From a5ffaca3e3208a580153bac121443bd33ee95f85 Mon Sep 17 00:00:00 2001
From: Yaman Umuroglu <maltanar@gmail.com>
Date: Wed, 2 Nov 2022 16:04:00 +0100
Subject: [PATCH] [FIFO] support FIFO splitting for pow2 size optimizations
---
.../fpgadataflow/set_fifo_depths.py | 99 +++++++++++++------
1 file changed, 69 insertions(+), 30 deletions(-)
diff --git a/src/finn/transformation/fpgadataflow/set_fifo_depths.py b/src/finn/transformation/fpgadataflow/set_fifo_depths.py
index dccd97020..9b882eeac 100644
--- a/src/finn/transformation/fpgadataflow/set_fifo_depths.py
+++ b/src/finn/transformation/fpgadataflow/set_fifo_depths.py
@@ -39,7 +39,6 @@ from qonnx.transformation.general import (
GiveUniqueNodeNames,
SortGraph,
)
-from qonnx.util.basic import get_by_name
from finn.analysis.fpgadataflow.dataflow_performance import dataflow_performance
from finn.transformation.fpgadataflow.annotate_cycles import AnnotateCycles
@@ -427,32 +426,80 @@ class SplitLargeFifos(Transformation):
"""Split FIFOs with a depth larger than 32768 into smaller ones
to ensure that they can be correctly generated."""
+ def __init__(
+ self,
+ max_qsrl_depth=256,
+ ):
+ super().__init__()
+ self.max_qsrl_depth = max_qsrl_depth
+
+ def get_split_configs(self, depth):
+ max_size = 32768
+
+ def floor_pow2(x):
+ if (x & (x - 1) == 0) and x != 0:
+ return x
+ else:
+ return 1 << ((x - 1).bit_length() - 1)
+
+ ret = []
+ # trivial case: for small FIFOs, return as-is with rtl style
+ if depth <= self.max_qsrl_depth:
+ return [(depth, "rtl")]
+ # first pass: ensure max depth of 32k is respected
+ # (restricted by Vivado AXIS infra IP)
+
+ remainder = depth
+ while remainder != 0:
+ if remainder > max_size:
+ ret.append(max_size)
+ remainder -= max_size
+ else:
+ ret.append(remainder)
+ remainder = 0
+ # second pass: break non-power-of-2 sized FIFOs
+ # into several ones
+ ret_pass2 = []
+
+ for cand_depth in ret:
+ cand_floor_pow2 = floor_pow2(cand_depth)
+ ret_pass2.append(cand_floor_pow2)
+ if cand_floor_pow2 < cand_depth:
+ ret_pass2.append(cand_depth - cand_floor_pow2)
+ # finally, add impl_style to each split FIFO
+ ret_final = []
+ for cand_depth in ret_pass2:
+ if cand_depth <= self.max_qsrl_depth:
+ ret_final.append((cand_depth, "rtl"))
+ else:
+ ret_final.append((cand_depth, "vivado"))
+
+ return ret_final
+
def apply(self, model):
graph = model.graph
node_ind = 0
graph_modified = False
- for n in graph.node:
+ for node in graph.node:
node_ind += 1
- if n.op_type == "StreamingFIFO":
- depth = get_by_name(n.attribute, "depth")
- if depth.i > 32768:
- n0 = getCustomOp(n)
- fld_shape = n0.get_folded_output_shape()
- dtype = n0.get_nodeattr("dataType")
- impl_style = n0.get_nodeattr("impl_style")
- ram_style = n0.get_nodeattr("ram_style")
- shape = model.get_tensor_shape(n.input[0])
- split_n = math.ceil(depth.i / 32768)
- fifo_depth = math.ceil(depth.i / split_n)
- for i in range(split_n):
+ if node.op_type == "StreamingFIFO":
+ n_inst = getCustomOp(node)
+ depth = n_inst.get_nodeattr("depth")
+ cfgs = self.get_split_configs(depth)
+ if len(cfgs) > 1:
+ fld_shape = n_inst.get_folded_output_shape()
+ dtype = n_inst.get_nodeattr("dataType")
+ ram_style = n_inst.get_nodeattr("ram_style")
+ shape = model.get_tensor_shape(node.input[0])
+ for i, (fifo_depth, impl_style) in enumerate(cfgs):
if i == 0:
- inp = n.input[0]
+ inp = node.input[0]
else:
- inp = n.name + "_" + str(i - 1) + "_out"
- if i == split_n - 1:
- outp = n.output[0]
+ inp = node.name + "_" + str(i - 1) + "_out"
+ if i == len(cfgs) - 1:
+ outp = node.output[0]
else:
- outp = n.name + "_" + str(i) + "_out"
+ outp = node.name + "_" + str(i) + "_out"
out_tensor = helper.make_tensor_value_info(
outp, TensorProto.FLOAT, shape
)
@@ -469,21 +516,13 @@ class SplitLargeFifos(Transformation):
dataType=dtype,
impl_style=impl_style,
ram_style=ram_style,
+ name=node.name + "_" + str(i),
)
graph.node.insert(node_ind + i, fifo_node)
- graph.node.remove(n)
- if n.output[0] != "global_out":
- consumer = model.find_consumer(n.output[0])
- n1 = getCustomOp(consumer)
- n1.set_nodeattr("outFIFODepth", fifo_depth)
- if n.input[0] != "global_in":
- producer = model.find_producer(n.input[0])
- n2 = getCustomOp(producer)
- n2.set_nodeattr("inFIFODepth", fifo_depth)
+ graph.node.remove(node)
graph_modified = True
if graph_modified:
model = model.transform(SortGraph())
- model = model.transform(GiveUniqueNodeNames())
model = model.transform(GiveReadableTensorNames())
- return (model, graph_modified)
+ return (model, False)
--
GitLab