| Message ID | alpine.LSU.2.20.1812031436000.1827@zhemvz.fhfr.qr |
|---|---|
| State | New |
| Headers | show |
| Series | Fix PR88301 | expand |
Hi Richard, On Mon, 3 Dec 2018 at 14:37, Richard Biener <rguenther@suse.de> wrote: > > > This fixes a missed optimization in EVRP by teaching the code > figuring out conditional asserts about conversions that preserve > the converted value. > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk > sofar. > > Richard. > > 2018-12-03 Richard Biener <rguenther@suse.de> > > PR tree-optimization/88301 > * tree-vrp.c (register_edge_assert_for_2): Handle conversions > that do not change the value by registering the same assert > for the operand. > > * gcc.dg/tree-ssa/evrp13.c: New testcase. > I noticed a regression in g++ on aarch64, and git bisect indicates this commits is the culprit. g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for warnings, line 14) g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for excess errors) g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for warnings, line 14) g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for excess errors) g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for warnings, line 14) g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for excess errors) g++.log says: Excess errors: /gcc/testsuite/g++.dg/warn/Wstringop-overflow-1.C:14:21: warning: 'char* __builtin_strncpy(char*, const char*, long unsigned int)' writing between 6 and 2147483647 bytes into a region of size 5 overflows the destination [-Wstringop-overflow=] Christophe > Index: gcc/tree-vrp.c > =================================================================== > --- gcc/tree-vrp.c (revision 266733) > +++ gcc/tree-vrp.c (working copy) > @@ -2966,6 +2966,23 @@ register_edge_assert_for_2 (tree name, e > add_assert_info (asserts, name2, tmp, new_comp_code, new_val); > } > > + /* If we have a conversion that doesn't change the value of the source > + simply register the same assert for it. */ > + if (CONVERT_EXPR_CODE_P (rhs_code)) > + { > + wide_int rmin, rmax; > + tree rhs1 = gimple_assign_rhs1 (def_stmt); > + if (INTEGRAL_TYPE_P (TREE_TYPE (rhs1)) > + && int_fits_type_p (val, TREE_TYPE (rhs1)) > + && ((TYPE_PRECISION (TREE_TYPE (name)) > + > TYPE_PRECISION (TREE_TYPE (rhs1))) > + || (get_range_info (rhs1, &rmin, &rmax) == VR_RANGE > + && wi::fits_to_tree_p (rmin, TREE_TYPE (name)) > + && wi::fits_to_tree_p (rmax, TREE_TYPE (name))))) > + add_assert_info (asserts, rhs1, rhs1, > + comp_code, fold_convert (TREE_TYPE (rhs1), val)); > + } > + > /* Add asserts for NAME cmp CST and NAME being defined as > NAME = NAME2 & CST2. > > Index: gcc/testsuite/gcc.dg/tree-ssa/evrp13.c > =================================================================== > --- gcc/testsuite/gcc.dg/tree-ssa/evrp13.c (revision 0) > +++ gcc/testsuite/gcc.dg/tree-ssa/evrp13.c (working copy) > @@ -0,0 +1,7 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O2 -fdump-tree-evrp" } */ > + > +#define ADD_NW(A,B) (__extension__({ __typeof(A+B) R; if(__builtin_add_overflow(A,B,&R)) __builtin_unreachable(); R ;})) > +_Bool a_b2(unsigned A, unsigned B) { return ADD_NW(A,B) >= B; } > + > +/* { dg-final { scan-tree-dump "return 1;" "evrp" } } */
On Tue, 4 Dec 2018, Christophe Lyon wrote: > Hi Richard, > On Mon, 3 Dec 2018 at 14:37, Richard Biener <rguenther@suse.de> wrote: > > > > > > This fixes a missed optimization in EVRP by teaching the code > > figuring out conditional asserts about conversions that preserve > > the converted value. > > > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk > > sofar. > > > > Richard. > > > > 2018-12-03 Richard Biener <rguenther@suse.de> > > > > PR tree-optimization/88301 > > * tree-vrp.c (register_edge_assert_for_2): Handle conversions > > that do not change the value by registering the same assert > > for the operand. > > > > * gcc.dg/tree-ssa/evrp13.c: New testcase. > > > > I noticed a regression in g++ on aarch64, and git bisect indicates > this commits is the culprit. > g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for warnings, line 14) > g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for excess errors) > g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for warnings, line 14) > g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for excess errors) > g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for warnings, line 14) > g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for excess errors) > > g++.log says: > Excess errors: > /gcc/testsuite/g++.dg/warn/Wstringop-overflow-1.C:14:21: warning: > 'char* __builtin_strncpy(char*, const char*, long unsigned int)' > writing between 6 and 2147483647 bytes into a region of size 5 > overflows the destination [-Wstringop-overflow=] Yes, sorry. r266773 should have fixed this. Richard.
