Added 1 in 1 out support for new simplified testbench and updated additional designs

aie_kernels/aie2/reduce_add.cc

@@ -6,16 +6,19 @@
 
 static void _reduce_add_scalar(int32_t *restrict in, int32_t *restrict out,
                                const int32_t input_size) {
+  event0();
   int32_t running_total = 0;
   for (int32_t i = 0; i < input_size; i++) {
     running_total = running_total + in[i];
   }
   *out = running_total;
+  event1();
   return;
 }
 
 static void _reduce_add_vector(int32_t *restrict in, int32_t *restrict out,
                                const int32_t input_size) {
+  event0();
   v16int32 zero = broadcast_to_v16int32((int32_t)0);
   const int32_t vector_size = 16;
   v16int32 after_vector;
@@ -37,6 +40,7 @@ static void _reduce_add_vector(int32_t *restrict in, int32_t *restrict out,
   v16int32 fifth = add(fourth, fourth_shift);
   int32_t last = extract_elem(fifth, 0U);
   *(int32_t *)out = last;
+  event1();
   return;
 }
 

aie_kernels/aie2/reduce_max.cc

@@ -8,6 +8,7 @@
 void _reduce_max_vector(int32_t *restrict in, int32_t *restrict out,
                         const int32_t input_size) {
 
+  event0();
   v16int32 tiny = broadcast_to_v16int32((int32_t)INT32_MIN);
   const int32_t vector_size = 16;
   v16int32 after_vector;
@@ -29,17 +30,20 @@ void _reduce_max_vector(int32_t *restrict in, int32_t *restrict out,
   v16int32 fifth = max(fourth, fourth_shift);
   int32_t last = extract_elem(fifth, 0U);
   *(int32_t *)out = last;
+  event1();
   return;
 }
 
 void _reduce_max_scalar(int32_t *restrict in, int32_t *restrict out,
                         const int32_t input_size) {
+  event0();
   int32_t running_max = (int32_t)INT32_MIN;
   for (int32_t i = 0; i < input_size; i++) {
     if (in[i] > running_max)
       running_max = in[i];
   }
   *(int32_t *)out = running_max;
+  event1();
 
   return;
 }

aie_kernels/aie2/reduce_min.cc

@@ -8,6 +8,7 @@
 void _reduce_min_vector(int32_t *restrict in, int32_t *restrict out,
                         const int32_t input_size) {
 
+  event0();
   v16int32 massive = broadcast_to_v16int32((int32_t)INT32_MAX);
   const int32_t vector_size = 16;
   v16int32 after_vector;
@@ -29,17 +30,20 @@ void _reduce_min_vector(int32_t *restrict in, int32_t *restrict out,
   v16int32 fifth = min(fourth, fourth_shift);
   int32_t last = extract_elem(fifth, 0U);
   *(int32_t *)out = last;
+  event1();
   return;
 }
 
 void _reduce_min_scalar(int32_t *restrict in, int32_t *restrict out,
                         const int32_t input_size) {
+  event0();
   int32_t running_min = (int32_t)INT32_MAX;
   for (int32_t i = 0; i < input_size; i++) {
     if (in[i] < running_min)
       running_min = in[i];
   }
   *(int32_t *)out = running_min;
+  event1();
 
   return;
 }

programming_examples/basic/passthrough_kernel/CMakeLists.txt

@@ -25,7 +25,8 @@ else()
     set(XRT_LIB_DIR C:/Technical/xrtNPUfromDLL CACHE STRING "Path to xrt_coreutil.lib")
 endif ()
 
-set(PASSTHROUGH_SIZE 4096 CACHE STRING "size")
+set(IN1_SIZE 4096 CACHE STRING "in1 buffer size")
+set(OUT_SIZE 4096 CACHE STRING "out buffer size")
 set(TARGET_NAME test CACHE STRING "Target to be built")
 
 SET (ProjectName ${TARGET_NAME})
@@ -46,8 +47,9 @@ add_executable(${currentTarget}
 )
 
 target_compile_definitions(${currentTarget} PUBLIC 
-        PASSTHROUGH_SIZE=${PASSTHROUGH_SIZE}
-        DISABLE_ABI_CHECK=1 
+    IN1_SIZE=${IN1_SIZE}
+    OUT_SIZE=${OUT_SIZE}
+    DISABLE_ABI_CHECK=1 
         )
 
 target_include_directories (${currentTarget} PUBLIC 

programming_examples/basic/passthrough_kernel/Makefile

@@ -11,14 +11,16 @@
 srcdir := $(shell dirname $(realpath $(firstword $(MAKEFILE_LIST))))
 
 include ${srcdir}/../../makefile-common
+VPATH := ${srcdir}/../../../aie_kernels/generic
 
 devicename ?= $(if $(filter 1,$(NPU2)),npu2,npu)
 targetname = passthrough_kernel
-VPATH := ${srcdir}/../../../aie_kernels/generic
-data_size = 4096
+in1_size = 4096 # in bytes
+out_size = 4096 # in bytes, should always be equal to in1_size
 trace_size = 8192
-PASSTHROUGH_SIZE = ${data_size}
+CHESS ?= false
 
