Add Vector Add Kernel

pim-os/src/bin/vadd.rs

@@ -6,36 +6,33 @@ extern crate alloc;
 use aarch64_cpu::asm::barrier;
 use alloc::boxed::Box;
 use core::fmt::Write;
-use nalgebra::{SMatrix, SVector};
+use half::f16;
-use num_traits::{One, Zero};
+use nalgebra::SVector;
 use pim_isa::BankMode;
 use pim_os::{
-    kernel::gemv,
+    kernel::vadd,
-    pim::{
+    pim::{self, vector::F16x1},
-        self, interleaved_array,
-        vector::{F16x1, F16x16},
-    },
     uart::Uart0,
 };
 
-const ROWS: usize = 32;
+const ROWS: usize = 2048;
-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);
+    pim::state::set_kernel(&vadd::KERNEL);
 
-    let mut matrix = SMatrix::<_, ROWS, COLUMNS>::zeros();
+    let a = Box::new(pim::continuous_array::Vector(
-    matrix.fill_lower_triangle(F16x1::one(), 0);
+        SVector::<F16x1, ROWS>::from_fn(|i, _| F16x1(f16::from_f32(i as _))),
+    ));
+    let b = Box::new(pim::continuous_array::Vector(
+        SVector::<F16x1, ROWS>::from_fn(|i, _| F16x1(f16::from_f32((ROWS - i) as _))),
+    ));
 
-    let pim_matrix = Box::new(pim::continuous_array::Matrix::<X16_ROWS, X16_COLUMNS>::from(matrix));
+    // writeln!(Uart0, "{}+{}=", a.0, b.0).unwrap();
 
-    let input_vector = SVector::<_, X16_COLUMNS>::from_element(F16x16::one());
+    let mut c = Box::new(pim::continuous_array::Vector(
-    let interleaved_input_vector = Box::new(interleaved_array::Vector::from(input_vector));
+        SVector::<F16x1, ROWS>::zeros(),
-
+    ));
-    let mut output_partial_sum_vector = Box::new(SVector::<F16x16, ROWS>::zeros());
 
     let dummy = Box::new(0);
 
@@ -46,24 +43,11 @@ pub extern "C" fn main() {
     {
         pim::state::set_bank_mode(BankMode::PimAllBank);
 
-        gemv::execute(
+        vadd::execute(&a.0, &b.0, &mut c.0, dummy.as_ref());
-            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();
+    // writeln!(Uart0, "{}", c.0).unwrap();
-
+    writeln!(Uart0, "Done").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

@@ -1,2 +1,3 @@
 pub mod legacy;
 pub mod gemv;
+pub mod vadd;

pim-os/src/kernel/vadd.rs

@@ -0,0 +1,145 @@
+use crate::pim::{operation::PimOperand, vector::F16x1};
+use nalgebra::SVector;
+use pim_isa::{File, Instruction, Kernel};
+
+pub const KERNEL: 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>(
+    a: &SVector<F16x1, R>,
+    b: &SVector<F16x1, R>,
+    c: &mut SVector<F16x1, R>,
+    dummy: &impl PimOperand,
+) {
+    a.fixed_rows_with_step::<8>(0, 16 * 16)
+        .iter()
+        .for_each(|entry| entry.execute_read());
+    b.fixed_rows_with_step::<8>(0, 16 * 16)
+        .iter()
+        .for_each(|entry| entry.execute_read());
+    c.fixed_rows_with_step_mut::<8>(0, 16 * 16)
+        .iter_mut()
+        .for_each(|entry| entry.execute_write());
+
+    dummy.execute_read();
+}