loop-interchange.c

Line Hotness Pass Source Function / Inlining Chain
1 
/* { dg-do run } */
2 
/* { dg-options "-O2 -floop-interchange -fassociative-math -fno-signed-zeros -fno-trapping-math -fdump-tree-linterchange-details" } */
3 
/* { dg-skip-if "too big data segment" { avr-*-* visium-*-* } } */
4 

    
  

  

    5
    
    
    
/* Copied from graphite/interchange-4.c */

    
  

  

    6
    
    
    

    
  

  

    7
    
    
    
#define DEBUG 0

    
  

  

    8
    
    
    
#if DEBUG

    
  

  

    9
    
    
    
#include <stdio.h>

    
  

  

    10
    
    
    
#endif

    
  

  

    11
    
    
    

    
  

  

    12
    
    
    
double u[1782225];

    
  

  

    13
    
    
    

    
  

  

    14
    
    
    
static int __attribute__((noinline))

    
  

  

    15
    
    
    
foo (int N, int *res)

    
  

  

    16
    
    
    
{

    
  

  

    17
    
    
    
  int i, j;

    
  

  

    18
    
    
    
  double sum = 0;

    
  

  

    19
    
    
    
  for (i = 0; i < N; i++)

    
  

  

    
    
0.01    
    
linterchange
    
  ^=== tree_loop_interchange::interchange ===
    Loop_pair<outer:2, inner:3> is interchanged
    loops interchanged in loop nest
  

    
    
  
  • foo.constprop
    • 
    

  

  

    
    
0.01    
    
vect
    

                        
  ^=== analyzing loop ===
    
    Analyzing loop at loop-interchange.c:19
    === analyze_loop_nest ===
      === vect_analyze_loop_form ===
        === vect_analyze_loop_form ===
          === get_loop_niters ===
        Considering outer-loop vectorization.
        === get_loop_niters ===
      === vect_analyze_data_refs ===
        analyze in outer loop: *(&u + (sizetype) (j_17 * 1335) * 8)
        	outer base_address: &u
        	outer offset from base address: 0
        	outer constant offset from base address: 0
        	outer step: 10680
        	outer base alignment: 16
        
        	outer base misalignment: 0
        
        	outer offset alignment: 128
        
        	outer step alignment: 8
        got vectype for stmt: _5 = u[_4];vector(2) double
      === vect_analyze_scalar_cycles ===
        Analyze phi: sum_42 = PHI <0.0(2), sum_37(10)>
        Access function of PHI: sum_42
        Analyze phi: j_17 = PHI <0(2), j_10(10)>
        Access function of PHI: {0, +, 1}_2
        step: 1,  init: 0
        Detected induction.
        Analyze phi: ivtmp_6 = PHI <1335(2), ivtmp_3(10)>
        Access function of PHI: {1335, +, 4294967295}_2
        step: 4294967295,  init: 1335
        Detected induction.
        Analyze phi: ivtmp_1 = PHI <1335(2), ivtmp_44(10)>
        Access function of PHI: {1335, +, 4294967295}_2
        step: 4294967295,  init: 1335
        Detected induction.
        Analyze phi: sum_42 = PHI <0.0(2), sum_37(10)>
        detected double reduction: sum_37 = PHI <sum_7(3)>
        Detected double reduction.
      === vect_analyze_scalar_cycles ===
        Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
        Access function of PHI: {sum_42, +, _5}_3
        step: _5,  init: sum_42
        step unknown.
        Analyze phi: i_26 = PHI <0(5), i_25(11)>
        Access function of PHI: {0, +, 1}_3
        step: 1,  init: 0
        Detected induction.
        Analyze phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
        Access function of PHI: {1335, +, 4294967295}_3
        step: 4294967295,  init: 1335
        Detected induction.
        Analyze phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
        Access function of PHI: {1335, +, 4294967295}_3
        step: 4294967295,  init: 1335
        Detected induction.
        Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
        detected reduction: sum_7 = _5 + sum_40;
        Detected vectorizable nested cycle.
      === vect_pattern_recog ===
        vect_is_simple_use: operand j_17
        def_stmt: j_17 = PHI <0(2), j_10(10)>
        type of def: induction
      === vect_analyze_data_ref_accesses ===
        grouped access in outer loop.
        not vectorized: complicated access pattern.
      bad data access.
  

