Master thesis simulation setup

pim-os/build.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+
+CARGO_TARGET_DIR=kernels/X1 cargo build --release --features X1
+CARGO_TARGET_DIR=kernels/X2 cargo build --release --features X2
+CARGO_TARGET_DIR=kernels/X3 cargo build --release --features X3
+CARGO_TARGET_DIR=kernels/X4 cargo build --release --features X4

pim-os/src/bin/classic_gemv.rs

@@ -0,0 +1,45 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use core::{arch::asm, fmt::Write};
+use nalgebra::{SMatrix, SVector};
+use pim_os::{pim::vector::F16x1, uart::Uart0};
+
+#[cfg(feature = "X1")]
+const ROWS: usize = 128;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 256;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 1024;
+
+const COLUMNS: usize = 128;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    let matrix = SMatrix::<F16x1, ROWS, COLUMNS>::zeros();
+    let input_vector = SVector::<F16x1, COLUMNS>::zeros();
+
+    // Flush cache
+    for element in matrix.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+    for element in input_vector.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+
+    pim_os::m5op::exit(0);
+    for _ in 0..10 {
+        let output_vector = matrix * input_vector;
+        core::hint::black_box(output_vector);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/classic_gemv_layers.rs

@@ -0,0 +1,46 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use core::fmt::Write;
+use half::f16;
+use nalgebra::{SMatrix, SVector};
+use pim_os::{pim::vector::F16x1, uart::Uart0};
+
+#[cfg(feature = "X1")]
+const DIMENSIONS: usize = 128;
+
+#[cfg(feature = "X2")]
+const DIMENSIONS: usize = 256;
+
+#[cfg(feature = "X3")]
+const DIMENSIONS: usize = 512;
+
+#[cfg(feature = "X4")]
+const DIMENSIONS: usize = 1024;
+
+const ROWS: usize = DIMENSIONS;
+const COLUMNS: usize = DIMENSIONS;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    let matrix = SMatrix::<F16x1, ROWS, COLUMNS>::zeros();
+    let mut input_vector = SVector::<F16x1, COLUMNS>::zeros();
+
+    pim_os::m5op::exit(0);
+    for _ in 0..1 {
+        input_vector = matrix * input_vector;
+        input_vector = input_vector.map(|element| {
+            if element.0 < f16::ZERO {
+                F16x1(f16::ZERO)
+            } else {
+                element
+            }
+        });
+        core::hint::black_box(input_vector);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/classic_haxpy.rs

@@ -0,0 +1,46 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use core::{arch::asm, fmt::Write};
+use nalgebra::SVector;
+use num_traits::identities::Zero;
+use pim_os::{pim::vector::F16x1, uart::Uart0};
+
+#[cfg(feature = "X1")]
+const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    let a = SVector::<F16x1, ROWS>::zeros();
+    let b = SVector::<F16x1, ROWS>::zeros();
+    let s = F16x1::zero();
+
+    // Flush cache
+    for element in a.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+    for element in b.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+
+    pim_os::m5op::exit(0);
+    for _ in 0..10 {
+        let a_s = a * s;
+        let c = a_s.component_mul(&b);
+        core::hint::black_box(c);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/classic_vadd.rs

@@ -0,0 +1,42 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use core::{arch::asm, fmt::Write};
+use nalgebra::SVector;
+use pim_os::{pim::vector::F16x1, uart::Uart0};
+
+#[cfg(feature = "X1")]
+const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    let a = SVector::<F16x1, ROWS>::zeros();
+    let b = SVector::<F16x1, ROWS>::zeros();
+
+    for element in a.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+    for element in b.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+
+    pim_os::m5op::exit(0);
+    for _ in 0..10 {
+        let c = a + b;
+        core::hint::black_box(c);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/classic_vmul.rs

@@ -0,0 +1,42 @@
+#![no_std]
+#![no_main]
+
+extern crate alloc;
+
+use core::{arch::asm, fmt::Write};
+use nalgebra::SVector;
+use pim_os::{pim::vector::F16x1, uart::Uart0};
+
+#[cfg(feature = "X1")]
+const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
+#[no_mangle]
+pub extern "C" fn main() {
+    let a = SVector::<F16x1, ROWS>::zeros();
+    let b = SVector::<F16x1, ROWS>::zeros();
+
+    for element in a.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+    for element in b.iter() {
+        unsafe { asm!("dc civac, {val}", val = in(reg) element) }
+    }
+
+    pim_os::m5op::exit(0);
+    for _ in 0..10 {
+        let c = a.component_mul(&b);
+        core::hint::black_box(c);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/gemv.rs

