Restructure pim-os project

pim-os/src/bin/gemv.rs

@@ -10,9 +10,9 @@ use nalgebra::{SMatrix, SVector};
 use num_traits::{One, Zero};
 use pim_isa::BankMode;
 use pim_os::{
+    kernel::gemv,
     pim::{
         self, interleaved_array,
-        kernel::samsung_matrix_vector_mul,
         vector::{F16x1, F16x16},
     },
     uart::Uart0,
@@ -25,7 +25,7 @@ const X16_COLUMNS: usize = COLUMNS / 16;
 
 #[no_mangle]
 pub extern "C" fn main() {
-    pim::state::set_kernel(&samsung_matrix_vector_mul::KERNEL);
+    pim::state::set_kernel(&gemv::KERNEL);
 
     let mut matrix = SMatrix::<_, ROWS, COLUMNS>::zeros();
     matrix.fill_lower_triangle(F16x1::one(), 0);
@@ -46,7 +46,7 @@ pub extern "C" fn main() {
     {
         pim::state::set_bank_mode(BankMode::PimAllBank);
 
-        samsung_matrix_vector_mul::execute(
+        gemv::execute(
             pim_matrix.as_ref(),
             interleaved_input_vector.as_ref(),
             output_partial_sum_vector.as_mut(),

pim-os/src/bin/vadd.rs

@@ -0,0 +1,69 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use aarch64_cpu::asm::barrier;
+use alloc::boxed::Box;
+use core::fmt::Write;
+use nalgebra::{SMatrix, SVector};
+use num_traits::{One, Zero};
+use pim_isa::BankMode;
+use pim_os::{
+    kernel::gemv,
+    pim::{
+        self, interleaved_array,
+        vector::{F16x1, F16x16},
+    },
+    uart::Uart0,
+};
+
+const ROWS: usize = 32;
+const COLUMNS: usize = 128;
+const X16_ROWS: usize = ROWS / 16;
+const X16_COLUMNS: usize = COLUMNS / 16;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    pim::state::set_kernel(&gemv::KERNEL);
+
+    let mut matrix = SMatrix::<_, ROWS, COLUMNS>::zeros();
+    matrix.fill_lower_triangle(F16x1::one(), 0);
+
+    let pim_matrix = Box::new(pim::continuous_array::Matrix::<X16_ROWS, X16_COLUMNS>::from(matrix));
+
+    let input_vector = SVector::<_, X16_COLUMNS>::from_element(F16x16::one());
+    let interleaved_input_vector = Box::new(interleaved_array::Vector::from(input_vector));
+
+    let mut output_partial_sum_vector = Box::new(SVector::<F16x16, ROWS>::zeros());
+
+    let dummy = Box::new(0);
+
+    // Verify everything is correctly initialized before PIM operation
+    barrier::dsb(barrier::SY);
+
+    // Execute kernel
+    {
+        pim::state::set_bank_mode(BankMode::PimAllBank);
+
+        gemv::execute(
+            pim_matrix.as_ref(),
+            interleaved_input_vector.as_ref(),
+            output_partial_sum_vector.as_mut(),
+            dummy.as_ref(),
+        );
+
+        pim::state::set_bank_mode(BankMode::SingleBank);
+    }
+
+    writeln!(Uart0, "{output_partial_sum_vector}").unwrap();
+
+    let output_vector = SVector::<F16x1, ROWS>::from_fn(|r, _| {
+        output_partial_sum_vector[r]
+            .0
+            .iter()
+            .fold(F16x1::zero(), |acc, val| acc + *val)
+    });
+
+    writeln!(Uart0, "{output_vector}").unwrap();
+}

pim-os/src/kernel.rs

@@ -0,0 +1,2 @@
+pub mod legacy;
+pub mod gemv;

pim-os/src/kernel/legacy.rs

@@ -0,0 +1,3 @@
+pub mod matrix_matrix_mul;
+pub mod matrix_scalar_mul;
+pub mod matrix_vector_mul;

