# Compiler
NVCC=nvcc
GPU=4090

# Conditional setup based on the target GPU
ifeq ($(GPU),4090)
NVCCFLAGS=-DNDEBUG -Xcompiler=-fPIE --expt-extended-lambda --expt-relaxed-constexpr -Xcompiler=-Wno-psabi -Xcompiler=-fno-strict-aliasing --use_fast_math -I../../../include -forward-unknown-to-host-compiler -O3 -Xnvlink=--verbose -Xptxas=--verbose -Xptxas=--warn-on-spills -std=c++20 -MD -MT -MF -x cu -lrt -lpthread -ldl -DKITTENS_4090 -arch=sm_89 -lcuda -lcudadevrt -lcudart_static -lcublas
TARGET=attn_fwd
SRC=4090.cu
else
NVCCFLAGS=-DNDEBUG -Xcompiler=-fPIE --expt-extended-lambda --expt-relaxed-constexpr -Xcompiler=-Wno-psabi -Xcompiler=-fno-strict-aliasing --use_fast_math -I../../../include -forward-unknown-to-host-compiler -O3 -Xnvlink=--verbose -Xptxas=--verbose -Xptxas=--warn-on-spills -std=c++20 -MD -MT -MF -x cu -lrt -lpthread -ldl -DKITTENS_HOPPER -arch=sm_90a -lcuda -lcudadevrt -lcudart_static -lcublas
TARGET=attn_fwd
SRC=h100_lcf.cu
endif

# Default target
all: $(TARGET)

$(TARGET): $(SRC)
	$(NVCC) $(SRC) $(NVCCFLAGS) -o $(TARGET)

# Clean target
clean:
	rm -f $(TARGET)