diff options
| -rw-r--r-- | lib/Target/X86/X86Instr64bit.td | 18 | ||||
| -rw-r--r-- | lib/Target/X86/X86InstrSSE.td | 717 |
2 files changed, 374 insertions, 361 deletions
diff --git a/lib/Target/X86/X86Instr64bit.td b/lib/Target/X86/X86Instr64bit.td index 5ca8555a0f..1fd7f3bbaf 100644 --- a/lib/Target/X86/X86Instr64bit.td +++ b/lib/Target/X86/X86Instr64bit.td @@ -2353,24 +2353,6 @@ def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src), // X86-64 SSE4.1 Instructions //===----------------------------------------------------------------------===// -/// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination -multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> { - def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst), - (ins VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set GR64:$dst, - (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W; - def mr : SS4AIi8<opc, MRMDestMem, (outs), - (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(store (extractelt (v2i64 VR128:$src1), imm:$src2), - addr:$dst)]>, OpSize, REX_W; -} - -defm PEXTRQ : SS41I_extract64<0x16, "pextrq">; - let Constraints = "$src1 = $dst" in { multiclass SS41I_insert64<bits<8> opc, string OpcodeStr> { def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 72619ad171..53f9e6812b 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -3900,7 +3900,376 @@ def : Pat<(store (v16i8 VR128:$src), addr:$dst), (MOVUPSmr addr:$dst, VR128:$src)>; //===----------------------------------------------------------------------===// -// SSE4.1 - Misc Instructions +// SSE4.1 - Packed Move with Sign/Zero Extend +//===----------------------------------------------------------------------===// + +multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> { + def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; + + def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, + (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>, + OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { +defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>, + VEX; +defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>, + VEX; +defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>, + VEX; +defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>, + VEX; +defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>, + VEX; +defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>, + VEX; +} + +defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>; +defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>; +defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>; +defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>; +defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>; +defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>; + +// Common patterns involving scalar load. +def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)), + (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)), + (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)), + (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)), + (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)), + (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)), + (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)), + (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)), + (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)), + (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)), + (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)), + (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)), + (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>; + + +multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> { + def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; + + def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, + (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>, + OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { +defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>, + VEX; +defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>, + VEX; +defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>, + VEX; +defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>, + VEX; +} + +defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>; +defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>; +defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>; +defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>; + +// Common patterns involving scalar load +def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)), + (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)), + (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)), + (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>; +def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)), + (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>; + + +multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> { + def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; + + // Expecting a i16 load any extended to i32 value. + def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (IntId (bitconvert + (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>, + OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { +defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>, + VEX; +defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>, + VEX; +} +defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>; +defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>; + +// Common patterns involving scalar load +def : Pat<(int_x86_sse41_pmovsxbq + (bitconvert (v4i32 (X86vzmovl + (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), + (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>; + +def : Pat<(int_x86_sse41_pmovzxbq + (bitconvert (v4i32 (X86vzmovl + (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), + (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>; + +//===----------------------------------------------------------------------===// +// SSE4.1 - Extract Instructions +//===----------------------------------------------------------------------===// + +/// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem +multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> { + def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), + (ins VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>, + OpSize; + def mr : SS4AIi8<opc, MRMDestMem, (outs), + (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, OpSize; +// FIXME: +// There's an AssertZext in the way of writing the store pattern +// (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst) +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX; + +defm PEXTRB : SS41I_extract8<0x14, "pextrb">; + + +/// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination +multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> { + def mr : SS4AIi8<opc, MRMDestMem, (outs), + (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, OpSize; +// FIXME: +// There's an AssertZext in the way of writing the store pattern +// (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst) +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX; + +defm PEXTRW : SS41I_extract16<0x15, "pextrw">; + + +/// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination +multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> { + def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), + (ins VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set GR32:$dst, + (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize; + def mr : SS4AIi8<opc, MRMDestMem, (outs), + (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(store (extractelt (v4i32 VR128:$src1), imm:$src2), + addr:$dst)]>, OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX; + +defm PEXTRD : SS41I_extract32<0x16, "pextrd">; + +/// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination +multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> { + def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst), + (ins VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set GR64:$dst, + (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W; + def mr : SS4AIi8<opc, MRMDestMem, (outs), + (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(store (extractelt (v2i64 VR128:$src1), imm:$src2), + addr:$dst)]>, OpSize, REX_W; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W; + +defm PEXTRQ : SS41I_extract64<0x16, "pextrq">; + +/// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory +/// destination +multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> { + def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), + (ins VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set GR32:$dst, + (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>, + OpSize; + def mr : SS4AIi8<opc, MRMDestMem, (outs), + (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2), + addr:$dst)]>, OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX; +defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">; + +// Also match an EXTRACTPS store when the store is done as f32 instead of i32. +def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)), + imm:$src2))), + addr:$dst), + (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>, + Requires<[HasSSE41]>; + +//===----------------------------------------------------------------------===// +// SSE4.1 - Insert Instructions +//===----------------------------------------------------------------------===// + +multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> { + def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, GR32:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize; + def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (X86pinsrb VR128:$src1, (extloadi8 addr:$src2), + imm:$src3))]>, OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V; +let Constraints = "$src1 = $dst" in + defm PINSRB : SS41I_insert8<0x20, "pinsrb">; + +multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> { + def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, GR32:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>, + OpSize; + def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2), + imm:$src3)))]>, OpSize; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V; +let Constraints = "$src1 = $dst" in + defm PINSRD : SS41I_insert32<0x22, "pinsrd">; + +multiclass SS41I_insert64_avx<bits<8> opc, string OpcodeStr> { + def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, GR64:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128:$dst, + (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>, + OpSize, REX_W; + def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128:$dst, + (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2), + imm:$src3)))]>, OpSize, REX_W; +} + +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VPINSRQ : SS41I_insert64_avx<0x22, "vpinsrq">, VEX_4V, VEX_W; + +// insertps has a few different modes, there's the first two here below which +// are optimized inserts that won't zero arbitrary elements in the destination +// vector. The next one matches the intrinsic and could zero arbitrary elements +// in the target vector. +multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> { + def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, VR128:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>, + OpSize; + def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3), + !if(Is2Addr, + !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + !strconcat(asm, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), + [(set VR128:$dst, + (X86insrtps VR128:$src1, + (v4f32 (scalar_to_vector (loadf32 addr:$src2))), + imm:$src3))]>, OpSize; +} + +let Constraints = "$src1 = $dst" in + defm INSERTPS : SS41I_insertf32<0x21, "insertps">; +let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in + defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V; + +def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3), + (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>; + +//===----------------------------------------------------------------------===// +// SSE4.1 - Round Instructions //===----------------------------------------------------------------------===// multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, @@ -4078,6 +4447,10 @@ let Constraints = "$src1 = $dst" in defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round", int_x86_sse41_round_ss, int_x86_sse41_round_sd>; +//===----------------------------------------------------------------------===// +// SSE4.1 - Misc Instructions +//===----------------------------------------------------------------------===// + // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16. multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr, Intrinsic IntId128> { @@ -4291,347 +4664,6 @@ defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>; defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>; defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>; - -multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; - - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, - (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>, - OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { -defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>, - VEX; -defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>, - VEX; -defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>, - VEX; -defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>, - VEX; -defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>, - VEX; -defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>, - VEX; -} - -defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>; -defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>; -defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>; -defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>; -defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>; -defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>; - -// Common patterns involving scalar load. -def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)), - (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)), - (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)), - (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)), - (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)), - (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)), - (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)), - (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)), - (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)), - (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)), - (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)), - (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)), - (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>; - - -multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; - - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, - (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>, - OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { -defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>, - VEX; -defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>, - VEX; -defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>, - VEX; -defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>, - VEX; -} - -defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>; -defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>; -defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>; -defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>; - -// Common patterns involving scalar load -def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)), - (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)), - (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)), - (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>; -def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)), - (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>; - - -multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> { - def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId VR128:$src))]>, OpSize; - - // Expecting a i16 load any extended to i32 value. - def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (IntId (bitconvert - (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>, - OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { -defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>, - VEX; -defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>, - VEX; -} -defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>; -defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>; - -// Common patterns involving scalar load -def : Pat<(int_x86_sse41_pmovsxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>; - -def : Pat<(int_x86_sse41_pmovzxbq - (bitconvert (v4i32 (X86vzmovl - (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), - (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>; - - -/// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem -multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> { - def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), - (ins VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>, - OpSize; - def mr : SS4AIi8<opc, MRMDestMem, (outs), - (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - []>, OpSize; -// FIXME: -// There's an AssertZext in the way of writing the store pattern -// (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst) -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX; - -defm PEXTRB : SS41I_extract8<0x14, "pextrb">; - - -/// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination -multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> { - def mr : SS4AIi8<opc, MRMDestMem, (outs), - (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - []>, OpSize; -// FIXME: -// There's an AssertZext in the way of writing the store pattern -// (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst) -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX; - -defm PEXTRW : SS41I_extract16<0x15, "pextrw">; - - -/// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination -multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> { - def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), - (ins VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set GR32:$dst, - (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize; - def mr : SS4AIi8<opc, MRMDestMem, (outs), - (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(store (extractelt (v4i32 VR128:$src1), imm:$src2), - addr:$dst)]>, OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX; - -defm PEXTRD : SS41I_extract32<0x16, "pextrd">; - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W; - -/// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory -/// destination -multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> { - def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst), - (ins VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set GR32:$dst, - (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>, - OpSize; - def mr : SS4AIi8<opc, MRMDestMem, (outs), - (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2), - addr:$dst)]>, OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX; -defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">; - -// Also match an EXTRACTPS store when the store is done as f32 instead of i32. -def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)), - imm:$src2))), - addr:$dst), - (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>, - Requires<[HasSSE41]>; - -multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> { - def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), - (ins VR128:$src1, GR32:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize; - def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (X86pinsrb VR128:$src1, (extloadi8 addr:$src2), - imm:$src3))]>, OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V; -let Constraints = "$src1 = $dst" in - defm PINSRB : SS41I_insert8<0x20, "pinsrb">; - -multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> { - def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), - (ins VR128:$src1, GR32:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>, - OpSize; - def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2), - imm:$src3)))]>, OpSize; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V; -let Constraints = "$src1 = $dst" in - defm PINSRD : SS41I_insert32<0x22, "pinsrd">; - -multiclass SS41I_insert64_avx<bits<8> opc, string OpcodeStr> { - def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), - (ins VR128:$src1, GR64:$src2, i32i8imm:$src3), - !strconcat(OpcodeStr, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - [(set VR128:$dst, - (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>, - OpSize, REX_W; - def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3), - !strconcat(OpcodeStr, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - [(set VR128:$dst, - (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2), - imm:$src3)))]>, OpSize, REX_W; -} - -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VPINSRQ : SS41I_insert64_avx<0x22, "vpinsrq">, VEX_4V, VEX_W; - -// insertps has a few different modes, there's the first two here below which -// are optimized inserts that won't zero arbitrary elements in the destination -// vector. The next one matches the intrinsic and could zero arbitrary elements -// in the target vector. -multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> { - def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst), - (ins VR128:$src1, VR128:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>, - OpSize; - def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3), - !if(Is2Addr, - !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), - !strconcat(asm, - "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [(set VR128:$dst, - (X86insrtps VR128:$src1, - (v4f32 (scalar_to_vector (loadf32 addr:$src2))), - imm:$src3))]>, OpSize; -} - -let Constraints = "$src1 = $dst" in - defm INSERTPS : SS41I_insertf32<0x21, "insertps">; -let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in - defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V; - -def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3), - (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>; - // ptest instruction we'll lower to this in X86ISelLowering primarily from // the intel intrinsic that corresponds to this. let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in { @@ -4666,7 +4698,6 @@ def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>, OpSize; - //===----------------------------------------------------------------------===// // SSE4.2 Instructions //===----------------------------------------------------------------------===// |