* [PATCH][4/4] SLP induction vectorization
@ 2017-06-02 13:22 Richard Biener
2017-06-02 23:38 ` Michael Meissner
0 siblings, 1 reply; 5+ messages in thread
From: Richard Biener @ 2017-06-02 13:22 UTC (permalink / raw)
To: gcc-patches
This implements vectorization of SLP inductions (in the not outer loop
vectorization case for now).
Bootstrapped and tested on x86_64-unknown-linux-gnu.
More testing is appreciated, I'm throwing it at SPEC2k6 now.
Richard.
2017-06-02 Richard Biener <rguenther@suse.de>
PR tree-optimization/80928
* tree-vect-loop.c (vect_update_vf_for_slp): Amend dumps.
(vect_analyze_loop_operations): Properly guard analysis for
pure SLP case.
(vect_transform_loop): Likewise.
(vect_analyze_loop_2): Also reset SLP type on PHIs.
(vect_model_induction_cost): Do not cost for pure SLP.
(vectorizable_induction): Pass in SLP node, implement SLP vectorization
of induction in inner loop vectorization.
* tree-vect-slp.c (vect_create_new_slp_node): Handle PHIs.
(vect_get_and_check_slp_defs): Handle vect_induction_def.
(vect_build_slp_tree): Likewise. Handle PHIs as terminating the
recursion.
(vect_analyze_slp_cost_1): Cost induction.
(vect_detect_hybrid_slp_stmts): Handle PHIs.
(vect_get_slp_vect_defs): Likewise.
* tree-vect-stmts.c (vect_analyze_stmt): Handle induction.
(vect_transform_stmt): Handle SLP reductions.
* tree-vectorizer.h (vectorizable_induction): Adjust.
* gcc.dg/vect/pr80928.c: New testcase.
* gcc.dg/vect/slp-13-big-array.c: Remove XFAILs.
* gcc.dg/vect/slp-13.c: Likewise.
* gcc.dg/vect/slp-perm-9.c: Prevent vectorization of check loop.
Index: gcc/testsuite/gcc.dg/vect/pr80928.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/pr80928.c (revision 0)
+++ gcc/testsuite/gcc.dg/vect/pr80928.c (working copy)
@@ -0,0 +1,37 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_int } */
+
+#include "tree-vect.h"
+
+int a[1020];
+
+void __attribute__((noinline))
+foo ()
+{
+ for (int i = 0; i < 1020; i += 5)
+ {
+ a[i] = i;
+ a[i+1] = i;
+ a[i+2] = i;
+ a[i+3] = i;
+ a[i+4] = i;
+ }
+}
+
+int main ()
+{
+ check_vect ();
+
+ foo ();
+
+ /* check results */
+ for (int i = 0; i < 1020; ++i)
+ if (a[i] != ((i + 4) / 5) * 5)
+ abort ();
+
+ return 0;
+}
+
+/* Make sure we are not triggering hybrid SLP due to the IV update. */
+/* { dg-final { scan-tree-dump "Loop contains only SLP stmts" "vect" } } */
+/* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" } } */
Index: gcc/testsuite/gcc.dg/vect/slp-13-big-array.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/slp-13-big-array.c (revision 248825)
+++ gcc/testsuite/gcc.dg/vect/slp-13-big-array.c (working copy)
@@ -22,7 +22,7 @@ main1 ()
abort ();
}
- /* Induction is not SLPable yet. */
+ /* Induction is SLPable. */
for (i = 0; i < N; i++)
{
out[i*8] = in[i*8] + i;
@@ -49,8 +49,7 @@ main1 ()
abort ();
}
- /* Induction is not SLPable yet and strided group size must be a power of 2
- to get vectorized. */
+ /* Induction is SLPable. */
for (i = 0; i < N/2; i++)
{
out2[i*12] = in2[i*12] + i;
@@ -134,7 +133,5 @@ int main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { target { vect_interleave && vect_extract_even_odd } } } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" { xfail *-*-* } } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" } } */
-
+/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { target { vect_interleave && vect_extract_even_odd } } } } */
