../../src/test.cc

^=== analyzing loop === Analyzing loop at ../../src/test.cc:4 === analyze_loop_nest === === vect_analyze_loop_form === === get_loop_niters === Symbolic number of iterations is (((((unsigned long) _10 - (unsigned long) _9) - 24) /[ex] 8) * 768614336404564651 & 2305843009213693951) + 1 === vect_analyze_data_refs === got vectype for stmt: _7 = MEM[(float * *)SR.8_23];vector(2) long unsigned int got vectype for stmt: _8 = MEM[(float * *)SR.8_23 + 8B];vector(2) long unsigned int === vect_analyze_scalar_cycles === Analyze phi: ret_17 = PHI <0(5), ret_6(6)> Access function of PHI: {0, +, _16}_1 step: _16, init: 0 step unknown. Analyze phi: SR.8_23 = PHI <_9(5), _12(6)> Access function of PHI: {_9, +, 24}_1 step: 24, init: _9 Detected induction. Analyze phi: ret_17 = PHI <0(5), ret_6(6)> detected reduction: ret_6 = _16 + ret_17; Detected reduction. === vect_pattern_recog === vect_is_simple_use: operand _16 def_stmt: _16 = (long unsigned int) _15; type of def: internal vect_is_simple_use: operand _16 def_stmt: _16 = (long unsigned int) _15; type of def: internal vect_is_simple_use: operand _15 def_stmt: _15 = _14 /[ex] 4; type of def: internal vect_is_simple_use: operand _16 def_stmt: _16 = (long unsigned int) _15; type of def: internal === vect_analyze_data_ref_accesses === Detected interleaving load MEM[(float * *)SR.8_23] and MEM[(float * *)SR.8_23 + 8B] Detected interleaving load of size 3 starting with _7 = MEM[(float * *)SR.8_23]; There is a gap of 1 elements after the group === vect_mark_stmts_to_be_vectorized === init: phi relevant? ret_17 = PHI <0(5), ret_6(6)> init: phi relevant? SR.8_23 = PHI <_9(5), _12(6)> init: stmt relevant? _7 = MEM[(float * *)SR.8_23]; init: stmt relevant? _8 = MEM[(float * *)SR.8_23 + 8B]; init: stmt relevant? _14 = _8 - _7; init: stmt relevant? _15 = _14 /[ex] 4; init: stmt relevant? _16 = (long unsigned int) _15; init: stmt relevant? ret_6 = _16 + ret_17; vec_stmt_relevant_p: used out of loop. vect_is_simple_use: operand _16 def_stmt: _16 = (long unsigned int) _15; type of def: internal vec_stmt_relevant_p: stmt live but not relevant. mark relevant 1, live 1: ret_6 = _16 + ret_17; init: stmt relevant? _12 = SR.8_23 + 24; init: stmt relevant? if (_10 != _12) worklist: examine stmt: ret_6 = _16 + ret_17; vect_is_simple_use: operand _16 def_stmt: _16 = (long unsigned int) _15; type of def: internal mark relevant 1, live 0: _16 = (long unsigned int) _15; vect_is_simple_use: operand ret_17 def_stmt: ret_17 = PHI <0(5), ret_6(6)> type of def: reduction mark relevant 1, live 0: ret_17 = PHI <0(5), ret_6(6)> worklist: examine stmt: ret_17 = PHI <0(5), ret_6(6)> vect_is_simple_use: operand 0 vect_is_simple_use: operand ret_6 def_stmt: ret_6 = _16 + ret_17; type of def: reduction reduc-stmt defining reduc-phi in the same nest. worklist: examine stmt: _16 = (long unsigned int) _15; vect_is_simple_use: operand _15 def_stmt: _15 = _14 /[ex] 4; type of def: internal mark relevant 1, live 0: _15 = _14 /[ex] 4; worklist: examine stmt: _15 = _14 /[ex] 4; vect_is_simple_use: operand _14 def_stmt: _14 = _8 - _7; type of def: internal mark relevant 1, live 0: _14 = _8 - _7; worklist: examine stmt: _14 = _8 - _7; vect_is_simple_use: operand _8 def_stmt: _8 = MEM[(float * *)SR.8_23 + 8B]; type of def: internal mark relevant 1, live 0: _8 = MEM[(float * *)SR.8_23 + 8B]; vect_is_simple_use: operand _7 def_stmt: _7 = MEM[(float * *)SR.8_23]; type of def: internal mark relevant 1, live 0: _7 = MEM[(float * *)SR.8_23]; worklist: examine stmt: _7 = MEM[(float * *)SR.8_23]; worklist: examine stmt: _8 = MEM[(float * *)SR.8_23 + 8B]; === vect_analyze_data_ref_dependences === === vect_determine_vectorization_factor === ==> examining phi: ret_17 = PHI <0(5), ret_6(6)> get