This patch enables the compiler to use inbranch simdclones when
generating masked loops in autovectorization.
gcc/ChangeLog:
* omp-simd-clone.cc (simd_clone_adjust_argument_types): Make function
compatible with mask parameters in clone.
* tree-vect-stmts.cc (vect_convert): New helper function.
(vect_build_all_ones_mask): Allow vector boolean typed masks.
(vectorizable_simd_clone_call): Enable the use of masked clones in
fully masked loops.
diff --git a/gcc/omp-simd-clone.cc b/gcc/omp-simd-clone.cc
index
a42643400ddcf10961633448b49d4caafb999f12..ef0b9b48c7212900023bc0eaebca5e1f9389db77
100644
--- a/gcc/omp-simd-clone.cc
+++ b/gcc/omp-simd-clone.cc
@@ -807,8 +807,14 @@ simd_clone_adjust_argument_types (struct cgraph_node *node)
{
ipa_adjusted_param adj;
memset (&adj, 0, sizeof (adj));
- tree parm = args[i];
- tree parm_type = node->definition ? TREE_TYPE (parm) : parm;
+ tree parm = NULL_TREE;
+ tree parm_type = NULL_TREE;
+ if(i < args.length())
+ {
+ parm = args[i];
+ parm_type = node->definition ? TREE_TYPE (parm) : parm;
+ }
+
adj.base_index = i;
adj.prev_clone_index = i;
@@ -1547,7 +1553,7 @@ simd_clone_adjust (struct cgraph_node *node)
mask = gimple_assign_lhs (g);
g = gimple_build_assign (make_ssa_name (TREE_TYPE (mask)),
BIT_AND_EXPR, mask,
- build_int_cst (TREE_TYPE (mask), 1));
+ build_one_cst (TREE_TYPE (mask)));
gsi_insert_after (&gsi, g, GSI_CONTINUE_LINKING);
mask = gimple_assign_lhs (g);
}
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index
664c3b5f7ca48fdb49383fb8a97f407465574479..7217f36a250d549b955c874d7c7644d94982b0b5
100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -1723,6 +1723,20 @@ check_load_store_for_partial_vectors (loop_vec_info
loop_vinfo, tree vectype,
}
}
+/* Return SSA name of the result of the conversion of OPERAND into type TYPE.
+ The conversion statement is inserted at GSI. */
+
+static tree
+vect_convert (vec_info *vinfo, stmt_vec_info stmt_info, tree type, tree
operand,
+ gimple_stmt_iterator *gsi)
+{
+ operand = build1 (VIEW_CONVERT_EXPR, type, operand);
+ gassign *new_stmt = gimple_build_assign (make_ssa_name (type),
+ operand);
+ vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi);
+ return gimple_get_lhs (new_stmt);
+}
+
/* Return the mask input to a masked load or store. VEC_MASK is the vectorized
form of the scalar mask condition and LOOP_MASK, if nonnull, is the mask
that needs to be applied to all loads and stores in a vectorized loop.
@@ -2666,7 +2680,8 @@ vect_build_all_ones_mask (vec_info *vinfo,
{
if (TREE_CODE (masktype) == INTEGER_TYPE)
return build_int_cst (masktype, -1);
- else if (TREE_CODE (TREE_TYPE (masktype)) == INTEGER_TYPE)
+ else if (VECTOR_BOOLEAN_TYPE_P (masktype)
+ || TREE_CODE (TREE_TYPE (masktype)) == INTEGER_TYPE)
{
tree mask = build_int_cst (TREE_TYPE (masktype), -1);
mask = build_vector_from_val (masktype, mask);
@@ -4018,7 +4033,7 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
size_t i, nargs;
tree lhs, rtype, ratype;
vec<constructor_elt, va_gc> *ret_ctor_elts = NULL;
- int arg_offset = 0;
+ int masked_call_offset = 0;
/* Is STMT a vectorizable call? */
gcall *stmt = dyn_cast <gcall *> (stmt_info->stmt);
@@ -4033,7 +4048,7 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
gcc_checking_assert (TREE_CODE (fndecl) == ADDR_EXPR);
fndecl = TREE_OPERAND (fndecl, 0);