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" } } */
Index: gcc/testsuite/gcc.dg/vect/slp-13.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/slp-13.c (revision 248825)
+++ gcc/testsuite/gcc.dg/vect/slp-13.c (working copy)
@@ -14,7 +14,7 @@ main1 ()
unsigned int in2[N*8] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63};
unsigned int out2[N*8];
- /* Induction is not SLPable yet. */
+ /* Induction is SLPable. */
for (i = 0; i < N; i++)
{
out[i*8] = in[i*8] + i;
@@ -41,8 +41,7 @@ main1 ()
abort ();
}
- /* Induction is not SLPable yet and strided group size must be a power of 2
- to get vectorized. */
+ /* Induction is SLPable. */
for (i = 0; i < N/2; i++)
{
out2[i*12] = in2[i*12] + i;
@@ -126,7 +125,5 @@ int main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { target { vect_interleave && vect_extract_even_odd } } } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" { xfail *-*-* } } } */
-/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" } } */
-
+/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" { target { vect_interleave && vect_extract_even_odd } } } } */
+/* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 3 "vect" } } */
Index: gcc/testsuite/gcc.dg/vect/slp-perm-9.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/slp-perm-9.c (revision 248825)
+++ gcc/testsuite/gcc.dg/vect/slp-perm-9.c (working copy)
@@ -42,6 +42,7 @@ int main (int argc, const char* argv[])
check_results[3*i] = 9 * i + 6;
check_results[3*i+1] = 9 * i + 15;
check_results[3*i+2] = 9 * i + 4;
+ __asm__ volatile ("" : : : "memory");
}
foo (input, output);
Index: gcc/tree-vect-loop.c
===================================================================
--- gcc/tree-vect-loop.c (revision 248825)
+++ gcc/tree-vect-loop.c (working copy)
@@ -1649,11 +1649,19 @@ vect_update_vf_for_slp (loop_vec_info lo
}
if (only_slp_in_loop)
- vectorization_factor = LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo);
+ {
+ dump_printf_loc (MSG_NOTE, vect_location,
+ "Loop contains only SLP stmts\n");
+ vectorization_factor = LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo);
+ }
else
- vectorization_factor
- = least_common_multiple (vectorization_factor,
- LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo));
+ {
+ dump_printf_loc (MSG_NOTE, vect_location,
+ "Loop contains SLP and non-SLP stmts\n");
+ vectorization_factor
+ = least_common_multiple (vectorization_factor,
+ LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo));
+ }
LOOP_VINFO_VECT_FACTOR (loop_vinfo) = vectorization_factor;
if (dump_enabled_p ())
@@ -1765,8 +1773,9 @@ vect_analyze_loop_operations (loop_vec_i
if (STMT_VINFO_RELEVANT_P (stmt_info))
{
need_to_vectorize = true;
- if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def)
- ok = vectorizable_induction (phi, NULL, NULL);
+ if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def
+ && ! PURE_SLP_STMT (stmt_info))
+ ok = vectorizable_induction (phi, NULL, NULL, NULL);
}
if (ok && STMT_VINFO_LIVE_P (stmt_info))
@@ -2275,6 +2284,12 @@ again:
for (i = 0; i < LOOP_VINFO_LOOP (loop_vinfo)->num_nodes; ++i)
{
basic_block bb = LOOP_VINFO_BBS (loop_vinfo)[i];
+ for (gimple_stmt_iterator si = gsi_start_phis (bb);
+ !gsi_end_p (si); gsi_next (&si))
+ {
+ stmt_vec_info stmt_info = vinfo_for_stmt (gsi_stmt (si));
+ STMT_SLP_TYPE (stmt_info) = loop_vect;
+ }
for (gimple_stmt_iterator si = gsi_start_bb (bb);