pim-os/src/kernel/legacy/matrix_matrix_add.rs

@@ -1,4 +1,4 @@
-use crate::pim::array::{DummyArray, PimMatrixArena, PimRegion};
+use crate::pim::legacy::array::{DummyArray, PimMatrixArena, PimRegion};
 use pim_isa::{File, Instruction, Kernel};
 
 pub const KERNEL: Kernel = Kernel([

pim-os/src/kernel/legacy/matrix_matrix_mul.rs

@@ -1,4 +1,4 @@
-use crate::pim::array::{DummyArray, PimMatrixArena, PimRegion};
+use crate::pim::legacy::array::{DummyArray, PimMatrixArena, PimRegion};
 use pim_isa::{File, Instruction, Kernel};
 
 pub const KERNEL: Kernel = Kernel([

pim-os/src/kernel/legacy/matrix_scalar_mul.rs

@@ -1,4 +1,4 @@
-use crate::pim::array::{DummyArray, PimMatrixArena, PimRegion, PimScalarArena};
+use crate::pim::legacy::array::{DummyArray, PimMatrixArena, PimRegion, PimScalarArena};
 use pim_isa::{File, Instruction, Kernel};
 
 pub const KERNEL: Kernel = Kernel([

pim-os/src/kernel/legacy/matrix_vector_mul.rs

@@ -1,4 +1,4 @@
-use crate::pim::array::{DummyArray, PimMatrixArena, PimRegion};
+use crate::pim::legacy::array::{DummyArray, PimMatrixArena, PimRegion};
 use pim_isa::{File, Instruction, Kernel};
 
 pub const KERNEL: Kernel = Kernel([

pim-os/src/lib.rs

@@ -4,10 +4,13 @@
 use core::sync::atomic::{compiler_fence, Ordering};
 
 mod alloc;
+mod panic;
+
 pub mod boot;
 pub mod critical_section;
+pub mod kernel;
 pub mod m5op;
-mod panic;
+pub mod memory_config;
 pub mod pim;
 pub mod uart;
 

pim-os/src/main.rs

@@ -1,86 +0,0 @@
-#![no_std]
-#![no_main]
-
-extern crate alloc;
-
-use aarch64_cpu::asm::barrier;
-use alloc::{boxed::Box, rc::Rc};
-use core::{cell::RefCell, fmt::Write};
-use half::f16;
-use nalgebra::Matrix;
-use pim_isa::BankMode;
-use pim_os::{
-    m5op,
-    pim::{
-        self,
-        array::{DummyArray, PimMatrixArena, PimStorage, NUMBER_OF_BANKS},
-        kernel::matrix_matrix_mul,
-        vector::{F16x1, F16x16},
-    },
-    uart::Uart0,
-};
-
-#[no_mangle]
-pub extern "C" fn main() {
-    pim::state::set_kernel(&matrix_matrix_mul::KERNEL);
-
-    let pim_matrix_arena0 = Rc::new(RefCell::new(PimMatrixArena(
-        [[[F16x16::default(); NUMBER_OF_BANKS]; 8]; 8],
-    )));
-    let pim_matrix_arena1 = Rc::new(RefCell::new(PimMatrixArena(
-        [[[F16x16::default(); NUMBER_OF_BANKS]; 8]; 8],
-    )));
-    let pim_matrix_arena2 = Rc::new(RefCell::new(PimMatrixArena(
-        [[[F16x16::default(); NUMBER_OF_BANKS]; 8]; 8],
-    )));
-
-    let mut matrix0 = Matrix::from_data(PimStorage {
-        arena: &pim_matrix_arena0,
-        index: 0,
-    });
-    matrix0.fill_lower_triangle(F16x1(f16::ONE), 0);
-
-    let mut matrix1 = Matrix::from_data(PimStorage {
-        arena: &pim_matrix_arena1,
-        index: 0,
-    });
-    matrix1.fill_lower_triangle(F16x1(f16::ONE), 0);
-
-    let matrix2 = Matrix::from_data(PimStorage {
-        arena: &pim_matrix_arena2,
-        index: 0,
-    });
-
-    writeln!(Uart0, "{matrix0} * {matrix1}\n=").unwrap();
-
-    let dummy_array = Box::new(DummyArray([F16x16::default(); NUMBER_OF_BANKS]));
-
-    // Verify everything is correctly initialized before PIM operation
-    barrier::dsb(barrier::SY);
-
-    // Execute kernel
-    {
-        let pim_matrix_arena0 = &pim_matrix_arena0.borrow();
-        let pim_matrix_arena1 = &pim_matrix_arena1.borrow();
-        let pim_matrix_arena2 = &mut pim_matrix_arena2.borrow_mut();
-
-        pim::state::set_bank_mode(BankMode::PimAllBank);
-
-        m5op::reset_stats();
-
-        for _ in 0..100 {
-            matrix_matrix_mul::execute(
-                pim_matrix_arena0,
-                pim_matrix_arena1,
-                pim_matrix_arena2,
-                dummy_array.as_ref(),
-            );
-        }
-
-        m5op::dump_stats();
-
-        pim::state::set_bank_mode(BankMode::SingleBank);
-    }
-
-    writeln!(Uart0, "{matrix2}").unwrap();
-}

