Merge pull request #238 from gwilymk/agbabi

Start using agbabi for memset and memcpy

agb/build.rs

@@ -1,7 +1,13 @@
 use std::path;
 
 fn main() {
-    let asm = &["crt0.s", "interrupt_handler.s", "src/sound/mixer/mixer.s"];
+    let asm = &[
+        "crt0.s",
+        "interrupt_handler.s",
+        "src/sound/mixer/mixer.s",
+        "src/agbabi/memset.s",
+        "src/agbabi/memcpy.s",
+    ];
 
     println!("cargo:rerun-if-changed=gba.ld");
     println!("cargo:rerun-if-changed=gba_mb.ld");
@@ -50,6 +56,7 @@ fn main() {
     }
 
     let archive = format!("{out_dir}/agb.a");
+    let _ = std::fs::remove_file(&archive);
     let ar_out = std::process::Command::new("arm-none-eabi-ar")
         .arg("-crs")
         .arg(&archive)

agb/gba.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

agb/gba_mb.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

agb/interrupt_handler.s

@@ -2,7 +2,7 @@
 @ An interrupt handler that simply acknowledges all interrupts
     .arm
     .global InterruptHandler
-    .section .iwram, "ax"
+    .section .iwram, "ax", %progbits
     .align
 InterruptHandler:
     mov r2, #0x04000000 @ interrupt enable register location
@@ -50,7 +50,7 @@ InterruptHandler:
 .pool
 
 
-.section .iwram
+.section .iwram.program_counter
     .global agb_rs__program_counter
     .balign 4
 agb_rs__program_counter:

agb/src/agbabi/LICENSE.md

@@ -0,0 +1,17 @@
+libagbabi is available under the [zlib license](https://www.zlib.net/zlib_license.html) :
+
+This software is provided 'as-is', without any express or implied
+warranty.  In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not
+   claim that you wrote the original software. If you use this software
+   in a product, an acknowledgment in the product documentation would be
+   appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be
+   misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.

agb/src/agbabi/README.md

@@ -0,0 +1,10 @@
+# agbabi
+
+GBA optimized library functions for common operations.
+
+Includes implementations for various aeabi functions.
+
+# Use in agb
+
+Restricted to the subset of replacement functions which make the biggest difference when using rust on the gba.
+Minor modifications have been made around adding extra symbols to reference in the linker script.