gcc_checking_assert (TREE_CODE (fndecl) == FUNCTION_DECL);
- arg_offset = 1;
+ masked_call_offset = 1;
}
if (fndecl == NULL_TREE)
return false;
@@ -4065,7 +4080,7 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
return false;
/* Process function arguments. */
- nargs = gimple_call_num_args (stmt) - arg_offset;
+ nargs = gimple_call_num_args (stmt) - masked_call_offset;
/* Bail out if the function has zero arguments. */
if (nargs == 0)
@@ -4083,7 +4098,7 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
thisarginfo.op = NULL_TREE;
thisarginfo.simd_lane_linear = false;
- op = gimple_call_arg (stmt, i + arg_offset);
+ op = gimple_call_arg (stmt, i + masked_call_offset);
if (!vect_is_simple_use (op, vinfo, &thisarginfo.dt,
&thisarginfo.vectype)
|| thisarginfo.dt == vect_uninitialized_def)
@@ -4161,14 +4176,6 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
}
poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
- if (!vf.is_constant ())
- {
- if (dump_enabled_p ())
- dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
- "not considering SIMD clones; not yet supported"
- " for variable-width vectors.\n");
- return false;
- }
unsigned int badness = 0;
struct cgraph_node *bestn = NULL;
@@ -4181,7 +4188,8 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
unsigned int this_badness = 0;
unsigned int num_calls;
if (!constant_multiple_p (vf, n->simdclone->simdlen, &num_calls)
- || n->simdclone->nargs != nargs)
+ || (!n->simdclone->inbranch && (masked_call_offset > 0))
+ || nargs != n->simdclone->nargs)
continue;
if (num_calls != 1)
this_badness += exact_log2 (num_calls) * 4096;
@@ -4198,7 +4206,8 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
case SIMD_CLONE_ARG_TYPE_VECTOR:
if (!useless_type_conversion_p
(n->simdclone->args[i].orig_type,
- TREE_TYPE (gimple_call_arg (stmt, i + arg_offset))))
+ TREE_TYPE (gimple_call_arg (stmt,
+ i + masked_call_offset))))
i = -1;
else if (arginfo[i].dt == vect_constant_def
|| arginfo[i].dt == vect_external_def
@@ -4243,6 +4252,17 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
}
if (i == (size_t) -1)
continue;
+ if (masked_call_offset == 0
+ && n->simdclone->inbranch
+ && n->simdclone->nargs > nargs)
+ {
+ gcc_assert (n->simdclone->args[n->simdclone->nargs - 1].arg_type ==
+ SIMD_CLONE_ARG_TYPE_MASK);
+ /* Penalize using a masked SIMD clone in a non-masked loop, that is
+ not in a branch, as we'd have to construct an all-true mask. */
+ if (!loop_vinfo || !LOOP_VINFO_FULLY_MASKED_P (loop_vinfo))
+ this_badness += 64;
+ }
if (bestn == NULL || this_badness < badness)
{
bestn = n;
@@ -4259,7 +4279,8 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
|| arginfo[i].dt == vect_external_def)
&& bestn->simdclone->args[i].arg_type == SIMD_CLONE_ARG_TYPE_VECTOR)
{
- tree arg_type = TREE_TYPE (gimple_call_arg (stmt, i + arg_offset));
+ tree arg_type = TREE_TYPE (gimple_call_arg (stmt,
+ i + masked_call_offset));
arginfo[i].vectype = get_vectype_for_scalar_type (vinfo, arg_type,
slp_node);
if (arginfo[i].vectype == NULL
@@ -4331,24 +4352,38 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
&& TREE_CODE (TREE_TYPE (TREE_TYPE (bestn->decl))) == ARRAY_TYPE)
vinfo->any_known_not_updated_vssa = true;
STMT_VINFO_SIMD_CLONE_INFO (stmt_info).safe_push (bestn->decl);
- for (i = 0; i < nargs; i++)
