Add support for Sve.UnzipEven/Odd & Sve.ZipHighLow (dotnet#101294)

* Add support for Sve.UnzipEven/Odd & Sve.ZipHighLow * Rename the test template --------- Co-authored-by: Kunal Pathak <[email protected]>
matouskozak · Apr 30, 2024 · 9a88fc2 · 9a88fc2
1 parent 2f23254
commit 9a88fc2
Show file tree

Hide file tree

Showing 7 changed files with 996 additions and 34 deletions.
diff --git a/src/coreclr/jit/hwintrinsiccodegenarm64.cpp b/src/coreclr/jit/hwintrinsiccodegenarm64.cpp
@@ -1443,6 +1443,15 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
                 break;
             }
 
+            case NI_Sve_UnzipEven:
+            case NI_Sve_UnzipOdd:
+            case NI_Sve_ZipHigh:
+            case NI_Sve_ZipLow:
+                // Use non-predicated version explicitly
+                GetEmitter()->emitIns_R_R_R(ins, emitSize, targetReg, op1Reg, op2Reg, opt,
+                                            INS_SCALABLE_OPTS_UNPREDICATED);
+                break;
+
             default:
                 unreached();
         }

diff --git a/src/coreclr/jit/hwintrinsiclistarm64sve.h b/src/coreclr/jit/hwintrinsiclistarm64sve.h
@@ -64,6 +64,11 @@ HARDWARE_INTRINSIC(Sve,           LoadVectorUInt16ZeroExtendToUInt64,
 HARDWARE_INTRINSIC(Sve,           LoadVectorUInt32ZeroExtendToInt64,                                -1,      2,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1w,       INS_invalid,        INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
 HARDWARE_INTRINSIC(Sve,           LoadVectorUInt32ZeroExtendToUInt64,                               -1,      2,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1w,       INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
 
+HARDWARE_INTRINSIC(Sve,           UnzipEven,                                                        -1,      2,      true,  {INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1,       INS_sve_uzp1},    HW_Category_SIMD,                  HW_Flag_Scalable|HW_Flag_SpecialCodeGen)
+HARDWARE_INTRINSIC(Sve,           UnzipOdd,                                                         -1,      2,      true,  {INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2,       INS_sve_uzp2},    HW_Category_SIMD,                  HW_Flag_Scalable|HW_Flag_SpecialCodeGen)
+HARDWARE_INTRINSIC(Sve,           ZipHigh,                                                          -1,      2,      true,  {INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2,       INS_sve_zip2},    HW_Category_SIMD,                  HW_Flag_Scalable|HW_Flag_SpecialCodeGen)
+HARDWARE_INTRINSIC(Sve,           ZipLow,                                                           -1,      2,      true,  {INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1,       INS_sve_zip1},    HW_Category_SIMD,                  HW_Flag_Scalable|HW_Flag_SpecialCodeGen)
+
 
 // ***************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************************
 //                 ISA            Function name                                              SIMD size  NumArg  EncodesExtraTypeArg                                                                                            Instructions                                                                                        Category                           Flags

diff --git a/...ries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Arm/Sve.PlatformNotSupported.cs b/...ries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Arm/Sve.PlatformNotSupported.cs
@@ -806,5 +806,260 @@ internal Arm64() { }
         public static unsafe Vector<ulong> LoadVectorUInt32ZeroExtendToUInt64(Vector<ulong> mask, uint* address) { throw new PlatformNotSupportedException(); }
 
