Introduce X1, X2, X3, X4 complexity levels for all workloads

pim-os/src/bin/classic_gemv.rs

@@ -7,7 +7,18 @@ use core::fmt::Write;
 use nalgebra::{SMatrix, SVector};
 use pim_os::{pim::vector::F16x1, uart::Uart0};
 
+#[cfg(feature = "X1")]
+const ROWS: usize = 16;
+
+#[cfg(feature = "X2")]
 const ROWS: usize = 32;
+
+#[cfg(feature = "X3")]
+const ROWS: usize = 64;
+
+#[cfg(feature = "X4")]
+const ROWS: usize = 128;
+
 const COLUMNS: usize = 128;
 
 #[no_mangle]

pim-os/src/bin/classic_haxpy.rs

@@ -8,6 +8,16 @@ 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]

pim-os/src/bin/classic_vadd.rs

@@ -7,6 +7,16 @@ use core::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]

pim-os/src/bin/classic_vmul.rs

@@ -7,6 +7,16 @@ use core::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]

pim-os/src/bin/haxpy.rs

@@ -18,11 +18,35 @@ use pim_os::{
     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 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 _))),
@@ -49,7 +73,7 @@ pub extern "C" fn main() {
     {
         pim::state::set_bank_mode(BankMode::PimAllBank);
 
-        haxpy::execute(
+        haxpy::execute::<ROWS, BLOCKS>(
             &a.0,
             &b.0,
             &interleaved_scalar_vector,

pim-os/src/bin/vadd.rs

@@ -15,11 +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 _))),
@@ -43,7 +67,7 @@ pub extern "C" fn main() {
     {
         pim::state::set_bank_mode(BankMode::PimAllBank);
 
-        vadd::execute(&a.0, &b.0, &mut c.0, dummy.as_ref());
+        vadd::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
 
         pim::state::set_bank_mode(BankMode::SingleBank);
     }

pim-os/src/bin/vmul.rs

@@ -15,11 +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 _))),
@@ -43,7 +67,7 @@ pub extern "C" fn main() {
     {
         pim::state::set_bank_mode(BankMode::PimAllBank);
 
-        vmul::execute(&a.0, &b.0, &mut c.0, dummy.as_ref());
+        vmul::execute::<ROWS, BLOCKS>(&a.0, &b.0, &mut c.0, dummy.as_ref());
 
         pim::state::set_bank_mode(BankMode::SingleBank);
     }

pim-os/src/kernel/gemv.rs

@@ -47,7 +47,7 @@ pub const KERNEL_X1: Kernel = Kernel([
     },
     Instruction::JUMP {
         offset: -1,
-        count: 15,
+        count: 7,
     },
     Instruction::FILL {
         src: File::GrfB { index: 0 },

pim-os/src/kernel/haxpy.rs

@@ -2,7 +2,119 @@ 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 },
+    },
+    Instruction::MOV {
+        src: File::Bank,
+        dst: File::GrfA { index: 0 },
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        src2: File::SrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        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,
+    Instruction::NOP,
+    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::GrfA { index: 0 },
+        src2: File::SrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        src2: File::SrfA { 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,
+]);
+
+pub const KERNEL_X3: Kernel = Kernel([
     Instruction::MOV {
         src: File::Bank,
         dst: File::SrfM { index: 0 },
@@ -23,51 +135,31 @@ pub const KERNEL: Kernel = Kernel([
         src: File::Bank,
         dst: File::GrfA { index: 3 },
     },
-    Instruction::MUL {
+    Instruction::MAD {
-        src0: File::SrfA { index: 0 },
-        src1: File::GrfA { index: 0 },
-        dst: File::GrfA { index: 0 },
-        aam: false,
-    },
-    Instruction::ADD {
         src0: File::Bank,
         src1: File::GrfA { index: 0 },
+        src2: File::SrfA { index: 0 },
         dst: File::GrfB { index: 0 },
         aam: false,
     },
-    Instruction::MUL {
+    Instruction::MAD {
-        src0: File::SrfA { index: 1 },
-        src1: File::GrfA { index: 1 },
-        dst: File::GrfA { index: 1 },
-        aam: false,
-    },
-    Instruction::ADD {
         src0: File::Bank,
         src1: File::GrfA { index: 1 },
+        src2: File::SrfA { index: 1 },
         dst: File::GrfB { index: 1 },
         aam: false,
     },
-    Instruction::MUL {
+    Instruction::MAD {
-        src0: File::SrfA { index: 2 },
-        src1: File::GrfA { index: 2 },
-        dst: File::GrfA { index: 2 },
-        aam: false,
-    },
-    Instruction::ADD {
         src0: File::Bank,
         src1: File::GrfA { index: 2 },
+        src2: File::SrfA { index: 2 },
         dst: File::GrfB { index: 2 },
         aam: false,
     },
-    Instruction::MUL {
+    Instruction::MAD {
-        src0: File::SrfA { index: 3 },
-        src1: File::GrfA { index: 3 },
-        dst: File::GrfA { index: 3 },
-        aam: false,
-    },
-    Instruction::ADD {
         src0: File::Bank,
         src1: File::GrfA { index: 3 },
+        src2: File::SrfA { index: 3 },
         dst: File::GrfB { index: 3 },
         aam: false,
     },
@@ -102,9 +194,147 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
     Instruction::NOP,
     Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
+    Instruction::NOP,
 ]);
 
