procedures), procedure merging is not to be used. The X reason is that the percentage of procedure and function calls is part X of the "Dhrystone distribution" of statements contained in [1]. This X restriction does not hold for the string functions of the C version X since ANSI C allows an implementation to use inline code for these X functions. X X X Xo Other optimizations are allowed, but they should be indicated X X It is often hard to draw an exact line between "normal code X generation" and "optimization" in compilers: Some compilers perform X operations by default that are invoked in other compilers only when X optimization is explicitly requested. Also, we cannot avoid that in X benchmarking people try to achieve results that look as good as X possible. Therefore, optimizations performed by compilers - other X than those listed above - are not forbidden when Dhrystone execution X times are measured. Dhrystone is not intended to be non-optimizable X but is intended to be similarly optimizable as normal programs. For X example, there are several places in Dhrystone where performance X benefits from optimizations like common subexpression elimination, X value propagation etc., but normal programs usually also benefit from X these optimizations. Therefore, no effort was made to artificially X prevent such optimizations. However, measurement reports should X indicate which compiler optimization levels have been used, and X reporting results with different levels of compiler optimization for X the same hardware is encouraged. X Xo Default results are those without "register" declarations (C version) X X When Dhrystone results are quoted without additional qualification, X they should be understood as results obtained without use of the X "register" attribute. Good compilers should be able to make good use X of registers even without explicit register declarations ([3], p. X 193). X XOf course, for experimental purposes, post-linkage optimization, Xprocedure merging and/or compilation in one unit can be done to Xdetermine their effects. However, Dhrystone numbers obtained under Xthese conditions should be explicitly marked as such; "normal" XDhrystone results should be understood as results obtained following Xthe ground rules listed above. X XIn any case, for serious performance evaluation, users are advised to Xask for code listings and to check them carefully. In this way, when Xresults for different systems are compared, the reader can get a Xfeeling how much performance difference is due to compiler optimization Xand how much is due to hardware speed. X X XThe C version 2.1 of Dhrystone has been developed in cooperation with XRick Richardson (Tinton Falls, NJ), it incorporates many ideas from the X"Version 1.1" distributed previously by him over the UNIX network XUsenet. Through his activity with Usenet, Rick Richardson has made a Xvery valuable contribution to the dissemination of the benchmark. I Xalso thank Chaim Benedelac (National Semiconductor), David Ditzel X(SUN), Earl Killian and John Mashey (MIPS), Alan Smith and Rafael XSaavedra-Barrera (UC at Berkeley) for their help with comments on Xearlier versions of the benchmark. X X X[1] X Reinhold P. Weicker: Dhrystone: A Synthetic Systems Programming X Benchmark. X Communications of the ACM 27, 10 (Oct. 1984), 1013-1030 X X[2] X Rick Richardson: Dhrystone 1.1 Benchmark Summary (and Program Text) X Informal Distribution via "Usenet", Last Version Known to me: Sept. X 21, 1987 X X[3] X Brian W. Kernighan and Dennis M. Ritchie: The C Programming X Language. X Prentice-Hall, Englewood Cliffs (NJ) 1978 X X X X X SHAR_EOF fi if test -f 'dhry.h' then echo shar: "will not over-write existing file 'dhry.h'" else sed 's/^X//' << \SHAR_EOF > 'dhry.h' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry.h (part 1 of 3) X * X * Date: May 17, 1988 X * X * Author: Reinhold P. Weicker X * Siemens AG, E STE 35 X * Postfach 3240 X * 8520 Erlangen X * Germany (West) X * Phone: [xxx-49]-9131-7-20330 X * (8-17 Central European Time) X * Usenet: ..!mcvax!unido!estevax!weicker X * X * Original Version (in Ada) published in X * "Communications of the ACM" vol. 27., no. 10 (Oct. 1984), X * pp. 1013 - 1030, together with the statistics X * on which the distribution of statements etc. is based. X * X * In this C version, the following C library functions are used: X * - strcpy, strcmp (inside the measurement loop) X * - printf, scanf (outside the measurement loop) X * In addition, Berkeley UNIX system calls "times ()" or "time ()" X * are used for execution time measurement. For measurements X * on other systems, these calls have to be changed. X * X * Collection of Results: X * Reinhold Weicker (address see above) and X * X * Rick Richardson X * PC Research. Inc. X * 94 Apple Orchard Drive X * Tinton Falls, NJ 07724 X * Phone: (201) 389-8963 (9-17 EST) X * Usenet: ...!uunet!pcrat!rick X * X * Please send results to Rick Richardson and/or Reinhold Weicker. X * Complete information should be given on hardware and software used. X * Hardware information includes: Machine type, CPU, type and size X * of caches; for microprocessors: clock frequency, memory speed X * (number of wait states). X * Software information includes: Compiler (and runtime library) X * manufacturer and version, compilation switches, OS version. X * The Operating System version may give an indication about the X * compiler; Dhrystone itself performs no OS calls in the measurement loop. X * X * The complete output generated by the program should be mailed X * such that at least some checks for correctness can be made. X * X *************************************************************************** X * X * History: This version C/2.1 has been made for two reasons: X * X * 1) There is an obvious need for a common C version of X * Dhrystone, since C is at present the most popular system X * programming language for the class of processors X * (microcomputers, minicomputers) where Dhrystone is used most. X * There should be, as far as possible, only one C version of X * Dhrystone such that results can be compared without X * restrictions. In the past, the C versions distributed X * by Rick Richardson (Version 1.1) and by Reinhold Weicker X * had small (though not significant) differences. X * X * 2) As far as it is possible without changes to the Dhrystone X * statistics, optimizing compilers should be prevented from X * removing significant statements. X * X * This C version has been developed in cooperation with X * Rick Richardson (Tinton Falls, NJ), it incorporates many X * ideas from the "Version 1.1" distributed previously by X * him over the UNIX network Usenet. X * I also thank Chaim Benedelac (National Semiconductor), X * David Ditzel (SUN), Earl Killian and John Mashey (MIPS), X * Alan Smith and Rafael Saavedra-Barrera (UC at Berkeley) X * for their help with comments on earlier versions of the X * benchmark. X * X * Changes: In the initialization part, this version follows mostly X * Rick Richardson's version distributed via Usenet, not the X * version distributed earlier via floppy disk by Reinhold Weicker. X * As a concession to older compilers, names have been made X * unique within the first 8 characters. X * Inside the measurement loop, this version follows the X * version previously distributed by Reinhold Weicker. X * X * At several places in the benchmark, code has been added, X * but within the measurement loop only in branches that X * are not executed. The intention is that optimizing compilers X * should be prevented from moving code out of the measurement X * loop, or from removing code altogether. Since the statements X * that are executed within the measurement loop have NOT been X * changed, the numbers defining the "Dhrystone distribution" X * (distribution of statements, operand types and locality) X * still hold. Except for sophisticated optimizing compilers, X * execution times for this version should be the same as X * for previous versions. X * X * Since it has proven difficult to subtract the time for the X * measurement loop overhead in a correct way, the loop check X * has been made a part of the benchmark. This does have X * an impact - though a very minor one - on the distribution X * statistics which have been updated for this version. X * X * All changes within the measurement loop are described X * and discussed in the companion paper "Rationale for X * Dhrystone version 2". X * X * Because of the self-imposed limitation that the order and X * distribution of the executed statements should not be X * changed, there are still cases where optimizing compilers X * may not generate code for some statements. To a certain X * degree, this is unavoidable for small synthetic benchmarks. X * Users of the benchmark are advised to check code listings X * whether code is generated for all statements of Dhrystone. X * X * Version 2.1 is identical to version 2.0 distributed via X * the UNIX network Usenet in March 1988 except that it corrects X * some minor deficiencies that were found by users of version 2.0. X * The following corrections have been made in the C version: X * - The assignment to Number_Of_Runs was changed X * - The constant Too_Small_Time was changed X * - An "else" part was added to the "if" statement in Func_3; X * for compensation, an "else" part was removed in Proc_3 X * - Shorter file names are used X * X *************************************************************************** X * X * Defines: The following "Defines" are possible: X * -DREG=register (default: Not defined) X * As an approximation to what an average C programmer X * might do, the "register" storage class is applied X * (if enabled by -DREG=register) X * - for local variables, if they are used (dynamically) X * five or more times X * - for parameters if they are used (dynamically) X * six or more times X * Note that an optimal "register" strategy is X * compiler-dependent, and that "register" declarations X * do not necessarily lead to faster execution. X * -DNOSTRUCTASSIGN (default: Not defined) X * Define if the C compiler does not support X * assignment of structures. X * -DNOENUMS (default: Not defined) X * Define if the C compiler does not support X * enumeration types. X * -DTIMES (default) X * -DTIME X * The "times" function of UNIX (returning process times) X * or the "time" function (returning wallclock time) X * is used for measurement. X * For single user machines, "time ()" is adequate. For X * multi-user machines where you cannot get single-user X * access, use the "times ()" function. If you have X * neither, use a stopwatch in the dead of night. X * "printf"s are provided marking the points "Start Timer" X * and "Stop Timer". DO NOT use the UNIX "time(1)" X * command, as this will measure the total time to X * run this program, which will (erroneously) include X * the time to allocate storage (malloc) and to perform X * the initialization. X * -DHZ=nnn X * In Berkeley UNIX, the function "times" returns process X * time in 1/HZ seconds, with HZ = 60 for most systems. X * CHECK YOUR SYSTEM DESCRIPTION BEFORE YOU JUST APPLY X * A VALUE. X * X *************************************************************************** X * X * Compilation model and measurement (IMPORTANT): X * X * This C version of Dhrystone consists of three files: X * - dhry.h (this file, containing global definitions and comments) X * - dhry_1.c (containing the code corresponding to Ada package Pack_1) X * - dhry_2.c (containing the code corresponding to Ada package Pack_2) X * X * The following "ground rules" apply for measurements: X * - Separate compilation X * - No procedure merging X * - Otherwise, compiler optimizations are allowed but should be indicated X * - Default results are those without register declarations X * See the companion paper "Rationale for Dhrystone Version 2" for a more X * detailed discussion of these ground rules. X * X * For 16-Bit processors (e.g. 80186, 80286), times for all compilation X * models ("small", "medium", "large" etc.) should be given if possible, X * together with a definition of these models for the compiler system used. X * X ************************************************************************** X * X * Dhrystone (C version) statistics: X * X * [Comment from the first distribution, updated for version 2. X * Note that because of language differences, the numbers are slightly X * different from the Ada version.] X * X * The following program contains statements of a high level programming X * language (here: C) in a distribution considered representative: X * X * assignments 52 (51.0 %) X * control statements 33 (32.4 %) X * procedure, function calls 17 (16.7 %) X * X * 103 statements are dynamically executed. The program is balanced with X * respect to the three aspects: X * X * - statement type X * - operand type X * - operand locality X * operand global, local, parameter, or constant. X * X * The combination of these three aspects is balanced only approximately. X * X * 1. Statement Type: X * ----------------- number X * X * V1 = V2 9 X * (incl. V1 = F(..) X * V = Constant 12 X * Assignment, 7 X * with array element X * Assignment, 6 X * with record component X * -- X * 34 34 X * X * X = Y +|-|"&&"|"|" Z 5 X * X = Y +|-|"==" Constant 6 X * X = X +|- 1 3 X * X = Y *|/ Z 2 X * X = Expression, 1 X * two operators X * X = Expression, 1 X * three operators X * -- X * 18 18 X * X * if .... 14 X * with "else" 7 X * without "else" 7 X * executed 3 X * not executed 4 X * for ... 7 | counted every time X * while ... 4 | the loop condition X * do ... while 1 | is evaluated X * switch ... 1 X * break 1 X * declaration with 1 X * initialization X * -- X * 34 34 X * X * P (...) procedure call 11 X * user procedure 10 X * library procedure 1 X * X = F (...) X * function call 6 X * user function 5 X * library function 1 X * -- X * 17 17 X * --- X * 103 X * X * The average number of parameters in procedure or function calls X * is 1.82 (not counting the function values as implicit parameters). X * X * X * 2. Operators X * ------------ X * number approximate X * percentage X * X * Arithmetic 32 50.8 X * X * + 21 33.3 X * - 7 11.1 X * * 3 4.8 X * / (int div) 1 1.6 X * X * Comparison 27 42.8 X * X * == 9 14.3 X * /= 4 6.3 X * > 1 1.6 X * < 3 4.8 X * >= 1 1.6 X * <= 9 14.3 X * X * Logic 4 6.3 X * X * && (AND-THEN) 1 1.6 X * | (OR) 1 1.6 X * ! (NOT) 2 3.2 X * X * -- ----- X * 63 100.1 X * X * X * 3. Operand Type (counted once per operand reference): X * --------------- X * number approximate X * percentage X * X * Integer 175 72.3 % X * Character 45 18.6 % X * Pointer 12 5.0 % X * String30 6 2.5 % X * Array 2 0.8 % X * Record 2 0.8 % X * --- ------- X * 242 100.0 % X * X * When there is an access path leading to the final operand (e.g. a record X * component), only the final data type on the access path is counted. X * X * X * 4. Operand Locality: X * ------------------- X * number approximate X * percentage X * X * local variable 114 47.1 % X * global variable 22 9.1 % X * parameter 45 18.6 % X * value 23 9.5 % X * reference 22 9.1 % X * function result 6 2.5 % X * constant 55 22.7 % X * --- ------- X * 242 100.0 % X * X * X * The program does not compute anything meaningful, but it is syntactically X * and semantically correct. All variables have a value assigned to them X * before they are used as a source operand. X * X * There has been no explicit effort to account for the effects of a X * cache, or to balance the use of long or short displacements for code or X * data. X * X *************************************************************************** X */ X X/* Compiler and system dependent definitions: */ X X#ifndef TIME X#ifndef TIMES X#define TIMES X#endif X#endif X /* Use times(2) time function unless */ X /* explicitly defined otherwise */ X X#ifdef MSC_CLOCK X#undef HZ X#undef TIMES X#include <time.h> X#define HZ CLK_TCK X#endif X /* Use Microsoft C hi-res clock */ X X#ifdef TIMES X#include <sys/types.h> X#include <sys/times.h> X /* for "times" */ X#endif X X#define Mic_secs_Per_Second 1000000.0 X /* Berkeley UNIX C returns process times in seconds/HZ */ X X#ifdef NOSTRUCTASSIGN X#define structassign(d, s) memcpy(&(d), &(s), sizeof(d)) X#else X#define structassign(d, s) d = s X#endif X X#ifdef NOENUM X#define Ident_1 0 X#define Ident_2 1 X#define Ident_3 2 X#define Ident_4 3 X#define Ident_5 4 X typedef int Enumeration; X#else X typedef enum {Ident_1, Ident_2, Ident_3, Ident_4, Ident_5} X Enumeration; X#endif X /* for boolean and enumeration types in Ada, Pascal */ X X/* General definitions: */ X X#include <stdio.h> X /* for strcpy, strcmp */ X X#define Null 0 X /* Value of a Null pointer */ X#define true 1 X#define false 0 X Xtypedef int One_Thirty; Xtypedef int One_Fifty; Xtypedef char Capital_Letter; Xtypedef int Boolean; Xtypedef char Str_30 [31]; Xtypedef int Arr_1_Dim [50]; Xtypedef int Arr_2_Dim [50] [50]; X Xtypedef struct record X { X struct record *Ptr_Comp; X Enumeration Discr; X union { X struct { X Enumeration Enum_Comp; X int Int_Comp; X char Str_Comp [31]; X } var_1; X struct { X Enumeration E_Comp_2; X char Str_2_Comp [31]; X } var_2; X struct { X char Ch_1_Comp; X char Ch_2_Comp; X } var_3; X } variant; X } Rec_Type, *Rec_Pointer; X X SHAR_EOF fi if test -f 'dhry_1.c' then echo shar: "will not over-write existing file 'dhry_1.c'" else sed 's/^X//' << \SHAR_EOF > 'dhry_1.c' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry_1.c (part 2 of 3) X * X * Date: May 17, 1988 X * X * Author: Reinhold P. Weicker X * X **************************************************************************** X */ X X#include "dhry.h" X X/* Global Variables: */ X XRec_Pointer Ptr_Glob, X Next_Ptr_Glob; Xint Int_Glob; XBoolean Bool_Glob; Xchar Ch_1_Glob, X Ch_2_Glob; Xint Arr_1_Glob [50]; Xint Arr_2_Glob [50] [50]; X Xextern char *malloc (); XEnumeration Func_1 (); X /* forward declaration necessary since Enumeration may not simply be int */ X X#ifndef REG X Boolean Reg = false; X#define REG X /* REG becomes defined as empty */ X /* i.e. no register variables */ X#else X Boolean Reg = true; X#endif X X/* variables for time measurement: */ X X#ifdef TIMES Xstruct tms time_info; Xextern int times (); X /* see library function "times" */ X#define Too_Small_Time (2*HZ) X /* Measurements should last at least about 2 seconds */ X#endif X#ifdef TIME Xextern long time(); X /* see library function "time" */ X#define Too_Small_Time 2 X /* Measurements should last at least 2 seconds */ X#endif X#ifdef MSC_CLOCK Xextern clock_t clock(); X#define Too_Small_Time (2*HZ) X#endif X Xlong Begin_Time, X End_Time, X User_Time; Xfloat Microseconds, X Dhrystones_Per_Second; X X/* end of variables for time measurement */ X X Xmain () X/*****/ X X /* main program, corresponds to procedures */ X /* Main and Proc_0 in the Ada version */ X{ X One_Fifty Int_1_Loc; X REG One_Fifty Int_2_Loc; X One_Fifty Int_3_Loc; X REG char Ch_Index; X Enumeration Enum_Loc; X Str_30 Str_1_Loc; X Str_30 Str_2_Loc; X REG int Run_Index; X REG int Number_Of_Runs; X X /* Initializations */ X X Next_Ptr_Glob = (Rec_Pointer) malloc (sizeof (Rec_Type)); X Ptr_Glob = (Rec_Pointer) malloc (sizeof (Rec_Type)); X X Ptr_Glob->Ptr_Comp = Next_Ptr_Glob; X Ptr_Glob->Discr = Ident_1; X Ptr_Glob->variant.var_1.Enum_Comp = Ident_3; X Ptr_Glob->variant.var_1.Int_Comp = 40; X strcpy (Ptr_Glob->variant.var_1.Str_Comp, X "DHRYSTONE PROGRAM, SOME STRING"); X strcpy (Str_1_Loc, "DHRYSTONE PROGRAM, 1'ST STRING"); X X Arr_2_Glob [8][7] = 10; X /* Was missing in published program. Without this statement, */ X /* Arr_2_Glob [8][7] would have an undefined value. */ X /* Warning: With 16-Bit processors and Number_Of_Runs > 32000, */ X /* overflow may occur for this array element. */ X X printf ("\n"); X printf ("Dhrystone Benchmark, Version 2.1 (Language: C)\n"); X printf ("\n"); X if (Reg) X { X printf ("Program compiled with 'register' attribute\n"); X printf ("\n"); X } X else X { X printf ("Program compiled without 'register' attribute\n"); X printf ("\n"); X } X printf ("Please give the number of runs through the benchmark: "); X { X int n; X scanf ("%d", &n); X Number_Of_Runs = n; X } X printf ("\n"); X X printf ("Execution starts, %d runs through Dhrystone\n", Number_Of_Runs); X X /***************/ X /* Start timer */ X /***************/ X X#ifdef TIMES X times (&time_info); X Begin_Time = (long) time_info.tms_utime; X#endif X#ifdef TIME X Begin_Time = time ( (long *) 0); X#endif X#ifdef MSC_CLOCK X Begin_Time = clock(); X#endif X X for (Run_Index = 1; Run_Index <= Number_Of_Runs; ++Run_Index) X { X X Proc_5(); X Proc_4(); X /* Ch_1_Glob == 'A', Ch_2_Glob == 'B', Bool_Glob == true */ X Int_1_Loc = 2; X Int_2_Loc = 3; X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 2'ND STRING"); X Enum_Loc = Ident_2; X Bool_Glob = ! Func_2 (Str_1_Loc, Str_2_Loc); X /* Bool_Glob == 1 */ X while (Int_1_Loc < Int_2_Loc) /* loop body executed once */ X { X Int_3_Loc = 5 * Int_1_Loc - Int_2_Loc; X /* Int_3_Loc == 7 */ X Proc_7 (Int_1_Loc, Int_2_Loc, &Int_3_Loc); X /* Int_3_Loc == 7 */ X Int_1_Loc += 1; X } /* while */ X /* Int_1_Loc == 3, Int_2_Loc == 3, Int_3_Loc == 7 */ X Proc_8 (Arr_1_Glob, Arr_2_Glob, Int_1_Loc, Int_3_Loc); X /* Int_Glob == 5 */ X Proc_1 (Ptr_Glob); X for (Ch_Index = 'A'; Ch_Index <= Ch_2_Glob; ++Ch_Index) X /* loop body executed twice */ X { X if (Enum_Loc == Func_1 (Ch_Index, 'C')) X /* then, not executed */ X { X Proc_6 (Ident_1, &Enum_Loc); X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 3'RD STRING"); X Int_2_Loc = Run_Index; X Int_Glob = Run_Index; X } X } X /* Int_1_Loc == 3, Int_2_Loc == 3, Int_3_Loc == 7 */ X Int_2_Loc = Int_2_Loc * Int_1_Loc; X Int_1_Loc = Int_2_Loc / Int_3_Loc; X Int_2_Loc = 7 * (Int_2_Loc - Int_3_Loc) - Int_1_Loc; X /* Int_1_Loc == 1, Int_2_Loc == 13, Int_3_Loc == 7 */ X Proc_2 (&Int_1_Loc); X /* Int_1_Loc == 5 */ X X } /* loop "for Run_Index" */ X X /**************/ X /* Stop timer */ X /**************/ X X#ifdef TIMES X times (&time_info); X End_Time = (long) time_info.tms_utime; X#endif X#ifdef TIME X End_Time = time ( (long *) 0); X#endif X#ifdef MSC_CLOCK X End_Time = clock(); X#endif X X printf ("Execution ends\n"); X printf ("\n"); X printf ("Final values of the variables used in the benchmark:\n"); X printf ("\n"); X printf ("Int_Glob: %d\n", Int_Glob); X printf (" should be: %d\n", 5); X printf ("Bool_Glob: %d\n", Bool_Glob); X printf (" should be: %d\n", 1); X printf ("Ch_1_Glob: %c\n", Ch_1_Glob); X printf (" should be: %c\n", 'A'); X printf ("Ch_2_Glob: %c\n", Ch_2_Glob); X printf (" should be: %c\n", 'B'); X printf ("Arr_1_Glob[8]: %d\n", Arr_1_Glob[8]); X printf (" should be: %d\n", 7); X printf ("Arr_2_Glob[8][7]: %d\n", Arr_2_Glob[8][7]); X printf (" should be: Number_Of_Runs + 10\n"); X printf ("Ptr_Glob->\n"); X printf (" Ptr_Comp: %d\n", (int) Ptr_Glob->Ptr_Comp); X printf (" should be: (implementation-dependent)\n"); X printf (" Discr: %d\n", Ptr_Glob->Discr); X printf (" should be: %d\n", 0); X printf (" Enum_Comp: %d\n", Ptr_Glob->variant.var_1.Enum_Comp); X printf (" should be: %d\n", 2); X printf (" Int_Comp: %d\n", Ptr_Glob->variant.var_1.Int_Comp); X printf (" should be: %d\n", 17); X printf (" Str_Comp: %s\n", Ptr_Glob->variant.var_1.Str_Comp); X printf (" should be: DHRYSTONE PROGRAM, SOME STRING\n"); X printf ("Next_Ptr_Glob->\n"); X printf (" Ptr_Comp: %d\n", (int) Next_Ptr_Glob->Ptr_Comp); X printf (" should be: (implementation-dependent), same as above\n"); X printf (" Discr: %d\n", Next_Ptr_Glob->Discr); X printf (" should be: %d\n", 0); X printf (" Enum_Comp: %d\n", Next_Ptr_Glob->variant.var_1.Enum_Comp); X printf (" should be: %d\n", 1); X printf (" Int_Comp: %d\n", Next_Ptr_Glob->variant.var_1.Int_Comp); X printf (" should be: %d\n", 18); X printf (" Str_Comp: %s\n", X Next_Ptr_Glob->variant.var_1.Str_Comp); X printf (" should be: DHRYSTONE PROGRAM, SOME STRING\n"); X printf ("Int_1_Loc: %d\n", Int_1_Loc); X printf (" should be: %d\n", 5); X printf ("Int_2_Loc: %d\n", Int_2_Loc); X printf (" should be: %d\n", 13); X printf ("Int_3_Loc: %d\n", Int_3_Loc); X printf (" should be: %d\n", 7); X printf ("Enum_Loc: %d\n", Enum_Loc); X printf (" should be: %d\n", 1); X printf ("Str_1_Loc: %s\n", Str_1_Loc); X printf (" should be: DHRYSTONE PROGRAM, 1'ST STRING\n"); X printf ("Str_2_Loc: %s\n", Str_2_Loc); X printf (" should be: DHRYSTONE PROGRAM, 2'ND STRING\n"); X printf ("\n"); X X User_Time = End_Time - Begin_Time; X X if (User_Time < Too_Small_Time) X { X printf ("Measured time too small to obtain meaningful results\n"); X printf ("Please increase number of runs\n"); X printf ("\n"); X } X else X { X#ifdef TIME X Microseconds = (float) User_Time * Mic_secs_Per_Second X / (float) Number_Of_Runs; X Dhrystones_Per_Second = (float) Number_Of_Runs / (float) User_Time; X#else X Microseconds = (float) User_Time * Mic_secs_Per_Second X / ((float) HZ * ((float) Number_Of_Runs)); X Dhrystones_Per_Second = ((float) HZ * (float) Number_Of_Runs) X / (float) User_Time; X#endif X printf ("Microseconds for one run through Dhrystone: "); X printf ("%6.1f \n", Microseconds); X printf ("Dhrystones per Second: "); X printf ("%6.1f \n", Dhrystones_Per_Second); X printf ("\n"); X } X X} X X XProc_1 (Ptr_Val_Par) X/******************/ X XREG Rec_Pointer Ptr_Val_Par; X /* executed once */ X{ X REG Rec_Pointer Next_Record = Ptr_Val_Par->Ptr_Comp; X /* == Ptr_Glob_Next */ X /* Local variable, initialized with Ptr_Val_Par->Ptr_Comp, */ X /* corresponds to "rename" in Ada, "with" in Pascal */ X X structassign (*Ptr_Val_Par->Ptr_Comp, *Ptr_Glob); X Ptr_Val_Par->variant.var_1.Int_Comp = 5; X Next_Record->variant.var_1.Int_Comp X = Ptr_Val_Par->variant.var_1.Int_Comp; X Next_Record->Ptr_Comp = Ptr_Val_Par->Ptr_Comp; X Proc_3 (&Next_Record->Ptr_Comp); X /* Ptr_Val_Par->Ptr_Comp->Ptr_Comp X == Ptr_Glob->Ptr_Comp */ X if (Next_Record->Discr == Ident_1) X /* then, executed */ X { X Next_Record->variant.var_1.Int_Comp = 6; X Proc_6 (Ptr_Val_Par->variant.var_1.Enum_Comp, X &Next_Record->variant.var_1.Enum_Comp); X Next_Record->Ptr_Comp = Ptr_Glob->Ptr_Comp; X Proc_7 (Next_Record->variant.var_1.Int_Comp, 10, X &Next_Record->variant.var_1.Int_Comp); X } X else /* not executed */ X structassign (*Ptr_Val_Par, *Ptr_Val_Par->Ptr_Comp); X} /* Proc_1 */ X X XProc_2 (Int_Par_Ref) X/******************/ X /* executed once */ X /* *Int_Par_Ref == 1, becomes 4 */ X XOne_Fifty *Int_Par_Ref; X{ X One_Fifty Int_Loc; X Enumeration Enum_Loc; X X Int_Loc = *Int_Par_Ref + 10; X do /* executed once */ X if (Ch_1_Glob == 'A') X /* then, executed */ X { X Int_Loc -= 1; X *Int_Par_Ref = Int_Loc - Int_Glob; X Enum_Loc = Ident_1; X } /* if */ X while (Enum_Loc != Ident_1); /* true */ X} /* Proc_2 */ X X XProc_3 (Ptr_Ref_Par) X/******************/ X /* executed once */ X /* Ptr_Ref_Par becomes Ptr_Glob */ X XRec_Pointer *Ptr_Ref_Par; X X{ X if (Ptr_Glob != Null) X /* then, executed */ X *Ptr_Ref_Par = Ptr_Glob->Ptr_Comp; X Proc_7 (10, Int_Glob, &Ptr_Glob->variant.var_1.Int_Comp); X} /* Proc_3 */ X X XProc_4 () /* without parameters */ X/*******/ X /* executed once */ X{ X Boolean Bool_Loc; X X Bool_Loc = Ch_1_Glob == 'A'; X Bool_Glob = Bool_Loc | Bool_Glob; X Ch_2_Glob = 'B'; X} /* Proc_4 */ X X XProc_5 () /* without parameters */ X/*******/ X /* executed once */ X{ X Ch_1_Glob = 'A'; X Bool_Glob = false; X} /* Proc_5 */ X X X /* Procedure for the assignment of structures, */ X /* if the C compiler doesn't support this feature */ X#ifdef NOSTRUCTASSIGN Xmemcpy (d, s, l) Xregister char *d; Xregister char *s; Xregister int l; X{ X while (l--) *d++ = *s++; X} X#endif X X SHAR_EOF fi if test -f 'dhry_2.c' then echo shar: "will not over-write existing file 'dhry_2.c'" else sed 's/^X//' << \SHAR_EOF > 'dhry_2.c' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry_2.c (part 3 of 3) X * X * Date: May 17, 1988 X * X * Author: Reinhold P. Weicker X * X **************************************************************************** X */ X X#include "dhry.h" X X#ifndef REG X#define REG X /* REG becomes defined as empty */ X /* i.e. no register variables */ X#endif X Xextern int Int_Glob; Xextern char Ch_1_Glob; X X XProc_6 (Enum_Val_Par, Enum_Ref_Par) X/*********************************/ X /* executed once */ X /* Enum_Val_Par == Ident_3, Enum_Ref_Par becomes Ident_2 */ X XEnumeration Enum_Val_Par; XEnumeration *Enum_Ref_Par; X{ X *Enum_Ref_Par = Enum_Val_Par; X if (! Func_3 (Enum_Val_Par)) X /* then, not executed */ X *Enum_Ref_Par = Ident_4; X switch (Enum_Val_Par) X { X case Ident_1: X *Enum_Ref_Par = Ident_1; X break; X case Ident_2: X if (Int_Glob > 100) X /* then */ X *Enum_Ref_Par = Ident_1; X else *Enum_Ref_Par = Ident_4; X break; X case Ident_3: /* executed */ X *Enum_Ref_Par = Ident_2; X break; X case Ident_4: break; X case Ident_5: X *Enum_Ref_Par = Ident_3; X break; X } /* switch */ X} /* Proc_6 */ X X XProc_7 (Int_1_Par_Val, Int_2_Par_Val, Int_Par_Ref) X/**********************************************/ X /* executed three times */ X /* first call: Int_1_Par_Val == 2, Int_2_Par_Val == 3, */ X /* Int_Par_Ref becomes 7 */ X /* second call: Int_1_Par_Val == 10, Int_2_Par_Val == 5, */ X /* Int_Par_Ref becomes 17 */ X /* third call: Int_1_Par_Val == 6, Int_2_Par_Val == 10, */ X /* Int_Par_Ref becomes 18 */ XOne_Fifty Int_1_Par_Val; XOne_Fifty Int_2_Par_Val; XOne_Fifty *Int_Par_Ref; X{ X One_Fifty Int_Loc; X X Int_Loc = Int_1_Par_Val + 2; X *Int_Par_Ref = Int_2_Par_Val + Int_Loc; X} /* Proc_7 */ X X XProc_8 (Arr_1_Par_Ref, Arr_2_Par_Ref, Int_1_Par_Val, Int_2_Par_Val) X/*********************************************************************/ X /* executed once */ X /* Int_Par_Val_1 == 3 */ X /* Int_Par_Val_2 == 7 */ XArr_1_Dim Arr_1_Par_Ref; XArr_2_Dim Arr_2_Par_Ref; Xint Int_1_Par_Val; Xint Int_2_Par_Val; X{ X REG One_Fifty Int_Index; X REG One_Fifty Int_Loc; X X Int_Loc = Int_1_Par_Val + 5; X Arr_1_Par_Ref [Int_Loc] = Int_2_Par_Val; X Arr_1_Par_Ref [Int_Loc+1] = Arr_1_Par_Ref [Int_Loc]; X Arr_1_Par_Ref [Int_Loc+30] = Int_Loc; X for (Int_Index = Int_Loc; Int_Index <= Int_Loc+1; ++Int_Index) X Arr_2_Par_Ref [Int_Loc] [Int_Index] = Int_Loc; X Arr_2_Par_Ref [Int_Loc] [Int_Loc-1] += 1; X Arr_2_Par_Ref [Int_Loc+20] [Int_Loc] = Arr_1_Par_Ref [Int_Loc]; X Int_Glob = 5; X} /* Proc_8 */ X X XEnumeration Func_1 (Ch_1_Par_Val, Ch_2_Par_Val) X/*************************************************/ X /* executed three times */ X /* first call: Ch_1_Par_Val == 'H', Ch_2_Par_Val == 'R' */ X /* second call: Ch_1_Par_Val == 'A', Ch_2_Par_Val == 'C' */ X /* third call: Ch_1_Par_Val == 'B', Ch_2_Par_Val == 'C' */ X XCapital_Letter Ch_1_Par_Val; XCapital_Letter Ch_2_Par_Val; X{ X Capital_Letter Ch_1_Loc; X Capital_Letter Ch_2_Loc; X X Ch_1_Loc = Ch_1_Par_Val; X Ch_2_Loc = Ch_1_Loc; X if (Ch_2_Loc != Ch_2_Par_Val) X /* then, executed */ X return (Ident_1); X else /* not executed */ X { X Ch_1_Glob = Ch_1_Loc; X return (Ident_2); X } X} /* Func_1 */ X X XBoolean Func_2 (Str_1_Par_Ref, Str_2_Par_Ref) X/*************************************************/ X /* executed once */ X /* Str_1_Par_Ref == "DHRYSTONE PROGRAM, 1'ST STRING" */ X /* Str_2_Par_Ref == "DHRYSTONE PROGRAM, 2'ND STRING" */ X XStr_30 Str_1_Par_Ref; XStr_30 Str_2_Par_Ref; X{ X REG One_Thirty Int_Loc; X Capital_Letter Ch_Loc; X X Int_Loc = 2; X while (Int_Loc <= 2) /* loop body executed once */ X if (Func_1 (Str_1_Par_Ref[Int_Loc], X Str_2_Par_Ref[Int_Loc+1]) == Ident_1) X /* then, executed */ X { X Ch_Loc = 'A'; X Int_Loc += 1; X } /* if, while */ X if (Ch_Loc >= 'W' && Ch_Loc < 'Z') X /* then, not executed */ X Int_Loc = 7; X if (Ch_Loc == 'R') X /* then, not executed */ X return (true); X else /* executed */ X { X if (strcmp (Str_1_Par_Ref, Str_2_Par_Ref) > 0) X /* then, not executed */ X { X Int_Loc += 7; X Int_Glob = Int_Loc; X return (true); X } X else /* executed */ X return (false); X } /* if Ch_Loc */ X} /* Func_2 */ X X XBoolean Func_3 (Enum_Par_Val) X/***************************/ X /* executed once */ X /* Enum_Par_Val == Ident_3 */ XEnumeration Enum_Par_Val; X{ X Enumeration Enum_Loc; X X Enum_Loc = Enum_Par_Val; X if (Enum_Loc == Ident_3) X /* then, executed */ X return (true); X else /* not executed */ X return (false); X} /* Func_3 */ X SHAR_EOF fi exit 0 # End of shell archive
peter@ficc.UUCP (Peter da Silva) (07/12/88)
One thing I noticed in this is that it makes the 80286 look a lot better than it really is, since it doesn't (even with lard... uh, large... model compilers) account for the problems you get as soon as objects get over 64K. There was an example in Byte recently, using the sieve benchmark, that showed about a factor of 11 difference between the 68000 and the 80286 as soon as arrays got over 64K. -- -- `-_-' Peter (have you hugged your wolf today) da Silva. -- U Ferranti International Controls Corporation. -- Phone: 713-274-5180. CI$: 70216,1076. ICBM: 29 37 N / 95 36 W. -- UUCP: {uunet,academ!uhnix1,bellcore!tness1}!sugar!ficc!peter.
