aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib')
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp6
-rw-r--r--llvm/lib/Target/X86/X86InstrAVX512.td193
-rw-r--r--llvm/lib/Target/X86/X86InstrSSE.td18
3 files changed, 173 insertions, 44 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 60b2599..fae877a 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1607,6 +1607,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::CONCAT_VECTORS, MVT::v64i1, Legal);
setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v32i1, Custom);
setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v64i1, Custom);
+ setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v32i16, Custom);
+ setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v64i8, Custom);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v32i16, Custom);
+ setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v64i8, Custom);
setOperationAction(ISD::SELECT, MVT::v32i1, Custom);
setOperationAction(ISD::SELECT, MVT::v64i1, Custom);
setOperationAction(ISD::SIGN_EXTEND, MVT::v32i8, Custom);
@@ -1618,6 +1622,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::ZERO_EXTEND, MVT::v64i8, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v32i1, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v64i1, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v32i16, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v64i8, Custom);
setOperationAction(ISD::VSELECT, MVT::v32i16, Legal);
setOperationAction(ISD::VSELECT, MVT::v64i8, Legal);
setOperationAction(ISD::TRUNCATE, MVT::v32i1, Custom);
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 85a53e1..8701522 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -677,53 +677,36 @@ multiclass vextract_for_size<int Opcode,
From.RC:$src1, imm:$idx)>;
}
-// This multiclass generates patterns for matching vextract with common types
-// (X86VectorVTInfo From , X86VectorVTInfo To) and alternative types
-// (X86VectorVTInfo AltFrom, X86VectorVTInfo AltTo)
-multiclass vextract_for_size_all<int Opcode,
- X86VectorVTInfo From, X86VectorVTInfo To,
- X86VectorVTInfo AltFrom, X86VectorVTInfo AltTo,
- PatFrag vextract_extract,
- SDNodeXForm EXTRACT_get_vextract_imm> :
- vextract_for_size<Opcode, From, To, vextract_extract>,
- vextract_for_size_first_position_lowering<AltFrom, AltTo> {
-
- // Codegen pattern with the alternative types.
- // Only add this if operation not supported natively via AVX512DQ
- let Predicates = [NoDQI] in
- def : Pat<(vextract_extract:$ext (AltFrom.VT AltFrom.RC:$src1), (iPTR imm)),
- (AltTo.VT (!cast<Instruction>(NAME # To.EltSize # "x" #
- To.NumElts # From.ZSuffix # "rr")
- AltFrom.RC:$src1,
- (EXTRACT_get_vextract_imm To.RC:$ext)))>;
+// Codegen pattern for the alternative types
+multiclass vextract_for_size_lowering<string InstrStr, X86VectorVTInfo From,
+ X86VectorVTInfo To, PatFrag vextract_extract,
+ SDNodeXForm EXTRACT_get_vextract_imm, list<Predicate> p> :
+ vextract_for_size_first_position_lowering<From, To> {
+
+ let Predicates = p in
+ def : Pat<(vextract_extract:$ext (From.VT From.RC:$src1), (iPTR imm)),
+ (To.VT (!cast<Instruction>(InstrStr#"rr")
+ From.RC:$src1,
+ (EXTRACT_get_vextract_imm To.RC:$ext)))>;
}
multiclass vextract_for_type<ValueType EltVT32, int Opcode128,
- ValueType EltVT64, int Opcode256> {
- defm NAME # "32x4Z" : vextract_for_size_all<Opcode128,
