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

Support wasm select instruction with V128-typed operands on AArch64.

Browse Source 
* this requires upgrading to wasmparser 0.67.0.

* There are no CLIF side changes because the CLIF `select` instruction is
  polymorphic enough.

* on aarch64, there is unfortunately no conditional-move (csel) instruction on
  vectors.  This patch adds a synthetic instruction `VecCSel` which *does*
  behave like that.  At emit time, this is emitted as an if-then-else diamond
  (4 insns).

* aarch64 implementation is otherwise straightforwards.
This commit is contained in:
Julian Seward
2020-11-11 17:09:57 +01:00
committed by julian-seward1
parent 9ced345aed
commit 41e87a2f99
15 changed files with 122 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
Cargo.lock generated
View File

@@ -542,7 +542,7 @@ dependencies = [
"smallvec",
"target-lexicon",
"thiserror",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wat",
]
@@ -1090,7 +1090,7 @@ dependencies = [
"smallvec",
"thiserror",
"typemap",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wat",
]
@@ -2223,6 +2223,12 @@ version = "0.66.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c97fc0456d6d09ca7b64bc33c34f4e8f29d5ccfa8e5595ef7a8957818339e6f6"
[[package]]
name = "wasmparser"
version = "0.67.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9f091cf3849e5fe76a60255bff169277459f2201435bc583b6656880553f0ad0"
[[package]]
name = "wasmprinter"
version = "0.2.13"
@@ -2250,7 +2256,7 @@ dependencies = [
"smallvec",
"target-lexicon",
"tempfile",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmtime-cache",
"wasmtime-environ",
"wasmtime-jit",
@@ -2328,7 +2334,7 @@ dependencies = [
"test-programs",
"tracing-subscriber",
"wasi-common",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmtime",
"wasmtime-cache",
"wasmtime-debug",
@@ -2363,7 +2369,7 @@ dependencies = [
"object",
"target-lexicon",
"thiserror",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmtime-environ",
]
@@ -2382,7 +2388,7 @@ dependencies = [
"more-asserts",
"serde",
"thiserror",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
]
[[package]]
@@ -2410,7 +2416,7 @@ dependencies = [
"log",
"rayon",
"wasm-smith",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmprinter",
"wasmtime",
"wasmtime-wast",
@@ -2437,7 +2443,7 @@ dependencies = [
"serde",
"target-lexicon",
"thiserror",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmtime-cranelift",
"wasmtime-debug",
"wasmtime-environ",
@@ -2454,7 +2460,7 @@ version = "0.21.0"
dependencies = [
"cranelift-codegen",
"lightbeam",
"wasmparser 0.66.0",
"wasmparser 0.67.0",
"wasmtime-environ",
]

2
Cargo.toml
View File

@@ -42,7 +42,7 @@ libc = "0.2.60"
log = "0.4.8"
rayon = "1.2.1"
humantime = "2.0.0"
wasmparser = "0.66"
wasmparser = "0.67"
[dev-dependencies]
env_logger = "0.8.1"

40
cranelift/codegen/src/isa/aarch64/inst/emit.rs
View File

@@ -1909,6 +1909,46 @@ impl MachInstEmit for Inst {
sink.put4(enc_ldst_vec(q, size, rn, rd));
}
&Inst::VecCSel { rd, rn, rm, cond } => {
/* Emit this:
b.cond else
mov rd, rm
b out
else:
mov rd, rn
out:
Note, we could do better in the cases where rd == rn or rd == rm.
*/
let else_label = sink.get_label();
let out_label = sink.get_label();
// b.cond else
let br_else_offset = sink.cur_offset();
sink.put4(enc_conditional_br(
BranchTarget::Label(else_label),
CondBrKind::Cond(cond),
));
sink.use_label_at_offset(br_else_offset, else_label, LabelUse::Branch19);
// mov rd, rm
sink.put4(enc_vecmov(/* 16b = */ true, rd, rm));
// b out
let b_out_offset = sink.cur_offset();
sink.use_label_at_offset(b_out_offset, out_label, LabelUse::Branch26);
sink.add_uncond_branch(b_out_offset, b_out_offset + 4, out_label);
sink.put4(enc_jump26(0b000101, 0 /* will be fixed up later */));
// else:
sink.bind_label(else_label);
// mov rd, rn
sink.put4(enc_vecmov(/* 16b = */ true, rd, rn));
// out:
sink.bind_label(out_label);
}
&Inst::MovToNZCV { rn } => {
sink.put4(0xd51b4200 | machreg_to_gpr(rn));
}

