8080325: SuperWord loop unrolling analysis

Determine loop unroll factor based on supported vectors sizes.

Reviewed-by: roland, kvn
This commit is contained in:
Michael Berg 2015-06-16 16:10:36 -07:00
parent fd10da6423
commit 7c7b91845f
7 changed files with 300 additions and 54 deletions

View file

@ -68,6 +68,7 @@ SuperWord::SuperWord(PhaseIdealLoop* phase) :
_bb(NULL), // basic block
_iv(NULL), // induction var
_race_possible(false), // cases where SDMU is true
_early_return(true), // analysis evaluations routine
_num_work_vecs(0), // amount of vector work we have
_num_reductions(0), // amount of reduction work we have
_do_vector_loop(phase->C->do_vector_loop()), // whether to do vectorization/simd style
@ -78,7 +79,7 @@ SuperWord::SuperWord(PhaseIdealLoop* phase) :
{}
//------------------------------transform_loop---------------------------
void SuperWord::transform_loop(IdealLoopTree* lpt) {
void SuperWord::transform_loop(IdealLoopTree* lpt, bool do_optimization) {
assert(UseSuperWord, "should be");
// Do vectors exist on this architecture?
if (Matcher::vector_width_in_bytes(T_BYTE) < 2) return;
@ -113,8 +114,158 @@ void SuperWord::transform_loop(IdealLoopTree* lpt) {
// For now, define one block which is the entire loop body
set_bb(cl);
assert(_packset.length() == 0, "packset must be empty");
SLP_extract();
if (do_optimization) {
assert(_packset.length() == 0, "packset must be empty");
SLP_extract();
}
}
//------------------------------early unrolling analysis------------------------------
void SuperWord::unrolling_analysis(CountedLoopNode *cl, int &local_loop_unroll_factor) {
bool is_slp = true;
ResourceMark rm;
size_t ignored_size = lpt()->_body.size();
int *ignored_loop_nodes = NEW_RESOURCE_ARRAY(int, ignored_size);
Node_Stack nstack((int)ignored_size);
Node *cl_exit = cl->loopexit();
// First clear the entries
for (uint i = 0; i < lpt()->_body.size(); i++) {
ignored_loop_nodes[i] = -1;
}
int max_vector = Matcher::max_vector_size(T_INT);
// Process the loop, some/all of the stack entries will not be in order, ergo
// need to preprocess the ignored initial state before we process the loop
for (uint i = 0; i < lpt()->_body.size(); i++) {
Node* n = lpt()->_body.at(i);
if (n == cl->incr() ||
n->is_reduction() ||
n->is_AddP() ||
n->is_Cmp() ||
n->is_IfTrue() ||
n->is_CountedLoop() ||
(n == cl_exit)) {
ignored_loop_nodes[i] = n->_idx;
continue;
}
if (n->is_If()) {
IfNode *iff = n->as_If();
if (iff->_fcnt != COUNT_UNKNOWN && iff->_prob != PROB_UNKNOWN) {
if (lpt()->is_loop_exit(iff)) {
ignored_loop_nodes[i] = n->_idx;
continue;
}
}
}
if (n->is_Phi() && (n->bottom_type() == Type::MEMORY)) {
Node* n_tail = n->in(LoopNode::LoopBackControl);
if (n_tail != n->in(LoopNode::EntryControl)) {
if (!n_tail->is_Mem()) {
is_slp = false;
break;
}
}
}
// This must happen after check of phi/if
if (n->is_Phi() || n->is_If()) {
ignored_loop_nodes[i] = n->_idx;
continue;
}
if (n->is_LoadStore() || n->is_MergeMem() ||
(n->is_Proj() && !n->as_Proj()->is_CFG())) {
is_slp = false;
break;
}
if (n->is_Mem()) {
Node* adr = n->in(MemNode::Address);
Node* n_ctrl = _phase->get_ctrl(adr);
// save a queue of post process nodes
if (n_ctrl != NULL && lpt()->is_member(_phase->get_loop(n_ctrl))) {
MemNode* current = n->as_Mem();
BasicType bt = current->memory_type();
if (is_java_primitive(bt) == false) {
ignored_loop_nodes[i] = n->_idx;
continue;
}
// Process the memory expression
int stack_idx = 0;
bool have_side_effects = true;
if (adr->is_AddP() == false) {
nstack.push(adr, stack_idx++);
} else {
// Mark the components of the memory operation in nstack
SWPointer p1(current, this, &nstack, true);
have_side_effects = p1.node_stack()->is_nonempty();
}
// Process the pointer stack
while (have_side_effects) {
Node* pointer_node = nstack.node();
for (uint j = 0; j < lpt()->_body.size(); j++) {
Node* cur_node = lpt()->_body.at(j);
if (cur_node == pointer_node) {
ignored_loop_nodes[j] = cur_node->_idx;
break;
}
}
nstack.pop();
have_side_effects = nstack.is_nonempty();
}
}
}
}
if (is_slp) {
// Now we try to find the maximum supported consistent vector which the machine
// description can use
for (uint i = 0; i < lpt()->_body.size(); i++) {
if (ignored_loop_nodes[i] != -1) continue;
BasicType bt;
Node* n = lpt()->_body.at(i);
if (n->is_Store()) {
bt = n->as_Mem()->memory_type();
}
else {
bt = n->bottom_type()->basic_type();
}
int cur_max_vector = Matcher::max_vector_size(bt);
// If a max vector exists which is not larger than _local_loop_unroll_factor
// stop looking, we already have the max vector to map to.
