Wasm (#9)

* webgpu : fix build on emscripten

* more debugging stuff

* test-backend-ops: force single thread on wasm

* fix single-thread case for init_tensor_uniform

* use jspi

* add pthread

* test: remember to set n_thread for cpu backend

* Add buffer label and enable dawn-specific toggles to turn off some checks

* Intermediate state

* Fast working f16/f32 vec4

* Working float fast mul mat

* Clean up naming of mul_mat to match logical model, start work on q mul_mat

* Setup for subgroup matrix mat mul

* Basic working subgroup matrix

* Working subgroup matrix tiling

* Handle weirder sg matrix sizes (but still % sg matrix size)

* Working start to gemv

* working f16 accumulation with shared memory staging

* Print out available subgroup matrix configurations

* Vectorize dst stores for sg matrix shader

* Gemv working scalar

* Minor set_rows optimization (#4)

* updated optimization, fixed errors

* non vectorized version now dispatches one thread per element

* Simplify

* Change logic for set_rows pipelines

---------

Co-authored-by: Neha Abbas <nehaabbas@macbookpro.lan>
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
Co-authored-by: Reese Levine <reeselevine1@gmail.com>

* Comment on dawn toggles

* Working subgroup matrix code for (semi)generic sizes

* Remove some comments

* Cleanup code

* Update dawn version and move to portable subgroup size

* Try to fix new dawn release

* Update subgroup size comment

* Only check for subgroup matrix configs if they are supported

* Add toggles for subgroup matrix/f16 support on nvidia+vulkan

* Make row/col naming consistent

* Refactor shared memory loading

* Move sg matrix stores to correct file

* Working q4_0

* Formatting

* Work with emscripten builds

* Fix test-backend-ops emscripten for f16/quantized types

* Use emscripten memory64 to support get_memory

* Add build flags and try ci

---------

Co-authored-by: Xuan Son Nguyen <son@huggingface.co>

Author: reeselevine

.github/workflows/build.yml

@@ -547,6 +547,46 @@ jobs:
           # This is using llvmpipe and runs slower than other backends
           ctest -L main --verbose --timeout 3600
 
+  ubuntu-24-wasm-webgpu:
+    runs-on: ubuntu-24.04
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.16
+        with:
+          key: ubuntu-latest-wasm-webgpu
+          evict-old-files: 1d
+
+      - name: Install Emscripten
+        run: |
+          git clone https://github.com/emscripten-core/emsdk.git
+          cd emsdk
+          ./emsdk install latest
+          ./emsdk activate latest
+
+      - name: Fetch emdawnwebgpu
+        run: |
+          DAWN_TAG="v20251027.212519"
+          EMDAWN_PKG="emdawnwebgpu_pkg-${DAWN_TAG}.zip"
+          echo "Downloading ${EMDAWN_PKG}"
+          curl -L -o emdawn.zip \
+            "https://github.com/google/dawn/releases/download/${DAWN_TAG}/${EMDAWN_PKG}"
+          unzip emdawn.zip
+
+      - name: Build WASM WebGPU
+        run: |
+          source emsdk/emsdk_env.sh
+          emcmake cmake -B build-wasm \
+            -DGGML_WEBGPU=ON \
+            -DLLAMA_CURL=OFF \
+            -DEMDAWNWEBGPU_DIR=emdawnwebgpu_pkg
+
+          cmake --build build-wasm --target test-backend-ops -j $(nproc)
+
   ubuntu-22-cmake-hip:
     runs-on: ubuntu-22.04
     container: rocm/dev-ubuntu-22.04:6.1.2

.gitignore

@@ -152,3 +152,5 @@ poetry.toml
 # IDE
 *.code-workspace
 .windsurf/
+# emscripten
+a.out.*

CMakeLists.txt

@@ -36,7 +36,17 @@ option(LLAMA_USE_SYSTEM_GGML "Use system libggml" OFF)
 if (EMSCRIPTEN)
     set(BUILD_SHARED_LIBS_DEFAULT OFF)
 