mahar@weitek.UUCP (07/13/88)
I've noticed that the way people use the Dhrystone benchmark is very different than the initial intent. It was supposed to be used to evaluate computer architectures. Instead, it is used to evaluate C compilers. In the RISC world, this makes a certain sense. Given the following rule of thumb. For a given RISC architecture, Vax780mips = .7 * Clock in Mhz A 10Mhz RISC is about 7 VAX780 mips. This is true for Mips Co., 88000, 29000, and Weitek XL8000. This doesn't seem to hold true for SPARC. That machine seems to be between 5 & 6 VAX780 mips. Given this the Dhrystone benchmark is used to evaluate compilers for the various RISC processors. A very good optimizing compiler will eliminate all the code in Dhrystone except the calls to times. This would give the result of a divide by zero exception since the two calls to times would take place in less than a 60th of a second. None of the compilers are doing that yet but they are making great strides is optimizing the code for Dhrystone. For example. The string copy can be done a word at a time if the strings are guaranteed to be word aligned. Since the string copys in Dhrystone are of constants to local arrays, the compiler can guarantee that the strings are aligned and even do the string copy in-line. There is a divide in every iteration that can be eliminated. Clever compilers can get rid of the divide. High speed string compares can be written that compare a word at a time. These three tricks alone can almost double the Dhrystone rating of a machine. Fast string copy and compare are usefull library routines but they don't account for 30% of a typical program's time. The divide optimization would be part of a general strength reduction and could be more useful but I don't know how much more usefull. Mike Mahar -- Mike Mahar UUCP: {turtlevax, cae780}!weitek!mahar
colwell@mfci.UUCP (Robert Colwell) (07/13/88)
In article <391@attila.weitek.UUCP> mahar@attila.UUCP (Mike Mahar) writes: >I've noticed that the way people use the Dhrystone benchmark is very different >than the initial intent. It was supposed to be used to evaluate computer >architectures. Instead, it is used to evaluate C compilers. > >Given this the Dhrystone benchmark is used to evaluate compilers for the >various RISC processors. A very good optimizing compiler will eliminate >all the code in Dhrystone except the calls to times. >... >These three tricks alone can almost double the >Dhrystone rating of a machine. Fast string copy and compare are usefull >library routines but they don't account for 30% of a typical program's time. I think this demonstrates why Dhrystone has been nearly destroyed as a useful benchmark. It's clear at this point that one's Dhrystone rating is at best only weakly correlated to one's realizable performance on interesting systems code. I suspect that it is fast becoming completely uncorrelated, due to the rampant compiler cheating it allows. I don't understand why you think "a very good optimizing compiler will eliminate all the code in Dhrystone". The dead-code removal phase of the compiler isn't really all that important in terms of overall object code performance -- it mostly cleans up conceptual errors on the part of the programmer. Putting a lot of time and energy into that part of the compiler is silly if: a) you want more than just high Dhrystone numbers, and b) you could have put that effort into other optimizations that really meant something. I'm even willing to grant that optimizing away whole routines (or auto-inlining them) is legitimate, if and only if your compiler does that in Dhrystone to the same extent that it does on real code. In fact, if your compiler is pretty good at those kinds of optimizations, the performance measurement on this benchmark would be more accurate rather than less. Under these circumstances, the only issues we'd still have to shake our heads about would be Dhrystone's unrealistic cache utilization and the string manipulation effects you noted above. Using Dhrystone as a measure of compiler cleverness, in my opinion, is very nearly as crazy as using it as an integer throughput predictor. (Sorry, Reinhold, I think it's of great usefulness when evaluating effects on performance while changing architecture, instruction sets, or certain compiler strategies, but in the commercial world it's Toto in a town full of wicked witches). Bob Colwell mfci!colwell@uunet.uucp Multiflow Computer 175 N. Main St. Branford, CT 06405 203-488-6090
jbs@fenchurch.MIT.EDU (Jeff Siegal) (07/13/88)
In article <465@m3.mfci.UUCP> colwell@mfci.UUCP (Robert Colwell) writes: >I don't understand why you think "a very good optimizing compiler >will eliminate all the code in Dhrystone". The dead-code removal >phase of the compiler isn't really all that important [...] >it mostly cleans up conceptual errors on the part of the programmer. I though that Dhrystone 2.1 had been modified to make large amounts of dead-code elimination impossible by: 1) Printing out the results at the end 2) Dividing up the routines into separate source files and specifing that inter-module optimizations not be done. Given that the results must be printed out at the end, wouldn't the compiler be doing lots of constant folding rather than dead code elimination. I suppose after doing lots of loop unrolling and removing the resulting invariants from the loops, the loops would turn out to be empty and could be eliminated, but this doesn't seem very significant. Isn't constant folding pretty important to overall performance? Jeff Siegal
reiter@babbage.harvard.edu (Ehud Reiter) (07/14/88)
We seem to be starting up another argument about how accurately Dhrystone serves as a predictor of performance of real programs. I think, though, that the main benefit of Dhrystone and other benchmarks is not to predict performance (that's impossible, since no single figure of merit can predict the performance of widely differing application programs), but rather to help architecture designers design good computers, and compiler writers design good compilers, by giving them some real code which is at least slightly representative to play with. I think this point was perhaps made most strongly by Richard Gabriel, in his book on LISP benchmarks, where he claimed that the availability of a good benchmark suite enabled some compiler writers to improve the execution speed of their output by a factor of two or more. So, I think it is unfair to criticize Dhrystone on the basis that compiler writers have optimized their compilers to produce fast Dhrystone code, which means that Dhrystone performance is not representative of real application performance. This is true, but misses the point that the purpose of Dhrystone is to encourage compiler writers to build good optimizations into their systems. If the optimizations required for good Dhrystone performance include optimizations that are not relevant to real application programs, who cares - it doesn't do any harm to include rarely-used optimizations in a compiler. If optimizations which would be very useful in general are not required for good Dhrystone performance, this is a problem, but our response should not be to throw out Dhrystone, but rather to add to it a new section (or perhaps a whole new benchmark) which does require these optimizations to obtain good performance. Ehud Reiter reiter@harvard (ARPA,BITNET,UUCP) reiter@harvard.harvard.EDU (new ARPA)
chase@Ozona.orc.olivetti.com (David Chase) (07/14/88)
In article <465@m3.mfci.UUCP> colwell@mfci.UUCP (Robert Colwell) writes: >I don't understand why you think "a very good optimizing compiler >will eliminate all the code in Dhrystone". The dead-code removal >phase of the compiler isn't really all that important in terms of >overall object code performance -- it mostly cleans up conceptual ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ >errors on the part of the programmer. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ That's dead wrong. Every account of dead code elimination that I have ever seen points out that DCE cleans up after the optimizer. I suggest that you read (for instance) _Program Flow Analysis_ (eds. Muchnick and Jones. Prentice-Hall), or _Compilers: Principles, Techniques, and Tools_ (Aho, Sethi, and Ullman. Addison-Wesley). Use of macros and inline-functions is one fruitful source of dead code. Substitution of constant actual parameters into inlined subroutines can easily yield dead code. The only possible contradiction that I have seen to my "that's dead wrong" is a paper by some people from HP in last month's Programming Language Design and Implementation Conference (proceedings recently appeared in SIGPLAN notices; they've got sort of a dark blue-green cover). There they talk about maintaining data flow information as transformations are applied to the program--one could conceivably also clean up dead code as it is created in such a system (it appears that they do, to some extent). Nevertheless, inlining and macros will both generate dead code without any conceptual errors on the part of the programmer. David Chase Olivetti Research Center, Menlo Park, CA
patc@tekcrl.CRL.TEK.COM (Pat Caudill) (07/14/88)
Dead code elimination is used in other code re-writing. In the original article on the IBM 801 they talked about the compiler for PL.8 which used this technique. They would hoist invariant code by copying to the highest common thread, fold constants and then let common subexpression and dead code elimination remove the orginal code. You might want to read these articles they are in some old sigplan-sigarch proceeding on hardware support for programming languages. Pat Caudill patc@tekcrl.TEK.COM
eugene@pioneer.arc.nasa.gov.arpa (Eugene N. Miya) (07/15/88)
Yet another dull benchmarking argument. Oh, Ehud makes an interesting point. You use the word "prediction." Like in the film Apocolypse Now, where Brando asks Sheen about his "method," Sheen says, "What method?" I say "What prediction?" These stones aren't predictors, at best they are descriptors, nothing more. I don't see many predictive tools out there. If I did, I would have a tool which tells me a Dhrystone will run XXX fast on a YYY. This is a prediction. I runs the Stones and I get ZZZ, this is a verification attempt and description. If I try to relate ZZZ to payroll program (or what ever), that's a extrapolation (at best). Prediction is held in high esteem. I would like to see more prediction tools. Another gross generalization from --eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov resident cynic at the Rock of Ages Home for Retired Hackers: "Mailers?! HA!", "If my mail does not reach you, please accept my apology." {uunet,hplabs,ncar,decwrl,allegra,tektronix}!ames!aurora!eugene "Send mail, avoid follow-ups. If enough, I'll summarize."
mat@amdahl.uts.amdahl.com (Mike Taylor) (07/15/88)
In article <11802@ames.arc.nasa.gov>, eugene@pioneer.arc.nasa.gov.arpa (Eugene N. Miya) writes: > Prediction is held in high esteem. I would like to see more > prediction tools. But prediction requires the scientific method. That's not so much fun as myth and dogma. Everybody knows that computer performance is only susceptible to religious discussions of the meaning of MIPS. -- Mike Taylor ...!{hplabs,amdcad,sun}!amdahl!mat [ This may not reflect my opinion, let alone anyone else's. ]
walter@garth.UUCP (Walter Bays) (07/16/88)
In article 4936 reiter@harvard.UUCP (Ehud Reiter) writes:
...the main benefit of Dhrystone and other benchmarks is not to
predict performance (that's impossible, since no single figure of
merit can predict the performance of widely differing application
programs), but rather to help architecture designers design good
computers, and compiler writers design good compilers, by giving
them some real code which is at least slightly representative to
play with.
That's true, though for these purposes I also like to look at real
applications and at customer-written benchmarks, which tend to be more
representative of real applications. The great advantages of Dhrystone
are that it's easy to get, simple to run, and comparative figures are
readily available for many machines and compilers. Richardson has
done a real public service in making it easy for different people to
run exactly the same code, and in making users more cognizent of
performance (resulting in demands for more performance, resulting in
more performance)
The danger of Dhrystone is when people start to view it as a kind of
"miles- per-gallon" single figure of merit. Weicker's and Richardson's
warn against this in their advice on the use of the Dhrystone benchmark
That advice is very good.
If the optimizations required for good Dhrystone performance
include optimizations that are not relevant to real application
programs, who cares - it doesn't do any harm to include rarely-used
optimizations in a compiler.
I agree: a compiler should have the largest possible battery of
optimizations to choose from. But judging from postings in
comp.lang.c, there are many users who would disagree. Optimizations
have a cost in compilation time, even when not applied. An applied
optimization may have no benefit on a particular application, yet have
a cost in memory size (i.e., excessive inlining). Worst would be if an
"optimization" helped Dhrystones but hurt many application programs.
There are many optimizations that help some programs but hurt others,
and the compiler cannot always tell the difference.
To the extent that Dhrystones is used as a miles-per-gallon figure, the
temptation exists for compiler writers (computer architects?) to design
for Dhrystones at the expense of real applications, or to spend
precious resources speeding up Dhrystones that could have been used to
benefit ordinary programs.
--
------------------------------------------------------------------------------
My opinions are my own. Objects in mirror are closer than they appear.
E-Mail route: ...!pyramid!garth!walter (415) 852-2384
USPS: Intergraph APD, 2400 Geng Road, Palo Alto, California 94303
------------------------------------------------------------------------------
colwell@mfci.UUCP (Robert Colwell) (07/18/88)
In article <25384@oliveb.olivetti.com> chase@Ozona.UUCP (David Chase) writes: >In article <465@m3.mfci.UUCP> colwell@mfci.UUCP (Robert Colwell) writes: >>The dead-code removal >>phase of the compiler isn't really all that important in terms of >>overall object code performance -- it mostly cleans up conceptual > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ >>errors on the part of the programmer. > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > >That's dead wrong. Every account of dead code elimination that I have >ever seen points out that DCE cleans up after the optimizer. I >suggest that you read (for instance) _Program Flow Analysis_ (eds. >Muchnick and Jones. Prentice-Hall), or _Compilers: Principles, >Techniques, and Tools_ (Aho, Sethi, and Ullman. Addison-Wesley). > >Use of macros and inline-functions is one fruitful source of dead >code. Substitution of constant actual parameters into inlined >subroutines can easily yield dead code. I sit corrected, having been already taken to task by our compiler writers. On the other hand, I notice you didn't take up the central point I was ineptly trying to make, to wit: why should I believe that a compiler which does a decent job of dead-code removal will exhibit this ability to the same extent on Dhrystone as it would on the normal sources, which do not require the compiler to look across procedure call/function invocation boundaries? The "conceptual errors" I was referring to were calls to functions or procedures that didn't return anything and made no changes to global state; it wasn't clear to me (and still isn't) why I should make sure my compiler is really adept at finding and removing these, when the only way they arise is via broken programming (or synthetic benchmarks). Bob Colwell mfci!colwell@uunet.uucp Multiflow Computer 175 N. Main St. Branford, CT 06405 203-488-6090
dry2@pcrat.UUCP (Dhrystone) (12/04/88)
12/04/88 Here comes the Dhrystone 2.1 Integer CPU benchmark sources, in three parts: dry2shar.1 (Stuff I added) dry2shar.2 (Reinhold's C version) dry2shar.3 (Reinhold's Pascal version) Look in newsgroup "comp.arch" for results to follow. #! /bin/sh # This is a shell archive, meaning: # 1. Remove everything above the #! /bin/sh line. # 2. Save the resulting text in a file. # 3. Execute the file with /bin/sh (not csh) to create: # README.RER # clarify.doc # Makefile # submit.frm # pure2_1.dif # dhry_c.dif # This archive created: Sun Dec 4 02:06:52 1988 export PATH; PATH=/bin:/usr/bin:$PATH echo shar: "extracting 'README.RER'" '(4419 characters)' if test -f 'README.RER' then echo shar: "will not over-write existing file 'README.RER'" else sed 's/^X//' << \SHAR_EOF > 'README.RER' X12/04/88 X XHere is Reinhold Weicker's DHRYSTONE 2.1 benchmark, slightly modified Xby me for instrumentation purposes only. This is an integer CPU Xbenchmark. The differences between 2.0 and 2.1 are minor, and XI believe that 2.0 and 2.1 results are comparable. X XFor purists, I've enclosed context diffs to turn what I'm posting Xas 2.1 back into what Reinhold sent me as 2.1. The changes Xare in pure2_1.dif. As you can see, it is only the instrumentation Xthat is slightly different. X XBoth the C and the Pascal versions are here. The Ada version I Xhave is marked preliminary (not to be posted). X XI've enclosed a new submission form (note new address for mailings). XPlease deluge this mailbox (..!uunet!pcrat!dry2) with your results. XI'll summarize and repost when the dust clears. Please do not Xassume that I will pull posted results off of the net (I won't, its Xtoo much work). X XI've attempted to include a Makefile for UNIX and Microsoft C (with ndmake). XPay particular attention to the HZ parameter, even though your power may Xbe 50 or 60 hertz, your computer may not be. You may have to ask someone, Xread the manual, or check: X /usr/include/sys/param.h X /usr/include/limits.h (CLK_TCK==HZ) Xfor this information. X XREPEAT: HZ IS NOT NECESSARILY THE FREQUENCY OF YOUR AC POWER !!!!!! X XThere are two versions to run, one with register variables, and one Xwithout. Please let the benchmark run for 30,000 loops on sixteen Xbit machines, and for much longer (a minute or two) on faster machines. XPlease note that "time(2)" has a resolution of 1 second, and may give Xvariable results. No matter how time is measured, a sanity check with Xa stopwatch is prudent. We've run into systems that lie about time, Xand there is always the configuration error problem. When it comes Xto time measurement on UNIX, there is no widely adhered to standard. X XFor segmented architectures, it is appropriate to submit results for Xall memory models, as shown below. X XI dropped the CODESIZE information that I attempted to add the last Xtime I posted this. People couldn't follow directions, and the Xinformation wasn't very useful anyway. X XHere's a sample submission of results: X XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: IBM XMODEL: PC/AT XPROC: 80286 XCLOCK: 8 XOS: Venix XOVERSION: SVr2.3 XCOMPILER: AT&T cc XCVERSION: 11/8/84 XOPTIONS: -O XNOREG: 1450 XREG: 1450 XNOTES: HZ=60; Small Model