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

8217561: X86: Add floating-point Math.min/max intrinsics

Browse Source 
Implementation taking care of +/-0.0 and NaN which uses a specific pattern for reductions

Reviewed-by: aph, kvn, neliasso, sviswanathan, adinn
This commit is contained in:
Jatin Bhateja 2019-03-07 15:27:42 +01:00 committed by Bernard Blaser
parent 367ae10733
commit c2ec1085e1
9 changed files with 718 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
src/hotspot/cpu/x86/assembler_x86.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@ -7765,9 +7765,43 @@ int Assembler::simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegis
}
}
void Assembler::vmaxss(XMMRegister dst, XMMRegister nds, XMMRegister src) {
assert(VM_Version::supports_avx(), "");
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_F3, VEX_OPCODE_0F, &attributes);
emit_int8(0x5F);
emit_int8((unsigned char)(0xC0 | encode));
}
void Assembler::vmaxsd(XMMRegister dst, XMMRegister nds, XMMRegister src) {
assert(VM_Version::supports_avx(), "");
InstructionAttr attributes(AVX_128bit, /* vex_w */ VM_Version::supports_evex(), /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
attributes.set_rex_vex_w_reverted();
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
emit_int8(0x5F);
emit_int8((unsigned char)(0xC0 | encode));
}
void Assembler::vminss(XMMRegister dst, XMMRegister nds, XMMRegister src) {
assert(VM_Version::supports_avx(), "");
InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_F3, VEX_OPCODE_0F, &attributes);
emit_int8(0x5D);
emit_int8((unsigned char)(0xC0 | encode));
}
void Assembler::vminsd(XMMRegister dst, XMMRegister nds, XMMRegister src) {
assert(VM_Version::supports_avx(), "");
InstructionAttr attributes(AVX_128bit, /* vex_w */ VM_Version::supports_evex(), /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
attributes.set_rex_vex_w_reverted();
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
emit_int8(0x5D);
emit_int8((unsigned char)(0xC0 | encode));
}
void Assembler::cmppd(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop, int vector_len) {
assert(VM_Version::supports_avx(), "");
assert(!VM_Version::supports_evex(), "");
assert(vector_len <= AVX_256bit, "");
InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);
int encode = simd_prefix_and_encode(dst, nds, src, VEX_SIMD_66, VEX_OPCODE_0F, &attributes);
emit_int8((unsigned char)0xC2);
@ -7777,7 +7811,7 @@ void Assembler::cmppd(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop
void Assembler::blendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMMRegister src2, int vector_len) {
assert(VM_Version::supports_avx(), "");
assert(!VM_Version::supports_evex(), "");
assert(vector_len <= AVX_256bit, "");
InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src1->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_3A, &attributes);
emit_int8((unsigned char)0x4B);
@ -7788,7 +7822,7 @@ void Assembler::blendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMM
void Assembler::cmpps(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop, int vector_len) {
assert(VM_Version::supports_avx(), "");
assert(!VM_Version::supports_evex(), "");
assert(vector_len <= AVX_256bit, "");
InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);
int encode = simd_prefix_and_encode(dst, nds, src, VEX_SIMD_NONE, VEX_OPCODE_0F, &attributes);
emit_int8((unsigned char)0xC2);
@ -7798,7 +7832,7 @@ void Assembler::cmpps(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop
void Assembler::blendvps(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMMRegister src2, int vector_len) {
assert(VM_Version::supports_avx(), "");
assert(!VM_Version::supports_evex(), "");
assert(vector_len <= AVX_256bit, "");
InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);
int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src1->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_3A, &attributes);
emit_int8((unsigned char)0x4A);

5
src/hotspot/cpu/x86/assembler_x86.hpp
View File

@ -1934,6 +1934,11 @@ private:
void vsubss(XMMRegister dst, XMMRegister nds, Address src);
void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src);
void vmaxss(XMMRegister dst, XMMRegister nds, XMMRegister src);
void vmaxsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
void vminss(XMMRegister dst, XMMRegister nds, XMMRegister src);
void vminsd(XMMRegister dst, XMMRegister nds, XMMRegister src);
void shlxl(Register dst, Register src1, Register src2);
void shlxq(Register dst, Register src1, Register src2);

