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

CL/aarch64: implement the wasm SIMD i32x4.dot_i16x8_s instruction

Browse Source 
This patch implements, for aarch64, the following wasm SIMD extensions

  i32x4.dot_i16x8_s instruction
  https://github.com/WebAssembly/simd/pull/127

It also updates dependencies as follows, in order that the new instruction can
be parsed, decoded, etc:

  wat          to  1.0.27
  wast         to  26.0.1
  wasmparser   to  0.65.0
  wasmprinter  to  0.2.12

The changes are straightforward:

* new CLIF instruction `widening_pairwise_dot_product_s`

* translation from wasm into `widening_pairwise_dot_product_s`

* new AArch64 instructions `smull`, `smull2` (part of the `VecRRR` group)

* translation from `widening_pairwise_dot_product_s` to `smull ; smull2 ; addv`

There is no testcase in this commit, because that is a separate repo.  The
implementation has been tested, nevertheless.
This commit is contained in:
Julian Seward
2020-10-27 15:04:32 +01:00
committed by julian-seward1
parent 54a97f784e
commit 5a5fb11979
26 changed files with 228 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
Cargo.lock generated
View File

@@ -349,7 +349,7 @@ dependencies = [
"souper-ir",
"target-lexicon",
"thiserror",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -549,7 +549,7 @@ dependencies = [
"smallvec",
"target-lexicon",
"thiserror",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wat",
]
@@ -1062,7 +1062,7 @@ dependencies = [
"staticvec",
"thiserror",
"typemap",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wat",
]
@@ -1233,7 +1233,7 @@ dependencies = [
"peepmatic-test-operator",
"peepmatic-traits",
"serde",
"wast 25.0.0",
"wast 26.0.1",
"z3",
]
@@ -1261,7 +1261,7 @@ dependencies = [
"peepmatic-traits",
"rand 0.7.3",
"serde",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -1286,7 +1286,7 @@ dependencies = [
"serde",
"serde_test",
"thiserror",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -1298,7 +1298,7 @@ dependencies = [
"peepmatic",
"peepmatic-test-operator",
"souper-ir",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -1319,7 +1319,7 @@ version = "0.67.0"
dependencies = [
"peepmatic-traits",
"serde",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -2290,18 +2290,18 @@ checksum = "32fddd575d477c6e9702484139cf9f23dcd554b06d185ed0f56c857dd3a47aa6"
[[package]]
name = "wasmparser"
version = "0.63.0"
version = "0.65.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "57da5d7300428d75d8b3cdfb736e41ee6af8926d69c1de2f201a1a22f234b7b5"
checksum = "87cc2fe6350834b4e528ba0901e7aa405d78b89dc1fa3145359eb4de0e323fcf"
[[package]]
name = "wasmprinter"
version = "0.2.10"
version = "0.2.12"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "fd1fea2e86fb107de33db6538db99f73e8d058f6fda2bbd8cceafff847dd9266"
checksum = "0fe87e79ad307dc73cb7caca2b57d049d2905c7db1317a9c21171e69aa9f93d3"
dependencies = [
"anyhow",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
]
[[package]]
@@ -2321,7 +2321,7 @@ dependencies = [
"smallvec",
"target-lexicon",
"tempfile",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmtime-cache",
"wasmtime-environ",
"wasmtime-jit",
@@ -2399,7 +2399,7 @@ dependencies = [
"test-programs",
"tracing-subscriber",
"wasi-common",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmtime",
"wasmtime-cache",
"wasmtime-debug",
@@ -2434,7 +2434,7 @@ dependencies = [
"object 0.21.1",
"target-lexicon",
"thiserror",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmtime-environ",
]
@@ -2453,7 +2453,7 @@ dependencies = [
"more-asserts",
"serde",
"thiserror",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
]
[[package]]
@@ -2481,7 +2481,7 @@ dependencies = [
"log",
"rayon",
"wasm-smith",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmprinter",
"wasmtime",
"wasmtime-wast",
@@ -2508,7 +2508,7 @@ dependencies = [
"serde",
"target-lexicon",
"thiserror",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmtime-cranelift",
"wasmtime-debug",
"wasmtime-environ",
@@ -2525,7 +2525,7 @@ version = "0.20.0"
dependencies = [
"cranelift-codegen",
"lightbeam",
"wasmparser 0.63.0",
"wasmparser 0.65.0",
"wasmtime-environ",
]
@@ -2617,7 +2617,7 @@ version = "0.20.0"
dependencies = [
"anyhow",
"wasmtime",
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]
@@ -2652,20 +2652,20 @@ dependencies = [
[[package]]
name = "wast"
version = "25.0.0"
version = "26.0.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d8c67a4386e4efe10563552848d8c6a4b7f941e69924a935495645c3f52b32d0"
checksum = "b3f174eed73e885ede6c8fcc3fbea8c3757afa521840676496cde56bb742ddab"
dependencies = [
"leb128",
]
[[package]]
name = "wat"
version = "1.0.26"
version = "1.0.27"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4766d466249e23279e92c52033429eb91141c5efea1c4478138fa6f6ef4efe3e"
checksum = "26b2dccbce4d0e14875091846e110a2369267b18ddd0d6423479b88dad914d71"
dependencies = [
"wast 25.0.0",
"wast 26.0.1",
]
[[package]]