- if ((bestn->simdclone->args[i].arg_type
- == SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP)
- || (bestn->simdclone->args[i].arg_type
- == SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP))
- {
- STMT_VINFO_SIMD_CLONE_INFO (stmt_info).safe_grow_cleared (i * 3
- + 1,
- true);
- STMT_VINFO_SIMD_CLONE_INFO (stmt_info).safe_push (arginfo[i].op);
- tree lst = POINTER_TYPE_P (TREE_TYPE (arginfo[i].op))
- ? size_type_node : TREE_TYPE (arginfo[i].op);
- tree ls = build_int_cst (lst, arginfo[i].linear_step);
- STMT_VINFO_SIMD_CLONE_INFO (stmt_info).safe_push (ls);
- tree sll = arginfo[i].simd_lane_linear
- ? boolean_true_node : boolean_false_node;
- STMT_VINFO_SIMD_CLONE_INFO (stmt_info).safe_push (sll);
- }
+
+ for (i = 0; i < bestn->simdclone->nargs; i++)
+ {
+ switch (bestn->simdclone->args[i].arg_type)
+ {
+ default:
+ continue;
+ case SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP:
+ case SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP:
+ {
+ auto &clone_info = STMT_VINFO_SIMD_CLONE_INFO (stmt_info);
+ clone_info.safe_grow_cleared (i * 3 + 1, true);
+ clone_info.safe_push (arginfo[i].op);
+ tree lst = POINTER_TYPE_P (TREE_TYPE (arginfo[i].op))
+ ? size_type_node : TREE_TYPE (arginfo[i].op);
+ tree ls = build_int_cst (lst, arginfo[i].linear_step);
+ clone_info.safe_push (ls);
+ tree sll = arginfo[i].simd_lane_linear
+ ? boolean_true_node : boolean_false_node;
+ clone_info.safe_push (sll);
+ }
+ break;
+ case SIMD_CLONE_ARG_TYPE_MASK:
+ if (loop_vinfo
+ && LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P (loop_vinfo))
+ vect_record_loop_mask (loop_vinfo,
+ &LOOP_VINFO_MASKS (loop_vinfo),
+ ncopies, vectype, op);
+
+ break;
+ }
+ }
if (!bestn->simdclone->inbranch)
{
@@ -4394,6 +4429,8 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
vec_oprnds_i.safe_grow_cleared (nargs, true);
for (j = 0; j < ncopies; ++j)
{
+ poly_uint64 callee_nelements;
+ poly_uint64 caller_nelements;
/* Build argument list for the vectorized call. */
if (j == 0)
vargs.create (nargs);
@@ -4404,8 +4441,7 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
{
unsigned int k, l, m, o;
tree atype;
- poly_uint64 callee_nelements, caller_nelements;
- op = gimple_call_arg (stmt, i + arg_offset);
+ op = gimple_call_arg (stmt, i + masked_call_offset);
switch (bestn->simdclone->args[i].arg_type)
{
case SIMD_CLONE_ARG_TYPE_VECTOR:
@@ -4482,16 +4518,9 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
if (k == 1)
if (!useless_type_conversion_p (TREE_TYPE (vec_oprnd0),
atype))
- {
- vec_oprnd0
- = build1 (VIEW_CONVERT_EXPR, atype, vec_oprnd0);
- gassign *new_stmt
- = gimple_build_assign (make_ssa_name (atype),
- vec_oprnd0);
- vect_finish_stmt_generation (vinfo, stmt_info,
- new_stmt, gsi);
- vargs.safe_push (gimple_assign_lhs (new_stmt));
- }
+ vargs.safe_push (vect_convert (vinfo, stmt_info,
+ atype, vec_oprnd0,
+ gsi));
else
vargs.safe_push (vec_oprnd0);
else
@@ -4544,6 +4573,24 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
vec_oprnds_i[i] = 0;
}
vec_oprnd0 = vec_oprnds[i][vec_oprnds_i[i]++];
+ if (loop_vinfo
+ && LOOP_VINFO_FULLY_MASKED_P (loop_vinfo))
+ {
+ vec_loop_masks *loop_masks
+ = &LOOP_VINFO_MASKS (loop_vinfo);
+ tree loop_mask
+ = vect_get_loop_mask (loop_vinfo, gsi,
+ loop_masks, ncopies,
+ vectype, j);
+ vec_oprnd0
+ = prepare_vec_mask (loop_vinfo,
