Add done_cra_writes function to the HAL

This commit adds the `simulation_streaming_interface_done_cra_writes()` function to the HAL. When called, the function signals the simulator that the runtime has finished writing all CRA args for a specific kernel with streaming control interface. This allows the kernell to begin computing and eventually finish any tasks it needs to complete.

Author: IlanTruanovsky

include/acl_hal.h

@@ -245,6 +245,9 @@ typedef struct {
   void (*simulation_streaming_kernel_done)(unsigned int physical_device_id,
                                            const std::string &signal_name,
                                            unsigned int &finish_counter);
+
+  void (*simulation_streaming_interface_done_cra_writes)(
+      const unsigned int physical_device_id, const int accel_id);
 } acl_hal_t;
 
 /// Linked list of MMD library names to load.

include/acl_hal_mmd.h

@@ -149,6 +149,11 @@ typedef struct {
   // accounted and returned in a subsequent invocation of this function.
   void (*aocl_mmd_simulation_streaming_kernel_done)(
       int handle, const std::string &signal_name, unsigned int &finish_counter);
+
+  // Signals a streaming kernel control that all of its CRA arguments have been
+  // written and that it can start
+  void (*aocl_mmd_simulation_streaming_interface_done_cra_writes)(
+      const int device_handle, const int accel_id);
 } acl_mmd_dispatch_t;
 
 typedef struct {

src/acl_hal_mmd.cpp

@@ -159,6 +159,8 @@ void acl_hal_mmd_simulation_streaming_kernel_start(
 void acl_hal_mmd_simulation_streaming_kernel_done(
     unsigned int physical_device_id, const std::string &kernel_name,
     unsigned int &finish_counter);
+void aocl_mmd_simulation_streaming_interface_done_cra_writes(
+    unsigned int physical_device_id, const int accel_id);
 
 static size_t acl_kernel_if_read(acl_bsp_io *io, dev_addr_t src, char *dest,
                                  size_t size);
@@ -353,6 +355,7 @@ static acl_hal_t acl_hal_mmd = {
     acl_hal_mmd_shared_alloc,                     // shared_alloc
     acl_hal_mmd_simulation_streaming_kernel_start, // simulation_streaming_kernel_start
     acl_hal_mmd_simulation_streaming_kernel_done, // simulation_streaming_kernel_done
+    aocl_mmd_simulation_streaming_interface_done_cra_writes, // simulation_streaming_interface_done_cra_writes
 };
 
 // This will contain the device physical id to tell us which device across all
@@ -2823,3 +2826,10 @@ void acl_hal_mmd_simulation_streaming_kernel_done(
       .mmd_dispatch->aocl_mmd_simulation_streaming_kernel_done(
           device_info[physical_device_id].handle, kernel_name, finish_counter);
 }
+
+void aocl_mmd_simulation_streaming_interface_done_cra_writes(
+    const unsigned int physical_device_id, const int accel_id) {
+  device_info[physical_device_id]
+      .mmd_dispatch->aocl_mmd_simulation_streaming_interface_done_cra_writes(
+          device_info[physical_device_id].handle, accel_id);
+}

src/acl_kernel_if.cpp

@@ -1252,9 +1252,8 @@ void acl_kernel_if_launch_kernel_on_custom_sof(
     offset = (unsigned int)(KERNEL_OFFSET_INVOCATION_IMAGE +
                             kern->cra_address_offset);
     image_p = (uintptr_t) & (image->work_dim);
-    image_size_static =
-        (size_t)((uintptr_t) & (image->arg_value) - (uintptr_t) &
-                 (image->work_dim));
+    image_size_static = (size_t)(
+        (uintptr_t) & (image->arg_value) - (uintptr_t) & (image->work_dim));
   }
 
   if ((kern->io.debug_verbosity) >= 2) {
@@ -1309,6 +1308,8 @@ void acl_kernel_if_launch_kernel_on_custom_sof(
     acl_get_hal()->simulation_streaming_kernel_start(
         kern->physical_device_id,
         kern->streaming_control_signal_names[accel_id]->start);
+    acl_get_hal()->simulation_streaming_interface_done_cra_writes(
+        kern->physical_device_id, accel_id);
     return;
   }