4
Cargo.toml
View File

@@ -37,12 +37,12 @@ anyhow = "1.0.19"
target-lexicon = { version = "0.11.0", default-features = false }
pretty_env_logger = "0.4.0"
file-per-thread-logger = "0.1.1"
wat = "1.0.26"
wat = "1.0.27"
libc = "0.2.60"
log = "0.4.8"
rayon = "1.2.1"
humantime = "1.3.0"
wasmparser = "0.63"
wasmparser = "0.65"
[dev-dependencies]
env_logger = "0.7.1"

2
cranelift/codegen/Cargo.toml
View File

@@ -30,7 +30,7 @@ peepmatic-traits = { path = "../peepmatic/crates/traits", optional = true, versi
peepmatic-runtime = { path = "../peepmatic/crates/runtime", optional = true, version = "0.67.0" }
regalloc = { version = "0.0.31" }
souper-ir = { version = "1", optional = true }
wast = { version = "25.0.0", optional = true }
wast = { version = "26.0.1", optional = true }
# It is a goal of the cranelift-codegen crate to have minimal external dependencies.
# Please don't add any unless they are essential to the task of creating binary
# machine code. Integration tests that need external dependencies can be

35
cranelift/codegen/meta/src/shared/instructions.rs
View File

@@ -4078,6 +4078,41 @@ pub(crate) fn define(
.operands_out(vec![a]),
);
let I16x8 = &TypeVar::new(
"I16x8",
"A SIMD vector type containing 8 integer lanes each 16 bits wide.",
TypeSetBuilder::new()
.ints(16..16)
.simd_lanes(8..8)
.includes_scalars(false)
.build(),
);
let x = &Operand::new("x", I16x8);
let y = &Operand::new("y", I16x8);
let a = &Operand::new("a", &I16x8.merge_lanes());
ig.push(
Inst::new(
"widening_pairwise_dot_product_s",
r#"
Takes corresponding elements in `x` and `y`, performs a sign-extending length-doubling
multiplication on them, then adds adjacent pairs of elements to form the result. For
example, if the input vectors are `[x3, x2, x1, x0]` and `[y3, y2, y1, y0]`, it produces
the vector `[r1, r0]`, where `r1 = sx(x3) * sx(y3) + sx(x2) * sx(y2)` and
`r0 = sx(x1) * sx(y1) + sx(x0) * sx(y0)`, and `sx(n)` sign-extends `n` to twice its width.
This will double the lane width and halve the number of lanes. So the resulting
vector has the same number of bits as `x` and `y` do (individually).
See https://github.com/WebAssembly/simd/pull/127 for background info.
"#,
&formats.binary,
)
.operands_in(vec![x, y])
.operands_out(vec![a]),
);
let IntTo = &TypeVar::new(
"IntTo",
"A larger integer type with the same number of lanes",

4
cranelift/codegen/src/isa/aarch64/inst/args.rs
View File

@@ -677,6 +677,9 @@ impl VectorSize {
}
}
/// Produces a `VectorSize` with lanes twice as wide. Note that if the resulting
/// size would exceed 128 bits, then the number of lanes is also halved, so as to
/// ensure that the result size is at most 128 bits.
pub fn widen(&self) -> VectorSize {
match self {
VectorSize::Size8x8 => VectorSize::Size16x8,
@@ -689,6 +692,7 @@ impl VectorSize {
}
}
/// Produces a `VectorSize` that has the same lane width, but half as many lanes.
pub fn halve(&self) -> VectorSize {
match self {
VectorSize::Size8x16 => VectorSize::Size8x8,

