[JitDiff X64] [tannergooding] Rework how EVEX+AVX512+AVX10 are supported by ...

[MihuBot]

Job completed in 20 minutes 53 seconds (remote runner delay: 40 seconds).

Using arguments: -NoPRLink -intel

Diffs

Found 260 files with textual diffs.

Summary of Code Size diffs:
(Lower is better)

Total bytes of base: 40546786
Total bytes of diff: 40546919
Total bytes of delta: 133 (0.00 % of base)
Total relative delta: Infinity
    diff is a regression.
    relative diff is a regression.


Top file regressions (bytes):
         109 : System.DirectoryServices.Protocols.dasm (0.08 % of base)
          24 : System.Private.CoreLib.dasm (0.00 % of base)

2 total files with Code Size differences (0 improved, 2 regressed), 258 unchanged.

Top method regressions (bytes):
         109 (Infinity of base) : System.DirectoryServices.Protocols.dasm - System.DirectoryServices.Protocols.PartialResultsRetriever:ThreadRoutine():this (FullOpts)
          12 (19.05 % of base) : System.Private.CoreLib.dasm - System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
           6 (17.14 % of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte] (FullOpts)
           6 (17.14 % of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)

Top method regressions (percentages):
         109 (Infinity of base) : System.DirectoryServices.Protocols.dasm - System.DirectoryServices.Protocols.PartialResultsRetriever:ThreadRoutine():this (FullOpts)
          12 (19.05 % of base) : System.Private.CoreLib.dasm - System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
           6 (17.14 % of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte] (FullOpts)
           6 (17.14 % of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)

4 total methods with Code Size differences (0 improved, 4 regressed), 245105 unchanged.

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

Artifacts:

• build-clr-checked-main.zip (88 MB)
• build-artifacts-main.zip (677 MB)
• build-clr-checked-pr.zip (87 MB)
• build-artifacts-pr.zip (676 MB)
• jit-diffs-pr.zip (135 MB)
• jit-diffs-main.zip (135 MB)
• diff-frameworks.txt (2 KB)
• ShortDiffsRegressions.md (7 KB)

[MihaZupan]

dotnet/runtime#115983

[MihuBot]

Top method regressions

12 (19.05 % of base) - System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte]

 ; Assembly listing for method System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
-; Emitting BLENDED_CODE for X64 with AVX512 - Unix
+; Emitting BLENDED_CODE for generic X64 + VEX + EVEX on Unix
 ; FullOpts code
 ; optimized code
 ; rsp based frame
 ; partially interruptible
 ; No PGO data
 ; Final local variable assignments
 ;
 ;  V00 RetBuf       [V00,T00] (  4,  4   )    long  ->  rdi         single-def
 ;  V01 arg0         [V01,T01] (  1,  1   )  simd64  ->  [rsp+0x08]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;  V02 arg1         [V02,T02] (  1,  1   )  simd64  ->  [rsp+0x48]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;  V03 arg2         [V03,T03] (  1,  1   )  simd64  ->  [rsp+0x88]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;  V04 arg3         [V04,T04] (  1,  1   )  simd64  ->  [rsp+0xC8]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;# V05 OutArgs      [V05    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace" <Empty>
 ;
 ; Lcl frame size = 0
 
 G_M43956_IG01:
 						;; size=0 bbWeight=1 PerfScore 0.00
 G_M43956_IG02:
        vmovups  zmm0, zmmword ptr [rsp+0x08]
-       vpshufb  zmm0, zmm0, zmmword ptr [rsp+0x88]
+       vmovups  zmm1, zmmword ptr [rsp+0x88]
+       vpshufb  zmm0, zmm0, zmm1
        vmovups  zmm1, zmmword ptr [rsp+0x48]
-       vpshufb  zmm1, zmm1, zmmword ptr [rsp+0xC8]
+       vmovups  zmm2, zmmword ptr [rsp+0xC8]
+       vpshufb  zmm1, zmm1, zmm2
        vpandd   zmm0, zmm1, zmm0
        vmovups  zmmword ptr [rdi], zmm0
        mov      rax, rdi
-						;; size=59 bbWeight=1 PerfScore 12.58
+						;; size=71 bbWeight=1 PerfScore 16.58
 G_M43956_IG03:
        vzeroupper 
        ret      
 						;; size=4 bbWeight=1 PerfScore 2.00
 
-; Total bytes of code 63, prolog size 0, PerfScore 14.58, instruction count 9, allocated bytes for code 63 (MethodHash=cdec544b) for method System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
+; Total bytes of code 75, prolog size 0, PerfScore 18.58, instruction count 11, allocated bytes for code 75 (MethodHash=cdec544b) for method System.Buffers.TeddyHelper:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
 ; ============================================================

