From e5404dca4d61cb75a9e8014350859aa75ce50357 Mon Sep 17 00:00:00 2001
From: Viviane Potocnik <vivianep@iis.ee.ethz.ch>
Date: Sun, 17 Sep 2023 00:35:39 +0200
Subject: [PATCH] sw: add generic DMA transfer and documentation

---
 sw/apps/transformer/src/transformer.h | 34 ++++++++++++++++++++++++++-
 1 file changed, 33 insertions(+), 1 deletion(-)

diff --git a/sw/apps/transformer/src/transformer.h b/sw/apps/transformer/src/transformer.h
index 0b25832a82..7301a0d5c1 100644
--- a/sw/apps/transformer/src/transformer.h
+++ b/sw/apps/transformer/src/transformer.h
@@ -83,6 +83,38 @@ dump_float(ifmap, 6);
 dump_float(weights, 10); // = 0xa
 dump_float(value, 12); // = 0xc
 
+/**
+ * @brief  Generic DMA transfer function
+ * @param dst        Pointer to destination
+ * @param src        Pointer to source
+ * @param size       How many bytes to transfer in total
+ * @param dst_stride Stride of the destination, i.e. how many bytes to jump
+ *                   between two consecutive transfers in the same row of 
+ *                   the destination
+ * @param src_stride Stride of the source, i.e. how many bytes to jump
+ *                   between two consecutive transfers in the same row of
+ *                   the source
+ * @param repetitions How many rows to transfer
+ */
+
+static inline void dma_transfer(void *dst, void *src, uint32_t size, uint32_t dst_stride, 
+                                uint32_t src_stride, uint32_t repetitions) {
+    if (!snrt_is_compute_core()) {
+        snrt_dma_txid_t txid =
+            snrt_dma_start_2d(
+                dst,        /* dst */
+                src,        /* src */
+                size,       /* size */
+                dst_stride, /* dst_stride */
+                src_stride, /* src_stride */
+                repetitions); /* repetitions */
+
+        snrt_dma_wait_all();
+    }
+    snrt_cluster_hw_barrier();
+}
+    
+
 /**
  * Implementation of the GELU layer
  */
@@ -216,7 +248,7 @@ static inline void transformer_layer(transformer_layer_t *const l) {
                     S_TILE * sizeof(float),       /* size */
                     l->embeddings,                /* dst_stride */
                     l->embeddings,                /* src_stride */
-                    l->seq_len);                  /* repetitions */
+                    S_TILE);                      /* repetitions */
 
                 snrt_dma_wait_all();