roll backing reinterpret_cast for merge conflict fix

cmake/external/abseil-cpp.cmake

@@ -27,14 +27,18 @@ FetchContent_Declare(
     URL ${DEP_URL_abseil_cpp}
     URL_HASH SHA1=${DEP_SHA1_abseil_cpp}
     PATCH_COMMAND ${ABSL_PATCH_COMMAND}
-    FIND_PACKAGE_ARGS NAMES absl
+    FIND_PACKAGE_ARGS 20240116 NAMES absl
 )
 
 onnxruntime_fetchcontent_makeavailable(abseil_cpp)
 FetchContent_GetProperties(abseil_cpp)
 set(ABSEIL_SOURCE_DIR ${abseil_cpp_SOURCE_DIR})
+# abseil_cpp_SOURCE_DIR is non-empty if we build it from source
 message(STATUS "Abseil source dir:" ${ABSEIL_SOURCE_DIR})
-
+# abseil_cpp_VERSION  is non-empty if we find a preinstalled ABSL
+if(abseil_cpp_VERSION)
+  message(STATUS "Abseil version:" ${abseil_cpp_VERSION})
+endif()
 if (GDK_PLATFORM)
   # Abseil considers any partition that is NOT in the WINAPI_PARTITION_APP a viable platform
   # for Win32 symbolize code (which depends on dbghelp.lib); this logic should really be flipped

docs/OperatorKernels.md

@@ -421,7 +421,7 @@ Do not modify directly.*
 |Transpose|*in* data:**T**<br> *out* transposed:**T**|21+|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e4m3fnuz), tensor(float8e5m2), tensor(float8e5m2fnuz), tensor(int16), tensor(int32), tensor(int4), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint4), tensor(uint64), tensor(uint8)|
 |||[13, 20]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
 |||[1, 12]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
-|Trilu|*in* input:**T**<br> *in* k:**tensor(int64)**<br> *out* output:**T**|14+|**T** = tensor(double), tensor(float), tensor(int64)|
+|Trilu|*in* input:**T**<br> *in* k:**tensor(int64)**<br> *out* output:**T**|14+|**T** = tensor(bool), tensor(double), tensor(float), tensor(int64)|
 |Unique|*in* X:**T**<br> *out* Y:**T**<br> *out* indices:**tensor(int64)**<br> *out* inverse_indices:**tensor(int64)**<br> *out* counts:**tensor(int64)**|11+|**T** = tensor(double), tensor(float), tensor(int64), tensor(int8), tensor(string)|
 |Unsqueeze|*in* data:**T**<br> *in* axes:**tensor(int64)**<br> *out* expanded:**T**<br><br>or<br><br>*in* data:**T**<br> *out* expanded:**T**|21+|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
 |||[13, 20]|**T** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|

include/onnxruntime/core/common/logging/isink.h

@@ -6,12 +6,15 @@
 #include <string>
 
 #include "core/common/logging/logging.h"
+#include "core/common/logging/sink_types.h"
 
 namespace onnxruntime {
 namespace logging {
 class ISink {
  public:
-  ISink() = default;
+  explicit ISink(SinkType type = SinkType::BaseSink) : type_(type) {}
+
+  SinkType GetType() const { return type_; }
 
   /**
      Sends the message to the sink.
@@ -32,6 +35,8 @@ class ISink {
   virtual ~ISink() = default;
 
  private:
+  SinkType type_;
+
   // Make Code Analysis happy by disabling all for now. Enable as needed.
   ORT_DISALLOW_COPY_ASSIGNMENT_AND_MOVE(ISink);
 

include/onnxruntime/core/common/logging/logging.h

@@ -14,10 +14,10 @@
 #include "core/common/common.h"
 #include "core/common/profiler_common.h"
 #include "core/common/logging/capture.h"
-#include "core/common/logging/severity.h"
-
 #include "core/common/logging/macros.h"
-
+#include "core/common/logging/severity.h"
+#include "core/common/logging/sink_types.h"
+#include "core/platform/ort_mutex.h"
 #include "date/date.h"
 
