Fix genPutArgStkFieldList for SIMD12 #88920

EgorBo · 2023-07-14T16:48:03Z

ghost · 2023-07-14T16:48:16Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

Fixes #88842
and, presumably, fixes #88842

Author:	EgorBo
Assignees:	-
Labels:	`area-CodeGen-coreclr`
Milestone:	-

filipnavara · 2023-07-14T16:57:53Z

Fixes #88842 and, presumably, fixes #88842

Is this a joke on a.SequenceEqual(a) == false (aka issue #65292)?

Seriously though, I assume you meant to link two different issues.

EgorBo · 2023-07-14T17:01:00Z

Fixes #88842 and, presumably, fixes #88842

Is this a joke on a.SequenceEqual(a) == false (aka issue #65292)?

Seriously though, I assume you meant to link two different issues.

took me a while to understand what you mean 😄 yeah, fixed

src/coreclr/jit/codegenlinear.cpp

jakobbotsch · 2023-07-14T20:47:24Z

I verified that this fixes #88842 as well, diff:

@@ -674,27 +674,31 @@ G_M13415_IG07:
 						;; size=10 bbWeight=1595.95 PerfScore 1462.95
 G_M13415_IG08:
        test     edx, edx
-       je       SHORT G_M13415_IG11
-						;; size=4 bbWeight=99.75 PerfScore 124.68
+       je       G_M13415_IG11
+						;; size=8 bbWeight=99.75 PerfScore 124.68
 G_M13415_IG09:
        mov      gword ptr [rbp-78H], rax
        mov      dword ptr [rbp-70H], edi
        vmovaps  xmmword ptr [rbp+C0H], xmm4
        vmovups  xmmword ptr [rbp+CCH], xmm2
        cmp      byte  ptr [rbx], bl
-       vmovups  xmmword ptr [rsp], xmm4
-       vmovups  xmmword ptr [rsp+0CH], xmm2
+       vmovsd   qword ptr [rsp], xmm4
+       vextractps dword ptr [rsp+08H], xmm4, 2
+       vmovsd   qword ptr [rsp+0CH], xmm2
+       vextractps dword ptr [rsp+14H], xmm2, 2
        mov      rdi, rbx
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:FindValue(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid):byref:this]
        test     rax, rax
        je       SHORT G_M13415_IG12
-						;; size=50 bbWeight=36.60 PerfScore 494.10
+						;; size=66 bbWeight=36.60 PerfScore 713.70
 G_M13415_IG10:
        mov      rsi, qword ptr [rax]
        vmovaps  xmm4, xmmword ptr [rbp+C0H]
-       vmovups  xmmword ptr [rsp], xmm4
+       vmovsd   qword ptr [rsp], xmm4
+       vextractps dword ptr [rsp+08H], xmm4, 2
        vmovups  xmm2, xmmword ptr [rbp+CCH]
-       vmovups  xmmword ptr [rsp+0CH], xmm2
+       vmovsd   qword ptr [rsp+0CH], xmm2
+       vextractps dword ptr [rsp+14H], xmm2, 2
        sub      rsi, r13
        mov      rdi, rbx
        mov      edx, 1
@@ -705,7 +709,7 @@ G_M13415_IG10:
        vmovups  xmm0, xmmword ptr [rbp-58H]
        vmovups  xmm1, xmmword ptr [rbp-68H]
        jmp      SHORT G_M13415_IG13
-						;; size=69 bbWeight=585.60 PerfScore 14493.55
+						;; size=85 bbWeight=585.60 PerfScore 18007.13
 G_M13415_IG11:
        mov      ecx, dword ptr [rbp-6CH]
        vmovups  xmm0, xmmword ptr [rbp-58H]
@@ -714,9 +718,11 @@ G_M13415_IG11:
 						;; size=15 bbWeight=49.87 PerfScore 448.86
 G_M13415_IG12:
        vmovaps  xmm4, xmmword ptr [rbp+C0H]
-       vmovups  xmmword ptr [rsp], xmm4
+       vmovsd   qword ptr [rsp], xmm4
+       vextractps dword ptr [rsp+08H], xmm4, 2
        vmovups  xmm2, xmmword ptr [rbp+CCH]
-       vmovups  xmmword ptr [rsp+0CH], xmm2
+       vmovsd   qword ptr [rsp+0CH], xmm2
+       vextractps dword ptr [rsp+14H], xmm2, 2
        mov      rsi, r13
        neg      rsi
        mov      rdi, rbx
@@ -727,7 +733,7 @@ G_M13415_IG12:
        mov      edi, dword ptr [rbp-70H]
        vmovups  xmm0, xmmword ptr [rbp-58H]
        vmovups  xmm1, xmmword ptr [rbp-68H]
-						;; size=67 bbWeight=585.60 PerfScore 12297.56
+						;; size=83 bbWeight=585.60 PerfScore 15811.14
 G_M13415_IG13:
        mov      r13d, 1
        mov      dword ptr [rbp-6CH], ecx
@@ -769,13 +775,15 @@ G_M13415_IG18:
        vmovups  xmmword ptr [rbp-38H], xmm0
        vmovups  xmmword ptr [rbp-48H], xmm1
        cmp      byte  ptr [rbx], bl
-       vmovups  xmmword ptr [rsp], xmm0
-       vmovups  xmmword ptr [rsp+0CH], xmm1
+       vmovsd   qword ptr [rsp], xmm0
+       vextractps dword ptr [rsp+08H], xmm0, 2
+       vmovsd   qword ptr [rsp+0CH], xmm1