11
cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
View File

@@ -3879,6 +3879,17 @@ fn test_aarch64_binemit() {
"ld1r { v0.8b }, [x25]",
));
insns.push((
Inst::VecCSel {
rd: writable_vreg(5),
rn: vreg(10),
rm: vreg(19),
cond: Cond::Gt,
},
"6C000054651EB34E02000014451DAA4E",
"vcsel v5.16b, v10.16b, v19.16b, gt (if-then-else diamond)",
));
insns.push((
Inst::Extend {
rd: writable_xreg(1),

31
cranelift/codegen/src/isa/aarch64/inst/mod.rs
View File

@@ -1016,6 +1016,15 @@ pub enum Inst {
size: VectorSize,
},
/// Vector conditional select, 128 bit. A synthetic instruction, which generates a 4-insn
/// control-flow diamond.
VecCSel {
rd: Writable<Reg>,
rn: Reg,
rm: Reg,
cond: Cond,
},
/// Move to the NZCV flags (actually a `MSR NZCV, Xn` insn).
MovToNZCV {
rn: Reg,
@@ -1732,6 +1741,11 @@ fn aarch64_get_regs(inst: &Inst, collector: &mut RegUsageCollector) {
collector.add_def(rd);
collector.add_use(rn);
}
&Inst::VecCSel { rd, rn, rm, .. } => {
collector.add_def(rd);
collector.add_use(rn);
collector.add_use(rm);
}
&Inst::FpuCmp32 { rn, rm } | &Inst::FpuCmp64 { rn, rm } => {
collector.add_use(rn);
collector.add_use(rm);
@@ -2343,6 +2357,16 @@ fn aarch64_map_regs<RUM: RegUsageMapper>(inst: &mut Inst, mapper: &RUM) {
map_def(mapper, rd);
map_use(mapper, rn);
}
&mut Inst::VecCSel {
ref mut rd,
ref mut rn,
ref mut rm,
..
} => {
map_def(mapper, rd);
map_use(mapper, rn);
map_use(mapper, rm);
}
&mut Inst::FpuCmp32 {
ref mut rn,
ref mut rm,
@@ -3591,6 +3615,13 @@ impl Inst {
format!("ld1r {{ {} }}, [{}]", rd, rn)
}
&Inst::VecCSel { rd, rn, rm, cond } => {
let rd = show_vreg_vector(rd.to_reg(), mb_rru, VectorSize::Size8x16);
let rn = show_vreg_vector(rn, mb_rru, VectorSize::Size8x16);
let rm = show_vreg_vector(rm, mb_rru, VectorSize::Size8x16);
let cond = cond.show_rru(mb_rru);
format!("vcsel {}, {}, {}, {} (if-then-else diamond)", rd, rn, rm, cond)
}
&Inst::MovToNZCV { rn } => {
let rn = rn.show_rru(mb_rru);
format!("msr nzcv, {}", rn)

2
cranelift/codegen/src/isa/aarch64/lower_inst.rs
View File

@@ -1412,6 +1412,8 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
ctx.emit(Inst::FpuCSel32 { cond, rd, rn, rm });
} else if is_float && bits == 64 {
ctx.emit(Inst::FpuCSel64 { cond, rd, rn, rm });
} else if is_float && bits == 128 {
ctx.emit(Inst::VecCSel { cond, rd, rn, rm });
} else {
ctx.emit(Inst::CSel { cond, rd, rn, rm });
}

14
cranelift/filetests/filetests/isa/aarch64/simd.clif
View File

@@ -47,3 +47,17 @@ block0:
; nextln: mov sp, fp
; nextln: ldp fp, lr, [sp], #16
; nextln: ret
function %f4(i32, i8x16, i8x16) -> i8x16 {
block0(v0: i32, v1: i8x16, v2: i8x16):
v3 = select v0, v1, v2
return v3
}
; check: stp fp, lr, [sp, #-16]!
; nextln: mov fp, sp
; nextln: subs wzr, w0, wzr
; nextln: vcsel v0.16b, v0.16b, v1.16b, ne (if-then-else diamond)
; nextln: mov sp, fp
; nextln: ldp fp, lr, [sp], #16
; nextln: ret

