[X64] [tannergooding] Change Vector2/3/4, Quaternion, Plane, Vector<T>, and Vector64/128/256/51 ... #400
Comments
Top method regressions451 (58.12 % of base) - System.Runtime.Intrinsics.Vector256`1[ubyte]:System.Runtime.Intrinsics.ISimdVector,T>.Dot(System.Runtime.Intrinsics.Vector256`1[ubyte],System.Runtime.Intrinsics.Vector256`1[ubyte]):ubyte ; Assembly listing for method System.Runtime.Intrinsics.Vector256`1[ubyte]:System.Runtime.Intrinsics.ISimdVector<System.Runtime.Intrinsics.Vector256<T>,T>.Dot(System.Runtime.Intrinsics.Vector256`1[ubyte],System.Runtime.Intrinsics.Vector256`1[ubyte]):ubyte (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX - Unix
; FullOpts code
; optimized code
; rbp based frame
-; partially interruptible
+; fully interruptible
; No PGO data
-; 0 inlinees with PGO data; 19 single block inlinees; 23 inlinees without PGO data
+; 0 inlinees with PGO data; 23 single block inlinees; 9 inlinees without PGO data
; Final local variable assignments
;
-; V00 arg0 [V00,T27] ( 2, 2 ) simd32 -> mm0 single-def <System.Runtime.Intrinsics.Vector256`1[ubyte]>
-; V01 arg1 [V01,T28] ( 2, 2 ) simd32 -> mm1 single-def <System.Runtime.Intrinsics.Vector256`1[ubyte]>
+; V00 arg0 [V00,T32] ( 2, 2 ) simd32 -> [rbp+0x10] single-def <System.Runtime.Intrinsics.Vector256`1[ubyte]>
+; V01 arg1 [V01,T33] ( 2, 2 ) simd32 -> [rbp+0x30] single-def <System.Runtime.Intrinsics.Vector256`1[ubyte]>
;# V02 OutArgs [V02 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
-; V03 tmp1 [V03,T12] ( 2, 4 ) int -> rax "impAppendStmt"
-; V04 tmp2 [V04,T23] ( 3, 6 ) simd16 -> [rbp-0x10] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V05 tmp3 [V05,T24] ( 3, 6 ) simd16 -> [rbp-0x20] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V06 tmp4 [V06,T13] ( 2, 4 ) int -> rax "impAppendStmt"
-; V07 tmp5 [V07,T04] ( 8, 8 ) ubyte -> rax ld-addr-op "Inline ldloca(s) first use temp"
-;* V08 tmp6 [V08,T19] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
-; V09 tmp7 [V09 ] ( 9, 18 ) struct ( 8) [rbp-0x28] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-; V10 tmp8 [V10 ] ( 9, 18 ) struct ( 8) [rbp-0x30] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-;* V11 tmp9 [V11 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-; V12 tmp10 [V12,T00] ( 8, 16 ) int -> registers "impAppendStmt"
-;* V13 tmp11 [V13 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-;* V14 tmp12 [V14 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
-;* V15 tmp13 [V15 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V16 tmp14 [V16 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V17 tmp15 [V17 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-;* V18 tmp16 [V18 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-; V19 tmp17 [V19,T08] ( 8, 8 ) ubyte -> registers "Inline return value spill temp"
-;* V20 tmp18 [V20 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V21 tmp19 [V21 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V22 tmp20 [V22 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V23 tmp21 [V23 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V24 tmp22 [V24 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V25 tmp23 [V25 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-; V26 tmp24 [V26,T05] ( 8, 8 ) ubyte -> rcx ld-addr-op "Inline ldloca(s) first use temp"
-;* V27 tmp25 [V27,T20] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
-; V28 tmp26 [V28 ] ( 9, 18 ) struct ( 8) [rbp-0x38] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-; V29 tmp27 [V29 ] ( 9, 18 ) struct ( 8) [rbp-0x40] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V03 tmp1 [V03,T12] ( 2, 4 ) int -> rbx "impAppendStmt"