 /*
@@ -167,6 +167,23 @@ class LoggingManager final {
   */
   static bool HasDefaultLogger() { return nullptr != s_default_logger_; }
 
+  /**
+    Gets the default instance of the LoggingManager.
+  */
+  static LoggingManager* GetDefaultInstance();
+
+  /**
+     Removes a Sink if one is present
+  */
+  void RemoveSink(SinkType sinkType);
+
+  /**
+     Adds a Sink to the current sink creating a CompositeSink if necessary
+     Sinks types must be unique
+     @param severity The severity level for the new Sink
+  */
+  bool AddSinkOfType(SinkType sinkType, std::function<std::unique_ptr<ISink>()> sinkFactory, logging::Severity severity);
+
   /**
      Change the minimum severity level for log messages to be output by the default logger.
      @param severity The severity.
@@ -214,7 +231,10 @@ class LoggingManager final {
   void CreateDefaultLogger(const std::string& logger_id);
 
   std::unique_ptr<ISink> sink_;
-  const Severity default_min_severity_;
+#ifdef _WIN32
+  mutable OrtMutex sink_mutex_;
+#endif
+  Severity default_min_severity_;
   const bool default_filter_user_data_;
   const int default_max_vlog_level_;
   bool owns_default_logger_;
@@ -362,8 +382,8 @@ unsigned int GetProcessId();
 /**
    If the ONNXRuntimeTraceLoggingProvider ETW Provider is enabled, then adds to the existing logger.
 */
-std::unique_ptr<ISink> EnhanceLoggerWithEtw(std::unique_ptr<ISink> existingLogger, logging::Severity originalSeverity,
-                                            logging::Severity etwSeverity);
+std::unique_ptr<ISink> EnhanceSinkWithEtw(std::unique_ptr<ISink> existingSink, logging::Severity originalSeverity,
+                                          logging::Severity etwSeverity);
 
 /**
   If the ONNXRuntimeTraceLoggingProvider ETW Provider is enabled, then can override the logging level.

include/onnxruntime/core/common/logging/sink_types.h

@@ -0,0 +1,11 @@
+#pragma once
+
+namespace onnxruntime {
+namespace logging {
+enum class SinkType {
+  BaseSink,
+  CompositeSink,
+  EtwSink
+};
+}  // namespace logging
+}  // namespace onnxruntime

js/web/docs/webgpu-operators.md

@@ -74,6 +74,7 @@ Do not modify directly.*
 | Not | ai.onnx(1+) |  |
 | Pad | ai.onnx(2-10,11-12,13-17,18,19+) |  |
 | Pow | ai.onnx(7-11,12,13-14,15+) |  |
+| QuickGelu | com.microsoft(1+) |  |
 | Range | ai.onnx(11+) |  |
 | Reciprocal | ai.onnx(6-12,13+) |  |
 | ReduceL1 | ai.onnx(1-10,11-12,13-17,18+) |  |