agb/src/agbabi/memcpy.s

@@ -0,0 +1,115 @@
+/*
+===============================================================================
+
+ ABI:
+    __aeabi_memcpy, __aeabi_memcpy4, __aeabi_memcpy8
+ Standard:
+    memcpy
+ Support:
+    __agbabi_memcpy2
+
+ Copyright (C) 2021-2022 agbabi contributors
+ For conditions of distribution and use, see copyright notice in LICENSE.md
+
+===============================================================================
+*/
+
+    .arm
+    .align 4
+
+    .section .iwram.__aeabi_memcpy, "ax", %progbits
+    .global __agbabi_memcpy
+__agbabi_memcpy:
+    .global __aeabi_memcpy
+__aeabi_memcpy:
+    // Check pointer alignment
+    eor     r3, r1, r0
+    // JoaoBapt carry & sign bit test
+    movs    r3, r3, lsl #31
+    bmi     .Lcopy1
+    bcs     .Lcopy2
+
+.Lcopy4:
+    // Copy half and byte head
+    rsb     r3, r0, #4
+    movs    r3, r3, lsl #31
+    ldrmib  r3, [r1], #1
+    strmib  r3, [r0], #1
+    submi   r2, r2, #1
+    ldrcsh  r3, [r1], #2
+    strcsh  r3, [r0], #2
+    subcs   r2, r2, #2
+    // Fallthrough
+
+    .global __aeabi_memcpy8
+__aeabi_memcpy8:
+    .global __aeabi_memcpy4
+__aeabi_memcpy4:
+    // Copy 8 words
+    movs    r12, r2, lsr #5
+    beq     .Lskip32
+    lsl     r3, r12, #5
+    sub     r2, r2, r3
+    push    {r4-r10}
+.LcopyWords8:
+    ldmia   r1!, {r3-r10}
+    stmia   r0!, {r3-r10}
+    subs    r12, r12, #1
+    bne     .LcopyWords8
+    pop     {r4-r10}
+.Lskip32:
+
+    // Copy words
+    movs    r12, r2, lsr #2
+.LcopyWords:
+    subs    r12, r12, #1
+    ldrhs   r3, [r1], #4
+    strhs   r3, [r0], #4
+    bhs     .LcopyWords
+
+    // Copy half and byte tail
+    movs    r3, r2, lsl #31
+    ldrcsh  r3, [r1], #2
+    strcsh  r3, [r0], #2
+    ldrmib  r3, [r1]
+    strmib  r3, [r0]
+    bx      lr
+
+.Lcopy2:
+    // Copy byte head
+    tst     r0, #1
+    ldrneb  r3, [r1], #1
+    strneb  r3, [r0], #1
+    subne   r2, r2, #1
+    // Fallthrough
+
+    .global __agbabi_memcpy2
+__agbabi_memcpy2:
+    // Copy halves
+    movs    r12, r2, lsr #1
+.LcopyHalves:
+    subs    r12, r12, #1
+    ldrhsh  r3, [r1], #2
+    strhsh  r3, [r0], #2
+    bhs     .LcopyHalves
+
+    // Copy byte tail
+    tst     r2, #1
+    ldrneb  r3, [r1]
+    strneb  r3, [r0]
+    bx      lr
+
+.Lcopy1:
+    subs    r2, r2, #1
+    ldrhsb  r3, [r1], #1
+    strhsb  r3, [r0], #1
+    bhs     .Lcopy1
+    bx      lr
+
+    .section .iwram.memcpy, "ax", %progbits
+    .global memcpy
+memcpy:
+    push    {r0, lr}
+    bl      __aeabi_memcpy
+    pop     {r0, lr}
+    bx      lr

agb/src/agbabi/memset.s

@@ -0,0 +1,107 @@
+/*
+===============================================================================
+
+ ABI:
+    __aeabi_memset, __aeabi_memset4, __aeabi_memset8,
+    __aeabi_memclr, __aeabi_memclr4, __aeabi_memclr8
+ Standard:
+    memset
+ Support:
+    __agbabi_wordset4
+
+ Copyright (C) 2021-2022 agbabi contributors
+ For conditions of distribution and use, see copyright notice in LICENSE.md
+
+===============================================================================
+*/
+
+    .arm
+    .balign 4
+
+    .section .iwram.__aeabi_memset, "ax", %progbits
+    .global __agbabi_memset
+__agbabi_memset:
+    .global __aeabi_memclr
+__aeabi_memclr:
+    mov     r2, #0
+    b       .LskipShifts
+
+    .global __aeabi_memset
+__aeabi_memset:
+    mov     r2, r2, lsl #24
+    orr     r2, r2, r2, lsr #8
+    orr     r2, r2, r2, lsr #16
+    // Fallthrough
+
+.LskipShifts:
+    // JoaoBapt carry & sign bit test
+    rsb     r3, r0, #4
+    movs    r3, r3, lsl #31
+    // Set half and byte head
+    strmib  r2, [r0], #1
+    submi   r1, r1, #1
+    strcsh  r2, [r0], #2
+    subcs   r1, r1, #2
+    b       __agbabi_wordset4
+
+    .global __aeabi_memclr8
+__aeabi_memclr8:
+    .global __aeabi_memclr4
+__aeabi_memclr4:
+    mov     r2, #0
+    b       __agbabi_wordset4
+
+    .global __aeabi_memset8
+__aeabi_memset8:
+    .global __aeabi_memset4
+__aeabi_memset4:
+    mov     r2, r2, lsl #24
+    orr     r2, r2, r2, lsr #8
+    orr     r2, r2, r2, lsr #16
+    // Fallthrough
+
+    .global __agbabi_wordset4
+__agbabi_wordset4:
+    // Set 8 words
+    movs    r12, r1, lsr #5
+    beq     .Lskip32
+    lsl     r3, r12, #5
+    sub     r1, r1, r3
+    push    {r4-r9}
+    mov     r3, r2
+    mov     r4, r2
+    mov     r5, r2
+    mov     r6, r2
+    mov     r7, r2
+    mov     r8, r2
+    mov     r9, r2
+.LsetWords8:
+    stmia   r0!, {r2-r9}
+    subs    r12, r12, #1
+    bne     .LsetWords8
+    pop     {r4-r9}
+.Lskip32:
+
+    // Set words
+    movs    r12, r1, lsr #2
+.LsetWords:
+    subs    r12, r12, #1
+    strhs   r2, [r0], #4
+    bhs     .LsetWords
+
+    // Set half and byte tail
+    movs    r3, r1, lsl #31
+    strcsh  r2, [r0], #2
+    strmib  r2, [r0]
+    bx      lr
+
+    .section .iwram.memset, "ax", %progbits
+    .global memset
+memset:
+    mov     r3, r1
+    mov     r1, r2
+    mov     r2, r3
+    push    {r0, lr}
+    bl      __aeabi_memset
+    pop     {r0, lr}
+    bx      lr