On Tue, 4 Dec 2018 at 14:17, Richard Biener <rguenther@suse.de> wrote: > > On Tue, 4 Dec 2018, Christophe Lyon wrote: > > > Hi Richard, > > On Mon, 3 Dec 2018 at 14:37, Richard Biener <rguenther@suse.de> wrote: > > > > > > > > > This fixes a missed optimization in EVRP by teaching the code > > > figuring out conditional asserts about conversions that preserve > > > the converted value. > > > > > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk > > > sofar. > > > > > > Richard. > > > > > > 2018-12-03 Richard Biener <rguenther@suse.de> > > > > > > PR tree-optimization/88301 > > > * tree-vrp.c (register_edge_assert_for_2): Handle conversions > > > that do not change the value by registering the same assert > > > for the operand. > > > > > > * gcc.dg/tree-ssa/evrp13.c: New testcase. > > > > > > > I noticed a regression in g++ on aarch64, and git bisect indicates > > this commits is the culprit. > > g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for warnings, line 14) > > g++.dg/warn/Wstringop-overflow-1.C -std=c++14 (test for excess errors) > > g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for warnings, line 14) > > g++.dg/warn/Wstringop-overflow-1.C -std=c++17 (test for excess errors) > > g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for warnings, line 14) > > g++.dg/warn/Wstringop-overflow-1.C -std=c++98 (test for excess errors) > > > > g++.log says: > > Excess errors: > > /gcc/testsuite/g++.dg/warn/Wstringop-overflow-1.C:14:21: warning: > > 'char* __builtin_strncpy(char*, const char*, long unsigned int)' > > writing between 6 and 2147483647 bytes into a region of size 5 > > overflows the destination [-Wstringop-overflow=] > > Yes, sorry. r266773 should have fixed this. > Indeed, thanks. > Richard.
Index: gcc/tree-vrp.c =================================================================== --- gcc/tree-vrp.c (revision 266733) +++ gcc/tree-vrp.c (working copy) @@ -2966,6 +2966,23 @@ register_edge_assert_for_2 (tree name, e add_assert_info (asserts, name2, tmp, new_comp_code, new_val); } + /* If we have a conversion that doesn't change the value of the source + simply register the same assert for it. */ + if (CONVERT_EXPR_CODE_P (rhs_code)) + { + wide_int rmin, rmax; + tree rhs1 = gimple_assign_rhs1 (def_stmt); + if (INTEGRAL_TYPE_P (TREE_TYPE (rhs1)) + && int_fits_type_p (val, TREE_TYPE (rhs1)) + && ((TYPE_PRECISION (TREE_TYPE (name)) + > TYPE_PRECISION (TREE_TYPE (rhs1))) + || (get_range_info (rhs1, &rmin, &rmax) == VR_RANGE + && wi::fits_to_tree_p (rmin, TREE_TYPE (name)) + && wi::fits_to_tree_p (rmax, TREE_TYPE (name))))) + add_assert_info (asserts, rhs1, rhs1, + comp_code, fold_convert (TREE_TYPE (rhs1), val)); + } + /* Add asserts for NAME cmp CST and NAME being defined as NAME = NAME2 & CST2. Index: gcc/testsuite/gcc.dg/tree-ssa/evrp13.c =================================================================== --- gcc/testsuite/gcc.dg/tree-ssa/evrp13.c (revision 0) +++ gcc/testsuite/gcc.dg/tree-ssa/evrp13.c (working copy) @@ -0,0 +1,7 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-evrp" } */ + +#define ADD_NW(A,B) (__extension__({ __typeof(A+B) R; if(__builtin_add_overflow(A,B,&R)) __builtin_unreachable(); R ;})) +_Bool a_b2(unsigned A, unsigned B) { return ADD_NW(A,B) >= B; } + +/* { dg-final { scan-tree-dump "return 1;" "evrp" } } */