// vinsertps - insert f32 to XMM
def VINSERTPSzrr : AVX512AIi8<0x21, MRMSrcReg, (outs VR128X:$dst),
- (ins VR128X:$src1, VR128X:$src2, u32u8imm:$src3),
+ (ins VR128X:$src1, VR128X:$src2, i8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insertps VR128X:$src1, VR128X:$src2, imm:$src3))]>,
EVEX_4V;
def VINSERTPSzrm: AVX512AIi8<0x21, MRMSrcMem, (outs VR128X:$dst),
- (ins VR128X:$src1, f32mem:$src2, u32u8imm:$src3),
+ (ins VR128X:$src1, f32mem:$src2, i8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128X:$dst, (X86insertps VR128X:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract128_extract:$ext VR512:$src1, (iPTR imm)),
- (v4i32 (VEXTRACTF32x4rr VR512:$src1,
+ (v4i32 (VEXTRACTI32x4rr VR512:$src1,
(EXTRACT_get_vextract128_imm VR128X:$ext)))>;
def : Pat<(vextract128_extract:$ext (v8f64 VR512:$src1), (iPTR imm)),
// vextractps - extract 32 bits from XMM
def VEXTRACTPSzrr : AVX512AIi8<0x17, MRMDestReg, (outs GR32:$dst),
- (ins VR128X:$src1, u32u8imm:$src2),
+ (ins VR128X:$src1, i32i8imm:$src2),
"vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set GR32:$dst, (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2))]>,
EVEX;
def VEXTRACTPSzmr : AVX512AIi8<0x17, MRMDestMem, (outs),
- (ins f32mem:$dst, VR128X:$src1, u32u8imm:$src2),
+ (ins f32mem:$dst, VR128X:$src1, i32i8imm:$src2),
"vextractps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(store (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2),
addr:$dst)]>, EVEX, EVEX_CD8<32, CD8VT1>;
}
multiclass avx512_icmp_packed_rmb<bits<8> opc, string OpcodeStr, SDNode OpNode,
- X86VectorVTInfo _> {
+ X86VectorVTInfo _> :
+ avx512_icmp_packed<opc, OpcodeStr, OpNode, _> {
let mayLoad = 1 in {
def rmb : AVX512BI<opc, MRMSrcMem,
(outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2),
avx512vl_i16_info, HasBWI>,
EVEX_CD8<16, CD8VF>;
-defm VPCMPEQD : avx512_icmp_packed_vl<0x76, "vpcmpeqd", X86pcmpeqm,
- avx512vl_i32_info, HasAVX512>,
- avx512_icmp_packed_rmb_vl<0x76, "vpcmpeqd", X86pcmpeqm,
+defm VPCMPEQD : avx512_icmp_packed_rmb_vl<0x76, "vpcmpeqd", X86pcmpeqm,
avx512vl_i32_info, HasAVX512>,
EVEX_CD8<32, CD8VF>;
-defm VPCMPEQQ : avx512_icmp_packed_vl<0x29, "vpcmpeqq", X86pcmpeqm,
- avx512vl_i64_info, HasAVX512>,
- avx512_icmp_packed_rmb_vl<0x29, "vpcmpeqq", X86pcmpeqm,
+defm VPCMPEQQ : avx512_icmp_packed_rmb_vl<0x29, "vpcmpeqq", X86pcmpeqm,
avx512vl_i64_info, HasAVX512>,
T8PD, VEX_W, EVEX_CD8<64, CD8VF>;
avx512vl_i16_info, HasBWI>,
EVEX_CD8<16, CD8VF>;
-defm VPCMPGTD : avx512_icmp_packed_vl<0x66, "vpcmpgtd", X86pcmpgtm,
- avx512vl_i32_info, HasAVX512>,
- avx512_icmp_packed_rmb_vl<0x66, "vpcmpgtd", X86pcmpgtm,
+defm VPCMPGTD : avx512_icmp_packed_rmb_vl<0x66, "vpcmpgtd", X86pcmpgtm,
avx512vl_i32_info, HasAVX512>,
EVEX_CD8<32, CD8VF>;
-defm VPCMPGTQ : avx512_icmp_packed_vl<0x37, "vpcmpgtq", X86pcmpgtm,
- avx512vl_i64_info, HasAVX512>,
- avx512_icmp_packed_rmb_vl<0x37, "vpcmpgtq", X86pcmpgtm,
+defm VPCMPGTQ : avx512_icmp_packed_rmb_vl<0x37, "vpcmpgtq", X86pcmpgtm,
avx512vl_i64_info, HasAVX512>,
T8PD, VEX_W, EVEX_CD8<64, CD8VF>;
def : Pat<(v8i1 (X86pcmpgtm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))),
