yuzu/unicorn
1
0
Fork 0
mirror of https://github.com/yuzu-emu/unicorn.git synced 2024-12-23 09:35:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

tcg/i386: Implement vector minmax arithmetic

Browse source 
The avx instruction set does not directly provide MO_64.
We can still implement 64-bit with comparison and vpblendvb.

Backports commit bc37faf4cb2baa77c44298c01558970b88d32808 from qemu
This commit is contained in:
Richard Henderson 2019-01-29 16:41:07 -05:00 committed by Lioncash
parent 5518b543ed
commit 63d1aae6b2
No known key found for this signature in database
GPG key ID: 4E3C3CC1031BA9C7
2 changed files with 82 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
qemu/tcg/i386/tcg-target.h
View file

@ -219,7 +219,7 @@ extern bool have_avx2;
#define TCG_TARGET_HAS_cmp_vec 1
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 0
#define TCG_TARGET_HAS_minmax_vec 1
#define TCG_TARGET_deposit_i32_valid(ofs, len) \
(((ofs) == 0 && (len) == 8) || ((ofs) == 8 && (len) == 8) || \

81
qemu/tcg/i386/tcg-target.inc.c
View file

@ -399,6 +399,18 @@ static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
#define OPC_PCMPGTW (0x65 | P_EXT | P_DATA16)
#define OPC_PCMPGTD (0x66 | P_EXT | P_DATA16)
#define OPC_PCMPGTQ (0x37 | P_EXT38 | P_DATA16)
#define OPC_PMAXSB (0x3c | P_EXT38 | P_DATA16)
#define OPC_PMAXSW (0xee | P_EXT | P_DATA16)
#define OPC_PMAXSD (0x3d | P_EXT38 | P_DATA16)
#define OPC_PMAXUB (0xde | P_EXT | P_DATA16)
#define OPC_PMAXUW (0x3e | P_EXT38 | P_DATA16)
#define OPC_PMAXUD (0x3f | P_EXT38 | P_DATA16)
#define OPC_PMINSB (0x38 | P_EXT38 | P_DATA16)
#define OPC_PMINSW (0xea | P_EXT | P_DATA16)
#define OPC_PMINSD (0x39 | P_EXT38 | P_DATA16)
#define OPC_PMINUB (0xda | P_EXT | P_DATA16)
#define OPC_PMINUW (0x3a | P_EXT38 | P_DATA16)
#define OPC_PMINUD (0x3b | P_EXT38 | P_DATA16)
#define OPC_PMOVSXBW (0x20 | P_EXT38 | P_DATA16)
#define OPC_PMOVSXWD (0x23 | P_EXT38 | P_DATA16)
#define OPC_PMOVSXDQ (0x25 | P_EXT38 | P_DATA16)
@ -2761,6 +2773,18 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
static int const packus_insn[4] = {
OPC_PACKUSWB, OPC_PACKUSDW, OPC_UD2, OPC_UD2
};
static int const smin_insn[4] = {
OPC_PMINSB, OPC_PMINSW, OPC_PMINSD, OPC_UD2
};
static int const smax_insn[4] = {
OPC_PMAXSB, OPC_PMAXSW, OPC_PMAXSD, OPC_UD2
};
static int const umin_insn[4] = {
OPC_PMINUB, OPC_PMINUW, OPC_PMINUD, OPC_UD2
};
static int const umax_insn[4] = {
OPC_PMAXUB, OPC_PMAXUW, OPC_PMAXUD, OPC_UD2
};
TCGType type = vecl + TCG_TYPE_V64;
int insn, sub;
@ -2801,6 +2825,18 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
case INDEX_op_xor_vec:
insn = OPC_PXOR;
goto gen_simd;
case INDEX_op_smin_vec:
insn = smin_insn[vece];
goto gen_simd;
case INDEX_op_umin_vec:
insn = umin_insn[vece];
goto gen_simd;
case INDEX_op_smax_vec:
insn = smax_insn[vece];
goto gen_simd;
case INDEX_op_umax_vec:
insn = umax_insn[vece];
goto gen_simd;
case INDEX_op_x86_punpckl_vec:
insn = punpckl_insn[vece];
goto gen_simd;
@ -3153,6 +3189,10 @@ static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
case INDEX_op_usadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_ussub_vec:
case INDEX_op_smin_vec:
case INDEX_op_umin_vec:
case INDEX_op_smax_vec:
case INDEX_op_umax_vec:
case INDEX_op_cmp_vec:
case INDEX_op_x86_shufps_vec:
case INDEX_op_x86_blend_vec:
@ -3225,6 +3265,11 @@ int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
case INDEX_op_sssub_vec:
case INDEX_op_ussub_vec:
return vece <= MO_16;
case INDEX_op_smin_vec:
case INDEX_op_smax_vec:
case INDEX_op_umin_vec:
case INDEX_op_umax_vec:
return vece <= MO_32 ? 1 : -1;
default:
return 0;
@ -3453,6 +3498,25 @@ static void expand_vec_cmp(TCGContext *s, TCGType type, unsigned vece, TCGv_vec
}
}
static void expand_vec_minmax(TCGContext *s, TCGType type, unsigned vece,
TCGCond cond, bool min,
TCGv_vec v0, TCGv_vec v1, TCGv_vec v2)
{
TCGv_vec t1 = tcg_temp_new_vec(s, type);
tcg_debug_assert(vece == MO_64);
tcg_gen_cmp_vec(s, cond, vece, t1, v1, v2);
if (min) {
TCGv_vec t2;
t2 = v1, v1 = v2, v2 = t2;
}
vec_gen_4(s, INDEX_op_x86_vpblendvb_vec, type, vece,
tcgv_vec_arg(s, v0), tcgv_vec_arg(s, v1),
tcgv_vec_arg(s, v2), tcgv_vec_arg(s, t1));
tcg_temp_free_vec(s, t1);
}
void tcg_expand_vec_op(TCGContext *s, TCGOpcode opc, TCGType type, unsigned vece,
TCGArg a0, ...)
{
@ -3485,6 +3549,23 @@ void tcg_expand_vec_op(TCGContext *s, TCGOpcode opc, TCGType type, unsigned vece
expand_vec_cmp(s, type, vece, v0, v1, v2, va_arg(va, TCGArg));
break;
case INDEX_op_smin_vec:
v2 = temp_tcgv_vec(s, arg_temp(a2));
expand_vec_minmax(s, type, vece, TCG_COND_GT, true, v0, v1, v2);
break;
case INDEX_op_smax_vec:
v2 = temp_tcgv_vec(s, arg_temp(a2));
expand_vec_minmax(s, type, vece, TCG_COND_GT, false, v0, v1, v2);
break;
case INDEX_op_umin_vec:
v2 = temp_tcgv_vec(s, arg_temp(a2));
expand_vec_minmax(s, type, vece, TCG_COND_GTU, true, v0, v1, v2);
break;
case INDEX_op_umax_vec:
v2 = temp_tcgv_vec(s, arg_temp(a2));
expand_vec_minmax(s, type, vece, TCG_COND_GTU, false, v0, v1, v2);
break;
default:
break;
}