@@ -23,9 +23,31 @@ const COLUMNS: usize = 128;
 const X16_ROWS: usize = ROWS / 16;
 const X16_COLUMNS: usize = COLUMNS / 16;
 
+#[cfg(feature = "X1")]
+const REPETITIONS: usize = 1;
+
+#[cfg(feature = "X2")]
+const REPETITIONS: usize = 2;
+
+#[cfg(feature = "X3")]
+const REPETITIONS: usize = 4;
+
+#[cfg(feature = "X4")]
+const REPETITIONS: usize = 8;
+
 #[no_mangle]
 pub extern "C" fn main() {
-    pim::state::set_kernel(&gemv::KERNEL);
+    // #[cfg(feature = "X1")]
+    // pim::state::set_kernel(&gemv::KERNEL_X1);
+
+    // #[cfg(feature = "X2")]
+    // pim::state::set_kernel(&gemv::KERNEL_X2);
+
+    // #[cfg(feature = "X3")]
+    // pim::state::set_kernel(&gemv::KERNEL_X3);
+
+    // #[cfg(feature = "X4")]
+    pim::state::set_kernel(&gemv::KERNEL_X4);
 
     let mut matrix = SMatrix::<_, ROWS, COLUMNS>::zeros();
     matrix.fill_lower_triangle(F16x1::one(), 0);
@@ -42,17 +64,25 @@ pub extern "C" fn main() {
     // Verify everything is correctly initialized before PIM operation
     barrier::dsb(barrier::SY);
 
+    pim_os::m5op::exit(0);
+    for _ in 0..10 {
         // Execute kernel
+        {
             pim::state::set_bank_mode(BankMode::PimAllBank);
+
+            for _ in 0..REPETITIONS {
                 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();
+            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]
@@ -61,7 +91,9 @@ pub extern "C" fn main() {
                 .fold(F16x1::zero(), |acc, val| acc + *val)
         });
         core::hint::black_box(output_vector);
+    }
+    pim_os::m5op::exit(0);
 
-    writeln!(Uart0, "{output_vector}").unwrap();
+    // writeln!(Uart0, "{output_vector}").unwrap();
     writeln!(Uart0, "Done").unwrap();
 }

pim-os/src/bin/gemv_layers.rs

@@ -0,0 +1,91 @@
+#![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,
+};
+
+#[cfg(feature = "X1")]
+const REPETITIONS: usize = 1;
+
+#[cfg(feature = "X2")]
+const REPETITIONS: usize = 2;
+
+#[cfg(feature = "X3")]
+const REPETITIONS: usize = 4;
+
+#[cfg(feature = "X4")]
+const REPETITIONS: usize = 8;
+
+const ROWS: usize = 128;
+const COLUMNS: usize = 128; // Has to be 128. Assume DIMENSIONS and fill rest with zeros.
+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_X4);
+
+    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 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);
+
+    pim_os::m5op::exit(0);
+    for _ in 0..1 {
+        let interleaved_input_vector =
+            Box::new(interleaved_array::Vector::from(input_vector.clone()));
+
+        // Execute kernel
+        {
+            pim::state::set_bank_mode(BankMode::PimAllBank);
+
+            for _ in 0..(REPETITIONS * REPETITIONS) {
+                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)
+        });
+
+        core::hint::black_box(output_vector);
+    }
+    pim_os::m5op::exit(0);
+
+    writeln!(Uart0, "Done").unwrap();
+}

pim-os/src/bin/haxpy.rs

@@ -18,14 +18,35 @@ use pim_os::{
     uart::Uart0,
 };
 
+#[cfg(feature = "X1")]
 const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
 const ELEMENTS_PER_BANK: usize = 16;
 const BANKS: usize = 16;
 const BLOCKS: usize = ROWS / (ELEMENTS_PER_BANK * BANKS);
 
