Enable NPU2 for All Programming Examples

.github/workflows/buildAndTestRyzenAI.yml

@@ -102,9 +102,9 @@ jobs:
           pushd build
 
           # -j here to reduce the number of parallel chess jobs.
-          # -j6 for 32GB RAM, -j12 for 64GB RAM
+          # -j4 for 32GB RAM, -j12 for 64GB RAM
           if [ x"${{ matrix.runner_type }}" == x"amdhx370" ]; then
-            LIT_OPTS="-j6 $LIT_OPTS"
+            LIT_OPTS="-j4 $LIT_OPTS"
           else
             LIT_OPTS="-j12 $LIT_OPTS"
           fi
@@ -129,7 +129,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        runner_type: [ amd7940hs ]
+        runner_type: [ amd7940hs, amdhx370 ]
     steps:
       - uses: actions/checkout@v4
         with:
@@ -166,9 +166,9 @@ jobs:
           pushd build
 
           # -j here to reduce the number of parallel chess jobs.
-          # -j6 for 32GB RAM, -j12 for 64GB RAM
+          # -j4 for 32GB RAM, -j12 for 64GB RAM
           if [ x"${{ matrix.runner_type }}" == x"amdhx370" ]; then
-            LIT_OPTS="-j6 $LIT_OPTS"
+            LIT_OPTS="-j4 $LIT_OPTS"
           else
             LIT_OPTS="-j12 $LIT_OPTS"
           fi

programming_examples/ml/softmax/Makefile

@@ -14,6 +14,7 @@ include ${srcdir}/../../makefile-common
 
 VPATH := ${srcdir}/../../../aie_kernels/aie2
 
+device ?= npu
 targetname = softmax
 trace_size = 8192
 
@@ -31,40 +32,67 @@ build/dut.cc: ${srcdir}/bf16_softmax.mlir
 	cd ${@D} &&	aie-opt $< -affine-super-vectorize="virtual-vector-size=16 test-fastest-varying=0 vectorize-reductions=true" --convert-vector-to-aievec="aie-target=aie2" -lower-affine | aie-translate -aie2=true --aievec-to-cpp -o ${@F}
 
 build/dut.o: build/dut.cc
+ifeq ($(device),npu)
 	cd ${@D} &&	${PEANO_INSTALL_DIR}/bin/clang++ ${PEANOWRAP2_FLAGS} -I../../../../aie_runtime_lib/AIE2 -c ${<F} -o ${@F}
+else ifeq ($(device),npu2)
+	cd ${@D} && xchesscc_wrapper ${CHESSCCWRAP2P_FLAGS} -I../../../../aie_runtime_lib/AIE2P -c ${<F} -o ${@F}
+else
+	echo "Device type not supported"
+endif
 
 build/lut_based_ops.o: ../../../aie_runtime_lib/AIE2/lut_based_ops.cpp
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && ${PEANO_INSTALL_DIR}/bin/clang++ ${PEANOWRAP2_FLAGS} -I. -c $(<:%=../%) -o ${@F}
+else ifeq ($(device),npu2)
+	cd ${@D} && xchesscc_wrapper ${CHESSCCWRAP2P_FLAGS} -I. -c $(<:%=../%) -o ${@F}
+else
+	echo "Device type not supported"
+endif
 
 build/softmax.o: bf16_softmax.cc
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && ${PEANO_INSTALL_DIR}/bin/clang++ ${PEANOWRAP2_FLAGS} -I. -I../../../../aie_runtime_lib/AIE2 -c $< -o ${@F}
+else ifeq ($(device),npu2)
+	cd ${@D} && xchesscc_wrapper ${CHESSCCWRAP2P_FLAGS} -I. -I../../../../aie_runtime_lib/AIE2P -c $< -o ${@F}
+else
+	echo "Device type not supported"
+endif
 
 build/kernels.a: build/softmax.o 
 	ar rvs $@ $+
 
 build/aie.mlir: ${srcdir}/${aie_py_src}
 	mkdir -p ${@D}
-	python3 $< > $@
+	python3 $< ${device} > $@
 
 build/aie_trace.mlir: ${srcdir}/${aie_py_src}
 	mkdir -p ${@D}
-	python3 $< ${trace_size} > $@
+	python3 $< ${device} ${trace_size} > $@
 
