Commit c12cc930 by Keith Besaw Committed by Ira Rosen

common.opt: Add option ftree-vect-loop-version.

        * common.opt: Add option ftree-vect-loop-version.
        * params.def: Add --param vect-max-version-checks.
        * doc/invoke.texi: Document ftree-vect-loop-version and
        --param vect-max-version-checks.
        * tree-vectorizer.h (_loop_vec_info): Add ptr_mask and
        may_misalign_stmts and defines for accessors.
        * tree-vectorizer.c : (new_loop_vec_info): VEC_alloc for
        LOOP_VINFO_MAY_MISALIGN_STMTS.
        (destroy_loop_vec_info): VEC_free for
        LOOP_VINFO_MAY_MISALIGN_STMTS.
        * tree-vect-analyze.c (vect_compute_data_ref_alignment):
        Update documentation.
        (vect_update_misalignment_for_peel): New.
        (vect_enhance_data_refs_alignment): Update to choose loop
        peeling or loop versioning if appropriate for the (potentially)
        unaligned data references in the loop.
        (vect_analyze_data_refs_alignment): Remove call to
        vect_enhance_data_refs_alignment so the checks can be done
        earlier.
        (vect_analyze_loop): Add call to vect_enhance_data_refs_alignment
        and move up call to vect_analyze_data_refs_alignment.
        * tree-vect-transform.c (vect_create_cond_for_align_checks): New.
        (vect_transform_loop): Add call to loop_version.

From-SVN: r103941
parent 4fe8db68
2005-09-06 Keith Besaw <kbesaw@us.ibm.com>
* common.opt: Add option ftree-vect-loop-version.
* params.def: Add --param vect-max-version-checks.
* doc/invoke.texi: Document ftree-vect-loop-version and
--param vect-max-version-checks.
* tree-vectorizer.h (_loop_vec_info): Add ptr_mask and
may_misalign_stmts and defines for accessors.
* tree-vectorizer.c : (new_loop_vec_info): VEC_alloc for
LOOP_VINFO_MAY_MISALIGN_STMTS.
(destroy_loop_vec_info): VEC_free for
LOOP_VINFO_MAY_MISALIGN_STMTS.
* tree-vect-analyze.c (vect_compute_data_ref_alignment):
Update documentation.
(vect_update_misalignment_for_peel): New.
(vect_enhance_data_refs_alignment): Update to choose loop
peeling or loop versioning if appropriate for the (potentially)
unaligned data references in the loop.
(vect_analyze_data_refs_alignment): Remove call to
vect_enhance_data_refs_alignment so the checks can be done
earlier.
(vect_analyze_loop): Add call to vect_enhance_data_refs_alignment
and move up call to vect_analyze_data_refs_alignment.
* tree-vect-transform.c (vect_create_cond_for_align_checks): New.
(vect_transform_loop): Add call to loop_version.
2005-09-06 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/23098
......
......@@ -998,6 +998,10 @@ ftree-vectorize
Common Report Var(flag_tree_vectorize)
Enable loop vectorization on trees
ftree-vect-loop-version
Common Report Var(flag_tree_vect_loop_version) Init(1)
Enable loop versioning when doing loop vectorization on trees
ftree-vectorizer-verbose=
Common RejectNegative Joined
-ftree-vectorizer-verbose=<number> Set the verbosity level of the vectorizer
......
......@@ -337,7 +337,7 @@ Objective-C and Objective-C++ Dialects}.
-ftree-loop-linear -ftree-loop-im -ftree-loop-ivcanon -fivopts @gol
-ftree-dominator-opts -ftree-dse -ftree-copyrename -ftree-sink @gol
-ftree-ch -ftree-sra -ftree-ter -ftree-lrs -ftree-fre -ftree-vectorize @gol
-ftree-salias -fweb @gol
-ftree-vect-loop-version -ftree-salias -fweb @gol
-ftree-copy-prop -ftree-store-ccp -ftree-store-copy-prop -fwhole-program @gol
--param @var{name}=@var{value}
-O -O0 -O1 -O2 -O3 -Os}
......@@ -4391,7 +4391,8 @@ optimizations designed to reduce code size.