+data_size = in1_size
 aie_py_src=${targetname}.py
 use_alt?=0
 
@@ -32,13 +34,13 @@ all: build/final_${data_size}.xclbin
 
 build/aie2_lineBased_8b_${data_size}.mlir: ${srcdir}/${aie_py_src}
 	mkdir -p ${@D}
-	python3 $< ${devicename} ${data_size} 0 > $@
+	python3 $< ${devicename} ${in1_size} ${out_size} 0 > $@
 
 build/aie2_trace_lineBased_8b_${data_size}.mlir: ${srcdir}/${aie_py_src}
 	mkdir -p ${@D}
-	python3 $< ${devicename} ${data_size} ${trace_size} > $@
+	python3 $< ${devicename} ${in1_size} ${out_size} ${trace_size} > $@
 
-build/passThrough.cc.o: passThrough.cc
+build/%.cc.o: %.cc
 	mkdir -p ${@D}
 ifeq ($(devicename),npu)
 	cd ${@D} && ${PEANO_INSTALL_DIR}/bin/clang++ ${PEANOWRAP2_FLAGS} -DBIT_WIDTH=8 -c $< -o ${@F}
@@ -64,7 +66,7 @@ build/final_trace_${data_size}.xclbin: build/aie2_trace_lineBased_8b_${data_size
 ${targetname}_${data_size}.exe: ${srcdir}/test.cpp
 	rm -rf _build
 	mkdir -p _build
-	cd _build && ${powershell} cmake ${srcdir} -DTARGET_NAME=${targetname} -DPASSTHROUGH_SIZE=${data_size}
+	cd _build && ${powershell} cmake ${srcdir} -DTARGET_NAME=${targetname}  -DIN1_SIZE=${in1_size} -DOUT_SIZE=${out_size}
 	cd _build && ${powershell} cmake --build . --config Release
 ifeq "${powershell}" "powershell.exe"
 	cp _build/${targetname}.exe $@
@@ -75,24 +77,21 @@ endif
 run: ${targetname}_${data_size}.exe build/final_${data_size}.xclbin build/insts_${data_size}.txt
 	${powershell} ./$< -x build/final_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE 
 
-#run-g: ${targetname}.exe build/final_${data_size}.xclbin build/insts.txt
-#	${powershell} ./$< -x build/final_${data_size}.xclbin -i build/insts.txt -k MLIR_AIE -t ${trace_size}
-
 run_py: build/final_${data_size}.xclbin build/insts_${data_size}.txt
-	${powershell} python3 ${srcdir}/test.py -s ${data_size} -x build/final_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE
+	${powershell} python3 ${srcdir}/test.py -x build/final_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE -i1s ${in1_size} -os ${out_size}
 
 trace: ${targetname}_${data_size}.exe build/final_trace_${data_size}.xclbin build/insts_${data_size}.txt 
 	${powershell} ./$< -x build/final_trace_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE -t ${trace_size}
 	${srcdir}/../../utils/parse_trace.py --filename trace.txt --mlir build/aie2_trace_lineBased_8b_${data_size}.mlir --colshift 1 > trace_passthrough_kernel.json
 	${srcdir}/../../utils/get_trace_summary.py --filename trace_passthrough_kernel.json
 
 trace_py: build/final_trace_${data_size}.xclbin build/insts_${data_size}.txt
-	${powershell} python3 ${srcdir}/test.py -x build/final_trace_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE -t ${trace_size} -s ${data_size}
+	${powershell} python3 ${srcdir}/test.py -x build/final_trace_${data_size}.xclbin -i build/insts_${data_size}.txt -k MLIR_AIE -t ${trace_size} -i1s ${in1_size} -os ${out_size}
 	${srcdir}/../../utils/parse_trace.py --filename trace.txt --mlir build/aie2_trace_lineBased_8b_${data_size}.mlir --colshift 1 > trace_passthrough_kernel.json
 	${srcdir}/../../utils/get_trace_summary.py --filename trace_passthrough_kernel.json
 