10
cranelift/codegen/src/isa/aarch64/inst/emit.rs
View File

@@ -1950,11 +1950,13 @@ impl MachInstEmit for Inst {
(0b001_01110_00_1 | enc_size << 1, 0b100000)
}
VecALUOp::Zip1 => (0b01001110_00_0 | enc_size << 1, 0b001110),
VecALUOp::Smull => (0b000_01110_00_1 | enc_size << 1, 0b110000),
VecALUOp::Smull2 => (0b010_01110_00_1 | enc_size << 1, 0b110000),
};
let top11 = if is_float {
top11 | (q << 9) | enc_float_size << 1
} else {
top11 | (q << 9)
let top11 = match alu_op {
VecALUOp::Smull | VecALUOp::Smull2 => top11,
_ if is_float => top11 | (q << 9) | enc_float_size << 1,
_ => top11 | (q << 9),
};
sink.put4(enc_vec_rrr(top11, rm, bit15_10, rn, rd));
}

72
cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
View File

@@ -3243,6 +3243,78 @@ fn test_aarch64_binemit() {
"zip1 v9.2d, v20.2d, v17.2d",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull,
rd: writable_vreg(16),
rn: vreg(12),
rm: vreg(1),
size: VectorSize::Size8x16,
},
"90C1210E",
"smull v16.8h, v12.8b, v1.8b",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull,
rd: writable_vreg(2),
rn: vreg(13),
rm: vreg(6),
size: VectorSize::Size16x8,
},
"A2C1660E",
"smull v2.4s, v13.4h, v6.4h",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull,
rd: writable_vreg(8),
rn: vreg(12),
rm: vreg(14),
size: VectorSize::Size32x4,
},
"88C1AE0E",
"smull v8.2d, v12.2s, v14.2s",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull2,
rd: writable_vreg(16),
rn: vreg(12),
rm: vreg(1),
size: VectorSize::Size8x16,
},
"90C1214E",
"smull2 v16.8h, v12.16b, v1.16b",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull2,
rd: writable_vreg(2),
rn: vreg(13),
rm: vreg(6),
size: VectorSize::Size16x8,
},
"A2C1664E",
"smull2 v2.4s, v13.8h, v6.8h",
));
insns.push((
Inst::VecRRR {
alu_op: VecALUOp::Smull2,
rd: writable_vreg(8),
rn: vreg(12),
rm: vreg(14),
size: VectorSize::Size32x4,
},
"88C1AE4E",
"smull2 v8.2d, v12.4s, v14.4s",
));
insns.push((
Inst::VecMisc {
op: VecMisc2::Not,

22
cranelift/codegen/src/isa/aarch64/inst/mod.rs
View File

@@ -291,6 +291,10 @@ pub enum VecALUOp {
Umlal,
/// Zip vectors (primary) [meaning, high halves]
Zip1,
/// Signed multiply long (low halves)
Smull,
/// Signed multiply long (high halves)
Smull2,
}
/// A Vector miscellaneous operation with two registers.
@@ -3546,15 +3550,21 @@ impl Inst {
VecALUOp::Addp => ("addp", size),
VecALUOp::Umlal => ("umlal", size),
VecALUOp::Zip1 => ("zip1", size),
VecALUOp::Smull => ("smull", size),
VecALUOp::Smull2 => ("smull2", size),
};
let rd_size = if alu_op == VecALUOp::Umlal {
size.widen()
} else {
size
let rd_size = match alu_op {
VecALUOp::Umlal | VecALUOp::Smull | VecALUOp::Smull2 => size.widen(),
_ => size
};
let rn_size = match alu_op {
VecALUOp::Smull => size.halve(),
_ => size
};
let rm_size = rn_size;
let rd = show_vreg_vector(rd.to_reg(), mb_rru, rd_size);
let rn = show_vreg_vector(rn, mb_rru, size);
let rm = show_vreg_vector(rm, mb_rru, size);
let rn = show_vreg_vector(rn, mb_rru, rn_size);
let rm = show_vreg_vector(rm, mb_rru, rm_size);
format!("{} {}, {}, {}", op, rd, rn, rm)
}
&Inst::VecMisc { op, rd, rn, size } => {

