Improve test firmware, increase testbench memory size to 128kB

Signed-off-by: Clifford Wolf <clifford@clifford.at>

8 files changed, 132 insertions, 16 deletions

diff --git a/Makefile b/Makefile
index a005920..8a52aa0 100644
--- a/Makefile
+++ b/Makefile
@@ -4,7 +4,7 @@ RISCV_GNU_TOOLCHAIN_INSTALL_PREFIX = /opt/riscv32
 
 SHELL = bash
 TEST_OBJS = $(addsuffix .o,$(basename $(wildcard tests/*.S)))
-FIRMWARE_OBJS = firmware/start.o firmware/irq.o firmware/print.o firmware/sieve.o firmware/multest.o firmware/stats.o
+FIRMWARE_OBJS = firmware/start.o firmware/irq.o firmware/print.o firmware/hello.o firmware/sieve.o firmware/multest.o firmware/stats.o
 GCC_WARNS  = -Werror -Wall -Wextra -Wshadow -Wundef -Wpointer-arith -Wcast-qual -Wcast-align -Wwrite-strings
 GCC_WARNS += -Wredundant-decls -Wstrict-prototypes -Wmissing-prototypes -pedantic # -Wconversion
 TOOLCHAIN_PREFIX = $(RISCV_GNU_TOOLCHAIN_INSTALL_PREFIX)i/bin/riscv32-unknown-elf-
@@ -92,7 +92,7 @@ synth.v: picorv32.v scripts/yosys/synth_sim.ys
 	yosys -qv3 -l synth.log scripts/yosys/synth_sim.ys
 
 firmware/firmware.hex: firmware/firmware.bin firmware/makehex.py
-	python3 firmware/makehex.py $< 16384 > $@
+	python3 firmware/makehex.py $< 32768 > $@
 
 firmware/firmware.bin: firmware/firmware.elf
 	$(TOOLCHAIN_PREFIX)objcopy -O binary $< $@
diff --git a/firmware/firmware.h b/firmware/firmware.h
index 59b5c75..59d3f11 100644
--- a/firmware/firmware.h
+++ b/firmware/firmware.h
@@ -20,6 +20,9 @@ void print_str(const char *p);
 void print_dec(unsigned int val);
 void print_hex(unsigned int val, int digits);
 
+// hello.c
+void hello(void);
+
 // sieve.c
 void sieve(void);
 
@@ -28,6 +31,10 @@ uint32_t hard_mul(uint32_t a, uint32_t b);
 uint32_t hard_mulh(uint32_t a, uint32_t b);
 uint32_t hard_mulhsu(uint32_t a, uint32_t b);
 uint32_t hard_mulhu(uint32_t a, uint32_t b);
+uint32_t hard_div(uint32_t a, uint32_t b);
+uint32_t hard_divu(uint32_t a, uint32_t b);
+uint32_t hard_rem(uint32_t a, uint32_t b);
+uint32_t hard_remu(uint32_t a, uint32_t b);
 void multest(void);
 
 // stats.c
diff --git a/firmware/hello.c b/firmware/hello.c
new file mode 100644
index 0000000..b7e0b96
--- /dev/null
+++ b/firmware/hello.c
@@ -0,0 +1,14 @@
+// This is free and unencumbered software released into the public domain.
+//
+// Anyone is free to copy, modify, publish, use, compile, sell, or
+// distribute this software, either in source code form or as a compiled
+// binary, for any purpose, commercial or non-commercial, and by any
+// means.
+
+#include "firmware.h"
+
+void hello(void)
+{
+	print_str("hello world\n");
+}
+
diff --git a/firmware/multest.c b/firmware/multest.c
index 93e48df..1706400 100644
--- a/firmware/multest.c
+++ b/firmware/multest.c
@@ -17,15 +17,35 @@ static uint32_t xorshift32(void) {
 
 void multest(void)
 {
-	for (int i = 0; i < 10; i++)
+	for (int i = 0; i < 15; i++)
 	{
 		uint32_t a = xorshift32();
 		uint32_t b = xorshift32();
 