!gsi_end_p (si); gsi_next (&si))
{
@@ -3797,6 +3812,9 @@ vect_model_induction_cost (stmt_vec_info
void *target_cost_data = LOOP_VINFO_TARGET_COST_DATA (loop_vinfo);
unsigned inside_cost, prologue_cost;
+ if (PURE_SLP_STMT (stmt_info))
+ return;
+
/* loop cost for vec_loop. */
inside_cost = add_stmt_cost (target_cost_data, ncopies, vector_stmt,
stmt_info, 0, vect_body);
@@ -6086,7 +6104,7 @@ vect_min_worthwhile_factor (enum tree_co
bool
vectorizable_induction (gimple *phi,
gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED,
- gimple **vec_stmt)
+ gimple **vec_stmt, slp_tree slp_node)
{
stmt_vec_info stmt_info = vinfo_for_stmt (phi);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
@@ -6125,14 +6143,13 @@ vectorizable_induction (gimple *phi,
if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_induction_def)
return false;
- /* FORNOW: SLP not supported. */
- if (STMT_SLP_TYPE (stmt_info))
- return false;
-
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
unsigned nunits = TYPE_VECTOR_SUBPARTS (vectype);
- ncopies = vf / nunits;
+ if (slp_node)
+ ncopies = 1;
+ else
+ ncopies = vf / nunits;
gcc_assert (ncopies >= 1);
/* FORNOW. These restrictions should be relaxed. */
@@ -6234,6 +6251,147 @@ vectorizable_induction (gimple *phi,
/* Find the first insertion point in the BB. */
si = gsi_after_labels (bb);
+ /* For SLP induction we have to generate several IVs as for example
+ with group size 3 we need [i, i, i, i + S] [i + S, i + S, i + 2*S, i + 2*S]
+ [i + 2*S, i + 3*S, i + 3*S, i + 3*S]. The step is the same uniform
+ [VF*S, VF*S, VF*S, VF*S] for all. */
+ if (slp_node)
+ {
+ /* Convert the init to the desired type. */
+ stmts = NULL;
+ init_expr = gimple_convert (&stmts, TREE_TYPE (vectype), init_expr);
+ if (stmts)
+ {
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
+ gcc_assert (!new_bb);
+ }
+
+ /* Generate [VF*S, VF*S, ... ]. */
+ if (SCALAR_FLOAT_TYPE_P (TREE_TYPE (step_expr)))
+ {
+ expr = build_int_cst (integer_type_node, vf);
+ expr = fold_convert (TREE_TYPE (step_expr), expr);
+ }
+ else
+ expr = build_int_cst (TREE_TYPE (step_expr), vf);
+ new_name = fold_build2 (MULT_EXPR, TREE_TYPE (step_expr),
+ expr, step_expr);
+ if (! CONSTANT_CLASS_P (new_name))
+ new_name = vect_init_vector (phi, new_name,
+ TREE_TYPE (step_expr), NULL);
+ new_vec = build_vector_from_val (vectype, new_name);
+ vec_step = vect_init_vector (phi, new_vec, vectype, NULL);
+
+ /* Now generate the IVs. */
+ unsigned group_size = SLP_TREE_SCALAR_STMTS (slp_node).length ();
+ unsigned nvects = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
+ unsigned elts = nunits * nvects;
+ unsigned nivs = least_common_multiple (group_size, nunits) / nunits;
+ gcc_assert (elts % group_size == 0);
+ tree elt = init_expr;
+ unsigned ivn;
+ for (ivn = 0; ivn < nivs; ++ivn)
+ {
+ tree *elts = XALLOCAVEC (tree, nunits);
+ bool constant_p = true;
+ for (unsigned eltn = 0; eltn < nunits; ++eltn)
+ {
+ if (ivn*nunits + eltn >= group_size
+ && (ivn*nunits + eltn) % group_size == 0)
+ {
+ stmts = NULL;
+ elt = gimple_build (&stmts, PLUS_EXPR, TREE_TYPE (elt),
+ elt, step_expr);
+ if (stmts)
+ {
+ new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
+ gcc_assert (!new_bb);
+ }
+ }
+ if (! CONSTANT_CLASS_P (elt))
+ constant_p = false;
+ elts[eltn] = elt;
+ }
+ if (constant_p)
+ new_vec = build_vector (vectype, elts);
+ else
+ {
+ vec<constructor_elt, va_gc> *v;