 
 build/final.xclbin: build/aie.mlir build/kernels.a
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && aiecc.py --aie-generate-cdo --no-compile-host --xclbin-name=${@F} \
     	--no-xchesscc --no-xbridge --peano ${PEANO_INSTALL_DIR} \
 				--aie-generate-npu --npu-insts-name=insts.txt $(<:%=../%)
+else
+	cd ${@D} && aiecc.py --aie-generate-cdo --no-compile-host --xclbin-name=${@F} \
+				--aie-generate-npu --npu-insts-name=insts.txt $(<:%=../%)
+endif
 
 build/final_trace.xclbin: build/aie_trace.mlir build/kernels.a
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && aiecc.py --aie-generate-cdo --no-compile-host --xclbin-name=${@F} \
     	--no-xchesscc --no-xbridge --peano ${PEANO_INSTALL_DIR} \
 				--aie-generate-npu --npu-insts-name=insts.txt $(<:%=../%)
-
+else
+	cd ${@D} && aiecc.py --aie-generate-cdo --no-compile-host --xclbin-name=${@F} \
+				--aie-generate-npu --npu-insts-name=insts.txt $(<:%=../%)
+endif
 
 ${targetname}.exe: ${srcdir}/test.cpp
 	rm -rf _build

programming_examples/ml/softmax/softmax.py

@@ -10,11 +10,11 @@
 
 from aie.iron import Kernel, ObjectFifo, Program, Runtime, Worker
 from aie.iron.placers import SequentialPlacer
-from aie.iron.device import NPU1Col1
+from aie.iron.device import NPU1Col1, NPU2
 from aie.iron.controlflow import range_
 
 
-def vector_softmax(trace_size):
+def vector_softmax(dev, trace_size):
     N = 262144  # *1024
 
     # Tile sizes
@@ -88,13 +88,20 @@ def core_fn(of_in, of_out, softmax_kernel):
         rt.drain(outC.cons(), C, wait=True)
 
     # Place components (assign them resources on the device) and generate an MLIR module
-    return Program(NPU1Col1(), rt).resolve_program(SequentialPlacer())
+    return Program(dev, rt).resolve_program(SequentialPlacer())
 
 
 try:
-    trace_size = 0 if (len(sys.argv) != 2) else int(sys.argv[1])
+    device_name = str(sys.argv[1])
+    if device_name == "npu":
+        dev = NPU1Col1()
+    elif device_name == "npu2":
+        dev = NPU2()
+    else:
+        raise ValueError("[ERROR] Device name {} is unknown".format(sys.argv[2]))
+    trace_size = 0 if (len(sys.argv) != 3) else int(sys.argv[2])
 except ValueError:
     print("Argument is not an integer")
 
-module = vector_softmax(trace_size)
+module = vector_softmax(dev, trace_size)
 print(module)

programming_examples/ml/softmax/softmax_alt.py

@@ -17,7 +17,7 @@
 import aie.utils.trace as trace_utils
 
 
-def vector_softmax(trace_size):
+def vector_softmax(dev, trace_size):
 
     word_size_in = 2
     N = 262144  # *1024
@@ -31,7 +31,7 @@ def vector_softmax(trace_size):
     tiles = N_div_n // n_cores
     buffer_depth = 2
 
-    @device(AIEDevice.npu1_1col)
+    @device(dev)
     def device_body():
         tile_ty = np.ndarray[(n,), np.dtype[bfloat16]]
 
@@ -134,12 +134,19 @@ def sequence(A, C):
 
 
 try:
-    trace_size = 0 if (len(sys.argv) != 2) else int(sys.argv[1])
+    device_name = str(sys.argv[1])
+    if device_name == "npu":
+        dev = AIEDevice.npu1_1col
+    elif device_name == "npu2":
+        dev = AIEDevice.npu2
+    else:
+        raise ValueError("[ERROR] Device name {} is unknown".format(sys.argv[2]))
+    trace_size = 0 if (len(sys.argv) != 3) else int(sys.argv[2])
 except ValueError:
     print("Argument is not an integer")
 
 with mlir_mod_ctx() as ctx:
-    vector_softmax(trace_size)
+    vector_softmax(dev, trace_size)
     res = ctx.module.operation.verify()
     if res == True:
         print(ctx.module)

