fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-09 08:58:02 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

teflon: Add support for tensor split and concatenation operations

Browse source 
These are often use to reuse the output from a previous operation, or to
implement convolution groups.

Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32498>
This commit is contained in:
Tomeu Vizoso 2024-10-24 11:19:41 +02:00 • committed by Marge Bot
parent 418f864ae4
commit 4ced6480ea
2 changed files with 44 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

42
src/gallium/frontends/teflon/tfl_device.c
View file

@ -155,6 +155,12 @@ fill_operation(struct teflon_delegate *delegate, TfLiteContext *tf_context, TfLi
case kTfLiteBuiltinAdd:
operation->type = PIPE_ML_OPERATION_TYPE_ADD;
break;
case kTfLiteBuiltinConcatenation:
operation->type = PIPE_ML_OPERATION_TYPE_CONCATENATION;
break;
case kTfLiteBuiltinSplit:
operation->type = PIPE_ML_OPERATION_TYPE_SPLIT;
break;
default:
unreachable("Unsupported ML operation type");
}
@ -224,9 +230,15 @@ dump_graph(struct pipe_tensor *tensors, unsigned tensor_count, struct pipe_ml_op
case PIPE_ML_OPERATION_TYPE_CONVOLUTION:
teflon_debug("%-6s ", operations[i].conv.depthwise ? "DWCONV" : "CONV");
break;
case PIPE_ML_OPERATION_TYPE_CONCATENATION:
teflon_debug("%-6s ", "CONCAT");
break;
case PIPE_ML_OPERATION_TYPE_POOLING:
teflon_debug("%-6s ", "POOL");
break;
case PIPE_ML_OPERATION_TYPE_SPLIT:
teflon_debug("%-6s ", "SPLIT");
break;
}
for (unsigned j = 0; j < operations[i].input_count; j++) {
@ -516,6 +528,36 @@ PrepareDelegate(TfLiteContext *context, TfLiteDelegate *delegate)
case kTfLiteBuiltinAdd:
supported = true;
break;
case kTfLiteBuiltinConcatenation: {
TfLiteConcatenationParams *params = node->builtin_data;
supported = true;
if (params->axis != 3 &&
params->axis != -1)
supported = false;
unsigned input_channels = context->tensors[node->inputs->data[0]].dims->data[3];
for (unsigned i = 1; i < node->inputs->size; i++)
if (input_channels != context->tensors[node->inputs->data[i]].dims->data[3])
supported = false;
break;
}
case kTfLiteBuiltinSplit: {
int32_t axis = context->tensors[node->inputs->data[0]].data.i32[0];
supported = true;
if (axis != 3 &&
axis != -1)
supported = false;
unsigned output_channels = context->tensors[node->outputs->data[0]].dims->data[3];
for (unsigned i = 1; i < node->outputs->size; i++)
if (output_channels != context->tensors[node->outputs->data[i]].dims->data[3])
supported = false;
break;
}
}
if (supported)

2
src/gallium/include/pipe/p_state.h
View file

@ -1059,6 +1059,8 @@ enum pipe_ml_operation_type {
PIPE_ML_OPERATION_TYPE_ADD,
PIPE_ML_OPERATION_TYPE_CONVOLUTION,
PIPE_ML_OPERATION_TYPE_POOLING,
PIPE_ML_OPERATION_TYPE_CONCATENATION,
PIPE_ML_OPERATION_TYPE_SPLIT,
};
/**