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

8292898: [vectorapi] Unify vector mask cast operation

Browse Source 
Co-authored-by: Quan Anh Mai <qamai@openjdk.org>
Reviewed-by: jbhateja, eliu
This commit is contained in:
Xiaohong Gong 2022-10-12 01:39:57 +00:00
parent 2ceb80c60f
commit ab8c1361bc
10 changed files with 470 additions and 322 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
src/hotspot/cpu/aarch64/aarch64_vector.ad
View File

@ -5087,6 +5087,48 @@ instruct vmaskcast_same_esize_neon(vReg dst_src) %{
ins_pipe(pipe_class_empty);
%}
instruct vmaskcast_extend_neon(vReg dst, vReg src) %{
predicate(UseSVE == 0 &&
Matcher::vector_length_in_bytes(n) > Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_extend_neon $dst, $src" %}
ins_encode %{
BasicType dst_bt = Matcher::vector_element_basic_type(this);
if (is_floating_point_type(dst_bt)) {
dst_bt = (dst_bt == T_FLOAT) ? T_INT : T_LONG;
}
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
BasicType src_bt = Matcher::vector_element_basic_type(this, $src);
if (is_floating_point_type(src_bt)) {
src_bt = (src_bt == T_FLOAT) ? T_INT : T_LONG;
}
__ neon_vector_extend($dst$$FloatRegister, dst_bt, length_in_bytes_dst,
$src$$FloatRegister, src_bt);
%}
ins_pipe(pipe_slow);
%}
instruct vmaskcast_narrow_neon(vReg dst, vReg src) %{
predicate(UseSVE == 0 &&
Matcher::vector_length_in_bytes(n) < Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_narrow_neon $dst, $src" %}
ins_encode %{
BasicType dst_bt = Matcher::vector_element_basic_type(this);
if (is_floating_point_type(dst_bt)) {
dst_bt = (dst_bt == T_FLOAT) ? T_INT : T_LONG;
}
BasicType src_bt = Matcher::vector_element_basic_type(this, $src);
if (is_floating_point_type(src_bt)) {
src_bt = (src_bt == T_FLOAT) ? T_INT : T_LONG;
}
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);
__ neon_vector_narrow($dst$$FloatRegister, dst_bt,
$src$$FloatRegister, src_bt, length_in_bytes_src);
%}
ins_pipe(pipe_slow);
%}
instruct vmaskcast_same_esize_sve(pReg dst_src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) == Matcher::vector_length_in_bytes(n->in(1)));
@ -5097,11 +5139,11 @@ instruct vmaskcast_same_esize_sve(pReg dst_src) %{
ins_pipe(pipe_class_empty);
%}
instruct vmaskcast_extend(pReg dst, pReg src) %{
instruct vmaskcast_extend_sve(pReg dst, pReg src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) > Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_extend $dst, $src" %}
format %{ "vmaskcast_extend_sve $dst, $src" %}
ins_encode %{
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);
@ -5114,11 +5156,11 @@ instruct vmaskcast_extend(pReg dst, pReg src) %{
ins_pipe(pipe_slow);
%}
instruct vmaskcast_narrow(pReg dst, pReg src) %{
instruct vmaskcast_narrow_sve(pReg dst, pReg src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) < Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_narrow $dst, $src" %}
format %{ "vmaskcast_narrow_sve $dst, $src" %}
ins_encode %{
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);

50
src/hotspot/cpu/aarch64/aarch64_vector_ad.m4
View File

@ -3503,6 +3503,48 @@ instruct vmaskcast_same_esize_neon(vReg dst_src) %{
ins_pipe(pipe_class_empty);
%}
instruct vmaskcast_extend_neon(vReg dst, vReg src) %{
predicate(UseSVE == 0 &&
Matcher::vector_length_in_bytes(n) > Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_extend_neon $dst, $src" %}
ins_encode %{
BasicType dst_bt = Matcher::vector_element_basic_type(this);
if (is_floating_point_type(dst_bt)) {
dst_bt = (dst_bt == T_FLOAT) ? T_INT : T_LONG;
}
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
BasicType src_bt = Matcher::vector_element_basic_type(this, $src);
if (is_floating_point_type(src_bt)) {
src_bt = (src_bt == T_FLOAT) ? T_INT : T_LONG;
}
__ neon_vector_extend($dst$$FloatRegister, dst_bt, length_in_bytes_dst,
$src$$FloatRegister, src_bt);
%}
ins_pipe(pipe_slow);
%}
instruct vmaskcast_narrow_neon(vReg dst, vReg src) %{
predicate(UseSVE == 0 &&
Matcher::vector_length_in_bytes(n) < Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_narrow_neon $dst, $src" %}
ins_encode %{
BasicType dst_bt = Matcher::vector_element_basic_type(this);
if (is_floating_point_type(dst_bt)) {
dst_bt = (dst_bt == T_FLOAT) ? T_INT : T_LONG;
}
BasicType src_bt = Matcher::vector_element_basic_type(this, $src);
if (is_floating_point_type(src_bt)) {
src_bt = (src_bt == T_FLOAT) ? T_INT : T_LONG;
}
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);
__ neon_vector_narrow($dst$$FloatRegister, dst_bt,
$src$$FloatRegister, src_bt, length_in_bytes_src);
%}
ins_pipe(pipe_slow);
%}
instruct vmaskcast_same_esize_sve(pReg dst_src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) == Matcher::vector_length_in_bytes(n->in(1)));
@ -3513,11 +3555,11 @@ instruct vmaskcast_same_esize_sve(pReg dst_src) %{
ins_pipe(pipe_class_empty);
%}
instruct vmaskcast_extend(pReg dst, pReg src) %{
instruct vmaskcast_extend_sve(pReg dst, pReg src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) > Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_extend $dst, $src" %}
format %{ "vmaskcast_extend_sve $dst, $src" %}
ins_encode %{
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);
@ -3530,11 +3572,11 @@ instruct vmaskcast_extend(pReg dst, pReg src) %{
ins_pipe(pipe_slow);
%}
instruct vmaskcast_narrow(pReg dst, pReg src) %{
instruct vmaskcast_narrow_sve(pReg dst, pReg src) %{
predicate(UseSVE > 0 &&
Matcher::vector_length_in_bytes(n) < Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vmaskcast_narrow $dst, $src" %}
format %{ "vmaskcast_narrow_sve $dst, $src" %}
ins_encode %{
uint length_in_bytes_dst = Matcher::vector_length_in_bytes(this);
uint length_in_bytes_src = Matcher::vector_length_in_bytes(this, $src);