@option{-Os} disables the following optimization flags:
@gccoptlist{-falign-functions -falign-jumps -falign-loops @gol
-falign-labels -freorder-blocks -freorder-blocks-and-partition -fprefetch-loop-arrays}
-falign-labels -freorder-blocks -freorder-blocks-and-partition @gol
-fprefetch-loop-arrays -ftree-vect-loop-version}
If you use multiple @option{-O} options, with or without level numbers,
the last such option is the one that is effective.
......@@ -5038,6 +5039,15 @@ optimization later. This is enabled by default at @option{-O} and higher.
@item -ftree-vectorize
Perform loop vectorization on trees.
@item -ftree-vect-loop-version
@opindex ftree-vect-loop-version
Perform loop versioning when doing loop vectorization on trees. When a loop
appears to be vectorizable except that data alignment or data dependence cannot
be determined at compile time then vectorized and non-vectorized versions of
the loop are generated along with runtime checks for alignment or dependence
to control which version is executed. This option is enabled by default
except at level @option{-Os} where it is disabled.
@item -ftree-vrp
Perform Value Range Propagation on trees. This is similar to the
constant propagation pass, but instead of values, ranges of values are
......@@ -5883,6 +5893,11 @@ optimization when a new iv is added to the set.
Bound on size of expressions used in the scalar evolutions analyzer.
Large expressions slow the analyzer.
@item vect-max-version-checks
The maxinum number of runtime checks that can be performed when doing
loop versioning in the vectorizer. See option ftree-vect-loop-version
for more information.
@item max-iterations-to-track
The maximum number of iterations of a loop the brute force algorithm
......
......@@ -398,6 +398,11 @@ DEFPARAM(PARAM_SCEV_MAX_EXPR_SIZE,
"Bound on size of expressions used in the scalar evolutions analyzer",
20, 0, 0)
DEFPARAM(PARAM_VECT_MAX_VERSION_CHECKS,
"vect-max-version-checks",
"Bound on number of runtime checks inserted by the vectorizer's loop versioning",
6, 0, 0)
/* The product of the next two is used to decide whether or not to
use .GLOBAL_VAR. See tree-dfa.c. */
DEFPARAM(PARAM_GLOBAL_VAR_THRESHOLD,
......
2005-09-06 Keith Besaw <kbesaw@us.ibm.com>
* gcc.dg/vect/vect-29.c: Update xfail testing.
* gcc.dg/vect/vect-44.c: Same.
* gcc.dg/vect/vect-48.c: Same.
* gcc.dg/vect/vect-50.c: Same.
* gcc.dg/vect/vect-72.c: Same.
* gcc.dg/vect/vect-77.c: Same.
* gcc.dg/vect/vect-78.c: Same.
* gcc.dg/vect/vect-80.c: Same.
* gcc.dg/vect/vect-96.c: Same.
* gcc.dg/vect/vect-100.c: Same.
* gcc.dg/vect/vect-exp: Add default flags for Os-vect-*.
* gfortran.dg/vect/vect-2.f90: Update xfail testing.
* gfortran.dg/vect/vect-5.f90: Same.
2005-09-06 Jakub Jelinek <jakub@redhat.com>
* gcc.dg/debug/dwarf2/dwarf-char1.c: Accept more assembler comment
......@@ -21,7 +21,7 @@ int main1 () {
p = (struct extraction *) malloc (sizeof (struct extraction));
/* Not vectorizable: p may alias a and/or b, since they are globals. */
/* Vectorizable: alias analysis determines that p can't point to a and/or b. */
for (i = 0; i < N; i++)
{
p->a[i] = a[i];
......@@ -73,6 +73,6 @@ int main (void)
}
/* Requires versioning. */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "vectorized 2 loops" 0 "vect" } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -43,7 +43,11 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the load.
(The store is aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 1 "vect" {target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -51,7 +51,12 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the
all three accesses (peeling to align the store will not force the
two loads to be aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 3 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -52,7 +52,11 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the two loads.
(The store is aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 2 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -49,7 +49,12 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the
all three accesses (peeling to align the store will not force the
two loads to be aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 3 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -51,7 +51,11 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the two loads.
(The store is aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 2 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -43,7 +43,11 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail { vect_no_align } } } } */
/* For targets that don't support misaligned loads we version for the load.
(The store is aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 1 "vect" { xfail { vect_no_align } } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 1 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -44,7 +44,11 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail { vect_no_align } } } } */
/* For targets that don't support misaligned loads we version for the load.
