From 00974c325bfdb24c6027a87ce391cfda0ab58fcb Mon Sep 17 00:00:00 2001 From: Mango0x45 Date: Thu, 9 Sep 2021 02:42:25 +0200 Subject: Genesis commit --- .gitignore | 6 + Makefile | 15 ++ UNLICENSE | 24 ++ extras/sizeof_struct_stat.c | 12 + macros.m4 | 2 + smallbrain.s | 538 ++++++++++++++++++++++++++++++++++++++++++++ 6 files changed, 597 insertions(+) create mode 100644 .gitignore create mode 100644 Makefile create mode 100644 UNLICENSE create mode 100644 extras/sizeof_struct_stat.c create mode 100644 macros.m4 create mode 100644 smallbrain.s diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..c430098 --- /dev/null +++ b/.gitignore @@ -0,0 +1,6 @@ +c-example/ +tests/ +*.bf +*.out +out.s +smallbrain diff --git a/Makefile b/Makefile new file mode 100644 index 0000000..475bf51 --- /dev/null +++ b/Makefile @@ -0,0 +1,15 @@ +target = smallbrain + +CC = cc +CFLAGS = -g -no-pie + +all: ${target} +${target}: ${target}.s + m4 macros.m4 $< >out.s + ${CC} ${CFLAGS} -o $@ out.s + +.PHONY: bench clean +bench: + hyperfine -S sh -w 200 -r 100000 "./smallbrain tests/towers-of-hanoi.bf" +clean: + rm -f ${target} out.s diff --git a/UNLICENSE b/UNLICENSE new file mode 100644 index 0000000..68a49da --- /dev/null +++ b/UNLICENSE @@ -0,0 +1,24 @@ +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. + +In jurisdictions that recognize copyright laws, the author or authors +of this software dedicate any and all copyright interest in the +software to the public domain. We make this dedication for the benefit +of the public at large and to the detriment of our heirs and +successors. We intend this dedication to be an overt act of +relinquishment in perpetuity of all present and future rights to this +software under copyright law. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR +OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR +OTHER DEALINGS IN THE SOFTWARE. + +For more information, please refer to diff --git a/extras/sizeof_struct_stat.c b/extras/sizeof_struct_stat.c new file mode 100644 index 0000000..67b6b9c --- /dev/null +++ b/extras/sizeof_struct_stat.c @@ -0,0 +1,12 @@ +#include + +#include +#include + +int +main(void) +{ + printf("Size of stat struct:\t%zu bytes\nOffset of st_size:\t%zu\n", sizeof(struct stat), + offsetof(struct stat, st_size)); + return 0; +} diff --git a/macros.m4 b/macros.m4 new file mode 100644 index 0000000..4f15035 --- /dev/null +++ b/macros.m4 @@ -0,0 +1,2 @@ +define(`COUNTER', 0) +define(`ENUM', `.equ $1, define(`COUNTER', incr(COUNTER))COUNTER') diff --git a/smallbrain.s b/smallbrain.s new file mode 100644 index 0000000..86e3114 --- /dev/null +++ b/smallbrain.s @@ -0,0 +1,538 @@ +.equ EOF, -1 +.equ O_RDONLY, 0 +.equ EXIT_FAILURE, 1 + +# Opcodes +ENUM(OP_ADD) +ENUM(OP_SUB) +ENUM(OP_RIGHT) +ENUM(OP_LEFT) +ENUM(OP_LOOP_START) +ENUM(OP_LOOP_END) +ENUM(OP_READ) +ENUM(OP_WRITE) +ENUM(OP_ZERO) +ENUM(OP_COPY) + +.global main + +.data +zero_pattern: + .asciz "[-]" +die_fmt: + .asciz "%s: %s\n" +usage_fmt: + .asciz "Usage: %s script\n" +func_open: + .asciz "open" +func_fstat: + .asciz "fstat" +func_read: + .asciz "read" +func_malloc: + .asciz "malloc" +memory: + .zero 30000 + +.bss +bytecode: + .quad 0 +program: + .quad 0 + +.text +# ================== +# Description: +# The entry point of the program. +# +# Args: +# %rdi: The number of command line arguments +# %rsi: An array of command line arguments ((%rsi) is the program name) +# +# Return: +# %rax: The programs exit code +# ================== +main: + # Make sure that the right number of arguments were passed + cmpl $2, %edi + jne usage + + movq 8(%rsi), %rdi # Move the specified script filename into %rdi + call read_file # Read the file + call compile # Compile into bytecode + #call execute # Execute the program + + # Return successfully + xorl %eax, %eax + ret + +# ================== +# Description: +# Read the specified file into a program buffer which the user must later free themselves. +# +# Args: +# %rdi: The scripts filename +# ================== +read_file: + pushq %rbp + movq %rsp, %rbp + + # Allocate space for local variables + # 4 bytes for the file descriptor + # 144 bytes for the struct stat (check `extras/sizeof_struct_stat.c` for numbers) + # 12 bytes of padding to align %rsp on a 16 byte boundary + subq $160, %rsp + + # Open the file in read-only mode with the open(2) syscall + movl $O_RDONLY, %esi + call open + + cmpl $-1, %eax # Check if open(1) returned -1 (it failed) + je open_die # If so, exit the program + movl %eax, -4(%rbp) # If not store the file descriptor + + movl %eax, %edi # Set the first argument to the file descriptor + leaq -160(%rbp), %rsi # Set the second argument to the address of the struct stat + call fstat # Call fstat(2) to populate the struct stat + + # Error check fstat(2) just like open(2) + cmpl $-1, %eax + je fstat_die + + # Allocate a buffer for the programs contents + movq -112(%rbp), %rdi # Put st_size (the filesize) in %rdi + incq %rdi # Make space for the NUL byte + call malloc # Allocate the memory + testq %rax, %rax # Check if %rax is NULL + je malloc_die # If it is then malloc failed and we exit + movq %rax, (program) # Store the address of the allocated memory in (program) + addq -112(%rbp), %rax # Point to the last element of the program array + movb $0, (%rax) # NULL terminate the program + + # Allocate a buffer for the bytecode + movq -112(%rbp), %rdi # Put st_size (the filesize) in %rdi + incq %rdi # Make space for the NUL byte + shlq $4, %rdi # Multiply the size of the buffer by 16 to make space for opcodes + call malloc # Allocate the memory + movq %rax, (bytecode) # Store the address of the allocated memory in (bytecode) + testq %rax, %rax # Check if %rax is NULL + je malloc_die # If it is then malloc failed and we exit + + # Read the program into the program buffer we allocated with read(2) + movq -112(%rbp), %rdx # Set the buffersize + movq (program), %rsi # Set the buffer + movl -4(%rbp), %edi # Set the file descriptor + call read + + # Error check read(2) just like open(2) + cmpl $-1, %eax + je read_die + + # Close the file, don't error check this + movl -4(%rbp), %eax + call close + + leave + ret + +# ================== +# Description: +# Compile the program into a bytecode which is an optimized version of the raw program. Each +# opcode is a "struct" where the higher 8 bytes are an opcode and the lower 8 are an option +# but of data. +# +# ================== +compile: + movq (program), %r15 # Store the address of the program pointer into %r15 + movq (bytecode), %r14 # Store the address of the bytecode pointer into %r14 +compile_loop: + # Load the current command into %rax + movq (%r15), %rax + + # Jump to a different label depending on which instruction we hit + cmpb $'+', %al + je compile_add + cmpb $'-', %al + je compile_sub + cmpb $'>', %al + je compile_right + cmpb $'<', %al + je compile_left + cmpb $'[', %al + je compile_loop_start + cmpb $']', %al + je compile_loop_end + cmpb $',', %al + je compile_read + cmpb $'.', %al + je compile_write + + # COMMENTS GET HERE + + incq %r15 # Increment the instruction pointer + subq $16, %r14 # Negate the effect of the later `addq` + jmp compile_out + +compile_add: + movq $OP_ADD, (%r14) # Specify the ADD opcode + movq $1, 8(%r14) # Write the count of '+'s to the data portion +compile_add_loop: + incq %r15 # Move to the next instruction + cmpb $'+', (%r15) # Check if there is another + + jne compile_out # If not, check for a newline + incq 