2
src/hotspot/cpu/x86/macroAssembler_x86.hpp
View File

@ -165,6 +165,7 @@ class MacroAssembler: public Assembler {
// Support optimal SSE move instructions.
void movflt(XMMRegister dst, XMMRegister src) {
if (dst-> encoding() == src->encoding()) return;
if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
else { movss (dst, src); return; }
}
@ -173,6 +174,7 @@ class MacroAssembler: public Assembler {
void movflt(Address dst, XMMRegister src) { movss(dst, src); }
void movdbl(XMMRegister dst, XMMRegister src) {
if (dst-> encoding() == src->encoding()) return;
if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
else { movsd (dst, src); return; }
}

11
src/hotspot/cpu/x86/x86.ad
View File

@ -1,5 +1,5 @@
//
// Copyright (c) 2011, 2018, Oracle and/or its affiliates. All rights reserved.
// Copyright (c) 2011, 2019, Oracle and/or its affiliates. All rights reserved.
// DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
//
// This code is free software; you can redistribute it and/or modify it
@ -1450,6 +1450,15 @@ const bool Matcher::match_rule_supported(int opcode) {
if (UseSSE < 2)
ret_value = false;
break;
#ifdef _LP64
case Op_MaxD:
case Op_MaxF:
case Op_MinD:
case Op_MinF:
if (UseAVX < 1) // enabled for AVX only
ret_value = false;
break;
#endif
}
return ret_value; // Per default match rules are supported.