+; V04 tmp2 [V04,T28] ( 3, 6 ) simd16 -> [rbp-0x30] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V05 tmp3 [V05,T29] ( 3, 6 ) simd16 -> [rbp-0x40] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V06 tmp4 [V06,T13] ( 2, 4 ) int -> rbx "impAppendStmt"
+;* V07 tmp5 [V07 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V08 tmp6 [V08 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V09 tmp7 [V09 ] ( 0, 0 ) struct ( 8) zero-ref "spilled call-like call argument" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V10 tmp8 [V10 ] ( 2, 5 ) struct ( 8) [rbp-0x48] do-not-enreg[XS] addr-exposed ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V11 tmp9 [V11,T04] ( 5, 17 ) int -> rsi "Inline stloc first use temp"
+; V12 tmp10 [V12 ] ( 2, 10 ) struct ( 8) [rbp-0x50] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V13 tmp11 [V13 ] ( 2, 10 ) struct ( 8) [rbp-0x58] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V14 tmp12 [V14 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+;* V15 tmp13 [V15 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
+;* V16 tmp14 [V16 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
+;* V17 tmp15 [V17 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
+;* V18 tmp16 [V18 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
+; V19 tmp17 [V19,T00] ( 16, 16 ) ubyte -> rbx ld-addr-op "Inline ldloca(s) first use temp"
+;* V20 tmp18 [V20,T24] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V21 tmp19 [V21 ] ( 9, 18 ) struct ( 8) [rbp-0x60] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V22 tmp20 [V22 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+;* V23 tmp21 [V23 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V24 tmp22 [V24 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V25 tmp23 [V25 ] ( 0, 0 ) struct ( 8) zero-ref "spilled call-like call argument" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V26 tmp24 [V26 ] ( 2, 5 ) struct ( 8) [rbp-0x68] do-not-enreg[XS] addr-exposed ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V27 tmp25 [V27,T05] ( 5, 17 ) int -> r15 "Inline stloc first use temp"
+; V28 tmp26 [V28 ] ( 2, 10 ) struct ( 8) [rbp-0x70] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V29 tmp27 [V29 ] ( 2, 10 ) struct ( 8) [rbp-0x78] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
;* V30 tmp28 [V30 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-; V31 tmp29 [V31,T01] ( 8, 16 ) int -> registers "impAppendStmt"
-;* V32 tmp30 [V32 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-;* V33 tmp31 [V33 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
-;* V34 tmp32 [V34 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V35 tmp33 [V35 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V36 tmp34 [V36 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-;* V37 tmp35 [V37 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-; V38 tmp36 [V38,T09] ( 8, 8 ) ubyte -> registers "Inline return value spill temp"
-;* V39 tmp37 [V39 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V40 tmp38 [V40 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V41 tmp39 [V41 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V42 tmp40 [V42 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V43 tmp41 [V43 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V44 tmp42 [V44 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V45 tmp43 [V45 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V46 tmp44 [V46 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V47 tmp45 [V47 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-; V48 tmp46 [V48,T25] ( 3, 6 ) simd16 -> [rbp-0x50] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V49 tmp47 [V49,T26] ( 3, 6 ) simd16 -> [rbp-0x60] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V50 tmp48 [V50,T14] ( 2, 4 ) int -> rcx "impAppendStmt"
