feat(stablediffusion): Enable SYCL (#5144)

* feat(sycl): Enable SYCL for stable diffusion

This is a pain because we compile with CGO, but SD is compiled with
CMake. I don't think we can easily use CMake to set the linker flags
necessary. Also I could not find pkg-config calls that would fully set
the flags, so some of them are set manually.

See https://www.intel.com/content/www/us/en/developer/tools/oneapi/onemkl-link-line-advisor.html
for reference. I also resorted to searching the shared object files in
MKLROOT/lib for the symbols.

Signed-off-by: Richard Palethorpe <io@richiejp.com>

* fix(ci): Don't set nproc on cmake

Signed-off-by: Richard Palethorpe <io@richiejp.com>

---------

Signed-off-by: Richard Palethorpe <io@richiejp.com>

Author: richiejp

Makefile

@@ -21,8 +21,8 @@ BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
 BARKCPP_VERSION?=v1.0.0
 
 # stablediffusion.cpp (ggml)
-STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=19d876ee300a055629926ff836489901f734f2b7
+STABLEDIFFUSION_GGML_REPO?=https://github.com/richiejp/stable-diffusion.cpp
+STABLEDIFFUSION_GGML_VERSION?=6c848a2e0053c86984023570e706aa212933d50d
 
 ONNX_VERSION?=1.20.0
 ONNX_ARCH?=x64
@@ -260,11 +260,7 @@ backend/go/image/stablediffusion-ggml/libsd.a: sources/stablediffusion-ggml.cpp
 	$(MAKE) -C backend/go/image/stablediffusion-ggml libsd.a
 
 backend-assets/grpc/stablediffusion-ggml: backend/go/image/stablediffusion-ggml/libsd.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ LIBRARY_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion-ggml ./backend/go/image/stablediffusion-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion-ggml
-endif
+	$(MAKE) -C backend/go/image/stablediffusion-ggml stablediffusion-ggml
 
 sources/onnxruntime:
 	mkdir -p sources/onnxruntime
@@ -809,6 +805,7 @@ docker-aio-all:
 
 docker-image-intel:
 	docker build \
+		--progress plain \
 		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.0-0-devel-ubuntu24.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \

backend/cpp/llama/Makefile

@@ -80,8 +80,8 @@ grpc-server: llama.cpp llama.cpp/examples/grpc-server
 	@echo "Building grpc-server with $(BUILD_TYPE) build type and $(CMAKE_ARGS)"
 ifneq (,$(findstring sycl,$(BUILD_TYPE)))
 	+bash -c "source $(ONEAPI_VARS); \
-	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET) -j$(nproc)"
+	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET)"
 else
-	+cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET) -j$(nproc)
+	+cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET)
 endif
 	cp llama.cpp/build/bin/grpc-server .

backend/go/image/stablediffusion-ggml/Makefile

@@ -8,6 +8,11 @@ ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
 # keep standard at C11 and C++11
 CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/ggml/include -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp -O3 -DNDEBUG -std=c++17 -fPIC
 
+GOCMD?=go
+CGO_LDFLAGS?=
+GO_TAGS?=
+LD_FLAGS?=
+
 # Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
 
@@ -21,7 +26,7 @@ else ifeq ($(BUILD_TYPE),openblas)
 # If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
+# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
 else ifeq ($(BUILD_TYPE),hipblas)
 	CMAKE_ARGS+=-DGGML_HIP=ON
 # If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
@@ -36,16 +41,35 @@ else ifeq ($(OS),Darwin)
 	endif
 endif
 
-# ifeq ($(BUILD_TYPE),sycl_f16)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON -DSD_SYCL=ON -DGGML_SYCL_F16=ON
-# endif
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON \
+		-DGGML_SYCL_F16=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
 
-# ifeq ($(BUILD_TYPE),sycl_f32)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DSD_SYCL=ON
-# endif
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
 
 # warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
+# CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
 
 # Find all .a archives in ARCHIVE_DIR
 # (ggml can have different backends cpu, cuda, etc., each backend generates a .a archive)
@@ -86,11 +110,24 @@ endif
 	$(MAKE) $(COMBINED_LIB)
 
 gosd.o:
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	+bash -c "source $(ONEAPI_VARS); \
+	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c"
+else
 	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
+endif
 
 libsd.a: gosd.o
 	cp $(INCLUDE_PATH)/build/libstable-diffusion.a ./libsd.a
 	$(AR) rcs libsd.a gosd.o
 
+stablediffusion-ggml:
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH="$(INCLUDE_PATH)" LIBRARY_PATH="$(LIBRARY_PATH)" \
+	CC="$(CC)" CXX="$(CXX)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o ../../../../backend-assets/grpc/stablediffusion-ggml ./
+ifneq ($(UPX),)
+	$(UPX) ../../../../backend-assets/grpc/stablediffusion-ggml
+endif
+
 clean:
-	rm -rf gosd.o libsd.a build $(COMBINED_LIB)
+	rm -rf gosd.o libsd.a build $(COMBINED_LIB)