user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add llvm.x86.sse2.cvtps2dq

Browse source
This commit is contained in:
Folkert 2024-06-06 22:33:43 +02:00
parent 632e5df38a
commit 63cb28ed48
No known key found for this signature in database
GPG key ID: 1F17F6FFD112B97C
2 changed files with 48 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

11
src/intrinsics/llvm_x86.rs
View file

@ -459,11 +459,20 @@ pub(crate) fn codegen_x86_llvm_intrinsic_call<'tcx>(
intrinsic_args!(fx, args => (a); intrinsic);
let a = a.load_scalar(fx);
let value = fx.bcx.ins().x86_cvtt2dq(types::I32X4, a);
let cvalue = CValue::by_val(value, ret.layout());
ret.write_cvalue(fx, cvalue);
}
"llvm.x86.sse2.cvtps2dq" => {
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtps_epi32
intrinsic_args!(fx, args => (a); intrinsic);
let a = a.load_scalar(fx);
// Using inline asm instead of fcvt_to_sint_sat as unrepresentable values are turned
// into 0x80000000 for which Cranelift doesn't have a native instruction.
codegen_inline_asm_inner(
fx,
&[InlineAsmTemplatePiece::String(format!("cvttps2dq xmm0, xmm0"))],
&[InlineAsmTemplatePiece::String(format!("cvtps2dq xmm0, xmm0"))],
&[CInlineAsmOperand::InOut {
reg: InlineAsmRegOrRegClass::Reg(InlineAsmReg::X86(X86InlineAsmReg::xmm0)),
_late: true,