-pub fn execute<const R: usize>(
+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::GrfA { index: 0 },
+        src2: File::SrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 1 },
+        src2: File::SrfA { index: 1 },
+        dst: File::GrfB { index: 1 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 2 },
+        src2: File::SrfA { index: 2 },
+        dst: File::GrfB { index: 2 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 3 },
+        src2: File::SrfA { index: 3 },
+        dst: File::GrfB { index: 3 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 4 },
+        src2: File::SrfA { index: 4 },
+        dst: File::GrfB { index: 4 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 5 },
+        src2: File::SrfA { index: 5 },
+        dst: File::GrfB { index: 5 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 6 },
+        src2: File::SrfA { index: 6 },
+        dst: File::GrfB { index: 6 },
+        aam: false,
+    },
+    Instruction::MAD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 7 },
+        src2: File::SrfA { 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,
+]);
+
+pub fn execute<const R: usize, const BLOCKS: usize>(
     a: &SVector<F16x1, R>,
     b: &SVector<F16x1, R>,
     interleaved_scalar: &interleaved_array::Vector<1>,
@@ -113,18 +343,15 @@ pub fn execute<const R: usize>(
 ) {
     interleaved_scalar.execute_read();
 
-    a.fixed_rows_with_step::<4>(0, 16 * 16)
+    a.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
         .for_each(|entry| entry.execute_read());
 
-    b.fixed_rows_with_step::<4>(0, 16 * 16)
+    b.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
-        .for_each(|entry| {
+        .for_each(|entry| entry.execute_read());
-            dummy.execute_read();
-            entry.execute_read();
-        });
 
-    c.fixed_rows_with_step_mut::<4>(0, 16 * 16)
+    c.fixed_rows_with_step_mut::<BLOCKS>(0, 256)
         .iter_mut()
         .for_each(|entry| entry.execute_write());
 

pim-os/src/kernel/vadd.rs

@@ -2,7 +2,189 @@ 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 },
+    },
+    Instruction::ADD {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        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,
+    Instruction::NOP,
+    Instruction::NOP,
+    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 },
@@ -125,19 +307,19 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
-pub fn execute<const R: usize>(
+pub fn execute<const R: usize, const BLOCKS: 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)
+    a.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
         .for_each(|entry| entry.execute_read());
-    b.fixed_rows_with_step::<8>(0, 16 * 16)
+    b.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
         .for_each(|entry| entry.execute_read());
-    c.fixed_rows_with_step_mut::<8>(0, 16 * 16)
+    c.fixed_rows_with_step_mut::<BLOCKS>(0, 256)
         .iter_mut()
         .for_each(|entry| entry.execute_write());
 

pim-os/src/kernel/vmul.rs

@@ -2,7 +2,189 @@ 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 },
+    },
+    Instruction::MUL {
+        src0: File::Bank,
+        src1: File::GrfA { index: 0 },
+        dst: File::GrfB { index: 0 },
+        aam: false,
+    },
+    Instruction::FILL {
+        src: File::GrfB { index: 0 },
+        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,
+    Instruction::NOP,
+    Instruction::NOP,
+    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 },
@@ -125,19 +307,19 @@ pub const KERNEL: Kernel = Kernel([
     Instruction::NOP,
 ]);
 
-pub fn execute<const R: usize>(
+pub fn execute<const R: usize, const BLOCKS: 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)
+    a.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
         .for_each(|entry| entry.execute_read());
-    b.fixed_rows_with_step::<8>(0, 16 * 16)
+    b.fixed_rows_with_step::<BLOCKS>(0, 256)
         .iter()
         .for_each(|entry| entry.execute_read());
-    c.fixed_rows_with_step_mut::<8>(0, 16 * 16)
+    c.fixed_rows_with_step_mut::<BLOCKS>(0, 256)
         .iter_mut()
         .for_each(|entry| entry.execute_write());