Archive/rpcs3
0
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-04-28 13:28:01 +03:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

BufferUtils: rewrite remaining intrinsic code with simd_builder

Browse source
This commit is contained in:
Nekotekina 2022-09-04 14:23:14 +03:00 • committed by Ivan
parent 11a1f090d3
commit 82258915da
3 changed files with 220 additions and 249 deletions
Download patch file Download diff file Expand all files Collapse all files

81
Utilities/JIT.cpp
View file

@ -508,6 +508,54 @@ void asmjit::simd_builder::vec_clobbering_test(u32 esize, const Operand& v, cons
}
}
void asmjit::simd_builder::vec_broadcast_gpr(u32 esize, const Operand& v, const x86::Gp& r)
{
if (esize == 2)
{
if (utils::has_avx512())
this->emit(x86::Inst::kIdVpbroadcastw, v, r.r32());
else if (utils::has_avx())
{
this->emit(x86::Inst::kIdVmovd, v, r.r32());
if (utils::has_avx2())
this->emit(x86::Inst::kIdVpbroadcastw, v, v);
else
{
this->emit(x86::Inst::kIdVpunpcklwd, v, v);
this->emit(x86::Inst::kIdVpshufd, v, v, Imm(0));
}
}
else
{
this->emit(x86::Inst::kIdMovd, v, r.r32());
this->emit(x86::Inst::kIdPunpcklwd, v, v);
this->emit(x86::Inst::kIdPshufd, v, v, Imm(0));
}
}
else if (esize == 4)
{
if (utils::has_avx512())
this->emit(x86::Inst::kIdVpbroadcastd, v, r.r32());
else if (utils::has_avx())
{
this->emit(x86::Inst::kIdVmovd, v, r.r32());
if (utils::has_avx2())
this->emit(x86::Inst::kIdVpbroadcastd, v, v);
else
this->emit(x86::Inst::kIdVpshufd, v, v, Imm(0));
}
else
{
this->emit(x86::Inst::kIdMovd, v, r.r32());
this->emit(x86::Inst::kIdPshufd, v, v, Imm(0));
}
}
else
{
fmt::throw_exception("Unimplemented");
}
}
asmjit::x86::Mem asmjit::simd_builder::ptr_scale_for_vec(u32 esize, const x86::Gp& base, const x86::Gp& index)
{
switch (ensure(esize))
@ -698,6 +746,39 @@ void asmjit::simd_builder::vec_umax(u32 esize, const Operand& dst, const Operand
fmt::throw_exception("Unimplemented");
}
void asmjit::simd_builder::vec_cmp_eq(u32 esize, const Operand& dst, const Operand& lhs, const Operand& rhs)
{
using enum x86::Inst::Id;
if (esize == 2)
{
if (vsize == 64)
{
this->evex().emit(kIdVpcmpeqw, x86::k0, lhs, rhs);
this->evex().emit(kIdVpmovm2w, dst, x86::k0);
}
else
{
_vec_binary_op(kIdPcmpeqw, kIdVpcmpeqw, kIdNone, dst, lhs, rhs);
}
}
else if (esize == 4)
{
if (vsize == 64)
{
this->evex().emit(kIdVpcmpeqd, x86::k0, lhs, rhs);
this->evex().emit(kIdVpmovm2d, dst, x86::k0);
}
else
{
_vec_binary_op(kIdPcmpeqw, kIdVpcmpeqw, kIdNone, dst, lhs, rhs);
}
}
else
{
fmt::throw_exception("Unimplemented");
}
}
void asmjit::simd_builder::vec_extract_high(u32, const Operand& dst, const Operand& src)
{
if (vsize == 32)

8
Utilities/JIT.h
View file

@ -232,6 +232,7 @@ namespace asmjit
void vec_set_all_ones(const Operand& v);
void vec_set_const(const Operand& v, const v128& value);
void vec_clobbering_test(u32 esize, const Operand& v, const Operand& rhs);
void vec_broadcast_gpr(u32 esize, const Operand& v, const x86::Gp& r);
// return x86::ptr(base, ctr, X, 0) where X is set for esize accordingly
x86::Mem ptr_scale_for_vec(u32 esize, const x86::Gp& base, const x86::Gp& index);
@ -260,8 +261,15 @@ namespace asmjit
_vec_binary_op(kIdPor, kIdVpor, kIdVpord, dst, lhs, rhs);
}
void vec_andn(u32, const Operand& dst, const Operand& lhs, const Operand& rhs)
{
using enum x86::Inst::Id;
_vec_binary_op(kIdPandn, kIdVpandn, kIdVpandnd, dst, lhs, rhs);
}
void vec_umin(u32 esize, const Operand& dst, const Operand& lhs, const Operand& rhs);
void vec_umax(u32 esize, const Operand& dst, const Operand& lhs, const Operand& rhs);
void vec_cmp_eq(u32 esize, const Operand& dst, const Operand& lhs, const Operand& rhs);
void vec_extract_high(u32 esize, const Operand& dst, const Operand& src);
void vec_extract_gpr(u32 esize, const x86::Gp& dst, const Operand& src);