if (cur_max_vector <= local_loop_unroll_factor) {
is_slp = false;
#ifndef PRODUCT
if (TraceSuperWordLoopUnrollAnalysis) {
tty->print_cr("slp analysis fails: unroll limit equals max vector\n");
}
#endif
break;
}
// Map the maximal common vector
if (VectorNode::implemented(n->Opcode(), cur_max_vector, bt)) {
if (cur_max_vector < max_vector) {
max_vector = cur_max_vector;
}
}
}
if (is_slp) {
local_loop_unroll_factor = max_vector;
}
cl->mark_passed_slp();
cl->set_slp_max_unroll(local_loop_unroll_factor);
}
}
//------------------------------SLP_extract---------------------------
@ -268,12 +419,12 @@ void SuperWord::find_adjacent_refs() {
best_iv_adjustment = iv_adjustment;
}
SWPointer align_to_ref_p(mem_ref, this);
SWPointer align_to_ref_p(mem_ref, this, NULL, false);
// Set alignment relative to "align_to_ref" for all related memory operations.
for (int i = memops.size() - 1; i >= 0; i--) {
MemNode* s = memops.at(i)->as_Mem();
if (isomorphic(s, mem_ref)) {
SWPointer p2(s, this);
SWPointer p2(s, this, NULL, false);
if (p2.comparable(align_to_ref_p)) {
int align = memory_alignment(s, iv_adjustment);
set_alignment(s, align);
@ -294,7 +445,7 @@ void SuperWord::find_adjacent_refs() {
// iterations in pre-loop will be not enough to align it.
create_pack = false;
} else {
SWPointer p2(best_align_to_mem_ref, this);
SWPointer p2(best_align_to_mem_ref, this, NULL, false);
if (align_to_ref_p.invar() != p2.invar()) {
// Do not vectorize memory accesses with different invariants
// if unaligned memory accesses are not allowed.
@ -411,7 +562,7 @@ MemNode* SuperWord::find_align_to_ref(Node_List &memops) {
// Count number of comparable memory ops
for (uint i = 0; i < memops.size(); i++) {
MemNode* s1 = memops.at(i)->as_Mem();
SWPointer p1(s1, this);
SWPointer p1(s1, this, NULL, false);
// Discard if pre loop can't align this reference
if (!ref_is_alignable(p1)) {
*cmp_ct.adr_at(i) = 0;
@ -420,7 +571,7 @@ MemNode* SuperWord::find_align_to_ref(Node_List &memops) {
for (uint j = i+1; j < memops.size(); j++) {
MemNode* s2 = memops.at(j)->as_Mem();
if (isomorphic(s1, s2)) {
SWPointer p2(s2, this);
SWPointer p2(s2, this, NULL, false);
if (p1.comparable(p2)) {
(*cmp_ct.adr_at(i))++;
(*cmp_ct.adr_at(j))++;
@ -441,7 +592,7 @@ MemNode* SuperWord::find_align_to_ref(Node_List &memops) {
if (s->is_Store()) {
int vw = vector_width_in_bytes(s);
assert(vw > 1, "sanity");
SWPointer p(s, this);
SWPointer p(s, this, NULL, false);
if (cmp_ct.at(j) > max_ct ||
cmp_ct.at(j) == max_ct &&
(vw > max_vw ||
@ -464,7 +615,7 @@ MemNode* SuperWord::find_align_to_ref(Node_List &memops) {
if (s->is_Load()) {
int vw = vector_width_in_bytes(s);
assert(vw > 1, "sanity");
SWPointer p(s, this);
SWPointer p(s, this, NULL, false);
if (cmp_ct.at(j) > max_ct ||
cmp_ct.at(j) == max_ct &&
(vw > max_vw ||
@ -575,7 +726,7 @@ bool SuperWord::ref_is_alignable(SWPointer& p) {
//---------------------------get_iv_adjustment---------------------------
// Calculate loop's iv adjustment for this memory ops.
