derek/pim-rs
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Layout matrix correctly for additional rows

Browse Source
This commit is contained in:
Derek Christ
2024-02-21 16:45:03 +01:00
parent cf83e27f50
commit f04ee8e603
6 changed files with 73 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
pim-os/src/pim/kernel/samsung_matrix_vector_mul.rs
View File

@@ -46,12 +46,16 @@ pub const KERNEL: Kernel = Kernel([
},
Instruction::JUMP {
offset: -1,
count: 7,
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,
@@ -72,12 +76,11 @@ pub const KERNEL: Kernel = Kernel([
Instruction::NOP,
Instruction::NOP,
Instruction::NOP,
Instruction::NOP,
]);
pub fn execute<const R: usize, const C: usize>(
matrix: &Matrix<R, C>,
input_vector: &interleaved_array::Vector<C>,
pub fn execute<const X16R: usize, const R: usize, const X16C: usize>(
matrix: &Matrix<X16R, X16C>,
input_vector: &interleaved_array::Vector<X16C>,
output_partial_sum_vector: &mut SVector<F16x16, R>,
dummy: &impl PimOperand,
) {
@@ -85,11 +88,18 @@ pub fn execute<const R: usize, const C: usize>(
block.execute_read();
}
for column_block in matrix.0.fixed_rows::<1>(0).iter() {
column_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();
}
}
output_partial_sum_vector.execute_write();
for chunk in output_partial_sum_vector
.fixed_rows_with_step_mut::<X16R>(0, 16)
.iter_mut()
{
chunk.execute_write();
}
dummy.execute_read();
}