Skip to content

A lightweight WebAssembly runtime that is fast, secure, and standards-compliant

winch(aarch64): v128 float ops (#13958)

Daniel Macovei33a607e1a270
diff --git a/tests/disas/winch/aarch64/v128_ops/f32x4_div.wat b/tests/disas/winch/aarch64/v128_ops/f32x4_div.wat
new file mode 100644
index 0000000000..34b23d7bcb
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f32x4_div.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f32x4_div") (param v128) (param v128) (result v128)
+ (f32x4.div (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fdiv v1.4s, v1.4s, v0.4s
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f32x4_max.wat b/tests/disas/winch/aarch64/v128_ops/f32x4_max.wat
new file mode 100644
index 0000000000..8c570af9ad
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f32x4_max.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f32x4_max") (param v128) (param v128) (result v128)
+ (f32x4.max (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fmax v1.4s, v1.4s, v0.4s
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f32x4_min.wat b/tests/disas/winch/aarch64/v128_ops/f32x4_min.wat
new file mode 100644
index 0000000000..e54ed26e6f
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f32x4_min.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f32x4_min") (param v128) (param v128) (result v128)
+ (f32x4.min (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fmin v1.4s, v1.4s, v0.4s
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f32x4_pmax.wat b/tests/disas/winch/aarch64/v128_ops/f32x4_pmax.wat
new file mode 100644
index 0000000000..5e45ce1ad2
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f32x4_pmax.wat
@@ -0,0 +1,41 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f32x4_pmax") (param v128) (param v128) (result v128)
+ (f32x4.pmax (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x78
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fcmgt v31.4s, v0.4s, v1.4s
+;; bsl v31.16b, v0.16b, v1.16b
+;; mov v1.16b, v31.16b
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 78: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f32x4_pmin.wat b/tests/disas/winch/aarch64/v128_ops/f32x4_pmin.wat
new file mode 100644
index 0000000000..b4ecc25bca
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f32x4_pmin.wat
@@ -0,0 +1,41 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f32x4_pmin") (param v128) (param v128) (result v128)
+ (f32x4.pmin (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x78
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fcmgt v31.4s, v1.4s, v0.4s
+;; bsl v31.16b, v0.16b, v1.16b
+;; mov v1.16b, v31.16b
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 78: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f64x2_div.wat b/tests/disas/winch/aarch64/v128_ops/f64x2_div.wat
new file mode 100644
index 0000000000..d231f2d940
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f64x2_div.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f64x2_div") (param v128) (param v128) (result v128)
+ (f64x2.div (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fdiv v1.2d, v1.2d, v0.2d
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f64x2_max.wat b/tests/disas/winch/aarch64/v128_ops/f64x2_max.wat
new file mode 100644
index 0000000000..76d6fb77cb
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f64x2_max.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f64x2_max") (param v128) (param v128) (result v128)
+ (f64x2.max (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fmax v1.2d, v1.2d, v0.2d
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f64x2_min.wat b/tests/disas/winch/aarch64/v128_ops/f64x2_min.wat
new file mode 100644
index 0000000000..489199fb18
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f64x2_min.wat
@@ -0,0 +1,39 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f64x2_min") (param v128) (param v128) (result v128)
+ (f64x2.min (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x70
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fmin v1.2d, v1.2d, v0.2d
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 70: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f64x2_pmax.wat b/tests/disas/winch/aarch64/v128_ops/f64x2_pmax.wat
new file mode 100644
index 0000000000..62f2b13dab
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f64x2_pmax.wat
@@ -0,0 +1,41 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f64x2_pmax") (param v128) (param v128) (result v128)
+ (f64x2.pmax (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x78
