Looks good to me. Please go ahead. 2012/2/13 Gang Yu <yugang...@gmail.com>: > Hi, > > Could a gatekeeper please help review the fix for > bug948(https://bugs.open64.net/show_bug.cgi?id=948)? > > A cutdown case below: > > typedef struct chunk_t chunk_t; > struct chunk_t { > unsigned char *ptr; > unsigned int len; > }; > extern chunk_t chunk_empty; > void add_sa(void *this, > unsigned int spi, > unsigned short int ipcomp, > unsigned short int cpi) > { > char *alg_name; > unsigned short icv_size = 64; > if (ipcomp != 241 && cpi != 0) > { > add_sa(this, > (__extension__ > ({ register unsigned int __v, __x = > ((__extension__ > ({ register unsigned short int __v, __x = (cpi); > __asm__ ("rorw $8, %w0" : "=r" (__v) : "0" (__x) : "cc"); > __v; }))); > __asm__ ("bswap %0" : "=r" (__v) : "0" (__x)); > __v; })) , > ipcomp,0); > ipcomp = 241; > } > } > with debug compiler: > opencc -S bug948-2.c -O3 > ### Assertion failure at line 1167 of > /fc/proj/ctires/open64/o64guru/src/Sun/trunk/osprey/be/cg/whirl2ops.cxx: > ### Compiler Error in file bug948-2.c during Code_Expansion phase: > ### PREG_To_TN: non-positive preg_num > opencc INTERNAL ERROR: > /fc/proj/ctires/open64/o64guru/OPEN64_X86_DBG/LATEST/bits//lib/gcc-lib/x86_64-open64-linux/5.0/be > returned non-zero status 1 > > Analysis: > WHIRL output after WOPT for the unexpected part is as below, marked with > red colors: > ASM_STMT 4 <2,11,rorw $8, %w0> # "rorw $8, %w0" (cc) {line: 1/25} > BLOCK {line: 0/0} > END_BLOCK > BLOCK {line: 0/0} > PRAGMA 0 175 <1,7,.preg_U2> -2199023252222 (0xfffffe0000000d02) # > ASM_CONSTRAINT, "=r", opnd:0 preg:-2 {line: 0/0} > END_BLOCK > ASM_INPUTS > U4U2LDID 51 <1,8,.preg_U4> T<8,.predef_U4,2> # cpi > ASM_INPUT opnd:1 <2,12,0> # "0" > END_ASM_INPUTS > U4U4LDID 0 <2,9,__v> T<8,.predef_U4,4> > U4U2LDID -2 <1,7,.preg_U2> T<7,.predef_U2,2> > U4COMPOSE_BITS <bofst:0 bsize:16> > U4STID 50 <1,8,.preg_U4> T<8,.predef_U4,1> # __v {line: 1/25} > U4U4LDID 50 <1,8,.preg_U4> T<8,.predef_U4,1> # __v > U4EXTRACT_BITS <bofst:0 bsize:16> > U2STID 0 <1,52,anon2> T<7,.predef_U2,2> {line: 1/26} > ASM_STMT 0 <2,14,bswap %0> # "bswap %0" {line: 1/27} > Since r3866 has enabled copy-propagation into the ASM_INPUTS, there are more > chances to do optimization for those ASM included programs. > at O3 level, snippet before the patch: > > U4U2LDID -2 <1,7,.preg_U2> T<7,.predef_U2,2> > U4CVTL 16 > U4STID 51 <1,8,.preg_U4> T<8,.predef_U4,2> # __v {line: 1/25} > U4U4LDID 51 <1,8,.preg_U4> T<8,.predef_U4,2> # __v > U2STID 0 <1,52,anon2> T<7,.predef_U2,2> {line: 1/26} > U4U4LDID 51 <1,8,.preg_U4> T<8,.predef_U4,2> # __v > U4STID 50 <1,8,.preg_U4> T<8,.predef_U4,4> # __x {line: 1/26} > is now more optimized to > U4U4LDID 0 <2,9,__v> T<8,.predef_U4,4> > U4U2LDID -2 <1,7,.preg_U2> T<7,.predef_U2,2> # (Gang) inner mutated > __v > U4COMPOSE_BITS <bofst:0 bsize:16> > U4STID 50 <1,8,.preg_U4> T<8,.predef_U4,1> # __v {line: 1/25} # (Gang) __v > here is the outer __v!! > U4U4LDID 50 <1,8,.preg_U4> T<8,.predef_U4,1> # __v > U4EXTRACT_BITS <bofst:0 bsize:16> > U2STID 0 <1,52,anon2> T<7,.predef_U2,2> {line: 1/26} > > reduced define and use of __x, this is still right and good. > Now, we could not assume the sink of