55
src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp
View File

@ -4750,6 +4750,61 @@ void C2_MacroAssembler::vector_unsigned_cast(XMMRegister dst, XMMRegister src, i
}
}
void C2_MacroAssembler::vector_mask_cast(XMMRegister dst, XMMRegister src,
BasicType dst_bt, BasicType src_bt, int vlen) {
int vlen_enc = vector_length_encoding(MAX2(type2aelembytes(src_bt), type2aelembytes(dst_bt)) * vlen);
assert(vlen_enc != AVX_512bit, "");
int dst_bt_size = type2aelembytes(dst_bt);
int src_bt_size = type2aelembytes(src_bt);
if (dst_bt_size > src_bt_size) {
switch (dst_bt_size / src_bt_size) {
case 2: vpmovsxbw(dst, src, vlen_enc); break;
case 4: vpmovsxbd(dst, src, vlen_enc); break;
case 8: vpmovsxbq(dst, src, vlen_enc); break;
default: ShouldNotReachHere();
}
} else {
assert(dst_bt_size < src_bt_size, "");
switch (src_bt_size / dst_bt_size) {
case 2: {
if (vlen_enc == AVX_128bit) {
vpacksswb(dst, src, src, vlen_enc);
} else {
vpacksswb(dst, src, src, vlen_enc);
vpermq(dst, dst, 0x08, vlen_enc);
}
break;
}
case 4: {
if (vlen_enc == AVX_128bit) {
vpackssdw(dst, src, src, vlen_enc);
vpacksswb(dst, dst, dst, vlen_enc);
} else {
vpackssdw(dst, src, src, vlen_enc);
vpermq(dst, dst, 0x08, vlen_enc);
vpacksswb(dst, dst, dst, AVX_128bit);
}
break;
}
case 8: {
if (vlen_enc == AVX_128bit) {
vpshufd(dst, src, 0x08, vlen_enc);
vpackssdw(dst, dst, dst, vlen_enc);
vpacksswb(dst, dst, dst, vlen_enc);
} else {
vpshufd(dst, src, 0x08, vlen_enc);
vpermq(dst, dst, 0x08, vlen_enc);
vpackssdw(dst, dst, dst, AVX_128bit);
vpacksswb(dst, dst, dst, AVX_128bit);
}
break;
}
default: ShouldNotReachHere();
}
}
}
void C2_MacroAssembler::evpternlog(XMMRegister dst, int func, KRegister mask, XMMRegister src2, XMMRegister src3,
bool merge, BasicType bt, int vlen_enc) {
if (bt == T_INT) {

2
src/hotspot/cpu/x86/c2_MacroAssembler_x86.hpp
View File

@ -362,6 +362,8 @@ public:
void vector_crosslane_doubleword_pack_avx(XMMRegister dst, XMMRegister src, XMMRegister zero,
XMMRegister xtmp, int index, int vec_enc);
void vector_mask_cast(XMMRegister dst, XMMRegister src, BasicType dst_bt, BasicType src_bt, int vlen);
#ifdef _LP64
void vector_round_double_evex(XMMRegister dst, XMMRegister src, AddressLiteral double_sign_flip, AddressLiteral new_mxcsr, int vec_enc,
Register tmp, XMMRegister xtmp1, XMMRegister xtmp2, KRegister ktmp1, KRegister ktmp2);

19
src/hotspot/cpu/x86/x86.ad
View File

@ -1480,6 +1480,7 @@ const bool Matcher::match_rule_supported(int opcode) {
case Op_VectorUCastB2X:
case Op_VectorUCastS2X:
case Op_VectorUCastI2X:
case Op_VectorMaskCast:
if (UseAVX < 1) { // enabled for AVX only
return false;
}
@ -1857,6 +1858,7 @@ const bool Matcher::match_rule_supported_vector(int opcode, int vlen, BasicType
}
break;
case Op_VectorLoadMask:
case Op_VectorMaskCast:
if (size_in_bits == 256 && UseAVX < 2) {
return false; // Implementation limitation
}
@ -8413,7 +8415,6 @@ instruct vstoreMask_evex(vec dst, kReg mask, immI size) %{
%}
instruct vmaskcast_evex(kReg dst) %{
predicate(Matcher::vector_length(n) == Matcher::vector_length(n->in(1)));
match(Set dst (VectorMaskCast dst));
ins_cost(0);
format %{ "vector_mask_cast $dst" %}
@ -8424,8 +8425,7 @@ instruct vmaskcast_evex(kReg dst) %{
%}
instruct vmaskcast(vec dst) %{
predicate((Matcher::vector_length(n) == Matcher::vector_length(n->in(1))) &&
(Matcher::vector_length_in_bytes(n) == Matcher::vector_length_in_bytes(n->in(1))));
predicate(Matcher::vector_length_in_bytes(n) == Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast dst));
ins_cost(0);
format %{ "vector_mask_cast $dst" %}
@ -8435,6 +8435,19 @@ instruct vmaskcast(vec dst) %{
ins_pipe(empty);
%}
instruct vmaskcast_avx(vec dst, vec src) %{
predicate(Matcher::vector_length_in_bytes(n) != Matcher::vector_length_in_bytes(n->in(1)));
match(Set dst (VectorMaskCast src));
format %{ "vector_mask_cast $dst, $src" %}
ins_encode %{
int vlen = Matcher::vector_length(this);
BasicType src_bt = Matcher::vector_element_basic_type(this, $src);
BasicType dst_bt = Matcher::vector_element_basic_type(this);
__ vector_mask_cast($dst$$XMMRegister, $src$$XMMRegister, dst_bt, src_bt, vlen);
%}
ins_pipe(pipe_slow);
%}
//-------------------------------- Load Iota Indices ----------------------------------
instruct loadIotaIndices(vec dst, immI_0 src) %{

