L4_HAL/pcd: Optimise writing to USB FIFO to fill it as much as possible.

The existing HAL code had one issue:

- If the write loop went around again it would use the old len32b to check
  if there was enough space, thereby missing some opportunities to write
  packets that were smaller than the previous len32b.

This patch fixes this issue.

Author: dpgeorge

STM32L4xx_HAL_Driver/Src/stm32l4xx_hal_pcd.c

@@ -2160,6 +2160,7 @@ static HAL_StatusTypeDef PCD_WriteEmptyTxFifo(PCD_HandleTypeDef *hpcd, uint32_t
     return HAL_ERROR;
   }
 
+  #if 0
   len = ep->xfer_len - ep->xfer_count;
 
   if (len > ep->maxpacket)
@@ -2187,6 +2188,36 @@ static HAL_StatusTypeDef PCD_WriteEmptyTxFifo(PCD_HandleTypeDef *hpcd, uint32_t
     ep->xfer_count += len;
   }
 
+  #else
+  /* dpgeorge: modified above loop to:
+   *    - recompute len32b before each check of this value against FIFO free space
+   */
+  for (;;)
+  {
+    len = ep->xfer_len - ep->xfer_count;
+    if (len <= 0U)
+    {
+      /* Finished sending all data */
+      break;
+    }
+    if (len > ep->maxpacket)
+    {
+      len = ep->maxpacket;
+    }
+
+    len32b = (len + 3U) / 4U;
+    if  (len32b > (USBx_INEP(epnum)->DTXFSTS & USB_OTG_DTXFSTS_INEPTFSAV))
+    {
+      /* No room left in FIFO */
+      break;
+    }
+
+    USB_WritePacket(USBx, ep->xfer_buff, epnum, len);
+    ep->xfer_buff  += len;
+    ep->xfer_count += len;
+  }
+  #endif
+
   if (ep->xfer_len <= ep->xfer_count)
   {
     fifoemptymsk = (uint32_t)(0x1UL << (epnum & EP_ADDR_MSK));