290
src/hotspot/cpu/x86/x86_64.ad
View File

@ -698,6 +698,87 @@ void emit_cmpfp3(MacroAssembler& _masm, Register dst) {
__ bind(done);
}
// Math.min() # Math.max()
// --------------------------
// ucomis[s/d] #
// ja -> b # a
// jp -> NaN # NaN
// jb -> a # b
// je #
// |-jz -> a | b # a & b
// | -> a #
void emit_fp_min_max(MacroAssembler& _masm, XMMRegister dst,
XMMRegister a, XMMRegister b,
XMMRegister xmmt, Register rt,
bool min, bool single) {
Label nan, zero, below, above, done;
if (single)
__ ucomiss(a, b);
else
__ ucomisd(a, b);
if (dst->encoding() != (min ? b : a)->encoding())
__ jccb(Assembler::above, above); // CF=0 & ZF=0
else
__ jccb(Assembler::above, done);
__ jccb(Assembler::parity, nan); // PF=1
__ jccb(Assembler::below, below); // CF=1
// equal
__ vpxor(xmmt, xmmt, xmmt, Assembler::AVX_128bit);
if (single) {
__ ucomiss(a, xmmt);
__ jccb(Assembler::equal, zero);
__ movflt(dst, a);
__ jmp(done);
}
else {
__ ucomisd(a, xmmt);
__ jccb(Assembler::equal, zero);
__ movdbl(dst, a);
__ jmp(done);
}
__ bind(zero);
if (min)
__ vpor(dst, a, b, Assembler::AVX_128bit);
else
__ vpand(dst, a, b, Assembler::AVX_128bit);
__ jmp(done);
__ bind(above);
if (single)
__ movflt(dst, min ? b : a);
else
__ movdbl(dst, min ? b : a);
__ jmp(done);
__ bind(nan);
if (single) {
__ movl(rt, 0x7fc00000); // Float.NaN
__ movdl(dst, rt);
}
else {
__ mov64(rt, 0x7ff8000000000000L); // Double.NaN
__ movdq(dst, rt);
}
__ jmp(done);
__ bind(below);
if (single)
__ movflt(dst, min ? a : b);
else
__ movdbl(dst, min ? a : b);
__ bind(done);
}
//=============================================================================
const RegMask& MachConstantBaseNode::_out_RegMask = RegMask::Empty;
@ -3547,6 +3628,15 @@ operand regF() %{
interface(REG_INTER);
%}
// Float register operands
operand legRegF() %{
constraint(ALLOC_IN_RC(float_reg_legacy));
match(RegF);
format %{ %}
interface(REG_INTER);
%}
// Float register operands
operand vlRegF() %{
constraint(ALLOC_IN_RC(float_reg_vl));
@ -3565,6 +3655,15 @@ operand regD() %{
interface(REG_INTER);
%}
// Double register operands
operand legRegD() %{
constraint(ALLOC_IN_RC(double_reg_legacy));
match(RegD);
format %{ %}
interface(REG_INTER);
%}
// Double register operands
operand vlRegD() %{
constraint(ALLOC_IN_RC(double_reg_vl));
@ -5303,6 +5402,16 @@ instruct MoveF2VL(vlRegF dst, regF src) %{
ins_pipe( fpu_reg_reg );
%}
// Load Float
instruct MoveF2LEG(legRegF dst, regF src) %{
match(Set dst src);
format %{ "movss $dst,$src\t# if src != dst load float (4 bytes)" %}
ins_encode %{
__ movflt($dst$$XMMRegister, $src$$XMMRegister);
%}
ins_pipe( fpu_reg_reg );
%}
// Load Float
instruct MoveVL2F(regF dst, vlRegF src) %{
match(Set dst src);
@ -5313,6 +5422,16 @@ instruct MoveVL2F(regF dst, vlRegF src) %{
ins_pipe( fpu_reg_reg );
%}
// Load Float
instruct MoveLEG2F(regF dst, legRegF src) %{
match(Set dst src);
format %{ "movss $dst,$src\t# if src != dst load float (4 bytes)" %}
ins_encode %{
__ movflt($dst$$XMMRegister, $src$$XMMRegister);
%}
ins_pipe( fpu_reg_reg );
%}
// Load Double
instruct loadD_partial(regD dst, memory mem)
%{
@ -5350,6 +5469,16 @@ instruct MoveD2VL(vlRegD dst, regD src) %{
ins_pipe( fpu_reg_reg );
%}
// Load Double
instruct MoveD2LEG(legRegD dst, regD src) %{
match(Set dst src);
format %{ "movsd $dst,$src\t# if src != dst load double (8 bytes)" %}
ins_encode %{
__ movdbl($dst$$XMMRegister, $src$$XMMRegister);
%}
ins_pipe( fpu_reg_reg );
%}
// Load Double
instruct MoveVL2D(regD dst, vlRegD src) %{
match(Set dst src);
@ -5360,6 +5489,167 @@ instruct MoveVL2D(regD dst, vlRegD src) %{
ins_pipe( fpu_reg_reg );
%}
// Load Double
instruct MoveLEG2D(regD dst, legRegD src) %{
match(Set dst src);
format %{ "movsd $dst,$src\t# if src != dst load double (8 bytes)" %}
ins_encode %{
__ movdbl($dst$$XMMRegister, $src$$XMMRegister);
%}
ins_pipe( fpu_reg_reg );
%}
// Following pseudo code describes the algorithm for max[FD]:
// Min algorithm is on similar lines
// btmp = (b < +0.0) ? a : b
// atmp = (b < +0.0) ? b : a
// Tmp = Max_Float(atmp , btmp)
// Res = (atmp == NaN) ? atmp : Tmp
// max = java.lang.Math.max(float a, float b)
instruct maxF_reg(legRegF dst, legRegF a, legRegF b, legRegF tmp, legRegF atmp, legRegF btmp) %{
predicate(UseAVX > 0 && !n->is_reduction());
match(Set dst (MaxF a b));
effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
format %{
"blendvps $btmp,$b,$a,$b \n\t"
"blendvps $atmp,$a,$b,$b \n\t"
"vmaxss $tmp,$atmp,$btmp \n\t"
"cmpps.unordered $btmp,$atmp,$atmp \n\t"
"blendvps $dst,$tmp,$atmp,$btmp \n\t"
%}
ins_encode %{
int vector_len = Assembler::AVX_128bit;
__ blendvps($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, vector_len);
__ blendvps($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $b$$XMMRegister, vector_len);
__ vmaxss($tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister);
__ cmpps($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, Assembler::_false, vector_len);
__ blendvps($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
%}
