Support WebAssembly Relaxed SIMD (#126)

Author: CryZe

.cargo/config.toml

@@ -1,2 +1,2 @@
-[target.wasm32-wasi]
+[target.wasm32-wasip1]
 runner = "wasmtime run --dir ."

.github/workflows/main.yml

@@ -66,23 +66,28 @@ jobs:
       uses: dtolnay/rust-toolchain@master
       with:
         toolchain: stable
-        target: wasm32-wasi
+        target: wasm32-wasip1
 
     - name: Install wasmtime
       run: |
         curl https://wasmtime.dev/install.sh -sSf | bash
         echo "$HOME/.wasmtime/bin" >> $GITHUB_PATH
 
     - name: Build with minimal features (no_std)
-      run: cargo build --target wasm32-wasi --verbose --no-default-features --features no-std-float
+      run: cargo build --target wasm32-wasip1 --verbose --no-default-features --features no-std-float
 
     - name: Run tests without SIMD
-      run: cargo test --target wasm32-wasi --verbose --no-default-features --features png-format
+      run: cargo test --target wasm32-wasip1 --verbose --no-default-features --features png-format
 
     - name: Run tests with SIMD128
       env:
         RUSTFLAGS: -Ctarget-feature=+simd128,+bulk-memory,+nontrapping-fptoint,+sign-ext
-      run: cargo test --target wasm32-wasi
+      run: cargo test --target wasm32-wasip1
+
+    - name: Run tests with Relaxed SIMD
+      env:
+        RUSTFLAGS: -Ctarget-feature=+simd128,+relaxed-simd,+bulk-memory,+nontrapping-fptoint,+sign-ext
+      run: cargo test --target wasm32-wasip1
 
   aarch64:
     runs-on: ubuntu-20.04

Cargo.toml

@@ -31,8 +31,8 @@ default = ["std", "simd", "png-format"]
 std = ["tiny-skia-path/std"]
 no-std-float = ["tiny-skia-path/no-std-float"]
 
-# Enables SIMD instructions on x86 (from SSE up to AVX2), WebAssembly (SIMD128)
-# and AArch64 (Neon).
+# Enables SIMD instructions on x86 (from SSE up to AVX2), WebAssembly (SIMD128,
+# Relaxed SIMD) and AArch64 (Neon).
 # Has no effect on other targets. Present mainly for testing.
 simd = []
 

src/wide/f32x4_t.rs

@@ -90,6 +90,8 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse2"))] {
                 Self(unsafe { _mm_max_ps(self.0, rhs.0) })
+            } else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                Self(f32x4_relaxed_max(self.0, rhs.0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 Self(f32x4_pmax(self.0, rhs.0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {
@@ -111,6 +113,8 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse2"))] {
                 Self(unsafe { _mm_min_ps(self.0, rhs.0) })
+            }  else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                Self(f32x4_relaxed_min(self.0, rhs.0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 Self(f32x4_pmin(self.0, rhs.0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {
@@ -245,6 +249,8 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse4.1"))] {
                 Self(unsafe { _mm_blendv_ps(f.0, t.0, self.0) })
+            } else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                Self(i32x4_relaxed_laneselect(t.0, f.0, self.0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 Self(v128_bitselect(t.0, f.0, self.0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {
@@ -302,6 +308,8 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse2"))] {
                 i32x4(unsafe { _mm_cvtps_epi32(self.0) })
+            } else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                i32x4(i32x4_relaxed_trunc_f32x4(self.round().0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 i32x4(i32x4_trunc_sat_f32x4(self.round().0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {
@@ -325,6 +333,8 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse2"))] {
                 i32x4(unsafe { _mm_cvttps_epi32(self.0) })
+            } else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                i32x4(i32x4_relaxed_trunc_f32x4(self.0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 i32x4(i32x4_trunc_sat_f32x4(self.0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {

src/wide/f32x8_t.rs

@@ -43,11 +43,17 @@ impl f32x8 {
     }
 
     pub fn floor(self) -> Self {
-        let roundtrip: f32x8 = cast(self.trunc_int().to_f32x8());
-        roundtrip
-            - roundtrip
-                .cmp_gt(self)
-                .blend(f32x8::splat(1.0), f32x8::default())
+        cfg_if::cfg_if! {
+            if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                Self(self.0.floor(), self.1.floor())
+            } else {
+                let roundtrip: f32x8 = cast(self.trunc_int().to_f32x8());
+                roundtrip
+                    - roundtrip
+                        .cmp_gt(self)
+                        .blend(f32x8::splat(1.0), f32x8::default())
+            }
+        }
     }
 
     pub fn fract(self) -> Self {

src/wide/i32x4_t.rs

@@ -56,6 +56,8 @@ impl i32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse4.1"))] {
                 Self(unsafe { _mm_blendv_epi8(f.0, t.0, self.0) })
+            } else if #[cfg(all(feature = "simd", target_feature = "relaxed-simd"))] {
+                Self(i32x4_relaxed_laneselect(t.0, f.0, self.0))
             } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
                 Self(v128_bitselect(t.0, f.0, self.0))
             } else if #[cfg(all(feature = "simd", target_arch = "aarch64", target_feature = "neon"))] {