add 1d mesh node and refactor 2d mesh node

+ Seems to decrease performance a little bit, presumably due to
  extra register copy, but simplifies code a lot and opens up more
  genericism so I'll consider it an upgrade for now. Copying packets
  around is rather slow though, might in the future move to some
  kind of pointer based packet handling

3 files changed, 173 insertions, 0 deletions

diff --git a/tests/simple_mesh1d/sim.c b/tests/simple_mesh1d/sim.c
new file mode 100644
index 0000000..c5f7290
--- /dev/null
+++ b/tests/simple_mesh1d/sim.c
@@ -0,0 +1,88 @@
+#include <assert.h>
+
+#include <gran/root.h>
+#include <gran/mem/simple_mem.h>
+#include <gran/bus/simple_bus.h>
+#include <gran/uart/simple_uart.h>
+#include <gran/mesh/node1d.h>
+#include <gran/cpu/riscv/simple_riscv64.h>
+
+#include "../build/tests/simple_mesh1d/test.inc"
+
+static stat build_node1d(struct clock_domain *clk, uint16_t x, uint16_t y)
+{
+	struct component **mesh = calloc(x + 1, sizeof(struct component *));
+	assert(mesh);
+
+	for (int i = 1; i < x + 1; ++i) {
+		/* 4 CPUs + 1 mem = 5 elems in total */
+		struct component *node = create_mesh_node1d(i, y + 1);
+		clock_domain_add(clk, node);
+		mesh[i] = node;
+
+		for (int j = 0; j < y; ++j) {
+			struct component *imem = create_simple_mem(4096);
+			init_simple_mem(imem, 0,
+					build_tests_simple_mesh1d_test_bin_len,
+					build_tests_simple_mesh1d_test_bin);
+
+			uint64_t rcv = mesh1d_addr(i, j, 0);
+			struct component *rv64 = create_simple_riscv64(rcv, 0, imem, node);
+			simple_riscv64_set_reg(rv64, 10, i); /* a0 */
+			simple_riscv64_set_reg(rv64, 11, j); /* a1 */
+			simple_riscv64_set_reg(rv64, 12, x); /* a2 */
+			simple_riscv64_set_reg(rv64, 13, y); /* a3 */
+
+			clock_domain_add(clk, rv64);
+			clock_domain_add(clk, imem);
+
+			mesh_node1d_connect(node, rv64, j);
+		}
+
+		struct component *dmem = create_simple_mem(4096);
+		clock_domain_add(clk, dmem);
+		mesh_node1d_connect(node, dmem, 4);
+	}
+
+	/* extra I/O node (kind of?) */
+	struct component *node = create_mesh_node1d(0, 2);
+	clock_domain_add(clk, node);
+	mesh[0] = node;
+
+	struct component *uart = create_simple_uart();
+	clock_domain_add(clk, uart);
+	mesh_node1d_connect(node, uart, 0);
+
+	struct component *dmem = create_simple_mem(4096);
+	clock_domain_add(clk, dmem);
+	mesh_node1d_connect(node, dmem, 1);
+
+	for (int i = 0; i < x + 1; ++i) {
+		if (i - 1 >= 0)
+			mesh_node1d_connect_right(mesh[i], mesh[i - 1]);
+
+		if (i + 1 < x + 1)
+			mesh_node1d_connect_left(mesh[i], mesh[i + 1]);
+	}
+
+	free(mesh);
+	return OK;
+}
+
+int main()
+{
+	struct clock_domain *clk = create_clock_domain(NS(1));
+
+	/* one cluster with four cores + mem for each core (should
+	 * instructions be fetched from cluster local mem?) */
+	stat r = build_node1d(clk, 8, 8);
+	assert(r == OK);
+
+	struct gran_root *root = create_root();