ins_pipe( pipe_slow );
%}
instruct maxF_reduction_reg(regF dst, regF a, regF b, regF xmmt, rRegI tmp, rFlagsReg cr) %{
predicate(UseAVX > 0 && n->is_reduction());
match(Set dst (MaxF a b));
effect(USE a, USE b, TEMP xmmt, TEMP tmp, KILL cr);
format %{ "$dst = max($a, $b)\t# intrinsic (float)" %}
ins_encode %{
emit_fp_min_max(_masm, $dst$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $xmmt$$XMMRegister, $tmp$$Register,
false /*min*/, true /*single*/);
%}
ins_pipe( pipe_slow );
%}
// max = java.lang.Math.max(double a, double b)
instruct maxD_reg(legRegD dst, legRegD a, legRegD b, legRegD tmp, legRegD atmp, legRegD btmp) %{
predicate(UseAVX > 0 && !n->is_reduction());
match(Set dst (MaxD a b));
effect(USE a, USE b, TEMP atmp, TEMP btmp, TEMP tmp);
format %{
"blendvpd $btmp,$b,$a,$b \n\t"
"blendvpd $atmp,$a,$b,$b \n\t"
"vmaxsd $tmp,$atmp,$btmp \n\t"
"cmppd.unordered $btmp,$atmp,$atmp \n\t"
"blendvpd $dst,$tmp,$atmp,$btmp \n\t"
%}
ins_encode %{
int vector_len = Assembler::AVX_128bit;
__ blendvpd($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, vector_len);
__ blendvpd($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $b$$XMMRegister, vector_len);
__ vmaxsd($tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister);
__ cmppd($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, Assembler::_false, vector_len);
__ blendvpd($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
%}
ins_pipe( pipe_slow );
%}
instruct maxD_reduction_reg(regD dst, regD a, regD b, regD xmmt, rRegL tmp, rFlagsReg cr) %{
predicate(UseAVX > 0 && n->is_reduction());
match(Set dst (MaxD a b));
effect(USE a, USE b, TEMP xmmt, TEMP tmp, KILL cr);
format %{ "$dst = max($a, $b)\t# intrinsic (double)" %}
ins_encode %{
emit_fp_min_max(_masm, $dst$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $xmmt$$XMMRegister, $tmp$$Register,
false /*min*/, false /*single*/);
%}
ins_pipe( pipe_slow );
%}
// min = java.lang.Math.min(float a, float b)
instruct minF_reg(legRegF dst, legRegF a, legRegF b, legRegF tmp, legRegF atmp, legRegF btmp) %{
predicate(UseAVX > 0 && !n->is_reduction());
match(Set dst (MinF a b));
effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
format %{
"blendvps $atmp,$a,$b,$a \n\t"
"blendvps $btmp,$b,$a,$a \n\t"
"vminss $tmp,$atmp,$btmp \n\t"
"cmpps.unordered $btmp,$atmp,$atmp \n\t"
"blendvps $dst,$tmp,$atmp,$btmp \n\t"
%}
ins_encode %{
int vector_len = Assembler::AVX_128bit;
__ blendvps($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, vector_len);
__ blendvps($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $a$$XMMRegister, vector_len);
__ vminss($tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister);
__ cmpps($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, Assembler::_false, vector_len);
__ blendvps($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
%}
ins_pipe( pipe_slow );
%}
instruct minF_reduction_reg(regF dst, regF a, regF b, regF xmmt, rRegI tmp, rFlagsReg cr) %{
predicate(UseAVX > 0 && n->is_reduction());
match(Set dst (MinF a b));
effect(USE a, USE b, TEMP xmmt, TEMP tmp, KILL cr);
format %{ "$dst = min($a, $b)\t# intrinsic (float)" %}
ins_encode %{
emit_fp_min_max(_masm, $dst$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $xmmt$$XMMRegister, $tmp$$Register,
true /*min*/, true /*single*/);
%}
ins_pipe( pipe_slow );
%}
// min = java.lang.Math.min(double a, double b)
instruct minD_reg(legRegD dst, legRegD a, legRegD b, legRegD tmp, legRegD atmp, legRegD btmp) %{
predicate(UseAVX > 0 && !n->is_reduction());
match(Set dst (MinD a b));
effect(USE a, USE b, TEMP tmp, TEMP atmp, TEMP btmp);
format %{
"blendvpd $atmp,$a,$b,$a \n\t"
"blendvpd $btmp,$b,$a,$a \n\t"
"vminsd $tmp,$atmp,$btmp \n\t"
"cmppd.unordered $btmp,$atmp,$atmp \n\t"
"blendvpd $dst,$tmp,$atmp,$btmp \n\t"
%}
ins_encode %{
int vector_len = Assembler::AVX_128bit;
__ blendvpd($atmp$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, vector_len);
__ blendvpd($btmp$$XMMRegister, $b$$XMMRegister, $a$$XMMRegister, $a$$XMMRegister, vector_len);
__ vminsd($tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister);
__ cmppd($btmp$$XMMRegister, $atmp$$XMMRegister, $atmp$$XMMRegister, Assembler::_false, vector_len);
__ blendvpd($dst$$XMMRegister, $tmp$$XMMRegister, $atmp$$XMMRegister, $btmp$$XMMRegister, vector_len);
%}
ins_pipe( pipe_slow );
%}
instruct minD_reduction_reg(regD dst, regD a, regD b, regD xmmt, rRegL tmp, rFlagsReg cr) %{
predicate(UseAVX > 0 && n->is_reduction());
match(Set dst (MinD a b));
effect(USE a, USE b, TEMP xmmt, TEMP tmp, KILL cr);
format %{ "$dst = min($a, $b)\t# intrinsic (double)" %}
ins_encode %{
emit_fp_min_max(_masm, $dst$$XMMRegister, $a$$XMMRegister, $b$$XMMRegister, $xmmt$$XMMRegister, $tmp$$Register,
true /*min*/, false /*single*/);
%}
ins_pipe( pipe_slow );
%}
// Load Effective Address
instruct leaP8(rRegP dst, indOffset8 mem)
%{