(The store is aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 1 "vect" { xfail { vect_no_align } } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 0 "vect" } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 1 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -43,7 +43,12 @@ int main (void)
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* For targets that don't support misaligned loads we version for the
all three accesses (peeling to align the store will not force the
two loads to be aligned). */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 3 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -37,7 +37,12 @@ int main (void)
return main1 (8);
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail vect_no_align } } } */
/* The store is unaligned, the load is aligned. For targets that support unaligned
loads, peel to align the store and generated unaligned access for the loads.
For targets that don't support unaligned loads, version for the store. */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail vect_no_align } } } */
/* { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 1 "vect" { target vect_no_align } } } */
/* { dg-final { cleanup-tree-dump "vect" } } */
......@@ -102,6 +102,11 @@ lappend DEFAULT_VECTCFLAGS "-ftrapv"
dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/trapv-vect*.\[cS\]]] \
"" $DEFAULT_VECTCFLAGS
# With -Os
lappend DEFAULT_VECTCFLAGS "-Os"
dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/Os-vect-*.\[cS\]]] \
"" $DEFAULT_VECTCFLAGS
# Clean up.
set dg-do-what-default ${save-dg-do-what-default}
......
......@@ -8,7 +8,15 @@ A = LOG(X); B = LOG(Y); C = A + B
PRINT*, C(500000)
END
! { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { xfail vect_no_align } } }
! First loop (A=LOG(X)) is vectorized using peeling to align the store.
! Same for the second loop (B=LOG(Y)).
! Third loop (C = A + B) is vectorized using versioning (for targets that don't
! support unaligned loads) or using peeling to align the store (on targets that
! support unaligned loads).
! { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" } }
! { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 3 "vect" { xfail vect_no_align } } }
! { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 2 "vect" { target vect_no_align } } }
! { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail vect_no_align } } }
! { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 3 "vect" {target vect_no_align } } }
! { dg-final { cleanup-tree-dump "vect" } }
......@@ -35,9 +35,10 @@
stop
end
! { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { xfail { vect_no_align || lp64 } } } }
! { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { xfail { lp64 } } } }
! { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail { vect_no_align || lp64 } } } }
! { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 2 "vect" { xfail { vect_no_align || lp64 } } } }
! { dg-final { scan-tree-dump-times "Alignment of access forced using versioning." 3 "vect" {target vect_no_align } } }
! We also expect to vectorize one loop for lp64 targets that support
! misaligned access:
......
......@@ -336,7 +336,7 @@ vect_create_data_ref_ptr (tree stmt, block_stmt_iterator *bsi, tree offset,
if (vect_print_dump_info (REPORT_DETAILS))
{
tree data_ref_base = base_name;
fprintf (vect_dump, "create array_ref of type: ");
fprintf (vect_dump, "create vector-pointer variable to type: ");
print_generic_expr (vect_dump, vectype, TDF_SLIM);
if (TREE_CODE (data_ref_base) == VAR_DECL)
fprintf (vect_dump, " vectorizing a one dimensional array ref: ");
......@@ -2697,6 +2697,128 @@ vect_do_peeling_for_alignment (loop_vec_info loop_vinfo, struct loops *loops)
}
/* Function vect_create_cond_for_align_checks.
Create a conditional expression that represents the alignment checks for
all of data references (array element references) whose alignment must be
checked at runtime.
Input:
LOOP_VINFO - two fields of the loop information are used.
LOOP_VINFO_PTR_MASK is the mask used to check the alignment.
LOOP_VINFO_MAY_MISALIGN_STMTS contains the refs to be checked.
Output:
COND_EXPR_STMT_LIST - statements needed to construct the conditional
expression.
The returned value is the conditional expression to be used in the if
statement that controls which version of the loop gets executed at runtime.
The algorithm makes two assumptions:
1) The number of bytes "n" in a vector is a power of 2.
