stan/jdk-24
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

8215483: Off heap memory accesses should be vectorized

Browse Source 
Reviewed-by: neliasso, kvn
This commit is contained in:
Roland Westrelin 2018-12-18 09:29:39 +01:00
parent 791ab38fcd
commit 6c640e1831
4 changed files with 22 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/hotspot/cpu/aarch64/aarch64.ad
View File

@ -2146,9 +2146,9 @@ const bool Matcher::pass_original_key_for_aes() {
return false;
}
// x86 supports misaligned vectors store/load.
// aarch64 supports misaligned vectors store/load.
const bool Matcher::misaligned_vectors_ok() {
return !AlignVector; // can be changed by flag
return true;
}
// false => size gets scaled to BytesPerLong, ok.

5
src/hotspot/cpu/ppc/ppc.ad
View File

@ -2338,9 +2338,10 @@ const int Matcher::min_vector_size(const BasicType bt) {
return max_vector_size(bt); // Same as max.
}
// PPC doesn't support misaligned vectors store/load.
// PPC implementation uses VSX load/store instructions (if
// SuperwordUseVSX) which support 4 byte but not arbitrary alignment
const bool Matcher::misaligned_vectors_ok() {
return !AlignVector; // can be changed by flag
return false;
}
// PPC AES support not yet implemented

2
src/hotspot/cpu/x86/x86.ad
View File

@ -1585,7 +1585,7 @@ const uint Matcher::vector_shift_count_ideal_reg(int size) {
// x86 supports misaligned vectors store/load.
const bool Matcher::misaligned_vectors_ok() {
return !AlignVector; // can be changed by flag
return true;
}
// x86 AES instructions are compatible with SunJCE expanded

22
src/hotspot/share/opto/superword.cpp
View File

@ -613,7 +613,7 @@ void SuperWord::find_adjacent_refs() {
// alignment is set and vectors will be aligned.
bool create_pack = true;
if (memory_alignment(mem_ref, best_iv_adjustment) == 0 || _do_vector_loop) {
if (!Matcher::misaligned_vectors_ok()) {
if (!Matcher::misaligned_vectors_ok() || AlignVector) {
int vw = vector_width(mem_ref);
int vw_best = vector_width(best_align_to_mem_ref);
if (vw > vw_best) {
@ -638,7 +638,7 @@ void SuperWord::find_adjacent_refs() {
} else {
// Allow independent (different type) unaligned memory operations
// if HW supports them.
if (!Matcher::misaligned_vectors_ok()) {
if (!Matcher::misaligned_vectors_ok() || AlignVector) {
create_pack = false;
} else {
// Check if packs of the same memory type but
@ -3372,9 +3372,9 @@ void SuperWord::align_initial_loop_index(MemNode* align_to_ref) {
_igvn.register_new_node_with_optimizer(e);
_phase->set_ctrl(e, pre_ctrl);
}
if (vw > ObjectAlignmentInBytes) {
if (vw > ObjectAlignmentInBytes || align_to_ref_p.base()->is_top()) {
// incorporate base e +/- base && Mask >>> log2(elt)
Node* xbase = new CastP2XNode(NULL, align_to_ref_p.base());
Node* xbase = new CastP2XNode(NULL, align_to_ref_p.adr());
_igvn.register_new_node_with_optimizer(xbase);
#ifdef _LP64
xbase = new ConvL2INode(xbase);
@ -3566,8 +3566,8 @@ SWPointer::SWPointer(MemNode* mem, SuperWord* slp, Node_Stack *nstack, bool anal
assert(!valid(), "base address is loop variant");
return;
}
//unsafe reference could not be aligned appropriately without runtime checking
if (base == NULL || base->bottom_type() == Type::TOP) {
// unsafe references require misaligned vector access support
if (base->is_top() && !Matcher::misaligned_vectors_ok()) {
assert(!valid(), "unsafe access");
return;
}
@ -3591,6 +3591,16 @@ SWPointer::SWPointer(MemNode* mem, SuperWord* slp, Node_Stack *nstack, bool anal
break; // stop looking at addp's
}
}
if (!invariant(adr)) {
assert(!valid(), "adr is loop variant");
return;
}
if (!base->is_top() && adr != base) {
assert(!valid(), "adr and base differ");
return;
}
NOT_PRODUCT(if(_slp->is_trace_alignment()) _tracer.restore_depth();)
NOT_PRODUCT(_tracer.ctor_6(mem);)