1802 lines
51 KiB
C
1802 lines
51 KiB
C
/////////////////////////////////////////////////////////////////////////
|
|
// $Id$
|
|
/////////////////////////////////////////////////////////////////////////
|
|
//
|
|
// Copyright (c) 2011-2017 Stanislav Shwartsman
|
|
// Written by Stanislav Shwartsman [sshwarts at sourceforge net]
|
|
//
|
|
// This library is free software; you can redistribute it and/or
|
|
// modify it under the terms of the GNU Lesser General Public
|
|
// License as published by the Free Software Foundation; either
|
|
// version 2 of the License, or (at your option) any later version.
|
|
//
|
|
// This library 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
|
|
// Lesser General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU Lesser General Public
|
|
// License along with this library; if not, write to the Free Software
|
|
// Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA B 02110-1301 USA
|
|
//
|
|
/////////////////////////////////////////////////////////////////////////
|
|
|
|
#ifndef BX_SIMD_INT_FUNCTIONS_H
|
|
#define BX_SIMD_INT_FUNCTIONS_H
|
|
|
|
// absolute value
|
|
|
|
BX_CPP_INLINE void xmm_pabsb(BxPackedXmmRegister *op)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if(op->xmmsbyte(n) < 0) op->xmmubyte(n) = -op->xmmsbyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pabsw(BxPackedXmmRegister *op)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if(op->xmm16s(n) < 0) op->xmm16u(n) = -op->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pabsd(BxPackedXmmRegister *op)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if(op->xmm32s(n) < 0) op->xmm32u(n) = -op->xmm32s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pabsq(BxPackedXmmRegister *op)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
if(op->xmm64s(n) < 0) op->xmm64u(n) = -op->xmm64s(n);
|
|
}
|
|
}
|
|
|
|
// min/max
|
|
|
|
BX_CPP_INLINE void xmm_pminsb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if(op2->xmmsbyte(n) < op1->xmmsbyte(n)) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminub(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if(op2->xmmubyte(n) < op1->xmmubyte(n)) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if(op2->xmm16s(n) < op1->xmm16s(n)) op1->xmm16s(n) = op2->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminuw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if(op2->xmm16u(n) < op1->xmm16u(n)) op1->xmm16s(n) = op2->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminsd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if(op2->xmm32s(n) < op1->xmm32s(n)) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminud(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if(op2->xmm32u(n) < op1->xmm32u(n)) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminsq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
if(op2->xmm64s(n) < op1->xmm64s(n)) op1->xmm64u(n) = op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pminuq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
if(op2->xmm64u(n) < op1->xmm64u(n)) op1->xmm64u(n) = op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxsb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if(op2->xmmsbyte(n) > op1->xmmsbyte(n)) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxub(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if(op2->xmmubyte(n) > op1->xmmubyte(n)) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if(op2->xmm16s(n) > op1->xmm16s(n)) op1->xmm16s(n) = op2->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxuw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if(op2->xmm16u(n) > op1->xmm16u(n)) op1->xmm16s(n) = op2->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxsd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if(op2->xmm32s(n) > op1->xmm32s(n)) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxud(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if(op2->xmm32u(n) > op1->xmm32u(n)) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxsq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
if(op2->xmm64s(n) > op1->xmm64s(n)) op1->xmm64u(n) = op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaxuq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
if(op2->xmm64u(n) > op1->xmm64u(n)) op1->xmm64u(n) = op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
// unpack
|
|
|
|
BX_CPP_INLINE void xmm_unpcklps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm32u(3) = op2->xmm32u(1);
|
|
op1->xmm32u(2) = op1->xmm32u(1);
|
|
op1->xmm32u(1) = op2->xmm32u(0);
|
|
//op1->xmm32u(0) = op1->xmm32u(0);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_unpckhps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm32u(0) = op1->xmm32u(2);
|
|
op1->xmm32u(1) = op2->xmm32u(2);
|
|
op1->xmm32u(2) = op1->xmm32u(3);
|
|
op1->xmm32u(3) = op2->xmm32u(3);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_unpcklpd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
//op1->xmm64u(0) = op1->xmm64u(0);
|
|
op1->xmm64u(1) = op2->xmm64u(0);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_unpckhpd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm64u(0) = op1->xmm64u(1);
|
|
op1->xmm64u(1) = op2->xmm64u(1);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_punpcklbw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmmubyte(0xF) = op2->xmmubyte(7);
|
|
op1->xmmubyte(0xE) = op1->xmmubyte(7);
|
|
op1->xmmubyte(0xD) = op2->xmmubyte(6);
|
|
op1->xmmubyte(0xC) = op1->xmmubyte(6);
|
|
op1->xmmubyte(0xB) = op2->xmmubyte(5);
|
|
op1->xmmubyte(0xA) = op1->xmmubyte(5);
|
|
op1->xmmubyte(0x9) = op2->xmmubyte(4);
|
|
op1->xmmubyte(0x8) = op1->xmmubyte(4);
|
|
op1->xmmubyte(0x7) = op2->xmmubyte(3);
|
|
op1->xmmubyte(0x6) = op1->xmmubyte(3);
|
|
op1->xmmubyte(0x5) = op2->xmmubyte(2);
|
|
op1->xmmubyte(0x4) = op1->xmmubyte(2);
|
|
op1->xmmubyte(0x3) = op2->xmmubyte(1);
|
|
op1->xmmubyte(0x2) = op1->xmmubyte(1);
|
|
op1->xmmubyte(0x1) = op2->xmmubyte(0);
|
|
//op1->xmmubyte(0x0) = op1->xmmubyte(0);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_punpckhbw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmmubyte(0x0) = op1->xmmubyte(0x8);
|
|
op1->xmmubyte(0x1) = op2->xmmubyte(0x8);
|
|
op1->xmmubyte(0x2) = op1->xmmubyte(0x9);
|
|
op1->xmmubyte(0x3) = op2->xmmubyte(0x9);
|
|
op1->xmmubyte(0x4) = op1->xmmubyte(0xA);
|
|
op1->xmmubyte(0x5) = op2->xmmubyte(0xA);
|
|