 #[no_mangle]
 pub extern "C" fn main() {
-    pim::state::set_kernel(&haxpy::KERNEL);
+    #[cfg(feature = "X1")]
+    pim::state::set_kernel(&haxpy::KERNEL_X1);
+
+    #[cfg(feature = "X2")]
+    pim::state::set_kernel(&haxpy::KERNEL_X2);
+
+    #[cfg(feature = "X3")]
+    pim::state::set_kernel(&haxpy::KERNEL_X3);
+
+    #[cfg(feature = "X4")]
+    pim::state::set_kernel(&haxpy::KERNEL_X4);
 
     let a = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::from_fn(|i, _| F16x1(f16::from_f32(i as _))),
@@ -37,7 +58,7 @@ pub extern "C" fn main() {
     let scalar_vector = SVector::<F16x16, 1>::from_element(F16x16([F16x1(f16::NEG_ONE); 16]));
     let interleaved_scalar_vector = Box::new(interleaved_array::Vector::from(scalar_vector));
 
-    writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
+    // writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
 
     let mut c = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::zeros(),
@@ -49,7 +70,11 @@ pub extern "C" fn main() {
     barrier::dsb(barrier::SY);
 
     // Execute kernel
+    {
         pim::state::set_bank_mode(BankMode::PimAllBank);
+
+        pim_os::m5op::exit(0);
+        for _ in 0..10 {
             haxpy::execute::<ROWS, BLOCKS>(
                 &a.0,
                 &b.0,
@@ -57,8 +82,12 @@ pub extern "C" fn main() {
                 &mut c.0,
                 dummy.as_ref(),
             );
-    pim::state::set_bank_mode(BankMode::SingleBank);
+        }
+        pim_os::m5op::exit(0);
 
-    writeln!(Uart0, "{}", c.0).unwrap();
+        pim::state::set_bank_mode(BankMode::SingleBank);
+    }
+
+    // writeln!(Uart0, "{}", c.0).unwrap();
     writeln!(Uart0, "Done").unwrap();
 }

pim-os/src/bin/vadd.rs

@@ -15,14 +15,35 @@ use pim_os::{
     uart::Uart0,
 };
 
+#[cfg(feature = "X1")]
 const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
 const ELEMENTS_PER_BANK: usize = 16;
 const BANKS: usize = 16;
 const BLOCKS: usize = ROWS / (ELEMENTS_PER_BANK * BANKS);
 
 #[no_mangle]
 pub extern "C" fn main() {
-    pim::state::set_kernel(&vadd::KERNEL);
+    #[cfg(feature = "X1")]
+    pim::state::set_kernel(&vadd::KERNEL_X1);
+
+    #[cfg(feature = "X2")]
+    pim::state::set_kernel(&vadd::KERNEL_X2);
+
+    #[cfg(feature = "X3")]
+    pim::state::set_kernel(&vadd::KERNEL_X3);
+
+    #[cfg(feature = "X4")]
+    pim::state::set_kernel(&vadd::KERNEL_X4);
 
     let a = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::from_fn(|i, _| F16x1(f16::from_f32(i as _))),
@@ -31,7 +52,7 @@ pub extern "C" fn main() {
         SVector::<F16x1, ROWS>::from_fn(|i, _| F16x1(f16::from_f32((ROWS - i) as _))),
     ));
 
-    writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
+    // writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
 
     let mut c = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::zeros(),
@@ -43,10 +64,18 @@ pub extern "C" fn main() {
     barrier::dsb(barrier::SY);
 
     // Execute kernel
+    {
         pim::state::set_bank_mode(BankMode::PimAllBank);
-    vadd::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
-    pim::state::set_bank_mode(BankMode::SingleBank);
 
-    writeln!(Uart0, "{}", c.0).unwrap();
+        pim_os::m5op::exit(0);
+        for _ in 0..10 {
+            vadd::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
+        }
+        pim_os::m5op::exit(0);
+
+        pim::state::set_bank_mode(BankMode::SingleBank);
+    }
+
+    // writeln!(Uart0, "{}", c.0).unwrap();
     writeln!(Uart0, "Done").unwrap();
 }

pim-os/src/bin/vmul.rs

@@ -15,14 +15,35 @@ use pim_os::{
     uart::Uart0,
 };
 