34
src/hotspot/share/opto/vectorIntrinsics.cpp
View File

@ -2488,24 +2488,15 @@ bool LibraryCallKit::inline_vector_convert() {
Node* op = opd1;
if (is_cast) {
BasicType new_elem_bt_to = elem_bt_to;
BasicType new_elem_bt_from = elem_bt_from;
if (is_mask && is_floating_point_type(elem_bt_from)) {
new_elem_bt_from = elem_bt_from == T_FLOAT ? T_INT : T_LONG;
}
if (is_mask && is_floating_point_type(elem_bt_to)) {
new_elem_bt_to = elem_bt_to == T_FLOAT ? T_INT : T_LONG;
}
int cast_vopc = VectorCastNode::opcode(new_elem_bt_from, !is_ucast);
assert(!is_mask || num_elem_from == num_elem_to, "vector mask cast needs the same elem num");
int cast_vopc = VectorCastNode::opcode(elem_bt_from, !is_ucast);
// Make sure that vector cast is implemented to particular type/size combination.
bool no_vec_cast_check = is_mask &&
((src_type->isa_vectmask() && dst_type->isa_vectmask()) ||
type2aelembytes(elem_bt_from) == type2aelembytes(elem_bt_to));
if (!no_vec_cast_check && !arch_supports_vector(cast_vopc, num_elem_to, new_elem_bt_to, VecMaskNotUsed)) {
// Make sure that vector cast is implemented to particular type/size combination if it is
// not a mask casting.
if (!is_mask && !arch_supports_vector(cast_vopc, num_elem_to, elem_bt_to, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=1 op=cast#%d/3 vlen2=%d etype2=%s ismask=%d",
cast_vopc, num_elem_to, type2name(new_elem_bt_to), is_mask);
cast_vopc, num_elem_to, type2name(elem_bt_to), is_mask);
}
return false;
}
@ -2552,12 +2543,15 @@ bool LibraryCallKit::inline_vector_convert() {
op = gvn().transform(VectorCastNode::make(cast_vopc, op, elem_bt_to, num_elem_to));
} else { // num_elem_from == num_elem_to
if (is_mask) {
if ((dst_type->isa_vectmask() && src_type->isa_vectmask()) ||
(type2aelembytes(elem_bt_from) == type2aelembytes(elem_bt_to))) {
op = gvn().transform(new VectorMaskCastNode(op, dst_type));
} else {
op = VectorMaskCastNode::makeCastNode(&gvn(), op, dst_type);
// Make sure that cast for vector mask is implemented to particular type/size combination.
if (!arch_supports_vector(Op_VectorMaskCast, num_elem_to, elem_bt_to, VecMaskNotUsed)) {
if (C->print_intrinsics()) {
tty->print_cr(" ** not supported: arity=1 op=maskcast vlen2=%d etype2=%s ismask=%d",
num_elem_to, type2name(elem_bt_to), is_mask);
}
return false;
}
op = gvn().transform(new VectorMaskCastNode(op, dst_type));
} else {
// Since input and output number of elements match, and since we know this vector size is
// supported, simply do a cast with no resize needed.

56
src/hotspot/share/opto/vectornode.cpp
View File

@ -1619,17 +1619,9 @@ Node* VectorUnboxNode::Ideal(PhaseGVN* phase, bool can_reshape) {
bool is_vector_mask = vbox_klass->is_subclass_of(ciEnv::current()->vector_VectorMask_klass());
bool is_vector_shuffle = vbox_klass->is_subclass_of(ciEnv::current()->vector_VectorShuffle_klass());
if (is_vector_mask) {
// VectorUnbox (VectorBox vmask) ==> VectorMaskCast vmask
const TypeVect* vmask_type = TypeVect::makemask(out_vt->element_basic_type(), out_vt->length());
if (in_vt->length_in_bytes() == out_vt->length_in_bytes() &&
Matcher::match_rule_supported_vector(Op_VectorMaskCast, out_vt->length(), out_vt->element_basic_type())) {
// Apply "VectorUnbox (VectorBox vmask) ==> VectorMaskCast (vmask)"
// directly. This could avoid the transformation ordering issue from
// "VectorStoreMask (VectorLoadMask vmask) => vmask".
return new VectorMaskCastNode(value, vmask_type);
}
// VectorUnbox (VectorBox vmask) ==> VectorLoadMask (VectorStoreMask vmask)
value = phase->transform(VectorStoreMaskNode::make(*phase, value, in_vt->element_basic_type(), in_vt->length()));
return new VectorLoadMaskNode(value, vmask_type);
return new VectorMaskCastNode(value, vmask_type);
} else if (is_vector_shuffle) {
if (!is_shuffle_to_vector()) {
// VectorUnbox (VectorBox vshuffle) ==> VectorLoadShuffle vshuffle
@ -1720,48 +1712,6 @@ Node* VectorMaskToLongNode::Identity(PhaseGVN* phase) {
return this;
}
Node* VectorMaskCastNode::makeCastNode(PhaseGVN* phase, Node* src, const TypeVect* dst_type) {
const TypeVect* src_type = src->bottom_type()->is_vect();
assert(src_type->length() == dst_type->length(), "");
int num_elem = src_type->length();
BasicType elem_bt_from = src_type->element_basic_type();
BasicType elem_bt_to = dst_type->element_basic_type();
if (dst_type->isa_vectmask() == NULL && src_type->isa_vectmask() == NULL &&
type2aelembytes(elem_bt_from) != type2aelembytes(elem_bt_to)) {
Node* op = src;
BasicType new_elem_bt_from = elem_bt_from;
BasicType new_elem_bt_to = elem_bt_to;
if (is_floating_point_type(elem_bt_from)) {
new_elem_bt_from = elem_bt_from == T_FLOAT ? T_INT : T_LONG;
}
if (is_floating_point_type(elem_bt_to)) {
new_elem_bt_to = elem_bt_to == T_FLOAT ? T_INT : T_LONG;
}
// Special handling for casting operation involving floating point types.
// Case A) F -> X := F -> VectorMaskCast (F->I/L [NOP]) -> VectorCast[I/L]2X
// Case B) X -> F := X -> VectorCastX2[I/L] -> VectorMaskCast ([I/L]->F [NOP])
// Case C) F -> F := VectorMaskCast (F->I/L [NOP]) -> VectorCast[I/L]2[L/I] -> VectotMaskCast (L/I->F [NOP])
if (new_elem_bt_from != elem_bt_from) {
const TypeVect* new_src_type = TypeVect::makemask(new_elem_bt_from, num_elem);
op = phase->transform(new VectorMaskCastNode(op, new_src_type));
}
op = phase->transform(VectorCastNode::make(VectorCastNode::opcode(new_elem_bt_from), op, new_elem_bt_to, num_elem));
if (new_elem_bt_to != elem_bt_to) {
op = phase->transform(new VectorMaskCastNode(op, dst_type));
}
return op;
} else {
return new VectorMaskCastNode(src, dst_type);
}
}
Node* VectorLongToMaskNode::Ideal(PhaseGVN* phase, bool can_reshape) {
const TypeVect* dst_type = bottom_type()->is_vect();
if (in(1)->Opcode() == Op_AndL &&
@ -1782,7 +1732,7 @@ Node* VectorLongToMaskNode::Ideal(PhaseGVN* phase, bool can_reshape) {
if (src_type->length() == dst_type->length() &&
((src_type->isa_vectmask() == NULL && dst_type->isa_vectmask() == NULL) ||
(src_type->isa_vectmask() && dst_type->isa_vectmask()))) {
return VectorMaskCastNode::makeCastNode(phase, src, dst_type);
return new VectorMaskCastNode(src, dst_type);
}
}
return NULL;

