From e6f5494242244520208f9d25e0ecb9d59908b3b9 Mon Sep 17 00:00:00 2001
From: Gwilym Kuiper <gw@ilym.me>
Date: Sat, 11 Jun 2022 17:22:29 +0100
Subject: [PATCH] Use the syscall to do the memset because the default
 implementation is really slow

---
 agb/src/sound/mixer/sw_mixer.rs |  4 +++-
 agb/src/syscall.rs              | 30 ++++++++++++++++++++++++++++--
 2 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/agb/src/sound/mixer/sw_mixer.rs b/agb/src/sound/mixer/sw_mixer.rs
index 5604672..629dd3e 100644
--- a/agb/src/sound/mixer/sw_mixer.rs
+++ b/agb/src/sound/mixer/sw_mixer.rs
@@ -5,6 +5,7 @@ use bare_metal::{CriticalSection, Mutex};
 use super::hw;
 use super::hw::LeftOrRight;
 use super::{SoundChannel, SoundPriority};
+use crate::syscall::cpu_fast_fill_i8;
 use crate::{
     fixnum::Num,
     interrupt::free,
@@ -284,7 +285,8 @@ impl MixerBuffer {
         let write_buffer_index = free(|cs| self.state.borrow(*cs).borrow_mut().active_advanced());
 
         let write_buffer = &mut self.buffers[write_buffer_index].0;
-        write_buffer.fill(0);
+        cpu_fast_fill_i8(write_buffer, 0);
+
         unsafe {
             agb_rs__mixer_collapse(write_buffer.as_mut_ptr(), buffer.as_ptr());
         }
diff --git a/agb/src/syscall.rs b/agb/src/syscall.rs
index 3e9492d..da94ee6 100644
--- a/agb/src/syscall.rs
+++ b/agb/src/syscall.rs
@@ -2,7 +2,6 @@ use core::arch::asm;
 
 // use crate::display::object::AffineMatrixAttributes;
 
-
 #[allow(non_snake_case)]
 
 pub fn halt() {
@@ -94,6 +93,9 @@ pub fn arc_tan(n: i16) -> i16 {
             "swi 0x09",
             in("r0") n,
             lateout("r0") result,
+            lateout("r1") _,
+            lateout("r2") _,
+            lateout("r3") _
         );
     }
     result
@@ -107,12 +109,36 @@ pub fn arc_tan2(x: i16, y: i32) -> i16 {
             in("r0") x,
             in("r1") y,
             lateout("r0") result,
-
+            lateout("r2") _,
+            lateout("r3") _
         );
     }
     result
 }
 
+pub(crate) fn cpu_fast_fill_i8(input: &mut [i8], new_content: i32) {
+    assert_eq!(
+        input.len() % (4 * 8),
+        0,
+        "Input length must be divisible by 32"
+    );
+
+    let input_ptr = [new_content].as_ptr();
+    let output_ptr = input.as_mut_ptr();
+    let length_mode = (1 << 24) | // copy
+        (input.len() / 4);
+
+    unsafe {
+        asm!(
+            "swi 0x0c",
+            in("r0") input_ptr,
+            in("r1") output_ptr,
+            in("r2") length_mode,
+            lateout("r3") _,
+        );
+    }
+}
+
 // pub fn affine_matrix(
 //     x_scale: Num<i16, 8>,
 //     y_scale: Num<i16, 8>,