    
    
  
  • foo.constprop
    • 
    

  

  

    20
    
    
    
    for (j = 0; j < N; j++)

    
  

  

    
    
6.96    
    
ldist
    
    ^Loop 3 not distributed.
    

    
    
  
  • foo.constprop
    • 
    

  

  

    
    
6.96    
    
vect
    

                        
    ^=== analyzing loop ===
      
      Analyzing loop at loop-interchange.c:20
      === analyze_loop_nest ===
        === vect_analyze_loop_form ===
          === get_loop_niters ===
        === vect_analyze_data_refs ===
          got vectype for stmt: _5 = u[_4];vector(2) double
        === vect_analyze_scalar_cycles ===
          Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
          Access function of PHI: {sum_42, +, _5}_3
          step: _5,  init: sum_42
          step unknown.
          Analyze phi: i_26 = PHI <0(5), i_25(11)>
          Access function of PHI: {0, +, 1}_3
          step: 1,  init: 0
          Detected induction.
          Analyze phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
          Access function of PHI: {1335, +, 4294967295}_3
          step: 4294967295,  init: 1335
          Detected induction.
          Analyze phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
          Access function of PHI: {1335, +, 4294967295}_3
          step: 4294967295,  init: 1335
          Detected induction.
          Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
          detected reduction: sum_7 = _5 + sum_40;
          Detected reduction.
        === vect_pattern_recog ===
          vect_is_simple_use: operand j_17
          def_stmt: j_17 = PHI <0(2), j_10(10)>
          type of def: external
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
        === vect_analyze_data_ref_accesses ===
        === vect_mark_stmts_to_be_vectorized ===
          init: phi relevant? sum_40 = PHI <sum_42(5), sum_7(11)>
          init: phi relevant? i_26 = PHI <0(5), i_25(11)>
          init: phi relevant? ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
          init: phi relevant? ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
          init: stmt relevant? # DEBUG sum => NULL
          init: stmt relevant? # DEBUG i => i_26
          init: stmt relevant? # DEBUG sum => NULL
          init: stmt relevant? # DEBUG j => 0
          init: stmt relevant? # DEBUG sum => sum_40
          init: stmt relevant? # DEBUG j => j_17
          init: stmt relevant? # DEBUG BEGIN_STMT
          init: stmt relevant? _2 = j_17 * 1335;
          init: stmt relevant? _4 = _2 + i_26;
          init: stmt relevant? _5 = u[_4];
          init: stmt relevant? sum_7 = _5 + sum_40;
          vec_stmt_relevant_p: used out of loop.
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
          vec_stmt_relevant_p: stmt live but not relevant.
          mark relevant 1, live 1: sum_7 = _5 + sum_40;
          init: stmt relevant? # DEBUG sum => sum_7
          init: stmt relevant? # DEBUG D#2 => j_17 + 1
          init: stmt relevant? # DEBUG j => D#2
          init: stmt relevant? # DEBUG sum => sum_7
          init: stmt relevant? # DEBUG j => D#2
          init: stmt relevant? # DEBUG i => NULL
          init: stmt relevant? # DEBUG sum => NULL
          init: stmt relevant? # DEBUG i => NULL
          init: stmt relevant? ivtmp_19 = ivtmp_24 - 1;
          init: stmt relevant? i_25 = i_26 + 1;
          init: stmt relevant? ivtmp_46 = ivtmp_45 - 1;
          init: stmt relevant? if (ivtmp_46 != 0)
          worklist: examine stmt: sum_7 = _5 + sum_40;
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
          mark relevant 1, live 0: _5 = u[_4];
          vect_is_simple_use: operand sum_40
          def_stmt: sum_40 = PHI <sum_42(5), sum_7(11)>
          type of def: reduction
          mark relevant 1, live 0: sum_40 = PHI <sum_42(5), sum_7(11)>
          worklist: examine stmt: sum_40 = PHI <sum_42(5), sum_7(11)>
          vect_is_simple_use: operand sum_42
          def_stmt: sum_42 = PHI <0.0(2), sum_37(10)>
          type of def: external
          def_stmt is out of loop.
          vect_is_simple_use: operand sum_7
          def_stmt: sum_7 = _5 + sum_40;
          type of def: reduction
          reduc-stmt defining reduc-phi in the same nest.
          worklist: examine stmt: _5 = u[_4];
        === vect_analyze_data_ref_dependences ===
        === vect_determine_vectorization_factor ===
          ==> examining phi: sum_40 = PHI <sum_42(5), sum_7(11)>
          get vectype for scalar type:  double
          vectype: vector(2) double
          nunits = 2
          ==> examining phi: i_26 = PHI <0(5), i_25(11)>
          ==> examining phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
          ==> examining phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
          ==> examining statement: # DEBUG sum => NULL
          skip.
          ==> examining statement: # DEBUG i => i_26
          skip.
          ==> examining statement: # DEBUG sum => NULL
          skip.
          ==> examining statement: # DEBUG j => 0
          skip.
          ==> examining statement: # DEBUG sum => sum_40