XDATE: 03/04/88 XSUBMITTER: pcrat!rick (Rick Richardson) XMAILTO: uunet!pcrat!dry2 X XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: IBM XMODEL: PC/AT XPROC: 80286 XCLOCK: 8 XOS: Venix XOVERSION: SVr2.3 XCOMPILER: AT&T cc XCVERSION: 11/8/84 XOPTIONS: -O -Ml XNOREG: 1043 XREG: 1043 XNOTES: HZ=60; Large Model XDATE: 03/04/88 XSUBMITTER: pcrat!rick (Rick Richardson) XMAILTO: uunet!pcrat!dry2 X XThe program that processes submission forms is rather dumb. Please Xdo not change the order, add or removes lines in the form. If your XNOTES are longer than the space provided, then they are too long for Xthe summary. Keep it terse, please. X XA form consists of all lines between: X DHRYSTONE 2.1 BENCHMARK REPORTING FORM Xand X MAILTO: uunet!pcrat!dry2 Xboth lines must be present for the form to be processed. If Xa field does not apply or is not known, leave it blank. The fields Xare: X MANUF: Computer manufacturer, e.g. AT&T, IBM X MODEL: Model number of computer X PROC: If a microprocessor CPU, the part number, e.g. 68030 X CLOCK: Clock in Mhz, if known. Numeric only, e.g. 16.67 X OS: Operating system, e.g. UNIX X OVERSION: OS version, e.g. SVR3 X COMPILER: Compiler name, e.g. cc, Microsoft, or Green Hills X CVERSION: Compiler version, e.g. 5.10 X OPTIONS: Relevant compiler options, e.g. -O3 X NOREG: Dhrystones/second, no register attribute X REG: Dhrystones/second, with register attribute X NOTES: Additional, terse comments on one line X DATE: Date of test, US format MM/DD/YY X SUBMITTER: uucp or domain address (full name) X X XHZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ X XPlease, if you are using times(2) to measure time, indicate Xthe HZ value you used in the NOTES field. Something like this is OK: X X NOTES: HZ=100; no inlining... X XHZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ HZ X X-- X Rick Richardson, President, PC Research, Inc. X X(201) 542-3734 (voice, nights) OR (201) 389-8963 (voice, days) Xuunet!pcrat!rick (UUCP) rick%pcrat.uucp@uunet.uu.net (INTERNET) X uunet!pcrat!dry2 (Dhrystone submission forms only) SHAR_EOF if test 4419 -ne "`wc -c < 'README.RER'`" then echo shar: "error transmitting 'README.RER'" '(should have been 4419 characters)' fi fi echo shar: "extracting 'clarify.doc'" '(3195 characters)' if test -f 'clarify.doc' then echo shar: "will not over-write existing file 'clarify.doc'" else sed 's/^X//' << \SHAR_EOF > 'clarify.doc' XCLARIFICATION XThere seems to have been a great deal of confusion over what this Xbenchmark measures, and how to use these results. Let me try to clarify Xthis: X X 1) DHRYSTONE is a measure of processor+compiler efficiency in X executing a 'typical' program. The 'typical' program was X designed by measuring statistics on a great number of X 'real' programs. The 'typical' program was then written X by Reinhold P. Weicker using these statistics. The X program is balanced according to statement type, as well X as data type. X X 2) DHRYSTONE does not use floating point. Typical programs don't. X X 3) DHRYSTONE does not do I/O. Typical programs do, but then X we'd have a whole can of worms opened up. X X 4) DHRYSTONE does not contain much code that can be optimized X by vector processors. That is why a CRAY doesn't look real X fast, they weren't built to do this sort of computing. X X 5) DHRYSTONE does not measure OS performance, as it avoids X calling the O.S. The O.S. is indicated in the results only X to help in identifying the compiler technology. X X 6) DHRYSTONE is not perfect, but is a hell of a lot better than X the "sieve", or "SI". X X 7) DHRYSTONE gives results in dhrystones/second. Bigger X numbers are better. As a baseline, the original IBM PC X gives around 300-400 dhrystones/second with a good compiler. X The fastest machines today are approaching 100,000. X XIf somebody asked me to pick out the best machine for the money, I Xwouldn't look at just the results of DHRYSTONE. I'd probably: X X 1) Run DHRYSTONE to get a feel for the compiler+processor X speed. X 2) Run any number of benchmarks to check disk I/O bandwidth, X using both sequential and random read/writes. X 3) Run a multitasking benchmark to check multi-user response X time. Typically, these benchmarks run several types of X programs such as editors, shell scripts, sorts, compiles, X and plot the results against the number of simulated users. X 4) If appropriate for the intended use, run something like X WHETSTONE, to determine floating point performance. X 5) If appropriate for intended use, run some programs which do X vector and matrix computations. X 6) Figure out what the box will: X - cost to buy X - cost to operate and maintain X - be worth when it is sold X - be worth if the manufacturer goes out of business X 7) Having done the above, I probably have a hand-full of X machines which meet my price/performance requirements. X Now, I find out if the applications programs I'd like X to use will run on any of these machines. I also find X out how much interest people have in writing new software X for the machine, and look carefully at the migration path X I will have to take when I reach the (inevitable) limits X of the machine. X XTo summarize, DHRYSTONES by themselves are not anything more than Xa way to win free beers when arguing 'Box-A versus Box-B' religion. XThey do provide insight into Box-A/Compiler-A versus Box-A/Compiler-B Xcomparisons. X X Rick Richardson X PC Research, Inc. X (201) 389-8963 (9-17 EST) X (201) 542-3734 (7-9,17-24 EST) X ...!uunet!pcrat!rick (normal mail) X ...!uunet!pcrat!dry2 (results only) SHAR_EOF if test 3195 -ne "`wc -c < 'clarify.doc'`" then echo shar: "error transmitting 'clarify.doc'" '(should have been 3195 characters)' fi fi echo shar: "extracting 'Makefile'" '(3956 characters)' if test -f 'Makefile' then echo shar: "will not over-write existing file 'Makefile'" else sed 's/^X//' << \SHAR_EOF > 'Makefile' X# X# Adjust for your system! X# X# Common options for generic UNIX and Microsoft C (under DOS) X# are listed here. You can change them by switching the order, X# placing the ones you want last. Pay particular attention to X# the HZ parameter, which may or may not be listed in some X# header file on your system, such as <sys/param.h> or <limits.h> X# (as CLK_TCK). Even if it is listed, it may be incorrect. X# Also, some operating systems (notably some (all?) versions X# of Microport UNIX) lie about the time. Sanity check with a X# stopwatch. X# X# For Microsoft C under DOS, you need a real make, not MSC make, X# to run this Makefile. The public domain "ndmake" will suffice. X# XCC= cl # C compiler name goes here (MSC) XCC= cc # C compiler name goes here (UNIX) X XPROGS= msc # Programs to build (MSC) XPROGS= unix # Programs to build (UNIX) X XTIME_FUNC= -DMSC_CLOCK # Use Microsoft clock() for measurement XTIME_FUNC= -DTIME # Use time(2) for measurement XTIME_FUNC= -DTIMES # Use times(2) for measurement XHZ= 50 # Frequency of times(2) clock ticks XHZ= 60 # Frequency of times(2) clock ticks XHZ= 100 # Frequency of times(2) clock ticks XHZ= 1 # Give bogus result unless changed! X XSTRUCTASSIGN= -DNOSTRUCTASSIGN # Compiler cannot assign structs XSTRUCTASSIGN= # Compiler can assign structs X XENUMS= -DNOENUMS # Compiler doesn't have enum type XENUMS= # Compiler does have enum type X XOPTIMIZE= -Ox -G2 # Optimization Level (MSC, 80286) XOPTIMIZE= -O # Optimization Level (generic UNIX) X XLFLAGS= #Loader Flags X XCFLAGS= $(OPTIMIZE) $(TIME_FUNC) -DHZ=$(HZ) $(ENUMS) $(STRUCTASSIGN) $(CFL) X X# X# You shouldn't need to touch the rest X# XSRC= dhry_1.c dhry_2.c XHDR= dhry.h X XUNIX_PROGS= dry2 dry2reg XMSC_PROGS= sdry2.exe sdry2reg.exe mdry2.exe mdry2reg.exe \ X ldry2.exe ldry2reg.exe cdry2.exe cdry2reg.exe \ X hdry2.exe hdry2reg.exe X X# Files added by rer: XFILES1= README.RER clarify.doc Makefile submit.frm pure2_1.dif \ X dhry_c.dif X# Reinhold's files: XFILES2= README RATIONALE $(HDR) $(SRC) XFILES3= dhry.p X Xall: $(PROGS) X Xunix: $(UNIX_PROGS) X Xmsc: $(MSC_PROGS) X Xdry2: $(SRC) $(HDR) X $(CC) $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xdry2reg: $(SRC) $(HDR) X $(CC) $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xsdry2.exe: $(SRC) $(HDR) X $(CC) -AS $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xsdry2reg.exe: $(SRC) $(HDR) X $(CC) -AS $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xmdry2.exe: $(SRC) $(HDR) X $(CC) -AM $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xmdry2reg.exe: $(SRC) $(HDR) X $(CC) -AM $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xldry2.exe: $(SRC) $(HDR) X $(CC) -AL $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xldry2reg.exe: $(SRC) $(HDR) X $(CC) -AL $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xcdry2.exe: $(SRC) $(HDR) X $(CC) -AC $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xcdry2reg.exe: $(SRC) $(HDR) X $(CC) -AC $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xhdry2.exe: $(SRC) $(HDR) X $(CC) -AH $(CFLAGS) $(SRC) $(LFLAGS) -o $@ X Xhdry2reg.exe: $(SRC) $(HDR) X $(CC) -AH $(CFLAGS) -DREG=register $(SRC) $(LFLAGS) -o $@ X Xshar: dry2shar.1 dry2shar.2 dry2shar.3 X Xdry2.arc: $(FILES1) $(FILES2) $(FILES3) X arc a dry2.arc $(FILES1) $(FILES2) $(FILES3) X Xdry2shar.1: $(FILES1) X shar -vc -p X $(FILES1) >$@ X Xdry2shar.2: $(FILES2) X shar -vc -p X $(FILES2) >$@ X Xdry2shar.3: $(FILES3) X shar -v -p X $(FILES3) >$@ X Xclean: X -rm -f *.o *.obj X Xclobber: clean X -rm -f $(UNIX_PROGS) $(MSC_PROGS) dry2shar.* dry2.arc X Xpost: dry2shar.1 dry2shar.2 dry2shar.3 X for i in 1 2 3;\ X do\ X cat HEADERS BOILER.$$i dry2shar.$$i |\ X inews -h -t "Dhrystone 2.1 ($$i of 3)" -n comp.arch;\ X done X Xrepost: dry2shar.1 dry2shar.2 dry2shar.3 X for i in 3;\ X do\ X cat HEADERS BOILER.$$i dry2shar.$$i |\ X inews -h -t "REPOST: Dhrystone 2.1 ($$i of 3)" -n comp.arch;\ X done X Xmail: dry2shar.1 dry2shar.2 dry2shar.3 X for i in 1 2 3;\ X do\ X cat BOILER.$$i dry2shar.$$i |\ X mailx -s "Dhrystone 2.1 ($$i of 3)" $(ADDR);\ X done X Xdos: X doscopy -a $(FILES1) $(FILES2) $(FILES3) dos!a: SHAR_EOF if test 3956 -ne "`wc -c < 'Makefile'`" then echo shar: "error transmitting 'Makefile'" '(should have been 3956 characters)' fi fi echo shar: "extracting 'submit.frm'" '(170 characters)' if test -f 'submit.frm' then echo shar: "will not over-write existing file 'submit.frm'" else sed 's/^X//' << \SHAR_EOF > 'submit.frm' XDHRYSTONE 2.1 BENCHMARK REPORTING FORM XMANUF: XMODEL: XPROC: XCLOCK: XOS: XOVERSION: XCOMPILER: XCVERSION: XOPTIONS: XNOREG: XREG: XNOTES: XDATE: XSUBMITTER: XMAILTO: uunet!pcrat!dry2 SHAR_EOF if test 170 -ne "`wc -c < 'submit.frm'`" then echo shar: "error transmitting 'submit.frm'" '(should have been 170 characters)' fi fi echo shar: "extracting 'pure2_1.dif'" '(3590 characters)' if test -f 'pure2_1.dif' then echo shar: "will not over-write existing file 'pure2_1.dif'" else sed 's/^X//' << \SHAR_EOF > 'pure2_1.dif' X*** rer/dhry.h Sun Dec 4 00:08:26 1988 X--- weicker/dhry.h Sun Dec 4 00:09:00 1988 X*************** X*** 38,45 X * PC Research. Inc. X * 94 Apple Orchard Drive X * Tinton Falls, NJ 07724 X! * Phone: (201) 389-8963 (9-17 EST) X! * Usenet: ...!uunet!pcrat!rick X * X * Please send results to Rick Richardson and/or Reinhold Weicker. X * Complete information should be given on hardware and software used. X X--- 38,45 ----- X * PC Research. Inc. X * 94 Apple Orchard Drive X * Tinton Falls, NJ 07724 X! * Phone: (201) 834-1378 (9-17 EST) X! * Usenet: ...!seismo!uunet!pcrat!rick X * X * Please send results to Rick Richardson and/or Reinhold Weicker. X * Complete information should be given on hardware and software used. X*************** X*** 348,354 X /* Compiler and system dependent definitions: */ X X #ifndef TIME X- #undef TIMES X #define TIMES X #endif X /* Use times(2) time function unless */ X X--- 348,353 ----- X /* Compiler and system dependent definitions: */ X X #ifndef TIME X #define TIMES X #endif X /* Use times(2) time function unless */ X*************** X*** 353,366 X #endif X /* Use times(2) time function unless */ X /* explicitly defined otherwise */ X- X- #ifdef MSC_CLOCK X- #undef HZ X- #undef TIMES X- #include <time.h> X- #define HZ CLK_TCK X- #endif X- /* Use Microsoft C hi-res clock */ X X #ifdef TIMES X #include <sys/types.h> X X--- 352,357 ----- X #endif X /* Use times(2) time function unless */ X /* explicitly defined otherwise */ X X #ifdef TIMES X #include <sys/types.h> X*** rer/dhry_1.c Sun Dec 4 00:08:36 1988 X--- weicker/dhry_1.c Sun Dec 4 00:09:09 1988 X*************** X*** 47,53 X struct tms time_info; X extern int times (); X /* see library function "times" */ X! #define Too_Small_Time (2*HZ) X /* Measurements should last at least about 2 seconds */ X #endif X #ifdef TIME X X--- 47,53 ----- X struct tms time_info; X extern int times (); X /* see library function "times" */ X! #define Too_Small_Time 120 X /* Measurements should last at least about 2 seconds */ X #endif X #ifdef TIME X*************** X*** 56,65 X #define Too_Small_Time 2 X /* Measurements should last at least 2 seconds */ X #endif X- #ifdef MSC_CLOCK X- extern clock_t clock(); X- #define Too_Small_Time (2*HZ) X- #endif X X long Begin_Time, X End_Time, X X--- 56,61 ----- X #define Too_Small_Time 2 X /* Measurements should last at least 2 seconds */ X #endif X X long Begin_Time, X End_Time, X*************** X*** 139,147 X #ifdef TIME X Begin_Time = time ( (long *) 0); X #endif X- #ifdef MSC_CLOCK X- Begin_Time = clock(); X- #endif X X for (Run_Index = 1; Run_Index <= Number_Of_Runs; ++Run_Index) X { X X--- 135,140 ----- X #ifdef TIME X Begin_Time = time ( (long *) 0); X #endif X X for (Run_Index = 1; Run_Index <= Number_Of_Runs; ++Run_Index) X { X*************** X*** 199,207 X #endif X #ifdef TIME X End_Time = time ( (long *) 0); X- #endif X- #ifdef MSC_CLOCK X- End_Time = clock(); X #endif X X printf ("Execution ends\n"); X X--- 192,197 ----- X #endif X #ifdef TIME X End_Time = time ( (long *) 0); X #endif X X printf ("Execution ends\n"); SHAR_EOF if test 3590 -ne "`wc -c < 'pure2_1.dif'`" then echo shar: "error transmitting 'pure2_1.dif'" '(should have been 3590 characters)' fi fi echo shar: "extracting 'dhry_c.dif'" '(4354 characters)' if test -f 'dhry_c.dif' then echo shar: "will not over-write existing file 'dhry_c.dif'" else sed 's/^X//' << \SHAR_EOF > 'dhry_c.dif' X7c7 X< * Version: C, Version 2.1 X--- X> * Version: C, Version 2.0 X9c9 X< * File: dhry.h (part 1 of 3) X--- X> * File: dhry_global.h (part 1 of 3) X11c11 X< * Date: May 25, 1988 X--- X> * Date: March 3, 1988 X30c30 X< * In addition, Berkeley UNIX system calls "times ()" or "time ()" X--- X> * In addition, UNIX system calls "times ()" or "time ()" X44c44 X< * Please send results to Rick Richardson and/or Reinhold Weicker. X--- X> * Please send results to Reinhold Weicker and/or Rick Richardson. X59c59 X< * History: This version C/2.1 has been made for two reasons: X--- X> * History: This version C/2.0 has been made for two reasons: X123,129d122 X< * Version 2.1 is identical to version 2.0 distributed via X< * the UNIX network Usenet in March 1988 except that it corrects X< * some minor deficiencies that were found by users of version 2.0. X< * The only change within the measurement loop is that a X< * non-executed "else" part was added to the "if" statement in X< * Func_3, and a non-executed "else" part removed from Proc_3. X< * X165,167c158,160 X< * -DHZ=nnn X< * In Berkeley UNIX, the function "times" returns process X< * time in 1/HZ seconds, with HZ = 60 for most systems. X--- X> * -DHZ=nnn (default: 60) X> * The function "times" returns process times in X> * 1/HZ seconds, with HZ = 60 for most systems. X169c162 X< * A VALUE. X--- X> * THE DEFAULT VALUE. X176,178c169,171 X< * - dhry.h (this file, containing global definitions and comments) X< * - dhry_1.c (containing the code corresponding to Ada package Pack_1) X< * - dhry_2.c (containing the code corresponding to Ada package Pack_2) X--- X> * - dhry_global.h (this file, containing global definitions and comments) X> * - dhry_pack_1.c (containing the code corresponding to Ada package Pack_1) X> * - dhry_pack_2.c (containing the code corresponding to Ada package Pack_2) X350a344 X> #ifndef TIMES X353,354c347,354 X< /* Use times(2) time function unless */ X< /* explicitly defined otherwise */ X--- X> #endif X> /* Use "times" function for measurement */ X> /* unless explicitly defined otherwise */ X> #ifndef HZ X> #define HZ 60 X> #endif X> /* Use HZ = 60 for "times" function */ X> /* unless explicitly defined otherwise */ X363c363 X< /* Berkeley UNIX C returns process times in seconds/HZ */ X--- X> /* UNIX C returns process times in seconds/HZ */ X7c7 X< * Version: C, Version 2.1 X--- X> * Version: C, Version 2.0 X9c9 X< * File: dhry_1.c (part 2 of 3) X--- X> * File: dhry_pack_1.c (part 2 of 3) X11c11 X< * Date: May 25, 1988 X--- X> * Date: March 3, 1988 X18c18 X< #include "dhry.h" X--- X> #include "dhry_global.h" X50,51d49 X< #define Too_Small_Time 120 X< /* Measurements should last at least about 2 seconds */ X55a54,55 X> #endif X> X58d57 X< #endif X73a73 X> X84a85 X> X99,100c100,102 X< /* Was missing in published program. Without this statement, */ X< /* Arr_2_Glob [8][7] would have an undefined value. */ X--- X> /* Was missing in published program. Without this */ X> /* initialization, Arr_2_Glob [8][7] would have an */ X> /* undefined value. */ X105c107 X< printf ("Dhrystone Benchmark, Version 2.1 (Language: C)\n"); X--- X> printf ("Dhrystone Benchmark, Version 2.0 (Language: C)\n"); X281c283 X< /******************/ X--- X> /**********************/ X338c340 X< /******************/ X--- X> /**********************/ X347a350,351 X> else /* not executed */ X> Int_Glob = 100; X349a354 X> X7c7 X< * Version: C, Version 2.1 X--- X> * Version: C, Version 2.0 X9c9 X< * File: dhry_2.c (part 3 of 3) X--- X> * File: dhry_pack_2.c (part 3 of 3) X11c11 X< * Date: May 25, 1988 X--- X> * Date: March 3, 1988 X18c18 X< #include "dhry.h" X--- X> #include "dhry_global.h" X189,190d188 X< else /* not executed */ X< return (false); SHAR_EOF if test 4354 -ne "`wc -c < 'dhry_c.dif'`" then echo shar: "error transmitting 'dhry_c.dif'" '(should have been 4354 characters)' fi fi exit 0 # End of shell archive
dry2@pcrat.UUCP (Dhrystone) (12/04/88)