op1->xmmubyte(0x6) = op1->xmmubyte(0xB);
|
|
op1->xmmubyte(0x7) = op2->xmmubyte(0xB);
|
|
op1->xmmubyte(0x8) = op1->xmmubyte(0xC);
|
|
op1->xmmubyte(0x9) = op2->xmmubyte(0xC);
|
|
op1->xmmubyte(0xA) = op1->xmmubyte(0xD);
|
|
op1->xmmubyte(0xB) = op2->xmmubyte(0xD);
|
|
op1->xmmubyte(0xC) = op1->xmmubyte(0xE);
|
|
op1->xmmubyte(0xD) = op2->xmmubyte(0xE);
|
|
op1->xmmubyte(0xE) = op1->xmmubyte(0xF);
|
|
op1->xmmubyte(0xF) = op2->xmmubyte(0xF);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_punpcklwd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16u(7) = op2->xmm16u(3);
|
|
op1->xmm16u(6) = op1->xmm16u(3);
|
|
op1->xmm16u(5) = op2->xmm16u(2);
|
|
op1->xmm16u(4) = op1->xmm16u(2);
|
|
op1->xmm16u(3) = op2->xmm16u(1);
|
|
op1->xmm16u(2) = op1->xmm16u(1);
|
|
op1->xmm16u(1) = op2->xmm16u(0);
|
|
//op1->xmm16u(0) = op1->xmm16u(0);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_punpckhwd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16u(0) = op1->xmm16u(4);
|
|
op1->xmm16u(1) = op2->xmm16u(4);
|
|
op1->xmm16u(2) = op1->xmm16u(5);
|
|
op1->xmm16u(3) = op2->xmm16u(5);
|
|
op1->xmm16u(4) = op1->xmm16u(6);
|
|
op1->xmm16u(5) = op2->xmm16u(6);
|
|
op1->xmm16u(6) = op1->xmm16u(7);
|
|
op1->xmm16u(7) = op2->xmm16u(7);
|
|
}
|
|
|
|
// pack
|
|
|
|
BX_CPP_INLINE void xmm_packuswb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmmubyte(0x0) = SaturateWordSToByteU(op1->xmm16s(0));
|
|
op1->xmmubyte(0x1) = SaturateWordSToByteU(op1->xmm16s(1));
|
|
op1->xmmubyte(0x2) = SaturateWordSToByteU(op1->xmm16s(2));
|
|
op1->xmmubyte(0x3) = SaturateWordSToByteU(op1->xmm16s(3));
|
|
op1->xmmubyte(0x4) = SaturateWordSToByteU(op1->xmm16s(4));
|
|
op1->xmmubyte(0x5) = SaturateWordSToByteU(op1->xmm16s(5));
|
|
op1->xmmubyte(0x6) = SaturateWordSToByteU(op1->xmm16s(6));
|
|
op1->xmmubyte(0x7) = SaturateWordSToByteU(op1->xmm16s(7));
|
|
|
|
op1->xmmubyte(0x8) = SaturateWordSToByteU(op2->xmm16s(0));
|
|
op1->xmmubyte(0x9) = SaturateWordSToByteU(op2->xmm16s(1));
|
|
op1->xmmubyte(0xA) = SaturateWordSToByteU(op2->xmm16s(2));
|
|
op1->xmmubyte(0xB) = SaturateWordSToByteU(op2->xmm16s(3));
|
|
op1->xmmubyte(0xC) = SaturateWordSToByteU(op2->xmm16s(4));
|
|
op1->xmmubyte(0xD) = SaturateWordSToByteU(op2->xmm16s(5));
|
|
op1->xmmubyte(0xE) = SaturateWordSToByteU(op2->xmm16s(6));
|
|
op1->xmmubyte(0xF) = SaturateWordSToByteU(op2->xmm16s(7));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_packsswb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmmsbyte(0x0) = SaturateWordSToByteS(op1->xmm16s(0));
|
|
op1->xmmsbyte(0x1) = SaturateWordSToByteS(op1->xmm16s(1));
|
|
op1->xmmsbyte(0x2) = SaturateWordSToByteS(op1->xmm16s(2));
|
|
op1->xmmsbyte(0x3) = SaturateWordSToByteS(op1->xmm16s(3));
|
|
op1->xmmsbyte(0x4) = SaturateWordSToByteS(op1->xmm16s(4));
|
|
op1->xmmsbyte(0x5) = SaturateWordSToByteS(op1->xmm16s(5));
|
|
op1->xmmsbyte(0x6) = SaturateWordSToByteS(op1->xmm16s(6));
|
|
op1->xmmsbyte(0x7) = SaturateWordSToByteS(op1->xmm16s(7));
|
|
|
|
op1->xmmsbyte(0x8) = SaturateWordSToByteS(op2->xmm16s(0));
|
|
op1->xmmsbyte(0x9) = SaturateWordSToByteS(op2->xmm16s(1));
|
|
op1->xmmsbyte(0xA) = SaturateWordSToByteS(op2->xmm16s(2));
|
|
op1->xmmsbyte(0xB) = SaturateWordSToByteS(op2->xmm16s(3));
|
|
op1->xmmsbyte(0xC) = SaturateWordSToByteS(op2->xmm16s(4));
|
|
op1->xmmsbyte(0xD) = SaturateWordSToByteS(op2->xmm16s(5));
|
|
op1->xmmsbyte(0xE) = SaturateWordSToByteS(op2->xmm16s(6));
|
|
op1->xmmsbyte(0xF) = SaturateWordSToByteS(op2->xmm16s(7));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_packusdw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16u(0) = SaturateDwordSToWordU(op1->xmm32s(0));
|
|
op1->xmm16u(1) = SaturateDwordSToWordU(op1->xmm32s(1));
|
|
op1->xmm16u(2) = SaturateDwordSToWordU(op1->xmm32s(2));
|
|
op1->xmm16u(3) = SaturateDwordSToWordU(op1->xmm32s(3));
|
|
|
|
op1->xmm16u(4) = SaturateDwordSToWordU(op2->xmm32s(0));
|
|
op1->xmm16u(5) = SaturateDwordSToWordU(op2->xmm32s(1));
|
|
op1->xmm16u(6) = SaturateDwordSToWordU(op2->xmm32s(2));
|
|
op1->xmm16u(7) = SaturateDwordSToWordU(op2->xmm32s(3));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_packssdw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16s(0) = SaturateDwordSToWordS(op1->xmm32s(0));
|
|
op1->xmm16s(1) = SaturateDwordSToWordS(op1->xmm32s(1));
|
|
op1->xmm16s(2) = SaturateDwordSToWordS(op1->xmm32s(2));
|
|
op1->xmm16s(3) = SaturateDwordSToWordS(op1->xmm32s(3));
|
|
|
|
op1->xmm16s(4) = SaturateDwordSToWordS(op2->xmm32s(0));
|
|
op1->xmm16s(5) = SaturateDwordSToWordS(op2->xmm32s(1));
|
|
op1->xmm16s(6) = SaturateDwordSToWordS(op2->xmm32s(2));
|
|
op1->xmm16s(7) = SaturateDwordSToWordS(op2->xmm32s(3));
|
|
}
|
|
|
|
// shuffle
|
|
|
|
BX_CPP_INLINE void xmm_pshufb(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++)
|
|
{
|
|
unsigned mask = op2->xmmubyte(n);
|
|
if (mask & 0x80)
|
|
r->xmmubyte(n) = 0;
|
|
else
|
|
r->xmmubyte(n) = op1->xmmubyte(mask & 0xf);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshufhw(BxPackedXmmRegister *r, const BxPackedXmmRegister *op, Bit8u order)
|
|
{
|
|
r->xmm64u(0) = op->xmm64u(0);
|
|
r->xmm16u(4) = op->xmm16u(4 + ((order >> 0) & 0x3));
|
|
r->xmm16u(5) = op->xmm16u(4 + ((order >> 2) & 0x3));
|
|
r->xmm16u(6) = op->xmm16u(4 + ((order >> 4) & 0x3));
|
|
r->xmm16u(7) = op->xmm16u(4 + ((order >> 6) & 0x3));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshuflw(BxPackedXmmRegister *r, const BxPackedXmmRegister *op, Bit8u order)
|
|
{
|
|
r->xmm16u(0) = op->xmm16u((order >> 0) & 0x3);
|
|
r->xmm16u(1) = op->xmm16u((order >> 2) & 0x3);
|
|
r->xmm16u(2) = op->xmm16u((order >> 4) & 0x3);
|
|
r->xmm16u(3) = op->xmm16u((order >> 6) & 0x3);
|
|
r->xmm64u(1) = op->xmm64u(1);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_shufps(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit8u order)
|
|
{
|
|
r->xmm32u(0) = op1->xmm32u((order >> 0) & 0x3);
|
|
r->xmm32u(1) = op1->xmm32u((order >> 2) & 0x3);
|
|
r->xmm32u(2) = op2->xmm32u((order >> 4) & 0x3);
|
|
r->xmm32u(3) = op2->xmm32u((order >> 6) & 0x3);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_shufpd(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit8u order)
|
|
{
|
|
r->xmm64u(0) = op1->xmm64u((order >> 0) & 0x1);
|
|
r->xmm64u(1) = op2->xmm64u((order >> 1) & 0x1);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_permilps(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