2) An address "a" is aligned if a%n is zero and that this
test can be done as a&(n-1) == 0. For example, for 16
byte vectors the test is a&0xf == 0. */
static tree
vect_create_cond_for_align_checks (loop_vec_info loop_vinfo,
tree *cond_expr_stmt_list)
{
VEC(tree,heap) *may_misalign_stmts
= LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo);
tree ref_stmt;
int mask = LOOP_VINFO_PTR_MASK (loop_vinfo);
tree mask_cst;
unsigned int i;
tree psize;
tree int_ptrsize_type;
char tmp_name[20];
tree or_tmp_name = NULL_TREE;
tree and_tmp, and_tmp_name, and_stmt;
tree ptrsize_zero;
/* Check that mask is one less than a power of 2, i.e., mask is
all zeros followed by all ones. */
gcc_assert ((mask != 0) && ((mask & (mask+1)) == 0));
/* CHECKME: what is the best integer or unsigned type to use to hold a
cast from a pointer value? */
psize = TYPE_SIZE (ptr_type_node);
int_ptrsize_type
= lang_hooks.types.type_for_size (tree_low_cst (psize, 1), 0);
/* Create expression (mask & (dr_1 || ... || dr_n)) where dr_i is the address
of the first vector of the i'th data reference. */
for (i = 0; VEC_iterate (tree, may_misalign_stmts, i, ref_stmt); i++)
{
tree new_stmt_list = NULL_TREE;
tree addr_base;
tree addr_tmp, addr_tmp_name, addr_stmt;
tree or_tmp, new_or_tmp_name, or_stmt;
/* create: addr_tmp = (int)(address_of_first_vector) */
addr_base = vect_create_addr_base_for_vector_ref (ref_stmt,
&new_stmt_list,
NULL_TREE);
if (new_stmt_list != NULL_TREE)
append_to_statement_list_force (new_stmt_list, cond_expr_stmt_list);
sprintf (tmp_name, "%s%d", "addr2int", i);
addr_tmp = create_tmp_var (int_ptrsize_type, tmp_name);
add_referenced_tmp_var (addr_tmp);
addr_tmp_name = make_ssa_name (addr_tmp, NULL_TREE);
addr_stmt = fold_convert (int_ptrsize_type, addr_base);
addr_stmt = build2 (MODIFY_EXPR, void_type_node,
addr_tmp_name, addr_stmt);
SSA_NAME_DEF_STMT (addr_tmp_name) = addr_stmt;
append_to_statement_list_force (addr_stmt, cond_expr_stmt_list);
/* The addresses are OR together. */
if (or_tmp_name != NULL_TREE)
{
/* create: or_tmp = or_tmp | addr_tmp */
sprintf (tmp_name, "%s%d", "orptrs", i);
or_tmp = create_tmp_var (int_ptrsize_type, tmp_name);
add_referenced_tmp_var (or_tmp);
new_or_tmp_name = make_ssa_name (or_tmp, NULL_TREE);
or_stmt = build2 (MODIFY_EXPR, void_type_node, new_or_tmp_name,
build2 (BIT_IOR_EXPR, int_ptrsize_type,
or_tmp_name,
addr_tmp_name));
SSA_NAME_DEF_STMT (new_or_tmp_name) = or_stmt;
append_to_statement_list_force (or_stmt, cond_expr_stmt_list);
or_tmp_name = new_or_tmp_name;
}
else
or_tmp_name = addr_tmp_name;
} /* end for i */
mask_cst = build_int_cst (int_ptrsize_type, mask);
/* create: and_tmp = or_tmp & mask */
and_tmp = create_tmp_var (int_ptrsize_type, "andmask" );
add_referenced_tmp_var (and_tmp);
and_tmp_name = make_ssa_name (and_tmp, NULL_TREE);
and_stmt = build2 (MODIFY_EXPR, void_type_node,
and_tmp_name,
build2 (BIT_AND_EXPR, int_ptrsize_type,
or_tmp_name, mask_cst));
SSA_NAME_DEF_STMT (and_tmp_name) = and_stmt;
append_to_statement_list_force (and_stmt, cond_expr_stmt_list);
/* Make and_tmp the left operand of the conditional test against zero.
if and_tmp has a non-zero bit then some address is unaligned. */
ptrsize_zero = build_int_cst (int_ptrsize_type, 0);
return build2 (EQ_EXPR, boolean_type_node,
and_tmp_name, ptrsize_zero);
}
/* Function vect_transform_loop.
The analysis phase has determined that the loop is vectorizable.