41
cranelift/codegen/src/isa/aarch64/lower_inst.rs
View File

@@ -2375,6 +2375,47 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
});
}
Opcode::WideningPairwiseDotProductS => {
let r_y = get_output_reg(ctx, outputs[0]);
let r_a = put_input_in_reg(ctx, inputs[0], NarrowValueMode::None);
let r_b = put_input_in_reg(ctx, inputs[1], NarrowValueMode::None);
let ty = ty.unwrap();
if ty == I32X4 {
let tmp = ctx.alloc_tmp(RegClass::V128, I8X16);
// The args have type I16X8.
// "y = i32x4.dot_i16x8_s(a, b)"
// => smull tmp, a, b
// smull2 y, a, b
// addp y, tmp, y
ctx.emit(Inst::VecRRR {
alu_op: VecALUOp::Smull,
rd: tmp,
rn: r_a,
rm: r_b,
size: VectorSize::Size16x8,
});
ctx.emit(Inst::VecRRR {
alu_op: VecALUOp::Smull2,
rd: r_y,
rn: r_a,
rm: r_b,
size: VectorSize::Size16x8,
});
ctx.emit(Inst::VecRRR {
alu_op: VecALUOp::Addp,
rd: r_y,
rn: tmp.to_reg(),
rm: r_y.to_reg(),
size: VectorSize::Size32x4,
});
} else {
return Err(CodegenError::Unsupported(format!(
"Opcode::WideningPairwiseDotProductS: unsupported laneage: {:?}",
ty
)));
}
}
Opcode::Fadd | Opcode::Fsub | Opcode::Fmul | Opcode::Fdiv | Opcode::Fmin | Opcode::Fmax => {
let ty = ty.unwrap();
let bits = ty_bits(ty);

BIN
cranelift/codegen/src/preopt.serialized
View File

Binary file not shown.

1
cranelift/interpreter/src/step.rs
View File

@@ -571,6 +571,7 @@ where
Opcode::AtomicLoad => unimplemented!("AtomicLoad"),
Opcode::AtomicStore => unimplemented!("AtomicStore"),
Opcode::Fence => unimplemented!("Fence"),
Opcode::WideningPairwiseDotProductS => unimplemented!("WideningPairwiseDotProductS"),
// TODO: these instructions should be removed once the new backend makes these obsolete
// (see https://github.com/bytecodealliance/wasmtime/issues/1936); additionally, the

2
cranelift/peepmatic/Cargo.toml
View File

@@ -15,7 +15,7 @@ peepmatic-macro = { version = "0.67.0", path = "crates/macro" }
peepmatic-runtime = { version = "0.67.0", path = "crates/runtime", features = ["construct"] }
peepmatic-traits = { version = "0.67.0", path = "crates/traits" }
serde = { version = "1.0.105", features = ["derive"] }
wast = "25.0.0"
wast = "26.0.1"
z3 = { version = "0.7.1", features = ["static-link-z3"] }
[dev-dependencies]

2
cranelift/peepmatic/crates/fuzzing/Cargo.toml
View File

@@ -21,4 +21,4 @@ peepmatic-test-operator = { path = "../test-operator" }
peepmatic-traits = { path = "../traits" }
rand = { version = "0.7.3", features = ["small_rng"] }
serde = "1.0.106"
wast = "25.0.0"
wast = "26.0.1"

2
cranelift/peepmatic/crates/runtime/Cargo.toml
View File

@@ -16,7 +16,7 @@ peepmatic-automata = { version = "0.67.0", path = "../automata", features = ["se
peepmatic-traits = { version = "0.67.0", path = "../traits" }
serde = { version = "1.0.105", features = ["derive"] }
thiserror = "1.0.15"
wast = { version = "25.0.0", optional = true }
wast = { version = "26.0.1", optional = true }
[dev-dependencies]
peepmatic-test-operator = { version = "0.67.0", path = "../test-operator" }

2
cranelift/peepmatic/crates/souper/Cargo.toml
View File

@@ -16,4 +16,4 @@ log = "0.4.8"
[dev-dependencies]
peepmatic = { path = "../..", version = "0.67.0" }
peepmatic-test-operator = { version = "0.67.0", path = "../test-operator" }
wast = "25.0.0"
wast = "26.0.1"

2
cranelift/peepmatic/crates/test-operator/Cargo.toml
View File

@@ -9,4 +9,4 @@ edition = "2018"
[dependencies]
peepmatic-traits = { version = "0.67.0", path = "../traits" }
serde = { version = "1.0.105", features = ["derive"] }
wast = "25.0.0"
wast = "26.0.1"