36
src/hotspot/share/opto/library_call.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@ -6609,6 +6609,40 @@ bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
//------------------------------inline_fp_min_max------------------------------
bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
/* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
// The intrinsic should be used only when the API branches aren't predictable,
// the last one performing the most important comparison. The following heuristic
// uses the branch statistics to eventually bail out if necessary.
ciMethodData *md = callee()->method_data();
if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
ciCallProfile cp = caller()->call_profile_at_bci(bci());
if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
// Bail out if the call-site didn't contribute enough to the statistics.
return false;
}
uint taken = 0, not_taken = 0;
for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
if (p->is_BranchData()) {
taken = ((ciBranchData*)p)->taken();
not_taken = ((ciBranchData*)p)->not_taken();
}
}
double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
balance = balance < 0 ? -balance : balance;
if ( balance > 0.2 ) {
// Bail out if the most important branch is predictable enough.
return false;
}
}
*/
Node *a = NULL;
Node *b = NULL;
Node *n = NULL;

5
src/hotspot/share/opto/loopTransform.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2000, 2019, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@ -2039,7 +2039,8 @@ void PhaseIdealLoop::mark_reductions(IdealLoopTree *loop) {
if (n_ctrl != NULL && loop->is_member(get_loop(n_ctrl))) {
// Now test it to see if it fits the standard pattern for a reduction operator.
int opc = def_node->Opcode();
if (opc != ReductionNode::opcode(opc, def_node->bottom_type()->basic_type())) {
if (opc != ReductionNode::opcode(opc, def_node->bottom_type()->basic_type())
|| opc == Op_MinD || opc == Op_MinF || opc == Op_MaxD || opc == Op_MaxF) {
if (!def_node->is_reduction()) { // Not marked yet
// To be a reduction, the arithmetic node must have the phi as input and provide a def to it
bool ok = false;

236
test/hotspot/jtreg/compiler/intrinsics/math/TestFpMinMaxIntrinsics.java
View File

