Merge remote-tracking branch 'origin/main' into user/linneamay/resize-18

.pipelines/OneBranch.Nuget-WindowsAI-Pipeline.Official.yml

@@ -53,10 +53,6 @@ extends:
           BuildArch: x86
           PythonPackageName: pythonx86
 
-      - template: .pipelines/windowsai-steps.yml@self
-        parameters:
-          BuildArch: arm
-
       - template: .pipelines/windowsai-steps.yml@self
         parameters:
           BuildArch: arm64
@@ -72,11 +68,6 @@ extends:
           PythonPackageName: pythonx86
           Runtime: static
 
-      - template: .pipelines/windowsai-steps.yml@self
-        parameters:
-          BuildArch: arm
-          Runtime: static
-
       - template: .pipelines/windowsai-steps.yml@self
         parameters:
           BuildArch: arm64
@@ -94,11 +85,9 @@ extends:
         dependsOn:
         - Windows_Packaging_x64_dynamic
         - Windows_Packaging_x86_dynamic
-        - Windows_Packaging_arm_dynamic
         - Windows_Packaging_arm64_dynamic
         - Windows_Packaging_x64_static
         - Windows_Packaging_x86_static
-        - Windows_Packaging_arm_static
         - Windows_Packaging_arm64_static
         condition: succeeded()
         steps:
@@ -120,12 +109,6 @@ extends:
             artifactName: 'drop_Windows_Build_Windows_Packaging_arm64_dynamic'
             targetPath: '$(Build.BinariesDirectory)/nuget-artifact-arm64'
 
-        - task: DownloadPipelineArtifact@0
-          displayName: 'Download Pipeline Artifact - NuGet DirectML arm'
-          inputs:
-            artifactName: 'drop_Windows_Build_Windows_Packaging_arm_dynamic'
-            targetPath: '$(Build.BinariesDirectory)/nuget-artifact-arm'
-
         - task: DownloadPipelineArtifact@0
           displayName: 'Download Pipeline Artifact - NuGet DirectML x64 StaticRuntime'
           inputs:
@@ -144,12 +127,6 @@ extends:
             artifactName: 'drop_Windows_Build_Windows_Packaging_arm64_static'
             targetPath: '$(Build.BinariesDirectory)/nuget-artifact-arm64-static-runtime'
 
-        - task: DownloadPipelineArtifact@0
-          displayName: 'Download Pipeline Artifact - NuGet DirectML arm StaticRuntime'
-          inputs:
-            artifactName: 'drop_Windows_Build_Windows_Packaging_arm_static'
-            targetPath: '$(Build.BinariesDirectory)/nuget-artifact-arm-static-runtime'
-
         - task: PowerShell@2
           displayName: 'Bundle NuGet and other binaries'
           inputs:
@@ -194,17 +171,7 @@ extends:
               $arm64_static_runtime_nupkg_unzipped_directory = [System.IO.Path]::Combine($arm64_static_runtime_nupkg_unzipped_directory_root, 'binaries', [System.IO.Path]::GetFileNameWithoutExtension($arm64_static_runtime_nuget_package))
               [System.IO.Compression.ZipFile]::ExtractToDirectory($arm64_static_runtime_nuget_package, $arm64_static_runtime_nupkg_unzipped_directory)
 
-              $nupkgs = (Get-ChildItem ..\nuget-artifact-arm -Filter Microsoft.AI.MachineLearning*.nupkg -Recurse)
-              $arm_nuget_package = $nupkgs[0].FullName
-              $arm_nupkg_unzipped_directory_root = $nupkgs[0].Directory.FullName
-              $arm_nupkg_unzipped_directory = [System.IO.Path]::Combine($arm_nupkg_unzipped_directory_root, 'binaries', [System.IO.Path]::GetFileNameWithoutExtension($arm_nuget_package))
-              [System.IO.Compression.ZipFile]::ExtractToDirectory($arm_nuget_package, $arm_nupkg_unzipped_directory)
-
-              $nupkgs = (Get-ChildItem ..\nuget-artifact-arm-static-runtime -Filter Microsoft.AI.MachineLearning*.nupkg -Recurse)
-              $arm_static_runtime_nuget_package = $nupkgs[0].FullName
-              $arm_static_runtime_nupkg_unzipped_directory_root = $nupkgs[0].Directory.FullName
-              $arm_static_runtime_nupkg_unzipped_directory = [System.IO.Path]::Combine($arm_static_runtime_nupkg_unzipped_directory_root, 'binaries', [System.IO.Path]::GetFileNameWithoutExtension($arm_static_runtime_nuget_package))
-              [System.IO.Compression.ZipFile]::ExtractToDirectory($arm_static_runtime_nuget_package, $arm_static_runtime_nupkg_unzipped_directory)
+             
 