agb/src/asm_include.s

@@ -1,5 +1,5 @@
 .macro agb_arm_func functionName:req
-.section .iwram
+.section .iwram.\functionName, "ax", %progbits
 .arm
 .align 2
 .global \functionName

agb/src/sound/mixer/mixer.s

@@ -1,6 +1,6 @@
 .include "src/asm_include.s"
 
-.section .iwram
+.section .iwram.buffer_size
     .global agb_rs__buffer_size
     .balign 4
 agb_rs__buffer_size:
@@ -26,7 +26,8 @@ modifications_fallback:
     orr r7, r7, r3, lsl #16   @ r7 now is the left channel followed by the right channel modifications.
 
     mov r5, #0                   @ current index we're reading from
-    ldr r8, agb_rs__buffer_size @ the number of steps left
+    ldr r8, =agb_rs__buffer_size @ the number of steps left
+    ldr r8, [r8]
 
 
 1:
@@ -64,7 +65,8 @@ same_modification:
     bne 1b
 
     mov r5, #0                   @ current index we're reading from
-    ldr r8, agb_rs__buffer_size @ the number of steps left
+    ldr r8, =agb_rs__buffer_size @ the number of steps left
+    ldr r8, [r8]
 
 1:
 .rept 4
@@ -99,7 +101,8 @@ agb_arm_func agb_rs__mixer_add_stereo
 
     ldr r5, =0x00000FFF
 
-    ldr r8, agb_rs__buffer_size
+    ldr r8, =agb_rs__buffer_size
+    ldr r8, [r8]
 1:
 .rept 4
     ldrsh r6, [r0], #2        @ load the current sound sample to r6
@@ -138,34 +141,6 @@ agb_arm_func agb_rs__mixer_add_stereo
 
 agb_arm_end agb_rs__mixer_add_stereo
 
-.section .iwram
-    .balign 4
-constant_zero:
-.rept 4
-    .word 0
-.endr
-
-agb_arm_func agb_rs__init_buffer
-    @ arguments:
-    @ r0 = target buffer
-    @ r1 = size in bytes (must be a multiple of 16)
-    push {r4-r5}
-
-    @ zero registers r3-r5
-    ldr r2, =constant_zero
-    ldm r2, {r3-r5,r12}
-
-1:
-    @ zero 4 words worth of the buffer
-    stmia r0!, {r3-r5,r12}
-    subs r1, r1, #(4 * 4)
-    @ loop if we haven't zeroed everything
-    bne 1b
-
-    pop {r4-r5}
-    bx lr
-agb_arm_end agb_rs__init_buffer
-
 agb_arm_func agb_rs__mixer_collapse
     @ Arguments:
     @ r0 = target buffer (i8)
@@ -184,7 +159,8 @@ SWAP_SIGN .req r11
     ldr CONST_127, =127
     ldr SWAP_SIGN, =0x80808080
 
