Added floor and ceiling intrinsics for System.Runtime.Intrinsics.Vector128 and System.Numerics.Vector (#83592)

Added floor and ceil intrinsics

Signed-off-by: Vlad - Alexandru Ionescu <[email protected]>

Author: LeVladIonescu

src/mono/mono/arch/arm64/arm64-codegen.h

@@ -1565,6 +1565,9 @@ arm_encode_arith_imm (int imm, guint32 *shift)
 #define arm_neon_fsqrt_4s(p, rd, rn) arm_neon_2mvec_opcode ((p), VREG_FULL, 0b1, 0b10 | SIZE_1, 0b11101, (rd), (rn))
 #define arm_neon_fsqrt_2d(p, rd, rn) arm_neon_2mvec_opcode ((p), VREG_FULL, 0b1, 0b10 | SIZE_2, 0b11101, (rd), (rn))
 
+#define arm_neon_frintm(p, width, type, rd, rn) arm_neon_2mvec_opcode ((p), (width), 0b0, (type), 0b11001, (rd), (rn))
+#define arm_neon_frintp(p, width, type, rd, rn) arm_neon_2mvec_opcode ((p), (width), 0b0, 0b10 | (type), 0b11000, (rd), (rn))
+
 /* NEON :: across lanes */
 #define arm_neon_xln_opcode(p, q, u, size, opcode, rd, rn) arm_neon_opcode_2reg ((p), (q), 0b00001110001100000000100000000000 | (u) << 29 | (size) << 22 | (opcode) << 12, (rd), (rn))
 #define arm_neon_addv(p, width, type, rd, rn) arm_neon_xln_opcode ((p), (width), 0b0, (type), 0b11011, (rd), (rn))

src/mono/mono/mini/cpu-arm64.mdesc

@@ -513,6 +513,7 @@ extract_i8: dest:i src1:x len:4
 extract_r4: dest:f src1:x len:4
 extract_r8: dest:f src1:x len:4
 arm64_xaddv: dest:x src1:x len:8
+xop_ovr_x_x: dest:x src1:x len:4
 
 generic_class_init: src1:a len:44 clob:c
 gc_safe_point: src1:i len:12 clob:c

src/mono/mono/mini/mini-arm64.c

@@ -38,6 +38,7 @@
 #define OPFMT_WDSS _w, dreg, sreg1, sreg2
 #define OPFMT_WTDS _w, _t, dreg, sreg1
 #define OPFMT_WTDSS _w, _t, dreg, sreg1, sreg2
+#define OPFMT_WTDS _w, _t, dreg, sreg1
 #define OPFMT_WTDSS_REV _w, _t, dreg, sreg2, sreg1
 #define _UNDEF(...) g_assert_not_reached ()
 #define _SKIP(...) goto manual_instruction_emit

src/mono/mono/mini/simd-arm64.h

@@ -64,4 +64,6 @@ SIMD_OP  (128, OP_XBINOP_FORCEINT,    XBINOP_FORCEINT_OR,     WDSS,      arm_neo
 SIMD_OP  (128, OP_XBINOP_FORCEINT,    XBINOP_FORCEINT_XOR,    WDSS,      arm_neon_eor,     arm_neon_eor,     arm_neon_eor,    arm_neon_eor,      arm_neon_eor,     arm_neon_eor)
 SIMD_OP  (128, OP_ARM64_XADDV, INTRINS_AARCH64_ADV_SIMD_UADDV, WTDS,     arm_neon_addv,    arm_neon_addv,    arm_neon_addv,   _SKIP,             _UNDEF,           _UNDEF)
 SIMD_OP  (128, OP_ARM64_XADDV, INTRINS_AARCH64_ADV_SIMD_SADDV, WTDS,     arm_neon_addv,    arm_neon_addv,    arm_neon_addv,   _SKIP,             _UNDEF,           _UNDEF)
-SIMD_OP  (128, OP_ARM64_XADDV, INTRINS_AARCH64_ADV_SIMD_FADDV, WTDS,     _UNDEF,           _UNDEF,           _UNDEF,          _UNDEF,            _SKIP,            _SKIP)
+SIMD_OP  (128, OP_ARM64_XADDV, INTRINS_AARCH64_ADV_SIMD_FADDV, WTDS,     _UNDEF,           _UNDEF,           _UNDEF,          _UNDEF,            _SKIP,            _SKIP)
+SIMD_OP  (128, OP_XOP_OVR_X_X,    INTRINS_AARCH64_ADV_SIMD_FRINTP,    WTDS,    _UNDEF,    _UNDEF,            _UNDEF,          _UNDEF,            arm_neon_frintp,  arm_neon_frintp)
+SIMD_OP  (128, OP_XOP_OVR_X_X,    INTRINS_AARCH64_ADV_SIMD_FRINTM,    WTDS,    _UNDEF,    _UNDEF,            _UNDEF,          _UNDEF,            arm_neon_frintm,  arm_neon_frintm)

src/mono/mono/mini/simd-intrinsics.c

@@ -1260,6 +1260,8 @@ emit_sri_vector (MonoCompile *cfg, MonoMethod *cmethod, MonoMethodSignature *fsi
 		case SN_Min:
 		case SN_Sum:
 		case SN_ToScalar:
+		case SN_Floor:
+		case SN_Ceiling:
 			break;
 		default: 
 			return NULL;