+ vec_alloc (v, nunits);
+ for (i = 0; i < nunits; ++i)
+ CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, elts[i]);
+ new_vec = build_constructor (vectype, v);
+ }
+ vec_init = vect_init_vector (phi, new_vec, vectype, NULL);
+
+ /* Create the induction-phi that defines the induction-operand. */
+ vec_dest = vect_get_new_vect_var (vectype, vect_simple_var, "vec_iv_");
+ induction_phi = create_phi_node (vec_dest, iv_loop->header);
+ set_vinfo_for_stmt (induction_phi,
+ new_stmt_vec_info (induction_phi, loop_vinfo));
+ induc_def = PHI_RESULT (induction_phi);
+
+ /* Create the iv update inside the loop */
+ vec_def = make_ssa_name (vec_dest);
+ new_stmt = gimple_build_assign (vec_def, PLUS_EXPR, induc_def, vec_step);
+ gsi_insert_before (&si, new_stmt, GSI_SAME_STMT);
+ set_vinfo_for_stmt (new_stmt, new_stmt_vec_info (new_stmt, loop_vinfo));
+
+ /* Set the arguments of the phi node: */
+ add_phi_arg (induction_phi, vec_init, pe, UNKNOWN_LOCATION);
+ add_phi_arg (induction_phi, vec_def, loop_latch_edge (iv_loop),
+ UNKNOWN_LOCATION);
+
+ SLP_TREE_VEC_STMTS (slp_node).quick_push (induction_phi);
+ }
+
+ /* Re-use IVs when we can. */
+ if (ivn < nvects)
+ {
+ unsigned vfp
+ = least_common_multiple (group_size, nunits) / group_size;
+ /* Generate [VF'*S, VF'*S, ... ]. */
+ if (SCALAR_FLOAT_TYPE_P (TREE_TYPE (step_expr)))
+ {
+ expr = build_int_cst (integer_type_node, vfp);
+ expr = fold_convert (TREE_TYPE (step_expr), expr);
+ }
+ else
+ expr = build_int_cst (TREE_TYPE (step_expr), vfp);
+ new_name = fold_build2 (MULT_EXPR, TREE_TYPE (step_expr),
+ expr, step_expr);
+ if (! CONSTANT_CLASS_P (new_name))
+ new_name = vect_init_vector (phi, new_name,
+ TREE_TYPE (step_expr), NULL);
+ new_vec = build_vector_from_val (vectype, new_name);
+ vec_step = vect_init_vector (phi, new_vec, vectype, NULL);
+ for (; ivn < nvects; ++ivn)
+ {
+ gimple *iv = SLP_TREE_VEC_STMTS (slp_node)[ivn - nivs];
+ tree def;
+ if (gimple_code (iv) == GIMPLE_PHI)
+ def = gimple_phi_result (iv);
+ else
+ def = gimple_assign_lhs (iv);
+ new_stmt = gimple_build_assign (make_ssa_name (vectype),
+ PLUS_EXPR,
+ def, vec_step);
+ if (gimple_code (iv) == GIMPLE_PHI)
+ gsi_insert_before (&si, new_stmt, GSI_SAME_STMT);
+ else
+ {
+ gimple_stmt_iterator tgsi = gsi_for_stmt (iv);
+ gsi_insert_after (&tgsi, new_stmt, GSI_CONTINUE_LINKING);
+ }
+ set_vinfo_for_stmt (new_stmt,
+ new_stmt_vec_info (new_stmt, loop_vinfo));
+ SLP_TREE_VEC_STMTS (slp_node).quick_push (new_stmt);
+ }
+ }
+
+ return true;
+ }
+
/* Create the vector that holds the initial_value of the induction. */
if (nested_in_vect_loop)
{
@@ -6842,7 +7000,8 @@ vect_transform_loop (loop_vec_info loop_
&& dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "multiple-types.\n");
- if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def)
+ if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def
+ && ! PURE_SLP_STMT (stmt_info))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "transform phi.\n");
Index: gcc/tree-vect-slp.c
===================================================================
--- gcc/tree-vect-slp.c (revision 248825)
+++ gcc/tree-vect-slp.c (working copy)
@@ -100,6 +100,8 @@ vect_create_new_slp_node (vec<gimple *>
if (gimple_assign_rhs_code (stmt) == COND_EXPR)
nops++;
}
+ else if (gimple_code (stmt) == GIMPLE_PHI)
+ nops = 0;
else
return NULL;
@@ -401,9 +403,10 @@ again:
{
case vect_constant_def:
case vect_external_def:
