bjoernager/rust
bjoernager/rust
1
Fork 0
Code Activity

intrinsics.fmuladdf{16,32,64,128}: expose llvm.fmuladd.* semantics

Browse source 
Add intrinsics `fmuladd{f16,f32,f64,f128}`. This computes `(a * b) +
c`, to be fused if the code generator determines that (i) the target
instruction set has support for a fused operation, and (ii) that the
fused operation is more efficient than the equivalent, separate pair
of `mul` and `add` instructions.

https://llvm.org/docs/LangRef.html#llvm-fmuladd-intrinsic

MIRI support is included for f32 and f64.

The codegen_cranelift uses the `fma` function from libc, which is a
correct implementation, but without the desired performance semantic. I
think this requires an update to cranelift to expose a suitable
instruction in its IR.

I have not tested with codegen_gcc, but it should behave the same
way (using `fma` from libc).
This commit is contained in:
Jed Brown 2024-01-05 21:04:41 -07:00
parent 01e2fff90c
commit 0d8a978e8a
11 changed files with 222 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

5
compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs
View file

@ -328,6 +328,9 @@ fn codegen_float_intrinsic_call<'tcx>(
sym::fabsf64 => ("fabs", 1, fx.tcx.types.f64, types::F64),
sym::fmaf32 => ("fmaf", 3, fx.tcx.types.f32, types::F32),
sym::fmaf64 => ("fma", 3, fx.tcx.types.f64, types::F64),
// FIXME: calling `fma` from libc without FMA target feature uses expensive sofware emulation
sym::fmuladdf32 => ("fmaf", 3, fx.tcx.types.f32, types::F32), // TODO: use cranelift intrinsic analogous to llvm.fmuladd.f32
sym::fmuladdf64 => ("fma", 3, fx.tcx.types.f64, types::F64), // TODO: use cranelift intrinsic analogous to llvm.fmuladd.f64
sym::copysignf32 => ("copysignf", 2, fx.tcx.types.f32, types::F32),
sym::copysignf64 => ("copysign", 2, fx.tcx.types.f64, types::F64),
sym::floorf32 => ("floorf", 1, fx.tcx.types.f32, types::F32),
@ -381,7 +384,7 @@ fn codegen_float_intrinsic_call<'tcx>(
let layout = fx.layout_of(ty);
let res = match intrinsic {
sym::fmaf32 | sym::fmaf64 => {
sym::fmaf32 | sym::fmaf64 | sym::fmuladdf32 | sym::fmuladdf64 => {
CValue::by_val(fx.bcx.ins().fma(args[0], args[1], args[2]), layout)
}
sym::copysignf32 | sym::copysignf64 => {