Skip to content

Add packet-switched flash attention programming example #1532

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

Draft
erwei-xilinx wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Add packet-switched flash attention programming example #1532

Show file tree
Hide file tree
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
101 changes: 101 additions & 0 deletions programming_examples/flash_attention/packet_switched/Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,101 @@
# Copyright (C) 2025, Advanced Micro Devices, Inc.
# SPDX-License-Identifier: MIT
#
# Packet-switched flash attention example.
# Uses dma_packet channels for Q and K routing through shared
# compute tile S2MM DMA channels.
# Supports both NPU2 (AIE2P) and NPU1 (AIE2).
#
srcdir := $(shell dirname $(realpath $(firstword $(MAKEFILE_LIST))))

# Attention parameters
LK ?= 512
LKP ?= 64
LQ ?= 512
LQP ?= 256
DK ?= 64
DV ?= 64
NUM_HEADS ?= 2
NUM_KV_HEADS ?= $(NUM_HEADS)
VAL_RANGE ?= 3

# Derived: kernel tile size = LQP / num_q_tiles (4)
NUM_Q_TILES ?= 4
LQP_TILE := $(shell echo $$(($(LQP) / $(NUM_Q_TILES))))

# Determine build dir based on whether PEANO_INSTALL_DIR is set
ifdef PEANO_INSTALL_DIR
BUILD_DIR := build_peano
else
BUILD_DIR := build_chess
endif

AIEOPT_DIR = $(shell realpath $(dir $(shell which aie-opt))/..)
WARNING_FLAGS = -Wno-parentheses -Wno-attributes -Wno-macro-redefined -Wno-empty-body
PEANOWRAP2P_FLAGS = -O2 -std=c++20 --target=aie2p-none-unknown-elf ${WARNING_FLAGS} -DNDEBUG -I ${AIEOPT_DIR}/include
PEANOWRAP2_FLAGS = -O2 -std=c++20 --target=aie2-none-unknown-elf ${WARNING_FLAGS} -DNDEBUG -I ${AIEOPT_DIR}/include

# ============================================================================
# NPU2 (AIE2P) targets — default
# ============================================================================

all: run

print:
${powershell} python3 ${srcdir}/attn.py -p --lk $(LK) --lkp $(LKP) --lq $(LQ) --lqp $(LQP) --dk $(DK) --dv $(DV) --num-heads $(NUM_HEADS) --num-kv-heads $(NUM_KV_HEADS)

run: compile-kernel
mkdir -p $(BUILD_DIR)
PEANO_INSTALL_DIR=$(PEANO_INSTALL_DIR) cd $(BUILD_DIR) && ${powershell} python3 ${srcdir}/attn.py --lk $(LK) --lkp $(LKP) --lq $(LQ) --lqp $(LQP) --dk $(DK) --dv $(DV) --num-heads $(NUM_HEADS) --num-kv-heads $(NUM_KV_HEADS) --val-range $(VAL_RANGE) $(EXTRA_PY_FLAGS)

compile-kernel:
mkdir -p $(BUILD_DIR)
@if [ -n "$(PEANO_INSTALL_DIR)" ]; then \
echo "Detected PEANO_INSTALL_DIR from environment: $(PEANO_INSTALL_DIR)"; \
if [ -x "$(PEANO_INSTALL_DIR)/bin/clang++" ]; then \
echo "Using clang++ from PEANO_INSTALL_DIR=$(PEANO_INSTALL_DIR)"; \
$(PEANO_INSTALL_DIR)/bin/clang++ ${PEANOWRAP2P_FLAGS} -DBIT_WIDTH=8 -c ${srcdir}/attn_pkt.cc -o $(BUILD_DIR)/attn_pkt.o -Dlqp=$(LQP_TILE) -Dlkp=$(LKP) -Ddk=$(DK) -Ddv=$(DV) -DAIE_API_EMULATE_BFLOAT16_MMUL_WITH_BFP16 -DROUND_CONV_EVEN $(EXTRA_KERNEL_FLAGS); \
else \
echo "Error: invalid PEANO_INSTALL_DIR, clang++ not found."; \
exit 1; \
fi; \
elif command -v xchesscc_wrapper >/dev/null 2>&1; then \
echo "Using xchesscc_wrapper from PATH"; \
cd $(BUILD_DIR) && ${powershell} xchesscc_wrapper aie2p -c ${srcdir}/attn_pkt.cc -o attn_pkt.o -Dlqp=$(LQP_TILE) -Dlkp=$(LKP) -Ddk=$(DK) -Ddv=$(DV); \
else \
echo "Error: Neither PEANO_INSTALL_DIR nor xchesscc_wrapper found."; \
exit 1; \
fi

# ============================================================================
# NPU1 (AIE2) targets
# ============================================================================

print-npu1:
${powershell} python3 ${srcdir}/attn_npu1.py -p --lk $(LK) --lkp $(LKP) --lq $(LQ) --lqp $(LQP) --dk $(DK) --dv $(DV) --num-heads $(NUM_HEADS) --num-kv-heads $(NUM_KV_HEADS)

run-npu1: compile-kernel-npu1
mkdir -p $(BUILD_DIR)
PEANO_INSTALL_DIR=$(PEANO_INSTALL_DIR) cd $(BUILD_DIR) && ${powershell} python3 ${srcdir}/attn_npu1.py --lk $(LK) --lkp $(LKP) --lq $(LQ) --lqp $(LQP) --dk $(DK) --dv $(DV) --num-heads $(NUM_HEADS) --num-kv-heads $(NUM_KV_HEADS) --val-range $(VAL_RANGE) $(EXTRA_PY_FLAGS)

compile-kernel-npu1:
mkdir -p $(BUILD_DIR)
@if [ -n "$(PEANO_INSTALL_DIR)" ]; then \
echo "Detected PEANO_INSTALL_DIR from environment: $(PEANO_INSTALL_DIR)"; \
if [ -x "$(PEANO_INSTALL_DIR)/bin/clang++" ]; then \
echo "Using clang++ from PEANO_INSTALL_DIR=$(PEANO_INSTALL_DIR) for AIE2 (NPU1)"; \
$(PEANO_INSTALL_DIR)/bin/clang++ ${PEANOWRAP2_FLAGS} -DBIT_WIDTH=8 -c ${srcdir}/attn_npu1.cc -o $(BUILD_DIR)/attn_npu1.o -I${srcdir}/../dataflow_based -Dlqp=$(LQP_TILE) -Dlkp=$(LKP) -Ddk=$(LKP) -Ddk_full=$(DK) -Ddv=$(LKP) -Ddv_full=$(DV) $(EXTRA_KERNEL_FLAGS); \
else \
echo "Error: invalid PEANO_INSTALL_DIR, clang++ not found."; \
exit 1; \
fi; \
elif command -v xchesscc_wrapper >/dev/null 2>&1; then \
echo "Using xchesscc_wrapper from PATH for AIE2 (NPU1)"; \
cd $(BUILD_DIR) && ${powershell} xchesscc_wrapper aie2 -c ${srcdir}/attn_npu1.cc -o attn_npu1.o -I${srcdir}/../dataflow_based -Dlqp=$(LQP_TILE) -Dlkp=$(LKP) -Ddk=$(LKP) -Ddk_full=$(DK) -Ddv=$(LKP) -Ddv_full=$(DV); \
else \
echo "Error: Neither PEANO_INSTALL_DIR nor xchesscc_wrapper found."; \
exit 1; \
fi

clean:
rm -rf $(BUILD_DIR) __pycache__
Loading
Loading