               $x64_static_runtime_path_old = [System.IO.Path]::Combine($x64_static_runtime_nupkg_unzipped_directory, 'runtimes', 'win-x64', '_native')
               $x64_static_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-x64', '_native', 'static')
@@ -216,10 +183,7 @@ extends:
               $arm64_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm64', '_native')
               $arm64_static_runtime_path_old = [System.IO.Path]::Combine($arm64_static_runtime_nupkg_unzipped_directory, 'runtimes', 'win-arm64', '_native')
               $arm64_static_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm64', '_native', 'static')
-              $arm_runtime_path_old = [System.IO.Path]::Combine($arm_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native')
-              $arm_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native')
-              $arm_static_runtime_path_old = [System.IO.Path]::Combine($arm_static_runtime_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native')
-              $arm_static_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native', 'static')
+              
               $uap_build_path_old = [System.IO.Path]::Combine($x64_static_runtime_nupkg_unzipped_directory, 'build', 'native')
               $uap_build_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'build', 'uap10.0')
 
@@ -228,8 +192,6 @@ extends:
               New-Item -Path $x86_static_runtime_path_new -ItemType Directory
               New-Item -Path $arm64_runtime_path_new -ItemType Directory
               New-Item -Path $arm64_static_runtime_path_new -ItemType Directory
-              New-Item -Path $arm_runtime_path_new -ItemType Directory
-              New-Item -Path $arm_static_runtime_path_new -ItemType Directory
 
               Copy-Item ([System.IO.Path]::Combine($x86_runtime_path_old, 'onnxruntime.dll'))                    $x86_runtime_path_new
               Copy-Item ([System.IO.Path]::Combine($x86_runtime_path_old, 'onnxruntime.lib'))                    $x86_runtime_path_new
@@ -241,11 +203,6 @@ extends:
               Copy-Item ([System.IO.Path]::Combine($arm64_runtime_path_old, 'microsoft.ai.machinelearning.dll')) $arm64_runtime_path_new
               Copy-Item ([System.IO.Path]::Combine($arm64_runtime_path_old, 'microsoft.ai.machinelearning.lib')) $arm64_runtime_path_new
 
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'onnxruntime.dll'))                    $arm_runtime_path_new
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'onnxruntime.lib'))                    $arm_runtime_path_new
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'microsoft.ai.machinelearning.dll'))   $arm_runtime_path_new
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'microsoft.ai.machinelearning.lib'))   $arm_runtime_path_new
-
               Copy-Item ([System.IO.Path]::Combine($x64_static_runtime_path_old, 'onnxruntime.dll'))                    ([System.IO.Path]::Combine($x64_static_runtime_path_new, 'onnxruntime.dll'))
               Copy-Item ([System.IO.Path]::Combine($x64_static_runtime_path_old, 'onnxruntime.lib'))                    ([System.IO.Path]::Combine($x64_static_runtime_path_new, 'onnxruntime.lib'))
               Copy-Item ([System.IO.Path]::Combine($x64_static_runtime_path_old, 'microsoft.ai.machinelearning.dll'))   ([System.IO.Path]::Combine($x64_static_runtime_path_new, 'microsoft.ai.machinelearning.dll'))
@@ -261,11 +218,6 @@ extends:
               Copy-Item ([System.IO.Path]::Combine($arm64_static_runtime_path_old, 'microsoft.ai.machinelearning.dll')) ([System.IO.Path]::Combine($arm64_static_runtime_path_new, 'microsoft.ai.machinelearning.dll'))
               Copy-Item ([System.IO.Path]::Combine($arm64_static_runtime_path_old, 'microsoft.ai.machinelearning.lib')) ([System.IO.Path]::Combine($arm64_static_runtime_path_new, 'microsoft.ai.machinelearning.lib'))
 