r->xmm32u(0) = op1->xmm32u(op2->xmm32u(0) & 0x3);
|
|
r->xmm32u(1) = op1->xmm32u(op2->xmm32u(1) & 0x3);
|
|
r->xmm32u(2) = op1->xmm32u(op2->xmm32u(2) & 0x3);
|
|
r->xmm32u(3) = op1->xmm32u(op2->xmm32u(3) & 0x3);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_permilpd(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
r->xmm64u(0) = op1->xmm64u((op2->xmm32u(0) >> 1) & 0x1);
|
|
r->xmm64u(1) = op1->xmm64u((op2->xmm32u(2) >> 1) & 0x1);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_permil2ps(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *op3, unsigned m2z)
|
|
{
|
|
for(unsigned n=0; n < 4; n++) {
|
|
Bit32u ctrl = op3->xmm32u(n);
|
|
if ((m2z ^ ((ctrl >> 3) & 0x1)) == 0x3)
|
|
r->xmm32u(n) = 0;
|
|
else
|
|
r->xmm32u(n) = (ctrl & 0x4) ? op1->xmm32u(ctrl & 0x3) : op2->xmm32u(ctrl & 0x3);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_permil2pd(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *op3, unsigned m2z)
|
|
{
|
|
for(unsigned n=0; n < 2; n++) {
|
|
Bit32u ctrl = op3->xmm32u(n*2);
|
|
if ((m2z ^ ((ctrl >> 3) & 0x1)) == 0x3)
|
|
r->xmm64u(n) = 0;
|
|
else
|
|
r->xmm64u(n) = (ctrl & 0x4) ? op1->xmm64u((ctrl >> 1) & 0x1) : op2->xmm64u((ctrl >> 1) & 0x1);
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_AVX
|
|
BX_CPP_INLINE void ymm_vpermq(BxPackedYmmRegister *r, const BxPackedYmmRegister *op, Bit8u control)
|
|
{
|
|
r->ymm64u(0) = op->ymm64u((control) & 0x3);
|
|
r->ymm64u(1) = op->ymm64u((control >> 2) & 0x3);
|
|
r->ymm64u(2) = op->ymm64u((control >> 4) & 0x3);
|
|
r->ymm64u(3) = op->ymm64u((control >> 6) & 0x3);
|
|
}
|
|
#endif
|
|
|
|
// sign
|
|
|
|
BX_CPP_INLINE void xmm_psignb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
int sign = (op2->xmmsbyte(n) > 0) - (op2->xmmsbyte(n) < 0);
|
|
op1->xmmsbyte(n) *= sign;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psignw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
int sign = (op2->xmm16s(n) > 0) - (op2->xmm16s(n) < 0);
|
|
op1->xmm16s(n) *= sign;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psignd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
int sign = (op2->xmm32s(n) > 0) - (op2->xmm32s(n) < 0);
|
|
op1->xmm32s(n) *= sign;
|
|
}
|
|
}
|
|
|
|
// mask creation
|
|
|
|
BX_CPP_INLINE Bit32u xmm_pmovmskb(const BxPackedXmmRegister *op)
|
|
{
|
|
Bit32u mask = 0;
|
|
|
|
if(op->xmmsbyte(0x0) < 0) mask |= 0x0001;
|
|
if(op->xmmsbyte(0x1) < 0) mask |= 0x0002;
|
|
if(op->xmmsbyte(0x2) < 0) mask |= 0x0004;
|
|
if(op->xmmsbyte(0x3) < 0) mask |= 0x0008;
|
|
if(op->xmmsbyte(0x4) < 0) mask |= 0x0010;
|
|
if(op->xmmsbyte(0x5) < 0) mask |= 0x0020;
|
|
if(op->xmmsbyte(0x6) < 0) mask |= 0x0040;
|
|
if(op->xmmsbyte(0x7) < 0) mask |= 0x0080;
|
|
if(op->xmmsbyte(0x8) < 0) mask |= 0x0100;
|
|
if(op->xmmsbyte(0x9) < 0) mask |= 0x0200;
|
|
if(op->xmmsbyte(0xA) < 0) mask |= 0x0400;
|
|
if(op->xmmsbyte(0xB) < 0) mask |= 0x0800;
|
|
if(op->xmmsbyte(0xC) < 0) mask |= 0x1000;
|
|
if(op->xmmsbyte(0xD) < 0) mask |= 0x2000;
|
|
if(op->xmmsbyte(0xE) < 0) mask |= 0x4000;
|
|
if(op->xmmsbyte(0xF) < 0) mask |= 0x8000;
|
|
|
|
return mask;
|
|
}
|
|
|
|
BX_CPP_INLINE Bit32u xmm_pmovmskw(const BxPackedXmmRegister *op)
|
|
{
|
|
Bit32u mask = 0;
|
|
|
|
if(op->xmm16s(0) < 0) mask |= 0x01;
|
|
if(op->xmm16s(1) < 0) mask |= 0x02;
|
|
if(op->xmm16s(2) < 0) mask |= 0x04;
|
|
if(op->xmm16s(3) < 0) mask |= 0x08;
|
|
if(op->xmm16s(4) < 0) mask |= 0x10;
|
|
if(op->xmm16s(5) < 0) mask |= 0x20;
|
|
if(op->xmm16s(6) < 0) mask |= 0x40;
|
|
if(op->xmm16s(7) < 0) mask |= 0x80;
|
|
|
|
return mask;
|
|
}
|
|
|
|
BX_CPP_INLINE Bit32u xmm_pmovmskd(const BxPackedXmmRegister *op)
|
|
{
|
|
Bit32u mask = 0;
|
|
|
|
if(op->xmm32s(0) < 0) mask |= 0x1;
|
|
if(op->xmm32s(1) < 0) mask |= 0x2;
|
|
if(op->xmm32s(2) < 0) mask |= 0x4;
|
|
if(op->xmm32s(3) < 0) mask |= 0x8;
|
|
|
|
return mask;
|
|
}
|
|
|
|
BX_CPP_INLINE Bit32u xmm_pmovmskq(const BxPackedXmmRegister *op)
|
|
{
|
|
Bit32u mask = 0;
|
|
|
|
if(op->xmm32s(1) < 0) mask |= 0x1;
|
|
if(op->xmm32s(3) < 0) mask |= 0x2;
|
|
|
|
return mask;
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmovm2b(BxPackedXmmRegister *dst, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 16; n++, mask >>= 1) {
|
|
dst->xmmsbyte(n) = - (mask & 0x1);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmovm2w(BxPackedXmmRegister *dst, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 8; n++, mask >>= 1) {
|
|
dst->xmm16s(n) = - (mask & 0x1);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmovm2d(BxPackedXmmRegister *dst, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 4; n++, mask >>= 1) {
|
|
dst->xmm32s(n) = - (mask & 0x1);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmovm2q(BxPackedXmmRegister *dst, Bit32u mask)
|
|
{
|
|
dst->xmm64s(0) = (mask & 0x1) ? (Bit64s) -1 : 0;
|
|
dst->xmm64s(1) = (mask & 0x2) ? (Bit64s) -1 : 0;
|
|
}
|
|
|
|
// blend
|
|
|
|
BX_CPP_INLINE void xmm_pblendb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 16; n++, mask >>= 1) {
|
|
if (mask & 0x1) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_zero_pblendb(BxPackedXmmRegister *dst, const BxPackedXmmRegister *op, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 16; n++, mask >>= 1) {
|
|
dst->xmmubyte(n) = (mask & 0x1) ? op->xmmubyte(n) : 0;
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
BX_CPP_INLINE void simd_pblendb(BxPackedAvxRegister *op1, const BxPackedAvxRegister *op2, Bit64u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
if (mask & 0x1) op1->vmmubyte(n) = op2->vmmubyte(n);
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_zero_pblendb(BxPackedAvxRegister *dst, const BxPackedAvxRegister *op, Bit64u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
dst->vmmubyte(n) = (mask & 0x1) ? op->vmmubyte(n) : 0;
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
BX_CPP_INLINE void xmm_pblendw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 8; n++, mask >>= 1) {
|
|
if (mask & 0x1) op1->xmm16u(n) = op2->xmm16u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_zero_pblendw(BxPackedXmmRegister *dst, const BxPackedXmmRegister *op, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 8; n++, mask >>= 1) {
|
|
dst->xmm16u(n) = (mask & 0x1) ? op->xmm16u(n) : 0;
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
BX_CPP_INLINE void simd_pblendw(BxPackedAvxRegister *op1, const BxPackedAvxRegister *op2, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