1
src/hotspot/share/opto/vectornode.hpp
View File

@ -1505,7 +1505,6 @@ class VectorMaskCastNode : public VectorNode {
const TypeVect* in_vt = in->bottom_type()->is_vect();
assert(in_vt->length() == vt->length(), "vector length must match");
}
static Node* makeCastNode(PhaseGVN* phase, Node* in1, const TypeVect * vt);
virtual int Opcode() const;
};

1
test/hotspot/jtreg/compiler/lib/ir_framework/IRNode.java
View File

@ -202,6 +202,7 @@ public class IRNode {
public static final String OR_V_MASK = START + "OrVMask" + MID + END;
public static final String XOR_V_MASK = START + "XorVMask" + MID + END;
public static final String VECTOR_MASK_CAST = START + "VectorMaskCast" + MID + END;
public static final String VECTOR_CAST_B2X = START + "VectorCastB2X" + MID + END;
public static final String VECTOR_CAST_S2X = START + "VectorCastS2X" + MID + END;
public static final String VECTOR_CAST_I2X = START + "VectorCastI2X" + MID + END;

524
test/hotspot/jtreg/compiler/vectorapi/VectorMaskCastTest.java
View File

@ -1,5 +1,5 @@
/*
* Copyright (c) 2021, Arm Limited. All rights reserved.
* Copyright (c) 2021, 2022, 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
@ -23,6 +23,8 @@
package compiler.vectorapi;
import compiler.lib.ir_framework.*;
import java.util.Random;
import jdk.incubator.vector.ByteVector;
@ -32,23 +34,21 @@ import jdk.incubator.vector.IntVector;
import jdk.incubator.vector.LongVector;
import jdk.incubator.vector.ShortVector;
import jdk.incubator.vector.VectorMask;
import jdk.test.lib.Utils;
import org.testng.Assert;
import org.testng.annotations.Test;
import jdk.test.lib.Asserts;
import jdk.test.lib.Utils;
/**
* @test
* @bug 8273264
* @bug 8273264 8292898
* @key randomness
* @library /test/lib
* @summary AArch64: [vector] Add missing rules for VectorMaskCast
* @library /test/lib /
* @summary Unify vector mask cast and add missing rules for VectorMaskCast
* @modules jdk.incubator.vector
*
* @run testng/othervm -XX:-TieredCompilation -XX:CompileThreshold=100 compiler.vectorapi.VectorMaskCastTest
* @run driver compiler.vectorapi.VectorMaskCastTest
*/
// Current vector mask cast test cases at test/jdk/jdk/incubator/vector/*ConversionTests.java
// could not be intrinsfied, hence not able to verify compiler codegen, see [1]. As a
// supplement, we add more tests for vector mask cast operations, which could be intrinsified
@ -56,413 +56,463 @@ import org.testng.annotations.Test;
//
// [1] https://bugs.openjdk.java.net/browse/JDK-8259610
public class VectorMaskCastTest{
public class VectorMaskCastTest {
private static final int NUM_ITER = 5000;
private static final Random rd = Utils.getRandomInstance();
public static boolean[] genMask() {
boolean[] mask = new boolean[64];
for (int i = 0; i < 64; i ++) {
mask[i] = rd.nextBoolean();
private static final boolean[] mask_arr;
static {
mask_arr = new boolean[64];
for (int i = 0; i < 64; i++) {
mask_arr[i] = rd.nextBoolean();
}
return mask;
}
// Byte
private static void testByte64ToShort128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testByte64ToShort128() {
VectorMask<Byte> mByte64 = VectorMask.fromArray(ByteVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mByte64.cast(ShortVector.SPECIES_128).toString(), mByte64.toString());
Asserts.assertEquals(mByte64.cast(ShortVector.SPECIES_128).toString(), mByte64.toString());
}
private static void testByte64ToInt256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testByte64ToInt256() {
VectorMask<Byte> mByte64 = VectorMask.fromArray(ByteVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mByte64.cast(IntVector.SPECIES_256).toString(), mByte64.toString());
Asserts.assertEquals(mByte64.cast(IntVector.SPECIES_256).toString(), mByte64.toString());
}
private static void testByte64ToFloat256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testByte64ToFloat256() {
VectorMask<Byte> mByte64 = VectorMask.fromArray(ByteVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mByte64.cast(FloatVector.SPECIES_256).toString(), mByte64.toString());
Asserts.assertEquals(mByte64.cast(FloatVector.SPECIES_256).toString(), mByte64.toString());
}
private static void testByte64ToLong512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testByte64ToLong512() {
VectorMask<Byte> mByte64 = VectorMask.fromArray(ByteVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mByte64.cast(LongVector.SPECIES_512).toString(), mByte64.toString());
Asserts.assertEquals(mByte64.cast(LongVector.SPECIES_512).toString(), mByte64.toString());
}
private static void testByte64ToDouble512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testByte64ToDouble512() {
VectorMask<Byte> mByte64 = VectorMask.fromArray(ByteVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mByte64.cast(DoubleVector.SPECIES_512).toString(), mByte64.toString());
Asserts.assertEquals(mByte64.cast(DoubleVector.SPECIES_512).toString(), mByte64.toString());
}
private static void testByte128ToShort256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testByte128ToShort256() {
VectorMask<Byte> mByte128 = VectorMask.fromArray(ByteVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mByte128.cast(ShortVector.SPECIES_256).toString(), mByte128.toString());
Asserts.assertEquals(mByte128.cast(ShortVector.SPECIES_256).toString(), mByte128.toString());
}
private static void testByte128ToInt512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testByte128ToInt512() {
VectorMask<Byte> mByte128 = VectorMask.fromArray(ByteVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mByte128.cast(IntVector.SPECIES_512).toString(), mByte128.toString());
Asserts.assertEquals(mByte128.cast(IntVector.SPECIES_512).toString(), mByte128.toString());
}
private static void testByte128ToFloat512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testByte128ToFloat512() {
VectorMask<Byte> mByte128 = VectorMask.fromArray(ByteVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mByte128.cast(FloatVector.SPECIES_512).toString(), mByte128.toString());
Asserts.assertEquals(mByte128.cast(FloatVector.SPECIES_512).toString(), mByte128.toString());
