Enable CPUINFO for all Windows build (microsoft#19655)

### Description It was disabled in PR microsoft#9065. And the reason was: " api-ms-win-core-kernel32-legacy-*.dll wasn't available in Windows 8 and was added in Windows 10, so cpuinfo breaks our Windows 8 support. I'm disabling it again." We no longer support Windows 8. Therefore we can add CPUINFO back. ### Motivation and Context To make the code simpler. If in any case the library doesn't work as expected, we can submit a PR to their code base and fix it.
zz002 · Mar 7, 2024 · e3944c9 · e3944c9
1 parent 15be056
commit e3944c9
Show file tree

Hide file tree

Showing 4 changed files with 42 additions and 73 deletions.
diff --git a/cmake/external/onnxruntime_external_deps.cmake b/cmake/external/onnxruntime_external_deps.cmake
@@ -256,14 +256,7 @@ if (onnxruntime_ENABLE_CPUINFO)
       set(CPUINFO_SUPPORTED TRUE)
     endif()
     if (WIN32)
-      # Exclude Windows ARM build and Windows Store
-      if (${onnxruntime_target_platform} MATCHES "^(ARM.*|arm.*)$" )
-        message(WARNING "Cpuinfo not included for compilation problems with Windows ARM.")
-        set(CPUINFO_SUPPORTED FALSE)
-      elseif (WIN32 AND NOT CMAKE_CXX_STANDARD_LIBRARIES MATCHES kernel32.lib)
-        message(WARNING "Cpuinfo not included non-Desktop builds")
-        set(CPUINFO_SUPPORTED FALSE)
-      endif()
+      set(CPUINFO_SUPPORTED TRUE)
     elseif (NOT ${onnxruntime_target_platform} MATCHES "^(i[3-6]86|AMD64|x86(_64)?|armv[5-8].*|aarch64|arm64)$")
       message(WARNING
         "Target processor architecture \"${onnxruntime_target_platform}\" is not supported in cpuinfo. "

diff --git a/cmake/onnxruntime_common.cmake b/cmake/onnxruntime_common.cmake
@@ -201,18 +201,13 @@ endif()
 
 
 if (RISCV64 OR ARM64 OR ARM OR X86 OR X64 OR X86_64)
-  if((WIN32 AND NOT CMAKE_CXX_STANDARD_LIBRARIES MATCHES kernel32.lib) OR ((ARM64 OR ARM) AND MSVC))
-    # msvc compiler report syntax error with cpuinfo arm source files
-    # and cpuinfo does not have code for getting arm uarch info under windows
-  else()
     # Link cpuinfo if supported
     # Using it mainly in ARM with Android.
     # Its functionality in detecting x86 cpu features are lacking, so is support for Windows.
     if (CPUINFO_SUPPORTED)
       onnxruntime_add_include_to_target(onnxruntime_common cpuinfo::cpuinfo)
       list(APPEND onnxruntime_EXTERNAL_LIBRARIES cpuinfo::cpuinfo ${ONNXRUNTIME_CLOG_TARGET_NAME})
     endif()
-  endif()
 endif()
 
 if (NOT onnxruntime_BUILD_SHARED_LIB)

diff --git a/onnxruntime/core/common/cpuid_info.cc b/onnxruntime/core/common/cpuid_info.cc
@@ -52,6 +52,13 @@
 
 #if defined(CPUINFO_SUPPORTED)
 #include <cpuinfo.h>
+#if defined(CPUIDINFO_ARCH_ARM)
+namespace onnxruntime {
+// The following function is declared in "core/common/cpuid_uarch.h" but we cannot include the whole header file because
+//  some of its symbols are conflict with <cpuinfo.h>
+void decodeMIDR(uint32_t midr, uint32_t uarch[1]);
+}  // namespace onnxruntime
+#endif
 #else
 #include "core/common/cpuid_uarch.h"
 #endif  // CPUINFO_SUPPORTED
@@ -142,11 +149,6 @@ void CPUIDInfo::ArmLinuxInit() {
   // Pytorch CPUINFO only works on ARM linux or android
   // Assuming no hyper-threading, no NUMA groups
 #ifdef CPUINFO_SUPPORTED
-  pytorch_cpuinfo_init_ = cpuinfo_initialize();
-  if (!pytorch_cpuinfo_init_) {
-    LOGS_DEFAULT(WARNING) << "Failed to init pytorch cpuinfo library, may cause CPU EP performance degradation due to undetected CPU features.";
-    return;
-  }
   is_hybrid_ = cpuinfo_get_uarchs_count() > 1;
   has_arm_neon_dot_ = cpuinfo_has_arm_neon_dot();
   has_fp16_ = cpuinfo_has_arm_neon_fp16_arith();
@@ -239,52 +241,24 @@ void CPUIDInfo::ArmWindowsInit() {
       lastUarch = uarch;
     }
   }
-
-  switch (lastUarch) {
-    case cpuinfo_uarch_cortex_a55:
-    case cpuinfo_uarch_cortex_a55r0:
-    case cpuinfo_uarch_cortex_a76:
-    case cpuinfo_uarch_neoverse_n1:
-    case cpuinfo_uarch_cortex_a77:
-    case cpuinfo_uarch_exynos_m4:
-    case cpuinfo_uarch_exynos_m5:
-      has_fp16_ = true;
-      break;
-    default:
-      break;
-  }
-  if (!has_fp16_) {
-    /*
-     * Detecting fp16 support. Different cores should have the same instruction set.
-     * So we just check the first ID_AA64PFR0_EL1
-     *  Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0100), Op2(0b000),
-     */
-    uint64_t ID_AA64PFR0_EL1;
-    unsigned long valsize = sizeof(uint64_t);
-    auto retCode = ::RegGetValueA(
-        HKEY_LOCAL_MACHINE,
-        "HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\0",
-        "CP 4020", RRF_RT_REG_QWORD, nullptr,
-        &ID_AA64PFR0_EL1, &valsize);
-    if (retCode == ERROR_SUCCESS) {
-      // AdvSIMD, bits [23:20]
-      auto advSimd = ID_AA64PFR0_EL1 >> 20;
-      if ((advSimd & 0xfULL) == 1) {
-        has_fp16_ = true;
-      }
-    }
-  }
 #endif /* Application Family or OneCore Family */
 