+		switch (i)
+		{
+		case 0:
+			a = 0x80000000;
+			b = 0xFFFFFFFF;
+			break;
+		case 1:
+			a = 0;
+			b = 0;
+			break;
+		case 2:
+			a |= 0x80000000;
+			b = 0;
+			break;
+		case 3:
+			a &= 0x7FFFFFFF;
+			b = 0;
+			break;
+		}
+
 		uint64_t au = a, bu = b;
 		int64_t as = (int32_t)a, bs = (int32_t)b;
 
-		print_str("input [");
+		print_str("input     [");
 		print_hex(as >> 32, 8);
 		print_str("] ");
 		print_hex(a, 8);
@@ -36,7 +56,7 @@ void multest(void)
 		print_chr('\n');
 
 		uint32_t h_mul, h_mulh, h_mulhsu, h_mulhu;
-		print_str("hard   ");
+		print_str("hard mul   ");
 
 		h_mul = hard_mul(a, b);
 		print_hex(h_mul, 8);
@@ -55,7 +75,7 @@ void multest(void)
 		print_chr('\n');
 
 		uint32_t s_mul, s_mulh, s_mulhsu, s_mulhu;
-		print_str("soft   ");
+		print_str("soft mul   ");
 
 		s_mul = a * b;
 		print_hex(s_mul, 8);
@@ -80,6 +100,52 @@ void multest(void)
 		}
 
 		print_str(" OK\n");
+
+		uint32_t h_div, h_divu, h_rem, h_remu;
+		print_str("hard div   ");
+
+		h_div = hard_div(a, b);
+		print_hex(h_div, 8);
+		print_str("  ");
+
+		h_divu = hard_divu(a, b);
+		print_hex(h_divu, 8);
+		print_str("  ");
+
+		h_rem = hard_rem(a, b);
+		print_hex(h_rem, 8);
+		print_str("  ");
+
+		h_remu = hard_remu(a, b);
+		print_hex(h_remu, 8);
+		print_chr('\n');
+
+		uint32_t s_div, s_divu, s_rem, s_remu;
+		print_str("soft div   ");
+
+		s_div = b ? as / bs : 0xffffffff;
+		print_hex(s_div, 8);
+		print_str("  ");
+
+		s_divu = b ? au / bu : 0xffffffff;
+		print_hex(s_divu, 8);
+		print_str("  ");
+
+		s_rem = b ? as % bs : a;
+		print_hex(s_rem, 8);
+		print_str("  ");
+
+		s_remu = b ? au % bu : a;
+		print_hex(s_remu, 8);
+		print_str("  ");
+
+		if (s_div != h_div || s_divu != h_divu || s_rem != h_rem || s_remu != h_remu) {
+			print_str("ERROR!\n");
+			__asm__ volatile ("ebreak");
+			return;
+		}
+
+		print_str(" OK\n");
 	}
 }
 
diff --git a/firmware/sections.lds b/firmware/sections.lds
index e1eb79f..645ce43 100644
--- a/firmware/sections.lds
+++ b/firmware/sections.lds
@@ -8,9 +8,10 @@ means.
 */
 
 MEMORY {
-	/* the memory in the testbench is 64k in size;
-	 * set LENGTH=48k and leave at least 16k for stack */
+	/* the memory in the testbench is 128k in size;
+	 * set LENGTH=96k and leave at least 32k for stack */
 	mem : ORIGIN = 0x00000000, LENGTH = 0x0000c000
+	mem : ORIGIN = 0x00000000, LENGTH = 0x00018000
 }
 
 SECTIONS {
diff --git a/firmware/start.S b/firmware/start.S
index 06d4744..d95f04c 100644
--- a/firmware/start.S
+++ b/firmware/start.S
@@ -6,6 +6,7 @@
 // means.
 