}
private static void testByte256ToShort512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testByte256ToShort512() {
VectorMask<Byte> mByte256 = VectorMask.fromArray(ByteVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mByte256.cast(ShortVector.SPECIES_512).toString(), mByte256.toString());
Asserts.assertEquals(mByte256.cast(ShortVector.SPECIES_512).toString(), mByte256.toString());
}
// Short
private static void testShort64ToInt128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testShort64ToInt128() {
VectorMask<Short> mShort64 = VectorMask.fromArray(ShortVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mShort64.cast(IntVector.SPECIES_128).toString(), mShort64.toString());
Asserts.assertEquals(mShort64.cast(IntVector.SPECIES_128).toString(), mShort64.toString());
}
private static void testShort64ToFloat128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testShort64ToFloat128() {
VectorMask<Short> mShort64 = VectorMask.fromArray(ShortVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mShort64.cast(FloatVector.SPECIES_128).toString(), mShort64.toString());
Asserts.assertEquals(mShort64.cast(FloatVector.SPECIES_128).toString(), mShort64.toString());
}
private static void testShort64ToLong256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testShort64ToLong256() {
VectorMask<Short> mShort64 = VectorMask.fromArray(ShortVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mShort64.cast(LongVector.SPECIES_256).toString(), mShort64.toString());
Asserts.assertEquals(mShort64.cast(LongVector.SPECIES_256).toString(), mShort64.toString());
}
private static void testShort64ToDouble256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testShort64ToDouble256() {
VectorMask<Short> mShort64 = VectorMask.fromArray(ShortVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mShort64.cast(DoubleVector.SPECIES_256).toString(), mShort64.toString());
Asserts.assertEquals(mShort64.cast(DoubleVector.SPECIES_256).toString(), mShort64.toString());
}
private static void testShort128ToByte64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testShort128ToByte64() {
VectorMask<Short> mShort128 = VectorMask.fromArray(ShortVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mShort128.cast(ByteVector.SPECIES_64).toString(), mShort128.toString());
Asserts.assertEquals(mShort128.cast(ByteVector.SPECIES_64).toString(), mShort128.toString());
}
private static void testShort128ToInt256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testShort128ToInt256() {
VectorMask<Short> mShort128 = VectorMask.fromArray(ShortVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mShort128.cast(IntVector.SPECIES_256).toString(), mShort128.toString());
Asserts.assertEquals(mShort128.cast(IntVector.SPECIES_256).toString(), mShort128.toString());
}
private static void testShort128ToFloat256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testShort128ToFloat256() {
VectorMask<Short> mShort128 = VectorMask.fromArray(ShortVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mShort128.cast(FloatVector.SPECIES_256).toString(), mShort128.toString());
Asserts.assertEquals(mShort128.cast(FloatVector.SPECIES_256).toString(), mShort128.toString());
}
private static void testShort128ToLong512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testShort128ToLong512() {
VectorMask<Short> mShort128 = VectorMask.fromArray(ShortVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mShort128.cast(LongVector.SPECIES_512).toString(), mShort128.toString());
Asserts.assertEquals(mShort128.cast(LongVector.SPECIES_512).toString(), mShort128.toString());
}
private static void testShort128ToDouble512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testShort128ToDouble512() {
VectorMask<Short> mShort128 = VectorMask.fromArray(ShortVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mShort128.cast(DoubleVector.SPECIES_512).toString(), mShort128.toString());
Asserts.assertEquals(mShort128.cast(DoubleVector.SPECIES_512).toString(), mShort128.toString());
}
private static void testShort256ToByte128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testShort256ToByte128() {
VectorMask<Short> mShort256 = VectorMask.fromArray(ShortVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mShort256.cast(ByteVector.SPECIES_128).toString(), mShort256.toString());
Asserts.assertEquals(mShort256.cast(ByteVector.SPECIES_128).toString(), mShort256.toString());
}
private static void testShort256ToInt512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testShort256ToInt512() {
VectorMask<Short> mShort256 = VectorMask.fromArray(ShortVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mShort256.cast(IntVector.SPECIES_512).toString(), mShort256.toString());
Asserts.assertEquals(mShort256.cast(IntVector.SPECIES_512).toString(), mShort256.toString());
}
private static void testShort256ToFloat512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testShort256ToFloat512() {
VectorMask<Short> mShort256 = VectorMask.fromArray(ShortVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mShort256.cast(FloatVector.SPECIES_512).toString(), mShort256.toString());
Asserts.assertEquals(mShort256.cast(FloatVector.SPECIES_512).toString(), mShort256.toString());
}
private static void testShort512ToByte256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testShort512ToByte256() {
VectorMask<Short> mShort512 = VectorMask.fromArray(ShortVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mShort512.cast(ByteVector.SPECIES_256).toString(), mShort512.toString());
Asserts.assertEquals(mShort512.cast(ByteVector.SPECIES_256).toString(), mShort512.toString());
}
// Int
private static void testInt64ToLong128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testInt64ToLong128() {
VectorMask<Integer> mInt64 = VectorMask.fromArray(IntVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mInt64.cast(LongVector.SPECIES_128).toString(), mInt64.toString());
Asserts.assertEquals(mInt64.cast(LongVector.SPECIES_128).toString(), mInt64.toString());
}
private static void testInt64ToDouble128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testInt64ToDouble128() {