- (COPY_TO_REGCLASS (VPCMPGTDZrr
+ (COPY_TO_REGCLASS (VPCMPGTDZrr
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>;
def : Pat<(v8i1 (X86pcmpeqm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))),
- (COPY_TO_REGCLASS (VPCMPEQDZrr
+ (COPY_TO_REGCLASS (VPCMPEQDZrr
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)),
(v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>;
-multiclass avx512_icmp_cc<bits<8> opc, RegisterClass WMRC, RegisterClass KRC,
- RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag,
- SDNode OpNode, ValueType vt, Operand CC, string Suffix> {
+multiclass avx512_icmp_cc<bits<8> opc, string Suffix, SDNode OpNode,
+ X86VectorVTInfo _> {
def rri : AVX512AIi8<opc, MRMSrcReg,
- (outs KRC:$dst), (ins RC:$src1, RC:$src2, CC:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, AVXCC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2), imm:$cc))],
+ [(set _.KRC:$dst, (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2),
+ imm:$cc))],
IIC_SSE_ALU_F32P_RR>, EVEX_4V;
+ let mayLoad = 1 in
def rmi : AVX512AIi8<opc, MRMSrcMem,
- (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, AVXCC:$cc),
!strconcat("vpcmp${cc}", Suffix,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set KRC:$dst, (OpNode (vt RC:$src1), (memop_frag addr:$src2),
- imm:$cc))], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
+ [(set _.KRC:$dst, (OpNode (_.VT _.RC:$src1),
+ (_.VT (bitconvert (_.LdFrag addr:$src2))),
+ imm:$cc))],
+ IIC_SSE_ALU_F32P_RM>, EVEX_4V;
+ def rrik : AVX512AIi8<opc, MRMSrcReg,
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2,
+ AVXCC:$cc),
+ !strconcat("vpcmp${cc}", Suffix,
+ "\t{$src2, $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, $src2}"),
+ [(set _.KRC:$dst, (and _.KRCWM:$mask,
+ (OpNode (_.VT _.RC:$src1), (_.VT _.RC:$src2),
+ imm:$cc)))],
+ IIC_SSE_ALU_F32P_RR>, EVEX_4V, EVEX_K;
+ let mayLoad = 1 in
+ def rmik : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2,
+ AVXCC:$cc),
+ !strconcat("vpcmp${cc}", Suffix,
+ "\t{$src2, $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, $src2}"),
+ [(set _.KRC:$dst, (and _.KRCWM:$mask,
+ (OpNode (_.VT _.RC:$src1),
+ (_.VT (bitconvert (_.LdFrag addr:$src2))),
+ imm:$cc)))],
+ IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K;
+
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : AVX512AIi8<opc, MRMSrcReg,
- (outs KRC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
- !strconcat("vpcmp", Suffix,
- "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.RC:$src2, i8imm:$cc),
+ !strconcat("vpcmp", Suffix, "\t{$cc, $src2, $src1, $dst|",
+ "$dst, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RR>, EVEX_4V;
+ def rmi_alt : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.MemOp:$src2, i8imm:$cc),
+ !strconcat("vpcmp", Suffix, "\t{$cc, $src2, $src1, $dst|",
+ "$dst, $src1, $src2, $cc}"),
+ [], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
def rrik_alt : AVX512AIi8<opc, MRMSrcReg,
- (outs KRC:$dst), (ins WMRC:$mask, RC:$src1, RC:$src2, i8imm:$cc),
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2,