          skip.
          ==> examining statement: # DEBUG j => j_17
          skip.
          ==> examining statement: # DEBUG BEGIN_STMT
          skip.
          ==> examining statement: _2 = j_17 * 1335;
          skip.
          ==> examining statement: _4 = _2 + i_26;
          skip.
          ==> examining statement: _5 = u[_4];
          get vectype for scalar type:  double
          vectype: vector(2) double
          nunits = 2
          ==> examining statement: sum_7 = _5 + sum_40;
          get vectype for scalar type:  double
          vectype: vector(2) double
          get vectype for scalar type:  double
          vectype: vector(2) double
          nunits = 2
          ==> examining statement: # DEBUG sum => sum_7
          skip.
          ==> examining statement: # DEBUG D#2 => j_17 + 1
          skip.
          ==> examining statement: # DEBUG j => D#2
          skip.
          ==> examining statement: # DEBUG sum => sum_7
          skip.
          ==> examining statement: # DEBUG j => D#2
          skip.
          ==> examining statement: # DEBUG i => NULL
          skip.
          ==> examining statement: # DEBUG sum => NULL
          skip.
          ==> examining statement: # DEBUG i => NULL
          skip.
          ==> examining statement: ivtmp_19 = ivtmp_24 - 1;
          skip.
          ==> examining statement: i_25 = i_26 + 1;
          skip.
          ==> examining statement: ivtmp_46 = ivtmp_45 - 1;
          skip.
          ==> examining statement: if (ivtmp_46 != 0)
          skip.
          vectorization factor = 2
        === vect_analyze_slp ===
        === vect_make_slp_decision ===
        vectorization_factor = 2, niters = 1335
        === vect_analyze_data_refs_alignment ===
          recording new base alignment for &u
            alignment:    16
            misalignment: 0
            based on:     _5 = u[_4];
          vect_compute_data_ref_alignment:
          Unknown alignment for access: u[_4]
        === vect_prune_runtime_alias_test_list ===
        === vect_enhance_data_refs_alignment ===
          Unknown misalignment, naturally aligned
          vect_can_advance_ivs_p:
          Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
          reduc or virtual phi. skip.
          Analyze phi: i_26 = PHI <0(5), i_25(11)>
          Analyze phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
          Analyze phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
          vect_model_load_cost: aligned.
          vect_get_data_access_cost: inside_cost = 12, outside_cost = 0.
          vect_model_load_cost: unaligned supported by hardware.
          vect_get_data_access_cost: inside_cost = 12, outside_cost = 0.
          Vectorizing an unaligned access.
        === vect_analyze_loop_operations ===
          examining phi: sum_40 = PHI <sum_42(5), sum_7(11)>
          examining phi: i_26 = PHI <0(5), i_25(11)>
          examining phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
          examining phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
          ==> examining statement: # DEBUG sum => NULL
          irrelevant.
          ==> examining statement: # DEBUG i => i_26
          irrelevant.
          ==> examining statement: # DEBUG sum => NULL
          irrelevant.
          ==> examining statement: # DEBUG j => 0
          irrelevant.
          ==> examining statement: # DEBUG sum => sum_40
          irrelevant.
          ==> examining statement: # DEBUG j => j_17
          irrelevant.
          ==> examining statement: # DEBUG BEGIN_STMT
          irrelevant.
          ==> examining statement: _2 = j_17 * 1335;
          irrelevant.
          ==> examining statement: _4 = _2 + i_26;
          irrelevant.
          ==> examining statement: _5 = u[_4];
          num. args = 4 (not unary/binary/ternary op).
          vect_is_simple_use: operand u[_4]
          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_model_load_cost: unaligned supported by hardware.
          vect_model_load_cost: inside_cost = 12, prologue_cost = 0 .
          ==> examining statement: sum_7 = _5 + sum_40;
          vect_is_simple_use: operand _5
          def_stmt: _5 = u[_4];
          type of def: internal
          vect_is_simple_use: operand sum_40
          def_stmt: sum_40 = PHI <sum_42(5), sum_7(11)>
          type of def: reduction
          reduc op not supported by target.
          vect_model_reduction_cost: inside_cost = 12, prologue_cost = 4, epilogue_cost = 28 .
          ==> examining statement: # DEBUG sum => sum_7
          irrelevant.
          ==> examining statement: # DEBUG D#2 => j_17 + 1
          irrelevant.
          ==> examining statement: # DEBUG j => D#2
          irrelevant.
          ==> examining statement: # DEBUG sum => sum_7
          irrelevant.
          ==> examining statement: # DEBUG j => D#2
          irrelevant.
          ==> examining statement: # DEBUG i => NULL
          irrelevant.
          ==> examining statement: # DEBUG sum => NULL
          irrelevant.
          ==> examining statement: # DEBUG i => NULL
          irrelevant.