 
+        ///  UnzipEven : Concatenate even elements from two inputs
+
+        /// <summary>
+        /// svint8_t svuzp1[_s8](svint8_t op1, svint8_t op2)
+        /// </summary>
+        public static unsafe Vector<sbyte> UnzipEven(Vector<sbyte> left, Vector<sbyte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint16_t svuzp1[_s16](svint16_t op1, svint16_t op2)
+        /// </summary>
+        public static unsafe Vector<short> UnzipEven(Vector<short> left, Vector<short> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint32_t svuzp1[_s32](svint32_t op1, svint32_t op2)
+        /// </summary>
+        public static unsafe Vector<int> UnzipEven(Vector<int> left, Vector<int> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint64_t svuzp1[_s64](svint64_t op1, svint64_t op2)
+        /// </summary>
+        public static unsafe Vector<long> UnzipEven(Vector<long> left, Vector<long> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint8_t svuzp1[_u8](svuint8_t op1, svuint8_t op2)
+        /// svbool_t svuzp1_b8(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<byte> UnzipEven(Vector<byte> left, Vector<byte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint16_t svuzp1[_u16](svuint16_t op1, svuint16_t op2)
+        /// svbool_t svuzp1_b16(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<ushort> UnzipEven(Vector<ushort> left, Vector<ushort> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint32_t svuzp1[_u32](svuint32_t op1, svuint32_t op2)
+        /// svbool_t svuzp1_b32(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<uint> UnzipEven(Vector<uint> left, Vector<uint> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint64_t svuzp1[_u64](svuint64_t op1, svuint64_t op2)
+        /// svbool_t svuzp1_b64(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<ulong> UnzipEven(Vector<ulong> left, Vector<ulong> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat32_t svuzp1[_f32](svfloat32_t op1, svfloat32_t op2)
+        /// </summary>
+        public static unsafe Vector<float> UnzipEven(Vector<float> left, Vector<float> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat64_t svuzp1[_f64](svfloat64_t op1, svfloat64_t op2)
+        /// </summary>
+        public static unsafe Vector<double> UnzipEven(Vector<double> left, Vector<double> right) { throw new PlatformNotSupportedException(); }
+
+
+        ///  UnzipOdd : Concatenate odd elements from two inputs
+
+        /// <summary>
+        /// svuint8_t svuzp2[_u8](svuint8_t op1, svuint8_t op2)
+        /// svbool_t svuzp2_b8(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<byte> UnzipOdd(Vector<byte> left, Vector<byte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat64_t svuzp2[_f64](svfloat64_t op1, svfloat64_t op2)
+        /// </summary>
+        public static unsafe Vector<double> UnzipOdd(Vector<double> left, Vector<double> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint16_t svuzp2[_s16](svint16_t op1, svint16_t op2)
+        ///   UZP2 Zresult.H, Zop1.H, Zop2.H
+        /// </summary>
+        public static unsafe Vector<short> UnzipOdd(Vector<short> left, Vector<short> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint32_t svuzp2[_s32](svint32_t op1, svint32_t op2)
+        /// </summary>
+        public static unsafe Vector<int> UnzipOdd(Vector<int> left, Vector<int> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint64_t svuzp2[_s64](svint64_t op1, svint64_t op2)
+        /// </summary>
+        public static unsafe Vector<long> UnzipOdd(Vector<long> left, Vector<long> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint8_t svuzp2[_s8](svint8_t op1, svint8_t op2)
+        /// </summary>
+        public static unsafe Vector<sbyte> UnzipOdd(Vector<sbyte> left, Vector<sbyte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat32_t svuzp2[_f32](svfloat32_t op1, svfloat32_t op2)
+        /// </summary>
+        public static unsafe Vector<float> UnzipOdd(Vector<float> left, Vector<float> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint16_t svuzp2[_u16](svuint16_t op1, svuint16_t op2)
+        /// svbool_t svuzp2_b16(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<ushort> UnzipOdd(Vector<ushort> left, Vector<ushort> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint32_t svuzp2[_u32](svuint32_t op1, svuint32_t op2)
+        /// svbool_t svuzp2_b32(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<uint> UnzipOdd(Vector<uint> left, Vector<uint> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint64_t svuzp2[_u64](svuint64_t op1, svuint64_t op2)
+        /// svbool_t svuzp2_b64(svbool_t op1, svbool_t op2)
+        /// </summary>
+        public static unsafe Vector<ulong> UnzipOdd(Vector<ulong> left, Vector<ulong> right) { throw new PlatformNotSupportedException(); }
+
+
+        ///  ZipHigh : Interleave elements from high halves of two inputs
+
+        /// <summary>
+        /// svuint8_t svzip2[_u8](svuint8_t op1, svuint8_t op2)
+        ///   ZIP2 Zresult.B, Zop1.B, Zop2.B
+        /// svbool_t svzip2_b8(svbool_t op1, svbool_t op2)
+        ///   ZIP2 Presult.B, Pop1.B, Pop2.B
+        /// </summary>
+        public static unsafe Vector<byte> ZipHigh(Vector<byte> left, Vector<byte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat64_t svzip2[_f64](svfloat64_t op1, svfloat64_t op2)
+        ///   ZIP2 Zresult.D, Zop1.D, Zop2.D
+        /// </summary>
+        public static unsafe Vector<double> ZipHigh(Vector<double> left, Vector<double> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint16_t svzip2[_s16](svint16_t op1, svint16_t op2)