   has_arm_neon_dot_ = (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE) != 0);
 #else
   has_arm_neon_dot_ = false;
 #endif
-  has_fp16_ |= has_arm_neon_dot_;
-  /* TODO: implement them when hw+sw is available for testing these features */
-  has_arm_neon_i8mm_ = false;
-  has_arm_sve_i8mm_ = false;
-  has_arm_neon_bf16_ = false;
+
+  if (pytorch_cpuinfo_init_) {
+    has_fp16_ = cpuinfo_has_arm_neon_fp16_arith();
+    has_arm_neon_i8mm_ = cpuinfo_has_arm_i8mm();
+    has_arm_sve_i8mm_ = cpuinfo_has_arm_sve() && cpuinfo_has_arm_i8mm();
+    has_arm_neon_bf16_ = cpuinfo_has_arm_neon_bf16();
+  } else {
+    has_fp16_ = false;
+    has_arm_neon_i8mm_ = false;
+    has_arm_sve_i8mm_ = false;
+    has_arm_neon_bf16_ = false;
+  }
 }
 
 #endif /* (arm or arm64) and windows */
@@ -304,5 +278,21 @@ uint32_t CPUIDInfo::GetCurrentCoreIdx() const {
   return 0xFFFFFFFF;  // don't know how to get core index
 #endif
 }
-
+CPUIDInfo::CPUIDInfo() {
+#ifdef CPUIDINFO_ARCH_X86
+  X86Init();
+#elif defined(CPUIDINFO_ARCH_ARM)
+#if CPUINFO_SUPPORTED
+  pytorch_cpuinfo_init_ = cpuinfo_initialize();
+  if (!pytorch_cpuinfo_init_) {
+    LOGS_DEFAULT(WARNING) << "Failed to init pytorch cpuinfo library, may cause CPU EP performance degradation due to undetected CPU features.";
+  }
+#endif
+#ifdef __linux__
+  ArmLinuxInit();
+#elif defined(_WIN32)
+  ArmWindowsInit();
+#endif /* (arm or arm64) and windows */
+#endif
+}
 }  // namespace onnxruntime
diff --git a/onnxruntime/core/common/cpuid_info.h b/onnxruntime/core/common/cpuid_info.h
@@ -93,17 +93,7 @@ class CPUIDInfo {
   }
 
  private:
-  CPUIDInfo() {
-#ifdef CPUIDINFO_ARCH_X86
-    X86Init();
-#elif defined(CPUIDINFO_ARCH_ARM)
-#ifdef __linux__
-    ArmLinuxInit();
-#elif defined(_WIN32)
-    ArmWindowsInit();
-#endif /* (arm or arm64) and windows */
-#endif
-  }
+  CPUIDInfo();
   bool has_amx_bf16_{false};
   bool has_avx_{false};
   bool has_avx2_{false};
@@ -131,19 +121,20 @@ class CPUIDInfo {
 #ifdef CPUIDINFO_ARCH_X86
 
   void X86Init();
-
 #elif defined(CPUIDINFO_ARCH_ARM)
+  // Now the following var is only used in ARM build, but later one we may expand the usage.
+  bool pytorch_cpuinfo_init_{false};
+#endif
+
 #ifdef __linux__
 
-  bool pytorch_cpuinfo_init_{false};
   void ArmLinuxInit();
 
 #elif defined(_WIN32)
 
   void ArmWindowsInit();
 
 #endif /* (arm or arm64) and windows */
-#endif
 };
 
 }  // namespace onnxruntime