          ==> examining statement: ivtmp_19 = ivtmp_24 - 1;
          irrelevant.
          ==> examining statement: i_25 = i_26 + 1;
          irrelevant.
          ==> examining statement: ivtmp_46 = ivtmp_45 - 1;
          irrelevant.
          ==> examining statement: if (ivtmp_46 != 0)
          irrelevant.
        not using a fully-masked loop.
        Cost model analysis: 
          Vector inside of loop cost: 24
          Vector prologue cost: 4
          Vector epilogue cost: 44
          Scalar iteration cost: 16
          Scalar outside cost: 0
          Vector outside cost: 48
          prologue iterations: 0
          epilogue iterations: 1
          Calculated minimum iters for profitability: 10
          Runtime profitability threshold = 10
          Static estimate profitability threshold = 10
        epilog loop required
        vect_can_advance_ivs_p:
        Analyze phi: sum_40 = PHI <sum_42(5), sum_7(11)>
        reduc or virtual phi. skip.
        Analyze phi: i_26 = PHI <0(5), i_25(11)>
        Analyze phi: ivtmp_24 = PHI <1335(5), ivtmp_19(11)>
        Analyze phi: ivtmp_45 = PHI <1335(5), ivtmp_46(11)>
      loop vectorized
      === vec_transform_loop ===
        vect_can_advance_ivs_p:
        Analyze phi: sum_40 = PHI <sum_7(11), sum_42(5)>
        reduc or virtual phi. skip.
        Analyze phi: i_26 = PHI <i_25(11), 0(5)>
        Analyze phi: ivtmp_24 = PHI <ivtmp_19(11), 1335(5)>
        Analyze phi: ivtmp_45 = PHI <ivtmp_46(11), 1335(5)>
        vect_update_ivs_after_vectorizer: phi: sum_40 = PHI <sum_7(11), sum_42(5)>
        reduc or virtual phi. skip.
        vect_update_ivs_after_vectorizer: phi: i_26 = PHI <i_25(11), 0(5)>
        vect_update_ivs_after_vectorizer: phi: ivtmp_24 = PHI <ivtmp_19(11), 1335(5)>
        vect_update_ivs_after_vectorizer: phi: ivtmp_45 = PHI <ivtmp_46(11), 1335(5)>
        ------>vectorizing phi: sum_40 = PHI <sum_7(11), sum_42(16)>
        transform phi.
        ------>vectorizing phi: i_26 = PHI <i_25(11), 0(16)>
        ------>vectorizing phi: ivtmp_24 = PHI <ivtmp_19(11), 1335(16)>
        ------>vectorizing phi: ivtmp_45 = PHI <ivtmp_46(11), 1335(16)>
        ------>vectorizing phi: vect_sum_7.15_51 = PHI <(11), (16)>
        ------>vectorizing statement: # DEBUG sum => NULL
        ------>vectorizing statement: # DEBUG i => i_26
        ------>vectorizing statement: # DEBUG sum => NULL
        ------>vectorizing statement: # DEBUG j => 0
        ------>vectorizing statement: # DEBUG sum => sum_40
        ------>vectorizing statement: # DEBUG j => j_17
        ------>vectorizing statement: # DEBUG BEGIN_STMT
        ------>vectorizing statement: _2 = j_17 * 1335;
        ------>vectorizing statement: _4 = _2 + i_26;
        ------>vectorizing statement: _5 = u[_4];
        transform statement.
        transform load. ncopies = 1
        create vector_type-pointer variable to type: vector(2) double  vectorizing an array ref: u
        created vectp_u.17_52
        add new stmt: vect__5.18_58 = MEM[(double *)vectp_u.16_56];
        ------>vectorizing statement: sum_7 = _5 + sum_40;
        transform statement.
        vect_is_simple_use: operand _5
        def_stmt: _5 = u[_4];
        type of def: internal
        vect_is_simple_use: operand sum_40
        def_stmt: sum_40 = PHI <sum_7(11), sum_42(16)>
        type of def: reduction
        reduc op not supported by target.
        transform reduction.
        vect_get_vec_def_for_operand: _5
        vect_is_simple_use: operand _5
        def_stmt: _5 = u[_4];
        type of def: internal
          def_stmt =  _5 = u[_4];
        vect_get_vec_def_for_operand: sum_40
        vect_is_simple_use: operand sum_40
        def_stmt: sum_40 = PHI <sum_7(11), sum_42(16)>
        type of def: reduction
          def_stmt =  sum_40 = PHI <sum_7(11), sum_42(16)>
        add new stmt: vect_sum_7.19_59 = vect__5.18_58 + vect_sum_7.15_51;
        vect_is_simple_use: operand sum_42
        def_stmt: sum_42 = PHI <0.0(2), sum_37(10)>
        type of def: external
        transform reduction: created def-use cycle: vect_sum_7.15_51 = PHI <vect_sum_7.19_59(11), { 0.0, 0.0 }(16)>vect_sum_7.19_59 = vect__5.18_58 + vect_sum_7.15_51;
        Reduce using vector shifts
        extract scalar result
        ------>vectorizing statement: # DEBUG sum => sum_7
        ------>vectorizing statement: # DEBUG D#2 => j_17 + 1
        ------>vectorizing statement: # DEBUG j => D#2
        ------>vectorizing statement: # DEBUG sum => sum_7
        ------>vectorizing statement: # DEBUG j => D#2
        ------>vectorizing statement: # DEBUG i => NULL
        ------>vectorizing statement: # DEBUG sum => NULL
        ------>vectorizing statement: # DEBUG i => NULL
        ------>vectorizing statement: ivtmp_19 = ivtmp_24 - 1;
        ------>vectorizing statement: i_25 = i_26 + 1;
        ------>vectorizing statement: ivtmp_46 = ivtmp_45 - 1;
        ------>vectorizing statement: vectp_u.16_57 = vectp_u.16_56 + 16;
        ------>vectorizing statement: if (ivtmp_46 != 0)
        New loop exit condition: if (ivtmp_66 < 667)
        LOOP VECTORIZED
    

