[Arm64] SIMD simple defines (#14628)

* [Arm64] SIMD simple defines

* Fix #else

5 files changed, 23 insertions, 6 deletions

diff --git a/src/jit/compiler.h b/src/jit/compiler.h
index d05b894c58..3ff6375a57 100644
--- a/src/jit/compiler.h
+++ b/src/jit/compiler.h
@@ -137,8 +137,12 @@ unsigned ReinterpretHexAsDecimal(unsigned);
 
 /*****************************************************************************/
 
-#ifdef FEATURE_SIMD
+#if defined(FEATURE_SIMD)
+#if defined(_TARGET_XARCH_)
 const unsigned TEMP_MAX_SIZE = YMM_REGSIZE_BYTES;
+#elif defined(_TARGET_ARM64_)
+const unsigned TEMP_MAX_SIZE = FP_REGSIZE_BYTES;
+#endif // defined(_TARGET_XARCH_) || defined(_TARGET_ARM64_)
 #else  // !FEATURE_SIMD
 const unsigned TEMP_MAX_SIZE = sizeof(double);
 #endif // !FEATURE_SIMD
@@ -7669,6 +7673,8 @@ private:
             assert(canUseSSE2());
             return TYP_SIMD16;
         }
+#elif defined(_TARGET_ARM64_)
+        return TYP_SIMD16;
 #else
         assert(!"getSIMDVectorType() unimplemented on target arch");
         unreached();
@@ -7705,6 +7711,8 @@ private:
             assert(canUseSSE2());
             return XMM_REGSIZE_BYTES;
         }
+#elif defined(_TARGET_ARM64_)
+        return FP_REGSIZE_BYTES;
 #else
         assert(!"getSIMDVectorRegisterByteLength() unimplemented on target arch");
         unreached();
diff --git a/src/jit/ee_il_dll.cpp b/src/jit/ee_il_dll.cpp
index 0cfc565679..822a2b9286 100644
--- a/src/jit/ee_il_dll.cpp
+++ b/src/jit/ee_il_dll.cpp
@@ -384,8 +384,7 @@ unsigned CILJit::getMaxIntrinsicSIMDVectorLength(CORJIT_FLAGS cpuCompileFlags)
     jitFlags.SetFromFlags(cpuCompileFlags);
 
 #ifdef FEATURE_SIMD
-#ifdef _TARGET_XARCH_
-#ifndef LEGACY_BACKEND
+#if defined(_TARGET_XARCH_) && !defined(LEGACY_BACKEND)
     if (!jitFlags.IsSet(JitFlags::JIT_FLAG_PREJIT) && jitFlags.IsSet(JitFlags::JIT_FLAG_FEATURE_SIMD) &&
         jitFlags.IsSet(JitFlags::JIT_FLAG_USE_AVX2))
     {
@@ -398,13 +397,12 @@ unsigned CILJit::getMaxIntrinsicSIMDVectorLength(CORJIT_FLAGS cpuCompileFlags)
             return 32;
         }
     }
-#endif // !LEGACY_BACKEND
+#endif // !(defined(_TARGET_XARCH_) && !defined(LEGACY_BACKEND))
     if (GetJitTls() != nullptr && JitTls::GetCompiler() != nullptr)
     {
         JITDUMP("getMaxIntrinsicSIMDVectorLength: returning 16\n");
     }
     return 16;
-#endif // _TARGET_XARCH_
 #else  // !FEATURE_SIMD
     if (GetJitTls() != nullptr && JitTls::GetCompiler() != nullptr)
     {
diff --git a/src/jit/lclvars.cpp b/src/jit/lclvars.cpp
index 78092344da..8ac57da52d 100644
--- a/src/jit/lclvars.cpp
+++ b/src/jit/lclvars.cpp
@@ -1460,9 +1460,13 @@ void Compiler::lvaCanPromoteStructType(CORINFO_CLASS_HANDLE    typeHnd,
         // In the future this may be changing to XMM_REGSIZE_BYTES.
         // Note: MaxOffset is used below to declare a local array, and therefore must be a compile-time constant.
         CLANG_FORMAT_COMMENT_ANCHOR;
-#ifdef FEATURE_SIMD
+#if defined(FEATURE_SIMD)
+#if defined(_TARGET_XARCH_)
         // This will allow promotion of 2 Vector<T> fields on AVX2, or 4 Vector<T> fields on SSE2.
         const int MaxOffset = MAX_NumOfFieldsInPromotableStruct * XMM_REGSIZE_BYTES;
+#elif defined(_TARGET_ARM64_)
+        const int MaxOffset = MAX_NumOfFieldsInPromotableStruct * FP_REGSIZE_BYTES;
+#endif // defined(_TARGET_XARCH_) || defined(_TARGET_ARM64_)
 #else  // !FEATURE_SIMD
         const int MaxOffset = MAX_NumOfFieldsInPromotableStruct * sizeof(double);
 #endif // !FEATURE_SIMD
diff --git a/src/jit/simd.cpp b/src/jit/simd.cpp
index 8605781c2d..d190cd29f4 100644
--- a/src/jit/simd.cpp
+++ b/src/jit/simd.cpp
@@ -98,6 +98,8 @@ int Compiler::getSIMDTypeAlignment(var_types simdType)
         assert(size == 32);
         return 32;
     }
+#elif defined(_TARGET_ARM64_)
+    return 16;
 #else
     assert(!"getSIMDTypeAlignment() unimplemented on target arch");
     unreached();
diff --git a/src/jit/target.h b/src/jit/target.h
index 1c352d0c1a..37919e2422 100644
--- a/src/jit/target.h
+++ b/src/jit/target.h
@@ -1529,6 +1529,11 @@ typedef unsigned short regPairNoSmall; // arm: need 12 bits
   #define CPBLK_UNROLL_LIMIT       64      // Upper bound to let the code generator to loop unroll CpBlk.
   #define INITBLK_UNROLL_LIMIT     64      // Upper bound to let the code generator to loop unroll InitBlk.
 
+#ifdef FEATURE_SIMD
+  #define ALIGN_SIMD_TYPES         1       // whether SIMD type locals are to be aligned
+  #define FEATURE_PARTIAL_SIMD_CALLEE_SAVE 1 // Whether SIMD registers are partially saved at calls
+#endif // FEATURE_SIMD
+
   #define FEATURE_WRITE_BARRIER    1       // Generate the proper WriteBarrier calls for GC    
   #define FEATURE_FIXED_OUT_ARGS   1       // Preallocate the outgoing arg area in the prolog
   #define FEATURE_STRUCTPROMOTE    1       // JIT Optimization to promote fields of structs into registers