pim-os/src/memory_config.rs

@@ -0,0 +1 @@
+pub const NUMBER_OF_BANKS: usize = 32;

pim-os/src/pim.rs

@@ -1,8 +1,7 @@
-pub mod array;
 pub mod config;
 pub mod continuous_array;
 pub mod interleaved_array;
-pub mod kernel;
+pub mod legacy;
 pub mod operation;
 pub mod state;
 pub mod vector;

pim-os/src/pim/continuous_array.rs

@@ -1,11 +1,15 @@
 use super::vector::{F16x1, F16x16};
 use core::fmt::Display;
-use nalgebra::SMatrix;
+use nalgebra::{SMatrix, SVector};
 
 #[repr(C, align(65536))]
 #[derive(Debug)]
 pub struct Matrix<const X16R: usize, const X16C: usize>(pub [SMatrix<F16x16, 16, X16C>; X16R]);
 
+#[repr(C, align(1024))]
+#[derive(Debug)]
+pub struct Vector<const R: usize>(pub SVector<F16x1, R>);
+
 impl<const X16R: usize, const X16C: usize> Display for Matrix<X16R, X16C> {
     fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
         for block in self.0.iter() {

pim-os/src/pim/interleaved_array.rs

@@ -1,4 +1,5 @@
-use super::{array::NUMBER_OF_BANKS, vector::F16x16};
+use super::vector::F16x16;
+use crate::memory_config::NUMBER_OF_BANKS;
 use nalgebra::SVector;
 
 #[repr(C, align(512))]

pim-os/src/pim/kernel.rs

@@ -1,5 +0,0 @@
-pub mod matrix_matrix_add;
-pub mod matrix_matrix_mul;
-pub mod matrix_scalar_mul;
-pub mod matrix_vector_mul;
-pub mod samsung_matrix_vector_mul;

pim-os/src/pim/legacy.rs

@@ -0,0 +1 @@
+pub mod array;

pim-os/src/pim/legacy/array.rs

@@ -1,10 +1,12 @@
-use super::vector::{F16x1, F16x16};
+use crate::{
+    memory_config::NUMBER_OF_BANKS,
+    pim::vector::{F16x1, F16x16},
+};
 use aarch64_cpu::asm::barrier;
 use core::{arch::asm, cell::RefCell};
 use half::f16;
 use nalgebra::{Const, Dyn, RawStorage, RawStorageMut};
 
-pub const NUMBER_OF_BANKS: usize = 32;
 const EVEN_BANK_INDEX: usize = 0;
 const ODD_BANK_INDEX: usize = 8;