-              Copy-Item ([System.IO.Path]::Combine($arm_static_runtime_path_old, 'onnxruntime.dll'))                    ([System.IO.Path]::Combine($arm_static_runtime_path_new, 'onnxruntime.dll'))
-              Copy-Item ([System.IO.Path]::Combine($arm_static_runtime_path_old, 'onnxruntime.lib'))                    ([System.IO.Path]::Combine($arm_static_runtime_path_new, 'onnxruntime.lib'))
-              Copy-Item ([System.IO.Path]::Combine($arm_static_runtime_path_old, 'microsoft.ai.machinelearning.dll'))   ([System.IO.Path]::Combine($arm_static_runtime_path_new, 'microsoft.ai.machinelearning.dll'))
-              Copy-Item ([System.IO.Path]::Combine($arm_static_runtime_path_old, 'microsoft.ai.machinelearning.lib'))   ([System.IO.Path]::Combine($arm_static_runtime_path_new, 'microsoft.ai.machinelearning.lib'))
-
               Copy-Item -Recurse $uap_build_path_old $uap_build_path_new
 
               $merged_nuget_path = [System.IO.Path]::Combine($Env:BUILD_ARTIFACTSTAGINGDIRECTORY, 'merged')
@@ -304,32 +256,20 @@ extends:
               $arm64_nupkg_unzipped_directory = [System.IO.Path]::Combine($arm64_nupkg_unzipped_directory_root, 'symbols', [System.IO.Path]::GetFileNameWithoutExtension($arm64_nuget_package))
               [System.IO.Compression.ZipFile]::ExtractToDirectory($arm64_nuget_package, $arm64_nupkg_unzipped_directory)
 
-              $nupkgs = (Get-ChildItem ..\nuget-artifact-arm -Filter Microsoft.AI.MachineLearning*.snupkg -Recurse)
-              $arm_nuget_package = $nupkgs[0].FullName
-              $arm_nupkg_unzipped_directory_root = $nupkgs[0].Directory.FullName
-              $arm_nupkg_unzipped_directory = [System.IO.Path]::Combine($arm_nupkg_unzipped_directory_root, 'symbols', [System.IO.Path]::GetFileNameWithoutExtension($arm_nuget_package))
-              [System.IO.Compression.ZipFile]::ExtractToDirectory($arm_nuget_package, $arm_nupkg_unzipped_directory)
-
               $x86_runtime_path_old = [System.IO.Path]::Combine($x86_nupkg_unzipped_directory, 'runtimes', 'win-x86', '_native')
               $x86_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-x86', '_native')
               $arm64_runtime_path_old = [System.IO.Path]::Combine($arm64_nupkg_unzipped_directory, 'runtimes', 'win-arm64', '_native')
               $arm64_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm64', '_native')
-              $arm_runtime_path_old = [System.IO.Path]::Combine($arm_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native')
-              $arm_runtime_path_new = [System.IO.Path]::Combine($x64_nupkg_unzipped_directory, 'runtimes', 'win-arm', '_native')
-
+              
               New-Item -Path $x86_runtime_path_new -ItemType Directory
               New-Item -Path $arm64_runtime_path_new -ItemType Directory
-              New-Item -Path $arm_runtime_path_new -ItemType Directory
 
               Copy-Item ([System.IO.Path]::Combine($x86_runtime_path_old, 'onnxruntime.pdb'))                    $x86_runtime_path_new
               Copy-Item ([System.IO.Path]::Combine($x86_runtime_path_old, 'microsoft.ai.machinelearning.pdb'))   $x86_runtime_path_new
 
               Copy-Item ([System.IO.Path]::Combine($arm64_runtime_path_old, 'onnxruntime.pdb'))                  $arm64_runtime_path_new
               Copy-Item ([System.IO.Path]::Combine($arm64_runtime_path_old, 'microsoft.ai.machinelearning.pdb')) $arm64_runtime_path_new
 
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'onnxruntime.pdb'))                    $arm_runtime_path_new
-              Copy-Item ([System.IO.Path]::Combine($arm_runtime_path_old, 'microsoft.ai.machinelearning.pdb'))   $arm_runtime_path_new
-
               $merged_nuget_path = [System.IO.Path]::Combine($Env:BUILD_ARTIFACTSTAGINGDIRECTORY, 'merged')
               if (!(Test-Path $merged_nuget_path)) {
                   New-Item -Path $merged_nuget_path -ItemType Directory

cmake/CMakeLists.txt

@@ -354,13 +354,7 @@ if (onnxruntime_USE_ROCM)
   endif()
 endif()
 
-if (APPLE)
-    if (NOT CMAKE_OSX_ARCHITECTURES)
-        message("Building ONNX Runtime for ${CMAKE_HOST_SYSTEM_PROCESSOR}")
-    endif()
-elseif (NOT WIN32 AND NOT APPLE)
-    message("Building ONNX Runtime for ${CMAKE_SYSTEM_PROCESSOR}")
-endif()
+
 