VectorMask<Integer> mInt64 = VectorMask.fromArray(IntVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mInt64.cast(DoubleVector.SPECIES_128).toString(), mInt64.toString());
Asserts.assertEquals(mInt64.cast(DoubleVector.SPECIES_128).toString(), mInt64.toString());
}
private static void testInt128ToShort64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testInt128ToShort64() {
VectorMask<Integer> mInt128 = VectorMask.fromArray(IntVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mInt128.cast(ShortVector.SPECIES_64).toString(), mInt128.toString());
Asserts.assertEquals(mInt128.cast(ShortVector.SPECIES_64).toString(), mInt128.toString());
}
private static void testInt128ToLong256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testInt128ToLong256() {
VectorMask<Integer> mInt128 = VectorMask.fromArray(IntVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mInt128.cast(LongVector.SPECIES_256).toString(), mInt128.toString());
Asserts.assertEquals(mInt128.cast(LongVector.SPECIES_256).toString(), mInt128.toString());
}
private static void testInt128ToDouble256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testInt128ToDouble256() {
VectorMask<Integer> mInt128 = VectorMask.fromArray(IntVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mInt128.cast(DoubleVector.SPECIES_256).toString(), mInt128.toString());
Asserts.assertEquals(mInt128.cast(DoubleVector.SPECIES_256).toString(), mInt128.toString());
}
private static void testInt256ToShort128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testInt256ToShort128() {
VectorMask<Integer> mInt256 = VectorMask.fromArray(IntVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mInt256.cast(ShortVector.SPECIES_128).toString(), mInt256.toString());
Asserts.assertEquals(mInt256.cast(ShortVector.SPECIES_128).toString(), mInt256.toString());
}
private static void testInt256ToByte64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testInt256ToByte64() {
VectorMask<Integer> mInt256 = VectorMask.fromArray(IntVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mInt256.cast(ByteVector.SPECIES_64).toString(), mInt256.toString());
Asserts.assertEquals(mInt256.cast(ByteVector.SPECIES_64).toString(), mInt256.toString());
}
private static void testInt256ToLong512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testInt256ToLong512() {
VectorMask<Integer> mInt256 = VectorMask.fromArray(IntVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mInt256.cast(LongVector.SPECIES_512).toString(), mInt256.toString());
Asserts.assertEquals(mInt256.cast(LongVector.SPECIES_512).toString(), mInt256.toString());
}
private static void testInt256ToDouble512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testInt256ToDouble512() {
VectorMask<Integer> mInt256 = VectorMask.fromArray(IntVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mInt256.cast(DoubleVector.SPECIES_512).toString(), mInt256.toString());
Asserts.assertEquals(mInt256.cast(DoubleVector.SPECIES_512).toString(), mInt256.toString());
}
private static void testInt512ToShort256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testInt512ToShort256() {
VectorMask<Integer> mInt512 = VectorMask.fromArray(IntVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mInt512.cast(ShortVector.SPECIES_256).toString(), mInt512.toString());
Asserts.assertEquals(mInt512.cast(ShortVector.SPECIES_256).toString(), mInt512.toString());
}
private static void testInt512ToByte128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testInt512ToByte128() {
VectorMask<Integer> mInt512 = VectorMask.fromArray(IntVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mInt512.cast(ByteVector.SPECIES_128).toString(), mInt512.toString());
Asserts.assertEquals(mInt512.cast(ByteVector.SPECIES_128).toString(), mInt512.toString());
}
// Float
private static void testFloat64ToLong128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testFloat64ToLong128() {
VectorMask<Float> mFloat64 = VectorMask.fromArray(FloatVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mFloat64.cast(LongVector.SPECIES_128).toString(), mFloat64.toString());
Asserts.assertEquals(mFloat64.cast(LongVector.SPECIES_128).toString(), mFloat64.toString());
}
private static void testFloat64ToDouble128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testFloat64ToDouble128() {
VectorMask<Float> mFloat64 = VectorMask.fromArray(FloatVector.SPECIES_64, mask_arr, 0);
Assert.assertEquals(mFloat64.cast(DoubleVector.SPECIES_128).toString(), mFloat64.toString());
Asserts.assertEquals(mFloat64.cast(DoubleVector.SPECIES_128).toString(), mFloat64.toString());
}
private static void testFloat128ToShort64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeatureOr = {"avx2", "true", "asimd", "true"})
public static void testFloat128ToShort64() {
VectorMask<Float> mFloat128 = VectorMask.fromArray(FloatVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mFloat128.cast(ShortVector.SPECIES_64).toString(), mFloat128.toString());
Asserts.assertEquals(mFloat128.cast(ShortVector.SPECIES_64).toString(), mFloat128.toString());
}
private static void testFloat128ToLong256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testFloat128ToLong256() {
VectorMask<Float> mFloat128 = VectorMask.fromArray(FloatVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mFloat128.cast(LongVector.SPECIES_256).toString(), mFloat128.toString());
Asserts.assertEquals(mFloat128.cast(LongVector.SPECIES_256).toString(), mFloat128.toString());
}
private static void testFloat128ToDouble256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testFloat128ToDouble256() {
VectorMask<Float> mFloat128 = VectorMask.fromArray(FloatVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mFloat128.cast(DoubleVector.SPECIES_256).toString(), mFloat128.toString());
Asserts.assertEquals(mFloat128.cast(DoubleVector.SPECIES_256).toString(), mFloat128.toString());
}
private static void testFloat256ToShort128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testFloat256ToShort128() {
