rlc: multi-user RLC kernel, bootrom toolchain fix, tile write-enable timing

Makefile

@@ -153,16 +153,16 @@ $(SNRT_BOOTINFO_H): $(SCRIPTS_DIR)/generate_bootdata.py $(HJSON_OUT) $(BOOTROM_D
 
 $(BOOTROM_DIR)/bootrom.elf $(BOOTROM_DIR)/bootrom.dump $(BOOTROM_DIR)/bootrom.bin: \
   $(BOOTROM_DIR)/bootrom.S $(BOOTROM_DIR)/bootdata_bootrom.cc $(BOOTROM_DIR)/bootrom.ld $(SNRT_BOOTINFO_H)
-	riscv -riscv64-gcc-9.5.0 riscv64-unknown-elf-gcc \
+	$(GCC_INSTALL_DIR)/bin/riscv32-unknown-elf-gcc \
 		-mabi=ilp32 -march=rv32imaf -static -nostartfiles \
 		-T$(BOOTROM_DIR)/bootrom.ld \
 		$(BOOTROM_DIR)/bootrom.S \
 		$(BOOTROM_DIR)/bootdata_bootrom.cc \
 		-I$(SPATZ_DIR)/hw/ip/snitch_test/src \
 		-I$(SOFTWARE_DIR)/snRuntime/include \
 		-o $(BOOTROM_DIR)/bootrom.elf
-	riscv -riscv64-gcc-9.5.0 riscv64-unknown-elf-objdump -D $(BOOTROM_DIR)/bootrom.elf > $(BOOTROM_DIR)/bootrom.dump
-	riscv -riscv64-gcc-9.5.0 riscv64-unknown-elf-objcopy -O binary $(BOOTROM_DIR)/bootrom.elf $(BOOTROM_DIR)/bootrom.bin
+	$(GCC_INSTALL_DIR)/bin/riscv32-unknown-elf-objdump -D $(BOOTROM_DIR)/bootrom.elf > $(BOOTROM_DIR)/bootrom.dump
+	$(GCC_INSTALL_DIR)/bin/riscv32-unknown-elf-objcopy -O binary $(BOOTROM_DIR)/bootrom.elf $(BOOTROM_DIR)/bootrom.bin
 
 $(BOOTROM_DIR)/bootrom.sv: $(BOOTROM_DIR)/bootrom.bin $(BOOTROM_DIR)/bootdata.cc
 	${PYTHON} $(SCRIPTS_DIR)/generate_bootrom.py \

hardware/src/cachepool_tile.sv

@@ -1342,8 +1342,14 @@ module cachepool_tile
             automatic logic sel_found;
             automatic int unsigned sel_part_idx;
 
-            bank_req[col][bank_sel] = 1'b0;
-            bank_we[col][bank_sel] = 1'b0;
+            // Timing (T1.1): bank_we/bank_req are value-identical to direct OR-reduces
+            // of the part bits (the arbiter always selects a write-part when any
+            // write exists, else a read-part), so compute them directly off the
+            // critical path tail (data-bank we -> meta-write clock-gate) instead
+            // of riding the priority cascade + 4:1 sel_part_idx mux.  addr/wdata/
+            // be still use the mux.  Matches gen_unfolded_data_banks (1303-1304).
+            bank_we[col][bank_sel]  = |part_we[col][bank_sel];
+            bank_req[col][bank_sel] = (|part_we[col][bank_sel]) | (|part_req[col][bank_sel]);
             bank_addr[col][bank_sel] = '0;
             bank_wdata[col][bank_sel] = '0;
             bank_be[col][bank_sel] = '0;
@@ -1366,8 +1372,6 @@ module cachepool_tile
             end
 
             if (sel_found) begin
-              bank_req[col][bank_sel] = 1'b1;
-              bank_we[col][bank_sel] = part_we[col][bank_sel][sel_part_idx];
               bank_addr[col][bank_sel] = part_addr[col][bank_sel][sel_part_idx];
               bank_wdata[col][bank_sel] = part_wdata[col][bank_sel][sel_part_idx];
               bank_be[col][bank_sel] = part_be[col][bank_sel][sel_part_idx];

software/tests/CMakeLists.txt

@@ -74,11 +74,9 @@ add_library(fmatmul-32b fmatmul-32b/kernel/fmatmul.c)
 
 
 ## RLC
-add_library(printf_lock multi_producer_single_consumer_double_linked_list/kernel/printf_lock.c)
-add_library(llist multi_producer_single_consumer_double_linked_list/kernel/llist.c)
-add_library(mm multi_producer_single_consumer_double_linked_list/kernel/mm.c)
-add_library(rlc multi_producer_single_consumer_double_linked_list/kernel/rlc.c)
-add_library(data_move_vec multi_producer_single_consumer_double_linked_list/kernel/data_move_vec.c)
+# (kernel sources are compiled into the test executable via main.c's textual
+#  #include, so no per-kernel libraries are needed here; rlc.c requires the
+#  per-variant DATAHEADER define, which a shared library target cannot get.)
 
 # Tests
 enable_testing()
@@ -107,6 +105,13 @@ add_spatz_test_zeroParam(cache-coverage-min cache-coverage-min/main.c)
 add_spatz_test_threeParam(multi_producer_single_consumer_double_linked_list multi_producer_single_consumer_double_linked_list/main.c 1 1350 300)
 add_spatz_test_threeParam(multi_producer_single_consumer_double_linked_list multi_producer_single_consumer_double_linked_list/main.c 1 1350 100)
 add_spatz_test_threeParam(multi_producer_single_consumer_double_linked_list multi_producer_single_consumer_double_linked_list/main.c 1 1350 10)
+# TC2 multi-user peak rate: 48 UEs, 800 B packets (data_48_800_*.h)
+add_spatz_test_threeParam(multi_producer_single_consumer_double_linked_list multi_producer_single_consumer_double_linked_list/main.c 48 800 300)
+add_spatz_test_threeParam(multi_producer_single_consumer_double_linked_list multi_producer_single_consumer_double_linked_list/main.c 48 800 1000)
+# TC2 with the payload self-check enabled (src vs tgt compare after the barrier)
+add_snitch_test(multi_producer_single_consumer_double_linked_list_M48_N800_K300_sc multi_producer_single_consumer_double_linked_list/main.c)
+target_link_libraries(test-${SNITCH_TEST_PREFIX}multi_producer_single_consumer_double_linked_list_M48_N800_K300_sc benchmark spin_lock mcs_lock ${SNITCH_RUNTIME})
+target_compile_definitions(test-${SNITCH_TEST_PREFIX}multi_producer_single_consumer_double_linked_list_M48_N800_K300_sc PUBLIC DATAHEADER="data/data_48_800_300.h" RLC_SELF_CHECK=1)
 
 ## Vector
 ### Floating-Point