ASM_STMT pregs as (defined > in Find_Asm_Out_Parameter_Load/whirl2ops.cxx) in: > > > // LDID preg_num > // STID > // ... > // or > // ... > // LDID preg_num > // CVT(CVTL) > // STID > forms. > > Suggested patch: > > --- a/osprey/be/cg/whirl2ops.cxx > +++ b/osprey/be/cg/whirl2ops.cxx > @@ -114,6 +114,7 @@ > #include "be_symtab.h" > #include "be_util.h" > #include "config_asm.h" > +#include "wn_tree_util.h" > #if defined(TARG_X8664) || defined(TARG_NVISA) || defined(TARG_SL) || > defined(TARG_PPC32) > #include "cgexp_internals.h" > @@ -6691,20 +6692,40 @@ Find_Asm_Out_Parameter_Load (const WN* stmt, > PREG_NUM preg_num, ST** ded_st) > #endif > if (OPERATOR_is_store(WN_operator(stmt))) { > WN* load = WN_kid0(stmt); > - OPERATOR opr = WN_operator(load); > - if (opr == OPR_CVT || opr == OPR_CVTL) { > - load = WN_kid0(load); > - opr = WN_operator(load); > - } > - if (OPERATOR_is_load(opr) || opr == OPR_LDA) { > - if (WN_has_sym(load) && > - WN_class(load) == CLASS_PREG && > - WN_offset(load) == preg_num) { > - ret_load = load; > - break; > + // open64.net bug948. Since we allow copy progration into the > + // ASM_INPUT. More aggressive optimizations could be made. > + // Now, we could not always assume the "LDID preg_num" for ASM_STMTS > output > + // appears in > + // ... > + // LDID preg_num > + // STID > + // ... > + // or > + // ... > + // LDID preg_num > + // CVT(CVTL) > + // STID > + // ... > + // forms in the succeeding stmts. > + // Instead, we now iterate on the store kid to find the "LDID > preg_num" > + // expressions. > + WN_TREE_CONTAINER<PRE_ORDER> lpre(load); > + WN_TREE_CONTAINER<PRE_ORDER> :: iterator lipre; > + for ( lipre = lpre.begin() ; lipre != lpre.end() ; ++ lipre) { > + WN* load_expr= lipre.Wn(); > + OPERATOR opr = WN_operator(load_expr); > + if (OPERATOR_is_load(opr) || opr == OPR_LDA) { > + if (WN_has_sym(load_expr) && > + WN_class(load_expr) == CLASS_PREG && > + WN_offset(load_expr) == preg_num) { > + ret_load = load_expr; > + break; > + } > } > } > } > + if (ret_load) > + break; > } > if (ret_load) { > if (OPERATOR_has_sym(WN_operator(stmt)) && > > Could a gatekeeper please help a review? thanks > > > Regards > Gang > > ------------------------------------------------------------------------------ > Try before you buy = See our experts in action! > The most comprehensive online learning library for Microsoft developers > is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3, > Metro Style Apps, more. Free future releases when you subscribe now! > http://p.sf.net/sfu/learndevnow-dev2 > _______________________________________________ > Open64-devel mailing list > Open64-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/open64-devel >
-- Regards, Lai Jian-Xin ------------------------------------------------------------------------------ Virtualization & Cloud Management Using Capacity Planning Cloud computing makes use of virtualization - but cloud computing also focuses on allowing computing to be delivered as a service. http://www.accelacomm.com/jaw/sfnl/114/51521223/ _______________________________________________ Open64-devel mailing list Open64-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/open64-devel
