| Message ID | 20241001185805.1332105-1-qing.zhao@oracle.com |
|---|---|
| State | New |
| Headers | show |
| Series | [Backport,GCC12] tree-optimization/116585 - SSA corruption with split_constant_offset | expand |
On Tue, 1 Oct 2024, Qing Zhao wrote: > From: Richard Biener <rguenther@suse.de> > > Hi, this is the backport of the fix for PR116585 to GCC12. > bootstrapped and regress tested on both X86 and aarch64. > > Okay for committing? OK. > thanks. > > Qing. > > ======================== > > split_constant_offset when looking through SSA defs can end up > picking SSA leafs that are subject to abnormal coalescing. This > can lead to downstream consumers to insert code based on the > result (like from dataref analysis) in places that violate constraints > for abnormal coalescing. It's best to not expand defs whose operands > are subject to abnormal coalescing - and not either do something when > a subexpression has operands like that already. > > PR tree-optimization/116585 > * tree-data-ref.cc (split_constant_offset_1): When either > operand is subject to abnormal coalescing do no further > processing. > > * gcc.dg/torture/pr116585.c: New testcase. > > (cherry picked from commit 1d0cb3b5fca69b81e69cfdb4aea0eebc1ac04750) > --- > gcc/testsuite/gcc.dg/torture/pr116585.c | 32 +++++++++++++++++++++++++ > gcc/tree-data-ref.cc | 11 ++++++--- > 2 files changed, 40 insertions(+), 3 deletions(-) > create mode 100644 gcc/testsuite/gcc.dg/torture/pr116585.c > > diff --git a/gcc/testsuite/gcc.dg/torture/pr116585.c b/gcc/testsuite/gcc.dg/torture/pr116585.c > new file mode 100644 > index 00000000000..108c481e104 > --- /dev/null > +++ b/gcc/testsuite/gcc.dg/torture/pr116585.c > @@ -0,0 +1,32 @@ > +/* { dg-do compile } */ > + > +char *s1, *s2; > +extern int* my_alloc (int); > +extern int _setjmp (); > +extern void bar(); > +void foo(int s1len, int s2len) > +{ > + int e; > + e = _setjmp (); > + { > + int l, i; > + int *md = my_alloc(((sizeof(int)) * (s1len + 1) * (s2len))); > + s1len++; > + for (; s1len; l) > + for (; s2len; l) > + for (; s1len; i) > + { > + int j = 1; > + for (; j < s2len; j++) > + { > + int cost; > + if (s1[1] == s2[1]) > + cost = 0; > + else > + cost = 1; > + md[j * s1len ] = ((cost)); > + } > + } > + bar(); > + } > +} > diff --git a/gcc/tree-data-ref.cc b/gcc/tree-data-ref.cc > index 706a49f226e..b7bca6a9d06 100644 > --- a/gcc/tree-data-ref.cc > +++ b/gcc/tree-data-ref.cc > @@ -761,6 +761,14 @@ split_constant_offset_1 (tree type, tree op0, enum tree_code code, tree op1, > if (INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_TRAPS (type)) > return false; > > + if (TREE_CODE (op0) == SSA_NAME > + && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op0)) > + return false; > + if (op1 > + && TREE_CODE (op1) == SSA_NAME > + && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op1)) > + return false; > + > switch (code) > { > case INTEGER_CST: > @@ -853,9 +861,6 @@ split_constant_offset_1 (tree type, tree op0, enum tree_code code, tree op1, > > case SSA_NAME: > { > - if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op0)) > - return false; > - > gimple *def_stmt = SSA_NAME_DEF_STMT (op0); > enum tree_code subcode; > >
diff --git a/gcc/testsuite/gcc.dg/torture/pr116585.c b/gcc/testsuite/gcc.dg/torture/pr116585.c new file mode 100644 index 00000000000..108c481e104 --- /dev/null +++ b/gcc/testsuite/gcc.dg/torture/pr116585.c @@ -0,0 +1,32 @@ +/* { dg-do compile } */ + +char *s1, *s2; +extern int* my_alloc (int); +extern int _setjmp (); +extern void bar(); +void foo(int s1len, int s2len) +{ + int e; + e = _setjmp (); + { + int l, i; + int *md = my_alloc(((sizeof(int)) * (s1len + 1) * (s2len))); + s1len++; + for (; s1len; l) + for (; s2len; l) + for (; s1len; i) + { + int j = 1; + for (; j < s2len; j++) + { + int cost; + if (s1[1] == s2[1]) + cost = 0; + else + cost = 1; + md[j * s1len ] = ((cost)); + } + } + bar(); + } +} diff --git a/gcc/tree-data-ref.cc b/gcc/tree-data-ref.cc index 706a49f226e..b7bca6a9d06 100644 --- a/gcc/tree-data-ref.cc +++ b/gcc/tree-data-ref.cc @@ -761,6 +761,14 @@ split_constant_offset_1 (tree type, tree op0, enum tree_code code, tree op1, if (INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_TRAPS (type)) return false; + if (TREE_CODE (op0) == SSA_NAME + && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op0)) + return false; + if (op1 + && TREE_CODE (op1) == SSA_NAME + && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op1)) + return false; + switch (code) { case INTEGER_CST: @@ -853,9 +861,6 @@ split_constant_offset_1 (tree type, tree op0, enum tree_code code, tree op1, case SSA_NAME: { - if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op0)) - return false; - gimple *def_stmt = SSA_NAME_DEF_STMT (op0); enum tree_code subcode;
From: Richard Biener <rguenther@suse.de> Hi, this is the backport of the fix for PR116585 to GCC12. bootstrapped and regress tested on both X86 and aarch64. Okay for committing? thanks. Qing. ======================== split_constant_offset when looking through SSA defs can end up picking SSA leafs that are subject to abnormal coalescing. This can lead to downstream consumers to insert code based on the result (like from dataref analysis) in places that violate constraints for abnormal coalescing. It's best to not expand defs whose operands are subject to abnormal coalescing - and not either do something when a subexpression has operands like that already. PR tree-optimization/116585 * tree-data-ref.cc (split_constant_offset_1): When either operand is subject to abnormal coalescing do no further processing. * gcc.dg/torture/pr116585.c: New testcase. (cherry picked from commit 1d0cb3b5fca69b81e69cfdb4aea0eebc1ac04750) --- gcc/testsuite/gcc.dg/torture/pr116585.c | 32 +++++++++++++++++++++++++ gcc/tree-data-ref.cc | 11 ++++++--- 2 files changed, 40 insertions(+), 3 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/torture/pr116585.c