......@@ -2720,6 +2842,30 @@ vect_transform_loop (loop_vec_info loop_vinfo,
if (vect_print_dump_info (REPORT_DETAILS))
fprintf (vect_dump, "=== vec_transform_loop ===");
/* If the loop has data references that may or may not be aligned then
two versions of the loop need to be generated, one which is vectorized
and one which isn't. A test is then generated to control which of the
loops is executed. The test checks for the alignment of all of the
data references that may or may not be aligned. */
if (VEC_length (tree, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)))
{
struct loop *nloop;
tree cond_expr;
tree cond_expr_stmt_list = NULL_TREE;
basic_block condition_bb;
block_stmt_iterator cond_exp_bsi;
cond_expr = vect_create_cond_for_align_checks (loop_vinfo,
&cond_expr_stmt_list);
initialize_original_copy_tables ();
nloop = loop_version (loops, loop, cond_expr, &condition_bb, true);
free_original_copy_tables();
update_ssa (TODO_update_ssa);
cond_exp_bsi = bsi_last (condition_bb);
bsi_insert_before (&cond_exp_bsi, cond_expr_stmt_list, BSI_SAME_STMT);
}
/* CHECKME: we wouldn't need this if we calles update_ssa once
for all loops. */
bitmap_zero (vect_vnames_to_rename);
......
......@@ -137,6 +137,7 @@ Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
#include "cfglayout.h"
#include "expr.h"
#include "optabs.h"
#include "params.h"
#include "toplev.h"
#include "tree-chrec.h"
#include "tree-data-ref.h"
......@@ -1420,6 +1421,8 @@ new_loop_vec_info (struct loop *loop)
VARRAY_GENERIC_PTR_INIT (LOOP_VINFO_DATAREFS (res), 20, "loop_datarefs");
VARRAY_GENERIC_PTR_INIT (LOOP_VINFO_DDRS (res), 20, "loop_ddrs");
LOOP_VINFO_UNALIGNED_DR (res) = NULL;
LOOP_VINFO_MAY_MISALIGN_STMTS (res)
= VEC_alloc (tree, heap, PARAM_VALUE (PARAM_VECT_MAX_VERSION_CHECKS));
return res;
}
......@@ -1480,6 +1483,7 @@ destroy_loop_vec_info (loop_vec_info loop_vinfo)
free (LOOP_VINFO_BBS (loop_vinfo));
varray_clear (LOOP_VINFO_DATAREFS (loop_vinfo));
varray_clear (LOOP_VINFO_DDRS (loop_vinfo));
VEC_free (tree, heap, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo));
free (loop_vinfo);
}
......@@ -2049,7 +2053,7 @@ vectorize_loops (struct loops *loops)
if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
continue;
vect_transform_loop (loop_vinfo, loops);
vect_transform_loop (loop_vinfo, loops);
num_vectorized_loops++;
}
......
......@@ -117,28 +117,37 @@ typedef struct _loop_vec_info {
unaligned_dr. */
int peeling_for_alignment;
/* The mask used to check the alignment of pointers or arrays. */
int ptr_mask;
/* All data references in the loop. */
varray_type datarefs;
/* All data dependences in the loop. */
varray_type ddrs;
/* Statements in the loop that have data references that are candidates for a
runtime (loop versioning) misalignment check. */
VEC(tree,heap) *may_misalign_stmts;
/* The loop location in the source. */
LOC loop_line_number;
} *loop_vec_info;
/* Access Functions. */
/* Access Functions. */
#define LOOP_VINFO_LOOP(L) (L)->loop
#define LOOP_VINFO_BBS(L) (L)->bbs
#define LOOP_VINFO_EXIT_COND(L) (L)->exit_cond
#define LOOP_VINFO_NITERS(L) (L)->num_iters
#define LOOP_VINFO_VECTORIZABLE_P(L) (L)->vectorizable
#define LOOP_VINFO_VECT_FACTOR(L) (L)->vectorization_factor
#define LOOP_VINFO_PTR_MASK(L) (L)->ptr_mask
#define LOOP_VINFO_DATAREFS(L) (L)->datarefs
#define LOOP_VINFO_DDRS(L) (L)->ddrs
#define LOOP_VINFO_INT_NITERS(L) (TREE_INT_CST_LOW ((L)->num_iters))
#define LOOP_PEELING_FOR_ALIGNMENT(L) (L)->peeling_for_alignment
#define LOOP_VINFO_UNALIGNED_DR(L) (L)->unaligned_dr
#define LOOP_VINFO_MAY_MISALIGN_STMTS(L) (L)->may_misalign_stmts
#define LOOP_VINFO_LOC(L) (L)->loop_line_number
#define LOOP_VINFO_NITERS_KNOWN_P(L) \
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment