vela: Added debug info to external API

 - Added optional name attributes to operators and tensors

Signed-off-by: Tim Hall <tim.hall@arm.com>
Change-Id: I3b5d881a7b1043a6ba4b58fff5d7532b271ba536

3 files changed, 14 insertions, 6 deletions

diff --git a/ethosu/vela/api.py b/ethosu/vela/api.py
index 3382ea9..399fd46 100644
--- a/ethosu/vela/api.py
+++ b/ethosu/vela/api.py
@@ -27,7 +27,7 @@ import numpy
 
 
 API_VERSION_MAJOR = 1
-API_VERSION_MINOR = 2
+API_VERSION_MINOR = 3
 API_VERSION = f"{API_VERSION_MAJOR}.{API_VERSION_MINOR}"
 
 
@@ -253,6 +253,8 @@ class NpuFeatureMap:
         self.layout: NpuLayout = NpuLayout.NHWC
         # x/y/c strides used by the NPU when traversing the feature map, if None, vela will use default strides
         self.strides: Optional[NpuShape3D] = None
+        # Used for debug
+        self.name: Optional[str] = None
 
 
 class NpuKernel:
@@ -290,6 +292,8 @@ class NpuOperation:
 
     def __init__(self, op_type: NpuOperationType):
         self.op_type = op_type
+        # Used for debug
+        self.name: Optional[str] = None
 
 
 class NpuDmaOperation(NpuOperation):
diff --git a/ethosu/vela/high_level_command_to_npu_op.py b/ethosu/vela/high_level_command_to_npu_op.py
index 8c5525b..e6bfc1c 100644
--- a/ethosu/vela/high_level_command_to_npu_op.py
+++ b/ethosu/vela/high_level_command_to_npu_op.py
@@ -296,6 +296,7 @@ def create_feature_map(tens: Tensor, box: Box, arch: ArchitectureFeatures, op_sh
     )
     strides = tens.get_strides(shape4D=op_shape4D)
     fm.strides = NpuShape3D(height=int(strides[2]), width=int(strides[3]), depth=int(strides[1]))
+    fm.name = tens.name
     return fm
 
 
@@ -539,6 +540,7 @@ def convert_command_to_npu_op(cmd: Command, arch: ArchitectureFeatures) -> NpuOp
     npu_op: NpuOperation
     if isinstance(cmd, DMA):
         npu_op = create_dma_op(cmd, arch)
+        npu_op.name = cmd.out_tensor.name
     elif isinstance(cmd, NpuStripe):
         npu_block_type = cmd.ps.primary_op.type.npu_block_type
         if npu_block_type in (NpuBlockType.ConvolutionMxN, NpuBlockType.VectorProduct):
@@ -551,6 +553,7 @@ def convert_command_to_npu_op(cmd: Command, arch: ArchitectureFeatures) -> NpuOp
             npu_op = create_npu_elementwise_op(cmd, arch)
         else:
             assert 0, f"Unknown command type {npu_block_type}"
+        npu_op.name = cmd.ps.primary_op.name
     return npu_op
 
 
diff --git a/ethosu/vela/register_command_stream_generator.py b/ethosu/vela/register_command_stream_generator.py
index be01a75..7858e70 100644
--- a/ethosu/vela/register_command_stream_generator.py
+++ b/ethosu/vela/register_command_stream_generator.py
@@ -824,19 +824,20 @@ def print_feature_map(fm: Optional[NpuFeatureMap], name: str):
         t = fm.tiles
         addresses = [hex(addr) for addr in t.addresses]
         print(f"         {stride_str}, tiles: w0={t.width_0}, h0={t.height_0}, h1={t.height_1}, base={addresses}")
+        print(f"         name={fm.name}")
 
 
 def print_operation(npu_op: NpuOperation, index: int = 0, cmd=None):
-    pass_info = f", {cmd}" if cmd else ""
+    pass_info = f" {cmd}" if cmd else ""
     if isinstance(npu_op, NpuOperation) and not isinstance(npu_op, (NpuDmaOperation, NpuBlockOperation)):
-        print(f"{index} {npu_op.op_type.name}{pass_info}")
+        print(f"{index} {npu_op.op_type.name}  name={npu_op.name}:{pass_info}")
         return
     if isinstance(npu_op, NpuDmaOperation):
-        print(f"{index} DMA_START src={npu_op.src}, dest={npu_op.dest}{pass_info}")
+        print(f"{index} {npu_op.op_type.name} name={npu_op.name}, src={npu_op.src}, dest={npu_op.dest}:{pass_info}")
         return
     k = None if npu_op.kernel is None else to_kernel(npu_op.kernel)
     if isinstance(npu_op, (NpuPoolingOperation, NpuElementWiseOperation)):
-        print(f"{index} {npu_op.sub_op_type.name} {npu_op.op_type.name}:{pass_info}")
+        print(f"{index} {npu_op.sub_op_type.name} {npu_op.op_type.name} name={npu_op.name}:{pass_info}")
     else:
         if (
             isinstance(npu_op, NpuConv2DOperation)
@@ -845,7 +846,7 @@ def print_operation(npu_op: NpuOperation, index: int = 0, cmd=None):
             fc = "FullyConnected "
         else:
             fc = ""
-        print(f"{index} {fc}{npu_op.op_type.name}{pass_info}")
+        print(f"{index} {fc}{npu_op.op_type.name} name={npu_op.name}:{pass_info}")
     print_feature_map(npu_op.ifm, "IFM")
     if npu_op.ifm2_scalar is not None:
         quant_val = quantise(npu_op.ifm2_scalar, npu_op.ifm2.quantization)