Skip to content

Commit

Permalink
use macroses to implement AVX convert functions
Browse files Browse the repository at this point in the history 
helps to reduce code duplication
  • Loading branch information
Stanislav Shwartsman committed Nov 5, 2024
1 parent 9114302 commit 4648bdf
Showing 1 changed file with 67 additions and 145 deletions.
212 changes: 67 additions & 145 deletions bochs/cpu/avx/avx_cvt.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -90,50 +90,79 @@ void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTSI2SD_VsdEqR(bxInstruction_c *i)
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.0F 5A (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTPS2PD_VpdWpsR(bxInstruction_c *i)
{
BxPackedAvxRegister result;
BxPackedYmmRegister op = BX_READ_YMM_REG(i->src());
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) {
result.vmm64u(n) = f32_to_f64(op.ymm32u(n), &status);
// packed

#define AVX_CVT32_TO_64(HANDLER, func) \
void BX_CPP_AttrRegparmN(1) BX_CPU_C:: HANDLER (bxInstruction_c *i) \
{ \
BxPackedAvxRegister result; \
BxPackedYmmRegister op = BX_READ_YMM_REG(i->src()); \
unsigned len = i->getVL(); \
\
softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR); \
softfloat_status_word_rc_override(status, i); \
\
for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) { \
result.vmm64u(n) = (func)(op.ymm32u(n), &status); \
} \
\
check_exceptionsSSE(softfloat_getExceptionFlags(&status)); \
\
BX_WRITE_AVX_REGZ(i->dst(), result, len); \
BX_NEXT_INSTR(i); \
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

BX_WRITE_AVX_REGZ(i->dst(), result, len);
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.66.0F 5A (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTPD2PS_VpsWpdR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src()), result;
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) {
result.vmm32u(n) = f64_to_f32(op.vmm64u(n), &status);
AVX_CVT32_TO_64(VCVTPS2PD_VpdWpsR, f32_to_f64)

#define AVX_CVT64_TO_32(HANDLER, func) \
void BX_CPP_AttrRegparmN(1) BX_CPU_C:: HANDLER (bxInstruction_c *i) \
{ \
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src()), result; \
unsigned len = i->getVL(); \
\
softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR); \
softfloat_status_word_rc_override(status, i); \
\
for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) { \
result.vmm32u(n) = (func)(op.vmm64u(n), &status); \
} \
\
check_exceptionsSSE(softfloat_getExceptionFlags(&status)); \
\
if (len == BX_VL128) { \
BX_WRITE_XMM_REG_LO_QWORD_CLEAR_HIGH(i->dst(), result.vmm64u(0)); \
} else { \
BX_WRITE_AVX_REGZ(i->dst(), result, len >> 1); /* write half vector */ \
} \
\
BX_NEXT_INSTR(i); \
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

if (len == BX_VL128) {
BX_WRITE_XMM_REG_LO_QWORD_CLEAR_HIGH(i->dst(), result.vmm64u(0));
}
else {
BX_WRITE_AVX_REGZ(i->dst(), result, len >> 1); // write half vector
AVX_CVT64_TO_32(VCVTPD2PS_VpsWpdR, f64_to_f32)
AVX_CVT64_TO_32(VCVTPD2DQ_VdqWpdR, f64_to_i32)
AVX_CVT64_TO_32(VCVTTPD2DQ_VdqWpdR, f64_to_i32_round_to_zero)

#define AVX_CVT32_TO_32(HANDLER, func) \
void BX_CPP_AttrRegparmN(1) BX_CPU_C:: HANDLER (bxInstruction_c *i) \
{ \
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src()); \
unsigned len = i->getVL(); \
\
softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR); \
softfloat_status_word_rc_override(status, i); \
\
for (unsigned n=0; n < DWORD_ELEMENTS(len); n++) { \
op.vmm32u(n) = (func)(op.vmm32u(n), &status); \
} \
\
check_exceptionsSSE(softfloat_getExceptionFlags(&status)); \
BX_WRITE_AVX_REGZ(i->dst(), op, len); \
BX_NEXT_INSTR(i); \
}

BX_NEXT_INSTR(i);
}
AVX_CVT32_TO_32(VCVTDQ2PS_VpsWdqR, i32_to_f32)
AVX_CVT32_TO_32(VCVTPS2DQ_VdqWpsR, f32_to_i32)
AVX_CVT32_TO_32(VCVTTPS2DQ_VdqWpsR, f32_to_i32_round_to_zero)

/* Opcode: VEX.F3.0F 5A (VEX.W ignore) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTSS2SD_VsdWssR(bxInstruction_c *i)
Expand Down Expand Up @@ -165,113 +194,6 @@ void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTSD2SS_VssWsdR(bxInstruction_c *i)
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.NDS.0F 5B (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTDQ2PS_VpsWdqR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src());
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < DWORD_ELEMENTS(len); n++) {
op.vmm32u(n) = i32_to_f32(op.vmm32s(n), &status);
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

BX_WRITE_AVX_REGZ(i->dst(), op, len);
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.NDS.66.0F 5B (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTPS2DQ_VdqWpsR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src());
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < DWORD_ELEMENTS(len); n++) {
op.vmm32s(n) = f32_to_i32(op.vmm32u(n), &status);
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

BX_WRITE_AVX_REGZ(i->dst(), op, len);
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.NDS.F3.0F 5B (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTTPS2DQ_VdqWpsR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src());
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < DWORD_ELEMENTS(len); n++) {
op.vmm32s(n) = f32_to_i32_round_to_zero(op.vmm32u(n), &status);
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

BX_WRITE_AVX_REGZ(i->dst(), op, len);
BX_NEXT_INSTR(i);
}

/* Opcode: VEX.66.0F.E6 (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTTPD2DQ_VdqWpdR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src()), result;
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) {
result.vmm32s(n) = f64_to_i32_round_to_zero(op.vmm64u(n), &status);
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

if (len == BX_VL128) {
BX_WRITE_XMM_REG_LO_QWORD_CLEAR_HIGH(i->dst(), result.vmm64u(0));
}
else {
BX_WRITE_AVX_REGZ(i->dst(), result, len >> 1); // write half vector
}

BX_NEXT_INSTR(i);
}

/* Opcode: VEX.F2.0F.E6 (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTPD2DQ_VdqWpdR(bxInstruction_c *i)
{
BxPackedAvxRegister op = BX_READ_AVX_REG(i->src()), result;
unsigned len = i->getVL();

softfloat_status_t status = mxcsr_to_softfloat_status_word(MXCSR);
softfloat_status_word_rc_override(status, i);

for (unsigned n=0; n < QWORD_ELEMENTS(len); n++) {
result.vmm32s(n) = f64_to_i32(op.vmm64u(n), &status);
}

check_exceptionsSSE(softfloat_getExceptionFlags(&status));

if (len == BX_VL128) {
BX_WRITE_XMM_REG_LO_QWORD_CLEAR_HIGH(i->dst(), result.vmm64u(0));
}
else {
BX_WRITE_AVX_REGZ(i->dst(), result, len >> 1); // write half vector
}

BX_NEXT_INSTR(i);
}

/* Opcode: VEX.F3.0F.E6 (VEX.W ignore, VEX.VVV #UD) */
void BX_CPP_AttrRegparmN(1) BX_CPU_C::VCVTDQ2PD_VpdWdqR(bxInstruction_c *i)
{
Expand Down

0 comments on commit 4648bdf

Please sign in to comment.