#! /bin/sh # This is a shell archive, meaning: # 1. Remove everything above the #! /bin/sh line. # 2. Save the resulting text in a file. # 3. Execute the file with /bin/sh (not csh) to create: # README # RATIONALE # dhry.h # dhry_1.c # dhry_2.c # This archive created: Sun Dec 4 00:42:33 1988 export PATH; PATH=/bin:/usr/bin:$PATH echo shar: "extracting 'README'" '(2308 characters)' if test -f 'README' then echo shar: "will not over-write existing file 'README'" else sed 's/^X//' << \SHAR_EOF > 'README' XThis floppy disk contains the source code of the Dhrystone benchmark X(version 2.1) in three languages: X XFiles containing the Ada version (*.s: Specifications, *.b: Bodies): X X d_global.s X d_main.b X d_pack_1.b X d_pack_1.s X d_pack_2.b X d_pack_2.s X X (The file names have been abbreviated to 8 characters plus extension X for MS-DOS) X XFiles containing the C version (*.h: Header File, *.c: C Modules) X X dhry.h X dhry_1.c X dhry_2.c X XFile containing the Pascal version: X X dhry.p X XThe file RATIONALE contains the article X X "Dhrystone Benchmark: Rationale for Version 2 and Measurement Rules" X Xwhich has been published, together with the C source code (Version 2.0), Xin SIGPLAN Notices vol. 23, no. 8 (Aug. 1988), pp. 49-62. XThis article explains all changes that have been made for Version 2, Xcompared with the version of the original publication Xin Communications of the ACM vol. 27, no. 10 (Oct. 1984), pp. 1013-1030. XIt also contains "ground rules" for benchmarking with Dhrystone Xwhich should be followed by everyone who uses the program and publishes XDhrystone results. X XCompared with the Version 2.0 published in SIGPLAN Notices, Version 2.1 Xcontains a few corrections that have been made after Version 2.0 was Xdistriobuted over the UNIX network Usenet. These small differences between XVersion 2.0 and 2.1 should not affect execution time measurements. XFor those who want to compare the exact contents of both versions, Xthe file "dhry_c.dif" contains the differences between the two versions, Xas generated by a file comparison of the corresponding files with the XUNIX utility "diff". X XRecipients of this floppy disk who perform measurements are asked Xto send measurement results to the author and/or to Rick Richardson. XRick Richardson publishes regularly Dhrystone results on the UNIX network XUsenet. For submissions of results to him (preferably by electronic mail, Xsee address in the program header), he has provided a form which is contained Xin the file "submit.frm". X XAugust 17, 1988 X X Reinhold P. Weicker X Siemens AG, E STE 35 X Postfach 3240 X D-8520 Erlangen X Germany (West) X X Phone: [xxx-49]-9131-7-20330 (8-17 Central European Time) X Usenet: ..!mcvax!unido!estevax!weicker SHAR_EOF if test 2308 -ne "`wc -c < 'README'`" then echo shar: "error transmitting 'README'" '(should have been 2308 characters)' fi fi echo shar: "extracting 'RATIONALE'" '(18793 characters)' if test -f 'RATIONALE' then echo shar: "will not over-write existing file 'RATIONALE'" else sed 's/^X//' << \SHAR_EOF > 'RATIONALE' X X X Dhrystone Benchmark: Rationale for Version 2 and Measurement Rules X X X Reinhold P. Weicker X Siemens AG, E STE 35 X Postfach 3240 X D-8520 Erlangen X Germany (West) X X X X X1. Why a Version 2 of Dhrystone? X XThe Dhrystone benchmark program [1] has become a popular benchmark for XCPU/compiler performance measurement, in particular in the area of Xminicomputers, workstations, PC's and microprocesors. It apparently satisfies Xa need for an easy-to-use integer benchmark; it gives a first performance Xindication which is more meaningful than MIPS numbers which, in their literal Xmeaning (million instructions per second), cannot be used across different Xinstruction sets (e.g. RISC vs. CISC). With the increasing use of the Xbenchmark, it seems necessary to reconsider the benchmark and to check whether Xit can still fulfill this function. Version 2 of Dhrystone is the result of Xsuch a re-evaluation, it has been made for two reasons: X Xo Dhrystone has been published in Ada [1], and Versions in Ada, Pascal and C X have been distributed by Reinhold Weicker via floppy disk. However, the X version that was used most often for benchmarking has been the version made X by Rick Richardson by another translation from the Ada version into the C X programming language, this has been the version distributed via the UNIX X network Usenet [2]. X X There is an obvious need for a common C version of Dhrystone, since C is at X present the most popular system programming language for the class of X systems (microcomputers, minicomputers, workstations) where Dhrystone is X used most. There should be, as far as possible, only one C version of X Dhrystone such that results can be compared without restrictions. In the X past, the C versions distributed by Rick Richardson (Version 1.1) and by X Reinhold Weicker had small (though not significant) differences. X X Together with the new C version, the Ada and Pascal versions have been X updated as well. X Xo As far as it is possible without changes to the Dhrystone statistics, X optimizing compilers should be prevented from removing significant X statements. It has turned out in the past that optimizing compilers X suppressed code generation for too many statements (by "dead code removal" X or "dead variable elimination"). This has lead to the danger that X benchmarking results obtained by a naive application of Dhrystone - without X inspection of the code that was generated - could become meaningless. X XThe overall policiy for version 2 has been that the distribution of Xstatements, operand types and operand locality described in [1] should remain Xunchanged as much as possible. (Very few changes were necessary; their impact Xshould be negligible.) Also, the order of statements should remain unchanged. XAlthough I am aware of some critical remarks on the benchmark - I agree with Xseveral of them - and know some suggestions for improvement, I didn't want to Xchange the benchmark into something different from what has become known as X"Dhrystone"; the confusion generated by such a change would probably outweight Xthe benefits. If I were to write a new benchmark program, I wouldn't give it Xthe name "Dhrystone" since this denotes the program published in [1]. XHowever, I do recognize the need for a larger number of representative Xprograms that can be used as benchmarks; users should always be encouraged to Xuse more than just one benchmark. X XThe new versions (version 2.1 for C, Pascal and Ada) will be distributed as Xwidely as possible. (Version 2.1 differs from version 2.0 distributed via the XUNIX Network Usenet in March 1988 only in a few corrections for minor Xdeficiencies found by users of version 2.0.) Readers who want to use the Xbenchmark for their own measurements can obtain a copy in machine-readable Xform on floppy disk (MS-DOS or XENIX format) from the author. X X X2. Overall Characteristics of Version 2 X XIn general, version 2 follows - in the parts that are significant for Xperformance measurement, i.e. within the measurement loop - the published X(Ada) version and the C versions previously distributed. Where the versions Xdistributed by Rick Richardson [2] and Reinhold Weicker have been different, Xit follows the version distributed by Reinhold Weicker. (However, the Xdifferences have been so small that their impact on execution time in all Xlikelihood has been negligible.) The initialization and UNIX instrumentation Xpart - which had been omitted in [1] - follows mostly the ideas of Rick XRichardson [2]. However, any changes in the initialization part and in the Xprinting of the result have no impact on performance measurement since they Xare outside the measaurement loop. As a concession to older compilers, names Xhave been made unique within the first 8 characters for the C version. X XThe original publication of Dhrystone did not contain any statements for time Xmeasurement since they are necessarily system-dependent. However, it turned Xout that it is not enough just to inclose the main procedure of Dhrystone in a Xloop and to measure the execution time. If the variables that are computed Xare not used somehow, there is the danger that the compiler considers them as X"dead variables" and suppresses code generation for a part of the statements. XTherefore in version 2 all variables of "main" are printed at the end of the Xprogram. This also permits some plausibility control for correct execution of Xthe benchmark. X XAt several places in the benchmark, code has been added, but only in branches Xthat are not executed. The intention is that optimizing compilers should be Xprevented from moving code out of the measurement loop, or from removing code Xaltogether. Statements that are executed have been changed in very few places Xonly. In these cases, only the role of some operands has been changed, and it Xwas made sure that the numbers defining the "Dhrystone distribution" X(distribution of statements, operand types and locality) still hold as much as Xpossible. Except for sophisticated optimizing compilers, execution times for Xversion 2.1 should be the same as for previous versions. X XBecause of the self-imposed limitation that the order and distribution of the Xexecuted statements should not be changed, there are still cases where Xoptimizing compilers may not generate code for some statements. To a certain Xdegree, this is unavoidable for small synthetic benchmarks. Users of the Xbenchmark are advised to check code listings whether code is generated for all Xstatements of Dhrystone. X XContrary to the suggestion in the published paper and its realization in the Xversions previously distributed, no attempt has been made to subtract the time Xfor the measurement loop overhead. (This calculation has proven difficult to Ximplement in a correct way, and its omission makes the program simpler.) XHowever, since the loop check is now part of the benchmark, this does have an Ximpact - though a very minor one - on the distribution statistics which have Xbeen updated for this version. X X X3. Discussion of Individual Changes X XIn this section, all changes are described that affect the measurement loop Xand that are not just renamings of variables. All remarks refer to the C Xversion; the other language versions have been updated similarly. X XIn addition to adding the measurement loop and the printout statements, Xchanges have been made at the following places: X Xo In procedure "main", three statements have been added in the non-executed X "then" part of the statement X X if (Enum_Loc == Func_1 (Ch_Index, 'C')) X X they are X X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 3'RD STRING"); X Int_2_Loc = Run_Index; X Int_Glob = Run_Index; X X The string assignment prevents movement of the preceding assignment to X Str_2_Loc (5'th statement of "main") out of the measurement loop (This X probably will not happen for the C version, but it did happen with another X language and compiler.) The assignment to Int_2_Loc prevents value X propagation for Int_2_Loc, and the assignment to Int_Glob makes the value of X Int_Glob possibly dependent from the value of Run_Index. X Xo In the three arithmetic computations at the end of the measurement loop in X "main ", the role of some variables has been exchanged, to prevent the X division from just cancelling out the multiplication as it was in [1]. A X very smart compiler might have recognized this and suppressed code X generation for the division. X Xo For Proc_2, no code has been changed, but the values of the actual parameter X have changed due to changes in "main". X Xo In Proc_4, the second assignment has been changed from X X Bool_Loc = Bool_Loc | Bool_Glob; X X to X X Bool_Glob = Bool_Loc | Bool_Glob; X X It now assigns a value to a global variable instead of a local variable X (Bool_Loc); Bool_Loc would be a "dead variable" which is not used X afterwards. X Xo In Func_1, the statement X X Ch_1_Glob = Ch_1_Loc; X X was added in the non-executed "else" part of the "if" statement, to prevent X the suppression of code generation for the assignment to Ch_1_Loc. X Xo In Func_2, the second character comparison statement has been changed to X X if (Ch_Loc == 'R') X X ('R' instead of 'X') because a comparison with 'X' is implied in the X preceding "if" statement. X X Also in Func_2, the statement X X Int_Glob = Int_Loc; X X has been added in the non-executed part of the last "if" statement, in order X to prevent Int_Loc from becoming a dead variable. X Xo In Func_3, a non-executed "else" part has been added to the "if" statement. X While the program would not be incorrect without this "else" part, it is X considered bad programming practice if a function can be left without a X return value. X X To compensate for this change, the (non-executed) "else" part in the "if" X statement of Proc_3 was removed. X XThe distribution statistics have been changed only by the addition of the Xmeasurement loop iteration (1 additional statement, 4 additional local integer Xoperands) and by the change in Proc_4 (one operand changed from local to Xglobal). The distribution statistics in the comment headers have been updated Xaccordingly. X X X4. String Operations X XThe string operations (string assignment and string comparison) have not been Xchanged, to keep the program consistent with the original version. X XThere has been some concern that the string operations are over-represented in Xthe program, and that execution time is dominated by these operations. This Xwas true in particular when optimizing compilers removed too much code in the Xmain part of the program, this should have been mitigated in version 2. X XIt should be noted that this is a language-dependent issue: Dhrystone was Xfirst published in Ada, and with Ada or Pascal semantics, the time spent in Xthe string operations is, at least in all implementations known to me, Xconsiderably smaller. In Ada and Pascal, assignment and comparison of strings Xare operators defined in the language, and the upper bounds of the strings Xoccuring in Dhrystone are part of the type information known at compilation Xtime. The compilers can therefore generate efficient inline code. In C, Xstring assignemt and comparisons are not part of the language, so the string Xoperations must be expressed in terms of the C library functions "strcpy" and X"strcmp". (ANSI C allows an implementation to use inline code for these Xfunctions.) In addition to the overhead caused by additional function calls, Xthese functions are defined for null-terminated strings where the length of Xthe strings is not known at compilation time; the function has to check every Xbyte for the termination condition (the null byte). X XObviously, a C library which includes efficiently coded "strcpy" and "strcmp" Xfunctions helps to obtain good Dhrystone results. However, I don't think that Xthis is unfair since string functions do occur quite frequently in real Xprograms (editors, command interpreters, etc.). If the strings functions are Ximplemented efficiently, this helps real programs as well as benchmark Xprograms. X XI admit that the string comparison in Dhrystone terminates later (after Xscanning 20 characters) than most string comparisons in real programs. For Xconsistency with the original benchmark, I didn't change the program despite Xthis weakness. X X X5. Intended Use of Dhrystone X XWhen Dhrystone is used, the following "ground rules" apply: X Xo Separate compilation (Ada and C versions) X X As mentioned in [1], Dhrystone was written to reflect actual programming X practice in systems programming. The division into several compilation X units (5 in the Ada version, 2 in the C version) is intended, as is the X distribution of inter-module and intra-module subprogram calls. Although on X many systems there will be no difference in execution time to a Dhrystone X version where all compilation units are merged into one file, the rule is X that separate compilation should be used. The intention is that real X programming practice, where programs consist of several independently X compiled units, should be reflected. This also has implies that the X compiler, while compiling one unit, has no information about the use of X variables, register allocation etc. occuring in other compilation units. X Although in real life compilation units will probably be larger, the X intention is that these effects of separate compilation are modeled in X Dhrystone. X X A few language systems have post-linkage optimization available (e.g., final X register allocation is performed after linkage). This is a borderline case: X Post-linkage optimization involves additional program preparation time X (although not as much as compilation in one unit) which may prevent its X general use in practical programming. I think that since it defeats the X intentions given above, it should not be used for Dhrystone. X X Unfortunately, ISO/ANSI Pascal does not contain language features for X separate compilation. Although most commercial Pascal compilers provide X separate compilation in some way, we cannot use it for Dhrystone since such X a version would not be portable. Therefore, no attempt has been made to X provide a Pascal version with several compilation units. X Xo No procedure merging X X Although Dhrystone contains some very short procedures where execution would X benefit from procedure merging (inlining, macro expansion of procedures), X procedure merging is not to be used. The reason is that the percentage of X procedure and function calls is part of the "Dhrystone distribution" of X statements contained in [1]. This restriction does not hold for the string X functions of the C version since ANSI C allows an implementation to use X inline code for these functions. X Xo Other optimizations are allowed, but they should be indicated X X It is often hard to draw an exact line between "normal code generation" and X "optimization" in compilers: Some compilers perform operations by default X that are invoked in other compilers only when optimization is explicitly X requested. Also, we cannot avoid that in benchmarking people try to achieve X results that look as good as possible. Therefore, optimizations performed X by compilers - other than those listed above - are not forbidden when X Dhrystone execution times are measured. Dhrystone is not intended to be X non-optimizable but is intended to be similarly optimizable as normal X programs. For example, there are several places in Dhrystone where X performance benefits from optimizations like common subexpression X elimination, value propagation etc., but normal programs usually also X benefit from these optimizations. Therefore, no effort was made to X artificially prevent such optimizations. However, measurement reports X should indicate which compiler optimization levels have been used, and X reporting results with different levels of compiler optimization for the X same hardware is encouraged. X Xo Default results are those without "register" declarations (C version) X X When Dhrystone results are quoted without additional qualification, they X should be understood as results obtained without use of the "register" X attribute. Good compilers should be able to make good use of registers even X without explicit register declarations ([3], p. 193). X XOf course, for experimental purposes, post-linkage optimization, procedure Xmerging and/or compilation in one unit can be done to determine their effects. XHowever, Dhrystone numbers obtained under these conditions should be Xexplicitly marked as such; "normal" Dhrystone results should be understood as Xresults obtained following the ground rules listed above. X XIn any case, for serious performance evaluation, users are advised to ask for Xcode listings and to check them carefully. In this way, when results for Xdifferent systems are compared, the reader can get a feeling how much Xperformance difference is due to compiler optimization and how much is due to Xhardware speed. X X X6. Acknowledgements X XThe C version 2.1 of Dhrystone has been developed in cooperation with Rick XRichardson (Tinton Falls, NJ), it incorporates many ideas from the "Version X1.1" distributed previously by him over the UNIX network Usenet. Through his Xactivity with Usenet, Rick Richardson has