T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

cranelift: Implement scalar FMA on x86 (#4460)

Browse Source 
x86 does not have dedicated instructions for scalar FMA, lower
to a libcall which seems to be what llvm does.
This commit is contained in:
Afonso Bordado
2022-08-03 18:29:10 +01:00
committed by GitHub
parent ff6082c0af
commit 709716bb8e
13 changed files with 167 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
cranelift/codegen/src/ir/libcall.rs
View File

@@ -1,6 +1,7 @@
//! Naming well-known routines in the runtime library.
use crate::ir::{types, ExternalName, FuncRef, Function, Opcode, Type};
use crate::ir::{types, AbiParam, ExternalName, FuncRef, Function, Opcode, Signature, Type};
use crate::isa::CallConv;
use core::fmt;
use core::str::FromStr;
#[cfg(feature = "enable-serde")]
@@ -50,6 +51,10 @@ pub enum LibCall {
NearestF32,
/// nearest.f64
NearestF64,
/// fma.f32
FmaF32,
/// fma.f64
FmaF64,
/// libc.memcpy
Memcpy,
/// libc.memset
@@ -91,6 +96,8 @@ impl FromStr for LibCall {
"TruncF64" => Ok(Self::TruncF64),
"NearestF32" => Ok(Self::NearestF32),
"NearestF64" => Ok(Self::NearestF64),
"FmaF32" => Ok(Self::FmaF32),
"FmaF64" => Ok(Self::FmaF64),
"Memcpy" => Ok(Self::Memcpy),
"Memset" => Ok(Self::Memset),
"Memmove" => Ok(Self::Memmove),
@@ -124,6 +131,7 @@ impl LibCall {
Opcode::Floor => Self::FloorF32,
Opcode::Trunc => Self::TruncF32,
Opcode::Nearest => Self::NearestF32,
Opcode::Fma => Self::FmaF32,
_ => return None,
},
types::F64 => match opcode {
@@ -131,6 +139,7 @@ impl LibCall {
Opcode::Floor => Self::FloorF64,
Opcode::Trunc => Self::TruncF64,
Opcode::Nearest => Self::NearestF64,
Opcode::Fma => Self::FmaF64,
_ => return None,
},
_ => return None,
@@ -157,6 +166,8 @@ impl LibCall {
TruncF64,
NearestF32,
NearestF64,
FmaF32,
FmaF64,
Memcpy,
Memset,
Memmove,
@@ -164,6 +175,50 @@ impl LibCall {
ElfTlsGetAddr,
]
}
/// Get a [Signature] for the function targeted by this [LibCall].
pub fn signature(&self, call_conv: CallConv) -> Signature {
use types::*;
let mut sig = Signature::new(call_conv);
match self {
LibCall::UdivI64
| LibCall::SdivI64
| LibCall::UremI64
| LibCall::SremI64
| LibCall::IshlI64
| LibCall::UshrI64
| LibCall::SshrI64 => {
sig.params.push(AbiParam::new(I64));
sig.params.push(AbiParam::new(I64));
sig.returns.push(AbiParam::new(I64));
}
LibCall::CeilF32 | LibCall::FloorF32 | LibCall::TruncF32 | LibCall::NearestF32 => {
sig.params.push(AbiParam::new(F32));
sig.returns.push(AbiParam::new(F32));
}
LibCall::TruncF64 | LibCall::FloorF64 | LibCall::CeilF64 | LibCall::NearestF64 => {
sig.params.push(AbiParam::new(F64));
sig.returns.push(AbiParam::new(F64));
}
LibCall::FmaF32 | LibCall::FmaF64 => {
let ty = if *self == LibCall::FmaF32 { F32 } else { F64 };
sig.params.push(AbiParam::new(ty));
sig.params.push(AbiParam::new(ty));
sig.params.push(AbiParam::new(ty));
sig.returns.push(AbiParam::new(ty));
}
LibCall::Probestack
| LibCall::Memcpy
| LibCall::Memset
| LibCall::Memmove
| LibCall::Memcmp
| LibCall::ElfTlsGetAddr => unimplemented!(),
}
sig
}
}
/// Get a function reference for the probestack function in `func`.