improved portability for Apple M-series chips

Makefile

@@ -1,4 +1,11 @@
-NUM_JOBS=$(shell nproc)
+OS := $(shell uname | tr -d '\n')
+
+ifeq ($(OS),Darwin)
+NUM_JOBS := $(shell sysctl -n hw.ncpu)
+else
+NUM_JOBS := $(shell nproc)
+endif
+
 CXX=clang++
 
 .PHONY: default examples/hello_world/build/hello_world tests libgpu debug build check-clang clean-build clean all watch-tests docs
@@ -20,7 +27,7 @@ pch:
 
 # TODO(avh): change extension based on platform
 # Get the current OS name
-OS = $(shell uname | tr -d '\n')
+
 # Set the specific variables for each platform
 LIB_PATH ?= /usr/lib
 HEADER_PATH ?= /usr/include
@@ -89,7 +96,13 @@ RELEASE_FLAGS = $(FLAGS) -DFASTBUILD:BOOL=OFF
 TARGET_LIB=gpu
 
 libgpu-cmake: check-clang check-cmake
-	$(CMAKE_CMD) $(RELEASE_FLAGS) && make -j$(NUM_JOBS) gpu
+	$(CMAKE_CMD) $(RELEASE_FLAGS) && cmake --build . --parallel $(NUM_JOBS) --target gpud
+	mkdir -p third_party/lib
+ifeq ($(OS),Darwin)
+	find third_party/fetchcontent -name "*.dylib" -exec cp {} third_party/lib/ \;
+else ifeq ($(OS),Linux)
+	find third_party/fetchcontent -name "*.so*" -exec cp {} third_party/lib/ \;
+endif
 
 debug-cmake: check-clang check-cmake
 	$(CMAKE_CMD) $(DEBUG_FLAGS) && make -j$(NUM_JOBS) $(TARGET_ALL)
@@ -102,7 +115,7 @@ all-cmake: check-clang check-cmake
 ################################################################################
 
 clean-dawnlib:
-	rm -f third_party/lib/libwebgpu_dawn.so third_party/lib/libwebgpu_dawn.dylib
+	rm -f third_party/lib/*.so third_party/lib/*.so.* third_party/lib/*.dylib
 
 clean:
 	read -r -p "This will delete the contents of build/*. Are you sure? [CTRL-C to abort] " response && rm -rf build/*

README.md

@@ -41,7 +41,11 @@ Optionally, Dawn can be built from scratch with gpu.cpp using the cmake build sc
 After cloning the repo, from the top-level gpu.cpp, you should be able to build and run the hello world GELU example by typing:
 
 ```
-make
+make libgpu-cmake
+cd examples
+make clean
+make hello_world
+make run_hello_world
 ```
 
 The first time you build and run the project this way, it will download a prebuilt shared library for the Dawn native WebGPU implementation automatically (using the setup.py script). This places the Dawn shared library in the `third_party/lib` directory. Afterwards you should see `libdawn.dylib` on MacOS or `libdawn.so` on Linux. This download only occurs once.

cmake/webgpu.cmake

@@ -39,19 +39,13 @@ endif()
 
 option(WEBGPU_TAG "WebGPU distribution tag to use")
 if(NOT WEBGPU_TAG)
-    set(WEBGPU_TAG "dawn")
+    set(WEBGPU_TAG "main")
 endif()
 message(STATUS "Using WebGPU distribution tag: ${WEBGPU_TAG}")
 
-if(WEBGPU_TAG STREQUAL "dawn")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DWEBGPU_BACKEND_DAWN")
-    # use specific commit set(WEBGPU_TAG
-    # "1025b977e1927b6d0327e67352f90feb4bcf8274") set(WEBGPU_TAG
-    # "acf972b7b909f52e183bdae3971b93bb13d4a29e")
-    # add_compile_options(-UABSL_INTERNAL_AT_LEAST_CXX20) set(CMAKE_CXX_FLAGS
-    # "${CMAKE_CXX_FLAGS} -UABSL_INTERNAL_AT_LEAST_CXX20")
-    message(STATUS "Using Dawn backend")
-endif()
+set(WEBGPU_BACKEND "DAWN" CACHE STRING "WebGPU backend")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DWEBGPU_BACKEND_DAWN")
+message(STATUS "Using Dawn backend")
 
 FetchContent_Declare(
     webgpu

examples/Makefile

@@ -14,7 +14,7 @@ else
 endif
 
 FLAGS=-std=c++17 $(STDLIB) -I$(GPUCPP) -I$(GPUCPP)/third_party/headers -L$(GPUCPP)/third_party/lib
-LFLAGS=-ldl -lwebgpu_dawn
+LFLAGS=-Wl,-rpath,$(GPUCPP)/third_party/lib -ldl -lwebgpu_dawn
 
 .PHONY: default all_release all_debug dawnlib run_setup check-python
 .PHONY: $(addsuffix _release, $(TARGETS))
@@ -57,7 +57,7 @@ $(foreach target,$(TARGETS),$(eval $(call CLEAN_RULES,$(target))))
 
 clean: $(addprefix clean_, $(TARGETS))
 
-dawnlib: $(if $(wildcard $(GPUCPP)/third_party/lib/libdawn.so $(GPUCPP)/third_party/lib/libdawn.dylib),,run_setup)
+dawnlib: $(if $(wildcard $(GPUCPP)/third_party/lib/libwebgpu_dawn.so $(GPUCPP)/third_party/lib/libwebgpu_dawn.dylib),,run_setup)
 
 run_setup: check-python
 	cd $(GPUCPP) && (command -v python3 >/dev/null 2>&1 && python3 setup.py || python setup.py)