Skip to content

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

Add f16 and f128 support to NaN canonicalization (#12337)

Zijie Zhaof79fc4aea270
diff --git a/cranelift/codegen/src/nan_canonicalization.rs b/cranelift/codegen/src/nan_canonicalization.rs
index e6a73deaf8..761277f1fd 100644
--- a/cranelift/codegen/src/nan_canonicalization.rs
+++ b/cranelift/codegen/src/nan_canonicalization.rs
@@ -4,7 +4,7 @@
use crate::cursor::{Cursor, FuncCursor};
use crate::ir::condcodes::FloatCC;
-use crate::ir::immediates::{Ieee32, Ieee64};
+use crate::ir::immediates::{Ieee16, Ieee32, Ieee64, Ieee128};
use crate::ir::types::{self};
use crate::ir::{Function, Inst, InstBuilder, InstructionData, Opcode, Value};
use crate::opts::MemFlags;
@@ -90,6 +90,10 @@ fn add_nan_canon_seq(pos: &mut FuncCursor, inst: Inst, has_vector_support: bool)
};
match val_type {
+ types::F16 => {
+ let canon_nan = pos.ins().f16const(Ieee16::NAN);
+ scalar_select(pos, canon_nan);
+ }
types::F32 => {
let canon_nan = pos.ins().f32const(Ieee32::NAN);
if has_vector_support {
@@ -116,6 +120,11 @@ fn add_nan_canon_seq(pos: &mut FuncCursor, inst: Inst, has_vector_support: bool)
let canon_nan = pos.ins().splat(types::F64X2, canon_nan);
vector_select(pos, canon_nan);
}
+ types::F128 => {
+ let nan_const = pos.func.dfg.constants.insert(Ieee128::NAN.into());
+ let canon_nan = pos.ins().f128const(nan_const);
+ scalar_select(pos, canon_nan);
+ }
_ => {
// Panic if the type given was not an IEEE floating point type.
panic!("Could not canonicalize NaN: Unexpected result type found.");
diff --git a/cranelift/filetests/filetests/isa/riscv64/nan-canonicalization-f16.clif b/cranelift/filetests/filetests/isa/riscv64/nan-canonicalization-f16.clif
new file mode 100644
index 0000000000..ab6f26fbb8
--- /dev/null
+++ b/cranelift/filetests/filetests/isa/riscv64/nan-canonicalization-f16.clif
@@ -0,0 +1,9 @@
+test compile
+set enable_nan_canonicalization=true
+target riscv64 has_zfhmin has_zfh
+
+function %fadd_f16(f16, f16) -> f16 {
+block0(v0: f16, v1: f16):
+ v2 = fadd v0, v1
+ return v2
+}
diff --git a/cranelift/filetests/filetests/isa/s390x/nan-canonicalization-f128.clif b/cranelift/filetests/filetests/isa/s390x/nan-canonicalization-f128.clif
new file mode 100644
index 0000000000..3917ee87e3
--- /dev/null
+++ b/cranelift/filetests/filetests/isa/s390x/nan-canonicalization-f128.clif
@@ -0,0 +1,10 @@
+test compile
+set enable_nan_canonicalization=true
+target s390x
+
+function %fadd_f128(i64, f128, f128) {
+block0(v0: i64, v1: f128, v2: f128):
+ v3 = fadd v1, v2
+ store v3, v0
+ return
+}