2
cranelift/wasm/Cargo.toml
View File

@@ -12,7 +12,7 @@ keywords = ["webassembly", "wasm"]
edition = "2018"
[dependencies]
wasmparser = { version = "0.63.0", default-features = false }
wasmparser = { version = "0.65.0", default-features = false }
cranelift-codegen = { path = "../codegen", version = "0.67.0", default-features = false }
cranelift-entity = { path = "../entity", version = "0.67.0" }
cranelift-frontend = { path = "../frontend", version = "0.67.0", default-features = false }

9
cranelift/wasm/src/code_translator.rs
View File

@@ -1782,9 +1782,18 @@ pub fn translate_operator<FE: FuncEnvironment + ?Sized>(
state.push1(builder.ins().nearest(arg));
}
Operator::I32x4DotI16x8S => {
let (a, b) = pop2_with_bitcast(state, I16X8, builder);
state.push1(builder.ins().widening_pairwise_dot_product_s(a, b));
}
Operator::ReturnCall { .. } | Operator::ReturnCallIndirect { .. } => {
return Err(wasm_unsupported!("proposed tail-call operator {:?}", op));
}
Operator::V128Load32Zero { .. } | Operator::V128Load64Zero { .. } => {
return Err(wasm_unsupported!("proposed SIMD operator {:?}", op));
}
};
Ok(())
}

2
crates/debug/Cargo.toml
View File

@@ -13,7 +13,7 @@ edition = "2018"
[dependencies]
gimli = "0.22.0"
wasmparser = "0.63.0"
wasmparser = "0.65.0"
object = { version = "0.21.1", default-features = false, features = ["read", "write"] }
wasmtime-environ = { path = "../environ", version = "0.20.0" }
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 = "0.67.0", features = ["enable-serde"] }
cranelift-entity = { path = "../../cranelift/entity", version = "0.67.0", features = ["enable-serde"] }
cranelift-wasm = { path = "../../cranelift/wasm", version = "0.67.0", features = ["enable-serde"] }
wasmparser = "0.63.0"
wasmparser = "0.65.0"
indexmap = { version = "1.0.2", features = ["serde-1"] }
thiserror = "1.0.4"
serde = { version = "1.0.94", features = ["derive"] }

4
crates/fuzzing/Cargo.toml
View File

@@ -12,8 +12,8 @@ arbitrary = { version = "0.4.1", features = ["derive"] }
env_logger = "0.7.1"
log = "0.4.8"
rayon = "1.2.1"
wasmparser = "0.63.0"
wasmprinter = "0.2.10"
wasmparser = "0.65.0"
wasmprinter = "0.2.12"
wasmtime = { path = "../wasmtime" }
wasmtime-wast = { path = "../wast" }
wasm-smith = "0.1.9"

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 = "0.63.0"
wasmparser = "0.65.0"
more-asserts = "0.2.1"
anyhow = "1.0"
cfg-if = "1.0"

2
crates/lightbeam/Cargo.toml
View File

@@ -24,7 +24,7 @@ smallvec = "1.0.0"
staticvec = "0.10"
thiserror = "1.0.9"
typemap = "0.3"
wasmparser = "0.63.0"
wasmparser = "0.65.0"
[dev-dependencies]
lazy_static = "1.2"

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

@@ -13,6 +13,6 @@ edition = "2018"
[dependencies]
lightbeam = { path = "..", version = "0.20.0" }
wasmparser = "0.63"
wasmparser = "0.65"
cranelift-codegen = { path = "../../../cranelift/codegen", version = "0.67.0" }
wasmtime-environ = { path = "../../environ", version = "0.20.0" }

2
crates/wasmtime/Cargo.toml
View File

@@ -16,7 +16,7 @@ wasmtime-jit = { path = "../jit", version = "0.20.0" }
wasmtime-cache = { path = "../cache", version = "0.20.0", optional = true }
wasmtime-profiling = { path = "../profiling", version = "0.20.0" }
target-lexicon = { version = "0.11.0", default-features = false }
wasmparser = "0.63.0"
wasmparser = "0.65.0"
anyhow = "1.0.19"
region = "2.2.0"
libc = "0.2"

2
crates/wast/Cargo.toml
View File

@@ -13,7 +13,7 @@ edition = "2018"
[dependencies]
anyhow = "1.0.19"
wasmtime = { path = "../wasmtime", version = "0.20.0", default-features = false }
wast = "25.0.0"
wast = "26.0.1"
[badges]
maintenance = { status = "actively-developed" }