-; V51 tmp49 [V51,T06] ( 8, 8 ) ubyte -> rcx ld-addr-op "Inline ldloca(s) first use temp"
-;* V52 tmp50 [V52,T21] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
-; V53 tmp51 [V53 ] ( 9, 18 ) struct ( 8) [rbp-0x68] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-; V54 tmp52 [V54 ] ( 9, 18 ) struct ( 8) [rbp-0x70] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-;* V55 tmp53 [V55 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-; V56 tmp54 [V56,T02] ( 8, 16 ) int -> registers "impAppendStmt"
-;* V57 tmp55 [V57 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-;* V58 tmp56 [V58 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
-;* V59 tmp57 [V59 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V60 tmp58 [V60 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V61 tmp59 [V61 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-;* V62 tmp60 [V62 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-; V63 tmp61 [V63,T10] ( 8, 8 ) ubyte -> registers "Inline return value spill temp"
-;* V64 tmp62 [V64 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V65 tmp63 [V65 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V66 tmp64 [V66 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V67 tmp65 [V67 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V68 tmp66 [V68 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V69 tmp67 [V69 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-; V70 tmp68 [V70,T07] ( 8, 8 ) ubyte -> rdx ld-addr-op "Inline ldloca(s) first use temp"
-;* V71 tmp69 [V71,T22] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
-; V72 tmp70 [V72 ] ( 9, 18 ) struct ( 8) [rbp-0x78] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-; V73 tmp71 [V73 ] ( 9, 18 ) struct ( 8) [rbp-0x80] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
-;* V74 tmp72 [V74 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-; V75 tmp73 [V75,T03] ( 8, 16 ) int -> registers "impAppendStmt"
-;* V76 tmp74 [V76 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
-;* V77 tmp75 [V77 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
-;* V78 tmp76 [V78 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V79 tmp77 [V79 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V80 tmp78 [V80 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-;* V81 tmp79 [V81 ] ( 0, 0 ) int -> zero-ref "Inlining Arg"
-; V82 tmp80 [V82,T11] ( 8, 8 ) ubyte -> registers "Inline return value spill temp"
-;* V83 tmp81 [V83 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V84 tmp82 [V84 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V85 tmp83 [V85 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V86 tmp84 [V86 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V87 tmp85 [V87 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V88 tmp86 [V88 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V89 tmp87 [V89 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V90 tmp88 [V90 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V91 tmp89 [V91 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-;* V92 tmp90 [V92 ] ( 0, 0 ) ubyte -> zero-ref "Inline return value spill temp"
-;* V93 tmp91 [V93 ] ( 0, 0 ) ubyte -> zero-ref "Inlining Arg"
-; V94 tmp92 [V94 ] ( 9, 17 ) long -> [rbp-0x28] do-not-enreg[X] addr-exposed "field V09._00 (fldOffset=0x0)" P-DEP
-; V95 tmp93 [V95 ] ( 9, 17 ) long -> [rbp-0x30] do-not-enreg[X] addr-exposed "field V10._00 (fldOffset=0x0)" P-DEP
-; V96 tmp94 [V96 ] ( 9, 17 ) long -> [rbp-0x38] do-not-enreg[X] addr-exposed "field V28._00 (fldOffset=0x0)" P-DEP
-; V97 tmp95 [V97 ] ( 9, 17 ) long -> [rbp-0x40] do-not-enreg[X] addr-exposed "field V29._00 (fldOffset=0x0)" P-DEP
-; V98 tmp96 [V98 ] ( 9, 17 ) long -> [rbp-0x68] do-not-enreg[X] addr-exposed "field V53._00 (fldOffset=0x0)" P-DEP
