| Message ID | 20220929050051.30330-1-hongtao.liu@intel.com |
|---|---|
| State | New |
| Headers | show |
| Series | Check nonlinear iv in vect_can_advance_ivs_p. | expand |
On Thu, Sep 29, 2022 at 7:03 AM liuhongt via Gcc-patches <gcc-patches@gcc.gnu.org> wrote: > > vectorizable_nonlinear_induction doesn't always guard > vect_peel_nonlinear_iv_init when it's called by > vect_update_ivs_after_vectorizer which is supposed to be guarded > by vect_can_advance_ivs_p. The patch put part codes from > vectorizable_nonlinear_induction into a new function > vect_can_peel_nonlinear_iv_p, and make vect_can_advcance_ivs_p call it. > > Also this patch fix a typo in my initial patch as bellow. > - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) > - && induction_type == vect_step_op_mul) > if (!INTEGRAL_TYPE_P (TREE_TYPE (vectype))) > > Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}. > Ok for trunk? OK. Thanks, Richard. > gcc/ChangeLog: > > PR tree-optimization/107055 > * tree-vect-loop-manip.cc (vect_can_advance_ivs_p): Check for > nonlinear induction variables. > * tree-vect-loop.cc (vect_can_peel_nonlinear_iv_p): New > functions. > (vectorizable_nonlinear_induction): Put part codes into > vect_can_peel_nonlinear_iv_p. > * tree-vectorizer.h (vect_can_peel_nonlinear_iv_p): Declare. > > gcc/testsuite/ChangeLog: > > * gcc.target/i386/pr107055.c: New test. > --- > gcc/testsuite/gcc.target/i386/pr107055.c | 4 ++ > gcc/tree-vect-loop-manip.cc | 10 +++ > gcc/tree-vect-loop.cc | 82 ++++++++++++++---------- > gcc/tree-vectorizer.h | 3 + > 4 files changed, 64 insertions(+), 35 deletions(-) > create mode 100644 gcc/testsuite/gcc.target/i386/pr107055.c > > diff --git a/gcc/testsuite/gcc.target/i386/pr107055.c b/gcc/testsuite/gcc.target/i386/pr107055.c > new file mode 100644 > index 00000000000..63bcb3d742e > --- /dev/null > +++ b/gcc/testsuite/gcc.target/i386/pr107055.c > @@ -0,0 +1,4 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O1 -fno-tree-dce -fno-vect-cost-model -ftree-vectorize -fprofile-arcs" } */ > + > +#include "../../gcc.dg/torture/pr24257.c" > diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc > index 74b221a973c..1d96130c985 100644 > --- a/gcc/tree-vect-loop-manip.cc > +++ b/gcc/tree-vect-loop-manip.cc > @@ -1413,6 +1413,7 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) > for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) > { > tree evolution_part; > + enum vect_induction_op_type induction_type; > > gphi *phi = gsi.phi (); > stmt_vec_info phi_info = loop_vinfo->lookup_stmt (phi); > @@ -1432,6 +1433,15 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) > continue; > } > > + induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); > + if (induction_type != vect_step_op_add) > + { > + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) > + return false; > + > + continue; > + } > + > /* Analyze the evolution function. */ > > evolution_part = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (phi_info); > diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc > index aabdc6f2d81..2536cc3cf49 100644 > --- a/gcc/tree-vect-loop.cc > +++ b/gcc/tree-vect-loop.cc > @@ -8558,6 +8558,50 @@ vect_update_nonlinear_iv (gimple_seq* stmts, tree vectype, > return vec_def; > > } > + > +/* Return true if vectorizer can peel for nonlinear iv. */ > +bool > +vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, > + enum vect_induction_op_type induction_type) > +{ > + tree niters_skip; > + /* Init_expr will be update by vect_update_ivs_after_vectorizer, > + if niters is unkown: > + For shift, when shift mount >= precision, there would be UD. > + For mult, don't known how to generate > + init_expr * pow (step, niters) for variable niters. > + For neg, it should be ok, since niters of vectorized main loop > + will always be multiple of 2. */ > + if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) > + && induction_type != vect_step_op_neg) > + { > + if (dump_enabled_p ()) > + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > + "Peeling for epilogue is not supported" > + " for nonlinear induction except neg" > + " when iteration count is unknown.\n"); > + return false; > + } > + > + /* Also doens't support peel for neg when niter is variable. > + ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ > + niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); > + if ((niters_skip != NULL_TREE > + && TREE_CODE (niters_skip) != INTEGER_CST) > + || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) > + && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) > + { > + if (dump_enabled_p ()) > + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > + "Peeling for alignement is not supported" > + " for nonlinear induction when niters_skip" > + " is not constant.