- case vect_reduction_def:
+ case vect_reduction_def:
break;
+ case vect_induction_def:
case vect_internal_def:
oprnd_info->def_stmts.quick_push (def_stmt);
break;
@@ -935,9 +938,23 @@ vect_build_slp_tree (vec_info *vinfo,
if (gimple_assign_rhs_code (stmt) == COND_EXPR)
nops++;
}
+ else if (gimple_code (stmt) == GIMPLE_PHI)
+ nops = 0;
else
return NULL;
+ /* If the SLP node is a PHI (induction), terminate the recursion. */
+ if (gimple_code (stmt) == GIMPLE_PHI)
+ {
+ FOR_EACH_VEC_ELT (stmts, i, stmt)
+ if (stmt != stmts[0])
+ /* Induction from different IVs is not supported. */
+ return NULL;
+ node = vect_create_new_slp_node (stmts);
+ return node;
+ }
+
+
bool two_operators = false;
unsigned char *swap = XALLOCAVEC (unsigned char, group_size);
if (!vect_build_slp_tree_1 (vinfo, swap,
@@ -987,7 +1004,8 @@ vect_build_slp_tree (vec_info *vinfo,
unsigned old_tree_size = this_tree_size;
unsigned int j;
- if (oprnd_info->first_dt != vect_internal_def)
+ if (oprnd_info->first_dt != vect_internal_def
+ && oprnd_info->first_dt != vect_induction_def)
continue;
if (++this_tree_size > max_tree_size)
@@ -1611,6 +1629,28 @@ vect_analyze_slp_cost_1 (slp_instance in
return;
}
}
+ else if (STMT_VINFO_TYPE (stmt_info) == induc_vec_info_type)
+ {
+ /* ncopies_for_cost is the number of IVs we generate. */
+ record_stmt_cost (body_cost_vec, ncopies_for_cost, vector_stmt,
+ stmt_info, 0, vect_body);
+
+ /* Prologue cost for the initial values and step vector. */
+ record_stmt_cost (prologue_cost_vec, ncopies_for_cost,
+ CONSTANT_CLASS_P
+ (STMT_VINFO_LOOP_PHI_EVOLUTION_BASE_UNCHANGED
+ (stmt_info))
+ ? vector_load : vec_construct,
+ stmt_info, 0, vect_prologue);
+ record_stmt_cost (prologue_cost_vec, 1,
+ CONSTANT_CLASS_P
+ (STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_info))
+ ? vector_load : vec_construct,
+ stmt_info, 0, vect_prologue);
+
+ /* ??? No easy way to get at the actual number of vector stmts
+ to be geneated and thus the derived IVs. */
+ }
else
{
record_stmt_cost (body_cost_vec, ncopies_for_cost, vector_stmt,
@@ -2169,8 +2209,13 @@ vect_detect_hybrid_slp_stmts (slp_tree n
if (! STMT_VINFO_IN_PATTERN_P (stmt_vinfo)
&& STMT_VINFO_RELATED_STMT (stmt_vinfo))
stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
- if (TREE_CODE (gimple_op (stmt, 0)) == SSA_NAME)
- FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, gimple_op (stmt, 0))
+ tree def;
+ if (gimple_code (stmt) == GIMPLE_PHI)
+ def = gimple_phi_result (stmt);
+ else
+ def = SINGLE_SSA_TREE_OPERAND (stmt, SSA_OP_DEF);
+ if (def)
+ FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, def)
{
if (!flow_bb_inside_loop_p (loop, gimple_bb (use_stmt)))
continue;
@@ -3277,7 +3322,10 @@ vect_get_slp_vect_defs (slp_tree slp_nod
FOR_EACH_VEC_ELT (SLP_TREE_VEC_STMTS (slp_node), i, vec_def_stmt)
{
gcc_assert (vec_def_stmt);
- vec_oprnd = gimple_get_lhs (vec_def_stmt);
+ if (gimple_code (vec_def_stmt) == GIMPLE_PHI)
+ vec_oprnd = gimple_phi_result (vec_def_stmt);
+ else
+ vec_oprnd = gimple_get_lhs (vec_def_stmt);
vec_oprnds->quick_push (vec_oprnd);
}
}
@@ -3331,8 +3379,13 @@ vect_get_slp_defs (vec<tree> ops, slp_tr
gimple *first_def = SLP_TREE_SCALAR_STMTS (child)[0];
gimple *related
= STMT_VINFO_RELATED_STMT (vinfo_for_stmt (first_def));
+ tree first_def_op;
- if (operand_equal_p (oprnd, gimple_get_lhs (first_def), 0)
+ if (gimple_code (first_def) == GIMPLE_PHI)
