sw: Add generic offload app

.gitlab-ci.yml

@@ -66,3 +66,5 @@ occamy-single-cluster-vsim:
     - cd target/sim
     - make CFG_OVERRIDE=cfg/single-cluster.hjson rtl
     - $QUESTA make bin/occamy_top.vsim
+    - $QUESTA ../../deps/snitch_cluster/util/sim/simulate.py sw/run.yaml
+      --simulator vsim --sim-bin bin/occamy_top.vsim

target/sim/Makefile

@@ -114,7 +114,9 @@ VLT_FLAGS += --prof-cfuncs # Enable code profiling
 # C testbench #
 ###############
 
-TB_CC_SOURCES  = $(SNITCH_TEST_DIR)/rtl_lib.cc
+TB_CC_SOURCES  = $(SNITCH_TEST_DIR)/bootrom.cc
+TB_CC_SOURCES += $(SNITCH_TEST_DIR)/ipc.cc
+TB_CC_SOURCES += $(SNITCH_TEST_DIR)/rtl_lib.cc
 TB_CC_SOURCES += $(SNITCH_TEST_DIR)/common_lib.cc
 TB_CC_SOURCES += $(TARGET_TEST_DIR)/bootdata.cc
 TB_CC_SOURCES += $(TARGET_TEST_DIR)/uartdpi/uartdpi.c
@@ -127,7 +129,9 @@ TB_CC_FLAGS +=-I$(SNITCH_TEST_DIR)
 TB_CC_FLAGS +=-I$(TARGET_TEST_DIR)/uartdpi
 
 # Required C sources for the verilator TB that are linked against the verilated model
-VLT_COBJ  = $(VLT_BUILDDIR)/tb/common_lib.o
+VLT_COBJ  = $(VLT_BUILDDIR)/tb/bootrom.o
+VLT_COBJ += $(VLT_BUILDDIR)/tb/ipc.o
+VLT_COBJ += $(VLT_BUILDDIR)/tb/common_lib.o
 VLT_COBJ += $(VLT_BUILDDIR)/tb/verilator_lib.o
 VLT_COBJ += $(VLT_BUILDDIR)/tb/tb_bin.o
 VLT_COBJ += $(VLT_BUILDDIR)/test/uartdpi/uartdpi.o
@@ -450,34 +454,29 @@ $(PLATFORM_HEADERS_DIR)/occamy_base_addr.h: $(CFG)
 # Traces #
 ##########
 
-# CVA6 trace should use binary written in logs/.rtlbinary
-# which is retrieved by the Makefrag in the BINARY variable,
-# but the Snitch traces use another binary which is expected to
-# be passed on the command-line, overriding the BINARY variable.
-# So we define a new variable to store the CVA6 binary
 CVA6_BINARY ?= $(shell cat $(LOGS_DIR)/.rtlbinary)
 
-$(LOGS_DIR)/hart_00000_perf.json: $(LOGS_DIR)/trace_hart_00000.txt $(ROOT)/util/trace/events.py
-	$(PYTHON) $(ROOT)/util/trace/events.py -f cva6 $< -o $@
+$(LOGS_DIR)/hart_00000_perf.json: $(LOGS_DIR)/trace_hart_00000.txt $(EVENTS_PY)
+	$(PYTHON) $(EVENTS_PY) -f cva6 $< -o $@
 
-$(LOGS_DIR)/trace_hart_00000.s: $(LOGS_DIR)/trace_hart_00000.txt ${ANNOTATE}
-	$(PYTHON) ${ANNOTATE} ${ANNOTATE_FLAGS} -o $@ $(CVA6_BINARY) $<
+$(LOGS_DIR)/trace_hart_00000.s: $(LOGS_DIR)/trace_hart_00000.txt $(ANNOTATE_PY)
+	$(PYTHON) $(ANNOTATE_PY) $(ANNOTATE_FLAGS) -o $@ $(CVA6_BINARY) $<
 
-$(LOGS_DIR)/trace_hart_00000.diff: $(LOGS_DIR)/trace_hart_00000.txt ${ANNOTATE}
-	$(PYTHON) ${ANNOTATE} ${ANNOTATE_FLAGS} -o $@ $(CVA6_BINARY) $< -d
+$(LOGS_DIR)/trace_hart_00000.diff: $(LOGS_DIR)/trace_hart_00000.txt $(ANNOTATE_PY)
+	$(PYTHON) $(ANNOTATE_PY) $(ANNOTATE_FLAGS) -o $@ $(CVA6_BINARY) $< -d
 
 traces: $(LOGS_DIR)/hart_00000_perf.json
 
 annotate: $(shell (echo $(LOGS_DIR)/trace_hart_00000.txt | sed 's/\.txt/\.s/') || echo "") \
           $(shell (echo $(LOGS_DIR)/trace_hart_00000.txt | sed 's/\.txt/\.diff/') || echo "")
 
 $(LOGS_DIR)/perf.csv: $(shell (ls $(LOGS_DIR)/trace_hart_*.dasm 2>/dev/null | sed 's/trace_hart/hart/' | sed 's/.dasm/_perf.json/')) \
-		  $(LOGS_DIR)/hart_00000_perf.json $(ROOT)/util/trace/perf_csv.py
-	$(PYTHON) $(ROOT)/util/trace/perf_csv.py -o $@ -i $(LOGS_DIR)/hart_*_perf.json
+		  $(LOGS_DIR)/hart_00000_perf.json $(PERF_CSV_PY)
+	$(PYTHON) $(PERF_CSV_PY) -o $@ -i $(LOGS_DIR)/hart_*_perf.json
 
