Bochs/bochs/cpu/avx/avx512_bf16.cc
Stanislav Shwartsman 1255a0c585 remove fpu/softfloat.h, reduce include deprendencies, fix warnings
regenerate dep lists in all Makefile.in
2024-04-25 11:50:48 +03:00

109 lines
3.9 KiB
C++

/////////////////////////////////////////////////////////////////////////
// $Id$
/////////////////////////////////////////////////////////////////////////
//
// Copyright (c) 2023 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
//
/////////////////////////////////////////////////////////////////////////
#define NEED_CPU_REG_SHORTCUTS 1
#include "bochs.h"
#include "cpu.h"
#define LOG_THIS BX_CPU_THIS_PTR
#if BX_SUPPORT_EVEX
#include "bf16.h"
#include "simd_int.h"
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTNEPS2BF16_MASK_Vbf16WpsR(bxInstruction_c *i)
{
BxPackedAvxRegister src = BX_READ_AVX_REG(i->src()), dst;
unsigned len = i->getVL();
dst.clear();
Bit32u opmask = (i->opmask() != 0) ? BX_READ_16BIT_OPMASK(i->opmask()) : 0xffff; // mask according to fp32 source
opmask &= CUT_OPMASK_TO(DWORD_ELEMENTS(len));
for (unsigned n=0, mask = 0x1; n < DWORD_ELEMENTS(len); n++, mask <<= 1) {
if (opmask & mask)
dst.vmm16u(n) = convert_ne_fp32_to_bfloat16(src.vmm32u(n));
}
if (! i->isZeroMasking()) {
for (unsigned n=0; n < len; n++, opmask >>= 8)
xmm_pblendw(&BX_READ_AVX_REG_LANE(i->dst(), n), &dst.vmm128(n), opmask);
}
BX_WRITE_AVX_REGZ(i->dst(), dst, len);
BX_NEXT_INSTR(i);
}
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTNE2PS2BF16_MASK_Vbf16HpsWpsR(bxInstruction_c *i)
{
BxPackedAvxRegister op1 = BX_READ_AVX_REG(i->src1()), op2 = BX_READ_AVX_REG(i->src2()), dst;
unsigned len = i->getVL();
unsigned n=0;
// half of the elements
for (n=0; n < DWORD_ELEMENTS(len); n++)
dst.vmm16u(n) = convert_ne_fp32_to_bfloat16(op2.vmm32u(n));
// other half of the elements
for (;n < WORD_ELEMENTS(len); n++)
dst.vmm16u(n) = convert_ne_fp32_to_bfloat16(op1.vmm32u(n));
if (i->opmask()) {
avx512_write_regw_masked(i, &dst, len, BX_READ_32BIT_OPMASK(i->opmask()));
}
else {
BX_WRITE_AVX_REGZ(i->dst(), dst, len);
}
BX_NEXT_INSTR(i);
}
extern softfloat_status_t prepare_ne_softfloat_status_helper();
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VDPBF16PS_MASK_VpsHdqWdqR(bxInstruction_c *i)
{
BxPackedAvxRegister op1 = BX_READ_AVX_REG(i->src1()), op2 = BX_READ_AVX_REG(i->src2()), dst = BX_READ_AVX_REG(i->dst());
unsigned len = i->getVL();
Bit32u mask = (i->opmask() != 0) ? BX_READ_16BIT_OPMASK(i->opmask()) : 0xffff;
// "round to nearest even" rounding mode is used when doing each accumulation of the FMA.
// output denormals are always flushed to zero and input denormals are always treated as zero.
softfloat_status_t status = prepare_ne_softfloat_status_helper();
status.softfloat_denormals_are_zeros = true;
for (unsigned n=0, tmp_mask = mask; n < DWORD_ELEMENTS(len); n++, tmp_mask >>= 1) {
if (tmp_mask & 0x1) {
dst.vmm32u(n) = f32_mulAdd(convert_bfloat16_to_fp32(op1.vmm32u(n) >> 16), convert_bfloat16_to_fp32(op2.vmm32u(n) >> 16), dst.vmm32u(n), 0, &status);
dst.vmm32u(n) = f32_mulAdd(convert_bfloat16_to_fp32(op1.vmm32u(n)), convert_bfloat16_to_fp32(op2.vmm32u(n)), dst.vmm32u(n), 0, &status);
}
else if (i->isZeroMasking()) {
dst.vmm32u(n) = 0;
}
}
BX_WRITE_AVX_REGZ(i->dst(), dst, len);
BX_NEXT_INSTR(i);
}
#endif