-    ldr r2, agb_rs__buffer_size @ loop counter
+    ldr r2, =agb_rs__buffer_size @ loop counter
+    ldr r2, [r2]
     mov r4, r2
 
 @ The idea for this solution came from pimpmobile:

agb/src/sound/mixer/sw_mixer.rs

@@ -1,6 +1,5 @@
 use core::cell::RefCell;
-use core::mem;
+use core::intrinsics::transmute;
-use core::mem::MaybeUninit;
 
 use bare_metal::{CriticalSection, Mutex};
 
@@ -29,8 +28,6 @@ extern "C" {
     fn agb_rs__mixer_add_stereo(sound_data: *const u8, sound_buffer: *mut Num<i16, 4>);
 
     fn agb_rs__mixer_collapse(sound_buffer: *mut i8, input_buffer: *const Num<i16, 4>);
-
-    fn agb_rs__init_buffer(buffer: *mut MaybeUninit<Num<i16, 4>>, size_in_bytes: usize);
 }
 
 pub struct Mixer {
@@ -235,28 +232,8 @@ impl MixerBuffer {
     }
 
     fn write_channels<'a>(&mut self, channels: impl Iterator<Item = &'a mut SoundChannel>) {
-        // This code is equivalent to:
+        let mut buffer: [Num<i16, 4>; constants::SOUND_BUFFER_SIZE * 2] =
-        // let mut buffer: [Num<i16, 4>; constants::SOUND_BUFFER_SIZE * 2] =
+            unsafe { transmute([0i16; constants::SOUND_BUFFER_SIZE * 2]) };
-        //     [Num::new(0); constants::SOUND_BUFFER_SIZE * 2];
-        // but the above uses approximately 7% of the CPU time if running at 32kHz
-        let mut buffer: [Num<i16, 4>; constants::SOUND_BUFFER_SIZE * 2] = {
-            // Create an uninitialized array of `MaybeUninit`. The `assume_init` is
-            // safe because the type we are claiming to have initialized here is a
-            // bunch of `MaybeUninit`s, which do not require initialization.
-            let mut data: [MaybeUninit<Num<i16, 4>>; constants::SOUND_BUFFER_SIZE * 2] =
-                unsafe { MaybeUninit::uninit().assume_init() };
-
-            // Actually init the array (by filling it with zeros) and then transmute it (which is safe because
-            // we have now zeroed everything)
-            unsafe {
-                agb_rs__init_buffer(
-                    data.as_mut_ptr(),
-                    mem::size_of::<Num<i16, 4>>() * data.len(),
-                );
-
-                mem::transmute(data)
-            }
-        };
 
         for channel in channels {
             if channel.is_done {

book/games/pong/gba.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

book/games/pong/gba_mb.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

examples/the-hat-chooses-the-wizard/gba.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

examples/the-purple-night/gba.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

examples/the-purple-night/gba_mb.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

justfile

@@ -50,6 +50,10 @@ build-book:
 update-lockfiles:
     bash .github/scripts/update-lockfiles.sh
 
+update-linker-scripts:
+    find -type f -name gba.ld | grep -v ./agb/gba.ld | xargs -n1 cp -v -- agb/gba.ld
+    find -type f -name gba_mb.ld | grep -v ./agb/gba_mb.ld | xargs -n1 cp -v -- agb/gba_mb.ld
+
 _build-rom folder name:
     #!/usr/bin/env bash
     set -euxo pipefail

template/gba.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K

template/gba_mb.ld

@@ -4,6 +4,9 @@ OUTPUT_ARCH(arm)
 ENTRY(__start)
 EXTERN(__RUST_INTERRUPT_HANDLER)
 
+EXTERN(__agbabi_memset)
+EXTERN(__agbabi_memcpy)
+
 MEMORY {
     ewram (w!x) : ORIGIN = 0x02000000, LENGTH = 256K
     iwram (w!x) : ORIGIN = 0x03000000, LENGTH = 32K