[OpenCLML] Transposed convolution support and other fixes (apache#14767)

* [OpenCLML] Transposed convolution support and other fixes

Added support for transposed convolution.
Epsilon support for batchnorm op added - part of v3.0.
CLML version query bug fixed.

* * review comments and various compilation warnings.

* * lint issues.

apps/cpp_rtvm/tvm_runner.h

@@ -38,8 +38,7 @@ namespace runtime {
 /*!
  * \brief various meta information related to the compiled TVM model.
  */
-typedef struct {
- public:
+typedef struct _TVMMetaInfo {
   int n_inputs;
   int n_outputs;
   std::map<std::string, std::pair<std::vector<int>, std::string>> input_info;

cmake/modules/LibInfo.cmake

@@ -120,6 +120,7 @@ function(add_lib_info src_file)
     TVM_INFO_USE_VULKAN="${USE_VULKAN}"
     TVM_INFO_USE_CLML="${USE_CLML}"
     TVM_INFO_USE_CLML_GRAPH_EXECUTOR="${USE_CLML_GRAPH_EXECUTOR}"
+    TVM_INFO_USE_TVM_CLML_VERSION="${CLML_VERSION_MAJOR}"
     TVM_INFO_USE_UMA="${USE_UMA}"
     TVM_INFO_USE_VERILATOR="${USE_VERILATOR}"
     TVM_INFO_USE_CCACHE="${USE_CCACHE}"

python/tvm/relay/op/contrib/clml.py

@@ -37,7 +37,7 @@
 def clml_sdk_version():
     """Utility function to get clml version version"""
 
-    return tvm.support.libinfo().get("TVM_CLML_VERSION", 2)
+    return int(tvm.support.libinfo().get("TVM_CLML_VERSION", 2))
 
 
 def is_clml_runtime_enabled():
@@ -155,6 +155,7 @@ def alter_conv(attrs, inputs, tinfos, out_type):
         seq = tvm.transform.Sequential(
             [
                 transform.ConvertLayout({"nn.conv2d": ["NCHW", "OIHW"]}),
+                transform.ConvertLayout({"nn.conv2d_transpose": ["NCHW", "OIHW"]}),
                 transform.AlterOpLayout(),
                 transform.FoldConstant(),
             ]
@@ -203,6 +204,22 @@ def conv_pattern():
         pattern = pattern.optional(is_op("clip"))
         return pattern
 
+    def conv_transpose_pattern():
+        """Create a transposed convolution pattern."""
+        pattern = is_op("nn.conv2d_transpose")(wildcard(), is_constant())
+        pattern = pattern.optional(lambda x: is_op("nn.bias_add")(x, is_constant()))
+        pattern = pattern.optional(lambda x: is_op("add")(x, is_constant()))
+        pattern = pattern.optional(
+            lambda x: is_tuple_get_item(
+                is_op("nn.batch_norm")(
+                    x, is_constant(), is_constant(), is_constant(), is_constant()
+                )
+            )
+        )
+        pattern = pattern.optional(is_op("nn.relu"))
+        pattern = pattern.optional(is_op("clip"))
+        return pattern
+
     def pad_conv_pattern():
         """Create a pad with convolution pattern."""
         pattern = is_op("nn.pad")(wildcard(), is_constant())
@@ -300,6 +317,31 @@ def check_conv(extract):
             return False
         return True
 
+    def check_conv_transpose(extract):
+        """Check transposed conv pattern is supported by CLML."""
+        call = extract
+        if isinstance(call, tvm.relay.expr.TupleGetItem):
+            call = call.tuple_value
+        elif call.op.name == "nn.relu":
+            call = call.args[0]
+            if isinstance(call, tvm.relay.expr.TupleGetItem):
+                call = call.tuple_value
+        elif call.op.name == "clip":
+            if call.attrs["a_min"] != 0.0 or call.attrs["a_max"] != 6.0:
+                return False
+            call = call.args[0]
+            if isinstance(call, tvm.relay.expr.TupleGetItem):
+                call = call.tuple_value
+
+        while call.op.name != "nn.conv2d_transpose":
+            call = call.args[0]
+
+        attrs = call.attrs
+        if attrs.data_layout != "NCHW":
+            return False
+
+        return True
+
     def check_binary_op(extract):
         call = extract
         if len(call.args[1].checked_type.shape) > 0:
@@ -340,6 +382,7 @@ def check_default_op(extract):
     return [
         ("clml.pad_conv2d", pad_conv_pattern(), check_conv),
         ("clml.conv2d", conv_pattern(), check_conv),
+        ("clml.conv2d_transpose", conv_transpose_pattern(), check_conv_transpose),
         ("clml.dense", dense_pattern(), check_default_op),
         ("clml.pad", pad_pattern(), check_pad_op),
         ("clml.concat", concat_pattern(), check_concat_op),