2
cranelift/wasm/Cargo.toml
View File

@@ -12,7 +12,7 @@ keywords = ["webassembly", "wasm"]
edition = "2018"
[dependencies]
wasmparser = { version = "0.66.0", default-features = false }
wasmparser = { version = "0.67.0", default-features = false }
cranelift-codegen = { path = "../codegen", version = "0.68.0", default-features = false }
cranelift-entity = { path = "../entity", version = "0.68.0" }
cranelift-frontend = { path = "../frontend", version = "0.68.0", default-features = false }

2
crates/debug/Cargo.toml
View File

@@ -13,7 +13,7 @@ edition = "2018"
[dependencies]
gimli = "0.23.0"
wasmparser = "0.66.0"
wasmparser = "0.67.0"
object = { version = "0.22.0", default-features = false, features = ["read", "write"] }
wasmtime-environ = { path = "../environ", version = "0.21.0" }
target-lexicon = { version = "0.11.0", default-features = false }

2
crates/environ/Cargo.toml
View File

@@ -16,7 +16,7 @@ anyhow = "1.0"
cranelift-codegen = { path = "../../cranelift/codegen", version = "0.68.0", features = ["enable-serde"] }
cranelift-entity = { path = "../../cranelift/entity", version = "0.68.0", features = ["enable-serde"] }
cranelift-wasm = { path = "../../cranelift/wasm", version = "0.68.0", features = ["enable-serde"] }
wasmparser = "0.66.0"
wasmparser = "0.67.0"
indexmap = { version = "1.0.2", features = ["serde-1"] }
thiserror = "1.0.4"
serde = { version = "1.0.94", features = ["derive"] }

2
crates/fuzzing/Cargo.toml
View File

@@ -12,7 +12,7 @@ arbitrary = { version = "0.4.1", features = ["derive"] }
env_logger = "0.8.1"
log = "0.4.8"
rayon = "1.2.1"
wasmparser = "0.66.0"
wasmparser = "0.67.0"
wasmprinter = "0.2.13"
wasmtime = { path = "../wasmtime" }
wasmtime-wast = { path = "../wast" }

2
crates/jit/Cargo.toml
View File

@@ -28,7 +28,7 @@ rayon = { version = "1.0", optional = true }
region = "2.1.0"
thiserror = "1.0.4"
target-lexicon = { version = "0.11.0", default-features = false }
wasmparser = "0.66.0"
wasmparser = "0.67.0"
more-asserts = "0.2.1"
anyhow = "1.0"
cfg-if = "1.0"

2
crates/lightbeam/Cargo.toml
View File

@@ -24,7 +24,7 @@ more-asserts = "0.2.1"
smallvec = "1.0.0"
thiserror = "1.0.9"
typemap = "0.3"
wasmparser = "0.66.0"
wasmparser = "0.67.0"
[dev-dependencies]
lazy_static = "1.2"

2
crates/lightbeam/wasmtime/Cargo.toml
View File

@@ -13,6 +13,6 @@ edition = "2018"
[dependencies]
lightbeam = { path = "..", version = "0.21.0" }
wasmparser = "0.66"
wasmparser = "0.67"
cranelift-codegen = { path = "../../../cranelift/codegen", version = "0.68.0" }
wasmtime-environ = { path = "../../environ", version = "0.21.0" }

2
crates/wasmtime/Cargo.toml
View File

@@ -16,7 +16,7 @@ wasmtime-jit = { path = "../jit", version = "0.21.0" }
wasmtime-cache = { path = "../cache", version = "0.21.0", optional = true }
wasmtime-profiling = { path = "../profiling", version = "0.21.0" }
target-lexicon = { version = "0.11.0", default-features = false }
wasmparser = "0.66.0"
wasmparser = "0.67.0"
anyhow = "1.0.19"
region = "2.2.0"
libc = "0.2"