+ ValueType EltVT64, int Opcode256> {
+ defm NAME # "32x4Z" : vextract_for_size<Opcode128,
X86VectorVTInfo<16, EltVT32, VR512>,
X86VectorVTInfo< 4, EltVT32, VR128X>,
- X86VectorVTInfo< 8, EltVT64, VR512>,
- X86VectorVTInfo< 2, EltVT64, VR128X>,
- vextract128_extract,
- EXTRACT_get_vextract128_imm>,
+ vextract128_extract>,
EVEX_V512, EVEX_CD8<32, CD8VT4>;
- defm NAME # "64x4Z" : vextract_for_size_all<Opcode256,
+ defm NAME # "64x4Z" : vextract_for_size<Opcode256,
X86VectorVTInfo< 8, EltVT64, VR512>,
X86VectorVTInfo< 4, EltVT64, VR256X>,
- X86VectorVTInfo<16, EltVT32, VR512>,
- X86VectorVTInfo< 8, EltVT32, VR256>,
- vextract256_extract,
- EXTRACT_get_vextract256_imm>,
+ vextract256_extract>,
VEX_W, EVEX_V512, EVEX_CD8<64, CD8VT4>;
let Predicates = [HasVLX] in
- defm NAME # "32x4Z256" : vextract_for_size_all<Opcode128,
+ defm NAME # "32x4Z256" : vextract_for_size<Opcode128,
X86VectorVTInfo< 8, EltVT32, VR256X>,
X86VectorVTInfo< 4, EltVT32, VR128X>,
- X86VectorVTInfo< 4, EltVT64, VR256X>,
- X86VectorVTInfo< 2, EltVT64, VR128X>,
- vextract128_extract,
- EXTRACT_get_vextract128_imm>,
+ vextract128_extract>,
EVEX_V256, EVEX_CD8<32, CD8VT4>;
let Predicates = [HasVLX, HasDQI] in
defm NAME # "64x2Z256" : vextract_for_size<Opcode128,
@@ -748,6 +731,34 @@ multiclass vextract_for_type<ValueType EltVT32, int Opcode128,
defm VEXTRACTF : vextract_for_type<f32, 0x19, f64, 0x1b>;
defm VEXTRACTI : vextract_for_type<i32, 0x39, i64, 0x3b>;
+// extract_subvector codegen patterns with the alternative types.
+// Only add this if 64x2 and its friends are not supported natively via AVX512DQ.
+defm : vextract_for_size_lowering<"VEXTRACTF32x4Z", v8f64_info, v2f64x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasAVX512, NoDQI]>;
+defm : vextract_for_size_lowering<"VEXTRACTI32x4Z", v8i64_info, v2i64x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasAVX512, NoDQI]>;
+
+defm : vextract_for_size_lowering<"VEXTRACTF64x4Z", v16f32_info, v8f32x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasAVX512, NoDQI]>;
+defm : vextract_for_size_lowering<"VEXTRACTI64x4Z", v16i32_info, v8i32x_info,
+ vextract256_extract, EXTRACT_get_vextract256_imm, [HasAVX512, NoDQI]>;
+
+defm : vextract_for_size_lowering<"VEXTRACTF32x4Z256", v4f64x_info, v2f64x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasVLX, NoDQI]>;
+defm : vextract_for_size_lowering<"VEXTRACTI32x4Z256", v4i64x_info, v2i64x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasVLX, NoDQI]>;
+
+// Codegen pattern with the alternative types extract VEC128 from VEC512
+defm : vextract_for_size_lowering<"VEXTRACTI32x4Z", v32i16_info, v8i16x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasAVX512]>;
+defm : vextract_for_size_lowering<"VEXTRACTI32x4Z", v64i8_info, v16i8x_info,
+ vextract128_extract, EXTRACT_get_vextract128_imm, [HasAVX512]>;
+// Codegen pattern with the alternative types extract VEC256 from VEC512
+defm : vextract_for_size_lowering<"VEXTRACTI64x4Z", v32i16_info, v16i16x_info,
+ vextract256_extract, EXTRACT_get_vextract256_imm, [HasAVX512]>;
+defm : vextract_for_size_lowering<"VEXTRACTI64x4Z", v64i8_info, v32i8x_info,
+ vextract256_extract, EXTRACT_get_vextract256_imm, [HasAVX512]>;
+
// A 128-bit subvector insert to the first 512-bit vector position
// is a subregister copy that needs no instruction.