-; V99 tmp97 [V99 ] ( 9, 17 ) long -> [rbp-0x70] do-not-enreg[X] addr-exposed "field V54._00 (fldOffset=0x0)" P-DEP
-; V100 tmp98 [V100 ] ( 9, 17 ) long -> [rbp-0x78] do-not-enreg[X] addr-exposed "field V72._00 (fldOffset=0x0)" P-DEP
-; V101 tmp99 [V101 ] ( 9, 17 ) long -> [rbp-0x80] do-not-enreg[X] addr-exposed "field V73._00 (fldOffset=0x0)" P-DEP
-; V102 cse0 [V102,T15] ( 2, 2 ) int -> rcx "CSE #01: moderate"
-; V103 cse1 [V103,T16] ( 2, 2 ) int -> rdx "CSE #02: moderate"
-; V104 cse2 [V104,T17] ( 2, 2 ) int -> rdx "CSE #03: moderate"
-; V105 cse3 [V105,T18] ( 2, 2 ) int -> rdi "CSE #04: moderate"
+;* V31 tmp29 [V31 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
+; V32 tmp30 [V32,T01] ( 16, 16 ) ubyte -> rax ld-addr-op "Inline ldloca(s) first use temp"
+;* V33 tmp31 [V33,T25] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V34 tmp32 [V34 ] ( 9, 18 ) struct ( 8) [rbp-0x80] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V35 tmp33 [V35 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+; V36 tmp34 [V36,T30] ( 3, 6 ) simd16 -> [rbp-0x90] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V37 tmp35 [V37,T31] ( 3, 6 ) simd16 -> [rbp-0xA0] do-not-enreg[SF] "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V38 tmp36 [V38,T14] ( 2, 4 ) int -> r15 "impAppendStmt"
+;* V39 tmp37 [V39 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V40 tmp38 [V40 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V41 tmp39 [V41 ] ( 0, 0 ) struct ( 8) zero-ref "spilled call-like call argument" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V42 tmp40 [V42 ] ( 2, 5 ) struct ( 8) [rbp-0xA8] do-not-enreg[XS] addr-exposed ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V43 tmp41 [V43,T06] ( 5, 17 ) int -> r15 "Inline stloc first use temp"
+; V44 tmp42 [V44 ] ( 2, 10 ) struct ( 8) [rbp-0xB0] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V45 tmp43 [V45 ] ( 2, 10 ) struct ( 8) [rbp-0xB8] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V46 tmp44 [V46 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+;* V47 tmp45 [V47 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
+; V48 tmp46 [V48,T02] ( 16, 16 ) ubyte -> r15 ld-addr-op "Inline ldloca(s) first use temp"
+;* V49 tmp47 [V49,T26] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V50 tmp48 [V50 ] ( 9, 18 ) struct ( 8) [rbp-0xC0] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V51 tmp49 [V51 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+;* V52 tmp50 [V52 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V53 tmp51 [V53 ] ( 0, 0 ) struct ( 8) zero-ref "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V54 tmp52 [V54 ] ( 0, 0 ) struct ( 8) zero-ref "spilled call-like call argument" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V55 tmp53 [V55 ] ( 2, 5 ) struct ( 8) [rbp-0xC8] do-not-enreg[XS] addr-exposed ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V56 tmp54 [V56,T07] ( 5, 17 ) int -> r14 "Inline stloc first use temp"
+; V57 tmp55 [V57 ] ( 2, 10 ) struct ( 8) [rbp-0xD0] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+; V58 tmp56 [V58 ] ( 2, 10 ) struct ( 8) [rbp-0xD8] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V59 tmp57 [V59 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+;* V60 tmp58 [V60 ] ( 0, 0 ) ubyte -> zero-ref "Inline stloc first use temp"
+; V61 tmp59 [V61,T03] ( 16, 16 ) ubyte -> rax ld-addr-op "Inline ldloca(s) first use temp"
+;* V62 tmp60 [V62,T27] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V63 tmp61 [V63 ] ( 9, 18 ) struct ( 8) [rbp-0xE0] do-not-enreg[XS] addr-exposed ld-addr-op "Inlining Arg" <System.Runtime.Intrinsics.Vector64`1[ubyte]>
+;* V64 tmp62 [V64 ] ( 0, 0 ) int -> zero-ref "impAppendStmt"
