Skip to content

fix(projection_asm): correct single-pass matmul emission #62

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
booth-algo merged 1 commit into from
Jul 2, 2026
Merged

fix(projection_asm): correct single-pass matmul emission #62

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 3 additions & 4 deletions asm_templates/projection_asm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,7 +198,9 @@ def projection_asm(
)
lines.append(f"S_ADDI_INT gp{w_actual_register}, gp0, 0 ")
else:
lines.append(f"S_ADDI_INT gp{w_actual_register}, gp0, {(weight_row % (mlen // blen)) * blen} ")
# weight rows are 1 output-feature each in matrix SRAM (row-granular), so the
# within-group offset must be in rows: (wr % (mlen//blen)) * blen * mlen.
lines.append(f"S_ADDI_INT gp{w_actual_register}, gp0, {(weight_row % (mlen // blen)) * blen * mlen} ")
lines.append(
f"S_ADDI_INT gp{intermediate_register}, gp{result_reg}, {(weight_row % (mlen // blen)) * blen} "
)
Expand All @@ -210,13 +212,10 @@ def projection_asm(
lines.append(f"M_MM 0, gp{w_temp_register}, gp{act_reg} ")
lines.append(f"S_ADDI_INT gp{w_temp_register}, gp{w_temp_register}, {mlen * mlen} ")
lines.append(f"S_ADDI_INT gp{act_reg}, gp{act_reg}, {mlen * batch} ")
break
lines.append(f"M_MM_WO gp{intermediate_register}, gp0, 0 ")
lines.append(f"S_ADDI_INT gp{intermediate_register}, gp{intermediate_register}, {blen * mlen} ")
break
if (weight_row + 1) % (mlen // blen) == 0 and weight_row != out_features // blen - 1:
lines.append(f"S_ADDI_INT gp{result_reg}, gp{result_reg}, {mlen * batch} ")
break
else:
# K-split path
chunks = _proj_k_chunks(num_k_tiles, MAX_K_TILES)
Expand Down
Loading