\n"); > + return false; > + } > + > + return true; > +} > + > /* Function vectorizable_induction > > Check if STMT_INFO performs an nonlinear induction computation that can be > @@ -8628,42 +8672,9 @@ vectorizable_nonlinear_induction (loop_vec_info loop_vinfo, > return false; > } > > - /* Init_expr will be update by vect_update_ivs_after_vectorizer, > - if niters is unkown: > - For shift, when shift mount >= precision, there would be UD. > - For mult, don't known how to generate > - init_expr * pow (step, niters) for variable niters. > - For neg, it should be ok, since niters of vectorized main loop > - will always be multiple of 2. */ > - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) > - && induction_type != vect_step_op_neg) > - { > - if (dump_enabled_p ()) > - dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > - "Peeling for epilogue is not supported" > - " for nonlinear induction except neg" > - " when iteration count is unknown.\n"); > - return false; > - } > - > - /* Also doens't support peel for neg when niter is variable. > - ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ > - niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); > - if ((niters_skip != NULL_TREE > - && TREE_CODE (niters_skip) != INTEGER_CST) > - || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) > - && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) > - { > - if (dump_enabled_p ()) > - dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > - "Peeling for alignement is not supported" > - " for nonlinear induction when niters_skip" > - " is not constant.\n"); > - return false; > - } > + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) > + return false; > > - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) > - && induction_type == vect_step_op_mul) > if (!INTEGRAL_TYPE_P (TREE_TYPE (vectype))) > { > if (dump_enabled_p ()) > @@ -8799,6 +8810,7 @@ vectorizable_nonlinear_induction (loop_vec_info loop_vinfo, > > gimple_seq stmts = NULL; > > + niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); > /* If we are using the loop mask to "peel" for alignment then we need > to adjust the start value here. */ > if (niters_skip != NULL_TREE) > diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h > index 5e75ed1532b..4870c754499 100644 > --- a/gcc/tree-vectorizer.h > +++ b/gcc/tree-vectorizer.h > @@ -2343,6 +2343,9 @@ extern tree cse_and_gimplify_to_preheader (loop_vec_info, tree); > /* Nonlinear induction. */ > extern tree vect_peel_nonlinear_iv_init (gimple_seq*, tree, tree, > tree, enum vect_induction_op_type); > +extern bool > +vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, > + enum vect_induction_op_type induction_type); > > /* In tree-vect-slp.cc. */ > extern void vect_slp_init (void); > -- > 2.18.1 >
diff --git a/gcc/testsuite/gcc.target/i386/pr107055.c b/gcc/testsuite/gcc.target/i386/pr107055.c new file mode 100644 index 00000000000..63bcb3d742e --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr107055.c @@ -0,0 +1,4 @@ +/* { dg-do compile } */ +/* { dg-options "-O1 -fno-tree-dce -fno-vect-cost-model -ftree-vectorize -fprofile-arcs" } */ + +#include "../../gcc.dg/torture/pr24257.c" diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index 74b221a973c..1d96130c985 100644 --- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -1413,6 +1413,7 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { tree evolution_part; + enum vect_induction_op_type induction_type; gphi *phi = gsi.phi (); stmt_vec_info phi_info = loop_vinfo->lookup_stmt (phi); @@ -1432,6 +1433,15 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) continue; } + induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); + if (induction_type != vect_step_op_add) + { + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) + return false; + + continue; + } + /* Analyze the evolution function. */ evolution_part = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (phi_info); diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc index aabdc6f2d81..2536cc3cf49 100644 --- a/gcc/tree-vect-loop.cc +++ b/gcc/tree-vect-loop.cc @@ -8558,6 +8558,50 @@ vect_update_nonlinear_iv (gimple_seq* stmts, tree vectype, return vec_def; } + +/* Return true if vectorizer can peel for nonlinear iv. */ +bool +vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, + enum vect_induction_op_type induction_type) +{ + tree niters_skip; + /* Init_expr will be update by vect_update_ivs_after_vectorizer, + if niters is unkown: + For shift, when shift mount >= precision, there would be UD. + For mult, don't known how to generate + init_expr * pow (step, niters) for variable niters. + For neg, it should be ok, since niters of vectorized main loop + will always be multiple of 2. */ + if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) + && induction_type != vect_step_op_neg) + { + if (dump_enabled_p ()) + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, + "Peeling for epilogue is not supported" + " for nonlinear induction except neg" + " when iteration count is unknown.