js/web/docs/webnn-operators.md

@@ -19,7 +19,7 @@ operators and the supported opset domain/versions in **WebNN EP** by ONNX Runtim
 | BatchNormalization | ai.onnx(7-8, 9-13, 14, 15+) | batchNormalization | ✗ | ✓ | Only supports 'training_mode' value is 0, one output |
 | Cast | ai.onnx(7-8, 9-12, 13-18, 19-20, 21+) | cast | ✗ | ✓ | |
 | Ceil | ai.onnx(7-12, 13+) | ceil | ✓ | ✓ | |
-| Clip | ai.onnx(7-10, 11, 12, 13+) | clamp | ✓ | ✓ | |
+| Clip | ai.onnx(7-10, 11, 12, 13+) | clamp | ✓ | ✓ | WebNN CPU backend only supports 3 specific ranges: [0.0, infinity], [-1.0, 1.0], [0.0, 6.0] (Chromium issue: https://issues.chromium.org/issues/326156496) |
 | Concat | ai.onnx(7-10, 11-12, 13+) | concat | ✓ | ✓ | |
 | Conv | ai.onnx(7-10, 11+) | conv2d | ✓ | ✓ | Only supports 3-D or 4-D input and 'W' (weight). WebNN CPU requires the 'W' (weight) input to be a constant |
 | ConvTranspose | ai.onnx(7-10, 11+) | convTranspose2d | ✓ | ✗ | Only supports 3-D or 4-D input and 'W' (weight). |
@@ -50,7 +50,7 @@ operators and the supported opset domain/versions in **WebNN EP** by ONNX Runtim
 | LessOrEqual | ai.onnx(12-15, 16+) | lesserOrEqual | ✗ | ✓ | |
 | Log | ai.onnx(7-12, 13+) | log | ✗ | ✓ | |
 | LpPool | ai.onnx(7-10, 11-17, 18+) | l2Pool2d | ✗ | ✓ | Only supports 4-D input, 2-D 'kernel_shape', 'p' value is 2 |
-| MatMul | ai.onnx(7-8, 9-12, 13+) | matmul | ✓ | ✓ | WebNN CPU doesn't support broadcasting for MatMul |
+| MatMul | ai.onnx(7-8, 9-12, 13+) | matmul | ✓ | ✓ | |
 | Max | ai.onnx(7, 8-11, 12, 13+) | max | ✓ | ✓ | |
 | MaxPool | ai.onnx(7, 8-9, 10, 11, 12+) | maxPool2d | ✓ | ✓ | Only supports 4-D input, 2-D 'kernel_shape', 'storage_order' != 1, one output |
 | Min | ai.onnx(7, 8-11, 12, 13+) | min | ✓ | ✓ | |
@@ -73,15 +73,15 @@ operators and the supported opset domain/versions in **WebNN EP** by ONNX Runtim
 | ReduceSumSquare | ai.onnx(7-10, 11-12, 13-17, 18+) | reduceSumSquare | ✗ | ✓ | Input 'axes' if present should be a constant |
 | Relu | ai.onnx(7-12, 13, 14+) | relu | ✓ | ✓ | |
 | Reshape | ai.onnx(7-12, 13, 14-18, 19-20, 21+) | reshape | ✓ | ✓ | Input 'shape' should be a constant, 0 dimension value in 'shape' is not supported |
-| Resize | ai.onnx(11-12, 13-17, 18, 19+) | resample2d | ✓ | ✓ | Only supports 4-D input, exclude_outside != 0, input 'scales' and 'sizes' if present must be a constant, WebNN CPU backend only supports 'linear' mode, WebNN GPU backend only supports 'linear' and 'nearest' modes |
+| Resize | ai.onnx(11-12, 13-17, 18, 19+) | resample2d | ✓ | ✓ | Only supports 4-D input, exclude_outside != 0, input 'scales' and 'sizes' if present must be a constant, 'linear' and 'nearest' modes |
 | Shape | ai.onnx(7-12, 13-14, 15-18, 19-20, 21+) | slice | ✓ | ✓ | |
 | Sigmoid | ai.onnx(7-12, 13+) | sigmoid | ✓ | ✓ | |
 | Softplus | ai.onnx(7+) | softplus | ✗ | ✓ | |
 | Softsign | ai.onnx(7+) | softsign | ✗ | ✓ | |
 | Sin | ai.onnx(7+) | sin | ✗ | ✓ | |
 | Slice | ai.onnx(7-9, 10, 11-12, 13+) | slice | ✓ | ✓ | Input 'starts', 'ends', 'axes', and 'steps' if present must be a constant, only supports 'steps' value 1 |
 | Softmax | ai.onnx(7-10, 11-12, 13+) | softmax | ✓ | ✓ | Only supports input rank >= 2 |
-| Split | ai.onnx(7-10, 11-12, 13-17, 18+) | split | ✓ | ✓ | Input 'split' if present should be a constant, WebNN CPU backend only supports up to 4 outputs |
+| Split | ai.onnx(7-10, 11-12, 13-17, 18+) | split | ✓ | ✓ | Input 'split' if present should be a constant |
 | Sqrt | ai.onnx(7-12, 13+) | sqrt | ✓ | ✓ | |
 | Squeeze | ai.onnx(7-10, 11-12, 13-20, 21+) | reshape | ✓ | ✓ | Input 'axes' if present should be a constant |
 | Sub | ai.onnx(7-12, 13, 14+) | sub | ✓ | ✓ | |