-    option(LLAMA_WASM_SINGLE_FILE "llama: embed WASM inside the generated llama.js" ON)
+    # Use 64-bit memory to support backend_get_memory queries
+    # TODO: analyze performance impact, see https://spidermonkey.dev/blog/2025/01/15/is-memory64-actually-worth-using
+    add_compile_options("-sMEMORY64=1")
+    add_link_options("-sMEMORY64=1")
+    add_link_options("-sALLOW_MEMORY_GROWTH=1")
+
+    option(LLAMA_WASM_SINGLE_FILE "llama: embed WASM inside the generated llama.js" OFF)
+    option(LLAMA_BUILD_HTML "llama: build HTML file" ON)
+    if (LLAMA_BUILD_HTML)
+        set(CMAKE_EXECUTABLE_SUFFIX ".html")
+    endif()
 else()
     if (MINGW)
         set(BUILD_SHARED_LIBS_DEFAULT OFF)

common/arg.cpp

@@ -39,6 +39,7 @@
 #include "http.h"
 #endif
 
+#ifndef __EMSCRIPTEN__
 #ifdef __linux__
 #include <linux/limits.h>
 #elif defined(_WIN32)
@@ -50,8 +51,11 @@
 #else
 #include <sys/syslimits.h>
 #endif
+#endif
+
 #define LLAMA_MAX_URL_LENGTH 2084 // Maximum URL Length in Chrome: 2083
 
+
 // isatty
 #if defined(_WIN32)
 #include <io.h>

common/common.cpp

@@ -889,6 +889,8 @@ std::string fs_get_cache_directory() {
         cache_directory = std::getenv("HOME") + std::string("/Library/Caches/");
 #elif defined(_WIN32)
         cache_directory = std::getenv("LOCALAPPDATA");
+#elif defined(__EMSCRIPTEN__)
+        GGML_ABORT("not implemented on this platform");
 #else
 #  error Unknown architecture
 #endif

ggml/CMakeLists.txt

@@ -224,7 +224,7 @@ option(GGML_WEBGPU                          "ggml: use WebGPU"
 option(GGML_WEBGPU_DEBUG                    "ggml: enable WebGPU debug output"                OFF)
 option(GGML_WEBGPU_CPU_PROFILE              "ggml: enable WebGPU profiling (CPU)"             OFF)
 option(GGML_WEBGPU_GPU_PROFILE              "ggml: enable WebGPU profiling (GPU)"             OFF)
-
+option(GGML_WEBGPU_JSPI                     "ggml: use JSPI for WebGPU"                       ON)
 option(GGML_ZDNN                            "ggml: use zDNN"                                  OFF)
 option(GGML_METAL                           "ggml: use Metal"                                 ${GGML_METAL_DEFAULT})
 option(GGML_METAL_NDEBUG                    "ggml: disable Metal debugging"                   OFF)

ggml/src/ggml-webgpu/CMakeLists.txt

@@ -39,15 +39,33 @@ add_dependencies(ggml-webgpu generate_shaders)
 if(EMSCRIPTEN)
     set(EMDAWNWEBGPU_DIR "" CACHE PATH "Path to emdawnwebgpu_pkg")
 
-    target_compile_options(ggml-webgpu PRIVATE "--use-port=${EMDAWNWEBGPU_DIR}/emdawnwebgpu.port.py")
-    target_link_options(ggml-webgpu PRIVATE "--use-port=${EMDAWNWEBGPU_DIR}/emdawnwebgpu.port.py")
+    if(NOT EMDAWNWEBGPU_DIR)
+        # default built-in port
+        target_compile_options(ggml-webgpu PRIVATE "--use-port=emdawnwebgpu")
+        target_link_options(ggml-webgpu INTERFACE "--use-port=emdawnwebgpu")
+    else()
+        # custom port
+        target_compile_options(ggml-webgpu PRIVATE "--use-port=${EMDAWNWEBGPU_DIR}/emdawnwebgpu.port.py")
+        target_link_options(ggml-webgpu INTERFACE "--use-port=${EMDAWNWEBGPU_DIR}/emdawnwebgpu.port.py")
+    endif()
+
+    if (GGML_WEBGPU_JSPI)
+        target_compile_options(ggml-webgpu PRIVATE "-fwasm-exceptions")
+        target_link_options(ggml-webgpu INTERFACE "-sJSPI" "-fwasm-exceptions")
+    else()
+        target_compile_options(ggml-webgpu PRIVATE "-fexceptions")
+        target_link_options(ggml-webgpu INTERFACE "-sASYNCIFY" "-exceptions")
+    endif()
 else()
     find_package(Dawn REQUIRED)
     set(DawnWebGPU_TARGET dawn::webgpu_dawn)
 endif()
 
 if (GGML_WEBGPU_DEBUG)
     target_compile_definitions(ggml-webgpu PRIVATE GGML_WEBGPU_DEBUG=1)
+    if(EMSCRIPTEN)
+        target_link_options(ggml-webgpu INTERFACE "-sASSERTIONS=2")
+    endif()
 endif()
 
 if (GGML_WEBGPU_CPU_PROFILE)