if (mask & 0x1) op1->vmm16u(n) = op2->vmm16u(n);
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_zero_pblendw(BxPackedAvxRegister *dst, const BxPackedAvxRegister *op, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
dst->vmm16u(n) = (mask & 0x1) ? op->vmm16u(n) : 0;
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
BX_CPP_INLINE void xmm_blendps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 4; n++, mask >>= 1) {
|
|
if (mask & 0x1) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_zero_blendps(BxPackedXmmRegister *dst, const BxPackedXmmRegister *op, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 4; n++, mask >>= 1) {
|
|
dst->xmm32u(n) = (mask & 0x1) ? op->xmm32u(n) : 0;
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
BX_CPP_INLINE void simd_blendps(BxPackedAvxRegister *op1, const BxPackedAvxRegister *op2, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
if (mask & 0x1) op1->vmm32u(n) = op2->vmm32u(n);
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_zero_blendps(BxPackedAvxRegister *dst, const BxPackedAvxRegister *op, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
dst->vmm32u(n) = (mask & 0x1) ? op->vmm32u(n) : 0;
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
BX_CPP_INLINE void xmm_blendpd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 2; n++, mask >>= 1) {
|
|
if (mask & 0x1) op1->xmm64u(n) = op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_zero_blendpd(BxPackedXmmRegister *dst, const BxPackedXmmRegister *op, Bit32u mask)
|
|
{
|
|
for (unsigned n=0; n < 2; n++, mask >>= 1) {
|
|
dst->xmm64u(n) = (mask & 0x1) ? op->xmm64u(n) : 0;
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
BX_CPP_INLINE void simd_blendpd(BxPackedAvxRegister *op1, const BxPackedAvxRegister *op2, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
if (mask & 0x1) op1->vmm64u(n) = op2->vmm64u(n);
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_zero_blendpd(BxPackedAvxRegister *dst, const BxPackedAvxRegister *op, Bit32u mask, unsigned len)
|
|
{
|
|
for (unsigned n=0; n < len; n++) {
|
|
dst->vmm64u(n) = (mask & 0x1) ? op->vmm64u(n) : 0;
|
|
mask >>= 1;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
BX_CPP_INLINE void xmm_pblendvb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *mask)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
if (mask->xmmsbyte(n) < 0) op1->xmmubyte(n) = op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pblendvw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *mask)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
if (mask->xmm16s(n) < 0) op1->xmm16u(n) = op2->xmm16u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_blendvps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *mask)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
if (mask->xmm32s(n) < 0) op1->xmm32u(n) = op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_blendvpd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *mask)
|
|
{
|
|
if (mask->xmm32s(1) < 0) op1->xmm64u(0) = op2->xmm64u(0);
|
|
if (mask->xmm32s(3) < 0) op1->xmm64u(1) = op2->xmm64u(1);
|
|
}
|
|
|
|
// arithmetic (logic)
|
|
|
|
BX_CPP_INLINE void xmm_andps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++)
|
|
op1->xmm64u(n) &= op2->xmm64u(n);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_andnps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++)
|
|
op1->xmm64u(n) = ~(op1->xmm64u(n)) & op2->xmm64u(n);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_orps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++)
|
|
op1->xmm64u(n) |= op2->xmm64u(n);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_xorps(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++)
|
|
op1->xmm64u(n) ^= op2->xmm64u(n);
|
|
}
|
|
|
|
// arithmetic (add/sub)
|
|
|
|
BX_CPP_INLINE void xmm_paddb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmubyte(n) += op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16u(n) += op2->xmm16u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
op1->xmm32u(n) += op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
op1->xmm64u(n) += op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmubyte(n) -= op2->xmmubyte(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16u(n) -= op2->xmm16u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
op1->xmm32u(n) -= op2->xmm32u(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
op1->xmm64u(n) -= op2->xmm64u(n);
|
|
}
|
|
}
|
|
|
|
// arithmetic (add/sub with saturation)
|
|
|
|
BX_CPP_INLINE void xmm_paddsb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmsbyte(n) = SaturateWordSToByteS(Bit16s(op1->xmmsbyte(n)) + Bit16s(op2->xmmsbyte(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16s(n) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(n)) + Bit32s(op2->xmm16s(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddusb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmubyte(n) = SaturateWordSToByteU(Bit16s(op1->xmmubyte(n)) + Bit16s(op2->xmmubyte(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_paddusw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16u(n) = SaturateDwordSToWordU(Bit32s(op1->xmm16u(n)) + Bit32s(op2->xmm16u(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubsb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmsbyte(n) = SaturateWordSToByteS(Bit16s(op1->xmmsbyte(n)) - Bit16s(op2->xmmsbyte(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16s(n) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(n)) - Bit32s(op2->xmm16s(n)));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubusb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++)
|
|
{
|
|
if(op1->xmmubyte(n) > op2->xmmubyte(n))
|
|
op1->xmmubyte(n) -= op2->xmmubyte(n);
|
|
else
|
|
op1->xmmubyte(n) = 0;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psubusw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++)
|
|
{
|
|
if(op1->xmm16u(n) > op2->xmm16u(n))
|
|
op1->xmm16u(n) -= op2->xmm16u(n);
|
|
else
|
|
op1->xmm16u(n) = 0;
|
|
}
|
|
}
|
|
|
|
// arithmetic (horizontal add/sub)
|
|
|
|
BX_CPP_INLINE void xmm_phaddw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16u(0) = op1->xmm16u(0) + op1->xmm16u(1);
|
|
op1->xmm16u(1) = op1->xmm16u(2) + op1->xmm16u(3);
|
|
op1->xmm16u(2) = op1->xmm16u(4) + op1->xmm16u(5);
|
|
op1->xmm16u(3) = op1->xmm16u(6) + op1->xmm16u(7);
|
|
|
|
op1->xmm16u(4) = op2->xmm16u(0) + op2->xmm16u(1);
|
|
op1->xmm16u(5) = op2->xmm16u(2) + op2->xmm16u(3);
|
|
op1->xmm16u(6) = op2->xmm16u(4) + op2->xmm16u(5);
|
|
op1->xmm16u(7) = op2->xmm16u(6) + op2->xmm16u(7);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_phaddd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm32u(0) = op1->xmm32u(0) + op1->xmm32u(1);
|
|
op1->xmm32u(1) = op1->xmm32u(2) + op1->xmm32u(3);
|
|
op1->xmm32u(2) = op2->xmm32u(0) + op2->xmm32u(1);
|
|
op1->xmm32u(3) = op2->xmm32u(2) + op2->xmm32u(3);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_phaddsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16s(0) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(0)) + Bit32s(op1->xmm16s(1)));
|
|
op1->xmm16s(1) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(2)) + Bit32s(op1->xmm16s(3)));
|
|
op1->xmm16s(2) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(4)) + Bit32s(op1->xmm16s(5)));
|
|
op1->xmm16s(3) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(6)) + Bit32s(op1->xmm16s(7)));
|
|
|
|
op1->xmm16s(4) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(0)) + Bit32s(op2->xmm16s(1)));
|
|
op1->xmm16s(5) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(2)) + Bit32s(op2->xmm16s(3)));
|
|
op1->xmm16s(6) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(4)) + Bit32s(op2->xmm16s(5)));
|
|
op1->xmm16s(7) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(6)) + Bit32s(op2->xmm16s(7)));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_phsubw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16u(0) = op1->xmm16u(0) - op1->xmm16u(1);
|
|
op1->xmm16u(1) = op1->xmm16u(2) - op1->xmm16u(3);
|
|
op1->xmm16u(2) = op1->xmm16u(4) - op1->xmm16u(5);
|
|
op1->xmm16u(3) = op1->xmm16u(6) - op1->xmm16u(7);
|
|
|
|
op1->xmm16u(4) = op2->xmm16u(0) - op2->xmm16u(1);
|
|
op1->xmm16u(5) = op2->xmm16u(2) - op2->xmm16u(3);
|
|
op1->xmm16u(6) = op2->xmm16u(4) - op2->xmm16u(5);
|
|
op1->xmm16u(7) = op2->xmm16u(6) - op2->xmm16u(7);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_phsubd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm32u(0) = op1->xmm32u(0) - op1->xmm32u(1);
|
|
op1->xmm32u(1) = op1->xmm32u(2) - op1->xmm32u(3);
|
|
op1->xmm32u(2) = op2->xmm32u(0) - op2->xmm32u(1);
|
|
op1->xmm32u(3) = op2->xmm32u(2) - op2->xmm32u(3);
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_phsubsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm16s(0) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(0)) - Bit32s(op1->xmm16s(1)));
|
|
op1->xmm16s(1) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(2)) - Bit32s(op1->xmm16s(3)));
|
|
op1->xmm16s(2) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(4)) - Bit32s(op1->xmm16s(5)));
|
|
op1->xmm16s(3) = SaturateDwordSToWordS(Bit32s(op1->xmm16s(6)) - Bit32s(op1->xmm16s(7)));
|
|
|
|
op1->xmm16s(4) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(0)) - Bit32s(op2->xmm16s(1)));
|
|
op1->xmm16s(5) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(2)) - Bit32s(op2->xmm16s(3)));
|
|
op1->xmm16s(6) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(4)) - Bit32s(op2->xmm16s(5)));
|
|
op1->xmm16s(7) = SaturateDwordSToWordS(Bit32s(op2->xmm16s(6)) - Bit32s(op2->xmm16s(7)));
|
|
}
|
|
|
|
// average
|
|
|
|
BX_CPP_INLINE void xmm_pavgb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op1->xmmubyte(n) = (op1->xmmubyte(n) + op2->xmmubyte(n) + 1) >> 1;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pavgw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16u(n) = (op1->xmm16u(n) + op2->xmm16u(n) + 1) >> 1;
|
|
}
|
|
}
|
|
|
|
// multiply
|
|
|
|
BX_CPP_INLINE void xmm_pmullw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16s(n) *= op2->xmm16s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmulhw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
Bit32s product = Bit32s(op1->xmm16s(n)) * Bit32s(op2->xmm16s(n));
|
|
op1->xmm16u(n) = (Bit16u)(product >> 16);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmulhuw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
Bit32u product = Bit32u(op1->xmm16u(n)) * Bit32u(op2->xmm16u(n));
|
|
op1->xmm16u(n) = (Bit16u)(product >> 16);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmulld(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
op1->xmm32s(n) *= op2->xmm32s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmullq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
op1->xmm64s(n) *= op2->xmm64s(n);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmuldq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm64s(0) = Bit64s(op1->xmm32s(0)) * Bit64s(op2->xmm32s(0));
|
|
op1->xmm64s(1) = Bit64s(op1->xmm32s(2)) * Bit64s(op2->xmm32s(2));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmuludq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
op1->xmm64u(0) = Bit64u(op1->xmm32u(0)) * Bit64u(op2->xmm32u(0));
|
|
op1->xmm64u(1) = Bit64u(op1->xmm32u(2)) * Bit64u(op2->xmm32u(2));
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmulhrsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op1->xmm16u(n) = (((Bit32s(op1->xmm16s(n)) * Bit32s(op2->xmm16s(n))) >> 14) + 1) >> 1;
|
|
}
|
|
}
|
|
|
|
// multiply/add
|
|
|
|
BX_CPP_INLINE void xmm_pmaddubsw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<8; n++)
|
|
{
|
|
Bit32s temp = Bit32s(op1->xmmubyte(n*2)) * Bit32s(op2->xmmsbyte(n*2)) +
|
|
Bit32s(op1->xmmubyte(n*2+1)) * Bit32s(op2->xmmsbyte(n*2+1));
|
|
|
|
op1->xmm16s(n) = SaturateDwordSToWordS(temp);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pmaddwd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++)
|
|
{
|
|
op1->xmm32u(n) = Bit32s(op1->xmm16s(n*2)) * Bit32s(op2->xmm16s(n*2)) +
|
|
Bit32s(op1->xmm16s(n*2+1)) * Bit32s(op2->xmm16s(n*2+1));
|
|
}
|
|
}
|
|
|
|
// broadcast
|
|
|
|
BX_CPP_INLINE void xmm_pbroadcastb(BxPackedXmmRegister *op, Bit8u val_8)
|
|
{
|
|
for(unsigned n=0; n<16; n++) {
|
|
op->xmmubyte(n) = val_8;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pbroadcastw(BxPackedXmmRegister *op, Bit16u val_16)
|
|
{
|
|
for(unsigned n=0; n<8; n++) {
|
|
op->xmm16u(n) = val_16;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pbroadcastd(BxPackedXmmRegister *op, Bit32u val_32)
|
|
{
|
|
for(unsigned n=0; n<4; n++) {
|
|
op->xmm32u(n) = val_32;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pbroadcastq(BxPackedXmmRegister *op, Bit64u val_64)