+#[cfg(feature = "X1")]
 const ROWS: usize = 256;
+
+#[cfg(feature = "X2")]
+const ROWS: usize = 512;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 1024;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 2048;
+
 const ELEMENTS_PER_BANK: usize = 16;
 const BANKS: usize = 16;
 const BLOCKS: usize = ROWS / (ELEMENTS_PER_BANK * BANKS);
 
 #[no_mangle]
 pub extern "C" fn main() {
-    pim::state::set_kernel(&vmul::KERNEL);
+    #[cfg(feature = "X1")]
+    pim::state::set_kernel(&vmul::KERNEL_X1);
+
+    #[cfg(feature = "X2")]
+    pim::state::set_kernel(&vmul::KERNEL_X2);
+
+    #[cfg(feature = "X3")]
+    pim::state::set_kernel(&vmul::KERNEL_X3);
+
+    #[cfg(feature = "X4")]
+    pim::state::set_kernel(&vmul::KERNEL_X4);
 
     let a = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::from_fn(|_, _| F16x1(f16::from_f32(2 as _))),
@@ -31,7 +52,7 @@ pub extern "C" fn main() {
         SVector::<F16x1, ROWS>::from_fn(|_, _| F16x1(f16::from_f32(3 as _))),
     ));
 
-    writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
+    // writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
 
     let mut c = Box::new(pim::continuous_array::Vector(
         SVector::<F16x1, ROWS>::zeros(),
@@ -43,10 +64,18 @@ pub extern "C" fn main() {
     barrier::dsb(barrier::SY);
 
     // Execute kernel
+    {
         pim::state::set_bank_mode(BankMode::PimAllBank);
-    vmul::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
-    pim::state::set_bank_mode(BankMode::SingleBank);
 
-    writeln!(Uart0, "{}", c.0).unwrap();
+        pim_os::m5op::exit(0);
+        for _ in 0..10 {
+            vmul::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
+        }
+        pim_os::m5op::exit(0);
+
+        pim::state::set_bank_mode(BankMode::SingleBank);
+    }
+
+    // writeln!(Uart0, "{}", c.0).unwrap();
     writeln!(Uart0, "Done").unwrap();
 }

pim-os/src/kernel/gemv.rs

@@ -5,7 +5,7 @@ use aarch64_cpu::asm::barrier;
 use nalgebra::SVector;
 use pim_isa::{File, Instruction, Kernel};
 
-pub const KERNEL: Kernel = Kernel([
+pub const KERNEL_X1: Kernel = Kernel([
     Instruction::MOV {
         src: File::Bank,
         dst: File::GrfA { index: 0 },
@@ -76,6 +76,252 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
+pub const KERNEL_X2: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::MAC {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        src2: File::GrfB { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: true,
+    },
+    Instruction::JUMP {
+        offset: -1,
+        count: 15,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X3: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::MAC {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        src2: File::GrfB { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: true,
+    },
+    Instruction::JUMP {
+        offset: -1,
+        count: 31,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X4: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::MAC {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        src2: File::GrfB { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: true,
+    },
+    Instruction::JUMP {
+        offset: -1,
+        count: 63,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 4 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 5 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 6 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 7 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
 pub fn execute<const X16R: usize, const R: usize>(
     matrix: &Matrix<X16R, 8>,
     input_vector: &interleaved_array::Vector<8>,

pim-os/src/kernel/haxpy.rs

@@ -2,7 +2,7 @@ use crate::pim::{interleaved_array, operation::PimOperand, vector::F16x1};
 use nalgebra::SVector;
 use pim_isa::{File, Instruction, Kernel};
 