6 (17.14 % of base) - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte]

 ; Assembly listing for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte] (FullOpts)
-; Emitting BLENDED_CODE for X64 with AVX512 - Unix
+; Emitting BLENDED_CODE for generic X64 + VEX + EVEX on Unix
 ; FullOpts code
 ; optimized code
 ; rsp based frame
 ; partially interruptible
 ; No PGO data
 ; Final local variable assignments
 ;
 ;  V00 RetBuf       [V00,T00] (  4,  4   )    long  ->  rdi         single-def
 ;  V01 arg0         [V01,T01] (  1,  1   )  simd64  ->  [rsp+0x08]  single-def <System.Runtime.Intrinsics.Vector512`1[byte]>
 ;  V02 arg1         [V02,T02] (  1,  1   )  simd64  ->  [rsp+0x48]  single-def <System.Runtime.Intrinsics.Vector512`1[byte]>
 ;# V03 OutArgs      [V03    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace" <Empty>
 ;
 ; Lcl frame size = 0
 
 G_M1718_IG01:
 						;; size=0 bbWeight=1 PerfScore 0.00
 G_M1718_IG02:
        vmovups  zmm0, zmmword ptr [rsp+0x08]
-       vpshufb  zmm0, zmm0, zmmword ptr [rsp+0x48]
+       vmovups  zmm1, zmmword ptr [rsp+0x48]
+       vpshufb  zmm0, zmm0, zmm1
        vmovups  zmmword ptr [rdi], zmm0
        mov      rax, rdi
-						;; size=31 bbWeight=1 PerfScore 7.25
+						;; size=37 bbWeight=1 PerfScore 9.25
 G_M1718_IG03:
        vzeroupper 
        ret      
 						;; size=4 bbWeight=1 PerfScore 2.00
 
-; Total bytes of code 35, prolog size 0, PerfScore 9.25, instruction count 6, allocated bytes for code 35 (MethodHash=11eef949) for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte] (FullOpts)
+; Total bytes of code 41, prolog size 0, PerfScore 11.25, instruction count 7, allocated bytes for code 41 (MethodHash=11eef949) for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[byte],System.Runtime.Intrinsics.Vector512`1[byte]):System.Runtime.Intrinsics.Vector512`1[byte] (FullOpts)
 ; ============================================================

6 (17.14 % of base) - System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte]

 ; Assembly listing for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
-; Emitting BLENDED_CODE for X64 with AVX512 - Unix
+; Emitting BLENDED_CODE for generic X64 + VEX + EVEX on Unix
 ; FullOpts code
 ; optimized code
 ; rsp based frame
 ; partially interruptible
 ; No PGO data
 ; Final local variable assignments
 ;
 ;  V00 RetBuf       [V00,T00] (  4,  4   )    long  ->  rdi         single-def
 ;  V01 arg0         [V01,T01] (  1,  1   )  simd64  ->  [rsp+0x08]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;  V02 arg1         [V02,T02] (  1,  1   )  simd64  ->  [rsp+0x48]  single-def <System.Runtime.Intrinsics.Vector512`1[ubyte]>
 ;# V03 OutArgs      [V03    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace" <Empty>
 ;
 ; Lcl frame size = 0
 
 G_M63011_IG01:
 						;; size=0 bbWeight=1 PerfScore 0.00
 G_M63011_IG02:
        vmovups  zmm0, zmmword ptr [rsp+0x08]
-       vpshufb  zmm0, zmm0, zmmword ptr [rsp+0x48]
+       vmovups  zmm1, zmmword ptr [rsp+0x48]
+       vpshufb  zmm0, zmm0, zmm1
        vmovups  zmmword ptr [rdi], zmm0
        mov      rax, rdi
-						;; size=31 bbWeight=1 PerfScore 7.25
+						;; size=37 bbWeight=1 PerfScore 9.25
 G_M63011_IG03:
        vzeroupper 
        ret      
 						;; size=4 bbWeight=1 PerfScore 2.00
 
-; Total bytes of code 35, prolog size 0, PerfScore 9.25, instruction count 6, allocated bytes for code 35 (MethodHash=116509dc) for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
+; Total bytes of code 41, prolog size 0, PerfScore 11.25, instruction count 7, allocated bytes for code 41 (MethodHash=116509dc) for method System.Runtime.Intrinsics.X86.Avx512BW:Shuffle(System.Runtime.Intrinsics.Vector512`1[ubyte],System.Runtime.Intrinsics.Vector512`1[ubyte]):System.Runtime.Intrinsics.Vector512`1[ubyte] (FullOpts)
 ; ============================================================

[MihuBot]

@MihaZupan