def : Pat<(insert_subvector undef, (v2i64 VR128X:$src), (iPTR 0)),
@@ -6940,6 +6951,118 @@ defm VPUNPCKLQDQ : avx512_binop_rm_vl_q<0x6C, "vpunpcklqdq", X86Unpckl,
SSE_INTALU_ITINS_P, HasAVX512>;
defm VPUNPCKHQDQ : avx512_binop_rm_vl_q<0x6D, "vpunpckhqdq", X86Unpckh,
SSE_INTALU_ITINS_P, HasAVX512>;
+
+//===----------------------------------------------------------------------===//
+// AVX-512 - Extract & Insert Integer Instructions
+//===----------------------------------------------------------------------===//
+
+multiclass avx512_extract_elt_bw_m<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ X86VectorVTInfo _> {
+ let mayStore = 1 in
+ def mr : AVX512Ii8<opc, MRMDestMem, (outs),
+ (ins _.ScalarMemOp:$dst, _.RC:$src1, u8imm:$src2),
+ OpcodeStr#"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(store (_.EltVT (trunc (assertzext (OpNode (_.VT _.RC:$src1),
+ imm:$src2)))),
+ addr:$dst)]>,
+ EVEX, EVEX_CD8<_.EltSize, CD8VT1>;
+}
+
+multiclass avx512_extract_elt_b<string OpcodeStr, X86VectorVTInfo _> {
+ let Predicates = [HasBWI] in {
+ def rr : AVX512Ii8<0x14, MRMDestReg, (outs GR32orGR64:$dst),
+ (ins _.RC:$src1, u8imm:$src2),
+ OpcodeStr#"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set GR32orGR64:$dst,
+ (X86pextrb (_.VT _.RC:$src1), imm:$src2))]>,
+ EVEX, TAPD;
+
+ defm NAME : avx512_extract_elt_bw_m<0x14, OpcodeStr, X86pextrb, _>, TAPD;
+ }
+}
+
+multiclass avx512_extract_elt_w<string OpcodeStr, X86VectorVTInfo _> {
+ let Predicates = [HasBWI] in {
+ def rr : AVX512Ii8<0xC5, MRMSrcReg, (outs GR32orGR64:$dst),
+ (ins _.RC:$src1, u8imm:$src2),
+ OpcodeStr#"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set GR32orGR64:$dst,
+ (X86pextrw (_.VT _.RC:$src1), imm:$src2))]>,
+ EVEX, PD;
+
+ defm NAME : avx512_extract_elt_bw_m<0x15, OpcodeStr, X86pextrw, _>, TAPD;
+ }
+}
+
+multiclass avx512_extract_elt_dq<string OpcodeStr, X86VectorVTInfo _,
+ RegisterClass GRC> {
+ let Predicates = [HasDQI] in {
+ def rr : AVX512Ii8<0x16, MRMDestReg, (outs GRC:$dst),
+ (ins _.RC:$src1, u8imm:$src2),
+ OpcodeStr#"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set GRC:$dst,
+ (extractelt (_.VT _.RC:$src1), imm:$src2))]>,
+ EVEX, TAPD;
+
+ let mayStore = 1 in
+ def mr : AVX512Ii8<0x16, MRMDestMem, (outs),
+ (ins _.ScalarMemOp:$dst, _.RC:$src1, u8imm:$src2),
+ OpcodeStr#"\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(store (extractelt (_.VT _.RC:$src1),
+ imm:$src2),addr:$dst)]>,
+ EVEX, EVEX_CD8<_.EltSize, CD8VT1>, TAPD;
+ }
+}
+
+defm VPEXTRBZ : avx512_extract_elt_b<"vpextrb", v16i8x_info>;
+defm VPEXTRWZ : avx512_extract_elt_w<"vpextrw", v8i16x_info>;
+defm VPEXTRDZ : avx512_extract_elt_dq<"vpextrd", v4i32x_info, GR32>;
+defm VPEXTRQZ : avx512_extract_elt_dq<"vpextrq", v2i64x_info, GR64>, VEX_W;
+
+multiclass avx512_insert_elt_m<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ X86VectorVTInfo _, PatFrag LdFrag> {
+ def rm : AVX512Ii8<opc, MRMSrcMem, (outs _.RC:$dst),
+ (ins _.RC:$src1, _.ScalarMemOp:$src2, u8imm:$src3),
+ OpcodeStr#"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set _.RC:$dst,
+ (_.VT (OpNode _.RC:$src1, (LdFrag addr:$src2), imm:$src3)))]>,
+ EVEX_4V, EVEX_CD8<_.EltSize, CD8VT1>;
+}
+
+multiclass avx512_insert_elt_bw<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ X86VectorVTInfo _, PatFrag LdFrag> {
