https://gcc.gnu.org/g:d90d9ba058fb54e1138efab273e06ec9cc6014d0
commit r16-323-gd90d9ba058fb54e1138efab273e06ec9cc6014d0
Author: Richard Biener <rguent...@suse.de>
Date: Wed Apr 30 14:57:03 2025 +0200
Remove non-SLP path from vectorizable_conversion
This removes the non-SLP paths from vectorizable_conversion and
in the process eliminates uses of 'ncopies' and 'STMT_VINFO_VECTYPE'
from the function.
* tree-vect-stmts.cc (vectorizable_conversion): Remove non-SLP
paths.
Diff:
---
gcc/tree-vect-stmts.cc | 88 +++++++++++---------------------------------------
1 file changed, 19 insertions(+), 69 deletions(-)
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 38612a166191..42b6059520ac 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -5528,7 +5528,6 @@ vectorizable_conversion (vec_info *vinfo,
tree vec_dest, cvt_op = NULL_TREE;
tree scalar_dest;
tree op0, op1 = NULL_TREE;
- loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
tree_code tc1;
code_helper code, code1, code2;
code_helper codecvt1 = ERROR_MARK, codecvt2 = ERROR_MARK;
@@ -5538,7 +5537,7 @@ vectorizable_conversion (vec_info *vinfo,
poly_uint64 nunits_in;
poly_uint64 nunits_out;
tree vectype_out, vectype_in;
- int ncopies, i;
+ int i;
tree lhs_type, rhs_type;
/* For conversions between floating point and integer, there're 2 NARROW
cases. NARROW_SRC is for FLOAT_EXPR, means
@@ -5605,7 +5604,7 @@ vectorizable_conversion (vec_info *vinfo,
/* Check types of lhs and rhs. */
scalar_dest = gimple_get_lhs (stmt);
lhs_type = TREE_TYPE (scalar_dest);
- vectype_out = STMT_VINFO_VECTYPE (stmt_info);
+ vectype_out = SLP_TREE_VECTYPE (slp_node);
/* Check the operands of the operation. */
slp_tree slp_op0, slp_op1 = NULL;
@@ -5703,20 +5702,6 @@ vectorizable_conversion (vec_info *vinfo,
modifier = WIDEN;
}
- /* Multiple types in SLP are handled by creating the appropriate number of
- vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in
- case of SLP. */
- if (slp_node)
- ncopies = 1;
- else if (modifier == NARROW_DST)
- ncopies = vect_get_num_copies (loop_vinfo, vectype_out);
- else
- ncopies = vect_get_num_copies (loop_vinfo, vectype_in);
-
- /* Sanity check: make sure that at least one copy of the vectorized stmt
- needs to be generated. */
- gcc_assert (ncopies >= 1);
-
bool found_mode = false;
scalar_mode lhs_mode = SCALAR_TYPE_MODE (lhs_type);
scalar_mode rhs_mode = SCALAR_TYPE_MODE (rhs_type);
@@ -5871,16 +5856,11 @@ vectorizable_conversion (vec_info *vinfo,
else if (code == FLOAT_EXPR)
{
wide_int op_min_value, op_max_value;
- if (slp_node)
- {
- tree def;
- /* ??? Merge ranges in case of more than one lane. */
- if (SLP_TREE_LANES (slp_op0) != 1
- || !(def = vect_get_slp_scalar_def (slp_op0, 0))
- || !vect_get_range_info (def, &op_min_value, &op_max_value))
- goto unsupported;
- }
- else if (!vect_get_range_info (op0, &op_min_value, &op_max_value))
+ tree def;
+ /* ??? Merge ranges in case of more than one lane. */
+ if (SLP_TREE_LANES (slp_op0) != 1
+ || !(def = vect_get_slp_scalar_def (slp_op0, 0))
+ || !vect_get_range_info (def, &op_min_value, &op_max_value))
goto unsupported;
cvt_type
@@ -5916,9 +5896,8 @@ vectorizable_conversion (vec_info *vinfo,
if (!vec_stmt) /* transformation not required. */
{
- if (slp_node
- && (!vect_maybe_update_slp_op_vectype (slp_op0, vectype_in)
- || !vect_maybe_update_slp_op_vectype (slp_op1, vectype_in)))
+ if (!vect_maybe_update_slp_op_vectype (slp_op0, vectype_in)
+ || !vect_maybe_update_slp_op_vectype (slp_op1, vectype_in))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
@@ -5929,16 +5908,14 @@ vectorizable_conversion (vec_info *vinfo,
if (modifier == NONE)
{
STMT_VINFO_TYPE (stmt_info) = type_conversion_vec_info_type;
- vect_model_simple_cost (vinfo, stmt_info,
- ncopies * (1 + multi_step_cvt),
+ vect_model_simple_cost (vinfo, stmt_info, (1 + multi_step_cvt),
dt, ndts, slp_node, cost_vec);
}