made a very valuable contribution to Xthe dissemination of the benchmark. I also thank Chaim Benedelac (National XSemiconductor), David Ditzel (SUN), Earl Killian and John Mashey (MIPS), Alan XSmith and Rafael Saavedra-Barrera (UC at Berkeley) for their help with Xcomments on earlier versions of the benchmark. X X X7. Bibliography X X[1] X Reinhold P. Weicker: Dhrystone: A Synthetic Systems Programming Benchmark. X Communications of the ACM 27, 10 (Oct. 1984), 1013-1030 X X[2] X Rick Richardson: Dhrystone 1.1 Benchmark Summary (and Program Text) X Informal Distribution via "Usenet", Last Version Known to me: Sept. 21, X 1987 X X[3] X Brian W. Kernighan and Dennis M. Ritchie: The C Programming Language. X Prentice-Hall, Englewood Cliffs (NJ) 1978 X SHAR_EOF if test 18793 -ne "`wc -c < 'RATIONALE'`" then echo shar: "error transmitting 'RATIONALE'" '(should have been 18793 characters)' fi fi echo shar: "extracting 'dhry.h'" '(18556 characters)' if test -f 'dhry.h' then echo shar: "will not over-write existing file 'dhry.h'" else sed 's/^X//' << \SHAR_EOF > 'dhry.h' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry.h (part 1 of 3) X * X * Date: May 25, 1988 X * X * Author: Reinhold P. Weicker X * Siemens AG, E STE 35 X * Postfach 3240 X * 8520 Erlangen X * Germany (West) X * Phone: [xxx-49]-9131-7-20330 X * (8-17 Central European Time) X * Usenet: ..!mcvax!unido!estevax!weicker X * X * Original Version (in Ada) published in X * "Communications of the ACM" vol. 27., no. 10 (Oct. 1984), X * pp. 1013 - 1030, together with the statistics X * on which the distribution of statements etc. is based. X * X * In this C version, the following C library functions are used: X * - strcpy, strcmp (inside the measurement loop) X * - printf, scanf (outside the measurement loop) X * In addition, Berkeley UNIX system calls "times ()" or "time ()" X * are used for execution time measurement. For measurements X * on other systems, these calls have to be changed. X * X * Collection of Results: X * Reinhold Weicker (address see above) and X * X * Rick Richardson X * PC Research. Inc. X * 94 Apple Orchard Drive X * Tinton Falls, NJ 07724 X * Phone: (201) 389-8963 (9-17 EST) X * Usenet: ...!uunet!pcrat!rick X * X * Please send results to Rick Richardson and/or Reinhold Weicker. X * Complete information should be given on hardware and software used. X * Hardware information includes: Machine type, CPU, type and size X * of caches; for microprocessors: clock frequency, memory speed X * (number of wait states). X * Software information includes: Compiler (and runtime library) X * manufacturer and version, compilation switches, OS version. X * The Operating System version may give an indication about the X * compiler; Dhrystone itself performs no OS calls in the measurement loop. X * X * The complete output generated by the program should be mailed X * such that at least some checks for correctness can be made. X * X *************************************************************************** X * X * History: This version C/2.1 has been made for two reasons: X * X * 1) There is an obvious need for a common C version of X * Dhrystone, since C is at present the most popular system X * programming language for the class of processors X * (microcomputers, minicomputers) where Dhrystone is used most. X * There should be, as far as possible, only one C version of X * Dhrystone such that results can be compared without X * restrictions. In the past, the C versions distributed X * by Rick Richardson (Version 1.1) and by Reinhold Weicker X * had small (though not significant) differences. X * X * 2) As far as it is possible without changes to the Dhrystone X * statistics, optimizing compilers should be prevented from X * removing significant statements. X * X * This C version has been developed in cooperation with X * Rick Richardson (Tinton Falls, NJ), it incorporates many X * ideas from the "Version 1.1" distributed previously by X * him over the UNIX network Usenet. X * I also thank Chaim Benedelac (National Semiconductor), X * David Ditzel (SUN), Earl Killian and John Mashey (MIPS), X * Alan Smith and Rafael Saavedra-Barrera (UC at Berkeley) X * for their help with comments on earlier versions of the X * benchmark. X * X * Changes: In the initialization part, this version follows mostly X * Rick Richardson's version distributed via Usenet, not the X * version distributed earlier via floppy disk by Reinhold Weicker. X * As a concession to older compilers, names have been made X * unique within the first 8 characters. X * Inside the measurement loop, this version follows the X * version previously distributed by Reinhold Weicker. X * X * At several places in the benchmark, code has been added, X * but within the measurement loop only in branches that X * are not executed. The intention is that optimizing compilers X * should be prevented from moving code out of the measurement X * loop, or from removing code altogether. Since the statements X * that are executed within the measurement loop have NOT been X * changed, the numbers defining the "Dhrystone distribution" X * (distribution of statements, operand types and locality) X * still hold. Except for sophisticated optimizing compilers, X * execution times for this version should be the same as X * for previous versions. X * X * Since it has proven difficult to subtract the time for the X * measurement loop overhead in a correct way, the loop check X * has been made a part of the benchmark. This does have X * an impact - though a very minor one - on the distribution X * statistics which have been updated for this version. X * X * All changes within the measurement loop are described X * and discussed in the companion paper "Rationale for X * Dhrystone version 2". X * X * Because of the self-imposed limitation that the order and X * distribution of the executed statements should not be X * changed, there are still cases where optimizing compilers X * may not generate code for some statements. To a certain X * degree, this is unavoidable for small synthetic benchmarks. X * Users of the benchmark are advised to check code listings X * whether code is generated for all statements of Dhrystone. X * X * Version 2.1 is identical to version 2.0 distributed via X * the UNIX network Usenet in March 1988 except that it corrects X * some minor deficiencies that were found by users of version 2.0. X * The only change within the measurement loop is that a X * non-executed "else" part was added to the "if" statement in X * Func_3, and a non-executed "else" part removed from Proc_3. X * X *************************************************************************** X * X * Defines: The following "Defines" are possible: X * -DREG=register (default: Not defined) X * As an approximation to what an average C programmer X * might do, the "register" storage class is applied X * (if enabled by -DREG=register) X * - for local variables, if they are used (dynamically) X * five or more times X * - for parameters if they are used (dynamically) X * six or more times X * Note that an optimal "register" strategy is X * compiler-dependent, and that "register" declarations X * do not necessarily lead to faster execution. X * -DNOSTRUCTASSIGN (default: Not defined) X * Define if the C compiler does not support X * assignment of structures. X * -DNOENUMS (default: Not defined) X * Define if the C compiler does not support X * enumeration types. X * -DTIMES (default) X * -DTIME X * The "times" function of UNIX (returning process times) X * or the "time" function (returning wallclock time) X * is used for measurement. X * For single user machines, "time ()" is adequate. For X * multi-user machines where you cannot get single-user X * access, use the "times ()" function. If you have X * neither, use a stopwatch in the dead of night. X * "printf"s are provided marking the points "Start Timer" X * and "Stop Timer". DO NOT use the UNIX "time(1)" X * command, as this will measure the total time to X * run this program, which will (erroneously) include X * the time to allocate storage (malloc) and to perform X * the initialization. X * -DHZ=nnn X * In Berkeley UNIX, the function "times" returns process X * time in 1/HZ seconds, with HZ = 60 for most systems. X * CHECK YOUR SYSTEM DESCRIPTION BEFORE YOU JUST APPLY X * A VALUE. X * X *************************************************************************** X * X * Compilation model and measurement (IMPORTANT): X * X * This C version of Dhrystone consists of three files: X * - dhry.h (this file, containing global definitions and comments) X * - dhry_1.c (containing the code corresponding to Ada package Pack_1) X * - dhry_2.c (containing the code corresponding to Ada package Pack_2) X * X * The following "ground rules" apply for measurements: X * - Separate compilation X * - No procedure merging X * - Otherwise, compiler optimizations are allowed but should be indicated X * - Default results are those without register declarations X * See the companion paper "Rationale for Dhrystone Version 2" for a more X * detailed discussion of these ground rules. X * X * For 16-Bit processors (e.g. 80186, 80286), times for all compilation X * models ("small", "medium", "large" etc.) should be given if possible, X * together with a definition of these models for the compiler system used. X * X ************************************************************************** X * X * Dhrystone (C version) statistics: X * X * [Comment from the first distribution, updated for version 2. X * Note that because of language differences, the numbers are slightly X * different from the Ada version.] X * X * The following program contains statements of a high level programming X * language (here: C) in a distribution considered representative: X * X * assignments 52 (51.0 %) X * control statements 33 (32.4 %) X * procedure, function calls 17 (16.7 %) X * X * 103 statements are dynamically executed. The program is balanced with X * respect to the three aspects: X * X * - statement type X * - operand type X * - operand locality X * operand global, local, parameter, or constant. X * X * The combination of these three aspects is balanced only approximately. X * X * 1. Statement Type: X * ----------------- number X * X * V1 = V2 9 X * (incl. V1 = F(..) X * V = Constant 12 X * Assignment, 7 X * with array element X * Assignment, 6 X * with record component X * -- X * 34 34 X * X * X = Y +|-|"&&"|"|" Z 5 X * X = Y +|-|"==" Constant 6 X * X = X +|- 1 3 X * X = Y *|/ Z 2 X * X = Expression, 1 X * two operators X * X = Expression, 1 X * three operators X * -- X * 18 18 X * X * if .... 14 X * with "else" 7 X * without "else" 7 X * executed 3 X * not executed 4 X * for ... 7 | counted every time X * while ... 4 | the loop condition X * do ... while 1 | is evaluated X * switch ... 1 X * break 1 X * declaration with 1 X * initialization X * -- X * 34 34 X * X * P (...) procedure call 11 X * user procedure 10 X * library procedure 1 X * X = F (...) X * function call 6 X * user function 5 X * library function 1 X * -- X * 17 17 X * --- X * 103 X * X * The average number of parameters in procedure or function calls X * is 1.82 (not counting the function values as implicit parameters). X * X * X * 2. Operators X * ------------ X * number approximate X * percentage X * X * Arithmetic 32 50.8 X * X * + 21 33.3 X * - 7 11.1 X * * 3 4.8 X * / (int div) 1 1.6 X * X * Comparison 27 42.8 X * X * == 9 14.3 X * /= 4 6.3 X * > 1 1.6 X * < 3 4.8 X * >= 1 1.6 X * <= 9 14.3 X * X * Logic 4 6.3 X * X * && (AND-THEN) 1 1.6 X * | (OR) 1 1.6 X * ! (NOT) 2 3.2 X * X * -- ----- X * 63 100.1 X * X * X * 3. Operand Type (counted once per operand reference): X * --------------- X * number approximate X * percentage X * X * Integer 175 72.3 % X * Character 45 18.6 % X * Pointer 12 5.0 % X * String30 6 2.5 % X * Array 2 0.8 % X * Record 2 0.8 % X * --- ------- X * 242 100.0 % X * X * When there is an access path leading to the final operand (e.g. a record X * component), only the final data type on the access path is counted. X * X * X * 4. Operand Locality: X * ------------------- X * number approximate X * percentage X * X * local variable 114 47.1 % X * global variable 22 9.1 % X * parameter 45 18.6 % X * value 23 9.5 % X * reference 22 9.1 % X * function result 6 2.5 % X * constant 55 22.7 % X * --- ------- X * 242 100.0 % X * X * X * The program does not compute anything meaningful, but it is syntactically X * and semantically correct. All variables have a value assigned to them X * before they are used as a source operand. X * X * There has been no explicit effort to account for the effects of a X * cache, or to balance the use of long or short displacements for code or X * data. X * X *************************************************************************** X */ X X/* Compiler and system dependent definitions: */ X X#ifndef TIME X#undef TIMES X#define TIMES X#endif X /* Use times(2) time function unless */ X /* explicitly defined otherwise */ X X#ifdef MSC_CLOCK X#undef HZ X#undef TIMES X#include <time.h> X#define HZ CLK_TCK X#endif X /* Use Microsoft C hi-res clock */ X X#ifdef TIMES X#include <sys/types.h> X#include <sys/times.h> X /* for "times" */ X#endif X X#define Mic_secs_Per_Second 1000000.0 X /* Berkeley UNIX C returns process times in seconds/HZ */ X X#ifdef NOSTRUCTASSIGN X#define structassign(d, s) memcpy(&(d), &(s), sizeof(d)) X#else X#define structassign(d, s) d = s X#endif X X#ifdef NOENUM X#define Ident_1 0 X#define Ident_2 1 X#define Ident_3 2 X#define Ident_4 3 X#define Ident_5 4 X typedef int Enumeration; X#else X typedef enum {Ident_1, Ident_2, Ident_3, Ident_4, Ident_5} X Enumeration; X#endif X /* for boolean and enumeration types in Ada, Pascal */ X X/* General definitions: */ X X#include <stdio.h> X /* for strcpy, strcmp */ X X#define Null 0 X /* Value of a Null pointer */ X#define true 1 X#define false 0 X Xtypedef int One_Thirty; Xtypedef int One_Fifty; Xtypedef char Capital_Letter; Xtypedef int Boolean; Xtypedef char Str_30 [31]; Xtypedef int Arr_1_Dim [50]; Xtypedef int Arr_2_Dim [50] [50]; X Xtypedef struct record X { X struct record *Ptr_Comp; X Enumeration Discr; X union { X struct { X Enumeration Enum_Comp; X int Int_Comp; X char Str_Comp [31]; X } var_1; X struct { X Enumeration E_Comp_2; X char Str_2_Comp [31]; X } var_2; X struct { X char Ch_1_Comp; X char Ch_2_Comp; X } var_3; X } variant; X } Rec_Type, *Rec_Pointer; X X SHAR_EOF if test 18556 -ne "`wc -c < 'dhry.h'`" then echo shar: "error transmitting 'dhry.h'" '(should have been 18556 characters)' fi fi echo shar: "extracting 'dhry_1.c'" '(11857 characters)' if test -f 'dhry_1.c' then echo shar: "will not over-write existing file 'dhry_1.c'" else sed 's/^X//' << \SHAR_EOF > 'dhry_1.c' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry_1.c (part 2 of 3) X * X * Date: May 25, 1988 X * X * Author: Reinhold P. Weicker X * X **************************************************************************** X */ X X#include "dhry.h" X X/* Global Variables: */ X XRec_Pointer Ptr_Glob, X Next_Ptr_Glob; Xint Int_Glob; XBoolean Bool_Glob; Xchar Ch_1_Glob, X Ch_2_Glob; Xint Arr_1_Glob [50]; Xint Arr_2_Glob [50] [50]; X Xextern char *malloc (); XEnumeration Func_1 (); X /* forward declaration necessary since Enumeration may not simply be int */ X X#ifndef REG X Boolean Reg = false; X#define REG X /* REG becomes defined as empty */ X /* i.e. no register variables */ X#else X Boolean Reg = true; X#endif X X/* variables for time measurement: */ X X#ifdef TIMES Xstruct tms time_info; Xextern int times (); X /* see library function "times" */ X#define Too_Small_Time (2*HZ) X /* Measurements should last at least about 2 seconds */ X#endif X#ifdef TIME Xextern long time(); X /* see library function "time" */ X#define Too_Small_Time 2 X /* Measurements should last at least 2 seconds */ X#endif X#ifdef MSC_CLOCK Xextern clock_t clock(); X#define Too_Small_Time (2*HZ) X#endif X Xlong Begin_Time, X End_Time, X User_Time; Xfloat Microseconds, X Dhrystones_Per_Second; X X/* end of variables for time measurement */ X X Xmain () X/*****/ X X /* main program, corresponds to procedures */ X /* Main and Proc_0 in the Ada version */ X{ X One_Fifty Int_1_Loc; X REG One_Fifty Int_2_Loc; X One_Fifty Int_3_Loc; X REG char Ch_Index; X Enumeration Enum_Loc; X Str_30 Str_1_Loc; X Str_30 Str_2_Loc; X REG int Run_Index; X REG int Number_Of_Runs; X X /* Initializations */ X X Next_Ptr_Glob = (Rec_Pointer) malloc (sizeof (Rec_Type)); X Ptr_Glob = (Rec_Pointer) malloc (sizeof (Rec_Type)); X X Ptr_Glob->Ptr_Comp = Next_Ptr_Glob; X Ptr_Glob->Discr = Ident_1; X Ptr_Glob->variant.var_1.Enum_Comp = Ident_3; X Ptr_Glob->variant.var_1.Int_Comp = 40; X strcpy (Ptr_Glob->variant.var_1.Str_Comp, X "DHRYSTONE PROGRAM, SOME STRING"); X strcpy (Str_1_Loc, "DHRYSTONE PROGRAM, 1'ST STRING"); X X Arr_2_Glob [8][7] = 10; X /* Was missing in published program. Without this statement, */ X /* Arr_2_Glob [8][7] would have an undefined value. */ X /* Warning: With 16-Bit processors and Number_Of_Runs > 32000, */ X /* overflow may occur for this array element. */ X X printf ("\n"); X printf ("Dhrystone Benchmark, Version 2.1 (Language: C)\n"); X printf ("\n"); X if (Reg) X { X printf ("Program compiled with 'register' attribute\n"); X printf ("\n"); X } X else X { X printf ("Program compiled without 'register' attribute\n"); X printf ("\n"); X } X printf ("Please give the number of runs through the benchmark: "); X { X int n; X scanf ("%d", &n); X Number_Of_Runs = n; X } X printf ("\n"); X X printf ("Execution starts, %d runs through Dhrystone\n", Number_Of_Runs); X X /***************/ X /* Start timer */ X /***************/ X X#ifdef TIMES X times (&time_info); X Begin_Time = (long) time_info.tms_utime; X#endif X#ifdef TIME X Begin_Time = time ( (long *) 0); X#endif X#ifdef MSC_CLOCK X Begin_Time = clock(); X#endif X X for (Run_Index = 1; Run_Index <= Number_Of_Runs; ++Run_Index) X { X X Proc_5(); X Proc_4(); X /* Ch_1_Glob == 'A', Ch_2_Glob == 'B', Bool_Glob == true */ X Int_1_Loc = 2; X Int_2_Loc = 3; X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 2'ND STRING"); X Enum_Loc = Ident_2; X Bool_Glob = ! Func_2 (Str_1_Loc, Str_2_Loc); X /* Bool_Glob == 1 */ X while (Int_1_Loc < Int_2_Loc) /* loop body executed once */ X { X Int_3_Loc = 5 * Int_1_Loc - Int_2_Loc; X /* Int_3_Loc == 7 */ X Proc_7 (Int_1_Loc, Int_2_Loc, &Int_3_Loc); X /* Int_3_Loc == 7 */ X Int_1_Loc += 1; X } /* while */ X /* Int_1_Loc == 3, Int_2_Loc == 3, Int_3_Loc == 7 */ X Proc_8 (Arr_1_Glob, Arr_2_Glob, Int_1_Loc, Int_3_Loc); X /* Int_Glob == 5 */ X Proc_1 (Ptr_Glob); X for (Ch_Index = 'A'; Ch_Index <= Ch_2_Glob; ++Ch_Index) X /* loop body executed twice */ X { X if (Enum_Loc == Func_1 (Ch_Index, 'C')) X /* then, not executed */ X { X Proc_6 (Ident_1, &Enum_Loc); X strcpy (Str_2_Loc, "DHRYSTONE PROGRAM, 3'RD STRING"); X Int_2_Loc = Run_Index; X Int_Glob = Run_Index; X } X } X /* Int_1_Loc == 3, Int_2_Loc == 3, Int_3_Loc == 7 */ X Int_2_Loc = Int_2_Loc * Int_1_Loc; X Int_1_Loc = Int_2_Loc / Int_3_Loc; X Int_2_Loc = 7 * (Int_2_Loc - Int_3_Loc) - Int_1_Loc; X /* Int_1_Loc == 1, Int_2_Loc == 13, Int_3_Loc == 7 */ X Proc_2 (&Int_1_Loc); X /* Int_1_Loc == 5 */ X X } /* loop "for Run_Index" */ X X /**************/ X /* Stop