 # Single output director for all binaries
 set(RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin CACHE PATH "Single output directory for all binaries.")
@@ -493,6 +487,14 @@ endif()
 
 include(adjust_global_compile_flags.cmake)
 
+if (APPLE)
+  if (NOT CMAKE_OSX_ARCHITECTURES)
+    message("Building ONNX Runtime for ${CMAKE_HOST_SYSTEM_PROCESSOR} CPU ARCH")
+  endif()
+elseif (NOT WIN32 AND NOT APPLE)
+  message("Building ONNX Runtime for ${onnxruntime_target_platform} CPU ARCH")
+endif()
+
 # We need to link with libatomic on systems that do not have built-in atomics, or
 # don't have built-in support for 8 byte atomics
 # Derived from https://github.com/protocolbuffers/protobuf/blob/master/cmake/CMakeLists.txt
@@ -639,7 +641,16 @@ else()
   check_cxx_compiler_flag(-Wunused-variable HAS_UNUSED_VARIABLE)
   check_cxx_compiler_flag(-Wuseless-cast HAS_USELESS_CAST)
   check_function_exists(reallocarray HAS_REALLOCARRAY)
-
+  if (NOT APPLE AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten" AND onnxruntime_target_platform STREQUAL "aarch64")
+   check_cxx_compiler_flag(-march=armv8.2-a+bf16 HAS_ARM64_BFLOAT16)
+   if(NOT HAS_ARM64_BFLOAT16)
+     message(FATAL_ERROR  "The compiler doesn't support BFLOAT16!!!")
+   endif()
+   check_cxx_compiler_flag(-march=armv8.2-a+fp16 HAS_ARM64_FLOAT16)
+   if(NOT HAS_ARM64_FLOAT16)
+     message(FATAL_ERROR  "The compiler doesn't support FLOAT16!!!")
+   endif()
+  endif()
   if (HAS_TAUTOLOGICAL_POINTER_COMPARE)
     #we may have extra null pointer checkings in debug build, it's not an issue
     list(APPEND ORT_WARNING_FLAGS -Wno-tautological-pointer-compare)

cmake/adjust_global_compile_flags.cmake

@@ -300,6 +300,31 @@ if (MSVC)
   endif()
 else()
   if (NOT APPLE)
+    #XXX: Sometimes the value of CMAKE_SYSTEM_PROCESSOR is set but it's wrong. For example, if you run an armv7 docker
+    #image on an aarch64 machine with an aarch64 Ubuntu host OS, in the docker instance cmake may still report
+    # CMAKE_SYSTEM_PROCESSOR as aarch64 by default. Given compiling this code may need more than 2GB memory, we do not
+    # support compiling for ARM32 natively(only support cross-compiling), we will ignore this issue for now.
+    if(NOT CMAKE_SYSTEM_PROCESSOR)
+      message(WARNING "CMAKE_SYSTEM_PROCESSOR is not set. Please set it in your toolchain cmake file.")
+      # Try to detect it
+      if("${CMAKE_C_COMPILER_ID}" STREQUAL "GNU" OR "${CMAKE_C_COMPILER_ID}" STREQUAL "Clang")
+        execute_process(
+		COMMAND "${CMAKE_C_COMPILER}" -dumpmachine
+		OUTPUT_VARIABLE GCC_DUMP_MACHINE_OUT OUTPUT_STRIP_TRAILING_WHITESPACE
+		ERROR_VARIABLE _err
+		RESULT_VARIABLE _res
+		)
+		if(NOT _res EQUAL 0)
+			message(SEND_ERROR "Failed to run 'gcc -dumpmachine':\n ${_res}")
+		endif()
+		string(REPLACE "-" ";" GCC_DUMP_MACHINE_OUT_LIST "${GCC_DUMP_MACHINE_OUT}")
+		list(LENGTH GCC_DUMP_MACHINE_OUT_LIST GCC_TRIPLET_LEN)
+		if(GCC_TRIPLET_LEN EQUAL 4)
+		  list(GET GCC_DUMP_MACHINE_OUT_LIST 0 CMAKE_SYSTEM_PROCESSOR)
+          message("Setting CMAKE_SYSTEM_PROCESSOR to ${CMAKE_SYSTEM_PROCESSOR}")
+        endif()
+      endif()
+    endif()
     set(onnxruntime_target_platform ${CMAKE_SYSTEM_PROCESSOR})
   endif()
   if (onnxruntime_BUILD_FOR_NATIVE_MACHINE)

