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

Add x86 SIMD encoding for icmp sgt

Browse Source
This commit is contained in:
Andrew Brown
2019-10-24 17:07:42 -07:00
parent 6f35273055
commit c454c3c771
4 changed files with 82 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
cranelift/codegen/meta/src/isa/x86/encodings.rs
View File

@@ -685,6 +685,7 @@ pub(crate) fn define(
let use_sse41 = settings.predicate_by_name("use_sse41");
let use_ssse3_simd = settings.predicate_by_name("use_ssse3_simd");
let use_sse41_simd = settings.predicate_by_name("use_sse41_simd");
let use_sse42_simd = settings.predicate_by_name("use_sse42_simd");
// Definitions.
let mut e = PerCpuModeEncodings::new();
@@ -2025,21 +2026,25 @@ pub(crate) fn define(
e.enc_32_64(x86_psra, rec_fa.opcodes(*opcodes));
}
// SIMD icmp using PCMPEQ*
for ty in ValueType::all_lane_types().filter(|t| t.is_int() && allowed_simd_type(t)) {
let (opcodes, isa_predicate): (&[_], _) = match ty.lane_bits() {
8 => (&PCMPEQB, None),
16 => (&PCMPEQW, None),
32 => (&PCMPEQD, None),
64 => (&PCMPEQQ, Some(use_sse41_simd)),
_ => panic!("invalid size for SIMD icmp"),
};
let instruction = icmp
.bind(Immediate::IntCC(IntCC::Equal))
.bind(vector(ty, sse_vector_size));
let template = rec_icscc_fpr.nonrex().opcodes(opcodes);
e.enc_32_64_maybe_isap(instruction, template, isa_predicate);
// SIMD integer comparisons
{
use IntCC::*;
for (ty, cc, opcodes, isa_predicate) in &[
(I8, Equal, &PCMPEQB[..], None),
(I16, Equal, &PCMPEQW[..], None),
(I32, Equal, &PCMPEQD[..], None),
(I64, Equal, &PCMPEQQ[..], Some(use_sse41_simd)),
(I8, SignedGreaterThan, &PCMPGTB[..], None),
(I16, SignedGreaterThan, &PCMPGTW[..], None),
(I32, SignedGreaterThan, &PCMPGTD[..], None),
(I64, SignedGreaterThan, &PCMPGTQ, Some(use_sse42_simd)),
] {
let instruction = icmp
.bind(Immediate::IntCC(*cc))
.bind(vector(*ty, sse_vector_size));
let template = rec_icscc_fpr.nonrex().opcodes(opcodes);
e.enc_32_64_maybe_isap(instruction, template, *isa_predicate);
}
}
// Reference type instructions

12
cranelift/codegen/meta/src/isa/x86/opcodes.rs
View File

@@ -281,6 +281,18 @@ pub static PCMPEQQ: [u8; 4] = [0x66, 0x0f, 0x38, 0x29];
/// Compare packed data for equal (SSE2).
pub static PCMPEQW: [u8; 3] = [0x66, 0x0f, 0x75];
/// Compare packed signed byte integers for greater than (SSE2).
pub static PCMPGTB: [u8; 3] = [0x66, 0x0f, 0x64];
/// Compare packed signed doubleword integers for greater than (SSE2).
pub static PCMPGTD: [u8; 3] = [0x66, 0x0f, 0x66];
/// Compare packed signed quadword integers for greater than (SSE4.2).
pub static PCMPGTQ: [u8; 4] = [0x66, 0x0f, 0x38, 0x37];
/// Compare packed signed word integers for greater than (SSE2).
pub static PCMPGTW: [u8; 3] = [0x66, 0x0f, 0x65];
/// Extract doubleword or quadword, depending on REX.W (SSE4.1).
pub static PEXTR: [u8; 4] = [0x66, 0x0f, 0x3a, 0x16];