+	root_add_clock(root, clk);
+
+	r = root_run(root);
+	assert(r == OK);
+
+	destroy_root(root);
+}
diff --git a/tests/simple_mesh1d/source.mk b/tests/simple_mesh1d/source.mk
new file mode 100644
index 0000000..5013c6e
--- /dev/null
+++ b/tests/simple_mesh1d/source.mk
@@ -0,0 +1,19 @@
+MESH1D_TEST_OBJ	!= ./scripts/gen-deps --sources "tests/simple_mesh1d/sim.c"
+TEST_PROGS	+= build/tests/simple_mesh1d/sim
+
+build/tests/simple_mesh1d/test.inc: tests/simple_mesh1d/test.c
+	riscv64-unknown-elf-gcc -O2 -Wall -Wextra -ffreestanding -nostdlib \
+		-march=rv64i -mabi=lp64 \
+		-fno-delete-null-pointer-checks \
+		-o build/tests/simple_mesh1d/test \
+		tests/simple_mesh1d/test.c
+	riscv64-unknown-elf-objcopy -Obinary \
+		build/tests/simple_mesh1d/test \
+		build/tests/simple_mesh1d/test.bin
+	xxd -i build/tests/simple_mesh1d/test.bin \
+		> build/tests/simple_mesh1d/test.inc
+
+build/tests/simple_mesh1d/sim.o: build/tests/simple_mesh1d/test.inc
+
+build/tests/simple_mesh1d/sim: $(MESH1D_TEST_OBJ) $(OBJS)
+	$(COMPILE) $(MESH1D_TEST_OBJ) $(OBJS) -o $@
diff --git a/tests/simple_mesh1d/test.c b/tests/simple_mesh1d/test.c
new file mode 100644
index 0000000..b3de347
--- /dev/null
+++ b/tests/simple_mesh1d/test.c
@@ -0,0 +1,66 @@
+#include <stdint.h>
+
+__attribute__((always_inline))
+static inline uint64_t extreme_numa_addr(uint16_t cluster, uint16_t elem, uint32_t off)
+{
+	return ((uint64_t)cluster << 48) | ((uint64_t)elem << 32) | off;
+}
+
+__attribute__((always_inline))
+static inline void print_int8(volatile char *uart, unsigned x)
+{
+	char lo_nibble = (x >> 0) & 0xf;
+	char hi_nibble = (x >> 4) & 0xf;
+
+	*uart = hi_nibble < 10 ? hi_nibble + '0' : (hi_nibble - 10) + 'a';
+	*uart = lo_nibble < 10 ? lo_nibble + '0' : (lo_nibble - 10) + 'a';
+}
+
+__attribute__((always_inline))
+static inline void print_addr(volatile char *uart, unsigned x, unsigned y)
+{
+	*uart = '(';
+	print_int8(uart, x);
+	*uart = ',';
+	*uart = ' ';
+	print_int8(uart, y);
+	*uart = ')';
+	*uart = '\n';
+}
+
+__attribute__((always_inline))
+static inline uint64_t wrap(unsigned x, unsigned X)
+{
+	return x + 1 >= X ? 0 : x + 1;
+}
+
+__attribute__((always_inline))
+static inline uint64_t next_idx(unsigned x, unsigned y, unsigned X, unsigned Y)
+{
+	unsigned yi = wrap(y, Y);
+	unsigned xi = yi < y ? wrap(x, X) : x;
+
+	return extreme_numa_addr(xi, yi, 0);
+}
+
+void _start(unsigned x, unsigned y, unsigned X, unsigned Y)
+{
+	volatile char *uart = (char *)extreme_numa_addr(0, 0, 0);
+	volatile uint64_t *control = (uint64_t *)extreme_numa_addr(0, 1, 0);
+
+	if (x == 1 && y == 0) {
+		goto do_work;
+	} else {
+		while (*control != extreme_numa_addr(x, y, 0)) {}
+	}
+
+do_work:
+	print_addr(uart, x, y);
+	*control = next_idx(x, y, X, Y);
+
+	if (x == X - 1 && y == Y - 1)
+		asm("ebreak");
+
+	/* otherwise just loop */
+	while (1) {}
+}