    
    
  
  • foo.constprop
    • 
    

  

  

    21
    
    
    
      sum = sum + u[i + 1335 * j];

    
  

  

    22
    
    
    

    
  

  

    23
    
    
    
  for (i = 0; i < N; i++)

    
  

  

    24
    
    
    
    u[1336 * i] *= 2;

    
  

  

    25
    
    
    

    
  

  

    26
    
    
    
  *res = sum + N + u[1336 * 2] + u[1336];

    
  

  

    27
    
    
    
}

    
  

  

    28
    
    
    

    
  

  

    29
    
    
    
extern void abort ();

    
  

  

    30
    
    
    

    
  

  

    31
    
    
    
int

    
  

  

    32
    
    
    
loop_interchange_c_main (void)

    
  

  

    33
    
    
    
{

    
  

  

    34
    
    
    
  int i, j, res;

    
  

  

    35
    
    
    

    
  

  

    36
    
    
    
  for (i = 0; i < 1782225; i++)

    
  

  

    
    
6.96    
    
ldist
    
  ^Loop 1 not distributed.
  

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    
    
6.96    
    
vect
    

                        
  ^=== analyzing loop ===
    
    Analyzing loop at loop-interchange.c:36
    === analyze_loop_nest ===
      === vect_analyze_loop_form ===
        === get_loop_niters ===
      === vect_analyze_data_refs ===
        got vectype for stmt: u[i_13] = 2.0e+0;vector(2) double
      === vect_analyze_scalar_cycles ===
        Analyze phi: i_13 = PHI <0(2), i_6(7)>
        Access function of PHI: {0, +, 1}_1
        step: 1,  init: 0
        Detected induction.
        Analyze phi: .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
        Analyze phi: ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
        Access function of PHI: {1782225, +, 4294967295}_1
        step: 4294967295,  init: 1782225
        Detected induction.
      === vect_pattern_recog ===
      === vect_analyze_data_ref_accesses ===
      === vect_mark_stmts_to_be_vectorized ===
        init: phi relevant? i_13 = PHI <0(2), i_6(7)>
        init: phi relevant? .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
        init: phi relevant? ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
        init: stmt relevant? # DEBUG i => i_13
        init: stmt relevant? # DEBUG BEGIN_STMT
        init: stmt relevant? u[i_13] = 2.0e+0;
        vec_stmt_relevant_p: stmt has vdefs.
        mark relevant 5, live 0: u[i_13] = 2.0e+0;
        init: stmt relevant? i_6 = i_13 + 1;
        init: stmt relevant? # DEBUG i => i_6
        init: stmt relevant? # DEBUG i => i_6
        init: stmt relevant? ivtmp_11 = ivtmp_2 - 1;
        init: stmt relevant? if (ivtmp_11 != 0)
        worklist: examine stmt: u[i_13] = 2.0e+0;
      === vect_analyze_data_ref_dependences ===
      === vect_determine_vectorization_factor ===
        ==> examining phi: i_13 = PHI <0(2), i_6(7)>
        ==> examining phi: .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
        ==> examining phi: ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
        ==> examining statement: # DEBUG i => i_13
        skip.
        ==> examining statement: # DEBUG BEGIN_STMT
        skip.
        ==> examining statement: u[i_13] = 2.0e+0;
        get vectype for scalar type:  double
        vectype: vector(2) double
        nunits = 2
        ==> examining statement: i_6 = i_13 + 1;
        skip.
        ==> examining statement: # DEBUG i => i_6
        skip.
        ==> examining statement: # DEBUG i => i_6
        skip.
        ==> examining statement: ivtmp_11 = ivtmp_2 - 1;
        skip.
        ==> examining statement: if (ivtmp_11 != 0)
        skip.
        vectorization factor = 2
      === vect_analyze_slp ===
      === vect_make_slp_decision ===
      vectorization_factor = 2, niters = 1782225
      === vect_analyze_data_refs_alignment ===
        recording new base alignment for &u
          alignment:    16
          misalignment: 0
          based on:     u[i_13] = 2.0e+0;
        vect_compute_data_ref_alignment:
        misalign = 0 bytes of ref u[i_13]