programming_examples/vision/color_detect/Makefile

@@ -12,6 +12,7 @@ include ${srcdir}/../../makefile-common
 
 VPATH := ${srcdir}/../../../aie_kernels/aie2
 
+device ?= npu
 COLORDETECT_WIDTH = 1920
 COLORDETECT_HEIGHT = 1080
 
@@ -36,21 +37,32 @@ mlir: build/aie2_lineBased_8b_${COLORDETECT_WIDTH}.mlir
 
 build/%.cc.o: %.cc
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && ${PEANO_INSTALL_DIR}/bin/clang++ ${PEANOWRAP2_FLAGS} -DBIT_WIDTH=8 -c $< -o ${@F}
+else ifeq ($(device),npu2)
+	cd ${@D} && xchesscc_wrapper ${CHESSCCWRAP2P_FLAGS} -DBIT_WIDTH=8 -c $< -o ${@F}
+else
+	echo "Device type not supported"
+endif
 
 build/combined_bitwiseOR_gray2rgba_bitwiseAND.a: build/bitwiseOR.cc.o build/gray2rgba.cc.o build/bitwiseAND.cc.o
 	mkdir -p ${@D}
 	ar rvs $@ $< $(word 2,$^) $(word 3,$^)
 
 build/aie2_lineBased_8b_${COLORDETECT_WIDTH}.mlir: ${srcdir}/${aie_py_src}
 	mkdir -p ${@D}
-	python3 $< ${COLORDETECT_WIDTH} ${COLORDETECT_HEIGHT} > $@
+	python3 $< ${device} ${COLORDETECT_WIDTH} ${COLORDETECT_HEIGHT} > $@
 
 build/final_${COLORDETECT_WIDTH}.xclbin: build/aie2_lineBased_8b_${COLORDETECT_WIDTH}.mlir build/rgba2hue.cc.o build/threshold.cc.o build/combined_bitwiseOR_gray2rgba_bitwiseAND.a
 	mkdir -p ${@D}
+ifeq ($(device),npu)
 	cd ${@D} && aiecc.py --aie-generate-cdo --aie-generate-npu --no-compile-host --alloc-scheme=basic-sequential \
 		--no-xchesscc --no-xbridge \
 		--xclbin-name=${@F} --npu-insts-name=insts.txt $(<:%=../%)
+else
+	cd ${@D} && aiecc.py --aie-generate-cdo --aie-generate-npu --no-compile-host --alloc-scheme=basic-sequential \
+		--xclbin-name=${@F} --npu-insts-name=insts.txt $(<:%=../%)
+endif
 
 ${targetname}.exe: ${srcdir}/test.cpp
 	mkdir -p ${@D}

programming_examples/vision/color_detect/color_detect.py

@@ -9,22 +9,15 @@
 
 from aie.iron import Kernel, ObjectFifo, Program, Runtime, Worker
 from aie.iron.placers import SequentialPlacer
-from aie.iron.device import NPU1Col1
+from aie.iron.device import NPU1Col1, NPU2
 
-width = 64
-height = 36
-if len(sys.argv) == 3:
-    width = int(sys.argv[1])
-    height = int(sys.argv[2])
 
-lineWidth = width
-lineWidthInBytes = width * 4
-tensorSize = width * height * 4  # 4 channels
+def color_detect(dev, width, height):
+    lineWidth = width
+    lineWidthInBytes = width * 4
+    tensorSize = width * height * 4  # 4 channels
 
-traceSize = 1024
-
-
-def color_detect():
+    traceSize = 1024
 
     # Define types
     line_bytes_ty = np.ndarray[(lineWidthInBytes,), np.dtype[np.uint8]]
@@ -212,8 +205,20 @@ def or_gray2rgba_and_fn(
         rt.drain(outOF_L2L3.cons(), O, wait=True)
 
     # Place components (assign them resources on the device) and generate an MLIR module
-    return Program(NPU1Col1(), rt).resolve_program(SequentialPlacer())
-
-
-module = color_detect()
+    return Program(dev, rt).resolve_program(SequentialPlacer())
+
+
+try:
+    device_name = str(sys.argv[1])
+    if device_name == "npu":
+        dev = NPU1Col1()
+    elif device_name == "npu2":
+        dev = NPU2()
+    else:
+        raise ValueError("[ERROR] Device name {} is unknown".format(sys.argv[1]))
+    width = 36 if (len(sys.argv) != 4) else int(sys.argv[2])
+    height = 64 if (len(sys.argv) != 4) else int(sys.argv[3])
+except ValueError:
+    print("Argument has inappropriate value")
+module = color_detect(dev, width, height)
 print(module)