+ first_def_op = gimple_phi_result (first_def);
+ else
+ first_def_op = gimple_get_lhs (first_def);
+ if (operand_equal_p (oprnd, first_def_op, 0)
|| (related
&& operand_equal_p (oprnd, gimple_get_lhs (related), 0)))
{
Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c (revision 248825)
+++ gcc/tree-vect-stmts.c (working copy)
@@ -8520,6 +8520,9 @@ vect_analyze_stmt (gimple *stmt, bool *n
break;
case vect_induction_def:
+ gcc_assert (!bb_vinfo);
+ break;
+
case vect_constant_def:
case vect_external_def:
case vect_unknown_def_type:
@@ -8598,6 +8601,7 @@ vect_analyze_stmt (gimple *stmt, bool *n
|| vectorizable_call (stmt, NULL, NULL, node)
|| vectorizable_store (stmt, NULL, NULL, node)
|| vectorizable_reduction (stmt, NULL, NULL, node)
+ || vectorizable_induction (stmt, NULL, NULL, node)
|| vectorizable_condition (stmt, NULL, NULL, NULL, 0, node)
|| vectorizable_comparison (stmt, NULL, NULL, NULL, node));
else
@@ -8681,8 +8685,7 @@ vect_transform_stmt (gimple *stmt, gimpl
break;
case induc_vec_info_type:
- gcc_assert (!slp_node);
- done = vectorizable_induction (stmt, gsi, &vec_stmt);
+ done = vectorizable_induction (stmt, gsi, &vec_stmt, slp_node);
gcc_assert (done);
break;
Index: gcc/tree-vectorizer.h
===================================================================
--- gcc/tree-vectorizer.h (revision 248825)
+++ gcc/tree-vectorizer.h (working copy)
@@ -1159,7 +1159,8 @@ extern bool vectorizable_live_operation
slp_tree, int, gimple **);
extern bool vectorizable_reduction (gimple *, gimple_stmt_iterator *,
gimple **, slp_tree);
-extern bool vectorizable_induction (gimple *, gimple_stmt_iterator *, gimple **);
+extern bool vectorizable_induction (gimple *, gimple_stmt_iterator *,
+ gimple **, slp_tree);
extern tree get_initial_def_for_reduction (gimple *, tree, tree *);
extern int vect_min_worthwhile_factor (enum tree_code);
extern int vect_get_known_peeling_cost (loop_vec_info, int, int *,
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH][4/4] SLP induction vectorization
2017-06-02 13:22 [PATCH][4/4] SLP induction vectorization Richard Biener
@ 2017-06-02 23:38 ` Michael Meissner
2017-06-03 7:32 ` Richard Biener
0 siblings, 1 reply; 5+ messages in thread
From: Michael Meissner @ 2017-06-02 23:38 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches
On Fri, Jun 02, 2017 at 03:22:27PM +0200, Richard Biener wrote:
>
> This implements vectorization of SLP inductions (in the not outer loop
> vectorization case for now).
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu.
>
> More testing is appreciated, I'm throwing it at SPEC2k6 now.
I was going to apply to the PowerPC and do a spec run but the patch for
tree-vect-loop.c doesn't apply. Could you regenerate the patch, and also tell
me the subversion id it is based off of?
Thanks in advance.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH][4/4] SLP induction vectorization
2017-06-02 23:38 ` Michael Meissner
@ 2017-06-03 7:32 ` Richard Biener
2017-06-06 7:38 ` Richard Biener
0 siblings, 1 reply; 5+ messages in thread
From: Richard Biener @ 2017-06-03 7:32 UTC (permalink / raw)
To: Michael Meissner; +Cc: gcc-patches
On June 3, 2017 1:38:14 AM GMT+02:00, Michael Meissner <meissner@linux.vnet.ibm.com> wrote:
>On Fri, Jun 02, 2017 at 03:22:27PM +0200, Richard Biener wrote:
>>
>> This implements vectorization of SLP inductions (in the not outer
>loop
>> vectorization case for now).
