miri/shims/x86/ssse3.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
use rustc_middle::mir;
use rustc_middle::ty::Ty;
use rustc_span::Symbol;
use rustc_target::callconv::{Conv, FnAbi};
use super::{horizontal_bin_op, int_abs, pmulhrsw, psign};
use crate::*;
impl<'tcx> EvalContextExt<'tcx> for crate::MiriInterpCx<'tcx> {}
pub(super) trait EvalContextExt<'tcx>: crate::MiriInterpCxExt<'tcx> {
fn emulate_x86_ssse3_intrinsic(
&mut self,
link_name: Symbol,
abi: &FnAbi<'tcx, Ty<'tcx>>,
args: &[OpTy<'tcx>],
dest: &MPlaceTy<'tcx>,
) -> InterpResult<'tcx, EmulateItemResult> {
let this = self.eval_context_mut();
this.expect_target_feature_for_intrinsic(link_name, "ssse3")?;
// Prefix should have already been checked.
let unprefixed_name = link_name.as_str().strip_prefix("llvm.x86.ssse3.").unwrap();
match unprefixed_name {
// Used to implement the _mm_abs_epi{8,16,32} functions.
// Calculates the absolute value of packed 8/16/32-bit integers.
"pabs.b.128" | "pabs.w.128" | "pabs.d.128" => {
let [op] = this.check_shim(abi, Conv::C, link_name, args)?;
int_abs(this, op, dest)?;
}
// Used to implement the _mm_shuffle_epi8 intrinsic.
// Shuffles bytes from `left` using `right` as pattern.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_shuffle_epi8
"pshuf.b.128" => {
let [left, right] = this.check_shim(abi, Conv::C, link_name, args)?;
let (left, left_len) = this.project_to_simd(left)?;
let (right, right_len) = this.project_to_simd(right)?;
let (dest, dest_len) = this.project_to_simd(dest)?;
assert_eq!(dest_len, left_len);
assert_eq!(dest_len, right_len);
for i in 0..dest_len {
let right = this.read_scalar(&this.project_index(&right, i)?)?.to_u8()?;
let dest = this.project_index(&dest, i)?;
let res = if right & 0x80 == 0 {
let j = right % 16; // index wraps around
this.read_scalar(&this.project_index(&left, j.into())?)?
} else {
// If the highest bit in `right` is 1, write zero.
Scalar::from_u8(0)
};
this.write_scalar(res, &dest)?;
}
}
// Used to implement the _mm_h{add,adds,sub}_epi{16,32} functions.
// Horizontally add / add with saturation / subtract adjacent 16/32-bit
// integer values in `left` and `right`.
"phadd.w.128" | "phadd.sw.128" | "phadd.d.128" | "phsub.w.128" | "phsub.sw.128"
| "phsub.d.128" => {
let [left, right] = this.check_shim(abi, Conv::C, link_name, args)?;
let (which, saturating) = match unprefixed_name {
"phadd.w.128" | "phadd.d.128" => (mir::BinOp::Add, false),
"phadd.sw.128" => (mir::BinOp::Add, true),
"phsub.w.128" | "phsub.d.128" => (mir::BinOp::Sub, false),
"phsub.sw.128" => (mir::BinOp::Sub, true),
_ => unreachable!(),
};
horizontal_bin_op(this, which, saturating, left, right, dest)?;
}
// Used to implement the _mm_maddubs_epi16 function.
// Multiplies packed 8-bit unsigned integers from `left` and packed
// signed 8-bit integers from `right` into 16-bit signed integers. Then,
// the saturating sum of the products with indices `2*i` and `2*i+1`
// produces the output at index `i`.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_maddubs_epi16
"pmadd.ub.sw.128" => {
let [left, right] = this.check_shim(abi, Conv::C, link_name, args)?;
let (left, left_len) = this.project_to_simd(left)?;
let (right, right_len) = this.project_to_simd(right)?;
let (dest, dest_len) = this.project_to_simd(dest)?;
assert_eq!(left_len, right_len);
assert_eq!(dest_len.strict_mul(2), left_len);
for i in 0..dest_len {
let j1 = i.strict_mul(2);
let left1 = this.read_scalar(&this.project_index(&left, j1)?)?.to_u8()?;
let right1 = this.read_scalar(&this.project_index(&right, j1)?)?.to_i8()?;
let j2 = j1.strict_add(1);
let left2 = this.read_scalar(&this.project_index(&left, j2)?)?.to_u8()?;
let right2 = this.read_scalar(&this.project_index(&right, j2)?)?.to_i8()?;
let dest = this.project_index(&dest, i)?;
// Multiplication of a u8 and an i8 into an i16 cannot overflow.
let mul1 = i16::from(left1).strict_mul(right1.into());
let mul2 = i16::from(left2).strict_mul(right2.into());
let res = mul1.saturating_add(mul2);
this.write_scalar(Scalar::from_i16(res), &dest)?;
}
}
// Used to implement the _mm_mulhrs_epi16 function.
// Multiplies packed 16-bit signed integer values, truncates the 32-bit
// product to the 18 most significant bits by right-shifting, and then
// divides the 18-bit value by 2 (rounding to nearest) by first adding
// 1 and then taking the bits `1..=16`.
// https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mulhrs_epi16
"pmul.hr.sw.128" => {
let [left, right] = this.check_shim(abi, Conv::C, link_name, args)?;
pmulhrsw(this, left, right, dest)?;
}
// Used to implement the _mm_sign_epi{8,16,32} functions.
// Negates elements from `left` when the corresponding element in
// `right` is negative. If an element from `right` is zero, zero
// is writen to the corresponding output element.
// Basically, we multiply `left` with `right.signum()`.
"psign.b.128" | "psign.w.128" | "psign.d.128" => {
let [left, right] = this.check_shim(abi, Conv::C, link_name, args)?;
psign(this, left, right, dest)?;
}
_ => return interp_ok(EmulateItemResult::NotSupported),
}
interp_ok(EmulateItemResult::NeedsReturn)
}
}