[JitDiff X64] [tannergooding] Accelerate System.Half arithmetic and conversi ...

[MihuBot]

Job completed in 23 minutes 38 seconds (remote runner delay: 56 seconds).
dotnet/runtime#130512

Main commit: dotnet/runtime@174a8af
PR commit: tannergooding/runtime@40b63e5

Diffs

Found 89 files with textual diffs.

Summary of Code Size diffs:
(Lower is better)

Total bytes of base: 158380445
Total bytes of diff: 158380500
Total bytes of delta: 55 (0.00 % of base)
Total relative delta: 1.00
    diff is a regression.
    relative diff is a regression.


Top file regressions (bytes):
          55 : System.Private.CoreLib.dasm (0.00 % of base)

1 total files with Code Size differences (0 improved, 1 regressed), 353 unchanged.

Top method regressions (bytes):
          55 (100.00 % of base) : System.Private.CoreLib.dasm - System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half (FullOpts)

Top method regressions (percentages):
          55 (100.00 % of base) : System.Private.CoreLib.dasm - System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half (FullOpts)

1 total methods with Code Size differences (0 improved, 1 regressed), 816794 unchanged.

--------------------------------------------------------------------------------

Artifacts:

• PackageList.json (32 KB)
• diff-frameworks.txt (999 B)
• ShortDiffsRegressions.md (4 KB)
• jit-diffs-pr.7z (140 MB)
• jit-diffs-main.7z (140 MB)

[MihuBot]

Top method regressions

55 (100.00 % of base) - System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half

 ; Assembly listing for method System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half (FullOpts)
 ; Emitting BLENDED_CODE for x64 + VEX + EVEX on Unix
 ; FullOpts code
 ; optimized code
-; rsp based frame
+; rbp based frame
 ; partially interruptible
 ; No PGO data
 ; Final local variable assignments
 ;
 ;* V00 arg0         [V00    ] (  0,  0   )  struct ( 8) zero-ref    single-def <System.Half>
 ;* V01 arg1         [V01    ] (  0,  0   )  struct ( 8) zero-ref    single-def <System.Half>
 ;* V02 arg2         [V02    ] (  0,  0   )  struct ( 8) zero-ref    single-def <System.Half>
 ;# V03 OutArgs      [V03    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace" <Empty>
-;* V04 tmp1         [V04    ] (  0,  0   )  struct ( 8) zero-ref    "System.Half tmp" <System.Half>
-;  V05 tmp2         [V05,T00] (  2,  2   )  ushort  ->  rdi         single-def "field V00._value (fldOffset=0x0)" P-INDEP
-;  V06 tmp3         [V06,T01] (  2,  2   )  ushort  ->  rsi         single-def "field V01._value (fldOffset=0x0)" P-INDEP
-;  V07 tmp4         [V07,T02] (  2,  2   )  ushort  ->  rdx         single-def "field V02._value (fldOffset=0x0)" P-INDEP
-;  V08 tmp5         [V08,T03] (  2,  2   )  ushort  ->  rax         "field V04._value (fldOffset=0x0)" P-INDEP
+;* V04 tmp1         [V04    ] (  0,  0   )  double  ->  zero-ref    "non-inline candidate call"
+;* V05 tmp2         [V05    ] (  0,  0   )  double  ->  zero-ref    "non-inline candidate call"
+;  V06 tmp3         [V06,T00] (  2,  2   )  ushort  ->  rdi         single-def "field V00._value (fldOffset=0x0)" P-INDEP
+;  V07 tmp4         [V07,T01] (  2,  2   )  ushort  ->  rbx         single-def "field V01._value (fldOffset=0x0)" P-INDEP
+;  V08 tmp5         [V08,T02] (  2,  2   )  ushort  ->  r15         single-def "field V02._value (fldOffset=0x0)" P-INDEP
+;  TEMP_02                                  double  ->  [rbp-0x18]
+;  TEMP_01                                  double  ->  [rbp-0x20]
 ;
-; Lcl frame size = 0
+; Lcl frame size = 16
 
 G_M3007_IG01:
-						;; size=0 bbWeight=1 PerfScore 0.00
+       push     rbp
+       push     r15
+       push     rbx
+       sub      rsp, 16
+       lea      rbp, [rsp+0x20]
+       mov      ebx, esi
+       mov      r15d, edx
+						;; size=18 bbWeight=1 PerfScore 4.25
 G_M3007_IG02:
-       movzx    rax, di
-       vmovd    xmm0, eax
-       vcvtph2ps xmm0, xmm0
-       movzx    rax, si
-       vmovd    xmm1, eax
-       vcvtph2ps xmm1, xmm1
-       movzx    rax, dx
-       vmovd    xmm2, eax
-       vcvtph2ps xmm2, xmm2
-       vfmadd213ss xmm0, xmm1, xmm2
-       vcvtps2ph xmm0, xmm0, 0
-       vmovd    eax, xmm0
-       movzx    rax, ax
-						;; size=54 bbWeight=1 PerfScore 28.00
+       movzx    rdi, di
+       mov      rax, 0xD1FFAB1E      ; code for System.Half:op_Explicit(System.Half):double
+       call     [rax]System.Half:op_Explicit(System.Half):double
+       vmovsd   qword ptr [rbp-0x18], xmm0
+       movzx    rdi, bx
+       mov      rax, 0xD1FFAB1E      ; code for System.Half:op_Explicit(System.Half):double
+       call     [rax]System.Half:op_Explicit(System.Half):double
+       vmovsd   qword ptr [rbp-0x20], xmm0
+       movzx    rdi, r15w
+       mov      rax, 0xD1FFAB1E      ; code for System.Half:op_Explicit(System.Half):double
+       call     [rax]System.Half:op_Explicit(System.Half):double
+       vmovsd   xmm1, qword ptr [rbp-0x18]
+       vfmadd132sd xmm1, xmm0, qword ptr [rbp-0x20]
+       vmovaps  xmm0, xmm1
+       mov      rax, 0xD1FFAB1E      ; code for System.Half:op_Explicit(double):System.Half
+						;; size=81 bbWeight=1 PerfScore 20.00
 G_M3007_IG03:
-       ret      
-						;; size=1 bbWeight=1 PerfScore 1.00
+       add      rsp, 16
+       pop      rbx
+       pop      r15
+       pop      rbp
+       tail.jmp [rax]System.Half:op_Explicit(double):System.Half
+						;; size=11 bbWeight=1 PerfScore 3.75
 
-; Total bytes of code 55, prolog size 0, PerfScore 29.00, instruction count 14, allocated bytes for code 55 (MethodHash=7841f440) for method System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half (FullOpts)
+; Total bytes of code 110, prolog size 13, PerfScore 28.00, instruction count 27, allocated bytes for code 110 (MethodHash=7841f440) for method System.Half:FusedMultiplyAdd(System.Half,System.Half,System.Half):System.Half (FullOpts)
 ; ============================================================

[MihuBot]

@tannergooding