|
|
{
|
|
for(unsigned n=0; n<2; n++) {
|
|
op->xmm64u(n) = val_64;
|
|
}
|
|
}
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
BX_CPP_INLINE void simd_pbroadcastb(BxPackedAvxRegister *op, Bit8u val_8, unsigned len)
|
|
{
|
|
for(unsigned n=0; n < len; n++) {
|
|
op->vmmubyte(n) = val_8;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_pbroadcastw(BxPackedAvxRegister *op, Bit16u val_16, unsigned len)
|
|
{
|
|
for(unsigned n=0; n < len; n++) {
|
|
op->vmm16u(n) = val_16;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_pbroadcastd(BxPackedAvxRegister *op, Bit32u val_32, unsigned len)
|
|
{
|
|
for(unsigned n=0; n < len; n++) {
|
|
op->vmm32u(n) = val_32;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void simd_pbroadcastq(BxPackedAvxRegister *op, Bit64u val_64, unsigned len)
|
|
{
|
|
for(unsigned n=0; n < len; n++) {
|
|
op->vmm64u(n) = val_64;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
// sum of absolute differences (SAD)
|
|
|
|
BX_CPP_INLINE void xmm_psadbw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
unsigned temp = 0;
|
|
for (unsigned n=0; n < 8; n++)
|
|
temp += abs(op1->xmmubyte(n) - op2->xmmubyte(n));
|
|
|
|
op1->xmm64u(0) = Bit64u(temp);
|
|
|
|
temp = 0;
|
|
for (unsigned n=8; n < 16; n++)
|
|
temp += abs(op1->xmmubyte(n) - op2->xmmubyte(n));
|
|
|
|
op1->xmm64u(1) = Bit64u(temp);
|
|
}
|
|
|
|
// multiple sum of absolute differences (MSAD)
|
|
|
|
BX_CPP_INLINE Bit16u sad_quadruple(const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, int op1_offset, int op2_offset)
|
|
{
|
|
Bit32u r = 0;
|
|
|
|
for (unsigned n=0; n < 4; n++) {
|
|
Bit8u temp1 = op1->xmmubyte(n + op1_offset);
|
|
Bit8u temp2 = op2->xmmubyte(n + op2_offset);
|
|
|
|
r += abs(temp1 - temp2);
|
|
}
|
|
|
|
return r;
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_mpsadbw(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, Bit8u control)
|
|
{
|
|
unsigned src_offset = (control & 0x3) * 4;
|
|
unsigned dst_offset = ((control >> 2) & 0x1) * 4;
|
|
|
|
for (unsigned j=0; j < 8; j++) {
|
|
r->xmm16u(j) = sad_quadruple(op1, op2, dst_offset + j, src_offset);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_dbpsadbw(BxPackedXmmRegister *r, const BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
// assuming shuffle of op2 was done outside
|
|
r->xmm16u(0) = sad_quadruple(op1, op2, 0, 0);
|
|
r->xmm16u(1) = sad_quadruple(op1, op2, 0, 1);
|
|
r->xmm16u(2) = sad_quadruple(op1, op2, 4, 2);
|
|
r->xmm16u(3) = sad_quadruple(op1, op2, 4, 3);
|
|
r->xmm16u(4) = sad_quadruple(op1, op2, 8, 8);
|
|
r->xmm16u(5) = sad_quadruple(op1, op2, 8, 9);
|
|
r->xmm16u(6) = sad_quadruple(op1, op2, 12, 10);
|
|
r->xmm16u(7) = sad_quadruple(op1, op2, 12, 11);
|
|
}
|
|
|
|
// conflict
|
|
|
|
#if BX_SUPPORT_EVEX
|
|
|
|
BX_CPP_INLINE Bit32u simd_pconflictd(const BxPackedAvxRegister *op, int index)
|
|
{
|
|
Bit32u result = 0;
|
|
// compare index element with all previous elements
|
|
for (int i=0; i<index-1; i++) {
|
|
if (op->vmm32u(index) == op->vmm32u(i)) result |= (1 << i);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
BX_CPP_INLINE Bit32u simd_pconflictq(const BxPackedAvxRegister *op, int index)
|
|
{
|
|
Bit32u result = 0;
|
|
// compare index element with all previous elements
|
|
for (int i=0; i<index-1; i++) {
|
|
if (op->vmm64u(index) == op->vmm64u(i)) result |= (1 << i);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
#endif
|
|
|
|
// bitwise select
|
|
|
|
BX_CPP_INLINE void xmm_pselect(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2, const BxPackedXmmRegister *op3)
|
|
{
|
|
for(unsigned n=0;n < 2;n++) {
|
|
op1->xmm64u(n) = (op3->xmm64u(n) & op1->xmm64u(n)) | (~op3->xmm64u(n) & op2->xmm64u(n));
|
|
}
|
|
}
|
|
|
|
// shift
|
|
|
|
BX_CPP_INLINE void xmm_psravw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 8; n++) {
|
|
unsigned shift = op2->xmm16u(n);
|
|
if(shift > 15)
|
|
op1->xmm16u(n) = (op1->xmm16s(n) < 0) ? 0xffff : 0;
|
|
else
|
|
op1->xmm16u(n) = (Bit16u)(op1->xmm16s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psravd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 4; n++) {
|
|
Bit32u shift = op2->xmm32u(n);
|
|
if(shift > 31)
|
|
op1->xmm32u(n) = (op1->xmm32s(n) < 0) ? 0xffffffff : 0;
|
|
else
|
|
op1->xmm32u(n) = (Bit32u)(op1->xmm32s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psravq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++) {
|
|
Bit64u shift = op2->xmm64u(n);
|
|
if(shift > 64)
|
|
op1->xmm64u(n) = (op1->xmm64s(n) < 0) ? BX_CONST64(0xffffffffffffffff) : 0;
|
|
else
|
|
op1->xmm64u(n) = (Bit64u)(op1->xmm64s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psllvw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 8; n++) {
|
|
unsigned shift = op2->xmm16u(n);
|
|
if(shift > 15)
|
|
op1->xmm16u(n) = 0;
|
|
else
|
|
op1->xmm16u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psllvd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 4; n++) {
|
|
Bit32u shift = op2->xmm32u(n);
|
|
if(shift > 31)
|
|
op1->xmm32u(n) = 0;
|
|
else
|
|
op1->xmm32u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psllvq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++) {
|
|
Bit64u shift = op2->xmm64u(n);
|
|
if(shift > 63)
|
|
op1->xmm64u(n) = 0;
|
|
else
|
|
op1->xmm64u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrlvw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 8; n++) {
|
|
unsigned shift = op2->xmm16u(n);
|
|
if(shift > 15)
|
|
op1->xmm16u(n) = 0;
|
|
else
|
|
op1->xmm16u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrlvd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 4; n++) {
|
|
Bit32u shift = op2->xmm32u(n);
|
|
if(shift > 31)
|
|
op1->xmm32u(n) = 0;
|
|
else
|
|
op1->xmm32u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrlvq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for (unsigned n=0; n < 2; n++) {
|
|
Bit64u shift = op2->xmm64u(n);
|
|
if(shift > 63)
|
|
op1->xmm64u(n) = 0;
|
|
else
|
|
op1->xmm64u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psraw(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 15) {
|
|
for (unsigned n=0; n < 8; n++)
|
|
op->xmm16u(n) = (op->xmm16s(n) < 0) ? 0xffff : 0;
|
|
}
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 8; n++)
|
|
op->xmm16u(n) = (Bit16u)(op->xmm16s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrad(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 31) {
