86 lines · plain
1#define WIN32_LEAN_AND_MEAN2#include <windows.h>3#include <processthreadsapi.h>4#include <stdint.h>5 6#ifndef PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE7#define PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE 43 8#endif9#ifndef PF_ARM_V83_JSCVT_INSTRUCTIONS_AVAILABLE10#define PF_ARM_V83_JSCVT_INSTRUCTIONS_AVAILABLE 4411#endif12#ifndef PF_ARM_V83_LRCPC_INSTRUCTIONS_AVAILABLE13#define PF_ARM_V83_LRCPC_INSTRUCTIONS_AVAILABLE 4514#endif15#ifndef PF_ARM_SVE_INSTRUCTIONS_AVAILABLE16#define PF_ARM_SVE_INSTRUCTIONS_AVAILABLE 4617#endif18#ifndef PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE19#define PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE 4720#endif21#ifndef PF_ARM_SVE_PMULL128_INSTRUCTIONS_AVAILABLE22#define PF_ARM_SVE_PMULL128_INSTRUCTIONS_AVAILABLE 5023#endif24#ifndef PF_ARM_SVE_SHA3_INSTRUCTIONS_AVAILABLE25#define PF_ARM_SVE_SHA3_INSTRUCTIONS_AVAILABLE 5526#endif27#ifndef PF_ARM_SVE_SM4_INSTRUCTIONS_AVAILABLE28#define PF_ARM_SVE_SM4_INSTRUCTIONS_AVAILABLE 5629#endif30#ifndef PF_ARM_SVE_I8MM_INSTRUCTIONS_AVAILABLE31#define PF_ARM_SVE_I8MM_INSTRUCTIONS_AVAILABLE 5732#endif33#ifndef PF_ARM_SVE_F32MM_INSTRUCTIONS_AVAILABLE34#define PF_ARM_SVE_F32MM_INSTRUCTIONS_AVAILABLE 5835#endif36#ifndef PF_ARM_SVE_F64MM_INSTRUCTIONS_AVAILABLE37#define PF_ARM_SVE_F64MM_INSTRUCTIONS_AVAILABLE 5938#endif39 40void __init_cpu_features_resolver(unsigned long hwcap,41 const __ifunc_arg_t *arg) {}42 43void CONSTRUCTOR_ATTRIBUTE __init_cpu_features(void) {44 if (__atomic_load_n(&__aarch64_cpu_features.features, __ATOMIC_RELAXED))45 return;46 47#define setCPUFeature(F) features |= 1ULL << F48 49 uint64_t features = 0;50 51 setCPUFeature(FEAT_INIT);52 setCPUFeature(FEAT_FP);53 54 // https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-isprocessorfeaturepresent55 if (IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE)) {56 setCPUFeature(FEAT_SHA2);57 setCPUFeature(FEAT_PMULL);58 }59 60 static const struct ProcessFeatureToFeatMap_t {61 int WinApiFeature;62 enum CPUFeatures CPUFeature;63 } FeatMap[] = {64 {PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE, FEAT_CRC},65 {PF_ARM_V81_ATOMIC_INSTRUCTIONS_AVAILABLE, FEAT_LSE},66 {PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE, FEAT_DOTPROD},67 {PF_ARM_V83_JSCVT_INSTRUCTIONS_AVAILABLE, FEAT_JSCVT},68 {PF_ARM_V83_LRCPC_INSTRUCTIONS_AVAILABLE, FEAT_RCPC},69 {PF_ARM_SVE_INSTRUCTIONS_AVAILABLE, FEAT_SVE},70 {PF_ARM_SVE2_INSTRUCTIONS_AVAILABLE, FEAT_SVE2},71 {PF_ARM_SVE_PMULL128_INSTRUCTIONS_AVAILABLE, FEAT_SVE_PMULL128},72 {PF_ARM_SVE_SHA3_INSTRUCTIONS_AVAILABLE, FEAT_SVE_SHA3},73 {PF_ARM_SVE_SM4_INSTRUCTIONS_AVAILABLE, FEAT_SVE_SM4},74 {PF_ARM_SVE_F32MM_INSTRUCTIONS_AVAILABLE, FEAT_SVE_F32MM},75 {PF_ARM_SVE_F64MM_INSTRUCTIONS_AVAILABLE, FEAT_SVE_F64MM},76 // There is no I8MM flag, but when SVE_I8MM is available, I8MM is too.77 {PF_ARM_SVE_I8MM_INSTRUCTIONS_AVAILABLE, FEAT_I8MM},78 };79 80 for (size_t I = 0, E = sizeof(FeatMap) / sizeof(FeatMap[0]); I != E; ++I)81 if (IsProcessorFeaturePresent(FeatMap[I].WinApiFeature))82 setCPUFeature(FeatMap[I].CPUFeature);83 84 __atomic_store(&__aarch64_cpu_features.features, &features, __ATOMIC_RELAXED);85}86