+; V65 tmp63 [V65,T16] ( 2, 2 ) long -> rsi "field V07._00 (fldOffset=0x0)" P-INDEP
+; V66 tmp64 [V66,T17] ( 2, 2 ) long -> rdi "field V08._00 (fldOffset=0x0)" P-INDEP
+;* V67 tmp65 [V67 ] ( 0, 0 ) long -> zero-ref "field V09._00 (fldOffset=0x0)" P-INDEP
+; V68 tmp66 [V68 ] ( 2, 5 ) long -> [rbp-0x48] do-not-enreg[X] addr-exposed "field V10._00 (fldOffset=0x0)" P-DEP
+; V69 tmp67 [V69 ] ( 2, 9 ) long -> [rbp-0x50] do-not-enreg[X] addr-exposed "field V12._00 (fldOffset=0x0)" P-DEP
+; V70 tmp68 [V70 ] ( 2, 9 ) long -> [rbp-0x58] do-not-enreg[X] addr-exposed "field V13._00 (fldOffset=0x0)" P-DEP
+; V71 tmp69 [V71 ] ( 9, 17 ) long -> [rbp-0x60] do-not-enreg[X] addr-exposed "field V21._00 (fldOffset=0x0)" P-DEP
+; V72 tmp70 [V72,T18] ( 2, 2 ) long -> rdi "field V23._00 (fldOffset=0x0)" P-INDEP
+; V73 tmp71 [V73,T19] ( 2, 2 ) long -> rsi "field V24._00 (fldOffset=0x0)" P-INDEP
+;* V74 tmp72 [V74 ] ( 0, 0 ) long -> zero-ref "field V25._00 (fldOffset=0x0)" P-INDEP
+; V75 tmp73 [V75 ] ( 2, 5 ) long -> [rbp-0x68] do-not-enreg[X] addr-exposed "field V26._00 (fldOffset=0x0)" P-DEP
+; V76 tmp74 [V76 ] ( 2, 9 ) long -> [rbp-0x70] do-not-enreg[X] addr-exposed "field V28._00 (fldOffset=0x0)" P-DEP
+; V77 tmp75 [V77 ] ( 2, 9 ) long -> [rbp-0x78] do-not-enreg[X] addr-exposed "field V29._00 (fldOffset=0x0)" P-DEP
+; V78 tmp76 [V78 ] ( 9, 17 ) long -> [rbp-0x80] do-not-enreg[X] addr-exposed "field V34._00 (fldOffset=0x0)" P-DEP
+; V79 tmp77 [V79,T20] ( 2, 2 ) long -> rdi "field V39._00 (fldOffset=0x0)" P-INDEP
+; V80 tmp78 [V80,T21] ( 2, 2 ) long -> rsi "field V40._00 (fldOffset=0x0)" P-INDEP
+;* V81 tmp79 [V81 ] ( 0, 0 ) long -> zero-ref "field V41._00 (fldOffset=0x0)" P-INDEP
+; V82 tmp80 [V82 ] ( 2, 5 ) long -> [rbp-0xA8] do-not-enreg[X] addr-exposed "field V42._00 (fldOffset=0x0)" P-DEP
+; V83 tmp81 [V83 ] ( 2, 9 ) long -> [rbp-0xB0] do-not-enreg[X] addr-exposed "field V44._00 (fldOffset=0x0)" P-DEP
+; V84 tmp82 [V84 ] ( 2, 9 ) long -> [rbp-0xB8] do-not-enreg[X] addr-exposed "field V45._00 (fldOffset=0x0)" P-DEP
+; V85 tmp83 [V85 ] ( 9, 17 ) long -> [rbp-0xC0] do-not-enreg[X] addr-exposed "field V50._00 (fldOffset=0x0)" P-DEP
+; V86 tmp84 [V86,T22] ( 2, 2 ) long -> rdi "field V52._00 (fldOffset=0x0)" P-INDEP
+; V87 tmp85 [V87,T23] ( 2, 2 ) long -> rsi "field V53._00 (fldOffset=0x0)" P-INDEP
+;* V88 tmp86 [V88 ] ( 0, 0 ) long -> zero-ref "field V54._00 (fldOffset=0x0)" P-INDEP
+; V89 tmp87 [V89 ] ( 2, 5 ) long -> [rbp-0xC8] do-not-enreg[X] addr-exposed "field V55._00 (fldOffset=0x0)" P-DEP
+; V90 tmp88 [V90 ] ( 2, 9 ) long -> [rbp-0xD0] do-not-enreg[X] addr-exposed "field V57._00 (fldOffset=0x0)" P-DEP
+; V91 tmp89 [V91 ] ( 2, 9 ) long -> [rbp-0xD8] do-not-enreg[X] addr-exposed "field V58._00 (fldOffset=0x0)" P-DEP
+; V92 tmp90 [V92 ] ( 9, 17 ) long -> [rbp-0xE0] do-not-enreg[X] addr-exposed "field V63._00 (fldOffset=0x0)" P-DEP
+; V93 tmp91 [V93,T15] ( 2, 4 ) int -> rsi "argument with side effect"
+; V94 cse0 [V94,T08] ( 4, 16 ) long -> rax "CSE #01: aggressive"
+; V95 cse1 [V95,T09] ( 4, 16 ) long -> r14 "CSE #02: aggressive"
+; V96 cse2 [V96,T10] ( 4, 16 ) long -> r14 "CSE #03: aggressive"
+; V97 cse3 [V97,T11] ( 4, 16 ) long -> r13 "CSE #04: aggressive"
;
-; Lcl frame size = 128
+; Lcl frame size = 192
G_M42821_IG01:
push rbp
- sub rsp, 128
- lea rbp, [rsp+0x80]
+ push r15
+ push r14
+ push r13
+ push rbx
+ sub rsp, 192
+ lea rbp, [rsp+0xE0]
vmovups ymm0, ymmword ptr [rbp+0x10]
vmovups ymm1, ymmword ptr [rbp+0x30]
- ;; size=26 bbWeight=1 PerfScore 9.75
+ ;; size=33 bbWeight=1 PerfScore 13.75
G_M42821_IG02:
+ vmovups ymmword ptr [rbp+0x10], ymm0
vmovaps ymm2, ymm0
- vmovaps xmmword ptr [rbp-0x10], xmm2
+ vmovaps xmmword ptr [rbp-0x30], xmm2
+ vmovups ymmword ptr [rbp+0x30], ymm1
vmovaps ymm2, ymm1