VectorMask<Float> mFloat256 = VectorMask.fromArray(FloatVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mFloat256.cast(ShortVector.SPECIES_128).toString(), mFloat256.toString());
Asserts.assertEquals(mFloat256.cast(ShortVector.SPECIES_128).toString(), mFloat256.toString());
}
private static void testFloat256ToByte64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testFloat256ToByte64() {
VectorMask<Float> mFloat256 = VectorMask.fromArray(FloatVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mFloat256.cast(ByteVector.SPECIES_64).toString(), mFloat256.toString());
Asserts.assertEquals(mFloat256.cast(ByteVector.SPECIES_64).toString(), mFloat256.toString());
}
private static void testFloat256ToLong512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testFloat256ToLong512() {
VectorMask<Float> mFloat256 = VectorMask.fromArray(FloatVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mFloat256.cast(LongVector.SPECIES_512).toString(), mFloat256.toString());
Asserts.assertEquals(mFloat256.cast(LongVector.SPECIES_512).toString(), mFloat256.toString());
}
private static void testFloat256ToDouble512(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testFloat256ToDouble512() {
VectorMask<Float> mFloat256 = VectorMask.fromArray(FloatVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mFloat256.cast(DoubleVector.SPECIES_512).toString(), mFloat256.toString());
Asserts.assertEquals(mFloat256.cast(DoubleVector.SPECIES_512).toString(), mFloat256.toString());
}
private static void testFloat512ToShort256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testFloat512ToShort256() {
VectorMask<Float> mFloat512 = VectorMask.fromArray(FloatVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mFloat512.cast(ShortVector.SPECIES_256).toString(), mFloat512.toString());
Asserts.assertEquals(mFloat512.cast(ShortVector.SPECIES_256).toString(), mFloat512.toString());
}
private static void testFloat512ToByte128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testFloat512ToByte128() {
VectorMask<Float> mFloat512 = VectorMask.fromArray(FloatVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mFloat512.cast(ByteVector.SPECIES_128).toString(), mFloat512.toString());
Asserts.assertEquals(mFloat512.cast(ByteVector.SPECIES_128).toString(), mFloat512.toString());
}
// Long
private static void testLong128ToInt64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testLong128ToInt64() {
VectorMask<Long> mLong128 = VectorMask.fromArray(LongVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mLong128.cast(IntVector.SPECIES_64).toString(), mLong128.toString());
Asserts.assertEquals(mLong128.cast(IntVector.SPECIES_64).toString(), mLong128.toString());
}
private static void testLong128ToFloat64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testLong128ToFloat64() {
VectorMask<Long> mLong128 = VectorMask.fromArray(LongVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mLong128.cast(FloatVector.SPECIES_64).toString(), mLong128.toString());
Asserts.assertEquals(mLong128.cast(FloatVector.SPECIES_64).toString(), mLong128.toString());
}
private static void testLong256ToInt128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testLong256ToInt128() {
VectorMask<Long> mLong256 = VectorMask.fromArray(LongVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mLong256.cast(IntVector.SPECIES_128).toString(), mLong256.toString());
Asserts.assertEquals(mLong256.cast(IntVector.SPECIES_128).toString(), mLong256.toString());
}
private static void testLong256ToFloat128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testLong256ToFloat128() {
VectorMask<Long> mLong256 = VectorMask.fromArray(LongVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mLong256.cast(FloatVector.SPECIES_128).toString(), mLong256.toString());
Asserts.assertEquals(mLong256.cast(FloatVector.SPECIES_128).toString(), mLong256.toString());
}
private static void testLong256ToShort64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testLong256ToShort64() {
VectorMask<Long> mLong256 = VectorMask.fromArray(LongVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mLong256.cast(ShortVector.SPECIES_64).toString(), mLong256.toString());
Asserts.assertEquals(mLong256.cast(ShortVector.SPECIES_64).toString(), mLong256.toString());
}
private static void testLong512ToInt256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testLong512ToInt256() {
VectorMask<Long> mLong512 = VectorMask.fromArray(LongVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mLong512.cast(IntVector.SPECIES_256).toString(), mLong512.toString());
Asserts.assertEquals(mLong512.cast(IntVector.SPECIES_256).toString(), mLong512.toString());
}
private static void testLong512ToFloat256(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testLong512ToFloat256() {
VectorMask<Long> mLong512 = VectorMask.fromArray(LongVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mLong512.cast(FloatVector.SPECIES_256).toString(), mLong512.toString());
Asserts.assertEquals(mLong512.cast(FloatVector.SPECIES_256).toString(), mLong512.toString());
}
private static void testLong512ToShort128(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testLong512ToShort128() {
VectorMask<Long> mLong512 = VectorMask.fromArray(LongVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mLong512.cast(ShortVector.SPECIES_128).toString(), mLong512.toString());
Asserts.assertEquals(mLong512.cast(ShortVector.SPECIES_128).toString(), mLong512.toString());
}
private static void testLong512ToByte64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testLong512ToByte64() {
VectorMask<Long> mLong512 = VectorMask.fromArray(LongVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mLong512.cast(ByteVector.SPECIES_64).toString(), mLong512.toString());
Asserts.assertEquals(mLong512.cast(ByteVector.SPECIES_64).toString(), mLong512.toString());
}
// Double
private static void testDouble128ToInt64(boolean[] mask_arr) {
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testDouble128ToInt64() {
VectorMask<Double> mDouble128 = VectorMask.fromArray(DoubleVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mDouble128.cast(IntVector.SPECIES_64).toString(), mDouble128.toString());