js/web/lib/build-def.d.ts

@@ -18,6 +18,10 @@ interface BuildDefinitions {
    * defines whether to disable the whole WebGpu backend in the build.
    */
   readonly DISABLE_WEBGPU: boolean;
+  /**
+   * defines whether to disable the whole WebNN backend in the build.
+   */
+  readonly DISABLE_WEBNN: boolean;
   /**
    * defines whether to disable the whole WebAssembly backend in the build.
    */

js/web/lib/index.ts

@@ -28,7 +28,9 @@ if (!BUILD_DEFS.DISABLE_WASM) {
   registerBackend('wasm', wasmBackend, 10);
   if (BUILD_DEFS.DISABLE_TRAINING) {
     registerBackend('xnnpack', wasmBackend, 9);
-    registerBackend('webnn', wasmBackend, 9);
+    if (!BUILD_DEFS.DISABLE_WEBNN) {
+      registerBackend('webnn', wasmBackend, 9);
+    }
   }
 }
 

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

@@ -2,7 +2,7 @@
 // Licensed under the MIT License.
 
 import {argMax, argMin, parseArgMinMaxAttributes} from './ops/argminmax';
-import {attention, parseAttentionAttributes} from './ops/attention';
+import {attention} from './ops/attention';
 import {batchNorm} from './ops/batch-norm';
 import {biasAdd} from './ops/bias-add';
 import {biasSplitGelu} from './ops/bias-split-gelu';
@@ -17,10 +17,10 @@ import {gather, parseGatherAttributes} from './ops/gather';
 import {gatherElements, parseGatherElementsAttributes} from './ops/gather-elements';
 import {gemm, parseGemmAttributes} from './ops/gemm';
 import {instanceNorm, parseInstanceNormAttributes} from './ops/instance-norm';
-import {layerNorm, parseLayerNormAttributes} from './ops/layer-norm';
+import {layerNorm} from './ops/layer-norm';
 import {matMul} from './ops/matmul';
 import {multiHeadAttention, parseMultiHeadAttentionAttributes} from './ops/multi-head-attentiion';
-import {pad, parsePadAttributes} from './ops/pad';
+import {pad} from './ops/pad';
 import * as pool from './ops/pool';
 import {range} from './ops/range';
 import {reduceL1, reduceL2, reduceLogSum, reduceLogSumExp, reduceMax, reduceMean, reduceMin, reduceProd, reduceSum, reduceSumSquare} from './ops/reduce';
@@ -50,7 +50,7 @@ export const WEBGPU_OP_RESOLVE_RULES: Map<string, OperatorImplementation> = new
   ['Asinh', [unaryOps.asinh]],
   ['Atan', [unaryOps.atan]],
   ['Atanh', [unaryOps.atanh]],
-  ['Attention', [attention, parseAttentionAttributes]],
+  ['Attention', [attention]],
   // TODO: support new attributes for AveragePool-10
   ['AveragePool', [pool.averagePool, pool.parseAveragePoolAttributes]],
   ['BatchNormalization', [batchNorm]],
@@ -83,7 +83,7 @@ export const WEBGPU_OP_RESOLVE_RULES: Map<string, OperatorImplementation> = new
   ['Greater', [binaryOps.greater]],
   ['GreaterOrEqual', [binaryOps.greaterOrEqual]],
   ['InstanceNormalization', [instanceNorm, parseInstanceNormAttributes]],
-  ['LayerNormalization', [layerNorm, parseLayerNormAttributes]],
+  ['LayerNormalization', [layerNorm]],
   ['LeakyRelu', [unaryOps.leakyRelu, unaryOps.parseAlphaAttributes]],
   ['Less', [binaryOps.less]],
   ['LessOrEqual', [binaryOps.lessOrEqual]],
@@ -95,7 +95,7 @@ export const WEBGPU_OP_RESOLVE_RULES: Map<string, OperatorImplementation> = new
   ['MultiHeadAttention', [multiHeadAttention, parseMultiHeadAttentionAttributes]],
   ['Neg', [unaryOps.neg]],
   ['Not', [unaryOps.not]],
-  ['Pad', [pad, parsePadAttributes]],
+  ['Pad', [pad]],
   ['Pow', [binaryOps.pow]],
   ['Range', [range]],
   ['Reciprocal', [unaryOps.reciprocal]],