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

ISLE: port iabs to ISLE for x64

Browse Source
This commit is contained in:
Nick Fitzgerald
2021-11-19 11:03:44 -08:00
parent ef8ea644f4
commit 94e0de45ed
6 changed files with 405 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
cranelift/codegen/src/isa/x64/lower.rs
View File

@@ -1504,6 +1504,14 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
return Ok(());
}
let implemented_in_isle = |ctx: &mut C| {
unreachable!(
"implemented in ISLE: inst = `{}`, type = `{:?}`",
ctx.dfg().display_inst(insn),
ty
)
};
match op {
Opcode::Iconst
| Opcode::Bconst
@@ -1520,54 +1528,8 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
| Opcode::Bor
| Opcode::Bxor
| Opcode::Imul
| Opcode::BandNot => {
unreachable!(
"implemented in ISLE: inst = `{}`, type = `{:?}`",
ctx.dfg().display_inst(insn),
ty
);
}
Opcode::Iabs => {
let src = input_to_reg_mem(ctx, inputs[0]);
let dst = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
let ty = ty.unwrap();
if ty == types::I64X2 {
if isa_flags.use_avx512vl_simd() && isa_flags.use_avx512f_simd() {
ctx.emit(Inst::xmm_unary_rm_r_evex(Avx512Opcode::Vpabsq, src, dst));
} else {
// If `VPABSQ` from AVX512 is unavailable, we use a separate register, `tmp`, to
// contain the results of `0 - src` and then blend in those results with
// `BLENDVPD` if the MSB of `tmp` was set to 1 (i.e. if `tmp` was negative or,
// conversely, if `src` was originally positive).
// Emit all 0s into the `tmp` register.
let tmp = ctx.alloc_tmp(ty).only_reg().unwrap();
ctx.emit(Inst::xmm_rm_r(SseOpcode::Pxor, RegMem::from(tmp), tmp));
// Subtract the lanes from 0 and set up `dst`.
ctx.emit(Inst::xmm_rm_r(SseOpcode::Psubq, src.clone(), tmp));
ctx.emit(Inst::gen_move(dst, tmp.to_reg(), ty));
// Choose the subtracted lanes when `tmp` has an MSB of 1. BLENDVPD's semantics
// require the "choice" mask to be in XMM0.
ctx.emit(Inst::gen_move(
Writable::from_reg(regs::xmm0()),
tmp.to_reg(),
ty,
));
ctx.emit(Inst::xmm_rm_r(SseOpcode::Blendvpd, src, dst));
}
} else if ty.is_vector() {
let opcode = match ty {
types::I8X16 => SseOpcode::Pabsb,
types::I16X8 => SseOpcode::Pabsw,
types::I32X4 => SseOpcode::Pabsd,
_ => panic!("Unsupported type for packed iabs instruction: {}", ty),
};
ctx.emit(Inst::xmm_unary_rm_r(opcode, src, dst));
} else {
unimplemented!("iabs is unimplemented for non-vector type: {}", ty);
}
}
| Opcode::BandNot
| Opcode::Iabs => implemented_in_isle(ctx),
Opcode::Imax | Opcode::Umax | Opcode::Imin | Opcode::Umin => {
let lhs = put_input_in_reg(ctx, inputs[0]);