+ let Predicates = [HasBWI] in {
+ def rr : AVX512Ii8<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.RC:$src1, GR32orGR64:$src2, u8imm:$src3),
+ OpcodeStr#"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set _.RC:$dst,
+ (OpNode _.RC:$src1, GR32orGR64:$src2, imm:$src3))]>, EVEX_4V;
+
+ defm NAME : avx512_insert_elt_m<opc, OpcodeStr, OpNode, _, LdFrag>;
+ }
+}
+
+multiclass avx512_insert_elt_dq<bits<8> opc, string OpcodeStr,
+ X86VectorVTInfo _, RegisterClass GRC> {
+ let Predicates = [HasDQI] in {
+ def rr : AVX512Ii8<opc, MRMSrcReg, (outs _.RC:$dst),
+ (ins _.RC:$src1, GRC:$src2, u8imm:$src3),
+ OpcodeStr#"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ [(set _.RC:$dst,
+ (_.VT (insertelt _.RC:$src1, GRC:$src2, imm:$src3)))]>,
+ EVEX_4V, TAPD;
+
+ defm NAME : avx512_insert_elt_m<opc, OpcodeStr, insertelt, _,
+ _.ScalarLdFrag>, TAPD;
+ }
+}
+
+defm VPINSRBZ : avx512_insert_elt_bw<0x20, "vpinsrb", X86pinsrb, v16i8x_info,
+ extloadi8>, TAPD;
+defm VPINSRWZ : avx512_insert_elt_bw<0xC4, "vpinsrw", X86pinsrw, v8i16x_info,
+ extloadi16>, PD;
+defm VPINSRDZ : avx512_insert_elt_dq<0x22, "vpinsrd", v4i32x_info, GR32>;
+defm VPINSRQZ : avx512_insert_elt_dq<0x22, "vpinsrq", v2i64x_info, GR64>, VEX_W;
//===----------------------------------------------------------------------===//
// VSHUFPS - VSHUFPD Operations
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index f390d8f..dac9e35 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -4611,7 +4611,7 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> {
}
// Extract
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
(outs GR32orGR64:$dst), (ins VR128:$src1, u8imm:$src2),
"vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
@@ -4626,7 +4626,7 @@ def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
Sched<[WriteShuffleLd, ReadAfterLd]>;
// Insert
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
defm VPINSRW : sse2_pinsrw<0>, PD, VEX_4V;
let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
@@ -6156,7 +6156,7 @@ multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
imm:$src2)))), addr:$dst)]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
@@ -6181,7 +6181,7 @@ multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
imm:$src2)))), addr:$dst)]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
@@ -6205,7 +6205,7 @@ multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
addr:$dst)]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
@@ -6228,7 +6228,7 @@ multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
addr:$dst)]>, REX_W;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
@@ -6296,7 +6296,7 @@ multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
imm:$src3))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoBWI] in
defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
@@ -6322,7 +6322,7 @@ multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
@@ -6348,7 +6348,7 @@ multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoDQI] in
defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
let Constraints = "$src1 = $dst" in
defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
generated by cgit v1.1 at 2025-10-17 21:01:30 +0000