|
|
for (unsigned n=0; n < 4; n++)
|
|
op->xmm32u(n) = (op->xmm32s(n) < 0) ? 0xffffffff : 0;
|
|
}
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 4; n++)
|
|
op->xmm32u(n) = (Bit32u)(op->xmm32s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psraq(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 63) {
|
|
for (unsigned n=0; n < 2; n++)
|
|
op->xmm64u(n) = (op->xmm64s(n) < 0) ? BX_CONST64(0xffffffffffffffff) : 0;
|
|
}
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 2; n++)
|
|
op->xmm64u(n) = (Bit64u)(op->xmm64s(n) >> shift);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrlw(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 15) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 8; n++)
|
|
op->xmm16u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrld(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 31) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 4; n++)
|
|
op->xmm32u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrlq(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 64) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 2; n++)
|
|
op->xmm64u(n) >>= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psllw(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 15) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 8; n++)
|
|
op->xmm16u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pslld(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 31) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 4; n++)
|
|
op->xmm32u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psllq(BxPackedXmmRegister *op, Bit64u shift_64)
|
|
{
|
|
if(shift_64 > 63) op->clear();
|
|
else
|
|
{
|
|
Bit8u shift = (Bit8u) shift_64;
|
|
|
|
for (unsigned n=0; n < 2; n++)
|
|
op->xmm64u(n) <<= shift;
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_psrldq(BxPackedXmmRegister *op, Bit8u shift)
|
|
{
|
|
if(shift > 15) op->clear();
|
|
else {
|
|
if (shift > 7) {
|
|
op->xmm64u(0) = op->xmm64u(1);
|
|
op->xmm64u(1) = 0;
|
|
shift -= 8;
|
|
}
|
|
|
|
shift <<= 3;
|
|
|
|
if (shift != 0) {
|
|
op->xmm64u(0) = (op->xmm64u(0) >> shift) | (op->xmm64u(1) << (64-shift));
|
|
op->xmm64u(1) = (op->xmm64u(1) >> shift);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pslldq(BxPackedXmmRegister *op, Bit8u shift)
|
|
{
|
|
if(shift > 15) op->clear();
|
|
else {
|
|
if (shift > 7) {
|
|
op->xmm64u(1) = op->xmm64u(0);
|
|
op->xmm64u(0) = 0;
|
|
shift -= 8;
|
|
}
|
|
|
|
shift <<= 3;
|
|
|
|
if (shift != 0) {
|
|
op->xmm64u(1) = (op->xmm64u(1) << shift) | (op->xmm64u(0) >> (64-shift));
|
|
op->xmm64u(0) = (op->xmm64u(0) << shift);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_palignr(BxPackedXmmRegister *op2, const BxPackedXmmRegister *op1, Bit8u shift)
|
|
{
|
|
// op2 = [op1:op2] >> shift
|
|
|
|
if (shift > 15) {
|
|
*op2 = *op1;
|
|
xmm_psrldq(op2, shift - 16);
|
|
return;
|
|
}
|
|
|
|
shift <<= 3;
|
|
|
|
if (shift > 64) {
|
|
shift -= 64;
|
|
op2->xmm64u(0) = (op2->xmm64u(1) >> shift) | (op1->xmm64u(0) << (64-shift));
|
|
op2->xmm64u(1) = (op1->xmm64u(0) >> shift) | (op1->xmm64u(1) << (64-shift));
|
|
}
|
|
else if (shift == 64) {
|
|
op2->xmm64u(0) = op2->xmm64u(1);
|
|
op2->xmm64u(1) = op1->xmm64u(0);
|
|
}
|
|
else if (shift != 0) {
|
|
op2->xmm64u(0) = (op2->xmm64u(0) >> shift) | (op2->xmm64u(1) << (64-shift));
|
|
op2->xmm64u(1) = (op2->xmm64u(1) >> shift) | (op1->xmm64u(0) << (64-shift));
|
|
}
|
|
}
|
|
|
|
// rotate (right)
|
|
|
|
BX_CPP_INLINE void xmm_prorb(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x7;
|
|
|
|
for(unsigned n=0;n<16;n++) {
|
|
op->xmmubyte(n) = (op->xmmubyte(n) >> shift) | (op->xmmubyte(n) << (8 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prorw(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0xf;
|
|
|
|
for(unsigned n=0;n<8;n++) {
|
|
op->xmm16u(n) = (op->xmm16u(n) >> shift) | (op->xmm16u(n) << (16 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prord(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x1f;
|
|
|
|
for(unsigned n=0;n<4;n++) {
|
|
op->xmm32u(n) = (op->xmm32u(n) >> shift) | (op->xmm32u(n) << (32 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prorq(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x3f;
|
|
|
|
for(unsigned n=0;n<2;n++) {
|
|
op->xmm64u(n) = (op->xmm64u(n) >> shift) | (op->xmm64u(n) << (64 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prorvd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n<4;n++) {
|
|
int shift = op2->xmm32u(n) & 0x1f;
|
|
op1->xmm32u(n) = (op1->xmm32u(n) >> shift) | (op1->xmm32u(n) << (32 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prorvq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n<2;n++) {
|
|
int shift = op2->xmm64u(n) & 0x3f;
|
|
op1->xmm64u(n) = (op1->xmm64u(n) >> shift) | (op1->xmm64u(n) << (64 - shift));
|
|
}
|
|
}
|
|
|
|
// rotate (left)
|
|
|
|
BX_CPP_INLINE void xmm_prolb(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x7;
|
|
|
|
for(unsigned n=0;n<16;n++) {
|
|
op->xmmubyte(n) = (op->xmmubyte(n) << shift) | (op->xmmubyte(n) >> (8 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prolw(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0xf;
|
|
|
|
for(unsigned n=0;n<8;n++) {
|
|
op->xmm16u(n) = (op->xmm16u(n) << shift) | (op->xmm16u(n) >> (16 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prold(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x1f;
|
|
|
|
for(unsigned n=0;n<4;n++) {
|
|
op->xmm32u(n) = (op->xmm32u(n) << shift) | (op->xmm32u(n) >> (32 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prolq(BxPackedXmmRegister *op, int shift)
|
|
{
|
|
shift &= 0x3f;
|
|
|
|
for(unsigned n=0;n<2;n++) {
|
|
op->xmm64u(n) = (op->xmm64u(n) << shift) | (op->xmm64u(n) >> (64 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prolvd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n<4;n++) {
|
|
int shift = op2->xmm32u(n) & 0x1f;
|
|
op1->xmm32u(n) = (op1->xmm32u(n) << shift) | (op1->xmm32u(n) >> (32 - shift));
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_prolvq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n<2;n++) {
|
|
int shift = op2->xmm64u(n) & 0x3f;
|
|
op1->xmm64u(n) = (op1->xmm64u(n) << shift) | (op1->xmm64u(n) >> (64 - shift));
|
|
}
|
|
}
|
|
|
|
// variable shift/rotate (XOP)
|
|
|
|