+ i8imm:$cc),
!strconcat("vpcmp", Suffix,
- "\t{$cc, $src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2, $cc}"),
+ "\t{$cc, $src2, $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RR>, EVEX_4V, EVEX_K;
- def rmi_alt : AVX512AIi8<opc, MRMSrcMem,
- (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
- !strconcat("vpcmp", Suffix,
- "\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}"),
- [], IIC_SSE_ALU_F32P_RM>, EVEX_4V;
def rmik_alt : AVX512AIi8<opc, MRMSrcMem,
- (outs KRC:$dst), (ins WMRC:$mask, RC:$src1, x86memop:$src2, i8imm:$cc),
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1, _.MemOp:$src2,
+ i8imm:$cc),
!strconcat("vpcmp", Suffix,
- "\t{$cc, $src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2, $cc}"),
+ "\t{$cc, $src2, $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, $src2, $cc}"),
[], IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K;
}
}
-defm VPCMPDZ : avx512_icmp_cc<0x1F, VK16WM, VK16, VR512, i512mem, memopv16i32,
- X86cmpm, v16i32, AVXCC, "d">,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VPCMPUDZ : avx512_icmp_cc<0x1E, VK16WM, VK16, VR512, i512mem, memopv16i32,
- X86cmpmu, v16i32, AVXCC, "ud">,
- EVEX_V512, EVEX_CD8<32, CD8VF>;
+multiclass avx512_icmp_cc_rmb<bits<8> opc, string Suffix, SDNode OpNode,
+ X86VectorVTInfo _> :
+ avx512_icmp_cc<opc, Suffix, OpNode, _> {
+ let mayLoad = 1 in {
+ def rmib : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2,
+ AVXCC:$cc),
+ !strconcat("vpcmp${cc}", Suffix,
+ "\t{${src2}", _.BroadcastStr, ", $src1, $dst|",
+ "$dst, $src1, ${src2}", _.BroadcastStr, "}"),
+ [(set _.KRC:$dst, (OpNode (_.VT _.RC:$src1),
+ (X86VBroadcast (_.ScalarLdFrag addr:$src2)),
+ imm:$cc))],
+ IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_B;
+ def rmibk : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1,
+ _.ScalarMemOp:$src2, AVXCC:$cc),
+ !strconcat("vpcmp${cc}", Suffix,
+ "\t{${src2}", _.BroadcastStr, ", $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, "}"),
+ [(set _.KRC:$dst, (and _.KRCWM:$mask,
+ (OpNode (_.VT _.RC:$src1),
+ (X86VBroadcast (_.ScalarLdFrag addr:$src2)),
+ imm:$cc)))],
+ IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K, EVEX_B;
+ }
+
+ // Accept explicit immediate argument form instead of comparison code.
+ let isAsmParserOnly = 1, hasSideEffects = 0 in {
+ def rmib_alt : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.RC:$src1, _.ScalarMemOp:$src2,
+ i8imm:$cc),
+ !strconcat("vpcmp", Suffix,
+ "\t{$cc, ${src2}", _.BroadcastStr, ", $src1, $dst|",
+ "$dst, $src1, ${src2}", _.BroadcastStr, ", $cc}"),
+ [], IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_B;
+ def rmibk_alt : AVX512AIi8<opc, MRMSrcMem,
+ (outs _.KRC:$dst), (ins _.KRCWM:$mask, _.RC:$src1,
+ _.ScalarMemOp:$src2, i8imm:$cc),
+ !strconcat("vpcmp", Suffix,
+ "\t{$cc, ${src2}", _.BroadcastStr, ", $src1, $dst {${mask}}|",
+ "$dst {${mask}}, $src1, ${src2}", _.BroadcastStr, ", $cc}"),
+ [], IIC_SSE_ALU_F32P_RM>, EVEX_4V, EVEX_K, EVEX_B;
+ }
+}
+
+multiclass avx512_icmp_cc_vl<bits<8> opc, string Suffix, SDNode OpNode,
+ AVX512VLVectorVTInfo VTInfo, Predicate prd> {
