archive/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2025-09-20 02:54:35 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

8151002: Make Assembler methods vextract and vinsert match actual instructions

Browse source 
Reviewed-by: kvn, vlivanov, mcberg
This commit is contained in:
Mikael Vidstedt 2016-03-07 15:03:48 -08:00
parent 1c2427a182
commit d4b4dca194
9 changed files with 342 additions and 197 deletions
Download patch file Download diff file Expand all files Collapse all files

20
hotspot/src/cpu/x86/vm/macroAssembler_x86.cpp
View file

@ -3445,7 +3445,7 @@ void MacroAssembler::movptr(Address dst, Register src) {
void MacroAssembler::movdqu(Address dst, XMMRegister src) {
if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (src->encoding() > 15)) {
Assembler::vextractf32x4h(dst, src, 0);
Assembler::vextractf32x4(dst, src, 0);
} else {
Assembler::movdqu(dst, src);
}
@ -3453,7 +3453,7 @@ void MacroAssembler::movdqu(Address dst, XMMRegister src) {
void MacroAssembler::movdqu(XMMRegister dst, Address src) {
if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (dst->encoding() > 15)) {
Assembler::vinsertf32x4h(dst, src, 0);
Assembler::vinsertf32x4(dst, dst, src, 0);
} else {
Assembler::movdqu(dst, src);
}
@ -3478,7 +3478,7 @@ void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src) {
void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (src->encoding() > 15)) {
Assembler::vextractf64x4h(dst, src, 0);
vextractf64x4_low(dst, src);
} else {
Assembler::vmovdqu(dst, src);
}
@ -3486,7 +3486,7 @@ void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
if (UseAVX > 2 && !VM_Version::supports_avx512vl() && (dst->encoding() > 15)) {
Assembler::vinsertf64x4h(dst, src, 0);
vinsertf64x4_low(dst, src);
} else {
Assembler::vmovdqu(dst, src);
}
@ -5649,14 +5649,14 @@ void MacroAssembler::fp_runtime_fallback(address runtime_entry, int nb_args, int
// Save upper half of ZMM registers
subptr(rsp, 32*num_xmm_regs);
for (int n = 0; n < num_xmm_regs; n++) {
vextractf64x4h(Address(rsp, n*32), as_XMMRegister(n), 1);
vextractf64x4_high(Address(rsp, n*32), as_XMMRegister(n));
}
}
assert(UseAVX > 0, "256 bit vectors are supported only with AVX");
// Save upper half of YMM registers
subptr(rsp, 16*num_xmm_regs);
for (int n = 0; n < num_xmm_regs; n++) {
vextractf128h(Address(rsp, n*16), as_XMMRegister(n));
vextractf128_high(Address(rsp, n*16), as_XMMRegister(n));
}
}
#endif
@ -5665,7 +5665,7 @@ void MacroAssembler::fp_runtime_fallback(address runtime_entry, int nb_args, int
#ifdef _LP64
if (VM_Version::supports_evex()) {
for (int n = 0; n < num_xmm_regs; n++) {
vextractf32x4h(Address(rsp, n*16), as_XMMRegister(n), 0);
vextractf32x4(Address(rsp, n*16), as_XMMRegister(n), 0);
}
} else {
for (int n = 0; n < num_xmm_regs; n++) {
@ -5753,7 +5753,7 @@ void MacroAssembler::fp_runtime_fallback(address runtime_entry, int nb_args, int
#ifdef _LP64
if (VM_Version::supports_evex()) {
for (int n = 0; n < num_xmm_regs; n++) {
vinsertf32x4h(as_XMMRegister(n), Address(rsp, n*16), 0);
vinsertf32x4(as_XMMRegister(n), as_XMMRegister(n), Address(rsp, n*16), 0);
}
} else {
for (int n = 0; n < num_xmm_regs; n++) {
@ -5771,12 +5771,12 @@ void MacroAssembler::fp_runtime_fallback(address runtime_entry, int nb_args, int
if (MaxVectorSize > 16) {
// Restore upper half of YMM registers.
for (int n = 0; n < num_xmm_regs; n++) {
vinsertf128h(as_XMMRegister(n), Address(rsp, n*16));
vinsertf128_high(as_XMMRegister(n), Address(rsp, n*16));
}
addptr(rsp, 16*num_xmm_regs);
if(UseAVX > 2) {
for (int n = 0; n < num_xmm_regs; n++) {
vinsertf64x4h(as_XMMRegister(n), Address(rsp, n*32), 1);
vinsertf64x4_high(as_XMMRegister(n), Address(rsp, n*32));
}
addptr(rsp, 32*num_xmm_regs);
}