>>
>> Bootstrapped and tested on x86_64-unknown-linux-gnu.
>>
>> More testing is appreciated, I'm throwing it at SPEC2k6 now.
>
>I was going to apply to the PowerPC and do a spec run but the patch for
>tree-vect-loop.c doesn't apply. Could you regenerate the patch, and
>also tell
>me the subversion id it is based off of?
Hum, it should apply cleanly (maybe one rejected hunk that I already committed separately). I'll double check on Tuesday.
Richard.
>Thanks in advance.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH][4/4] SLP induction vectorization
2017-06-03 7:32 ` Richard Biener
@ 2017-06-06 7:38 ` Richard Biener
2017-06-07 4:17 ` Michael Meissner
0 siblings, 1 reply; 5+ messages in thread
From: Richard Biener @ 2017-06-06 7:38 UTC (permalink / raw)
To: Michael Meissner; +Cc: gcc-patches
On Sat, 3 Jun 2017, Richard Biener wrote:
> On June 3, 2017 1:38:14 AM GMT+02:00, Michael Meissner <meissner@linux.vnet.ibm.com> wrote:
> >On Fri, Jun 02, 2017 at 03:22:27PM +0200, Richard Biener wrote:
> >>
> >> This implements vectorization of SLP inductions (in the not outer
> >loop
> >> vectorization case for now).
> >>
> >> Bootstrapped and tested on x86_64-unknown-linux-gnu.
> >>
> >> More testing is appreciated, I'm throwing it at SPEC2k6 now.
> >
> >I was going to apply to the PowerPC and do a spec run but the patch for
> >tree-vect-loop.c doesn't apply. Could you regenerate the patch, and
> >also tell
> >me the subversion id it is based off of?
>
> Hum, it should apply cleanly (maybe one rejected hunk that I already
> committed separately). I'll double check on Tuesday.
My SPEC 2k6 testing went ok and I've committed the patch as r248909
now.
Richard.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH][4/4] SLP induction vectorization
2017-06-06 7:38 ` Richard Biener
@ 2017-06-07 4:17 ` Michael Meissner
0 siblings, 0 replies; 5+ messages in thread
From: Michael Meissner @ 2017-06-07 4:17 UTC (permalink / raw)
To: Richard Biener; +Cc: Michael Meissner, gcc-patches
On Tue, Jun 06, 2017 at 09:38:04AM +0200, Richard Biener wrote:
> On Sat, 3 Jun 2017, Richard Biener wrote:
>
> > On June 3, 2017 1:38:14 AM GMT+02:00, Michael Meissner <meissner@linux.vnet.ibm.com> wrote:
> > >On Fri, Jun 02, 2017 at 03:22:27PM +0200, Richard Biener wrote:
> > >>
> > >> This implements vectorization of SLP inductions (in the not outer
> > >loop
> > >> vectorization case for now).
> > >>
> > >> Bootstrapped and tested on x86_64-unknown-linux-gnu.
> > >>
> > >> More testing is appreciated, I'm throwing it at SPEC2k6 now.
> > >
> > >I was going to apply to the PowerPC and do a spec run but the patch for
> > >tree-vect-loop.c doesn't apply. Could you regenerate the patch, and
> > >also tell
> > >me the subversion id it is based off of?
> >
> > Hum, it should apply cleanly (maybe one rejected hunk that I already
> > committed separately). I'll double check on Tuesday.
>
> My SPEC 2k6 testing went ok and I've committed the patch as r248909
> now.
FWIW, I saw no significant differeneces in Spec 2006 CPU on a little endian
power8 system comparing spec built with GCC svn id 248908 and 248909 using the
same switches and libraries.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2017-06-07 4:17 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-06-02 13:22 [PATCH][4/4] SLP induction vectorization Richard Biener
2017-06-02 23:38 ` Michael Meissner
2017-06-03 7:32 ` Richard Biener
2017-06-06 7:38 ` Richard Biener
2017-06-07 4:17 ` Michael Meissner
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).