int SuperWord::get_iv_adjustment(MemNode* mem_ref) {
SWPointer align_to_ref_p(mem_ref, this);
SWPointer align_to_ref_p(mem_ref, this, NULL, false);
int offset = align_to_ref_p.offset_in_bytes();
int scale = align_to_ref_p.scale_in_bytes();
int elt_size = align_to_ref_p.memory_size();
@ -649,13 +800,13 @@ void SuperWord::dependence_graph() {
if (_dg.dep(s1)->in_cnt() == 0) {
_dg.make_edge(slice, s1);
}
SWPointer p1(s1->as_Mem(), this);
SWPointer p1(s1->as_Mem(), this, NULL, false);
bool sink_dependent = true;
for (int k = j - 1; k >= 0; k--) {
Node* s2 = _nlist.at(k);
if (s1->is_Load() && s2->is_Load())
continue;
SWPointer p2(s2->as_Mem(), this);
SWPointer p2(s2->as_Mem(), this, NULL, false);
int cmp = p1.cmp(p2);
if (SuperWordRTDepCheck &&
@ -795,8 +946,8 @@ bool SuperWord::are_adjacent_refs(Node* s1, Node* s2) {
if (_phase->C->get_alias_index(s1->as_Mem()->adr_type()) !=
_phase->C->get_alias_index(s2->as_Mem()->adr_type()))
return false;
SWPointer p1(s1->as_Mem(), this);
SWPointer p2(s2->as_Mem(), this);
SWPointer p1(s1->as_Mem(), this, NULL, false);
SWPointer p2(s2->as_Mem(), this, NULL, false);
if (p1.base() != p2.base() || !p1.comparable(p2)) return false;
int diff = p2.offset_in_bytes() - p1.offset_in_bytes();
return diff == data_size(s1);
@ -1615,13 +1766,13 @@ void SuperWord::output() {
if (n->is_Load()) {
Node* ctl = n->in(MemNode::Control);
Node* mem = first->in(MemNode::Memory);
SWPointer p1(n->as_Mem(), this);
SWPointer p1(n->as_Mem(), this, NULL, false);
// Identify the memory dependency for the new loadVector node by
// walking up through memory chain.
// This is done to give flexibility to the new loadVector node so that
// it can move above independent storeVector nodes.
while (mem->is_StoreVector()) {
SWPointer p2(mem->as_Mem(), this);
SWPointer p2(mem->as_Mem(), this, NULL, false);
int cmp = p1.cmp(p2);
if (SWPointer::not_equal(cmp) || !SWPointer::comparable(cmp)) {
mem = mem->in(MemNode::Memory);
@ -2138,7 +2289,7 @@ void SuperWord::compute_vector_element_type() {
//------------------------------memory_alignment---------------------------
// Alignment within a vector memory reference
int SuperWord::memory_alignment(MemNode* s, int iv_adjust) {
SWPointer p(s, this);
SWPointer p(s, this, NULL, false);
if (!p.valid()) {
return bottom_align;
}
@ -2315,7 +2466,7 @@ void SuperWord::align_initial_loop_index(MemNode* align_to_ref) {
Node *orig_limit = pre_opaq->original_loop_limit();
assert(orig_limit != NULL && _igvn.type(orig_limit) != Type::TOP, "");
SWPointer align_to_ref_p(align_to_ref, this);
SWPointer align_to_ref_p(align_to_ref, this, NULL, false);
assert(align_to_ref_p.valid(), "sanity");
// Given:
@ -2489,6 +2640,7 @@ void SuperWord::init() {
_bb = NULL;
_iv = NULL;
_race_possible = 0;
_early_return = false;
_num_work_vecs = 0;
_num_reductions = 0;
}
@ -2559,9 +2711,11 @@ char* SuperWord::blank(uint depth) {
//==============================SWPointer===========================
//----------------------------SWPointer------------------------
SWPointer::SWPointer(MemNode* mem, SuperWord* slp) :
SWPointer::SWPointer(MemNode* mem, SuperWord* slp, Node_Stack *nstack, bool analyze_only) :
_mem(mem), _slp(slp), _base(NULL), _adr(NULL),
_scale(0), _offset(0), _invar(NULL), _negate_invar(false) {
_scale(0), _offset(0), _invar(NULL), _negate_invar(false),
_nstack(nstack), _analyze_only(analyze_only),
_stack_idx(0) {
Node* adr = mem->in(MemNode::Address);
if (!adr->is_AddP()) {
@ -2599,7 +2753,9 @@ SWPointer::SWPointer(MemNode* mem, SuperWord* slp) :
// the pattern match of an address expression.
SWPointer::SWPointer(SWPointer* p) :
_mem(p->_mem), _slp(p->_slp), _base(NULL), _adr(NULL),
_scale(0), _offset(0), _invar(NULL), _negate_invar(false) {}
_scale(0), _offset(0), _invar(NULL), _negate_invar(false),
_nstack(p->_nstack), _analyze_only(p->_analyze_only),
_stack_idx(p->_stack_idx) {}
//------------------------scaled_iv_plus_offset--------------------
// Match: k*iv + offset
@ -2642,6 +2798,9 @@ bool SWPointer::scaled_iv(Node* n) {
_scale = 1;
return true;
}
if (_analyze_only && (invariant(n) == false)) {
_nstack->push(n, _stack_idx++);
}
int opc = n->Opcode();
if (opc == Op_MulI) {
if (n->in(1) == iv() && n->in(2)->is_Con()) {
@ -2699,6 +2858,9 @@ bool SWPointer::offset_plus_k(Node* n, bool negate) {
return false;
}
if (_invar != NULL) return false; // already have an invariant
if (_analyze_only && (invariant(n) == false)) {
_nstack->push(n, _stack_idx++);
}
if (opc == Op_AddI) {
if (n->in(2)->is_Con() && invariant(n->in(1))) {
_negate_invar = negate;