From ce4d70c9cb0bf982e4dbbda61fb522be8a3f415b Mon Sep 17 00:00:00 2001
From: Yaman Umuroglu <maltanar@gmail.com>
Date: Mon, 10 Feb 2020 22:43:29 +0100
Subject: [PATCH] [StreamingFC] fix pragmas for array partitioning
---
.../fpgadataflow/streamingfclayer_batch.py | 23 +++++++++++++------
1 file changed, 16 insertions(+), 7 deletions(-)
diff --git a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
index a42397414..33a5d963d 100644
--- a/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
+++ b/src/finn/custom_op/fpgadataflow/streamingfclayer_batch.py
@@ -628,18 +628,27 @@ class StreamingFCLayer_Batch(HLSCustomOp):
self.code_gen_dict["$PRAGMAS$"].append(
"#pragma HLS INTERFACE ap_ctrl_none port=return"
)
+ # the weight tensor is ap_uint<simd*prec> [PE][WMEM]
+ # partition for parallel access along the PE dimension (dim 1)
self.code_gen_dict["$PRAGMAS$"].append(
- "DO_PRAGMA(HLS ARRAY_PARTITION variable=weights complete dim=1)"
- )
-
- self.code_gen_dict["$PRAGMAS$"].append(
- "DO_PRAGMA(HLS ARRAY_PARTITION variable=weights complete dim=2)"
+ (
+ "DO_PRAGMA(HLS ARRAY_PARTITION "
+ "variable=weights.m_weights complete dim=1)"
+ )
)
+ # the threshold tensor is acc_type [PE][TMEM][N_THRES]
+ # partition for parallel access along PE and N_THRES dimensions (dims 1 and 3)
if self.calc_tmem() != 0:
# TODO find a better way of checking for no pregenerated thresholds
self.code_gen_dict["$PRAGMAS$"].append(
- "DO_PRAGMA(HLS ARRAY_PARTITION variable=threshs complete dim=1)"
+ (
+ "DO_PRAGMA(HLS ARRAY_PARTITION variable=threshs.m_thresholds "
+ "complete dim=1)"
+ )
)
self.code_gen_dict["$PRAGMAS$"].append(
- "DO_PRAGMA(HLS ARRAY_PARTITION variable=threshs complete dim=3)"
+ (
+ "DO_PRAGMA(HLS ARRAY_PARTITION variable=threshs.m_thresholds "
+ "complete dim=3)"
+ )
)
--
GitLab