| Message ID | alpine.LSU.2.11.1412011402020.5894@zhemvz.fhfr.qr |
|---|---|
| State | New |
| Headers | show |
On Mon, 1 Dec 2014, Richard Biener wrote: > The following fixes the oldest bug in my list of assigned PRs, namely > combining A / CST / CST' to A / (CST * CST'). extract_muldiv does > this in fold-const.c (but it fails to optimize to zero when CST * CST' > overflows). > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk. > > Richard. > > 2014-12-01 Richard Biener <rguenther@suse.de> > > PR tree-optimization/15346 > * Makefile.in (gimple-match.o-warn): Remove -Wno-unused-parameter, > add -Wno-unused-but-set-variable. > * match.pd: Combine two successive divisions. > > * gcc.dg/tree-ssa/forwprop-32.c: New testcase. > > Index: gcc/match.pd > =================================================================== > --- gcc/match.pd (revision 218140) > +++ gcc/match.pd (working copy) > @@ -129,6 +129,19 @@ (define_operator_list inverted_tcc_compa > && TYPE_UNSIGNED (type)) > (trunc_div @0 @1))) > > +/* Combine two successive divisions. */ > +(for div (trunc_div ceil_div floor_div round_div exact_div) > + (simplify > + (div (div @0 INTEGER_CST@1) INTEGER_CST@2) > + (with { > + bool overflow_p; > + wide_int mul = wi::mul (@1, @2, TYPE_SIGN (type), &overflow_p); > + } > + (if (!overflow_p) > + (div @0 { wide_int_to_tree (type, mul); })) > + (if (overflow_p) > + { build_zero_cst (type); })))) > + Can't you have something like: INT_MIN / 2 / (INT_MIN / -2) == -1 where 2 * (INT_MIN / -2) overflows?
On Mon, 1 Dec 2014, Richard Biener wrote: > +/* Combine two successive divisions. */ > +(for div (trunc_div ceil_div floor_div round_div exact_div) This doesn't seem correct for all kinds of division and signedness of arguments. TRUNC_DIV_EXPR (C division) and EXACT_DIV_EXPR should be OK regardless. But for CEIL_DIV_EXPR and FLOOR_DIV_EXPR, in ((a / b) / c) you need c to be positive so that both roundings are in the same direction (consider e.g. 3 FLOOR_DIV -2 FLOOR_DIV -2 == 1, but 3 FLOOR_DIV 4 == 0). And for ROUND_DIV_EXPR, it can be incorrect whatever the signs (e.g. 2 ROUND_DIV 3 ROUND_DIV 2 == 1, but 2 ROUND_DIV 6 == 0). I'm not sure if these forms of division actually occur in places where this could cause a problem, but it does look like Ada may enable you to generate ROUND_DIV_EXPR.
Index: gcc/match.pd =================================================================== --- gcc/match.pd (revision 218140) +++ gcc/match.pd (working copy) @@ -129,6 +129,19 @@ (define_operator_list inverted_tcc_compa && TYPE_UNSIGNED (type)) (trunc_div @0 @1))) +/* Combine two successive divisions. */ +(for div (trunc_div ceil_div floor_div round_div exact_div) + (simplify + (div (div @0 INTEGER_CST@1) INTEGER_CST@2) + (with { + bool overflow_p; + wide_int mul = wi::mul (@1, @2, TYPE_SIGN (type), &overflow_p); + } + (if (!overflow_p) + (div @0 { wide_int_to_tree (type, mul); })) + (if (overflow_p) + { build_zero_cst (type); })))) + /* Optimize A / A to 1.0 if we don't care about NaNs or Infinities. */ (simplify Index: gcc/testsuite/gcc.dg/tree-ssa/forwprop-32.c =================================================================== --- gcc/testsuite/gcc.dg/tree-ssa/forwprop-32.c (revision 0) +++ gcc/testsuite/gcc.dg/tree-ssa/forwprop-32.c (working copy) @@ -0,0 +1,18 @@ +/* { dg-do compile } */ +/* { dg-options "-O -fdump-tree-forwprop1 -fdump-tree-ccp1" } */ + +int foo (int x) +{ + int tem = x / 3; + return tem / 5; +} +int bar (int x) +{ + int tem = x / 3; + return tem / (__INT_MAX__ / 2); +} + +/* { dg-final { scan-tree-dump "x_.\\(D\\) / 15" "forwprop1" } } */ +/* { dg-final { scan-tree-dump "return 0;" "ccp1" } } */ +/* { dg-final { cleanup-tree-dump "forwprop1" } } */ +/* { dg-final { cleanup-tree-dump "ccp1" } } */ Index: gcc/Makefile.in =================================================================== --- gcc/Makefile.in (revision 218142) +++ gcc/Makefile.in (working copy) @@ -209,8 +209,8 @@ gengtype-lex.o-warn = -Wno-error libgcov-util.o-warn = -Wno-error libgcov-driver-tool.o-warn = -Wno-error libgcov-merge-tool.o-warn = -Wno-error -gimple-match.o-warn = -Wno-unused-variable -Wno-unused-parameter -generic-match.o-warn = -Wno-unused-variable -Wno-unused-parameter +gimple-match.o-warn = -Wno-unused-variable -Wno-unused-but-set-variable +generic-match.o-warn = -Wno-unused-variable -Wno-unused-but-set-variable # All warnings have to be shut off in stage1 if the compiler used then # isn't gcc; configure determines that. WARN_CFLAGS will be either