+ TREE_TYPE (loop_mask),
+ loop_mask, vec_oprnd0,
+ gsi);
+ loop_vinfo->vec_cond_masked_set.add ({ vec_oprnd0,
+ loop_mask
});
+
+ }
vec_oprnd0
= build3 (VEC_COND_EXPR, atype, vec_oprnd0,
build_vector_from_val (atype, one),
@@ -4641,6 +4688,64 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
}
}
+ if (masked_call_offset == 0
+ && bestn->simdclone->inbranch
+ && bestn->simdclone->nargs > nargs)
+ {
+ unsigned long m, o;
+ size_t mask_i = bestn->simdclone->nargs - 1;
+ tree mask;
+ gcc_assert (bestn->simdclone->args[mask_i].arg_type ==
+ SIMD_CLONE_ARG_TYPE_MASK);
+
+ tree masktype = bestn->simdclone->args[mask_i].vector_type;
+ callee_nelements = TYPE_VECTOR_SUBPARTS (masktype);
+ o = vector_unroll_factor (nunits, callee_nelements);
+ for (m = j * o; m < (j + 1) * o; m++)
+ {
+ if (loop_vinfo && LOOP_VINFO_FULLY_MASKED_P (loop_vinfo))
+ {
+ vec_loop_masks *loop_masks = &LOOP_VINFO_MASKS (loop_vinfo);
+ mask = vect_get_loop_mask (loop_vinfo, gsi, loop_masks,
+ ncopies, vectype, j);
+ }
+ else
+ mask = vect_build_all_ones_mask (vinfo, stmt_info, masktype);
+
+ if (!useless_type_conversion_p (TREE_TYPE (mask), masktype))
+ {
+ gassign *new_stmt;
+ if (bestn->simdclone->mask_mode != VOIDmode)
+ {
+ /* This means we are dealing with integer mask modes.
+ First convert to an integer type with the same size as
+ the current vector type. */
+ unsigned HOST_WIDE_INT intermediate_size
+ = tree_to_uhwi (TYPE_SIZE (TREE_TYPE (mask)));
+ tree mid_int_type =
+ build_nonstandard_integer_type (intermediate_size, 1);
+ mask = build1 (VIEW_CONVERT_EXPR, mid_int_type, mask);
+ new_stmt
+ = gimple_build_assign (make_ssa_name (mid_int_type),
+ mask);
+ gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
+ /* Then zero-extend to the mask mode. */
+ mask = fold_build1 (NOP_EXPR, masktype,
+ gimple_get_lhs (new_stmt));
+ }
+ else
+ mask = build1 (VIEW_CONVERT_EXPR, masktype, mask);
+
+ new_stmt = gimple_build_assign (make_ssa_name (masktype),
+ mask);
+ vect_finish_stmt_generation (vinfo, stmt_info,
+ new_stmt, gsi);
+ mask = gimple_assign_lhs (new_stmt);
+ }
+ vargs.safe_push (mask);
+ }
+ }
+
gcall *new_call = gimple_build_call_vec (fndecl, vargs);
if (vec_dest)
{
@@ -4659,13 +4764,13 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
if (vec_dest)
{
- if (!multiple_p (TYPE_VECTOR_SUBPARTS (vectype), nunits))
+ caller_nelements = TYPE_VECTOR_SUBPARTS (vectype);
+ if (!multiple_p (caller_nelements, nunits))
{
unsigned int k, l;
poly_uint64 prec = GET_MODE_BITSIZE (TYPE_MODE (vectype));
poly_uint64 bytes = GET_MODE_SIZE (TYPE_MODE (vectype));
- k = vector_unroll_factor (nunits,
- TYPE_VECTOR_SUBPARTS (vectype));
+ k = vector_unroll_factor (nunits, caller_nelements);
gcc_assert ((k & (k - 1)) == 0);
for (l = 0; l < k; l++)
{
@@ -4691,11 +4796,11 @@ vectorizable_simd_clone_call (vec_info *vinfo,
stmt_vec_info stmt_info,
vect_clobber_variable (vinfo, stmt_info, gsi, new_temp);
continue;
}
- else if (!multiple_p (nunits, TYPE_VECTOR_SUBPARTS (vectype)))
+ else if (!multiple_p (nunits, caller_nelements))
{
unsigned int k;
- if (!constant_multiple_p (TYPE_VECTOR_SUBPARTS (rtype),
- TYPE_VECTOR_SUBPARTS (vectype), &k))
+ if (!constant_multiple_p (caller_nelements,
+ TYPE_VECTOR_SUBPARTS (rtype), &k))
gcc_unreachable ();
gcc_assert ((k & (k - 1)) == 0);
if ((j & (k - 1)) == 0)