@ -1,6 +1,6 @@
/*
* Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2018, Arm Limited. All rights reserved.
* Copyright (c) 2018, 2019, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2018, 2019, Arm Limited. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@ -27,20 +27,41 @@
* @bug 8212043
* @summary Test compiler intrinsics of floating-point Math.min/max
*
* @run main/othervm -Xint compiler.intrinsics.math.TestFpMinMaxIntrinsics
* @run main/othervm -Xint compiler.intrinsics.math.TestFpMinMaxIntrinsics sanityTests 1
* @run main/othervm -XX:+UnlockDiagnosticVMOptions
* -Xcomp -XX:TieredStopAtLevel=1
* -XX:CompileOnly=java/lang/Math
* compiler.intrinsics.math.TestFpMinMaxIntrinsics
* compiler.intrinsics.math.TestFpMinMaxIntrinsics sanityTests 1
* @run main/othervm -XX:+UnlockDiagnosticVMOptions
* -Xcomp -XX:-TieredCompilation
* -XX:CompileOnly=java/lang/Math
* compiler.intrinsics.math.TestFpMinMaxIntrinsics
* compiler.intrinsics.math.TestFpMinMaxIntrinsics sanityTests 1
* @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+UnlockDiagnosticVMOptions
* -XX:-TieredCompilation -XX:CompileThresholdScaling=0.1
* -XX:CompileCommand=print,compiler/intrinsics/math/TestFpMinMaxIntrinsics.*Test*
* compiler.intrinsics.math.TestFpMinMaxIntrinsics sanityTests 10000
* @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+UnlockDiagnosticVMOptions
* -XX:-TieredCompilation -Xcomp
* -XX:CompileCommand=print,compiler/intrinsics/math/TestFpMinMaxIntrinsics.*Test*
* -XX:CompileCommand=compileonly,compiler/intrinsics/math/TestFpMinMaxIntrinsics.*Test*
* compiler.intrinsics.math.TestFpMinMaxIntrinsics reductionTests 100
* @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+UnlockDiagnosticVMOptions
* -XX:+TieredCompilation
* -XX:CompileCommand=print,compiler/intrinsics/math/TestFpMinMaxIntrinsics.min*
* -XX:CompileCommand=dontinline,compiler/intrinsics/math/TestFpMinMaxIntrinsics.min*
* compiler.intrinsics.math.TestFpMinMaxIntrinsics randomSearchTree 1
* @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+UnlockDiagnosticVMOptions
* -XX:+TieredCompilation
* -XX:CompileCommand=print,compiler/intrinsics/math/TestFpMinMaxIntrinsics.min*
* -XX:CompileCommand=dontinline,compiler/intrinsics/math/TestFpMinMaxIntrinsics.min*
* compiler.intrinsics.math.TestFpMinMaxIntrinsics sortedSearchTree 1
*/
package compiler.intrinsics.math;
import java.util.Arrays;
import java.util.Random;
import java.lang.reflect.Method;
public class TestFpMinMaxIntrinsics {
@ -63,63 +84,220 @@ public class TestFpMinMaxIntrinsics {
private static final float[][] f_cases = {
// a b min max
{ fPos, fPos, fPos, fPos },
{ fNeg, fNeg, fNeg, fNeg },
{ fPos, fNeg, fNeg, fPos },
{ fNeg, fPos, fNeg, fPos },
{ fPosZero, fNegZero, fNegZero, fPosZero },
{ fNegZero, fPosZero, fNegZero, fPosZero },
{ fNegZero, fNegZero, fNegZero, fNegZero },
{ fPos, fPosInf, fPos, fPosInf },
{ fNeg, fNegInf, fNegInf, fNeg },
{ fPos, fNaN, fNaN, fNaN },
{ fNaN, fPos, fNaN, fNaN },
{ fNeg, fNaN, fNaN, fNaN },
{ fNaN, fNeg, fNaN, fNaN },
{ fPosInf, fNaN, fNaN, fNaN },
{ fNaN, fPosInf, fNaN, fNaN },
{ fNegInf, fNaN, fNaN, fNaN },
{ fNaN, fNegInf, fNaN, fNaN }
};
private static final double[][] d_cases = {
// a b min max
{ dPos, dPos, dPos, dPos },
{ dNeg, dNeg, dNeg, dNeg },
{ dPos, dNeg, dNeg, dPos },
{ dNeg, dPos, dNeg, dPos },
{ dPosZero, dNegZero, dNegZero, dPosZero },
{ dNegZero, dPosZero, dNegZero, dPosZero },
{ dNegZero, dNegZero, dNegZero, dNegZero },
{ dPos, dPosInf, dPos, dPosInf },
{ dNeg, dNegInf, dNegInf, dNeg },
{ dPos, dNaN, dNaN, dNaN },
{ dNaN, dPos, dNaN, dNaN },
{ dNeg, dNaN, dNaN, dNaN },
{ dNaN, dNeg, dNaN, dNaN },
{ dPosInf, dNaN, dNaN, dNaN },
{ dNaN, dPosInf, dNaN, dNaN },
{ dNegInf, dNaN, dNaN, dNaN },
{ dNaN, dNegInf, dNaN, dNaN }
};
private static void fTest(float[] row) {
float min = Math.min(row[0], row[1]);
float max = Math.max(row[0], row[1]);
if (Float.isNaN(min) && Float.isNaN(max)
&& Float.isNaN(row[2]) && Float.isNaN(row[3])) {
// Return if all of them are NaN
return;
fCheck(row[0], row[1], Math.min(row[0], row[1]), Math.max(row[0], row[1]), row[2], row[3]);
}
private static void fReductionTest(float[] row) {
float fmin = row[0], fmax = row[0];
for (int i=0; i<100; i++) {
fmin = Math.min(fmin, row[1]);
fmax = Math.max(fmax, row[1]);
}
if (min != row[2] || max != row[3]) {
fCheck(row[0], row[1], fmin, fmax, row[2], row[3]);
}
private static void fCheck(float a, float b, float fmin, float fmax, float efmin, float efmax) {