      === vect_prune_runtime_alias_test_list ===
      === vect_enhance_data_refs_alignment ===
        vect_can_advance_ivs_p:
        Analyze phi: i_13 = PHI <0(2), i_6(7)>
        Analyze phi: .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
        reduc or virtual phi. skip.
        Analyze phi: ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
        vect_model_store_cost: aligned.
        vect_get_data_access_cost: inside_cost = 12, outside_cost = 0.
      === vect_analyze_loop_operations ===
        examining phi: i_13 = PHI <0(2), i_6(7)>
        examining phi: .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
        examining phi: ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
        ==> examining statement: # DEBUG i => i_13
        irrelevant.
        ==> examining statement: # DEBUG BEGIN_STMT
        irrelevant.
        ==> examining statement: u[i_13] = 2.0e+0;
        vect_is_simple_use: operand 2.0e+0
        can't use a fully-masked loop because the target doesn't have the appropriate masked load or store.
        vect_model_store_cost: aligned.
        vect_model_store_cost: inside_cost = 12, prologue_cost = 4 .
        ==> examining statement: i_6 = i_13 + 1;
        irrelevant.
        ==> examining statement: # DEBUG i => i_6
        irrelevant.
        ==> examining statement: # DEBUG i => i_6
        irrelevant.
        ==> examining statement: ivtmp_11 = ivtmp_2 - 1;
        irrelevant.
        ==> examining statement: if (ivtmp_11 != 0)
        irrelevant.
      not using a fully-masked loop.
      Cost model analysis: 
        Vector inside of loop cost: 12
        Vector prologue cost: 4
        Vector epilogue cost: 12
        Scalar iteration cost: 12
        Scalar outside cost: 0
        Vector outside cost: 16
        prologue iterations: 0
        epilogue iterations: 1
        Calculated minimum iters for profitability: 2
        Runtime profitability threshold = 2
        Static estimate profitability threshold = 2
      epilog loop required
      vect_can_advance_ivs_p:
      Analyze phi: i_13 = PHI <0(2), i_6(7)>
      Analyze phi: .MEM_14 = PHI <.MEM_4(D)(2), .MEM_5(7)>
      reduc or virtual phi. skip.
      Analyze phi: ivtmp_2 = PHI <1782225(2), ivtmp_11(7)>
    loop vectorized
    === vec_transform_loop ===
      vect_can_advance_ivs_p:
      Analyze phi: i_13 = PHI <i_6(7), 0(2)>
      Analyze phi: .MEM_14 = PHI <.MEM_5(7), .MEM_4(D)(2)>
      reduc or virtual phi. skip.
      Analyze phi: ivtmp_2 = PHI <ivtmp_11(7), 1782225(2)>
      vect_update_ivs_after_vectorizer: phi: i_13 = PHI <i_6(7), 0(2)>
      vect_update_ivs_after_vectorizer: phi: .MEM_14 = PHI <.MEM_5(7), .MEM_4(D)(2)>
      reduc or virtual phi. skip.
      vect_update_ivs_after_vectorizer: phi: ivtmp_2 = PHI <ivtmp_11(7), 1782225(2)>
      ------>vectorizing phi: i_13 = PHI <i_6(7), 0(12)>
      ------>vectorizing phi: .MEM_14 = PHI <.MEM_5(7), .MEM_4(D)(12)>
      ------>vectorizing phi: ivtmp_2 = PHI <ivtmp_11(7), 1782225(12)>
      ------>vectorizing statement: # DEBUG i => i_13
      ------>vectorizing statement: # DEBUG BEGIN_STMT
      ------>vectorizing statement: u[i_13] = 2.0e+0;
      transform statement.
      vect_is_simple_use: operand 2.0e+0
      transform store. ncopies = 1
      vect_get_vec_def_for_operand: 2.0e+0
      vect_is_simple_use: operand 2.0e+0
      created new init_stmt: vect_cst__20 = { 2.0e+0, 2.0e+0 };
      create vector_type-pointer variable to type: vector(2) double  vectorizing an array ref: u
      created &u
      add new stmt: MEM[(double *)vectp_u.3_21] = vect_cst__20;
      ------>vectorizing statement: i_6 = i_13 + 1;
      ------>vectorizing statement: # DEBUG i => i_6
      ------>vectorizing statement: # DEBUG i => i_6
      ------>vectorizing statement: ivtmp_11 = ivtmp_2 - 1;
      ------>vectorizing statement: vectp_u.3_22 = vectp_u.3_21 + 16;
      ------>vectorizing statement: if (ivtmp_11 != 0)
      New loop exit condition: if (ivtmp_25 < 891112)
      LOOP VECTORIZED
  