vectype for scalar type: size_t vectype: vector(2) long unsigned int nunits = 2 ==> examining phi: SR.8_23 = PHI <_9(5), _12(6)> ==> examining statement: _7 = MEM[(float * *)SR.8_23]; get vectype for scalar type: float * vectype: vector(2) long unsigned int nunits = 2 ==> examining statement: _8 = MEM[(float * *)SR.8_23 + 8B]; get vectype for scalar type: float * vectype: vector(2) long unsigned int nunits = 2 ==> examining statement: _14 = _8 - _7; get vectype for scalar type: long int vectype: vector(2) long int get vectype for scalar type: long int vectype: vector(2) long int nunits = 2 ==> examining statement: _15 = _14 /[ex] 4; get vectype for scalar type: long int vectype: vector(2) long int get vectype for scalar type: long int vectype: vector(2) long int nunits = 2 ==> examining statement: _16 = (long unsigned int) _15; get vectype for scalar type: long unsigned int vectype: vector(2) long unsigned int get vectype for scalar type: long unsigned int vectype: vector(2) long unsigned int nunits = 2 ==> examining statement: ret_6 = _16 + ret_17; get vectype for scalar type: size_t vectype: vector(2) long unsigned int get vectype for scalar type: size_t vectype: vector(2) long unsigned int nunits = 2 ==> examining statement: _12 = SR.8_23 + 24; skip. ==> examining statement: if (_10 != _12) skip. vectorization factor = 2 === vect_analyze_slp === === vect_make_slp_decision === === vect_analyze_data_refs_alignment === recording new base alignment for _9 alignment: 8 misalignment: 0 based on: _7 = MEM[(float * *)SR.8_23]; vect_compute_data_ref_alignment: can't force alignment of ref: MEM[(float * *)SR.8_23] vect_compute_data_ref_alignment: can't force alignment of ref: MEM[(float * *)SR.8_23 + 8B] === vect_prune_runtime_alias_test_list === === vect_enhance_data_refs_alignment === vector alignment may not be reachable vect_can_advance_ivs_p: Analyze phi: ret_17 = PHI <0(5), ret_6(6)> reduc or virtual phi. skip. Analyze phi: SR.8_23 = PHI <_9(5), _12(6)> Vectorizing an unaligned access. === vect_analyze_loop_operations === examining phi: ret_17 = PHI <0(5), ret_6(6)> examining phi: SR.8_23 = PHI <_9(5), _12(6)> ==> examining statement: _7 = MEM[(float * *)SR.8_23]; vect_is_simple_use: operand MEM[(float * *)SR.8_23] not ssa-name. use not simple. vect_is_simple_use: operand MEM[(float * *)SR.8_23] not ssa-name. use not simple. no array mode for V2DI[3] Data access with gaps requires scalar epilogue loop can't use a fully-masked loop because the target doesn't have the appropriate masked load or store. vect_model_load_cost: strided group_size = 3 . vect_model_load_cost: unaligned supported by hardware. vect_model_load_cost: inside_cost = 36, prologue_cost = 0 . ==> examining statement: _8 = MEM[(float * *)SR.8_23 + 8B]; vect_is_simple_use: operand MEM[(float * *)SR.8_23 + 8B] not ssa-name. use not simple. vect_is_simple_use: operand MEM[(float * *)SR.8_23 + 8B] not ssa-name. use not simple. no array mode for V2DI[3] Data access with gaps requires scalar epilogue loop vect_model_load_cost: unaligned supported by hardware. vect_model_load_cost: inside_cost = 12, prologue_cost = 0 . ==> examining statement: _14 = _8 - _7; vect_is_simple_use: operand _8 def_stmt: _8 = MEM[(float * *)SR.8_23 + 8B]; type of def: internal vect_is_simple_use: operand _7 def_stmt: _7 = MEM[(float * *)SR.8_23]; type of def: internal === vectorizable_operation === vect_model_simple_cost: inside_cost = 4, prologue_cost = 0 . ==> examining statement: _15 = _14 /[ex] 4; vect_is_simple_use: operand _14 def_stmt: _14 = _8 - _7; type of def: internal vect_is_simple_use: operand 4 op not supported by target. not vectorized: relevant stmt not supported: _15 = _14 /[ex] 4; bad operation or unsupported loop bound.