timer */ X /**************/ X X#ifdef TIMES X times (&time_info); X End_Time = (long) time_info.tms_utime; X#endif X#ifdef TIME X End_Time = time ( (long *) 0); X#endif X#ifdef MSC_CLOCK X End_Time = clock(); X#endif X X printf ("Execution ends\n"); X printf ("\n"); X printf ("Final values of the variables used in the benchmark:\n"); X printf ("\n"); X printf ("Int_Glob: %d\n", Int_Glob); X printf (" should be: %d\n", 5); X printf ("Bool_Glob: %d\n", Bool_Glob); X printf (" should be: %d\n", 1); X printf ("Ch_1_Glob: %c\n", Ch_1_Glob); X printf (" should be: %c\n", 'A'); X printf ("Ch_2_Glob: %c\n", Ch_2_Glob); X printf (" should be: %c\n", 'B'); X printf ("Arr_1_Glob[8]: %d\n", Arr_1_Glob[8]); X printf (" should be: %d\n", 7); X printf ("Arr_2_Glob[8][7]: %d\n", Arr_2_Glob[8][7]); X printf (" should be: Number_Of_Runs + 10\n"); X printf ("Ptr_Glob->\n"); X printf (" Ptr_Comp: %d\n", (int) Ptr_Glob->Ptr_Comp); X printf (" should be: (implementation-dependent)\n"); X printf (" Discr: %d\n", Ptr_Glob->Discr); X printf (" should be: %d\n", 0); X printf (" Enum_Comp: %d\n", Ptr_Glob->variant.var_1.Enum_Comp); X printf (" should be: %d\n", 2); X printf (" Int_Comp: %d\n", Ptr_Glob->variant.var_1.Int_Comp); X printf (" should be: %d\n", 17); X printf (" Str_Comp: %s\n", Ptr_Glob->variant.var_1.Str_Comp); X printf (" should be: DHRYSTONE PROGRAM, SOME STRING\n"); X printf ("Next_Ptr_Glob->\n"); X printf (" Ptr_Comp: %d\n", (int) Next_Ptr_Glob->Ptr_Comp); X printf (" should be: (implementation-dependent), same as above\n"); X printf (" Discr: %d\n", Next_Ptr_Glob->Discr); X printf (" should be: %d\n", 0); X printf (" Enum_Comp: %d\n", Next_Ptr_Glob->variant.var_1.Enum_Comp); X printf (" should be: %d\n", 1); X printf (" Int_Comp: %d\n", Next_Ptr_Glob->variant.var_1.Int_Comp); X printf (" should be: %d\n", 18); X printf (" Str_Comp: %s\n", X Next_Ptr_Glob->variant.var_1.Str_Comp); X printf (" should be: DHRYSTONE PROGRAM, SOME STRING\n"); X printf ("Int_1_Loc: %d\n", Int_1_Loc); X printf (" should be: %d\n", 5); X printf ("Int_2_Loc: %d\n", Int_2_Loc); X printf (" should be: %d\n", 13); X printf ("Int_3_Loc: %d\n", Int_3_Loc); X printf (" should be: %d\n", 7); X printf ("Enum_Loc: %d\n", Enum_Loc); X printf (" should be: %d\n", 1); X printf ("Str_1_Loc: %s\n", Str_1_Loc); X printf (" should be: DHRYSTONE PROGRAM, 1'ST STRING\n"); X printf ("Str_2_Loc: %s\n", Str_2_Loc); X printf (" should be: DHRYSTONE PROGRAM, 2'ND STRING\n"); X printf ("\n"); X X User_Time = End_Time - Begin_Time; X X if (User_Time < Too_Small_Time) X { X printf ("Measured time too small to obtain meaningful results\n"); X printf ("Please increase number of runs\n"); X printf ("\n"); X } X else X { X#ifdef TIME X Microseconds = (float) User_Time * Mic_secs_Per_Second X / (float) Number_Of_Runs; X Dhrystones_Per_Second = (float) Number_Of_Runs / (float) User_Time; X#else X Microseconds = (float) User_Time * Mic_secs_Per_Second X / ((float) HZ * ((float) Number_Of_Runs)); X Dhrystones_Per_Second = ((float) HZ * (float) Number_Of_Runs) X / (float) User_Time; X#endif X printf ("Microseconds for one run through Dhrystone: "); X printf ("%6.1f \n", Microseconds); X printf ("Dhrystones per Second: "); X printf ("%6.1f \n", Dhrystones_Per_Second); X printf ("\n"); X } X X} X X XProc_1 (Ptr_Val_Par) X/******************/ X XREG Rec_Pointer Ptr_Val_Par; X /* executed once */ X{ X REG Rec_Pointer Next_Record = Ptr_Val_Par->Ptr_Comp; X /* == Ptr_Glob_Next */ X /* Local variable, initialized with Ptr_Val_Par->Ptr_Comp, */ X /* corresponds to "rename" in Ada, "with" in Pascal */ X X structassign (*Ptr_Val_Par->Ptr_Comp, *Ptr_Glob); X Ptr_Val_Par->variant.var_1.Int_Comp = 5; X Next_Record->variant.var_1.Int_Comp X = Ptr_Val_Par->variant.var_1.Int_Comp; X Next_Record->Ptr_Comp = Ptr_Val_Par->Ptr_Comp; X Proc_3 (&Next_Record->Ptr_Comp); X /* Ptr_Val_Par->Ptr_Comp->Ptr_Comp X == Ptr_Glob->Ptr_Comp */ X if (Next_Record->Discr == Ident_1) X /* then, executed */ X { X Next_Record->variant.var_1.Int_Comp = 6; X Proc_6 (Ptr_Val_Par->variant.var_1.Enum_Comp, X &Next_Record->variant.var_1.Enum_Comp); X Next_Record->Ptr_Comp = Ptr_Glob->Ptr_Comp; X Proc_7 (Next_Record->variant.var_1.Int_Comp, 10, X &Next_Record->variant.var_1.Int_Comp); X } X else /* not executed */ X structassign (*Ptr_Val_Par, *Ptr_Val_Par->Ptr_Comp); X} /* Proc_1 */ X X XProc_2 (Int_Par_Ref) X/******************/ X /* executed once */ X /* *Int_Par_Ref == 1, becomes 4 */ X XOne_Fifty *Int_Par_Ref; X{ X One_Fifty Int_Loc; X Enumeration Enum_Loc; X X Int_Loc = *Int_Par_Ref + 10; X do /* executed once */ X if (Ch_1_Glob == 'A') X /* then, executed */ X { X Int_Loc -= 1; X *Int_Par_Ref = Int_Loc - Int_Glob; X Enum_Loc = Ident_1; X } /* if */ X while (Enum_Loc != Ident_1); /* true */ X} /* Proc_2 */ X X XProc_3 (Ptr_Ref_Par) X/******************/ X /* executed once */ X /* Ptr_Ref_Par becomes Ptr_Glob */ X XRec_Pointer *Ptr_Ref_Par; X X{ X if (Ptr_Glob != Null) X /* then, executed */ X *Ptr_Ref_Par = Ptr_Glob->Ptr_Comp; X Proc_7 (10, Int_Glob, &Ptr_Glob->variant.var_1.Int_Comp); X} /* Proc_3 */ X X XProc_4 () /* without parameters */ X/*******/ X /* executed once */ X{ X Boolean Bool_Loc; X X Bool_Loc = Ch_1_Glob == 'A'; X Bool_Glob = Bool_Loc | Bool_Glob; X Ch_2_Glob = 'B'; X} /* Proc_4 */ X X XProc_5 () /* without parameters */ X/*******/ X /* executed once */ X{ X Ch_1_Glob = 'A'; X Bool_Glob = false; X} /* Proc_5 */ X X X /* Procedure for the assignment of structures, */ X /* if the C compiler doesn't support this feature */ X#ifdef NOSTRUCTASSIGN Xmemcpy (d, s, l) Xregister char *d; Xregister char *s; Xregister int l; X{ X while (l--) *d++ = *s++; X} X#endif X X SHAR_EOF if test 11857 -ne "`wc -c < 'dhry_1.c'`" then echo shar: "error transmitting 'dhry_1.c'" '(should have been 11857 characters)' fi fi echo shar: "extracting 'dhry_2.c'" '(5273 characters)' if test -f 'dhry_2.c' then echo shar: "will not over-write existing file 'dhry_2.c'" else sed 's/^X//' << \SHAR_EOF > 'dhry_2.c' X/* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: C, Version 2.1 X * X * File: dhry_2.c (part 3 of 3) X * X * Date: May 25, 1988 X * X * Author: Reinhold P. Weicker X * X **************************************************************************** X */ X X#include "dhry.h" X X#ifndef REG X#define REG X /* REG becomes defined as empty */ X /* i.e. no register variables */ X#endif X Xextern int Int_Glob; Xextern char Ch_1_Glob; X X XProc_6 (Enum_Val_Par, Enum_Ref_Par) X/*********************************/ X /* executed once */ X /* Enum_Val_Par == Ident_3, Enum_Ref_Par becomes Ident_2 */ X XEnumeration Enum_Val_Par; XEnumeration *Enum_Ref_Par; X{ X *Enum_Ref_Par = Enum_Val_Par; X if (! Func_3 (Enum_Val_Par)) X /* then, not executed */ X *Enum_Ref_Par = Ident_4; X switch (Enum_Val_Par) X { X case Ident_1: X *Enum_Ref_Par = Ident_1; X break; X case Ident_2: X if (Int_Glob > 100) X /* then */ X *Enum_Ref_Par = Ident_1; X else *Enum_Ref_Par = Ident_4; X break; X case Ident_3: /* executed */ X *Enum_Ref_Par = Ident_2; X break; X case Ident_4: break; X case Ident_5: X *Enum_Ref_Par = Ident_3; X break; X } /* switch */ X} /* Proc_6 */ X X XProc_7 (Int_1_Par_Val, Int_2_Par_Val, Int_Par_Ref) X/**********************************************/ X /* executed three times */ X /* first call: Int_1_Par_Val == 2, Int_2_Par_Val == 3, */ X /* Int_Par_Ref becomes 7 */ X /* second call: Int_1_Par_Val == 10, Int_2_Par_Val == 5, */ X /* Int_Par_Ref becomes 17 */ X /* third call: Int_1_Par_Val == 6, Int_2_Par_Val == 10, */ X /* Int_Par_Ref becomes 18 */ XOne_Fifty Int_1_Par_Val; XOne_Fifty Int_2_Par_Val; XOne_Fifty *Int_Par_Ref; X{ X One_Fifty Int_Loc; X X Int_Loc = Int_1_Par_Val + 2; X *Int_Par_Ref = Int_2_Par_Val + Int_Loc; X} /* Proc_7 */ X X XProc_8 (Arr_1_Par_Ref, Arr_2_Par_Ref, Int_1_Par_Val, Int_2_Par_Val) X/*********************************************************************/ X /* executed once */ X /* Int_Par_Val_1 == 3 */ X /* Int_Par_Val_2 == 7 */ XArr_1_Dim Arr_1_Par_Ref; XArr_2_Dim Arr_2_Par_Ref; Xint Int_1_Par_Val; Xint Int_2_Par_Val; X{ X REG One_Fifty Int_Index; X REG One_Fifty Int_Loc; X X Int_Loc = Int_1_Par_Val + 5; X Arr_1_Par_Ref [Int_Loc] = Int_2_Par_Val; X Arr_1_Par_Ref [Int_Loc+1] = Arr_1_Par_Ref [Int_Loc]; X Arr_1_Par_Ref [Int_Loc+30] = Int_Loc; X for (Int_Index = Int_Loc; Int_Index <= Int_Loc+1; ++Int_Index) X Arr_2_Par_Ref [Int_Loc] [Int_Index] = Int_Loc; X Arr_2_Par_Ref [Int_Loc] [Int_Loc-1] += 1; X Arr_2_Par_Ref [Int_Loc+20] [Int_Loc] = Arr_1_Par_Ref [Int_Loc]; X Int_Glob = 5; X} /* Proc_8 */ X X XEnumeration Func_1 (Ch_1_Par_Val, Ch_2_Par_Val) X/*************************************************/ X /* executed three times */ X /* first call: Ch_1_Par_Val == 'H', Ch_2_Par_Val == 'R' */ X /* second call: Ch_1_Par_Val == 'A', Ch_2_Par_Val == 'C' */ X /* third call: Ch_1_Par_Val == 'B', Ch_2_Par_Val == 'C' */ X XCapital_Letter Ch_1_Par_Val; XCapital_Letter Ch_2_Par_Val; X{ X Capital_Letter Ch_1_Loc; X Capital_Letter Ch_2_Loc; X X Ch_1_Loc = Ch_1_Par_Val; X Ch_2_Loc = Ch_1_Loc; X if (Ch_2_Loc != Ch_2_Par_Val) X /* then, executed */ X return (Ident_1); X else /* not executed */ X { X Ch_1_Glob = Ch_1_Loc; X return (Ident_2); X } X} /* Func_1 */ X X XBoolean Func_2 (Str_1_Par_Ref, Str_2_Par_Ref) X/*************************************************/ X /* executed once */ X /* Str_1_Par_Ref == "DHRYSTONE PROGRAM, 1'ST STRING" */ X /* Str_2_Par_Ref == "DHRYSTONE PROGRAM, 2'ND STRING" */ X XStr_30 Str_1_Par_Ref; XStr_30 Str_2_Par_Ref; X{ X REG One_Thirty Int_Loc; X Capital_Letter Ch_Loc; X X Int_Loc = 2; X while (Int_Loc <= 2) /* loop body executed once */ X if (Func_1 (Str_1_Par_Ref[Int_Loc], X Str_2_Par_Ref[Int_Loc+1]) == Ident_1) X /* then, executed */ X { X Ch_Loc = 'A'; X Int_Loc += 1; X } /* if, while */ X if (Ch_Loc >= 'W' && Ch_Loc < 'Z') X /* then, not executed */ X Int_Loc = 7; X if (Ch_Loc == 'R') X /* then, not executed */ X return (true); X else /* executed */ X { X if (strcmp (Str_1_Par_Ref, Str_2_Par_Ref) > 0) X /* then, not executed */ X { X Int_Loc += 7; X Int_Glob = Int_Loc; X return (true); X } X else /* executed */ X return (false); X } /* if Ch_Loc */ X} /* Func_2 */ X X XBoolean Func_3 (Enum_Par_Val) X/***************************/ X /* executed once */ X /* Enum_Par_Val == Ident_3 */ XEnumeration Enum_Par_Val; X{ X Enumeration Enum_Loc; X X Enum_Loc = Enum_Par_Val; X if (Enum_Loc == Ident_3) X /* then, executed */ X return (true); X else /* not executed */ X return (false); X} /* Func_3 */ X SHAR_EOF if test 5273 -ne "`wc -c < 'dhry_2.c'`" then echo shar: "error transmitting 'dhry_2.c'" '(should have been 5273 characters)' fi fi exit 0 # End of shell archive
dry2@pcrat.UUCP (Dhrystone) (12/04/88)
#! /bin/sh # This is a shell archive, meaning: # 1. Remove everything above the #! /bin/sh line. # 2. Save the resulting text in a file. # 3. Execute the file with /bin/sh (not csh) to create: # dhry.p # This archive created: Sun Dec 4 00:42:38 1988 export PATH; PATH=/bin:/usr/bin:$PATH echo shar: "extracting 'dhry.p'" '(37108 characters)' if test -f 'dhry.p' then echo shar: "will not over-write existing file 'dhry.p'" else sed 's/^X//' << \SHAR_EOF > 'dhry.p' X(* X **************************************************************************** X * X * "DHRYSTONE" Benchmark Program X * ----------------------------- X * X * Version: Pascal, Version 2.1 X * X * File: dhry.p X * X * Date: May 25, 1988 X * X * Author: Reinhold P. Weicker X * Siemens AG, E STE 35 X * Postfach 3240 X * 8520 Erlangen X * Germany (West) X * Phone: [xxx-49]-9131-7-20330 X * (8-17 Central European Time) X * Usenet: ..!mcvax!unido!estevax!weicker X * X * Original Version (in Ada) published in X * "Communications of the ACM" vol. 27., no. 10 (Oct. 1984), X * pp. 1013 - 1030, together with the statistics X * on which the distribution of statements etc. is based, X * X * This version uses calls to the Pascal runtime library of the X * Berkeley UNIX system (4.3 bsd) for time measurement. X * For measurements on other systems, these calls need to be X * modified. X * X * Collection of Results: X * Reinhold Weicker (address see above) and X * X * Rick Richardson X * PC Research. Inc. X * 94 Apple Orchard Drive X * Tinton Falls, NJ 07724 X * Phone: (201) 834-1378 (9-17 EST) X * Usenet: ...!seismo!uunet!pcrat!rick X * X * Please send results to Rick Richardson and/or Reinhold Weicker. X * Complete information should be given on hardware and software used. X * Hardware information includes: Machine type, CPU, type and size X * of caches; for microprocessors: clock frequency, memory speed X * (number of wait states). X * Software information includes: Compiler (and runtime library) X * manufacturer and version, compilation switches, OS version. X * The Operating System version may give an indication about the X * compiler; Dhrystone itself performs no OS calls in the measurement loop. X * X * The complete output generated by the program should be mailed X * such that at least some checks for correctness can be made. X * X **************************************************************************** X * X * History: This version Pascal/2.1 has been made for two reasons: X * X * 1) There is a need for a common Pascal version of X * Dhrystone. Although translation from the published (Ada) X * version to Pascal is straightforward in most aspects, X * there are cases where it may not be obvious to everyone. X * There should be, as far as possible, only one Pascal version X * of Dhrystone such that results can be compared without X * restrictions. Also, a Pascal version of Dhrystone has not yet X * found a network distribution comparable to the C version X * (version 1.1) distributed by Rick Richardson. X * X * 2) As far as it is possible without changes to the Dhrystone X * statistics, optimizing compilers should be prevented from X * removing significant statements. X * X * This Pascal version 2.1 has been made consistent with the X * C version 2.1; therefore the acknowledgments for the C version X * are due for the Pascal version as well: I thank X * Rick Richardson (Tinton Falls, NJ), Chaim Benedelac (Nat. X * Semi.), David Ditzel (SUN), Earl Killian and John Mashey (MIPS), X * Alan Smith and Rafael Saavedra-Barrera (UC at Berkeley) X * for their help with comments on earlier versions of the X * benchmark. X * X * Changes: In the initialization part, this version differs X * from the Pascal version previously distributed by Reinhold X * Weicker, the number of runs through Dhrystone is obtained X * interactively from the terminal. Output of the result X * has been changed to conform to the C version (2.1). X * The changes in the initialization part and in the printing X * of the result have no impact on performance measurement X * since they are outside the measaurement loop. X * X * Inside the measurement loop, this version follows the X * version previously distributed by Reinhold Weicker. X * As a correction to the published version, a statement X * initializing Array2Glob [8,7] (outside the measurement X * loop) has been added. Otherwise, this array element would X * have an undefined value. X * X * At several places in the benchmark, code has been added, X * but within the measurement loop only in branches that X * are not executed. The intention is that optimizing compilers X * should be prevented from moving code out of the measurement X * loop, or from removing code altogether. Since the statements X * that are executed within the measurement loop have NOT been X * changed, all numbers defining the "Dhrystone distribution" X * (distribution of statements, operand types and locality) X * still hold. Except for sophisticated optimizing compilers, X * execution times for this version should be the same as X * for previous versions. X * X * Since it has proven difficult to subtract the time for the X * measurement loop overhead in a correct way, the loop check X * has been made a part of the benchmark. This does have X * an impact - though a very minor one - on the distribution X * statistics which have been updated for this version. X * X * All changes within the measurement loop are described X * and discussed in the companion paper "Rationale for X * Dhrystone version 2". X * X * Because of the self-imposed limitation that the order and X * distribution of the executed statements should not be X * changed, there are still cases where optimizing compilers X * may not generate code for some statements. To a certain X * degree, this is unavoidable for small synthetic benchmarks. X * Users of the benchmark are advised to check code listings X * whether code is generated for all statements of Dhrystone. X * X * Version 2.1 is identical to version 2.0 distributed via X * the UNIX network Usenet in March 1988 except that it corrects X * some minor deficiencies that were found by users of version 2.0. X * The only change within the measurement loop is that a X * non-executed "else" part was added to the "if" statement in X * Func3, and a non-executed "else" part removed from Proc3. X * X *************************************************************************** X * X * Compilation model and measurement (IMPORTANT): X * X * This program contains the Dhrystone program, including measurement setup, X * in one file. The original (Ada) program contained three packages, X * - a package with global type definitions, X * - Pack_1, containing the main program (Proc_0 in Ada) and procedures X * Proc_1, ... , Proc_5, X * - Pack_2, containing Proc_6, ... , Proc_8, Func_1, ..., Func_3. X * Since ISO/ANSI Standard Pascal provides no means to express separate X * compilation (although many Pascal implementations provide such a feature), X * it is not possible to formulate a portable Pascal version with the program X * in several modules, corresponding more closely to the Ada and C versions. X * Therefore, no attempt has been made to construct a Pascal version with X * the program consisting of several modules. X * X * This difference may impact execution time because the compiler can X * perform more data flow analysis for a single-module program; X * sophisticated compilers may be able to suppress code generation for X * some parts of the program. X * Users should check machine code listings generated by the compiler X * to ensure that code is generated for all parts of the program. X * X * The following "ground rules" apply for measurements: X * - No procedure merging X * - Otherwise, compiler optimizations are allowed but should be indicated X * See the companion paper "Rationale for Dhrystone Version 2" for a more X * detailed discussion of these ground rules. X * X * For 16-Bit processors (e.g. 80x86), times for all compilation models X * ("small", "medium", "large") should be given if possible, together X * with a definition of these models for the compiler system used. X * X ************************************************************************** X * X * Dhrystone (Pascal version) statistics: X * X * [Comment from the first distribution by Reinhold Weicker, X * the distribution statistics have been updated for Version 2.1. X * Note that because of language differences, the numbers are different X * from the Ada version. The main difference is that the variables that X * are local variables of "Proc_0" (Ada) or "main" (C) are global X * variables in the Pascal version.] X * X * The following program contains statements of a high level programming X * language (here: Pascal) in a distribution considered representative: X * X * assignments 58 X * control statements 28 X * procedure, function calls 15 X * X * 100 statements are dynamically executed. The program is balanced with X * respect to the three aspects: X * X * - statement type X * - operand type (for simple data types) X * - operand access X * operand global, local, parameter, or constant. X * There is no static nesting of blocks or procedures, X * therefore all variables are either global or local. X * X * The combination of these three aspects is balanced only approximately. X * X * 1. Statement Type: X * ----------------- number X * X * V1 := V2 15 X * V := Constant 12 X * (incl. V1 := F(..) X * Assignment, 7 X * with array element X * Assignment, 6 X * with record component X * -- X * 40 40 X * X * X := Y +|-|and|or Z 5 X * X := Y +|-|"=" Constant 6 X * X := X +|- 1 3 X * X := Y *|/ Z 2 X * X := Expression, 1 X * two operators X * X := Expression, 1 X * three operators X * -- X * 18 18 X * X * if .... then .... 