-pub const KERNEL: Kernel = Kernel([
+pub const KERNEL_X1: Kernel = Kernel([
     Instruction::MOV {
         src: File::Bank,
         dst: File::SrfM { index: 0 },
@@ -52,6 +52,288 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
+pub const KERNEL_X2: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::SrfM { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 0 },
+        src2: File::GrfA { index: 0 },
+        dst: File::GrfA { index: 0 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 1 },
+        src2: File::GrfA { index: 1 },
+        dst: File::GrfA { index: 1 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X3: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::SrfM { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 0 },
+        src2: File::GrfA { index: 0 },
+        dst: File::GrfA { index: 0 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 1 },
+        src2: File::GrfA { index: 1 },
+        dst: File::GrfA { index: 1 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 2 },
+        src2: File::GrfA { index: 2 },
+        dst: File::GrfA { index: 2 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 3 },
+        src2: File::GrfA { index: 3 },
+        dst: File::GrfA { index: 3 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X4: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::SrfM { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 0 },
+        src2: File::GrfA { index: 0 },
+        dst: File::GrfA { index: 0 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 1 },
+        src2: File::GrfA { index: 1 },
+        dst: File::GrfA { index: 1 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 2 },
+        src2: File::GrfA { index: 2 },
+        dst: File::GrfA { index: 2 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 3 },
+        src2: File::GrfA { index: 3 },
+        dst: File::GrfA { index: 3 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 4 },
+        src2: File::GrfA { index: 4 },
+        dst: File::GrfA { index: 4 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 5 },
+        src2: File::GrfA { index: 5 },
+        dst: File::GrfA { index: 5 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 6 },
+        src2: File::GrfA { index: 6 },
+        dst: File::GrfA { index: 6 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::SrfA { index: 7 },
+        src2: File::GrfA { index: 7 },
+        dst: File::GrfA { index: 7 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 4 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 5 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 6 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfA { index: 7 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
 pub fn execute<const R: usize, const BLOCKS: usize>(
     a: &SVector<F16x1, R>,
     b: &SVector<F16x1, R>,

pim-os/src/kernel/vadd.rs

@@ -2,7 +2,7 @@ use crate::pim::{operation::PimOperand, vector::F16x1};
 use nalgebra::SVector;
 use pim_isa::{File, Instruction, Kernel};
 
-pub const KERNEL: Kernel = Kernel([
+pub const KERNEL_X1: Kernel = Kernel([
     Instruction::MOV {
         src: File::Bank,
         dst: File::GrfA { index: 0 },
@@ -48,6 +48,265 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
+pub const KERNEL_X2: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X3: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 2 },
+        dst: File::GrfB { index: 2 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 3 },
+        dst: File::GrfB { index: 3 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X4: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 2 },
+        dst: File::GrfB { index: 2 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 3 },
+        dst: File::GrfB { index: 3 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 4 },
+        dst: File::GrfB { index: 4 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 5 },
+        dst: File::GrfB { index: 5 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 6 },
+        dst: File::GrfB { index: 6 },
+        aam: false,
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 7 },
+        dst: File::GrfB { index: 7 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 4 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 5 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 6 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 7 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
 pub fn execute<const R: usize, const BLOCKS: usize>(
     a: &SVector<F16x1, R>,
     b: &SVector<F16x1, R>,

pim-os/src/kernel/vmul.rs

@@ -2,7 +2,7 @@ use crate::pim::{operation::PimOperand, vector::F16x1};
 use nalgebra::SVector;
 use pim_isa::{File, Instruction, Kernel};
 
-pub const KERNEL: Kernel = Kernel([
+pub const KERNEL_X1: Kernel = Kernel([
     Instruction::MOV {
         src: File::Bank,
         dst: File::GrfA { index: 0 },
@@ -48,6 +48,265 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
+pub const KERNEL_X2: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X3: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 2 },
+        dst: File::GrfB { index: 2 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 3 },
+        dst: File::GrfB { index: 3 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
+pub const KERNEL_X4: Kernel = Kernel([
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 1 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 2 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 3 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 4 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 5 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 6 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 7 },
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 2 },
+        dst: File::GrfB { index: 2 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 3 },
+        dst: File::GrfB { index: 3 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 4 },
+        dst: File::GrfB { index: 4 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 5 },
+        dst: File::GrfB { index: 5 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 6 },
+        dst: File::GrfB { index: 6 },
+        aam: false,
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 7 },
+        dst: File::GrfB { index: 7 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 1 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 2 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 3 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 4 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 5 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 6 },
+        dst: File::Bank,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 7 },
+        dst: File::Bank,
+    },
+    Instruction::EXIT,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+]);
+
 pub fn execute<const R: usize, const BLOCKS: usize>(
     a: &SVector<F16x1, R>,
     b: &SVector<F16x1, R>,