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fcmgt v31.2d, v0.2d, v1.2d
+;; bsl v31.16b, v0.16b, v1.16b
+;; mov v1.16b, v31.16b
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 78: udf #0xc11f
diff --git a/tests/disas/winch/aarch64/v128_ops/f64x2_pmin.wat b/tests/disas/winch/aarch64/v128_ops/f64x2_pmin.wat
new file mode 100644
index 0000000000..cd41d8149c
--- /dev/null
+++ b/tests/disas/winch/aarch64/v128_ops/f64x2_pmin.wat
@@ -0,0 +1,41 @@
+;;! target = "aarch64"
+;;! test = "winch"
+
+(module
+ (memory (data "\00\00\00\00\00\00\00\00\00\00\00\00\00\00\a0\7f"))
+
+ (func (export "f64x2_pmin") (param v128) (param v128) (result v128)
+ (f64x2.pmin (local.get 0) (local.get 1)))
+)
+;; wasm[0]::function[0]:
+;; stp x29, x30, [sp, #-0x10]!
+;; mov x29, sp
+;; str x28, [sp, #-0x10]!
+;; mov x28, sp
+;; ldur x16, [x0, #8]
+;; ldur x16, [x16, #0x18]
+;; mov x17, #0
+;; movk x17, #0x30
+;; add x16, x16, x17
+;; cmp sp, x16
+;; b.lo #0x78
+;; 2c: mov x9, x0
+;; sub x28, x28, #0x30
+;; mov sp, x28
+;; stur x0, [x28, #0x28]
+;; stur x1, [x28, #0x20]
+;; stur q0, [x28, #0x10]
+;; stur q1, [x28]
+;; ldur q0, [x28]
+;; ldur q1, [x28, #0x10]
+;; fcmgt v31.2d, v1.2d, v0.2d
+;; bsl v31.16b, v0.16b, v1.16b
+;; mov v1.16b, v31.16b
+;; mov v0.16b, v1.16b
+;; add x28, x28, #0x30
+;; mov sp, x28
+;; mov sp, x28
+;; ldr x28, [sp], #0x10
+;; ldp x29, x30, [sp], #0x10
+;; ret
+;; 78: udf #0xc11f
diff --git a/winch/codegen/src/isa/aarch64/masm.rs b/winch/codegen/src/isa/aarch64/masm.rs
index 663830b241..419252ab16 100644
--- a/winch/codegen/src/isa/aarch64/masm.rs
+++ b/winch/codegen/src/isa/aarch64/masm.rs
@@ -2090,9 +2090,8 @@ impl Masm for MacroAssembler {
kind: V128MinKind,
) -> Result<()> {
let (op, size) = match kind {
- V128MinKind::F32x4 | V128MinKind::F64x2 => {
- bail!(CodeGenError::unimplemented_masm_instruction())
- }
+ V128MinKind::F32x4 => (VecALUOp::Fmin, VectorSize::Size32x4),
+ V128MinKind::F64x2 => (VecALUOp::Fmin, VectorSize::Size64x2),
V128MinKind::I8x16S => (VecALUOp::Smin, VectorSize::Size8x16),
V128MinKind::I8x16U => (VecALUOp::Umin, VectorSize::Size8x16),
V128MinKind::I16x8S => (VecALUOp::Smin, VectorSize::Size16x8),
@@ -2112,9 +2111,8 @@ impl Masm for MacroAssembler {
kind: V128MaxKind,
) -> Result<()> {
let (op, size) = match kind {
- V128MaxKind::F32x4 | V128MaxKind::F64x2 => {
- bail!(CodeGenError::unimplemented_masm_instruction())
- }
+ V128MaxKind::F32x4 => (VecALUOp::Fmax, VectorSize::Size32x4),
+ V128MaxKind::F64x2 => (VecALUOp::Fmax, VectorSize::Size64x2),
V128MaxKind::I8x16S => (VecALUOp::Smax, VectorSize::Size8x16),
V128MaxKind::I8x16U => (VecALUOp::Umax, VectorSize::Size8x16),
V128MaxKind::I16x8S => (VecALUOp::Smax, VectorSize::Size16x8),
@@ -2193,14 +2191,15 @@ impl Masm for MacroAssembler {
Ok(())
}
- fn v128_div(
- &mut self,
- _lhs: Reg,
- _rhs: Reg,
- _dst: WritableReg,
- _size: OperandSize,
- ) -> Result<()> {
- bail!(CodeGenError::unimplemented_masm_instruction())
+ fn v128_div(&mut self, lhs: Reg, rhs: Reg, dst: WritableReg, size: OperandSize) -> Result<()> {
+ self.asm.vec_rrr(
+ VecALUOp::Fdiv,
+ lhs,
+ rhs,
+ dst,
+ VectorSize::from_lane_size(size.into(), true),
+ );
+ Ok(())
}
fn v128_sqrt(&mut self, src: Reg, dst: WritableReg, size: OperandSize) -> Result<()> {
@@ -2243,24 +2242,28 @@ impl Masm for MacroAssembler {
Ok(())
}
- fn v128_pmin(
- &mut self,
- _lhs: Reg,
- _rhs: Reg,
- _dst: WritableReg,
- _size: OperandSize,
- ) -> Result<()> {
- bail!(CodeGenError::unimplemented_masm_instruction())
+ fn v128_pmin(&mut self, lhs: Reg, rhs: Reg, dst: WritableReg, size: OperandSize) -> Result<()> {
+ let size = VectorSize::from_lane_size(size.into(), true);
+ self.with_scratch::<FloatScratch, _>(|masm, mask| {
+ masm.asm
+ .vec_rrr(VecALUOp::Fcmgt, lhs, rhs, mask.writable(), size);
+ masm.asm
+ .vec_rrr_mod(VecALUModOp::Bsl, rhs, lhs, mask.writable(), size);
+ masm.asm.fmov_rr(mask.inner(), dst, OperandSize::S128);
+ });
+ Ok(())
}
- fn v128_pmax(
- &mut self,
- _lhs: Reg,
- _rhs: Reg,
- _dst: WritableReg,
- _size: OperandSize,
- ) -> Result<()> {
- bail!(CodeGenError::unimplemented_masm_instruction())
+ fn v128_pmax(&mut self, lhs: Reg, rhs: Reg, dst: WritableReg, size: OperandSize) -> Result<()> {
+ let size = VectorSize::from_lane_size(size.into(), true);
+ self.with_scratch::<FloatScratch, _>(|masm, mask| {
+ masm.asm
+ .vec_rrr(VecALUOp::Fcmgt, rhs, lhs, mask.writable(), size);
+ masm.asm
+ .vec_rrr_mod(VecALUModOp::Bsl, rhs, lhs, mask.writable(), size);
+ masm.asm.fmov_rr(mask.inner(), dst, OperandSize::S128);
+ });
+ Ok(())
}
}