Skip to content

Add f16 and f128 inline ASM support for x86 and x86-64 #126417

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 15, 2024
Merged

Add f16 and f128 inline ASM support for x86 and x86-64 #126417

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
100 changes: 100 additions & 0 deletions compiler/rustc_codegen_llvm/src/asm.rs
View file
Original file line number Diff line number Diff line change
@@ -959,6 +959,43 @@ fn llvm_fixup_input<'ll, 'tcx>(
InlineAsmRegClass::X86(X86InlineAsmRegClass::xmm_reg | X86InlineAsmRegClass::zmm_reg),
Abi::Vector { .. },
) if layout.size.bytes() == 64 => bx.bitcast(value, bx.cx.type_vector(bx.cx.type_f64(), 8)),
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if bx.sess().asm_arch == Some(InlineAsmArch::X86)
&& s.primitive() == Primitive::Float(Float::F128) =>
{
bx.bitcast(value, bx.type_vector(bx.type_i32(), 4))
}
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if s.primitive() == Primitive::Float(Float::F16) => {
let value = bx.insert_element(
bx.const_undef(bx.type_vector(bx.type_f16(), 8)),
value,
bx.const_usize(0),
);
bx.bitcast(value, bx.type_vector(bx.type_i16(), 8))
}
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Vector { element, count: count @ (8 | 16) },
) if element.primitive() == Primitive::Float(Float::F16) => {
bx.bitcast(value, bx.type_vector(bx.type_i16(), count))
}
(
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg | ArmInlineAsmRegClass::sreg_low16),
Abi::Scalar(s),
@@ -1036,6 +1073,39 @@ fn llvm_fixup_output<'ll, 'tcx>(
InlineAsmRegClass::X86(X86InlineAsmRegClass::xmm_reg | X86InlineAsmRegClass::zmm_reg),
Abi::Vector { .. },
) if layout.size.bytes() == 64 => bx.bitcast(value, layout.llvm_type(bx.cx)),
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if bx.sess().asm_arch == Some(InlineAsmArch::X86)
&& s.primitive() == Primitive::Float(Float::F128) =>
{
bx.bitcast(value, bx.type_f128())
}
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if s.primitive() == Primitive::Float(Float::F16) => {
let value = bx.bitcast(value, bx.type_vector(bx.type_f16(), 8));
bx.extract_element(value, bx.const_usize(0))
}
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Vector { element, count: count @ (8 | 16) },
) if element.primitive() == Primitive::Float(Float::F16) => {
bx.bitcast(value, bx.type_vector(bx.type_f16(), count))
}
(
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg | ArmInlineAsmRegClass::sreg_low16),
Abi::Scalar(s),
@@ -1109,6 +1179,36 @@ fn llvm_fixup_output_type<'ll, 'tcx>(
InlineAsmRegClass::X86(X86InlineAsmRegClass::xmm_reg | X86InlineAsmRegClass::zmm_reg),
Abi::Vector { .. },
) if layout.size.bytes() == 64 => cx.type_vector(cx.type_f64(), 8),
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if cx.sess().asm_arch == Some(InlineAsmArch::X86)
&& s.primitive() == Primitive::Float(Float::F128) =>
{
cx.type_vector(cx.type_i32(), 4)
}
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Scalar(s),
) if s.primitive() == Primitive::Float(Float::F16) => cx.type_vector(cx.type_i16(), 8),
(
InlineAsmRegClass::X86(
X86InlineAsmRegClass::xmm_reg
| X86InlineAsmRegClass::ymm_reg
| X86InlineAsmRegClass::zmm_reg,
),
Abi::Vector { element, count: count @ (8 | 16) },
) if element.primitive() == Primitive::Float(Float::F16) => {
cx.type_vector(cx.type_i16(), count)
}
(
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg | ArmInlineAsmRegClass::sreg_low16),
Abi::Scalar(s),
4 changes: 4 additions & 0 deletions compiler/rustc_hir_analysis/src/check/intrinsicck.rs
View file
Original file line number Diff line number Diff line change
@@ -62,8 +62,10 @@ impl<'a, 'tcx> InlineAsmCtxt<'a, 'tcx> {
ty::Int(IntTy::I64) | ty::Uint(UintTy::U64) => Some(InlineAsmType::I64),
ty::Int(IntTy::I128) | ty::Uint(UintTy::U128) => Some(InlineAsmType::I128),
ty::Int(IntTy::Isize) | ty::Uint(UintTy::Usize) => Some(asm_ty_isize),
ty::Float(FloatTy::F16) => Some(InlineAsmType::F16),
ty::Float(FloatTy::F32) => Some(InlineAsmType::F32),
ty::Float(FloatTy::F64) => Some(InlineAsmType::F64),
ty::Float(FloatTy::F128) => Some(InlineAsmType::F128),
ty::FnPtr(_) => Some(asm_ty_isize),
ty::RawPtr(ty, _) if self.is_thin_ptr_ty(ty) => Some(asm_ty_isize),
ty::Adt(adt, args) if adt.repr().simd() => {
@@ -105,8 +107,10 @@ impl<'a, 'tcx> InlineAsmCtxt<'a, 'tcx> {
width => bug!("unsupported pointer width: {width}"),
})
}
ty::Float(FloatTy::F16) => Some(InlineAsmType::VecF16(size)),
ty::Float(FloatTy::F32) => Some(InlineAsmType::VecF32(size)),
ty::Float(FloatTy::F64) => Some(InlineAsmType::VecF64(size)),
ty::Float(FloatTy::F128) => Some(InlineAsmType::VecF128(size)),
_ => None,
}
}
12 changes: 12 additions & 0 deletions compiler/rustc_target/src/asm/mod.rs
View file
Original file line number Diff line number Diff line change
@@ -707,15 +707,19 @@ pub enum InlineAsmType {
I32,
I64,
I128,
F16,
F32,
F64,
F128,
VecI8(u64),
VecI16(u64),
VecI32(u64),
VecI64(u64),
VecI128(u64),
VecF16(u64),
VecF32(u64),
VecF64(u64),
VecF128(u64),
}

