Make wasm +sign-ext and +nontrapping-fptoint the default

These have been supported in ~all wasm runtimes for a while now, and +nontrapping-fptoint in particular can make a big performance difference. We should enable these by default, and add a new backdoor (wasm_mvponly) for code paths that need to use the original wasm Minimum Viable Product spec only.

README_webassembly.md

@@ -6,10 +6,11 @@ backend.
 As WebAssembly itself is still under active development, Halide's support has
 some limitations. Some of the most important:
 
+-   Sign-extension operations are enabled by default (but can be avoided via
+    Target::WasmMvpOnly).
+-   Non-trapping float-to-int conversions are enabled by default (but can be
+    avoided via Target::WasmMvpOnly).
 -   Fixed-width SIMD (128 bit) can be enabled via Target::WasmSimd128.
--   Sign-extension operations can be enabled via Target::WasmSignExt.
--   Non-trapping float-to-int conversions can be enabled via
-    Target::WasmSatFloatToInt.
 -   Threads have very limited support via Target::WasmThreads; see
     [below](#using-threads) for more details.
 -   Halide's JIT for Wasm is extremely limited and really useful only for
@@ -152,9 +153,8 @@ cmake -DLLVM_ENABLE_PROJECTS="clang;lld" ...
 ```
 
 -   To run the JIT tests, set `HL_JIT_TARGET=wasm-32-wasmrt` (possibly adding
-    `wasm_simd128`, `wasm_signext`, and/or `wasm_sat_float_to_int`) and run
-    CMake/CTest normally. Note that wasm testing is only support under CMake
-    (not via Make).
+    `wasm_simd128`) and run CMake/CTest normally. Note that wasm testing is
+    only supported under CMake (not via Make).
 
 ## Enabling wasm AOT
 
@@ -165,9 +165,8 @@ will), you need to install Emscripten locally.
     (https://emscripten.org/docs/getting_started/downloads.html).
 
 -   To run the AOT tests, set `HL_TARGET=wasm-32-wasmrt` (possibly adding
-    `wasm_simd128`, `wasm_signext`, and/or `wasm_sat_float_to_int`) and run
-    CMake/CTest normally. Note that wasm testing is only support under CMake
-    (not via Make).
+    `wasm_simd128`) and run CMake/CTest normally. Note that wasm testing is
+    only supported under CMake (not via Make).
 
 # Running benchmarks
 

python_bindings/src/halide/halide_/PyEnums.cpp

@@ -165,9 +165,8 @@ void define_enums(py::module &m) {
         .value("HexagonDma", Target::Feature::HexagonDma)
         .value("EmbedBitcode", Target::Feature::EmbedBitcode)
         .value("EnableLLVMLoopOpt", Target::Feature::EnableLLVMLoopOpt)
+        .value("WasmMvpOnly", Target::Feature::WasmMvpOnly)
         .value("WasmSimd128", Target::Feature::WasmSimd128)
-        .value("WasmSignExt", Target::Feature::WasmSignExt)
-        .value("WasmSatFloatToInt", Target::Feature::WasmSatFloatToInt)
         .value("WasmThreads", Target::Feature::WasmThreads)
         .value("WasmBulkMemory", Target::Feature::WasmBulkMemory)
         .value("SVE", Target::Feature::SVE)

src/CodeGen_WebAssembly.cpp

@@ -333,21 +333,17 @@ string CodeGen_WebAssembly::mattrs() const {
     std::ostringstream s;
     string sep;
 
-    if (target.has_feature(Target::WasmSignExt)) {
+    if (!target.has_feature(Target::WasmMvpOnly)) {
         s << sep << "+sign-ext";
         sep = ",";
+        s << sep << "+nontrapping-fptoint";
     }
 
     if (target.has_feature(Target::WasmSimd128)) {
         s << sep << "+simd128";
         sep = ",";
     }
 
-    if (target.has_feature(Target::WasmSatFloatToInt)) {
-        s << sep << "+nontrapping-fptoint";
-        sep = ",";
-    }
-
     if (target.has_feature(Target::WasmThreads)) {
         // "WasmThreads" doesn't directly affect LLVM codegen,
         // but it does end up requiring atomics, so be sure to enable them.

src/Target.cpp

@@ -533,8 +533,7 @@ const std::map<std::string, Target::Feature> feature_name_map = {