Line	Hotness	Pass	Source	Function / Inlining Chain
1			#include <vector>
2			std::size_t f(std::vector<std::vector<float>> const & v) {
3			std::size_t ret = 0;
4			for (auto const & w: v)
	100.00	ldist	^Loop 1 not distributed.	`std::size_t f(const std::vector<std::vector<float> >&)`
	100.00	vect	^=== analyzing loop === Analyzing loop at ../../src/test.cc:4 === analyze_loop_nest === === vect_analyze_loop_form === === get_loop_niters === Symbolic number of iterations is `(((((unsigned long) _10 - (unsigned long) _9) - 24) /[ex] 8) * 768614336404564651 & 2305843009213693951) + 1` === vect_analyze_data_refs === got vectype for stmt: `_7 = MEM[(float * )SR.8_23];vector(2) long unsigned int` got vectype for stmt: `_8 = MEM[(float )SR.8_23 + 8B];vector(2) long unsigned int` === vect_analyze_scalar_cycles === Analyze phi: `ret_17 = PHI <0(5), ret_6(6)>` Access function of PHI: `{0, +, _16}_1` step: `_16`, init: `0` step unknown. Analyze phi: `SR.8_23 = PHI <_9(5), _12(6)>` Access function of PHI: `{_9, +, 24}_1` step: `24`, init: `_9` Detected induction. Analyze phi: `ret_17 = PHI <0(5), ret_6(6)>` detected reduction: `ret_6 = _16 + ret_17;` Detected reduction. === vect_pattern_recog === vect_is_simple_use: operand `_16` def_stmt: `_16 = (long unsigned int) _15;` type of def: internal vect_is_simple_use: operand `_16` def_stmt: `_16 = (long unsigned int) _15;` type of def: internal vect_is_simple_use: operand `_15` def_stmt: `_15 = _14 /[ex] 4;` type of def: internal vect_is_simple_use: operand `_16` def_stmt: `_16 = (long unsigned int) _15;` type of def: internal === vect_analyze_data_ref_accesses === Detected interleaving load `MEM[(float )SR.8_23]` and `MEM[(float )SR.8_23 + 8B]` Detected interleaving load of size 3 starting with `_7 = MEM[(float )SR.8_23];` There is a gap of 1 elements after the group === vect_mark_stmts_to_be_vectorized === init: phi relevant? `ret_17 = PHI <0(5), ret_6(6)>` init: phi relevant? `SR.8_23 = PHI <_9(5), _12(6)>` init: stmt relevant? `_7 = MEM[(float )SR.8_23];` init: stmt relevant? `_8 = MEM[(float )SR.8_23 + 8B];` init: stmt relevant? `_14 = _8 - _7;` init: stmt relevant? `_15 = _14 /[ex] 4;` init: stmt relevant? `_16 = (long unsigned int) _15;` init: stmt relevant? `ret_6 = _16 + ret_17;` vec_stmt_relevant_p: used out of loop. vect_is_simple_use: operand `_16` def_stmt: `_16 = (long unsigned int) _15;` type of def: internal vec_stmt_relevant_p: stmt live but not relevant. mark relevant 1, live 1: `ret_6 = _16 + ret_17;` init: stmt relevant? `_12 = SR.8_23 + 24;` init: stmt relevant? `if (_10 != _12)` worklist: examine stmt: `ret_6 = _16 + ret_17;` vect_is_simple_use: operand `_16` def_stmt: `_16 = (long unsigned int) _15;` type of def: internal mark relevant 1, live 0: `_16 = (long unsigned int) _15;` vect_is_simple_use: operand `ret_17` def_stmt: `ret_17 = PHI <0(5), ret_6(6)>` type of def: reduction mark relevant 1, live 0: `ret_17 = PHI <0(5), ret_6(6)>` worklist: examine stmt: `ret_17 = PHI <0(5), ret_6(6)>` vect_is_simple_use: operand `0` vect_is_simple_use: operand `ret_6` def_stmt: `ret_6 = _16 + ret_17;` type of def: reduction reduc-stmt defining reduc-phi in the same nest. worklist: examine stmt: `_16 = (long unsigned int) _15;` vect_is_simple_use: operand `_15` def_stmt: `_15 = _14 /[ex] 4;` type of def: internal mark relevant 1, live 0: `_15 = _14 /[ex] 4;` worklist: examine stmt: `_15 = _14 /[ex] 4;` vect_is_simple_use: operand `_14` def_stmt: `_14 = _8 - _7;` type of def: internal mark relevant 1, live 0: `_14 = _8 - _7;` worklist: examine stmt: `_14 = _8 - _7;` vect_is_simple_use: operand `_8` def_stmt: `_8 = MEM[(float )SR.8_23 + 8B];` type of def: internal mark relevant 1, live 0: `_8 = MEM[(float )SR.8_23 + 8B];` vect_is_simple_use: operand `_7` def_stmt: `_7 = MEM[(float )SR.8_23];` type of def: internal mark relevant 1, live 0: `_7 = MEM[(float )SR.8_23];` worklist: examine stmt: `_7 = MEM[(float )SR.8_23];` worklist: examine stmt: `_8 = MEM[(float )SR.8_23 + 8B];` === vect_analyze_data_ref_dependences === === vect_determine_vectorization_factor === ==> examining phi: `ret_17 = PHI <0(5), ret_6(6)>` get vectype for scalar type: `size_t` vectype: `vector(2) long unsigned int` nunits = 2 ==> examining phi: `SR.8_23 = PHI <_9(5), _12(6)>` ==> examining statement: `_7 = MEM[(float )SR.8_23];` get vectype for scalar type: `float ` vectype: `vector(2) long unsigned int` nunits = 2 ==> examining statement: `_8 = MEM[(float * )SR.8_23 + 8B];` get vectype for scalar type: `float ` vectype: `vector(2) long unsigned int` nunits = 2 ==> examining statement: `_14 = _8 - _7;` get vectype for scalar type: `long int` vectype: `vector(2) long int` get vectype for scalar type: `long int` vectype: `vector(2) long int` nunits = 2 ==> examining statement: `_15 = _14 /[ex] 4;` get vectype for scalar type: `long int` vectype: `vector(2) long int` get vectype for scalar type: `long int` vectype: `vector(2) long int` nunits = 2 ==> examining statement: `_16 = (long unsigned int) _15;` get vectype for scalar type: `long unsigned int` vectype: `vector(2) long unsigned int` get vectype for scalar type: `long unsigned int` vectype: `vector(2) long unsigned int` nunits = 2 ==> examining statement: `ret_6 = _16 + ret_17;` get vectype for scalar type: `size_t` vectype: `vector(2) long unsigned int` get vectype for scalar type: `size_t` vectype: `vector(2) long unsigned int` nunits = 2 ==> examining statement: `_12 = SR.8_23 + 24;` skip. ==> examining statement: `if (_10 != _12)` skip. vectorization factor = 2 === vect_analyze_slp === === vect_make_slp_decision === === vect_analyze_data_refs_alignment === recording new base alignment for `_9` alignment: 8 misalignment: 0 based on: `_7 = MEM[(float * )SR.8_23];` vect_compute_data_ref_alignment: can't force alignment of ref: `MEM[(float )SR.8_23]` vect_compute_data_ref_alignment: can't force alignment of ref: `MEM[(float )SR.8_23 + 8B]` === vect_prune_runtime_alias_test_list === === vect_enhance_data_refs_alignment === vector alignment may not be reachable vect_can_advance_ivs_p: Analyze phi: `ret_17 = PHI <0(5), ret_6(6)>` reduc or virtual phi. skip. Analyze phi: `SR.8_23 = PHI <_9(5), _12(6)>` Vectorizing an unaligned access. === vect_analyze_loop_operations === examining phi: `ret_17 = PHI <0(5), ret_6(6)>` examining phi: `SR.8_23 = PHI <_9(5), _12(6)>` ==> examining statement: `_7 = MEM[(float )SR.8_23];` vect_is_simple_use: operand `MEM[(float )SR.8_23]` not ssa-name. use not simple. vect_is_simple_use: operand `MEM[(float )SR.8_23]` not ssa-name. use not simple. no array mode for V2DI[3] Data access with gaps requires scalar epilogue loop can't use a fully-masked loop because the target doesn't have the appropriate masked load or store. vect_model_load_cost: strided group_size = 3 . vect_model_load_cost: unaligned supported by hardware. vect_model_load_cost: inside_cost = 36, prologue_cost = 0 . ==> examining statement: `_8 = MEM[(float )SR.8_23 + 8B];` vect_is_simple_use: operand `MEM[(float )SR.8_23 + 8B]` not ssa-name. use not simple. vect_is_simple_use: operand `MEM[(float )SR.8_23 + 8B]` not ssa-name. use not simple. no array mode for V2DI[3] Data access with gaps requires scalar epilogue loop vect_model_load_cost: unaligned supported by hardware. vect_model_load_cost: inside_cost = 12, prologue_cost = 0 . ==> examining statement: `_14 = _8 - _7;` vect_is_simple_use: operand `_8` def_stmt: `_8 = MEM[(float )SR.8_23 + 8B];` type of def: internal vect_is_simple_use: operand `_7` def_stmt: `_7 = MEM[(float *)SR.8_23];` type of def: internal === vectorizable_operation === vect_model_simple_cost: inside_cost = 4, prologue_cost = 0 . ==> examining statement: `_15 = _14 /[ex] 4;` vect_is_simple_use: operand `_14` def_stmt: `_14 = _8 - _7;` type of def: internal vect_is_simple_use: operand `4` op not supported by target. not vectorized: relevant stmt not supported: `_15 = _14 /[ex] 4;` bad operation or unsupported loop bound.	`std::size_t f(const std::vector<std::vector<float> >&)`
5			ret += w.size();
6			return ret;
7			}