add example with run-time-parametrized matmul

programming_examples/basic/matrix_multiplication/rtp/Makefile

@@ -0,0 +1,113 @@
+##===- Makefile -----------------------------------------------------------===##
+# 
+# This file licensed under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# 
+##===----------------------------------------------------------------------===##
+
+# ---
+
+# The following environment variables that point to the Xilinx runtime (XRT)
+# should be set up by an environment setup script already.
+XILINX_XRT?=/opt/xilinx/xrt
+XILINX_VITIS?=/tools/Xilinx/Vitis/2023.2
+BOOST_ROOT?=/usr/include/boost
+Boost_INCLUDE_DIRS=${BOOST_ROOT}
+
+# ---
+
+XILINX_XRT_INCLUDE?=${XILINX_XRT}/include
+XILINX_XRT_LIB?=${XILINX_XRT}/lib
+
+CXX=g++-13
+
+CXXFLAGS+=-std=gnu++23 -ggdb -I${XILINX_XRT_INCLUDE} ${Boost_INCLUDE_DIRS:%=-I%}
+LDFLAGS+=-L${XILINX_XRT_LIB} ${Boost_LIBRARY_DIRS:%=-L%}
+LDLIBS+=-lxrt_coreutil -lboost_program_options -lboost_filesystem -luuid
+
+CHESSCCWRAP2_FLAGS = aie2 -I${XILINX_VITIS}/aietools/include 
+
+mlir_target?=build/aie.mlir
+xclbin_target?=build/final.xclbin
+insts_target?=build/insts.txt
+kernel_target?=build/mm_${m}x${k}x${n}.o
+
+M=256
+K=3072
+N=50304
+m=64
+k=64
+n=32
+n_aie_cols=4
+dtype_in=bf16
+dtype_out=f32
+
+aieargs+=-m $m -k $k -n $n --n-aie-cols ${n_aie_cols} --dtype_in ${dtype_in} --dtype_out ${dtype_out}
+
+.PHONY: all
+all: test ${xclbin_target} ${host_target} \
+	build/insts_256x768x2304.txt \
+	build/insts_256x768x768.txt \
+	build/insts_256x768x3072.txt \
+	build/insts_256x3072x768.txt \
+	build/insts_768x256x3072.txt \
+	build/insts_3072x256x768.txt \
+	build/insts_768x256x768.txt \
+	build/insts_256x2304x768.txt \
+	build/insts_2304x256x768.txt
+
+${kernel_target}: kernel.cc await_rtp.cc zero.cc
+	mkdir -p ${@D}
+	cd ${@D} && xchesscc_wrapper ${CHESSCCWRAP2_FLAGS} -DDIM_M=$m -DDIM_K=$k -DDIM_N=$n -DBIT_WIDTH=8 -c ${<:%=../%} -o ${@F}
+
+${xclbin_target}: build/aie_256x768x2304.mlir ${kernel_target} 
+	mkdir -p ${@D}
+	cd ${@D} && aiecc.py -v --aie-generate-cdo --no-compile-host --xclbin-name=${@F}  $(<:%=../%)
+
+build/aie_256x768x2304.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 256 -K 768 -N 2304 $(aieargs) > $@
+build/aie_256x768x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 256 -K 768 -N 768 $(aieargs) > $@
+build/aie_256x768x3072.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 256 -K 768 -N 3072 $(aieargs) > $@
+build/aie_256x3072x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 256 -K 3072 -N 768 $(aieargs) > $@
+build/aie_768x256x3072.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 768 -K 256 -N 3072 $(aieargs) > $@
+build/aie_3072x256x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 3072 -K 256 -N 768 $(aieargs) > $@
+build/aie_768x256x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 768 -K 256 -N 768 $(aieargs) > $@
+build/aie_256x2304x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 256 -K 2304 -N 768 $(aieargs) > $@
+build/aie_2304x256x768.mlir: aie2.py
+	mkdir -p ${@D}
+	python3 $< -M 2304 -K 256 -N 768 $(aieargs) > $@
+
+build/insts_%.txt: build/aie_%.mlir 
+	cd ${@D} && aiecc.py -v --aie-only-generate-npu --npu-insts-name=${@:build/%=%} $(<:%=../%)
+
+xclbin_sign=${XILINX_XRT}/amdxdna/setup_xclbin_firmware.sh 
+.PHONY: sign
+sign: ${xclbin_target}
+	${xclbin_sign} -dev Phoenix -xclbin $<
+
+.PHONY: clean
+clean:
+	@rm -r build
+
+test: test.cpp 
+	${CXX} ${CXXFLAGS} $^ -o $@ ${LDFLAGS} ${LDLIBS}
+
+.PHONY: run
+run: all
+	./test