8(%r14) # Increment the accumulator + jmp compile_add_loop # Loop again + +compile_sub: + movq $OP_SUB, (%r14) # Specify the SUB opcode + movq $1, 8(%r14) # Write the count of '-'s to the data portion +compile_sub_loop: + incq %r15 # Move to the next instruction + cmpb $'-', (%r15) # Check if there is another - + jne compile_out # If not, exit this loop + incq 8(%r14) # Increment the accumulator + jmp compile_sub_loop # Loop again + +compile_right: + movq $OP_RIGHT, (%r14) # Specify the RIGHT opcode + movq $1, 8(%r14) # Write the count of '>'s to the data portion +compile_right_loop: + incq %r15 # Move to the next instruction + cmpb $'>', (%r15) # Check if there is another > + jne compile_out # If not, exit this loop + incq 8(%r14) # Increment the accumulator + jmp compile_right_loop # Loop again + +compile_left: + movq $OP_LEFT, (%r14) # Specify the RIGHT opcode + movq $1, 8(%r14) # Write the count of '<'s to the data portion +compile_left_loop: + incq %r15 # Move to the next instruction + cmpb $'<', (%r15) # Check if there is another < + jne compile_out # If not, exit this loop + incq 8(%r14) # Increment the accumulator + jmp compile_left_loop # Loop again + +compile_loop_start: + # When we reach a '[' the first thing we want to do is check to see if it matches the + # pattern '[-]'. This pattern is one that sets a memory cell to 0, so we can optimize that. + movq %r15, %rdi # Compare the current position in the program string + movq $zero_pattern, %rsi # Compare it against the zero pattern '[-]' + movl $3, %edx # We want to compare 3 bytes + call strncmp # Compare them with strncmp(3) + testl %eax, %eax # Check if the comparison returned 0 + jz compile_zero # If it did, we matched the pattern + + movq %r15, %rdi # Move the current instruction pointer into %rdi + call copy_loop_checker # Call the copy loop checker + testq %rax, %rax # Check to see if we hit a copy loop + jz compile_loop_start_normal # If we didn't, this is a regular loop + movq %rax, %r15 # Otherwise, set %r15 to the new location + incq %r15 # Then point to the next instruction + jmp compile_out + +compile_loop_start_normal: + # Push the address of the loop start to the stack for the next ']' + pushq %r14 + + movq $OP_LOOP_START, (%r14) # Specify the LOOP_START opcode + movq $0, 8(%r14) # Zero the data section + incq %r15 # Increment the instruction pointer + jmp compile_out + +compile_loop_end: + popq 8(%r14) # Pop the address of the previous loop start to the data section + movq $OP_LOOP_END, (%r14) # Push the address of the loop end to the stack + incq %r15 # Increment the instruction pointer + jmp compile_out + +compile_read: + movq $OP_READ, (%r14) # Specify the READ opcode + movq $0, 8(%r14) # Zero the data section + incq %r15 # Increment the instruction pointer + jmp compile_out + +compile_write: + movq $OP_WRITE, (%r14) # Specify the WRITE opcode + movq $0, 8(%r14) # Zero the data section + incq %r15 # Increment the instruction pointer + jmp compile_out + +compile_zero: + movq $OP_ZERO, (%r14) # Specify the ZERO opcode + movq $0, 8(%r14) # Zero the data section + addq $3, %r15 # '[-]' is a 3 byte instruction + # FALLTHROUGH + +compile_out: + addq $16, %r14 # Move to the next opcode + movb (%r15), %al # Move the current instruction into %al + testb %al, %al # Check if we have reached the NUL byte + jne compile_loop # If we haven't, loop + movq $0, (%r14) # Otherwise, NUL terminate the bytecode + + # Now that we have traversed the entire program, we do a 2nd pass backwards so that we can + # set the jump addresses for the '[' commands now that the ']' commands have the addresses + # set. +compile_backwards: + # We are at the NUL terminator, so move backwards + subq $16, %r14 + + cmpq $OP_LOOP_END, (%r14) # Check if we hit a ']' + jne compile_backwards_next # If we didn't move to the nex check + pushq %r14 # Otherwise push the address of the opcode to the stack + jmp compile_backwards_out + +compile_backwards_next: + cmpq $OP_LOOP_START, (%r14) # Check if we hit a '[' + jne compile_backwards_out # If we didn't just keep looping + popq 8(%r14) # If we did then pop the corresponding ']' address + +compile_backwards_out: + cmpq %r14, (bytecode) # Check if we've seen every opcode + jne compile_backwards # If not keep looping + ret # Otherwise return + +# ================== +# Description: +# Try to figure out if we are at a copy loop and optimize it. A copy loop follows the pattern +# of a loop ([]) beginning with a '-' followed by N occurances of '>' followed by a '+' and N +# occurances of '<'. This sequence copies the current cell to the cell at offset N and clears +# the current cell afterwards. +# +# Args: +# %rdi: A pointer to the first '[' of the potential copy loop +# +# Return: +# 0 if not a copy loop, otherwise the new position of the instruction pointer. +# ================== +copy_loop_checker: + # Skip '[' + incq %rdi + + # All copy loops must begin with a '-' + cmpb $'-', (%rdi) + jne copy_loop_fail + incq %rdi + + # Zero %rax so we can use it to count the copy offset + xorl %eax, %eax +copy_loop_count_offset: + cmpb $'>', (%rdi) # Check for '>' + jne copy_loop_next # If we don't match anymore then move to the next step + incq %rax # Increment the offset counter + incq %rdi # Increment the instruction pointer + jmp copy_loop_count_offset # Loop again + +copy_loop_next: + cmpb $'+', (%rdi) # Check if we see the mandatory '+' + jne copy_loop_fail # If we don't then fail + incq %rdi # Otherwise move to the next instruction + + # The following code is the exact same as what we just did to count the offset but we are + # now using %rcx and decrementing for each '<'. This is so we can make sure that the copy + # loop is a working one. + cmpb $'<', (%rdi) + jne copy_loop_fail + movq %rax, %rcx +copy_loop_verify_offset: + cmpb $'<', (%rdi) + jne copy_loop_next_2 + decq %rcx + incq %rdi + +copy_loop_next_2: + cmpb $']', (%rdi) # Ensure this is the end of the loop + jne copy_loop_fail # If its not then fail + testq %rcx, %rcx # Otherwise make sure that our offsets line up + jnz copy_loop_fail # If they don't then fail + + movq $OP_COPY, (%r14) # Create an OP_COPY opcode + movq %rax, 8(%r14) # Set the offset to copy to + movq $OP_ZERO, 16(%r14) # Create an OP_ZERO opcode + movq $0, 24(%r14) # Set an empty data section + addq $16, %r14 # Increment the opcode pointer + + # Return the address of the instruction pointer + movq %rdi, %rax + ret +copy_loop_fail: + xorl %eax, %eax + ret + +# ================== +# Description: +# Execute the brainfuck program. This is done the slow and naive way by just looping over +# command and executing it (although the '[-]' pattern does get optimized for that towers of +# hanoi speed!). I have written a *much* faster C version which compiles the brainfuck into a +# more optimized bytecode, but porting that to ASM was just too frustrating. +# ================== +execute: + movq (bytecode), %r15 # Store the address of the program pointer into %r15 + movq $memory, %r14 # Store the address of the first memory cell into %r14 # TODO make sure this handles overflows normally +execute_loop: + # Load the current command into %rax + movq (%r15), %rax + + # Jump to a different label depending on which instruction we hit + cmpq $OP_ADD, %rax + je execute_add + cmpq $OP_SUB, %rax + je execute_sub + cmpq $OP_RIGHT, %rax + je execute_right + cmpq $OP_LEFT, %rax + je execute_left + cmpq $OP_LOOP_START, %rax + je execute_loop_start + cmpq $OP_LOOP_END, %rax + je execute_loop_end + cmpq $OP_READ, %rax + je execute_read + cmpq $OP_WRITE, %rax + je execute_write + cmpq $OP_ZERO, %rax + je execute_zero + cmpq $OP_COPY, %rax + je execute_copy + +execute_add: + # Increment the current memory cell + movq 8(%r15), %rax + addb %al, (%r14) + jmp execute_out + +execute_sub: + # Decrement the current memory cell + movq 8(%r15), %rax + subb %al, (%r14) + jmp execute_out + +execute_right: + # Move the memory pointer right by 1 + addq 8(%r15), %r14 + jmp execute_out + +execute_left: + # Move the memory pointer left by 1 + subq 8(%r15), %r14 + jmp execute_out + +execute_loop_start: + cmpb $0, (%r14) + cmovzq 8(%r15), %r15 + jmp execute_out + +execute_loop_end: + # If the current memory cell is 0 move to the next command + cmpb $0, (%r14) + cmovnzq 8(%r15), %r15 + jmp execute_out + +execute_read: + call getchar # Read a character with getchar(3) + cmpb $EOF, %al # Check if the EOF was read + je execute_read_eof # If EOF was read, jump to a special handler for that + movb %al, (%r14) # Otherwise move the read character into the current memory cell + jmp execute_out + +execute_read_eof: + # If EOF was read, set the current cell to 0 + movb $0, (%r14) + jmp execute_out + +execute_write: + movl (%r14), %edi # Move the current memory cell into %edi + call putchar # Print it with putchar(3) + jmp execute_out + +execute_zero: + # This is the one optimzation we actually do. If the string [-] is matched then just zero + # the current cell. + movb $0, (%r14) + jmp execute_out + +execute_copy: + movq 8(%r15), %rax + movb (%r14), %cl + movb %cl, (%rax, %r14, 1) + # FALLTHROUGH + +execute_out: + addq $16 , %r15 # Increment the instruction pointer + movq (%r15), %rax # Move the current instruction into %rax + testq %rax, %rax # Check if we have reached the NUL byte + jne execute_loop # If we haven't, loop + ret # Otherwise, return + +# ================== +# Description: +# The following die functions all work the same. They simply take the name of the +# corresponding function and store it in %rdi. Then the die function is called to print a +# message to stderr and terminate the program. +# ================== + +# Set the function name to "open" and call die +open_die: + movq $func_open, %rdi + jmp die + +# Same as open_die but for fstat(2) +fstat_die: + movq $func_fstat, %rdi + jmp die + +# Same as open_die but for read(2) +read_die: + movq $func_read, %rdi + jmp die + +# Same as open_die but for malloc(3) +malloc_die: + movq $func_malloc, %rdi + jmp die + + +# ================== +# Description: +# Print out an error message in the format ": " then exit via `_exit` +# +# Args: +# %rdi: The function name +# ================== +die: + # Store the function name temporarily in %r15 + movq %rdi, %r15 + + call __errno_location # Call __errno_location to get a pointer to errno + movq (%rax), %rdi # Move errno into %rdi + call strerror # Get the error string with strerror(3) + + movq %rax, %rcx # Set the error string + xorl %eax, %eax # Zero %rax + movq %r15, %rdx # Set the function name + movq $die_fmt, %rsi # Set the format string + movq stderr, %rdi # Get stderr + call fprintf # Print the message to stderr + + jmp _exit + +# ================== +# Description: +# Print a usage message to standard error and exit the program via `_exit` +# ================== +usage: + xorl %eax, %eax # Set rax to 0 + movq (%rsi), %rdx # Get argv[0] + movq $usage_fmt, %rsi # Set the format string + movq stderr, %rdi # Get stderr + call fprintf # Print the error + # FALLTHROUGH + +# ================== +# Description: +# Exit the program with the return code EXIT_FAILURE +# ================== +_exit: + movl $EXIT_FAILURE, %eax + call exit -- cgit v1.2.3