^=== analyzing loop ===
=== analyze_loop_nest ===
=== vect_analyze_loop_form ===
=== get_loop_niters ===
symbolic number of iterations is
(unsigned int) n_8(D)
=== vect_analyze_data_refs ===
got vectype for stmt:
_4 = *_3;vector(4) int
=== vect_analyze_scalar_cycles ===
Analyze phi:
i_17 = PHI <0(5), i_10(6)>
Access function of PHI:
{0, +, 1}_1
step:
1, init:
0
Detected induction
Analyze phi:
sum_20 = PHI <0(5), _12(6)>
Access function of PHI:
{0, +, _4}_1
step:
_4, init:
0
step unknown
Analyze phi:
sum_20 = PHI <0(5), _12(6)>
detected reduction:
_12 = _4 + sum_20;
Detected reduction
=== vect_pattern_recog ===
vect_is_simple_use: operand
_1
def_stmt:
_1 = (long unsigned int) i_17;
type of def: internal
vect_is_simple_use: operand
i_17
def_stmt:
i_17 = PHI <0(5), i_10(6)>
type of def: induction
vect_is_simple_use: operand
4
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
=== vect_analyze_data_ref_accesses ===
=== vect_mark_stmts_to_be_vectorized ===
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
sum_20
def_stmt:
sum_20 = PHI <0(5), _12(6)>
type of def: reduction
vect_is_simple_use: operand
0
vect_is_simple_use: operand
_12
def_stmt:
_12 = _4 + sum_20;
type of def: reduction
=== vect_analyze_data_ref_dependences ===
=== vect_determine_vectorization_factor ===
==> examining phi:
i_17 = PHI <0(5), i_10(6)>
==> examining phi:
sum_20 = PHI <0(5), _12(6)>
get vectype for scalar type:
int
vectype:
vector(4) int
nunits = FIXME: poly_int
==> examining statement:
_1 = (long unsigned int) i_17;
skip
==> examining statement:
_2 = _1 * 4;
skip
==> examining statement:
_3 = arr_9(D) + _2;
skip
==> examining statement:
_4 = *_3;
get vectype for scalar type:
int
vectype:
vector(4) int
nunits = FIXME: poly_int
==> examining statement:
_12 = _4 + sum_20;
get vectype for scalar type:
int
vectype:
vector(4) int
get vectype for scalar type:
int
vectype:
vector(4) int
nunits = FIXME: poly_int
==> examining statement:
i_10 = i_17 + 1;
skip
==> examining statement:
if (n_8(D) > i_10)
skip
vectorization factor = FIXME: poly_int
=== vect_analyze_slp ===
=== vect_make_slp_decision ===
=== vect_analyze_data_refs_alignment ===
recording new base alignment for
arr_9(D)
alignment: 4
misalignment: 0
based on:
_4 = *_3;
vect_compute_data_ref_alignment:
can't force alignment of ref:
*_3
=== vect_prune_runtime_alias_test_list ===
=== vect_enhance_data_refs_alignment ===
Unknown misalignment, naturally aligned
Analyze phi:
i_17 = PHI <0(5), i_10(6)>
Analyze phi:
sum_20 = PHI <0(5), _12(6)>
reduc or virtual phi. skip
vect_get_data_access_cost: inside_cost = 12, outside_cost = 0.
cost model: epilogue peel iters set to vf/2 because loop iterations are unknown
vect_get_data_access_cost: inside_cost = 12, outside_cost = 0.
cost model: epilogue peel iters set to vf/2 because loop iterations are unknown
Vectorizing an unaligned access
=== vect_analyze_loop_operations ===
examining phi:
i_17 = PHI <0(5), i_10(6)>
examining phi:
sum_20 = PHI <0(5), _12(6)>
vect_is_simple_use: operand
*_3
not ssa-name
use not simple
vect_is_simple_use: operand
*_3
not ssa-name
use not simple
can't use a fully-masked loop because the target doesn't have the appropriate masked load or store.
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
sum_20
def_stmt:
sum_20 = PHI <0(5), _12(6)>
type of def: reduction
reduc op not supported by target
not using a fully-masked loop
cost model: epilogue peel iters set to vf/2 because loop iterations are unknown
Cost model analysis:
Vector inside of loop cost: 16
Vector prologue cost: 36
Vector epilogue cost: 52
Scalar iteration cost: 16
Scalar outside cost: 32
Vector outside cost: 88
prologue iterations: 0
epilogue iterations: 2
Runtime profitability threshold = 5
Static estimate profitability threshold = 9
epilog loop required
Analyze phi:
i_17 = PHI <0(5), i_10(6)>
Analyze phi:
sum_20 = PHI <0(5), _12(6)>
reduc or virtual phi. skip
loop vectorized
=== vec_transform_loop ===
Profitability threshold is 5 loop iterations
Analyze phi:
i_17 = PHI <i_10(6), 0(9)>
Analyze phi:
sum_20 = PHI <_12(6), 0(9)>
reduc or virtual phi. skip
vect_update_ivs_after_vectorizer: phi:
i_17 = PHI <i_10(6), 0(9)>
vect_update_ivs_after_vectorizer: phi:
sum_20 = PHI <_12(6), 0(9)>
reduc or virtual phi. skip.
------>vectorizing phi:
i_17 = PHI <i_10(6), 0(16)>
------>vectorizing phi:
sum_20 = PHI <_12(6), 0(16)>
transform phi
------>vectorizing phi:
vect__12.4_33 = PHI <(6), (16)>
------>vectorizing statement:
_1 = (long unsigned int) i_17;
------>vectorizing statement:
_2 = _1 * 4;
------>vectorizing statement:
_3 = arr_9(D) + _2;
------>vectorizing statement:
_4 = *_3;
transform statement
create vector_type-pointer variable to type:
vector(4) int vectorizing a pointer ref:
*arr_9(D)
created
arr_9(D)
------>vectorizing statement:
_12 = _4 + sum_20;
transform statement
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
vect_is_simple_use: operand
sum_20
def_stmt:
sum_20 = PHI <_12(6), 0(16)>
type of def: reduction
reduc op not supported by target
transform reduction
vect_get_vec_def_for_operand:
_4
vect_is_simple_use: operand
_4
def_stmt:
_4 = *_3;
type of def: internal
def_stmt =
_4 = *_3;
vect_get_vec_def_for_operand:
sum_20
vect_is_simple_use: operand
sum_20
def_stmt:
sum_20 = PHI <_12(6), 0(16)>
type of def: reduction
def_stmt =
sum_20 = PHI <_12(6), 0(16)>
vect_is_simple_use: operand
0
transform reduction: created def-use cycle:
vect__12.4_33 = PHI <vect__12.8_37(6), { 0, 0, 0, 0 }(16)>vect__12.8_37 = vect__4.7_36 + vect__12.4_33;
Reduce using vector shifts
extract scalar result
------>vectorizing statement:
i_10 = i_17 + 1;
------>vectorizing statement:
vectp_arr.5_35 = vectp_arr.5_34 + 16;
------>vectorizing statement:
if (n_8(D) > i_10)
New loop exit condition:
if (ivtmp_46 < bnd.1_29)
LOOP VECTORIZED