+ let Predicates = [prd] in
+ defm Z : avx512_icmp_cc<opc, Suffix, OpNode, VTInfo.info512>, EVEX_V512;
+
+ let Predicates = [prd, HasVLX] in {
+ defm Z256 : avx512_icmp_cc<opc, Suffix, OpNode, VTInfo.info256>, EVEX_V256;
+ defm Z128 : avx512_icmp_cc<opc, Suffix, OpNode, VTInfo.info128>, EVEX_V128;
+ }
+}
+
+multiclass avx512_icmp_cc_rmb_vl<bits<8> opc, string Suffix, SDNode OpNode,
+ AVX512VLVectorVTInfo VTInfo, Predicate prd> {
+ let Predicates = [prd] in
+ defm Z : avx512_icmp_cc_rmb<opc, Suffix, OpNode, VTInfo.info512>,
+ EVEX_V512;
+
+ let Predicates = [prd, HasVLX] in {
+ defm Z256 : avx512_icmp_cc_rmb<opc, Suffix, OpNode, VTInfo.info256>,
+ EVEX_V256;
+ defm Z128 : avx512_icmp_cc_rmb<opc, Suffix, OpNode, VTInfo.info128>,
+ EVEX_V128;
+ }
+}
+
+defm VPCMPB : avx512_icmp_cc_vl<0x3F, "b", X86cmpm, avx512vl_i8_info,
+ HasBWI>, EVEX_CD8<8, CD8VF>;
+defm VPCMPUB : avx512_icmp_cc_vl<0x3E, "ub", X86cmpmu, avx512vl_i8_info,
+ HasBWI>, EVEX_CD8<8, CD8VF>;
+
+defm VPCMPW : avx512_icmp_cc_vl<0x3F, "w", X86cmpm, avx512vl_i16_info,
+ HasBWI>, VEX_W, EVEX_CD8<16, CD8VF>;
+defm VPCMPUW : avx512_icmp_cc_vl<0x3E, "uw", X86cmpmu, avx512vl_i16_info,
+ HasBWI>, VEX_W, EVEX_CD8<16, CD8VF>;
+
+defm VPCMPD : avx512_icmp_cc_rmb_vl<0x1F, "d", X86cmpm, avx512vl_i32_info,
+ HasAVX512>, EVEX_CD8<32, CD8VF>;
+defm VPCMPUD : avx512_icmp_cc_rmb_vl<0x1E, "ud", X86cmpmu, avx512vl_i32_info,
+ HasAVX512>, EVEX_CD8<32, CD8VF>;
-defm VPCMPQZ : avx512_icmp_cc<0x1F, VK8WM, VK8, VR512, i512mem, memopv8i64,
- X86cmpm, v8i64, AVXCC, "q">,
- VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
-defm VPCMPUQZ : avx512_icmp_cc<0x1E, VK8WM, VK8, VR512, i512mem, memopv8i64,
- X86cmpmu, v8i64, AVXCC, "uq">,
- VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
+defm VPCMPQ : avx512_icmp_cc_rmb_vl<0x1F, "q", X86cmpm, avx512vl_i64_info,
+ HasAVX512>, VEX_W, EVEX_CD8<64, CD8VF>;
+defm VPCMPUQ : avx512_icmp_cc_rmb_vl<0x1E, "uq", X86cmpmu, avx512vl_i64_info,
+ HasAVX512>, VEX_W, EVEX_CD8<64, CD8VF>;
// avx512_cmp_packed - compare packed instructions
multiclass avx512_cmp_packed<RegisterClass KRC, RegisterClass RC,
i512mem, v16i32>, PD, EVEX_V512, EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedSingle in
-defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilp,
+defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilpi,
memopv16f32, i512mem, v16f32>, TAPD, EVEX_V512,
EVEX_CD8<32, CD8VF>;
let ExeDomain = SSEPackedDouble in
-defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilp,
+defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilpi,
memopv8f64, i512mem, v8f64>, TAPD, EVEX_V512,
VEX_W, EVEX_CD8<32, CD8VF>;
-def : Pat<(v16i32 (X86VPermilp VR512:$src1, (i8 imm:$imm))),
+def : Pat<(v16i32 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
(VPERMILPSZri VR512:$src1, imm:$imm)>;
-def : Pat<(v8i64 (X86VPermilp VR512:$src1, (i8 imm:$imm))),
+def : Pat<(v8i64 (X86VPermilpi VR512:$src1, (i8 imm:$imm))),
(VPERMILPDZri VR512:$src1, imm:$imm)>;
//===----------------------------------------------------------------------===//
projects / oota-llvm.git / blobdiff