pulley: Implement float arithmetic operations (bytecodealliance#9808)

* pulley: Implement float arithmetic operations

Fill out enough to get `f32.wast` and `f64.wast` spec tests working. A
minor ABI issue was discovered along the way which is also required to
get a new test working on both 32 and 64-bit platforms.

cc bytecodealliance#9783

* Centralize handling of float operations

Add a new `wasmtime-math` crate which is tasked with providing the float
math functions needed by Wasm with with, in theory, most optimal
platform implementation available.

* Fix riscv64 tests

Cargo.toml

@@ -234,6 +234,7 @@ wasi-common = { path = "crates/wasi-common", version = "=29.0.0", default-featur
 wasmtime-fuzzing = { path = "crates/fuzzing" }
 wasmtime-jit-icache-coherence = { path = "crates/jit-icache-coherence", version = "=29.0.0" }
 wasmtime-wit-bindgen = { path = "crates/wit-bindgen", version = "=29.0.0" }
+wasmtime-math = { path = "crates/math", version = "=29.0.0" }
 test-programs-artifacts = { path = 'crates/test-programs/artifacts' }
 
 pulley-interpreter = { path = 'pulley', version = "=29.0.0" }

cranelift/codegen/src/isa/pulley_shared/abi.rs

@@ -564,15 +564,15 @@ where
             || clobber_size > 0
             || fixed_frame_storage_size > 0
         {
-            16 // FP, LR
+            P::pointer_width().bytes() * 2 // FP, LR
         } else {
             0
         };
 
         FrameLayout {
             incoming_args_size,
             tail_args_size,
-            setup_area_size,
+            setup_area_size: setup_area_size.into(),
             clobber_size,
             fixed_frame_storage_size,
             outgoing_args_size,

cranelift/codegen/src/isa/pulley_shared/lower.isle

@@ -589,3 +589,58 @@
 
 (rule (lower (has_type $F64 (fcopysign a b)))
   (pulley_fcopysign64 a b))
+
+;;;; Rules for `fadd` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fadd a b))) (pulley_fadd32 a b))
+(rule (lower (has_type $F64 (fadd a b))) (pulley_fadd64 a b))
+
+;;;; Rules for `fsub` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fsub a b))) (pulley_fsub32 a b))
+(rule (lower (has_type $F64 (fsub a b))) (pulley_fsub64 a b))
+
+;;;; Rules for `fmul` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fmul a b))) (pulley_fmul32 a b))
+(rule (lower (has_type $F64 (fmul a b))) (pulley_fmul64 a b))
+
+;;;; Rules for `fdiv` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fdiv a b))) (pulley_fdiv32 a b))
+(rule (lower (has_type $F64 (fdiv a b))) (pulley_fdiv64 a b))
+
+;;;; Rules for `fmax` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fmax a b))) (pulley_fmaximum32 a b))
+(rule (lower (has_type $F64 (fmax a b))) (pulley_fmaximum64 a b))
+
+;;;; Rules for `fmin` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (fmin a b))) (pulley_fminimum32 a b))
+(rule (lower (has_type $F64 (fmin a b))) (pulley_fminimum64 a b))
+
+;;;; Rules for `trunc` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (trunc a))) (pulley_ftrunc32 a))
+(rule (lower (has_type $F64 (trunc a))) (pulley_ftrunc64 a))
+
+;;;; Rules for `floor` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (floor a))) (pulley_ffloor32 a))
+(rule (lower (has_type $F64 (floor a))) (pulley_ffloor64 a))
+
+;;;; Rules for `ceil` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (ceil a))) (pulley_fceil32 a))
+(rule (lower (has_type $F64 (ceil a))) (pulley_fceil64 a))
+
+;;;; Rules for `nearest` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (nearest a))) (pulley_fnearest32 a))
+(rule (lower (has_type $F64 (nearest a))) (pulley_fnearest64 a))
+
+;;;; Rules for `sqrt` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+(rule (lower (has_type $F32 (sqrt a))) (pulley_fsqrt32 a))
+(rule (lower (has_type $F64 (sqrt a))) (pulley_fsqrt64 a))

cranelift/codegen/src/isa/pulley_shared/mod.rs

@@ -54,6 +54,10 @@ impl PointerWidth {
             PointerWidth::PointerWidth64 => 64,
         }
     }
+
+    pub fn bytes(self) -> u8 {
+        self.bits() / 8
+    }
 }
 
 /// A Pulley backend.

crates/math/Cargo.toml

@@ -0,0 +1,22 @@
+[package]
+name = "wasmtime-math"
+version.workspace = true
+authors.workspace = true
+description = "Low-level math routines used in Wasmtime"
+documentation = "https://docs.rs/wasmtime-math"
+license = "Apache-2.0 WITH LLVM-exception"
+repository = "https://github.com/bytecodealliance/wasmtime"
+edition.workspace = true
+rust-version.workspace = true
+
+[lints]
+workspace = true
+
+[package.metadata.docs.rs]
+all-features = true
+
+[features]
+std = []
+
+[dependencies]
+libm = { workspace = true }