+ def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
+ (ins RC:$src1, i32i8imm:$src2),
+ "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
+ TA, OpSize, VEX;
+ let neverHasSideEffects = 1, mayLoad = 1 in
+ def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
+ (ins RC:$src1, i32i8imm:$src2),
+ "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
+ TA, OpSize, VEX;
+}
+}
+
+defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
+defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
+defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
+defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
+
+//===----------------------------------------------------------------------===//
+// AVX2 Instructions
+//===----------------------------------------------------------------------===//
+
+/// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
+multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
+ Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
+ X86MemOperand x86memop> {
+ let isCommutable = 1 in
+ def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2, u32u8imm:$src3),
+ !strconcat(OpcodeStr,
+ "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
+ [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
+ VEX_4V;
+ def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
+ (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
+ !strconcat(OpcodeStr,
+ "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
+ [(set RC:$dst,
+ (IntId RC:$src1,
+ (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
+ VEX_4V;
+}
+
+let isCommutable = 0 in {
+defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
+ VR128, memopv2i64, i128mem>;
+defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
+ VR256, memopv4i64, i256mem>;
+}
+
+//===----------------------------------------------------------------------===//
+// VPBROADCAST - Load from memory and broadcast to all elements of the
+// destination operand
+//
+multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
+ X86MemOperand x86memop, PatFrag ld_frag,
+ Intrinsic Int128, Intrinsic Int256> {
+ def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
+ def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst,
+ (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
+ def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
+ def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst,
+ (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
+}
+
+defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
+ int_x86_avx2_pbroadcastb_128,
+ int_x86_avx2_pbroadcastb_256>;
+defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
+ int_x86_avx2_pbroadcastw_128,
+ int_x86_avx2_pbroadcastw_256>;
+defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
+ int_x86_avx2_pbroadcastd_128,
+ int_x86_avx2_pbroadcastd_256>;
+defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
+ int_x86_avx2_pbroadcastq_128,
+ int_x86_avx2_pbroadcastq_256>;
+
+let Predicates = [HasAVX2] in {
+ def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
+ (VPBROADCASTBrm addr:$src)>;
+ def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
+ (VPBROADCASTBYrm addr:$src)>;
+ def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
+ (VPBROADCASTWrm addr:$src)>;
+ def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
+ (VPBROADCASTWYrm addr:$src)>;
+ def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VPBROADCASTDrm addr:$src)>;
+ def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VPBROADCASTDYrm addr:$src)>;
+ def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VPBROADCASTQrm addr:$src)>;
+ def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VPBROADCASTQYrm addr:$src)>;
+}
+
+// AVX1 broadcast patterns
+let Predicates = [HasAVX] in {
+def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSSYrm addr:$src)>;
+def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VBROADCASTSDrm addr:$src)>;
+def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSSYrm addr:$src)>;
+def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
+ (VBROADCASTSDrm addr:$src)>;
+
+def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
+ (VBROADCASTSSrm addr:$src)>;
+def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
+ (VBROADCASTSSrm addr:$src)>;
+}
+
+//===----------------------------------------------------------------------===//
+// VPERM - Permute instructions
+//
+
+multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
+ Intrinsic Int> {
+ def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
+ def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
+ (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1,
+ (bitconvert (mem_frag addr:$src2))))]>,
+ VEX_4V;
+}
+
+defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
+let ExeDomain = SSEPackedSingle in
+defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
+
+multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
+ Intrinsic Int> {
+ def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
+ def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
+ (ins i256mem:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
+ VEX;
+}
+
+defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
+ VEX_W;
+let ExeDomain = SSEPackedDouble in
+defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
+ VEX_W;
+
+//===----------------------------------------------------------------------===//
+// VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
+//
+let AddedComplexity = 1 in {
+def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+ "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
+ (i8 imm:$src3))))]>, VEX_4V;
+def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
+ (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
+ "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
+ (i8 imm:$src3)))]>, VEX_4V;
+}
+
+let Predicates = [HasAVX2], AddedComplexity = 1 in {
+def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+ (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
+
+def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
+ (i8 imm:$imm))),
+ (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
+def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
+ (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
+def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
+ (i8 imm:$imm))),
+ (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
+}
+
+
+//===----------------------------------------------------------------------===//
+// VINSERTI128 - Insert packed integer values
+//
+def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, VR128:$src2, i8imm:$src3),
+ "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst,
+ (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
+ VEX_4V;
+def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
+ (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
+ "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set VR256:$dst,
+ (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
+ imm:$src3))]>, VEX_4V;
+
+let Predicates = [HasAVX2] in {
+def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
+ (i32 imm)),
+ (VINSERTI128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
+ (i32 imm)),
+ (VINSERTI128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
+ (i32 imm)),
+ (VINSERTI128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
+ (i32 imm)),
+ (VINSERTI128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+}
+
+// AVX1 patterns
+let Predicates = [HasAVX] in {
+def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
+ (i32 imm)),
+ (VINSERTF128rr VR256:$src1, VR128:$src2,
+ (INSERT_get_vinsertf128_imm VR256:$ins))>;
+}
+
+//===----------------------------------------------------------------------===//
+// VEXTRACTI128 - Extract packed integer values
+//
+def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
+ (ins VR256:$src1, i8imm:$src2),
+ "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set VR128:$dst,
+ (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
+ VEX;
+let neverHasSideEffects = 1, mayStore = 1 in
+def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
+ (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
+ "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
+
+let Predicates = [HasAVX2] in {
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v2i64 (VEXTRACTI128rr
+ (v4i64 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v4i32 (VEXTRACTI128rr
+ (v8i32 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v8i16 (VEXTRACTI128rr
+ (v16i16 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v16i8 (VEXTRACTI128rr
+ (v32i8 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+}
+
+// AVX1 patterns
+let Predicates = [HasAVX] in {
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v4f32 (VEXTRACTF128rr
+ (v8f32 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v2f64 (VEXTRACTF128rr
+ (v4f64 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v2i64 (VEXTRACTF128rr
+ (v4i64 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v4i32 (VEXTRACTF128rr
+ (v8i32 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v8i16 (VEXTRACTF128rr
+ (v16i16 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
+ (v16i8 (VEXTRACTF128rr
+ (v32i8 VR256:$src1),
+ (EXTRACT_get_vextractf128_imm VR128:$ext)))>;