Asserts.assertEquals(mDouble128.cast(IntVector.SPECIES_64).toString(), mDouble128.toString());
}
private static void testDouble128ToFloat64(boolean[] mask_arr) {
VectorMask<Double> mDouble128 = VectorMask.fromArray(DoubleVector.SPECIES_128, mask_arr, 0);
Assert.assertEquals(mDouble128.cast(FloatVector.SPECIES_64).toString(), mDouble128.toString());
}
private static void testDouble256ToInt128(boolean[] mask_arr) {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mDouble256.cast(IntVector.SPECIES_128).toString(), mDouble256.toString());
}
private static void testDouble256ToFloat128(boolean[] mask_arr) {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mDouble256.cast(FloatVector.SPECIES_128).toString(), mDouble256.toString());
}
private static void testDouble256ToShort64(boolean[] mask_arr) {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Assert.assertEquals(mDouble256.cast(ShortVector.SPECIES_64).toString(), mDouble256.toString());
};
private static void testDouble512ToInt256(boolean[] mask_arr) {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mDouble512.cast(IntVector.SPECIES_256).toString(), mDouble512.toString());
}
private static void testDouble512ToFloat256(boolean[] mask_arr) {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mDouble512.cast(FloatVector.SPECIES_256).toString(), mDouble512.toString());
}
private static void testDouble512ToShort128(boolean[] mask_arr) {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mDouble512.cast(ShortVector.SPECIES_128).toString(), mDouble512.toString());
}
private static void testDouble512ToByte64(boolean[] mask_arr) {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Assert.assertEquals(mDouble512.cast(ByteVector.SPECIES_64).toString(), mDouble512.toString());
}
@Test
public static void testMaskCast() {
for (int i = 0; i < NUM_ITER; i++) {
boolean[] mask = genMask();
// Byte
testByte64ToShort128(mask);
testByte64ToInt256(mask);
testByte64ToFloat256(mask);
testByte64ToLong512(mask);
testByte64ToDouble512(mask);
testByte128ToShort256(mask);
testByte128ToInt512(mask);
testByte128ToFloat512(mask);
testByte256ToShort512(mask);
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"asimd", "true"})
public static void testDouble128ToFloat64() {
VectorMask<Double> mDouble128 = VectorMask.fromArray(DoubleVector.SPECIES_128, mask_arr, 0);
Asserts.assertEquals(mDouble128.cast(FloatVector.SPECIES_64).toString(), mDouble128.toString());
}
// Short
testShort64ToInt128(mask);
testShort64ToFloat128(mask);
testShort64ToLong256(mask);
testShort64ToDouble256(mask);
testShort128ToByte64(mask);
testShort128ToInt256(mask);
testShort128ToFloat256(mask);
testShort128ToLong512(mask);
testShort128ToDouble512(mask);
testShort256ToByte128(mask);
testShort256ToInt512(mask);
testShort256ToFloat512(mask);
testShort512ToByte256(mask);
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testDouble256ToInt128() {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Asserts.assertEquals(mDouble256.cast(IntVector.SPECIES_128).toString(), mDouble256.toString());
}
// Int
testInt64ToLong128(mask);
testInt64ToDouble128(mask);
testInt128ToShort64(mask);
testInt128ToLong256(mask);
testInt128ToDouble256(mask);
testInt256ToShort128(mask);
testInt256ToByte64(mask);
testInt256ToLong512(mask);
testInt256ToDouble512(mask);
testInt512ToShort256(mask);
testInt512ToByte128(mask);
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testDouble256ToFloat128() {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Asserts.assertEquals(mDouble256.cast(FloatVector.SPECIES_128).toString(), mDouble256.toString());
}
// Float
testFloat64ToLong128(mask);
testFloat64ToDouble128(mask);
testFloat128ToShort64(mask);
testFloat128ToLong256(mask);
testFloat128ToDouble256(mask);
testFloat256ToShort128(mask);
testFloat256ToByte64(mask);
testFloat256ToLong512(mask);
testFloat256ToDouble512(mask);
testFloat512ToShort256(mask);
testFloat512ToByte128(mask);
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx2", "true"})
public static void testDouble256ToShort64() {
VectorMask<Double> mDouble256 = VectorMask.fromArray(DoubleVector.SPECIES_256, mask_arr, 0);
Asserts.assertEquals(mDouble256.cast(ShortVector.SPECIES_64).toString(), mDouble256.toString());
}
// Long
testLong128ToInt64(mask);
testLong128ToFloat64(mask);
testLong256ToInt128(mask);
testLong256ToFloat128(mask);
testLong256ToShort64(mask);
testLong512ToInt256(mask);
testLong512ToFloat256(mask);
testLong512ToShort128(mask);
testLong512ToByte64(mask);
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testDouble512ToInt256() {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Asserts.assertEquals(mDouble512.cast(IntVector.SPECIES_256).toString(), mDouble512.toString());
}
// Double
testDouble128ToInt64(mask);
testDouble128ToFloat64(mask);
testDouble256ToInt128(mask);
testDouble256ToFloat128(mask);
testDouble256ToShort64(mask);
testDouble512ToInt256(mask);
testDouble512ToFloat256(mask);
testDouble512ToShort128(mask);
testDouble512ToByte64(mask);
}
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testDouble512ToFloat256() {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Asserts.assertEquals(mDouble512.cast(FloatVector.SPECIES_256).toString(), mDouble512.toString());
}
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testDouble512ToShort128() {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Asserts.assertEquals(mDouble512.cast(ShortVector.SPECIES_128).toString(), mDouble512.toString());
}
@Test
@IR(counts = { IRNode.VECTOR_MASK_CAST, "> 0" }, applyIfCPUFeature = {"avx512vl", "true"})
public static void testDouble512ToByte64() {
VectorMask<Double> mDouble512 = VectorMask.fromArray(DoubleVector.SPECIES_512, mask_arr, 0);
Asserts.assertEquals(mDouble512.cast(ByteVector.SPECIES_64).toString(), mDouble512.toString());
}
public static void main(String[] args) {
TestFramework testFramework = new TestFramework();
testFramework.setDefaultWarmup(5000)
.addFlags("--add-modules=jdk.incubator.vector")
.start();
}
}