src/relay/backend/contrib/clml/codegen.cc

@@ -83,7 +83,7 @@ class CLMLJSONSerializer : public backend::contrib::JSONSerializer {
     ICHECK(comp.defined()) << "CLML JSON runtime only supports composite functions.";
     const std::string name = comp.value();
     std::shared_ptr<JSONGraphNode> json_node;
-    if (name == "clml.conv2d" || name == "clml.pad_conv2d") {
+    if (name == "clml.conv2d" || name == "clml.pad_conv2d" || name == "clml.conv2d_transpose") {
       json_node = CreateCompositeConvJSONNode(cn);
     } else if (name == "clml.batch_norm") {
       json_node = CreateBatchNormJSONNode(cn);
@@ -169,7 +169,10 @@ class CLMLJSONSerializer : public backend::contrib::JSONSerializer {
       current_call = current_call->args[0].as<CallNode>();
     }
     // Enforce a convolution node exists at this point during traversal
-    ICHECK(backend::IsOp(current_call, "nn.conv2d"));
+    if (!backend::IsOp(current_call, "nn.conv2d") &&
+        !backend::IsOp(current_call, "nn.conv2d_transpose")) {
+      LOG(FATAL) << "Can't find primary op in Convolution node";
+    }
     nodes.conv = current_call;
     if (!current_call->args.empty() && current_call->args[0]->IsInstance<CallNode>()) {
       current_call = current_call->args[0].as<CallNode>();
@@ -189,22 +192,27 @@ class CLMLJSONSerializer : public backend::contrib::JSONSerializer {
   std::shared_ptr<JSONGraphNode> CreateCompositeConvJSONNode(const CallNode* cn) {
     CompositeConvNode nodes = UnpackCompositeConvolution(cn);
 
-    const auto* conv_attr = nodes.conv->attrs.as<Conv2DAttrs>();
-    ICHECK(conv_attr);
-
     std::string name;
     std::string name_prefix = "nn";
-
-    // Distinguish between normal and depth-wise convolution
-    if (conv_attr->channels.defined() &&
-        tvm::tir::ExprDeepEqual()(conv_attr->channels, conv_attr->groups) &&
-        conv_attr->groups != 1) {
-      name = "depthwise_conv2d";
-      ICHECK(conv_attr->kernel_layout == "IOHW")
-          << "Kernel layout must be IHWO, has the module been pre-processed correctly?";
-    } else {
-      name = "conv2d";
-      ICHECK(conv_attr->kernel_layout == "OIHW")
+    if (backend::IsOp(nodes.conv, "nn.conv2d")) {
+      const auto* conv_attr = nodes.conv->attrs.as<Conv2DAttrs>();
+      ICHECK(conv_attr);
+      if (conv_attr->channels.defined() &&
+          tvm::tir::ExprDeepEqual()(conv_attr->channels, conv_attr->groups) &&
+          conv_attr->groups != 1) {
+        name = "depthwise_conv2d";
+        ICHECK(conv_attr->kernel_layout == "IOHW")
+            << "Kernel layout must be IHWO, has the module been pre-processed correctly?";
+      } else {
+        name = "conv2d";
+        ICHECK(conv_attr->kernel_layout == "OIHW")
+            << "Kernel layout must be OHWI, has the module been pre-processed correctly?";
+      }
+    } else if (backend::IsOp(nodes.conv, "nn.conv2d_transpose")) {
+      name = "conv2d_transpose";
+      const auto* conv_transpose_attr = nodes.conv->attrs.as<Conv2DTransposeAttrs>();
+      ICHECK(conv_transpose_attr);
+      ICHECK(conv_transpose_attr->kernel_layout == "OIHW")
           << "Kernel layout must be OHWI, has the module been pre-processed correctly?";
     }