+        ///   ZIP2 Zresult.H, Zop1.H, Zop2.H
+        /// </summary>
+        public static unsafe Vector<short> ZipHigh(Vector<short> left, Vector<short> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint32_t svzip2[_s32](svint32_t op1, svint32_t op2)
+        ///   ZIP2 Zresult.S, Zop1.S, Zop2.S
+        /// </summary>
+        public static unsafe Vector<int> ZipHigh(Vector<int> left, Vector<int> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint64_t svzip2[_s64](svint64_t op1, svint64_t op2)
+        ///   ZIP2 Zresult.D, Zop1.D, Zop2.D
+        /// </summary>
+        public static unsafe Vector<long> ZipHigh(Vector<long> left, Vector<long> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint8_t svzip2[_s8](svint8_t op1, svint8_t op2)
+        ///   ZIP2 Zresult.B, Zop1.B, Zop2.B
+        /// </summary>
+        public static unsafe Vector<sbyte> ZipHigh(Vector<sbyte> left, Vector<sbyte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat32_t svzip2[_f32](svfloat32_t op1, svfloat32_t op2)
+        ///   ZIP2 Zresult.S, Zop1.S, Zop2.S
+        /// </summary>
+        public static unsafe Vector<float> ZipHigh(Vector<float> left, Vector<float> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint16_t svzip2[_u16](svuint16_t op1, svuint16_t op2)
+        ///   ZIP2 Zresult.H, Zop1.H, Zop2.H
+        /// svbool_t svzip2_b16(svbool_t op1, svbool_t op2)
+        ///   ZIP2 Presult.H, Pop1.H, Pop2.H
+        /// </summary>
+        public static unsafe Vector<ushort> ZipHigh(Vector<ushort> left, Vector<ushort> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint32_t svzip2[_u32](svuint32_t op1, svuint32_t op2)
+        ///   ZIP2 Zresult.S, Zop1.S, Zop2.S
+        /// svbool_t svzip2_b32(svbool_t op1, svbool_t op2)
+        ///   ZIP2 Presult.S, Pop1.S, Pop2.S
+        /// </summary>
+        public static unsafe Vector<uint> ZipHigh(Vector<uint> left, Vector<uint> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint64_t svzip2[_u64](svuint64_t op1, svuint64_t op2)
+        ///   ZIP2 Zresult.D, Zop1.D, Zop2.D
+        /// svbool_t svzip2_b64(svbool_t op1, svbool_t op2)
+        ///   ZIP2 Presult.D, Pop1.D, Pop2.D
+        /// </summary>
+        public static unsafe Vector<ulong> ZipHigh(Vector<ulong> left, Vector<ulong> right) { throw new PlatformNotSupportedException(); }
+
+
+        ///  ZipLow : Interleave elements from low halves of two inputs
+
+        /// <summary>
+        /// svuint8_t svzip1[_u8](svuint8_t op1, svuint8_t op2)
+        ///   ZIP1 Zresult.B, Zop1.B, Zop2.B
+        /// svbool_t svzip1_b8(svbool_t op1, svbool_t op2)
+        ///   ZIP1 Presult.B, Pop1.B, Pop2.B
+        /// </summary>
+        public static unsafe Vector<byte> ZipLow(Vector<byte> left, Vector<byte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat64_t svzip1[_f64](svfloat64_t op1, svfloat64_t op2)
+        ///   ZIP1 Zresult.D, Zop1.D, Zop2.D
+        /// </summary>
+        public static unsafe Vector<double> ZipLow(Vector<double> left, Vector<double> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint16_t svzip1[_s16](svint16_t op1, svint16_t op2)
+        ///   ZIP1 Zresult.H, Zop1.H, Zop2.H
+        /// </summary>
+        public static unsafe Vector<short> ZipLow(Vector<short> left, Vector<short> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint32_t svzip1[_s32](svint32_t op1, svint32_t op2)
+        ///   ZIP1 Zresult.S, Zop1.S, Zop2.S
+        /// </summary>
+        public static unsafe Vector<int> ZipLow(Vector<int> left, Vector<int> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint64_t svzip1[_s64](svint64_t op1, svint64_t op2)
+        ///   ZIP1 Zresult.D, Zop1.D, Zop2.D
+        /// </summary>
+        public static unsafe Vector<long> ZipLow(Vector<long> left, Vector<long> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svint8_t svzip1[_s8](svint8_t op1, svint8_t op2)
+        ///   ZIP1 Zresult.B, Zop1.B, Zop2.B
+        /// </summary>
+        public static unsafe Vector<sbyte> ZipLow(Vector<sbyte> left, Vector<sbyte> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svfloat32_t svzip1[_f32](svfloat32_t op1, svfloat32_t op2)
+        ///   ZIP1 Zresult.S, Zop1.S, Zop2.S
+        /// </summary>
+        public static unsafe Vector<float> ZipLow(Vector<float> left, Vector<float> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint16_t svzip1[_u16](svuint16_t op1, svuint16_t op2)
+        ///   ZIP1 Zresult.H, Zop1.H, Zop2.H
+        /// svbool_t svzip1_b16(svbool_t op1, svbool_t op2)
+        ///   ZIP1 Presult.H, Pop1.H, Pop2.H
+        /// </summary>
+        public static unsafe Vector<ushort> ZipLow(Vector<ushort> left, Vector<ushort> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint32_t svzip1[_u32](svuint32_t op1, svuint32_t op2)
+        ///   ZIP1 Zresult.S, Zop1.S, Zop2.S
+        /// svbool_t svzip1_b32(svbool_t op1, svbool_t op2)
+        ///   ZIP1 Presult.S, Pop1.S, Pop2.S
+        /// </summary>
+        public static unsafe Vector<uint> ZipLow(Vector<uint> left, Vector<uint> right) { throw new PlatformNotSupportedException(); }
+
+        /// <summary>
+        /// svuint64_t svzip1[_u64](svuint64_t op1, svuint64_t op2)
+        ///   ZIP1 Zresult.D, Zop1.D, Zop2.D
+        /// svbool_t svzip1_b64(svbool_t op1, svbool_t op2)
+        ///   ZIP1 Presult.D, Pop1.D, Pop2.D
+        /// </summary>
+        public static unsafe Vector<ulong> ZipLow(Vector<ulong> left, Vector<ulong> right) { throw new PlatformNotSupportedException(); }
     }
 }