 #define ENABLE_QREGS
+#define ENABLE_HELLO
 #define ENABLE_RVTST
 #define ENABLE_SIEVE
 #define ENABLE_MULTST
@@ -24,12 +25,17 @@
 
 	.section .text
 	.global irq
+	.global hello
 	.global sieve
 	.global multest
 	.global hard_mul
 	.global hard_mulh
 	.global hard_mulhsu
 	.global hard_mulhu
+	.global hard_div
+	.global hard_divu
+	.global hard_rem
+	.global hard_remu
 	.global stats
 
 reset_vec:
@@ -372,6 +378,14 @@ start:
 	addi x30, zero, 0
 	addi x31, zero, 0
 
+#ifdef ENABLE_HELLO
+	/* set stack pointer */
+	lui sp,(128*1024)>>12
+
+	/* call hello C code */
+	jal ra,hello
+#endif
+
 	/* running tests from riscv-tests */
 
 #ifdef ENABLE_RVTST
@@ -443,7 +457,7 @@ start:
 	TEST(simple)
 
 	/* set stack pointer */
-	lui sp,(64*1024)>>12
+	lui sp,(128*1024)>>12
 
 	/* set gp and tp */
 	lui gp, %hi(0xdeadbeef)
@@ -505,3 +519,19 @@ hard_mulhu:
 	mulhu a0, a0, a1
 	ret
 
+hard_div:
+	div a0, a0, a1
+	ret
+
+hard_divu:
+	divu a0, a0, a1
+	ret
+
+hard_rem:
+	rem a0, a0, a1
+	ret
+
+hard_remu:
+	remu a0, a0, a1
+	ret
+
diff --git a/testbench.v b/testbench.v
index 6451252..eee2676 100644
--- a/testbench.v
+++ b/testbench.v
@@ -305,7 +305,7 @@ module axi4_memory #(
 
 	output reg        tests_passed
 );
-	reg [31:0]   memory [0:64*1024/4-1] /* verilator public */;
+	reg [31:0]   memory [0:128*1024/4-1] /* verilator public */;
 	reg verbose;
 	initial verbose = $test$plusargs("verbose") || VERBOSE;
 
@@ -383,7 +383,7 @@ module axi4_memory #(
 	task handle_axi_rvalid; begin
 		if (verbose)
 			$display("RD: ADDR=%08x DATA=%08x%s", latched_raddr, memory[latched_raddr >> 2], latched_rinsn ? " INSN" : "");
-		if (latched_raddr < 64*1024) begin
+		if (latched_raddr < 128*1024) begin
 			mem_axi_rdata <= memory[latched_raddr >> 2];
 			mem_axi_rvalid <= 1;
 			latched_raddr_en = 0;
@@ -396,7 +396,7 @@ module axi4_memory #(
 	task handle_axi_bvalid; begin
 		if (verbose)
 			$display("WR: ADDR=%08x DATA=%08x STRB=%04b", latched_waddr, latched_wdata, latched_wstrb);
-		if (latched_waddr < 64*1024) begin
+		if (latched_waddr < 128*1024) begin
 			if (latched_wstrb[0]) memory[latched_waddr >> 2][ 7: 0] <= latched_wdata[ 7: 0];
 			if (latched_wstrb[1]) memory[latched_waddr >> 2][15: 8] <= latched_wdata[15: 8];
 			if (latched_wstrb[2]) memory[latched_waddr >> 2][23:16] <= latched_wdata[23:16];
diff --git a/testbench_wb.v b/testbench_wb.v
index 789ec78..e78dbfa 100644
--- a/testbench_wb.v
+++ b/testbench_wb.v
@@ -56,8 +56,6 @@ endmodule
 `endif
 
 module picorv32_wrapper #(
-	parameter BOOTROM_MEMFILE = "",
-	parameter BOOTROM_MEMDEPTH = 16384 * 4,
 	parameter VERBOSE = 0
 ) (
 	input wb_clk,
@@ -89,7 +87,7 @@ module picorv32_wrapper #(
 	wire wb_s2m_ack;
 
 	wb_ram #(
-		.depth (16384 * 4),
+		.depth (128*1024),
 		.VERBOSE (VERBOSE)
 	) ram ( // Wishbone interface
 		.wb_clk_i(wb_clk),
@@ -266,7 +264,7 @@ module wb_ram #(
 	end
 
 	always @(posedge wb_clk_i) begin
-		if (waddr2 < 64 * 1024 / 4) begin
+		if (waddr2 < 128 * 1024 / 4) begin
 			if (we[0])
 				mem[waddr2][7:0] <= wb_dat_i[7:0];