int min = Float.floatToRawIntBits(fmin);
int max = Float.floatToRawIntBits(fmax);
int emin = Float.floatToRawIntBits(efmin);
int emax = Float.floatToRawIntBits(efmax);
if (min != emin || max != emax) {
throw new AssertionError("Unexpected result of float min/max: " +
"a = " + row[0] + ", b = " + row[1] + ", " +
"result = (" + min + ", " + max + "), " +
"expected = (" + row[2] + ", " + row[3] + ")");
"a = " + a + ", b = " + b + ", " +
"result = (" + fmin + ", " + fmax + "), " +
"expected = (" + efmin + ", " + efmax + ")");
}
}
private static void dTest(double[] row) {
double min = Math.min(row[0], row[1]);
double max = Math.max(row[0], row[1]);
if (Double.isNaN(min) && Double.isNaN(max)
&& Double.isNaN(row[2]) && Double.isNaN(row[3])) {
// Return if all of them are NaN
return;
dCheck(row[0], row[1], Math.min(row[0], row[1]), Math.max(row[0], row[1]), row[2], row[3]);
}
private static void dReductionTest(double[] row) {
double dmin = row[0], dmax = row[0];
for (int i=0; i<100; i++) {
dmin = Math.min(dmin, row[1]);
dmax = Math.max(dmax, row[1]);
}
if (min != row[2] || max != row[3]) {
throw new AssertionError("Unexpected result of double min/max" +
"a = " + row[0] + ", b = " + row[1] + ", " +
"result = (" + min + ", " + max + "), " +
"expected = (" + row[2] + ", " + row[3] + ")");
dCheck(row[0], row[1], dmin, dmax, row[2], row[3]);
}
private static void dCheck(double a, double b, double dmin, double dmax, double edmin, double edmax) {
double min = Double.doubleToRawLongBits(dmin);
double max = Double.doubleToRawLongBits(dmax);
double emin = Double.doubleToRawLongBits(edmin);
double emax = Double.doubleToRawLongBits(edmax);
if (min != emin || max != emax) {
throw new AssertionError("Unexpected result of double min/max: " +
"a = " + a + ", b = " + b + ", " +
"result = (" + dmin + ", " + dmax + "), " +
"expected = (" + edmin + ", " + edmax + ")");
}
}
public static void main(String[] args) {
public static void sanityTests() {
Arrays.stream(f_cases).forEach(TestFpMinMaxIntrinsics::fTest);
Arrays.stream(d_cases).forEach(TestFpMinMaxIntrinsics::dTest);
System.out.println("PASS");
}
public static void reductionTests() {
Arrays.stream(f_cases).forEach(TestFpMinMaxIntrinsics::fReductionTest);
Arrays.stream(d_cases).forEach(TestFpMinMaxIntrinsics::dReductionTest);
}
public static void main(String[] args) throws Exception {
Method m = TestFpMinMaxIntrinsics.class.getDeclaredMethod(args[0]);
for (int i = 0 ; i < Integer.parseInt(args[1]) ; i++)
m.invoke(null);
}
private static final int COUNT = 1000;
private static final int LOOPS = 100;
private static Random r = new Random();
private static Node[] pool = new Node[COUNT];
private static long time = 0;
private static long times = 0;
public static void init() {
for (int i=0; i<COUNT; i++)
pool[i] = new Node(Double.NaN);
}
public static void finish() {
// String sorted = pool[0].toString();
// System.out.println("Sorted: {" + sorted.substring(0, Math.min(sorted.length(), 180)) + "... }");
System.out.println("Average time: " + (time/times) + " ns");
}
public static void randomSearchTree() {
init();
for (int l=0; l < LOOPS; l++) {
Node root = pool[0].reset(r.nextDouble());
for (int i=1; i<COUNT; i++)
insert(root, pool[i].reset(r.nextDouble()));
}
finish();
}
public static void sortedSearchTree() {
init();
for (int l=0; l < LOOPS; l++) {
Node root = pool[0].reset(-0.0);
for (int i=1; i<COUNT; i++)
insert(root, pool[i].reset(i-1));
}
finish();
}
private static class Node {
private double value;
private Node min;
private Node max;
public Node(double d) { value = d; }
public Node reset(double d) { value = d; min = max = null; return this; }
@Override
public String toString() {
return (min != null ? min + ", " : "") +
value +
(max != null ? ", " + max : "");
}
}
private static Node insert(Node root, Node d) {
for ( ; ; ) {
long rootBits = Double.doubleToRawLongBits(root.value);
long dBits = Double.doubleToRawLongBits(d.value);
// No duplicates
if (rootBits == dBits)
return root;
long delta = System.nanoTime();
double dmin = min(root.value, d.value);
time += System.nanoTime() - delta;
times++;
long minBits = Double.doubleToRawLongBits(dmin);
if (minBits == dBits)
if (root.min != null)
root = root.min;
else
return root.min = d;
else
if (root.max != null)
root = root.max;
else
return root.max = d;
}
}
// Wrapper method to prevent code reordering from affecting measures (JLS 17.4).
private static double min(double a, double b) {
return Math.min(a, b);
}
}