 $(LOGS_DIR)/event.csv: $(shell (ls $(LOGS_DIR)/trace_hart_*.dasm 2>/dev/null | sed 's/trace_hart/hart/' | sed 's/.dasm/_perf.json/')) \
-		  $(LOGS_DIR)/hart_00000_perf.json $(ROOT)/util/trace/perf_csv.py
-	$(PYTHON) $(ROOT)/util/trace/perf_csv.py -o $@ -i $(LOGS_DIR)/hart_*_perf.json --filter tstart tend
+		  $(LOGS_DIR)/hart_00000_perf.json $(PERF_CSV_PY)
+	$(PYTHON) $(PERF_CSV_PY) -o $@ -i $(LOGS_DIR)/hart_*_perf.json --filter tstart tend
 
 #############
 # Verilator #
@@ -575,7 +574,7 @@ clean-work:
 	rm -rf work
 
 clean-bender:
-	rm -rf Bender.lock .bender/ work/
+	rm -rf $(ROOT)/Bender.lock $(ROOT)/.bender/ $(ROOT)/deps
 
 clean-logs:
 	rm -rf $(LOGS_DIR)/

target/sim/sw/device/Makefile

@@ -5,7 +5,7 @@
 # Luca Colagrande <colluca@iis.ee.ethz.ch>
 
 # Add user applications to APPS variable
-APPS = offload
+APPS = blas/axpy
 
 TARGET ?= all
 

target/sim/sw/device/apps/offload/Makefile → target/sim/sw/device/apps/blas/axpy/Makefile

@@ -4,8 +4,7 @@
 #
 # Luca Colagrande <colluca@iis.ee.ethz.ch>
 
-APP     = offload
-SRCS    = src/offload.c
-INCDIRS = ../../../host/apps/offload/src
+include ../../../../../../../deps/snitch_cluster/sw/blas/axpy/Makefile
+include ../../common.mk
 
-include ../common.mk
+$(DEP): $(DATA_H)

target/sim/sw/device/apps/common.mk

@@ -4,18 +4,21 @@
 #
 # Luca Colagrande <colluca@iis.ee.ethz.ch>
 
-include ../../toolchain.mk
+# Usage of absolute paths is required to externally include
+# this Makefile from multiple different locations
+MK_DIR := $(dir $(realpath $(lastword $(MAKEFILE_LIST))))
+include $(MK_DIR)/../toolchain.mk
 
 ###################
 # Build variables #
 ###################
 
 # Directories
 BUILDDIR    = $(abspath build)
-APPSDIR     = $(abspath ../)
-RUNTIME_DIR = $(abspath ../../runtime)
+APPSDIR     = $(abspath $(MK_DIR))
+RUNTIME_DIR = $(abspath $(MK_DIR)/../runtime)
 SNRT_DIR    = $(shell bender path snitch_cluster)/sw/snRuntime
-SW_DIR      = $(abspath ../../../)
+SW_DIR      = $(abspath $(MK_DIR)/../../)
 
 # Dependencies
 INCDIRS += $(RUNTIME_DIR)/src

target/sim/sw/device/runtime/src/occamy_memory.c

@@ -17,3 +17,5 @@ extern volatile uint32_t* snrt_cluster_clint_set_ptr();
 extern volatile uint32_t* snrt_cluster_clint_clr_ptr();
 
 extern uint32_t snrt_cluster_hw_barrier_addr();
+
+extern volatile uint32_t* snrt_zero_memory_ptr();

target/sim/sw/device/runtime/src/occamy_memory.h

@@ -32,3 +32,7 @@ inline volatile uint32_t* __attribute__((const)) snrt_cluster_clint_clr_ptr() {
 inline uint32_t __attribute__((const)) snrt_cluster_hw_barrier_addr() {
     return _snrt_cluster_hw_barrier;
 }
+
+inline volatile uint32_t* __attribute__((const)) snrt_zero_memory_ptr() {
+    return cluster_zero_memory_ptr(snrt_cluster_idx());
+}

target/sim/sw/device/runtime/src/occamy_start.c

@@ -10,9 +10,12 @@
 #define SNRT_CRT0_PRE_BARRIER
 #define SNRT_INVOKE_MAIN
 #define SNRT_CRT0_POST_BARRIER
+#define SNRT_CRT0_CALLBACK7
 
 static inline void snrt_crt0_callback3() {
     _snrt_cluster_hw_barrier = cluster_hw_barrier_addr(snrt_cluster_idx());
 }
 
+static inline void snrt_crt0_callback7() { return_to_cva6(SYNC_CLUSTERS); }
+
 #include "start.c"

target/sim/sw/device/runtime/src/snrt.c

@@ -13,4 +13,5 @@
 #include "occamy_memory.c"
 #include "occamy_start.c"
 #include "sync.c"
+#include "sys_dma.c"
 #include "team.c"

target/sim/sw/device/runtime/src/snrt.h

@@ -2,12 +2,16 @@
 // Licensed under the Apache License, Version 2.0, see LICENSE for details.
 // SPDX-License-Identifier: Apache-2.0
 
+#ifndef SNRT_H
+#define SNRT_H
+
 #include <stddef.h>
 #include <stdint.h>
 
 // Occamy specific definitions
 #include "occamy_defs.h"
 #include "occamy_memory_map.h"
+#include "sys_dma.h"
 
 // Forward declarations
 #include "alloc_decls.h"
@@ -30,3 +34,5 @@
 #include "ssr.h"
 #include "sync.h"
 #include "team.h"
+
+#endif  // SNRT_H