\n"); + return false; + } + + /* Also doens't support peel for neg when niter is variable. + ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ + niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); + if ((niters_skip != NULL_TREE + && TREE_CODE (niters_skip) != INTEGER_CST) + || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) + && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) + { + if (dump_enabled_p ()) + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, + "Peeling for alignement is not supported" + " for nonlinear induction when niters_skip" + " is not constant.\n"); + return false; + } + + return true; +} + /* Function vectorizable_induction Check if STMT_INFO performs an nonlinear induction computation that can be @@ -8628,42 +8672,9 @@ vectorizable_nonlinear_induction (loop_vec_info loop_vinfo, return false; } - /* Init_expr will be update by vect_update_ivs_after_vectorizer, - if niters is unkown: - For shift, when shift mount >= precision, there would be UD. - For mult, don't known how to generate - init_expr * pow (step, niters) for variable niters. - For neg, it should be ok, since niters of vectorized main loop - will always be multiple of 2. */ - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) - && induction_type != vect_step_op_neg) - { - if (dump_enabled_p ()) - dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, - "Peeling for epilogue is not supported" - " for nonlinear induction except neg" - " when iteration count is unknown.\n"); - return false; - } - - /* Also doens't support peel for neg when niter is variable. - ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ - niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); - if ((niters_skip != NULL_TREE - && TREE_CODE (niters_skip) != INTEGER_CST) - || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) - && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) - { - if (dump_enabled_p ()) - dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, - "Peeling for alignement is not supported" - " for nonlinear induction when niters_skip" - " is not constant.\n"); - return false; - } + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) + return false; - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) - && induction_type == vect_step_op_mul) if (!INTEGRAL_TYPE_P (TREE_TYPE (vectype))) { if (dump_enabled_p ()) @@ -8799,6 +8810,7 @@ vectorizable_nonlinear_induction (loop_vec_info loop_vinfo, gimple_seq stmts = NULL; + niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); /* If we are using the loop mask to "peel" for alignment then we need to adjust the start value here. */ if (niters_skip != NULL_TREE) diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h index 5e75ed1532b..4870c754499 100644 --- a/gcc/tree-vectorizer.h +++ b/gcc/tree-vectorizer.h @@ -2343,6 +2343,9 @@ extern tree cse_and_gimplify_to_preheader (loop_vec_info, tree); /* Nonlinear induction. */ extern tree vect_peel_nonlinear_iv_init (gimple_seq*, tree, tree, tree, enum vect_induction_op_type); +extern bool +vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, + enum vect_induction_op_type induction_type); /* In tree-vect-slp.cc. */ extern void vect_slp_init (void);
vectorizable_nonlinear_induction doesn't always guard vect_peel_nonlinear_iv_init when it's called by vect_update_ivs_after_vectorizer which is supposed to be guarded by vect_can_advance_ivs_p. The patch put part codes from vectorizable_nonlinear_induction into a new function vect_can_peel_nonlinear_iv_p, and make vect_can_advcance_ivs_p call it. Also this patch fix a typo in my initial patch as bellow. - if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) - && induction_type == vect_step_op_mul) if (!INTEGRAL_TYPE_P (TREE_TYPE (vectype))) Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}. Ok for trunk? gcc/ChangeLog: PR tree-optimization/107055 * tree-vect-loop-manip.cc (vect_can_advance_ivs_p): Check for nonlinear induction variables. * tree-vect-loop.cc (vect_can_peel_nonlinear_iv_p): New functions. (vectorizable_nonlinear_induction): Put part codes into vect_can_peel_nonlinear_iv_p. * tree-vectorizer.h (vect_can_peel_nonlinear_iv_p): Declare. gcc/testsuite/ChangeLog: * gcc.target/i386/pr107055.c: New test. --- gcc/testsuite/gcc.target/i386/pr107055.c | 4 ++ gcc/tree-vect-loop-manip.cc | 10 +++ gcc/tree-vect-loop.cc | 82 ++++++++++++++---------- gcc/tree-vectorizer.h | 3 + 4 files changed, 64 insertions(+), 35 deletions(-) create mode 100644 gcc/testsuite/gcc.target/i386/pr107055.c