else if (modifier == NARROW_SRC || modifier == NARROW_DST)
{
STMT_VINFO_TYPE (stmt_info) = type_demotion_vec_info_type;
/* The final packing step produces one vector result per copy. */
- unsigned int nvectors
- = (slp_node ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) : ncopies);
+ unsigned int nvectors = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
vect_model_promotion_demotion_cost (stmt_info, dt, nvectors,
multi_step_cvt, cost_vec,
widen_arith);
@@ -5950,9 +5927,7 @@ vectorizable_conversion (vec_info *vinfo,
per copy. MULTI_STEP_CVT is 0 for a single conversion,
so >> MULTI_STEP_CVT divides by 2^(number of steps - 1). */
unsigned int nvectors
- = (slp_node
- ? SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) >> multi_step_cvt
- : ncopies * 2);
+ = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node) >> multi_step_cvt;
vect_model_promotion_demotion_cost (stmt_info, dt, nvectors,
multi_step_cvt, cost_vec,
widen_arith);
@@ -5963,8 +5938,7 @@ vectorizable_conversion (vec_info *vinfo,
/* Transform. */
if (dump_enabled_p ())
- dump_printf_loc (MSG_NOTE, vect_location,
- "transform conversion. ncopies = %d.\n", ncopies);
+ dump_printf_loc (MSG_NOTE, vect_location, "transform conversion.\n");
if (op_type == binary_op)
{
@@ -6003,23 +5977,10 @@ vectorizable_conversion (vec_info *vinfo,
widen_or_narrow_float_p
? vectype_out : cvt_type);
- int ninputs = 1;
- if (!slp_node)
- {
- if (modifier == WIDEN)
- ;
- else if (modifier == NARROW_SRC || modifier == NARROW_DST)
- {
- if (multi_step_cvt)
- ninputs = vect_pow2 (multi_step_cvt);
- ninputs *= 2;
- }
- }
-
switch (modifier)
{
case NONE:
- vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies,
+ vect_get_vec_defs (vinfo, stmt_info, slp_node, 1,
op0, vectype_in, &vec_oprnds0);
/* vec_dest is intermediate type operand when multi_step_cvt. */
if (multi_step_cvt)
@@ -6046,10 +6007,7 @@ vectorizable_conversion (vec_info *vinfo,
gimple_set_lhs (new_stmt, new_temp);
vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi);
- if (slp_node)
- slp_node->push_vec_def (new_stmt);
- else
- STMT_VINFO_VEC_STMTS (stmt_info).safe_push (new_stmt);
+ slp_node->push_vec_def (new_stmt);
}
break;
@@ -6058,7 +6016,7 @@ vectorizable_conversion (vec_info *vinfo,
of elements that we can fit in a vectype (nunits), we have to
generate more than one vector stmt - i.e - we need to "unroll"
the vector stmt by a factor VF/nunits. */
- vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies * ninputs,
+ vect_get_vec_defs (vinfo, stmt_info, slp_node, 1,
op0, vectype_in, &vec_oprnds0,
code == WIDEN_LSHIFT_EXPR ? NULL_TREE : op1,
vectype_in, &vec_oprnds1);
@@ -6102,10 +6060,7 @@ vectorizable_conversion (vec_info *vinfo,
else
new_stmt = SSA_NAME_DEF_STMT (vop0);
- if (slp_node)
- slp_node->push_vec_def (new_stmt);
- else
- STMT_VINFO_VEC_STMTS (stmt_info).safe_push (new_stmt);
+ slp_node->push_vec_def (new_stmt);
}
break;
@@ -6115,7 +6070,7 @@ vectorizable_conversion (vec_info *vinfo,
of elements that we can fit in a vectype (nunits), we have to
generate more than one vector stmt - i.e - we need to "unroll"
the vector stmt by a factor VF/nunits. */
- vect_get_vec_defs (vinfo, stmt_info, slp_node, ncopies * ninputs,
+ vect_get_vec_defs (vinfo, stmt_info, slp_node, 1,
op0, vectype_in, &vec_oprnds0);
/* Arguments are ready. Create the new vector stmts. */
if (cvt_type && modifier == NARROW_DST)
@@ -6148,16 +6103,11 @@ vectorizable_conversion (vec_info *vinfo,
/* This is the last step of the conversion sequence. Store the
vectors in SLP_NODE or in vector info of the scalar statement
(or in STMT_VINFO_RELATED_STMT chain). */
- if (slp_node)
- slp_node->push_vec_def (new_stmt);
- else
- STMT_VINFO_VEC_STMTS (stmt_info).safe_push (new_stmt);
+ slp_node->push_vec_def (new_stmt);
}
}
break;
}
- if (!slp_node)
- *vec_stmt = STMT_VINFO_VEC_STMTS (stmt_info)[0];
vec_oprnds0.release ();
vec_oprnds1.release ();