- vmovaps xmmword ptr [rbp-0x20], xmm2
- mov rax, qword ptr [rbp-0x10]
- mov qword ptr [rbp-0x28], rax
- mov rax, qword ptr [rbp-0x20]
- mov qword ptr [rbp-0x30], rax
- movzx rax, byte ptr [rbp-0x28]
- movzx rcx, byte ptr [rbp-0x30]
- imul eax, ecx
- movzx rax, al
- movzx rcx, byte ptr [rbp-0x27]
- movzx rdx, byte ptr [rbp-0x2F]
- imul ecx, edx
- movzx rcx, cl
- add eax, ecx
- movzx rcx, al
- movzx rax, cl
- movzx rcx, byte ptr [rbp-0x26]
- movzx rdx, byte ptr [rbp-0x2E]
- imul ecx, edx
- movzx rcx, cl
- movzx rdx, byte ptr [rbp-0x25]
- movzx rdi, byte ptr [rbp-0x2D]
- imul edx, edi
- movzx rdx, dl
- add ecx, edx
- add eax, ecx
- movzx rax, al
- movzx rdx, byte ptr [rbp-0x24]
- movzx rcx, byte ptr [rbp-0x2C]
- imul ecx, edx
- movzx rcx, cl
- movzx rdx, byte ptr [rbp-0x23]
- movzx rdi, byte ptr [rbp-0x2B]
- imul edx, edi
- movzx rdx, dl
- add ecx, edx
- add ecx, eax
- movzx rax, cl
- movzx rdx, byte ptr [rbp-0x22]
- movzx rcx, byte ptr [rbp-0x2A]
- imul ecx, edx
- movzx rcx, cl
- movzx rdx, byte ptr [rbp-0x21]
- movzx rdi, byte ptr [rbp-0x29]
- imul edx, edi
- movzx rdx, dl
- add ecx, edx
- add ecx, eax
- movzx rax, cl
- mov rcx, qword ptr [rbp-0x08]
- mov qword ptr [rbp-0x38], rcx
- mov rcx, qword ptr [rbp-0x18]
- mov qword ptr [rbp-0x40], rcx
- movzx rcx, byte ptr [rbp-0x38]
- movzx rdx, byte ptr [rbp-0x40]
- imul ecx, edx
- movzx rcx, cl
- movzx rdx, byte ptr [rbp-0x37]
- movzx rdi, byte ptr [rbp-0x3F]
- imul edx, edi
- movzx rdx, dl
- add ecx, edx
- movzx rdx, cl
- movzx rcx, dl
- movzx rdx, byte ptr [rbp-0x36]
- movzx rdi, byte ptr [rbp-0x3E]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x35]
- ;; size=251 bbWeight=1 PerfScore 62.00
+ vmovaps xmmword ptr [rbp-0x40], xmm2
+ mov rsi, qword ptr [rbp-0x30]
+ mov rdi, qword ptr [rbp-0x40]
+ mov qword ptr [rbp-0x50], rsi
+ mov qword ptr [rbp-0x58], rdi
+ xor esi, esi
+ align [0 bytes for IG03]
+ ;; size=46 bbWeight=1 PerfScore 8.75
G_M42821_IG03:
- movzx rsi, byte ptr [rbp-0x3D]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add ecx, edx
- movzx rcx, cl
- movzx rdi, byte ptr [rbp-0x34]
- movzx rdx, byte ptr [rbp-0x3C]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x33]
- movzx rsi, byte ptr [rbp-0x3B]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add edx, ecx
- movzx rcx, dl
- movzx rdi, byte ptr [rbp-0x32]
- movzx rdx, byte ptr [rbp-0x3A]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x31]
- movzx rsi, byte ptr [rbp-0x39]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add edx, ecx
- movzx rcx, dl
- add eax, ecx
- movzx rax, al
- vextractf128 xmm0, ymm0, 1
- vmovaps xmmword ptr [rbp-0x50], xmm0
- vextractf128 xmm0, ymm1, 1
- vmovaps xmmword ptr [rbp-0x60], xmm0
- mov rcx, qword ptr [rbp-0x50]
- mov qword ptr [rbp-0x68], rcx
- mov rcx, qword ptr [rbp-0x60]
- mov qword ptr [rbp-0x70], rcx
- movzx rcx, byte ptr [rbp-0x68]
- movzx rdx, byte ptr [rbp-0x70]
- imul ecx, edx
- movzx rcx, cl
- movzx rdx, byte ptr [rbp-0x67]
- movzx rdi, byte ptr [rbp-0x6F]
- imul edx, edi
- movzx rdx, dl
- add ecx, edx
- movzx rdx, cl
- movzx rcx, dl
- movzx rdx, byte ptr [rbp-0x66]
- movzx rdi, byte ptr [rbp-0x6E]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x65]
- movzx rsi, byte ptr [rbp-0x6D]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add ecx, edx
- movzx rcx, cl
- movzx rdi, byte ptr [rbp-0x64]
- movzx rdx, byte ptr [rbp-0x6C]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x63]
- movzx rsi, byte ptr [rbp-0x6B]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add edx, ecx
- movzx rcx, dl
- movzx rdi, byte ptr [rbp-0x62]
- movzx rdx, byte ptr [rbp-0x6A]
- imul edx, edi
- ;; size=267 bbWeight=1 PerfScore 64.75
+ lea rdi, [rbp-0x50]
+ movsxd rax, esi
+ movzx rdi, byte ptr [rdi+rax]
+ lea rcx, [rbp-0x58]
+ movzx rcx, byte ptr [rcx+rax]