     {"embed_bitcode", Target::EmbedBitcode},
     {"enable_llvm_loop_opt", Target::EnableLLVMLoopOpt},
     {"wasm_simd128", Target::WasmSimd128},
-    {"wasm_signext", Target::WasmSignExt},
-    {"wasm_sat_float_to_int", Target::WasmSatFloatToInt},
+    {"wasm_mvponly", Target::WasmMvpOnly},
     {"wasm_threads", Target::WasmThreads},
     {"wasm_bulk_memory", Target::WasmBulkMemory},
     {"webgpu", Target::WebGPU},

src/Target.h

@@ -143,9 +143,8 @@ struct Target {
         CheckUnsafePromises = halide_target_feature_check_unsafe_promises,
         EmbedBitcode = halide_target_feature_embed_bitcode,
         EnableLLVMLoopOpt = halide_target_feature_enable_llvm_loop_opt,
+        WasmMvpOnly = halide_target_feature_wasm_mvponly,
         WasmSimd128 = halide_target_feature_wasm_simd128,
-        WasmSignExt = halide_target_feature_wasm_signext,
-        WasmSatFloatToInt = halide_target_feature_wasm_sat_float_to_int,
         WasmThreads = halide_target_feature_wasm_threads,
         WasmBulkMemory = halide_target_feature_wasm_bulk_memory,
         WebGPU = halide_target_feature_webgpu,

src/WasmExecutor.cpp

@@ -1308,15 +1308,13 @@ wabt::interp::HostFunc::Ptr make_extern_callback(wabt::interp::Store &store,
 
 wabt::Features calc_features(const Target &target) {
     wabt::Features f;
-    if (target.has_feature(Target::WasmSignExt)) {
+    if (!target.has_feature(Target::WasmMvpOnly)) {
         f.enable_sign_extension();
+        f.enable_sat_float_to_int();
     }
     if (target.has_feature(Target::WasmSimd128)) {
         f.enable_simd();
     }
-    if (target.has_feature(Target::WasmSatFloatToInt)) {
-        f.enable_sat_float_to_int();
-    }
     return f;
 }
 #endif  // WITH_WABT

src/runtime/HalideRuntime.h

@@ -1386,9 +1386,8 @@ typedef enum halide_target_feature_t {
     halide_target_feature_hexagon_dma,            ///< Enable Hexagon DMA buffers.
     halide_target_feature_embed_bitcode,          ///< Emulate clang -fembed-bitcode flag.
     halide_target_feature_enable_llvm_loop_opt,   ///< Enable loop vectorization + unrolling in LLVM. Overrides halide_target_feature_disable_llvm_loop_opt. (Ignored for non-LLVM targets.)
+    halide_target_feature_wasm_mvponly,           ///< Disable all extensions to WebAssembly codegen (including +sign-ext and +nontrapping-fptoint, which are on by default).
     halide_target_feature_wasm_simd128,           ///< Enable +simd128 instructions for WebAssembly codegen.
-    halide_target_feature_wasm_signext,           ///< Enable +sign-ext instructions for WebAssembly codegen.
-    halide_target_feature_wasm_sat_float_to_int,  ///< Enable saturating (nontrapping) float-to-int instructions for WebAssembly codegen.
     halide_target_feature_wasm_threads,           ///< Enable use of threads in WebAssembly codegen. Requires the use of a wasm runtime that provides pthread-compatible wrappers (typically, Emscripten with the -pthreads flag). Unsupported under WASI.
     halide_target_feature_wasm_bulk_memory,       ///< Enable +bulk-memory instructions for WebAssembly codegen.
     halide_target_feature_webgpu,                 ///< Enable the WebGPU runtime.

test/correctness/simd_op_check_wasm.cpp

@@ -16,8 +16,8 @@ class SimdOpCheckWASM : public SimdOpCheckTest {
     SimdOpCheckWASM(Target t, int w = 768, int h = 128)
         : SimdOpCheckTest(t, w, h) {
         use_wasm_simd128 = target.has_feature(Target::WasmSimd128);
-        use_wasm_sat_float_to_int = target.has_feature(Target::WasmSatFloatToInt);
-        use_wasm_sign_ext = target.has_feature(Target::WasmSignExt);
+        use_wasm_sign_ext = !target.has_feature(Target::WasmMvpOnly);
+        use_wasm_sat_float_to_int = !target.has_feature(Target::WasmMvpOnly);
     }
 
     void add_tests() override {