Read AA64ISAR0_EL1 to check dot product support (#16082)

skandhurkat · Skand Hurkat · web-flow · commit b28e927ca4ee · 2023-05-25T17:05:30.000-07:00
### Description Use an assembly instruction to read the `AA64ISAR0_EL1` register for dot product support. ### Motivation and Context The only reliable way to check for supported instruction extensions in ARM is to query the instruction set attribute registers. [Dot product instructions can be checked using bits 47:44 in the AA64ISAR0_EL1 register](https://developer.arm.com/documentation/ddi0601/2021-12/AArch64-Registers/ID-AA64ISAR0-EL1--AArch64-Instruction-Set-Attribute-Register-0?lang=en#fieldset_0-47_44). On `qemu-aarch64` with the `a64fx` cpu which does not support the dot product instructions, running a quantized BERT-Large (from MLPerf) results in `SIGILL`. With the change, the program continues without using the dot product instructions. Also verified that `S8S8_SDOT` kernels are invoked when running on hardware that supports dot product instructions. --------- Co-authored-by: Skand Hurkat <skhurkat@microsoft.com>
diff --git a/onnxruntime/core/mlas/lib/platform.cpp b/onnxruntime/core/mlas/lib/platform.cpp
@@ -55,7 +55,7 @@ MLASCPUIDInfo::MLASCPUIDInfo()
 #if defined(BUILD_MLAS_NO_ONNXRUNTIME)
 MLASCPUIDInfo::MLASCPUIDInfo()
 {
-    has_arm_neon_dot_ = ((getauxval(AT_HWCAP) & HWCAP_ASIMDDP) != 0); 
+    has_arm_neon_dot_ = ((getauxval(AT_HWCAP) & HWCAP_ASIMDDP) != 0);
 
     // raw hack! Need CPUIDInfo implementation for more precise detection
     has_fp16_ = has_arm_neon_dot_;
@@ -447,10 +447,10 @@ Return Value:
 
 #if defined(_WIN32)
     HasDotProductInstructions = (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE) != 0);
-#elif defined(__linux__)
-    HasDotProductInstructions = MLAS_CPUIDINFO::GetCPUIDInfo().HasArmNeonDot();
 #else
-    HasDotProductInstructions = false;
+    uint64_t isar0_el1;
+    asm("mrs %[reg], ID_AA64ISAR0_EL1\n" : [reg] "=r"(isar0_el1) : :);
+    HasDotProductInstructions = ((isar0_el1 >> 44) & 0xfu) == 0x1u;
 #endif
 
     if (HasDotProductInstructions) {