 clean_trace:
 	rm -rf tmpTrace trace.txt parse*json trace*json
 
-clean:
+clean: clean_trace
 	rm -rf build _build ${targetname}*.exe

programming_examples/basic/passthrough_kernel/passthrough_kernel.py

@@ -20,18 +20,21 @@
         dev = NPU2()
     else:
         raise ValueError("[ERROR] Device name {} is unknown".format(sys.argv[1]))
-    vector_size = int(sys.argv[2])
-    if vector_size % 64 != 0 or vector_size < 512:
-        print("Vector size must be a multiple of 64 and greater than or equal to 512")
+    in1_size = int(sys.argv[2])
+    if in1_size % 64 != 0 or in1_size < 512:
+        print(
+            "In1 buffer size must be a multiple of 64 and greater than or equal to 512"
+        )
         raise ValueError
-    trace_size = 0 if (len(sys.argv) != 4) else int(sys.argv[3])
+    out_size = int(sys.argv[3])
+    trace_size = 0 if (len(sys.argv) != 5) else int(sys.argv[4])
 except ValueError:
     print("Argument has inappropriate value")
 
 # Define tensor types
-line_size = vector_size // 4
+line_size = in1_size // 4
 line_type = np.ndarray[(line_size,), np.dtype[np.uint8]]
-vector_type = np.ndarray[(vector_size,), np.dtype[np.uint8]]
+vector_type = np.ndarray[(in1_size,), np.dtype[np.uint8]]
 
 # Dataflow with ObjectFifos
 of_in = ObjectFifo(line_type, name="in")

programming_examples/basic/passthrough_kernel/passthrough_kernel_alt.py

@@ -16,8 +16,8 @@
 import aie.utils.trace as trace_utils
 
 
-def passthroughKernel(dev, vector_size, trace_size):
-    N = vector_size
+def passthrough_kernel(dev, in1_size, out_size, trace_size):
+    N = in1_size
     lineWidthInBytes = N // 4  # chop input in 4 sub-tensors
 
     @device(dev)
@@ -36,7 +36,7 @@ def device_body():
         ComputeTile2 = tile(0, 2)
 
         # Set up a packet-switched flow from core to shim for tracing information
-        tiles_to_trace = [ComputeTile2]
+        tiles_to_trace = [ComputeTile2, ShimTile]
         if trace_size > 0:
             trace_utils.configure_packet_tracing_flow(tiles_to_trace, ShimTile)
 
@@ -56,17 +56,13 @@ def core_body():
                 of_in.release(ObjectFifoPort.Consume, 1)
                 of_out.release(ObjectFifoPort.Produce, 1)
 
-        #    print(ctx.module.operation.verify())
-
         @runtime_sequence(vector_ty, vector_ty, vector_ty)
         def sequence(inTensor, outTensor, notUsed):
             if trace_size > 0:
                 trace_utils.configure_packet_tracing_aie2(
                     tiles_to_trace=tiles_to_trace,
                     shim=ShimTile,
                     trace_size=trace_size,
-                    trace_offset=N,
-                    ddr_id=1,
                 )
 
             in_task = shim_dma_single_bd_task(
@@ -82,21 +78,30 @@ def sequence(inTensor, outTensor, notUsed):
             trace_utils.gen_trace_done_aie2(ShimTile)
 
 
-try:
-    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[1]))
-    vector_size = int(sys.argv[2])
-    if vector_size % 64 != 0 or vector_size < 512:
-        print("Vector size must be a multiple of 64 and greater than or equal to 512")
-        raise ValueError
-    trace_size = 0 if (len(sys.argv) != 4) else int(sys.argv[3])
-except ValueError:
-    print("Argument has inappropriate value")
+if len(sys.argv) < 4:
+    raise ValueError("[ERROR] Need at least 4 arguments (dev, in1_size, out_size)")
+
+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[1]))
+in1_size = int(sys.argv[2])
+if in1_size % 64 != 0 or in1_size < 512:
+    raise ValueError(
+        "[ERROR] In1 buffer size ("
+        + str(in1_size)
+        + ") must be a multiple of 64 and greater than or equal to 512"
+    )
+out_size = int(sys.argv[3])
+trace_size = 0 if (len(sys.argv) != 5) else int(sys.argv[4])
+
 with mlir_mod_ctx() as ctx:
-    passthroughKernel(dev, vector_size, trace_size)
-    print(ctx.module)
+    passthrough_kernel(dev, in1_size, out_size, trace_size)
+    res = ctx.module.operation.verify()
+    if res == True:
+        print(ctx.module)
+    else:
+        print(res)

programming_examples/basic/passthrough_kernel/run_makefile_alt.lit

@@ -8,4 +8,7 @@
 // RUN: make -f %S/Makefile clean
 // RUN: env use_alt=1 make -f %S/Makefile 
 // RUN: %run_on_npu make -f %S/Makefile run_py
+// RUN: make -f %S/Makefile clean
+// RUN: env CHESS=false use_alt=1 %run_on_npu make -f %S/Makefile trace 
+// RUN: env CHESS=false use_alt=1 %run_on_npu make -f %S/Makefile trace_py