From ec3ad2e497a7a6fec0386fdd4828dc7fcb94c126 Mon Sep 17 00:00:00 2001
From: Derek Christ <christ.derek@gmail.com>
Date: Wed, 21 Feb 2024 23:43:11 +0100
Subject: [PATCH] Introduce async execution API without barriers

---
 pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs | 9 ++++++---
 pim-os/src/pim/operation.rs                        | 4 ++++
 2 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs b/pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs
index 6f23c5b..42c6e16 100644
--- a/pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs
+++ b/pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs
@@ -1,6 +1,7 @@
 use crate::pim::{
     continuous_array::Matrix, interleaved_array, operation::PimOperand, vector::F16x16,
 };
+use aarch64_cpu::asm::barrier;
 use nalgebra::SVector;
 use pim_isa::{File, Instruction, Kernel};
 
@@ -88,12 +89,14 @@ pub fn execute<const X16R: usize, const R: usize, const X16C: usize>(
         block.execute_read();
     }
 
-    for row_block in matrix.0.iter() {
-        for column_block in row_block.fixed_rows::<1>(0).iter() {
-            column_block.execute_read();
+    for sub_matrix in matrix.0.iter() {
+        for column_block in sub_matrix.fixed_rows::<1>(0).iter() {
+            column_block.execute_read_async();
         }
     }
 
+    barrier::dsb(barrier::SY);
+
     for chunk in output_partial_sum_vector
         .fixed_rows_with_step_mut::<X16R>(0, 16)
         .iter_mut()
diff --git a/pim-os/src/pim/operation.rs b/pim-os/src/pim/operation.rs
index 146a480..a0c59b4 100644
--- a/pim-os/src/pim/operation.rs
+++ b/pim-os/src/pim/operation.rs
@@ -9,6 +9,10 @@ pub trait PimOperand {
         barrier::dsb(barrier::SY);
     }
 
+    fn execute_read_async(&self) {
+        unsafe { core::ptr::read_volatile(self.ptr()) };
+    }
+
     fn execute_write(&mut self) {
         unsafe { core::ptr::write_volatile(self.ptr_mut(), Default::default()) };
         barrier::dsb(barrier::SY);