127
test/micro/org/openjdk/bench/vm/compiler/FpMinMaxIntrinsics.java Normal file
View File

@ -0,0 +1,127 @@
/*
* Copyright (c) 2019, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*/
package org.openjdk.bench.vm.compiler;
import org.openjdk.jmh.annotations.*;
import org.openjdk.jmh.infra.*;
import java.util.concurrent.TimeUnit;
import java.util.Random;
@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
@State(Scope.Thread)
public class FpMinMaxIntrinsics {
private static final int COUNT = 1000;
private double[] doubles = new double[COUNT];
private float[] floats = new float[COUNT];
private int c1, c2, s1, s2;
private Random r = new Random();
@Setup
public void init() {
c1 = s1 = step();
c2 = COUNT - (s2 = step());
for (int i=0; i<COUNT; i++) {
floats[i] = r.nextFloat();
doubles[i] = r.nextDouble();
}
}
private int step() {
return (r.nextInt() & 0xf) + 1;
}
@Benchmark
public void dMax(Blackhole bh) {
for (int i=0; i<COUNT; i++)
bh.consume(dMaxBench());
}
@Benchmark
public void dMin(Blackhole bh) {
for (int i=0; i<COUNT; i++)
bh.consume(dMinBench());
}
@Benchmark
public void fMax(Blackhole bh) {
for (int i=0; i<COUNT; i++)
bh.consume(fMaxBench());
}
@Benchmark
public void fMin(Blackhole bh) {
for (int i=0; i<COUNT; i++)
bh.consume(fMinBench());
}
private double dMaxBench() {
inc();
return Math.max(doubles[c1], doubles[c2]);
}
private double dMinBench() {
inc();
return Math.min(doubles[c1], doubles[c2]);
}
private float fMaxBench() {
inc();
return Math.max(floats[c1], floats[c2]);
}
private float fMinBench() {
inc();
return Math.min(floats[c1], floats[c2]);
}
private void inc() {
c1 = c1 + s1 < COUNT ? c1 + s1 : (s1 = step());
c2 = c2 - s2 > 0 ? c2 - s2 : COUNT - (s2 = step());
}
@Benchmark
public float fMinReduce() {
float result = Float.MAX_VALUE;
for (int i=0; i<COUNT; i++)
result = Math.min(result, floats[i]);
return result;
}
@Benchmark
public double dMinReduce() {
double result = Double.MAX_VALUE;
for (int i=0; i<COUNT; i++)
result = Math.min(result, doubles[i]);
return result;
}
}