impl InlineAsmType {
@@ -730,15 +734,19 @@ impl InlineAsmType {
Self::I32 => 4,
Self::I64 => 8,
Self::I128 => 16,
Self::F16 => 2,
Self::F32 => 4,
Self::F64 => 8,
Self::F128 => 16,
Self::VecI8(n) => n * 1,
Self::VecI16(n) => n * 2,
Self::VecI32(n) => n * 4,
Self::VecI64(n) => n * 8,
Self::VecI128(n) => n * 16,
Self::VecF16(n) => n * 2,
Self::VecF32(n) => n * 4,
Self::VecF64(n) => n * 8,
Self::VecF128(n) => n * 16,
})
}
}
@@ -751,15 +759,19 @@ impl fmt::Display for InlineAsmType {
Self::I32 => f.write_str("i32"),
Self::I64 => f.write_str("i64"),
Self::I128 => f.write_str("i128"),
Self::F16 => f.write_str("f16"),
Self::F32 => f.write_str("f32"),
Self::F64 => f.write_str("f64"),
Self::F128 => f.write_str("f128"),
Self::VecI8(n) => write!(f, "i8x{n}"),
Self::VecI16(n) => write!(f, "i16x{n}"),
Self::VecI32(n) => write!(f, "i32x{n}"),
Self::VecI64(n) => write!(f, "i64x{n}"),
Self::VecI128(n) => write!(f, "i128x{n}"),
Self::VecF16(n) => write!(f, "f16x{n}"),
Self::VecF32(n) => write!(f, "f32x{n}"),
Self::VecF64(n) => write!(f, "f64x{n}"),
Self::VecF128(n) => write!(f, "f128x{n}"),
}
}
}
22 changes: 11 additions & 11 deletions compiler/rustc_target/src/asm/x86.rs
View file
Original file line number Diff line number Diff line change
@@ -107,26 +107,26 @@ impl X86InlineAsmRegClass {
match self {
Self::reg | Self::reg_abcd => {
if arch == InlineAsmArch::X86_64 {
types! { _: I16, I32, I64, F32, F64; }
types! { _: I16, I32, I64, F16, F32, F64; }
} else {
types! { _: I16, I32, F32; }
types! { _: I16, I32, F16, F32; }
}
}
Self::reg_byte => types! { _: I8; },
Self::xmm_reg => types! {
sse: I32, I64, F32, F64,
VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF32(4), VecF64(2);
sse: I32, I64, F16, F32, F64, F128,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would hold back on adding f128 as an allowed input type since there aren't any instructions that actually support that type.

But for f16 this seems fine.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

While there aren't any instructions that take f128 as an operand, f128 is passed and returned in XMM registers as part of the x86-64 System V C ABI so seems worth including.

VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF16(8), VecF32(4), VecF64(2);
},
Self::ymm_reg => types! {
avx: I32, I64, F32, F64,
VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF32(4), VecF64(2),
VecI8(32), VecI16(16), VecI32(8), VecI64(4), VecF32(8), VecF64(4);
avx: I32, I64, F16, F32, F64, F128,
VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF16(8), VecF32(4), VecF64(2),
VecI8(32), VecI16(16), VecI32(8), VecI64(4), VecF16(16), VecF32(8), VecF64(4);
},
Self::zmm_reg => types! {
avx512f: I32, I64, F32, F64,
VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF32(4), VecF64(2),
VecI8(32), VecI16(16), VecI32(8), VecI64(4), VecF32(8), VecF64(4),
VecI8(64), VecI16(32), VecI32(16), VecI64(8), VecF32(16), VecF64(8);
avx512f: I32, I64, F16, F32, F64, F128,
VecI8(16), VecI16(8), VecI32(4), VecI64(2), VecF16(8), VecF32(4), VecF64(2),
VecI8(32), VecI16(16), VecI32(8), VecI64(4), VecF16(16), VecF32(8), VecF64(4),
VecI8(64), VecI16(32), VecI32(16), VecI64(8), VecF16(32), VecF32(16), VecF64(8);
},
Self::kreg => types! {
avx512f: I8, I16;
Loading