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    37
    
    
    
    u[i] = 2;

    
  

  

    38
    
    
    

    
  

  

    39
    
    
    
  foo (1335, &res);

    
  

  

    
    
0.00    
    
inline
    
  ^  not inlinable: loop_interchange_c_main/2 -> foo.constprop/4, function not inlinable
  

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    
    
0.00    
    
inline
    
  ^  not inlinable: loop_interchange_c_main/2 -> foo.constprop/4, function not inlinable
  

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    
    
0.00    
    
inline
    
  ^  not inlinable: loop_interchange_c_main/2 -> foo.constprop/4, function not inlinable
  

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    40
    
    
    

    
  

  

    41
    
    
    
#if DEBUG

    
  

  

    42
    
    
    
  fprintf (stderr, "res = %d \n", res);

    
  

  

    43
    
    
    
#endif

    
  

  

    44
    
    
    

    
  

  

    45
    
    
    
  if (res != 3565793)

    
  

  

    46
    
    
    
    abort ();

    
  

  

    
    
0.00    
    
inline
    
    ^  not inlinable: loop_interchange_c_main/2 -> abort/3, function body not available
    

    
    
  
  • loop_interchange_c_main
    • 
    

  

  

    47
    
    
    

    
  

  

    48
    
    
    
  return 0;

    
  

  

    49
    
    
    
}

    
  

  

    50
    
    
    

    
  

  

    51
    
    
    
/* { dg-final { scan-tree-dump-times "Loop_pair<outer:., inner:.> is interchanged" 1 "linterchange"} } */