BX_CPP_INLINE void xmm_protb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 16;n++) {
|
|
int shift = op2->xmmsbyte(n);
|
|
if (shift > 0) {
|
|
// rotate left
|
|
shift &= 0x7;
|
|
op1->xmmubyte(n) = (op1->xmmubyte(n) << shift) | (op1->xmmubyte(n) >> (8 - shift));
|
|
}
|
|
else if (shift < 0) {
|
|
// rotate right
|
|
shift = -shift & 0x7;
|
|
op1->xmmubyte(n) = (op1->xmmubyte(n) >> shift) | (op1->xmmubyte(n) << (8 - shift));
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_protw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 8;n++) {
|
|
int shift = op2->xmmsbyte(n*2);
|
|
if (shift > 0) {
|
|
// rotate left
|
|
shift &= 0xf;
|
|
op1->xmm16u(n) = (op1->xmm16u(n) << shift) | (op1->xmm16u(n) >> (16 - shift));
|
|
}
|
|
else if (shift < 0) {
|
|
// rotate right
|
|
shift = -shift & 0xf;
|
|
op1->xmm16u(n) = (op1->xmm16u(n) >> shift) | (op1->xmm16u(n) << (16 - shift));
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_protd(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 4;n++) {
|
|
int shift = op2->xmmsbyte(n*4);
|
|
if (shift > 0) {
|
|
// rotate left
|
|
shift &= 0x1f;
|
|
op1->xmm32u(n) = (op1->xmm32u(n) << shift) | (op1->xmm32u(n) >> (32 - shift));
|
|
}
|
|
else if (shift < 0) {
|
|
// rotate right
|
|
shift = -shift & 0x1f;
|
|
op1->xmm32u(n) = (op1->xmm32u(n) >> shift) | (op1->xmm32u(n) << (32 - shift));
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_protq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 2;n++) {
|
|
int shift = op2->xmmsbyte(n*8);
|
|
if (shift > 0) {
|
|
// rotate left
|
|
shift &= 0x3f;
|
|
op1->xmm64u(n) = (op1->xmm64u(n) << shift) | (op1->xmm64u(n) >> (64 - shift));
|
|
}
|
|
else if (shift < 0) {
|
|
// rotate right
|
|
shift = -shift & 0x3f;
|
|
op1->xmm64u(n) = (op1->xmm64u(n) >> shift) | (op1->xmm64u(n) << (64 - shift));
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshab(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 16;n++) {
|
|
int shift = op2->xmmsbyte(n);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmmsbyte(n) <<= (shift & 0x7);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmmsbyte(n) >>= (-shift & 0x7);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshaw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 8;n++) {
|
|
int shift = op2->xmmsbyte(n*2);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm16s(n) <<= (shift & 0xf);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm16s(n) >>= (-shift & 0xf);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshad(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 4;n++) {
|
|
int shift = op2->xmmsbyte(n*4);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm32s(n) <<= (shift & 0x1f);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm32s(n) >>= (-shift & 0x1f);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshaq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 2;n++) {
|
|
int shift = op2->xmmsbyte(n*8);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm64s(n) <<= (shift & 0x3f);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm64s(n) >>= (-shift & 0x3f);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshlb(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 16;n++) {
|
|
int shift = op2->xmmsbyte(n);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmmubyte(n) <<= (shift & 0x7);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmmubyte(n) >>= (-shift & 0x7);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshlw(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 8;n++) {
|
|
int shift = op2->xmmubyte(n*2);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm16u(n) <<= (shift & 0xf);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm16u(n) >>= (-shift & 0xf);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshld(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 4;n++) {
|
|
int shift = op2->xmmsbyte(n*4);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm32u(n) <<= (shift & 0x1f);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm32u(n) >>= (-shift & 0x1f);
|
|
}
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pshlq(BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0;n < 2;n++) {
|
|
int shift = op2->xmmsbyte(n*8);
|
|
if (shift > 0) {
|
|
// shift left
|
|
op1->xmm64u(n) <<= (shift & 0x3f);
|
|
}
|
|
else if (shift < 0) {
|
|
// shift right
|
|
op1->xmm64u(n) >>= (-shift & 0x3f);
|
|
}
|
|
}
|
|
}
|
|
|
|
// VNNI
|
|
|
|
BX_CPP_INLINE void xmm_pdpbusd(BxPackedXmmRegister *dst, BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++)
|
|
{
|
|
Bit32s p1word = (Bit32u) op1->xmmubyte(n*4) * (Bit32s) op2->xmmsbyte(n*4);
|
|
Bit32s p2word = (Bit32u) op1->xmmubyte(n*4+1) * (Bit32s) op2->xmmsbyte(n*4+1);
|
|
Bit32s p3word = (Bit32u) op1->xmmubyte(n*4+2) * (Bit32s) op2->xmmsbyte(n*4+2);
|
|
Bit32s p4word = (Bit32u) op1->xmmubyte(n*4+3) * (Bit32s) op2->xmmsbyte(n*4+3);
|
|
|
|
dst->xmm32s(n) += (p1word + p2word + p3word + p4word);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pdpbusds(BxPackedXmmRegister *dst, BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++)
|
|
{
|
|
Bit32s p1word = (Bit32u) op1->xmmubyte(n*4) * (Bit32s) op2->xmmsbyte(n*4);
|
|
Bit32s p2word = (Bit32u) op1->xmmubyte(n*4+1) * (Bit32s) op2->xmmsbyte(n*4+1);
|
|
Bit32s p3word = (Bit32u) op1->xmmubyte(n*4+2) * (Bit32s) op2->xmmsbyte(n*4+2);
|
|
Bit32s p4word = (Bit32u) op1->xmmubyte(n*4+3) * (Bit32s) op2->xmmsbyte(n*4+3);
|
|
|
|
Bit64s result = (Bit64s) dst->xmm32s(n) + (p1word + p2word + p3word + p4word);
|
|
dst->xmm32s(n) = SaturateQwordSToDwordS(result);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pdpwssd(BxPackedXmmRegister *dst, BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++)
|
|
{
|
|
Bit32s p1_dword = (Bit32s) op1->xmm16s(n*2) * (Bit32s) op2->xmm16s(n*2);
|
|
Bit32s p2_dword = (Bit32s) op1->xmm16s(n*2+1) * (Bit32s) op2->xmm16s(n*2+1);
|
|
|
|
dst->xmm32s(n) += (p1_dword + p2_dword);
|
|
}
|
|
}
|
|
|
|
BX_CPP_INLINE void xmm_pdpwssds(BxPackedXmmRegister *dst, BxPackedXmmRegister *op1, const BxPackedXmmRegister *op2)
|
|
{
|
|
for(unsigned n=0; n<4; n++)
|
|
{
|
|
Bit32s p1_dword = (Bit32s) op1->xmm16s(n*2) * (Bit32s) op2->xmm16s(n*2);
|
|
Bit32s p2_dword = (Bit32s) op1->xmm16s(n*2+1) * (Bit32s) op2->xmm16s(n*2+1);
|
|
|
|
Bit64s result = (Bit64s) dst->xmm32s(n) + (p1_dword + p2_dword);
|
|
dst->xmm32s(n) = SaturateQwordSToDwordS(result);
|
|
}
|
|
}
|
|
|
|
#endif
|