14 X * with "else" 7 X * without "else" 7 X * executed 3 X * not executed 4 X * for I in 1..N do ... 7 | counted every time X * while ... do ... 4 | the loop condition X * repeat ... until 1 | is evaluated X * case ... end 1 X * with 1 X * -- X * 28 28 X * X * P (...) procedure call 10 X * X := F (...) X * function call 5 X * -- X * 15 15 X * --- X * 101 X * X * 22 of the 58 assignments have a variable of a constrained X * (sub-)type as their destination. In general, discriminant checks X * will be necessary in these cases; however, the compiler may X * optimize out some of these checks. X * X * The average number of parameters in procedure or function calls X * is 1.80 (not counting the function values as implicit parameters). X * X * X * 2. Operators X * ------------ X * number approximate X * percentage X * X * Arithmetic 27 52.9 X * X * + 16 31.4 X * - 7 13.7 X * * 3 5.9 X * div 1 2.0 X * X * Comparison 20 39.2 X * X * = 9 17.6 X * <> 4 7.8 X * > 1 2.0 X * < 3 5.9 X * >= 1 2.0 X * <= 2 3.9 X * X * Logic 4 7.8 X * X * AND 1 2.0 X * OR 1 2.0 X * NOT 2 3.9 X * X * -- ----- X * 51 99.9 X * X * X * 3. Operand Type (counted once per operand reference): X * --------------- X * number approximate X * percentage X * X * Integer 135 54.7 % X * Character 47 19.0 % X * Enumeration 31 12.6 % X * Boolean 13 5.3 % X * Pointer 11 4.5 % X * String30 6 2.4 % X * Array 2 0.8 % X * Record 2 0.8 % X * --- ------- X * 247 100.1 % X * X * When there is an access path leading to the final operand (e.g. a record X * component), only the final data type on the access path is counted. X * X * There are 16 accesses to components of a record, 9 of them go to X * a component in a variant part. For some of these accesses, the X * compiler may suppress generation of code checking the tag field X * during optimization. X * X * X * 3. Operand Locality: X * ------------------- X * X * local variable 84 34.0 % X * global variable 58 23.5 % X * parameter 45 18.2 % X * value 23 9.3 % X * reference 22 8.9 % X * function result 5 2.0 % X * constant 55 22.3 % X * --- ------- X * 247 100.0 % X * X * X * The program does not compute anything meaningful, but it is syntactically X * and semantically correct. All variables have a value assigned to them X * before they are used as a source operand. X * X * There may be cases where a highly optimizing compiler may recognize X * unnecessary statements and may not generate code for them. X * X * There has been no explicit effort to account for the effects of a X * cache, or to balance the use of long or short displacements for code or X * data. X * X **************************************************************************** X *) X Xprogram Dhrystone (input, output); X(***************) X Xconst (* for measurement *) X X MicrosecondsPerClock = 1000; X ClocksPerSecond = 1000; X (* In Berkeley UNIX Pascal, the function "clock" *) X (* returns milliseconds *) X TooSmallTime = 2000; X (* Measurements should last at least 2 seconds *) X Xtype X X (* Global type definitions *) X X Enumeration = (Ident1, Ident2, Ident3, Ident4, Ident5); X X OneToThirty = 1..30; X OneToFifty = 1..50; X CapitalLetter = 'A'..'Z'; X X String30 = packed array [OneToThirty] of char; X X Array1DimInteger = array [OneToFifty] of integer; X Array2DimInteger = array [OneToFifty, OneToFifty] of integer; X X RecordPointer = ^RecordType; X X RecordType = X record X PointerComp: RecordPointer; X case Discr: Enumeration of X Ident1: (* only this variant is used, *) X (* but in some cases discriminant *) X (* checks are necessary *) X (EnumComp: Enumeration; X IntComp: OneToFifty; X StringComp: String30); X Ident2: X (Enum2Comp: Enumeration; X String2Comp: String30); X Ident3, Ident4, Ident5: X (Char1Comp, X Char2Comp: char); X end; (* record *) X Xvar X X (* Ada version: Variables local in Proc_0 *) X X Int1Glob, X Int2Glob, X Int3Glob: OneToFifty; X CharIndex: char; X EnumGlob: Enumeration; X String1Glob, X String2Glob: String30; X X (* Ada version: Variables global in Pack_1 *) X X PointerGlob, X NextPointerGlob: RecordPointer; X IntGlob: integer; X X BoolGlob: boolean; X Char1Glob, X Char2Glob: char; X Array1Glob: Array1DimInteger; X Array2Glob: Array2DimInteger; X X (* Variables for measurement *) X X RunIndex, X NumberOfRuns, X BeginClock, X EndClock, X SumClocks: integer; X Microseconds, X DhrystonesPerSecond: real; X I: integer; X X (* end of variables for measurement *) X Xprocedure Proc1 ( PointerParVal: RecordPointer); forward; X Xprocedure Proc2 (var IntParRef: OneToFifty); forward; X Xprocedure Proc3 (var PointerParRef: RecordPointer); forward; X Xprocedure Proc4; forward; X (* without parameters *) X Xprocedure Proc5; forward; X (* without parameters *) X Xprocedure Proc6 ( EnumParVal: Enumeration; X var EnumParRef: Enumeration); forward; X Xprocedure Proc7 ( Int1ParVal, X Int2ParVal: OneToFifty; X var IntParRef: OneToFifty); forward; X Xprocedure Proc8 (var Array1ParRef: Array1DimInteger; X var Array2ParRef: Array2DimInteger; X Int1ParVal, X Int2ParVal: integer); forward; X Xfunction Func1 ( Char1ParVal, X Char2ParVal: CapitalLetter): X Enumeration; forward; X Xfunction Func2 (var String1ParRef, X String2ParRef: String30): X boolean; forward; X Xfunction Func3 ( EnumParVal: Enumeration): X boolean; forward; X X Xprocedure Proc1; (* (PointerParVal: RecordPointer) *) X (* executed once *) Xbegin X with PointerParVal^.PointerComp^ (* = PointerGlobNext *) do X begin X PointerParVal^.PointerComp^ := PointerGlob^; X PointerParVal^.IntComp := 5; X IntComp := PointerParVal^.IntComp; X PointerComp := PointerParVal^.PointerComp; X Proc3 (PointerComp); X (* PointerParVal^.PointerComp^.PointerComp = PointerGlob^.PointerComp *) X if Discr = Ident1 X then (* executed *) X begin X IntComp := 6; X Proc6 (PointerParVal^.EnumComp, EnumComp); X PointerComp := PointerGlob^.PointerComp; X Proc7 (IntComp, 10, IntComp); X end (* then *) X else (* not executed *) X PointerParVal^ := PointerParVal^.PointerComp^; X end; (* with *) Xend; (* Proc1 *) X X Xprocedure Proc2; (* (var IntParRef: OneToFifty) *) X (* executed once *) X (* InParRef = 3, becomes 7 *) Xvar X IntLoc: OneToFifty; X EnumLoc: Enumeration; Xbegin X IntLoc := IntParRef + 10; X repeat (* executed once *) X if Char1Glob = 'A' X then (* executed *) X begin X IntLoc := IntLoc - 1; X IntParRef := IntLoc - IntGlob; X EnumLoc := Ident1; X end (* if *) X until EnumLoc = Ident1; (* true *) Xend; (* Proc2 *) X X Xprocedure Proc3; (* (var PointerParRef: RecordPointer) *) X (* executed once *) X (* PointerParRef becomes PointerGlob *) Xbegin X if PointerGlob <> nil X then (* executed *) X PointerParRef := PointerGlob^.PointerComp; X Proc7 (10, IntGlob, PointerGlob^.IntComp); Xend; (* Proc3 *) X X Xprocedure Proc4; (* without parameters *) X (* executed once *) Xvar X BoolLoc: boolean; Xbegin X BoolLoc := Char1Glob = 'A'; X BoolGlob := BoolLoc or BoolGlob; X Char2Glob := 'B'; Xend; (* Proc4 *) X X Xprocedure Proc5; (* without parameters *) X (* executed once *) Xbegin X Char1Glob := 'A'; X BoolGlob := false; Xend; (* Proc5 *) X X Xprocedure Proc6; (* ( EnumParVal: Enumeration; X var EnumParRef: Enumeration) *) X (* executed once *) X (* EnumParVal = Ident3, EnumParRef becomes Ident2 *) Xbegin X EnumParRef := EnumParVal; X if not Func3 (EnumParVal) X then (* not executed *) X EnumParRef := Ident4; X case EnumParVal of X Ident1: EnumParRef := Ident1; X Ident2: if IntGlob > 100 X then EnumParRef := Ident1 X else EnumParRef := Ident4; X Ident3: EnumParRef := Ident2; (* executed *) X Ident4: ; X Ident5: EnumParRef := Ident3; X end; (* case *) Xend; (* Proc6 *) X X Xprocedure Proc7; (* ( Int1ParVal, X Int2ParVal: OneToFifty; X var IntParRef: OneToFifty) *) X (* executed three times *) X (* first call: Int1ParVal = 2, Int2ParVal = 3, *) X (* IntParRef becomes 7 *) X (* second call: Int1ParVal = 10, Int2ParVal = 5, *) X (* IntParRef becomes 17 *) X (* third call: Int1ParVal = 6, Int2ParVal = 10, *) X (* IntParRef becomes 18 *) Xvar X IntLoc: OneToFifty; Xbegin X IntLoc := Int1ParVal + 2; X IntParRef := Int2ParVal + IntLoc; Xend; (* Proc7 *) X X Xprocedure Proc8; (* (var Array1ParRef: Array1DimInteger; X var Array2ParRef: Array2DimInteger; X Int1ParVal, X Int2ParVal: integer) *) X (* executed once *) X (* Int1ParVal = 3 *) X (* Int2ParVal = 7 *) Xvar X IntIndex, X IntLoc: OneToFifty; Xbegin X IntLoc := Int1ParVal + 5; X Array1ParRef [IntLoc] := Int2ParVal; X Array1ParRef [IntLoc+1] := Array1ParRef [IntLoc]; X Array1ParRef [IntLoc+30] := IntLoc; X for IntIndex := IntLoc to IntLoc+1 do X Array2ParRef [IntLoc, IntIndex] := IntLoc; X Array2ParRef [IntLoc, IntLoc-1] := Array2ParRef [IntLoc, IntLoc-1] + 1; X Array2ParRef [IntLoc+20, IntLoc] := Array1ParRef [IntLoc]; X IntGlob := 5; Xend; (* Proc8 *) X X Xfunction Func1; (* (Char1ParVal, X Char2ParVal: CapitalLetter): Enumeration *) X (* executed three times, returns always Ident1 *) X (* first call: Char1ParVal = 'H', Char2ParVal = 'R' *) X (* second call: Char1ParVal = 'A', Char2ParVal = 'C' *) X (* third call: Char1ParVal = 'B', Char2ParVal = 'C' *) Xvar X Char1Loc, Char2Loc: CapitalLetter; Xbegin X Char1Loc := Char1ParVal; X Char2Loc := Char1Loc; X if Char2Loc <> Char2ParVal X then (* executed *) X Func1 := Ident1 X else (* not executed *) X begin X Char1Glob := Char1Loc; X Func1 := Ident2; X end; Xend; (* Func1 *) X X Xfunction Func2; (* (var String1ParRef, X String2ParRef: String30): boolean *) X (* executed once, returns false *) X (* String1ParRef = 'DHRYSTONE PROGRAM, 1''ST STRING' *) X (* String2ParRef = 'DHRYSTONE PROGRAM, 2''ND STRING' *) Xvar X IntLoc: OneToThirty; X CharLoc: CapitalLetter; Xbegin X IntLoc := 2; X while IntLoc <= 2 do (* loop body executed once *) X if Func1 (String1ParRef[IntLoc], X String2ParRef[IntLoc+1]) = Ident1 X then (* executed *) X begin X CharLoc := 'A'; X IntLoc := IntLoc + 1; X end; (* if, while *) X if (CharLoc >= 'W') and (CharLoc < 'Z') X then (* not executed *) X IntLoc := 7; X if CharLoc = 'R' X then (* not executed *) X Func2 := true X else (* executed *) X begin X if String1ParRef > String2ParRef X then (* not executed *) X begin X IntLoc := IntLoc + 7; X IntGlob := IntLoc; X Func2 := true X end X else (* executed *) X Func2 := false; X end; (* if CharLoc *) Xend; (* Func2 *) X X Xfunction Func3; (* (EnumParVal: Enumeration): boolean *) X (* executed once, returns true *) X (* EnumParVal = Ident3 *) Xvar X EnumLoc: Enumeration; Xbegin X EnumLoc := EnumParVal; X if EnumLoc = Ident3 X then (* executed *) X Func3 := true X else (* not executed *) X Func3 := false; Xend; (* Func3 *) X X Xbegin (* main program, corresponds to procedures *) X (* Main and Proc_0 in the Ada version *) X X (* Initializations *) X X new (NextPointerGlob); X X new (PointerGlob); X X PointerGlob^.PointerComp := NextPointerGlob; X PointerGlob^.Discr := Ident1; X PointerGlob^.EnumComp := Ident3; X PointerGlob^.IntComp := 40; X PointerGlob^.StringComp := 'DHRYSTONE PROGRAM, SOME STRING'; X X String1Glob := 'DHRYSTONE PROGRAM, 1''ST STRING'; X X Array2Glob [8,7] := 10; X X writeln; X writeln ('Dhrystone Benchmark, Version 2.1 (Language: Pascal)'); X writeln; X writeln ('Please give the number of runs through the benchmark: '); X readln (NumberOfRuns); X writeln; X writeln ('Execution starts, ', NumberOfRuns : 7, ' runs through Dhrystone'); X X BeginClock := clock; X X (***************) X (* Start timer *) X (***************) X X for RunIndex := 1 to NumberOfRuns do X begin X X Proc5; X Proc4; X (* Char1Glob = 'A', Char2Glob = 'B', BoolGlob = false *) X Int1Glob := 2; X Int2Glob := 3; X String2Glob := 'DHRYSTONE PROGRAM, 2''ND STRING'; X EnumGlob := Ident2; X BoolGlob := not Func2 (String1Glob, String2Glob); X (* BoolGlob = true *) X while Int1Glob < Int2Glob do (* loop body executed once *) X begin X Int3Glob := 5 * Int1Glob - Int2Glob; X (* Int3Glob = 7 *) X Proc7 (Int1Glob, Int2Glob, Int3Glob); X (* Int3Glob = 7 *) X Int1Glob := Int1Glob + 1; X end; (* while *) X (* Int1Glob = 3 *) X Proc8 (Array1Glob, Array2Glob, Int1Glob, Int3Glob); X (* IntGlob = 5 *) X Proc1 (PointerGlob); X for CharIndex := 'A' to Char2Glob do (* loop body executed twice *) X if EnumGlob = Func1 (CharIndex, 'C') X then (* not executed *) X begin X Proc6 (Ident1, EnumGlob); X String2Glob := 'DHRYSTONE PROGRAM, 3''RD STRING'; X Int2Glob := RunIndex; X IntGlob := RunIndex; X end; X (* Int1Glob = 3, Int2Glob = 3, Int3Glob = 7 *) X Int2Glob := Int2Glob * Int1Glob; X Int1Glob := Int2Glob div Int3Glob; X Int2Glob := 7 * (Int2Glob - Int3Glob) - Int1Glob; X (* Int1Glob = 1, Int2Glob = 13, Int3Glob = 7 *) X Proc2 (Int1Glob); X (* Int1Glob = 5 *) X X end; (* for RunIndex *) X X EndClock := clock; X X (**************) X (* Stop timer *) X (**************) X X writeln ('Execution ends'); X writeln; X writeln ('Final values of the variables used in the benchmark:'); X writeln; X X writeln ('IntGlob: ', IntGlob : 5); X writeln (' should be: 5'); X write ('BoolGlob: '); X if BoolGlob = true X then X writeln ('TRUE') X else X writeln ('FALSE'); X writeln (' should be: TRUE'); X writeln ('Char1Glob: ', Char1Glob); X writeln (' should be: A'); X writeln ('Char2Glob: ', Char2Glob); X writeln (' should be: B'); X writeln ('Array1Glob [8]: ', Array1Glob [8] : 5); X writeln (' should be: 7'); X writeln ('Array2Glob [8,7]: ', Array2Glob [8,7] : 5); X writeln (' should be: NumberOfRuns + 10'); X writeln ('PointerGlob^.Discr: ', ord (PointerGlob^.Discr) : 5); X writeln (' should be: 0'); X writeln ('PointerGlob^.EnumComp: ', ord (PointerGlob^.EnumComp) : 5); X writeln (' should be: 2'); X writeln ('PointerGlob^.IntComp : ', PointerGlob^.IntComp : 5); X writeln (' should be: 17'); X write ('PointerGlob^.StringComp: '); X for I := 1 to 30 do X write (PointerGlob^.StringComp [I]); X writeln; X writeln (' should be: DHRYSTONE PROGRAM, SOME STRING'); X writeln ('NextPointerGlob^.Discr: ', ord (NextPointerGlob^.Discr) : 5); X writeln (' should be: 0'); X writeln ('NextPointerGlob^.EnumComp: ', X ord (NextPointerGlob^.EnumComp) : 5); X writeln (' should be: 1'); X writeln ('NextPointerGlob^.IntComp: ', NextPointerGlob^.IntComp : 5); X writeln (' should be: 18'); X write ('NextPointerGlob^.StringComp: '); X for I := 1 to 30 do X write (NextPointerGlob^.StringComp [I]); X writeln; X writeln (' should be: DHRYSTONE PROGRAM, SOME STRING'); X writeln ('Int1Glob: ', Int1Glob : 5); X writeln (' should be: 5'); X writeln ('Int2Glob: ', Int2Glob : 5); X writeln (' should be: 13'); X writeln ('Int3Glob: ', Int3Glob : 5); X writeln (' should be: 7'); X writeln ('EnumGlob: ', ord (EnumGlob) : 5); X writeln (' should be: 1'); X write ('String1Glob: '); X for I := 1 to 30 do X write (String1Glob [I]); X writeln; X writeln (' should be: DHRYSTONE PROGRAM, 1''ST STRING'); X write ('String2Glob: '); X for I := 1 to 30 do X write (String2Glob [I]); X writeln; X writeln (' should be: DHRYSTONE PROGRAM, 2''ND STRING'); X writeln; X writeln; X X SumClocks := EndClock - BeginClock; X X if SumClocks < TooSmallTime X then X begin X writeln ('Measured time too small to obtain meaningful results'); X writeln ('Please increase number of runs'); X writeln; X end X else X begin X Microseconds := SumClocks * (MicrosecondsPerClock / NumberOfRuns); X (* Brackets to prevent integer overflow *) X DhrystonesPerSecond := NumberOfRuns * (ClocksPerSecond / SumClocks); X write ('Microseconds for one run through Dhrystone: '); X writeln (Microseconds : 8 : 1); X write ('Dhrystones per Second: '); X writeln (DhrystonesPerSecond : 8 : 1); X writeln; X end; X Xend. SHAR_EOF fi exit 0 # End of shell archive
donegan@stanton.TCC.COM (Steven P. Donegan) (12/06/88)
After unpacking the 3 parts of the Dhrystone benchmark version 2.1 and compiling via make without any errors or warnings, I attempted to run dry2. If less than 1000000 iterations are chosen it immediately exits, displaying a message that indicates the benchmark ran too fast (too few iterations). If 1000000 iterations are chosen it indicates 16k iterations are to be run then says my system has a whopping 346.1 dhrystones per second (which is at least an order of magnitude less than the earlier dhrystone 1.1 program). What the hell gives? I am running a 80286(16mhz 0wait) system with SCO XENIX 2.2.1, the full development system (Microsoft compiler, by sheer volume probably one of the most used C compilers other than PCC) and have no problems with the earlier dhrystone benchmark program. Any productive comments would be appreciated. -- Steven P. Donegan These opinions are given on MY time, not Sr. Telecommunications Analyst Western Digital's Western Digital Corp. stanton!donegan || donegan@stanton.TCC.COM || donegan%stanton@tcc.com
daveh@cbmvax.UUCP (Dave Haynie) (12/08/88)
in article <88@stanton.TCC.COM>, donegan@stanton.TCC.COM (Steven P. Donegan) says: > Keywords: integer benchmark > Summary: Bogus Results? > If less than 1000000 iterations are chosen it immediately exits, > displaying a message that indicates the benchmark ran too fast (too > few iterations). If 1000000 iterations are chosen it indicates 16k > iterations are to be run then says my system has a whopping 346.1 > dhrystones per second (which is at least an order of magnitude less than > the earlier dhrystone 1.1 program). What the hell gives? Did you try something really small, like say, less than 32K iterations? Based on the last version I tried at least, the Dhrystone code is sensitive to the length of your "int". So specifying small integers, or running it on a 16 bit machine, may run you into some strange problems when the iterations go over 32K or so. Whether or not it runs or kicks out is probably based on whether the number % 65536 comes out positive or negative. Try either a smaller number of iterations, or compile it for a 32 bit machine model, if you have that option. > Steven P. Donegan These opinions are given on MY time, not > Sr. Telecommunications Analyst Western Digital's > Western Digital Corp. > stanton!donegan || donegan@stanton.TCC.COM || donegan%stanton@tcc.com -- Dave Haynie "The 32 Bit Guy" Commodore-Amiga "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: D-DAVE H BIX: hazy Amiga -- It's not just a job, it's an obsession