Update Intel Thread Counts (microsoft#22894)

### Description
The default thread count methodology by onnxruntime did not account for
new upcoming Intel microarchitectures leading to a suboptimal thread
count. Optimizing the thread count for new Intel microarchitectures
reveal gains on the majority of models across datatypes and shows gains
up to ~1.5x speedup.


### Motivation and Context
Applications should run on Intel with the most performant thread
configuration for the majority of models. With new microarchitectures,
adjusting the thread count methodology is required to take advantage of
their differences.
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->

Author: A-Satti

onnxruntime/core/platform/windows/hardware_core_enumerator.cc

@@ -1,7 +1,8 @@
-// Copyright (c) Microsoft Corporation. All rights reserved.
+// Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
 #include "hardware_core_enumerator.h"
+#include "core/platform/windows/env.h"
 #include <memory>
 #include <Windows.h>
 #include <assert.h>
@@ -83,6 +84,38 @@ uint32_t HardwareCoreEnumerator::DefaultIntraOpNumThreads() {
   // # of physical cores = # of P cores + # of E Cores + # of Soc Cores.
   // # of logical cores = # of P cores x 2 (if hyper threading is enabled) + # of E cores + # of Soc Cores.
   auto cores = GetCoreInfo();
+#if !defined(_M_ARM64EC) && !defined(_M_ARM64) && !defined(__aarch64__)
+  const int kVendorID_Intel[3] = {0x756e6547, 0x6c65746e, 0x49656e69};  // "GenuntelineI"
+  bool isIntelSpecifiedPlatform = false;
+  const int kVendorID_IntelSpecifiedPlatformIDs[3] = {
+      // ExtendedModel, ExtendedFamily, Family Code, and Model Number
+      0xa06a,  // MTL
+      0xc065,  // ARL-H
+      0xb065   // ARL-U
+  };
+
+  int regs_leaf0[4];
+  int regs_leaf1[4];
+  __cpuid(regs_leaf0, 0);
+  __cpuid(regs_leaf1, 0x1);
+
+  auto isIntel = (kVendorID_Intel[0] == regs_leaf0[1]) && (kVendorID_Intel[1] == regs_leaf0[2]) && (kVendorID_Intel[2] == regs_leaf0[3]);
+
+  for (int intelSpecifiedPlatform : kVendorID_IntelSpecifiedPlatformIDs) {
+    if ((regs_leaf1[0] >> 4) == intelSpecifiedPlatform) {
+      isIntelSpecifiedPlatform = true;
+    }
+  }
+
+  if (isIntel) {
+    if (isIntelSpecifiedPlatform) {
+      // We want to exclude cores without an LLC
+      return cores.LLCCores;
+    } else {
+      return cores.PhysicalCores;
+    }
+  }
+#endif
 
   return cores.LLCCores;
 }

tools/ci_build/build.py

@@ -1559,10 +1559,6 @@ def generate_build_tree(
                 # The "/profile" flag implies "/DEBUG:FULL /DEBUGTYPE:cv,fixup /OPT:REF /OPT:NOICF /INCREMENTAL:NO /FIXED:NO". We set it for satisfying a Microsoft internal compliance requirement. External users
                 # do not need to have it.
                 ldflags = ["/profile", "/DYNAMICBASE"]
-                # Address Sanitizer libs do not have a Qspectre version. So they two cannot be both enabled.
-                if not args.enable_address_sanitizer:
-                    # Also enable a special perf patch that was made for Intel Meteor Lake mobile CPUs
-                    cflags += ["/Qspectre", "/DONNXRUNTIME_ENABLE_INTEL_METEOR_LAKE_MOBILE_PLATFORM_PERF_PATCH"]
                 if config == "Release":
                     cflags += ["/O2", "/Ob2", "/DNDEBUG"]
                 elif config == "RelWithDebInfo":

winml/lib/Api/HardwareCoreEnumerator.cpp

@@ -1,8 +1,8 @@
-// Copyright (c) Microsoft Corporation. All rights reserved.
+
+// Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
 #include "lib/Api/pch/pch.h"
-
 #include "HardwareCoreEnumerator.h"
 
 namespace WINMLP {
@@ -88,22 +88,33 @@ uint32_t HardwareCoreEnumerator::DefaultIntraOpNumThreads() {
 
 #if !defined(_M_ARM64EC) && !defined(_M_ARM64) && !defined(__aarch64__)
   const int kVendorID_Intel[3] = {0x756e6547, 0x6c65746e, 0x49656e69};  // "GenuntelineI"
+  bool isIntelSpecifiedPlatform = false;
+  const int kVendorID_IntelSpecifiedPlatformIDs[3] = {
+    // ExtendedModel,ExtendedFamily,Family Code, and Model Number
+    0xa06a,  // MTL
+    0xc065,  // ARL-H
+    0xb065   // ARL-U
+  };
+
   int regs_leaf0[4];
-  int regs_leaf7[4];
+  int regs_leaf1[4];
   __cpuid(regs_leaf0, 0);
-  __cpuid(regs_leaf7, 0x7);
+  __cpuid(regs_leaf1, 0x1);
 
   auto isIntel = (kVendorID_Intel[0] == regs_leaf0[1]) && (kVendorID_Intel[1] == regs_leaf0[2]) &&
     (kVendorID_Intel[2] == regs_leaf0[3]);
 
-  auto isHybrid = (regs_leaf7[3] & (1 << 15));
+  for (int intelSpecifiedPlatform : kVendorID_IntelSpecifiedPlatformIDs) {
+    if ((regs_leaf1[0] >> 4) == intelSpecifiedPlatform) {
+      isIntelSpecifiedPlatform = true;
+    }
+  }
 
-  if (isIntel && isHybrid) {
+  if (isIntel && isIntelSpecifiedPlatform) {
     // We want to use the number of physical cores, but exclude cores without an LLC
     return cores.LLCCores;
   }
 #endif
-
   return cores.PhysicalCores;
 }