+ imul edi, ecx
+ lea rcx, [rbp-0x48]
+ mov byte ptr [rcx+rax], dil
+ inc esi
+ cmp esi, 8
+ jl SHORT G_M42821_IG03
+ ;; size=38 bbWeight=4 PerfScore 41.00
G_M42821_IG04:
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x61]
- movzx rsi, byte ptr [rbp-0x69]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- add edx, ecx
- movzx rcx, dl
- mov rdx, qword ptr [rbp-0x48]
- mov qword ptr [rbp-0x78], rdx
- mov rdx, qword ptr [rbp-0x58]
- mov qword ptr [rbp-0x80], rdx
- movzx rdx, byte ptr [rbp-0x78]
- movzx rdi, byte ptr [rbp-0x80]
- imul edx, edi
- movzx rdx, dl
- movzx rdi, byte ptr [rbp-0x77]
+ mov rsi, qword ptr [rbp-0x48]
+ mov qword ptr [rbp-0x60], rsi
+ movzx rsi, byte ptr [rbp-0x60]
+ xor edi, edi
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5F]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5E]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5D]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5C]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5B]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x5A]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ movzx rsi, byte ptr [rbp-0x59]
+ mov edi, ebx
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ mov rdi, qword ptr [rbp-0x28]
+ mov rsi, qword ptr [rbp-0x38]
+ mov qword ptr [rbp-0x70], rdi
+ mov qword ptr [rbp-0x78], rsi
+ xor r15d, r15d
+ ;; size=195 bbWeight=1 PerfScore 44.25
+G_M42821_IG05:
+ lea rdi, [rbp-0x70]
+ movsxd r14, r15d
+ movzx rdi, byte ptr [rdi+r14]
+ lea rsi, [rbp-0x78]
+ movzx rsi, byte ptr [rsi+r14]
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ lea rsi, [rbp-0x68]
+ mov byte ptr [rsi+r14], al
+ inc r15d
+ cmp r15d, 8
+ jl SHORT G_M42821_IG05
+ ;; size=50 bbWeight=4 PerfScore 46.00
+G_M42821_IG06:
+ mov rsi, qword ptr [rbp-0x68]
+ mov qword ptr [rbp-0x80], rsi
+ movzx rsi, byte ptr [rbp-0x80]
+ xor edi, edi
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
movzx rsi, byte ptr [rbp-0x7F]
- imul edi, esi
- movzx rdi, dil
- add edx, edi
- movzx rdi, dl
- movzx rdx, dil
- movzx rdi, byte ptr [rbp-0x76]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
movzx rsi, byte ptr [rbp-0x7E]
- imul edi, esi
- movzx rdi, dil
- movzx rsi, byte ptr [rbp-0x75]
- movzx r8, byte ptr [rbp-0x7D]
- imul esi, r8d
- movzx rsi, sil
- add edi, esi
- add edx, edi
- movzx rdx, dl
- movzx rsi, byte ptr [rbp-0x74]
- movzx rdi, byte ptr [rbp-0x7C]
- imul edi, esi
- movzx rdi, dil
- movzx rsi, byte ptr [rbp-0x73]
- movzx r8, byte ptr [rbp-0x7B]
- imul esi, r8d
- movzx rsi, sil
- add edi, esi
- add edi, edx
- movzx rdx, dil
- movzx rsi, byte ptr [rbp-0x72]
- movzx rdi, byte ptr [rbp-0x7A]
- imul edi, esi
- movzx rdi, dil
- movzx rsi, byte ptr [rbp-0x71]
- movzx r8, byte ptr [rbp-0x79]
- imul esi, r8d
- movzx rsi, sil
- add edi, esi
- add edi, edx
- movzx rdx, dil
- add ecx, edx
- add eax, ecx
- movzx rax, al
- ;; size=220 bbWeight=1 PerfScore 47.00
-G_M42821_IG05:
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0x7D]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0x7C]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0x7B]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0x7A]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0x79]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rdi, bl
+ mov esi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov ebx, eax
+ vmovups ymm0, ymmword ptr [rbp+0x10]
+ vextractf128 xmm0, ymm0, 1
+ vmovaps xmmword ptr [rbp-0x90], xmm0
+ vmovups ymm1, ymmword ptr [rbp+0x30]
+ vextractf128 xmm0, ymm1, 1
+ vmovaps xmmword ptr [rbp-0xA0], xmm0
+ mov rdi, qword ptr [rbp-0x90]
+ mov rsi, qword ptr [rbp-0xA0]
+ mov qword ptr [rbp-0xB0], rdi
+ mov qword ptr [rbp-0xB8], rsi
+ xor r15d, r15d
+ ;; size=249 bbWeight=1 PerfScore 60.25
+G_M42821_IG07:
+ lea rdi, [rbp-0xB0]
+ movsxd r14, r15d
+ movzx rdi, byte ptr [rdi+r14]
+ lea rsi, [rbp-0xB8]
+ movzx rsi, byte ptr [rsi+r14]
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ lea rsi, [rbp-0xA8]
+ mov byte ptr [rsi+r14], al
+ inc r15d
+ cmp r15d, 8
+ jl SHORT G_M42821_IG07
+ ;; size=59 bbWeight=4 PerfScore 46.00
+G_M42821_IG08:
+ mov rsi, qword ptr [rbp-0xA8]
+ mov qword ptr [rbp-0xC0], rsi
+ movzx rsi, byte ptr [rbp-0xC0]
+ xor edi, edi
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBF]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBE]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBD]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBC]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBB]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xBA]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ movzx rsi, byte ptr [rbp-0xB9]
+ mov edi, r15d
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov r15d, eax
+ mov rdi, qword ptr [rbp-0x88]
+ mov rsi, qword ptr [rbp-0x98]
+ mov qword ptr [rbp-0xD0], rdi
+ mov qword ptr [rbp-0xD8], rsi
+ xor r14d, r14d
+ ;; size=252 bbWeight=1 PerfScore 44.25
+G_M42821_IG09:
+ lea rdi, [rbp-0xD0]
+ movsxd r13, r14d
+ movzx rdi, byte ptr [rdi+r13]
+ lea rsi, [rbp-0xD8]
+ movzx rsi, byte ptr [rsi+r13]
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Multiply(ubyte,ubyte):ubyte
+ lea rsi, [rbp-0xC8]
+ mov byte ptr [rsi+r13], al
+ inc r14d
+ cmp r14d, 8
+ jl SHORT G_M42821_IG09
+ ;; size=59 bbWeight=4 PerfScore 46.00
+G_M42821_IG10:
+ mov rsi, qword ptr [rbp-0xC8]
+ mov qword ptr [rbp-0xE0], rsi
+ movzx rsi, byte ptr [rbp-0xE0]
+ xor edi, edi
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDF]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDE]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDD]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDC]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDB]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xDA]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rsi, byte ptr [rbp-0xD9]
+ mov edi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ movzx rdi, r15b
+ mov esi, eax
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ mov esi, eax
+ movzx rdi, bl
+ mov rax, 0xD1FFAB1E ; code for System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ call [rax]System.Runtime.Intrinsics.Scalar`1[ubyte]:Add(ubyte,ubyte):ubyte
+ nop
+ ;; size=227 bbWeight=1 PerfScore 45.75
+G_M42821_IG11:
vzeroupper
- add rsp, 128
+ add rsp, 192
+ pop rbx
+ pop r13
+ pop r14
+ pop r15
pop rbp
ret
- ;; size=12 bbWeight=1 PerfScore 2.75
+ ;; size=19 bbWeight=1 PerfScore 4.75
-; Total bytes of code 776, prolog size 16, PerfScore 186.25, instruction count 214, allocated bytes for code 776 (MethodHash=d63158ba) for method System.Runtime.Intrinsics.Vector256`1[ubyte]:System.Runtime.Intrinsics.ISimdVector<System.Runtime.Intrinsics.Vector256<T>,T>.Dot(System.Runtime.Intrinsics.Vector256`1[ubyte],System.Runtime.Intrinsics.Vector256`1[ubyte]):ubyte (FullOpts)
+; Total bytes of code 1227, prolog size 33, PerfScore 400.75, instruction count 263, allocated bytes for code 1227 (MethodHash=d63158ba) for method System.Runtime.Intrinsics.Vector256`1[ubyte]:System.Runtime.Intrinsics.ISimdVector<System.Runtime.Intrinsics.Vector256<T>,T>.Dot(System.Runtime.Intrinsics.Vector256`1[ubyte],System.Runtime.Intrinsics.Vector256`1[ubyte]):ubyte (FullOpts)Note: some changes were skipped as they were too large to fit into a comment. Larger list of diffs: https://gist.github.com/MihuBot/f6ef2f25cbdf1a6fa108ff971f43d624 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Job completed in 29 minutes.
dotnet/runtime#102301
Diffs
Diffs
Artifacts:
The text was updated successfully, but these errors were encountered: