[js/web] JSEP Gemm for opset 13 (microsoft#16936)

### Description
Added JSEP Gemm registration for opset 13. It was falling back to CPU
provider as CPU has it for 13

---------

Co-authored-by: Guenther Schmuelling <[email protected]>

Author: dakenf

js/web/docs/webgpu-operators.md

@@ -38,7 +38,7 @@ Do not modify directly.*
 | Floor | ai.onnx(6-12,13+) |  |
 | Gather | ai.onnx(1-10,11-12,13+) |  |
 | Gelu | com.microsoft(1+) |  |
-| Gemm | ai.onnx(7-8,9-10,11+) |  |
+| Gemm | ai.onnx(7-8,9-10,11-12,13+) |  |
 | GlobalAveragePool | ai.onnx(1+); com.ms.internal.nhwc(1+) |  |
 | GlobalMaxPool | ai.onnx(1+); com.ms.internal.nhwc(1+) |  |
 | InstanceNormalization | ai.onnx(6+); com.ms.internal.nhwc(6+) |  |

onnxruntime/core/providers/js/js_execution_provider.cc

@@ -231,7 +231,8 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnn
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, float, ConvTranspose);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 7, 8, float, Gemm);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 9, 10, float, Gemm);
-class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, float, Gemm);
+class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, 12, float, Gemm);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 13, float, Gemm);
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 1, 12, MatMul);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 13, MatMul);
 
@@ -464,7 +465,8 @@ std::unique_ptr<KernelRegistry> RegisterKernels() {
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, float, ConvTranspose)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 7, 8, float, Gemm)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 9, 10, float, Gemm)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, float, Gemm)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 11, 12, float, Gemm)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 13, float, Gemm)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 1, 12, MatMul)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kJsExecutionProvider, kOnnxDomain, 13, MatMul)>,
 

onnxruntime/core/providers/js/operators/gemm.cc

@@ -12,7 +12,15 @@ namespace js {
   ONNX_OPERATOR_TYPED_KERNEL_EX(                                                           \
       Gemm,                                                                                \
       kOnnxDomain,                                                                         \
-      11,                                                                                  \
+      13,                                                                                  \
+      T,                                                                                   \
+      kJsExecutionProvider,                                                                \
+      (*KernelDefBuilder::Create()).TypeConstraint("T", DataTypeImpl::GetTensorType<T>()), \
+      Gemm<T>);                                                                            \
+  ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_EX(                                                 \
+      Gemm,                                                                                \
+      kOnnxDomain,                                                                         \
+      11, 12,                                                                              \
       T,                                                                                   \
       kJsExecutionProvider,                                                                \
       (*KernelDefBuilder::Create()).TypeConstraint("T", DataTypeImpl::GetTensorType<T>()), \