js/web/lib/wasm/jsep/webgpu/op-resolve-rules.ts

@@ -107,6 +107,7 @@ export const WEBGPU_OP_RESOLVE_RULES: Map<string, OperatorImplementation> = new
   ['Not', [unaryOps.not]],
   ['Pad', [pad]],
   ['Pow', [binaryOps.pow]],
+  ['QuickGelu', [unaryOps.quickgelu, unaryOps.parseAlphaAttributes]],
   ['Range', [range]],
   ['Reciprocal', [unaryOps.reciprocal]],
   ['ReduceMin', [reduceMin]],

js/web/lib/wasm/jsep/webgpu/ops/unary-op.ts

@@ -314,3 +314,31 @@ export const thresholdedRelu = (context: ComputeContext, attributes: AlphaAttrib
 export const log = (context: ComputeContext): void => {
   context.compute(createElementwiseProgramInfo(context.inputs[0], 'Log', 'log'));
 };
+
+export const quickGeluImpl = (varType: string, alpha: number) => `
+const alpha = vec4<${varType}>(${alpha});
+const one = ${varType}(1.0);
+const zero = ${varType}(0.0);
+
+fn quick_gelu_impl(x: vec4<${varType}>) -> vec4<${varType}> {
+  let v = x *alpha;
+  var x1 : vec4<${varType}>;
+  for (var i = 0; i < 4; i = i + 1) {
+    if (v[i] >= zero) {
+      x1[i] = one / (one + exp(-v[i]));
+    } else {
+      x1[i] = one - one / (one + exp(v[i]));
+    }
+  }
+  return x * x1;
+}
+`;
+
+export const quickGeluExpression = (x: string) => `quick_gelu_impl(${x})`;
+
+export const quickgelu = (context: ComputeContext, attributes: AlphaAttributes): void => {
+  const dType = tensorTypeToWsglValueType(context.inputs[0].dataType);
+  context.compute(createElementwiseProgramInfo(
+      context.inputs[0], 'QuickGelu', quickGeluExpression, quickGeluImpl(dType, attributes.alpha), attributes.cacheKey,
+      context.inputs[0].dataType));
+};

js/web/test/data/ops/quick-gelu.jsonc

@@ -0,0 +1,46 @@
+[
+  {
+    "name": "QuickGelu test",
+    "operator": "QuickGelu",
+    "opset": { "domain": "com.microsoft", "version": 1 },
+    "cases": [
+      {
+        "name": "[2x4]",
+        "inputs": [
+          {
+            "data": [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, -0.8],
+            "dims": [2, 4],
+            "type": "float32"
+          }
+        ],
+        "outputs": [
+          {
+            "data": [0.0542447, 0.116857, 0.187484, 0.265566, 0.350388, 0.441123, 0.53689, 0.636815],
+            "dims": [2, 4],
+            "type": "float32"
+          }
+        ]
+      },
+      {
+        "name": "[3x5]",
+        "inputs": [
+          {
+            "data": [0.1, 0.2, 0.3, 0.4, 0.5, 1, 2, 3, 4, 5, 1.1, 1.2, 1.3, 1.4, -1.5],
+            "dims": [3, 5],
+            "type": "float32"
+          }
+        ],
+        "outputs": [
+          {
+            "data": [
+              0.0542447, 0.116857, 0.187484, 0.265566, 0.350388, 0.845795, 1.9356, 2.98192, 3.99558, 4.99899, 0.953383,
+              1.0622, 1.17178, 1.2817, 1.39166
+            ],
+            "dims": [3, 5],
+            "type": "float32"
+          }
+        ]
+      }
+    ]
+  }
+]