+       vextractps dword ptr [rsp+14H], xmm1, 2
        mov      rdi, rbx
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:FindValue(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid):byref:this]
        test     rax, rax
        je       G_M13415_IG25
-						;; size=41 bbWeight=0.19 PerfScore 2.15
+						;; size=57 bbWeight=0.19 PerfScore 3.28
 G_M13415_IG19:
        mov      rsi, qword ptr [rax]
        jmp      G_M13415_IG24
@@ -801,47 +809,55 @@ G_M13415_IG22:
        vmovups  xmmword ptr [rbp+A8H], xmm2
        vmovups  xmmword ptr [rbp+B4H], xmm3
        mov      qword ptr [rbp+A0H], rdx
-       vmovups  xmmword ptr [rsp], xmm2
-       vmovups  xmmword ptr [rsp+0CH], xmm3
+       vmovsd   qword ptr [rsp], xmm2
+       vextractps dword ptr [rsp+08H], xmm2, 2
+       vmovsd   qword ptr [rsp+0CH], xmm3
+       vextractps dword ptr [rsp+14H], xmm3, 2
        mov      rdi, r15
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:FindValue(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid):byref:this]
        test     rax, rax
        je       G_M13415_IG33
-						;; size=52 bbWeight=20.63 PerfScore 196.03
+						;; size=68 bbWeight=20.63 PerfScore 319.83
 G_M13415_IG23:
        mov      rsi, qword ptr [rax]
        vmovups  xmm2, xmmword ptr [rbp+A8H]
-       vmovups  xmmword ptr [rsp], xmm2
+       vmovsd   qword ptr [rsp], xmm2
+       vextractps dword ptr [rsp+08H], xmm2, 2
        vmovups  xmm3, xmmword ptr [rbp+B4H]
-       vmovups  xmmword ptr [rsp+0CH], xmm3
+       vmovsd   qword ptr [rsp+0CH], xmm3
+       vextractps dword ptr [rsp+14H], xmm3, 2
        add      rsi, qword ptr [rbp+A0H]
        mov      rdi, r15
        mov      edx, 1
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:TryInsert(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long,ubyte):bool:this]
        mov      eax, dword ptr [rbp-70H]
        jmp      G_M13415_IG34
-						;; size=59 bbWeight=330.15 PerfScore 6107.73
+						;; size=75 bbWeight=330.15 PerfScore 8088.61
 G_M13415_IG24:
        vmovups  xmm1, xmmword ptr [rbp-38H]
-       vmovups  xmmword ptr [rsp], xmm1
+       vmovsd   qword ptr [rsp], xmm1
+       vextractps dword ptr [rsp+08H], xmm1, 2
        vmovups  xmm2, xmmword ptr [rbp-48H]
-       vmovups  xmmword ptr [rsp+0CH], xmm2
+       vmovsd   qword ptr [rsp+0CH], xmm2
+       vextractps dword ptr [rsp+14H], xmm2, 2
        inc      rsi
        mov      rdi, rbx
        mov      edx, 1
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:TryInsert(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long,ubyte):bool:this]
        jmp      SHORT G_M13415_IG26
-						;; size=40 bbWeight=0.19 PerfScore 2.57
+						;; size=56 bbWeight=0.19 PerfScore 3.70
 G_M13415_IG25:
        vmovups  xmm1, xmmword ptr [rbp-38H]
-       vmovups  xmmword ptr [rsp], xmm1
+       vmovsd   qword ptr [rsp], xmm1
+       vextractps dword ptr [rsp+08H], xmm1, 2
        vmovups  xmm2, xmmword ptr [rbp-48H]
-       vmovups  xmmword ptr [rsp+0CH], xmm2
+       vmovsd   qword ptr [rsp+0CH], xmm2
+       vextractps dword ptr [rsp+14H], xmm2, 2
        mov      rdi, rbx
        mov      esi, 1
        mov      edx, 1
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:TryInsert(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long,ubyte):bool:this]
-						;; size=40 bbWeight=0.75 PerfScore 8.80
+						;; size=56 bbWeight=0.75 PerfScore 13.29
 G_M13415_IG26:
        mov      edi, dword ptr [rbx+38H]
        sub      edi, dword ptr [rbx+40H]
@@ -893,15 +909,17 @@ G_M13415_IG32:
 						;; size=22 bbWeight=0.23 PerfScore 1.33
 G_M13415_IG33:
        vmovups  xmm2, xmmword ptr [rbp+A8H]
-       vmovups  xmmword ptr [rsp], xmm2
+       vmovsd   qword ptr [rsp], xmm2
+       vextractps dword ptr [rsp+08H], xmm2, 2
        vmovups  xmm3, xmmword ptr [rbp+B4H]
-       vmovups  xmmword ptr [rsp+0CH], xmm3
+       vmovsd   qword ptr [rsp+0CH], xmm3
+       vextractps dword ptr [rsp+14H], xmm3, 2
        mov      rdi, r15
        mov      rsi, qword ptr [rbp+A0H]
        mov      edx, 1
        call     [System.Collections.Generic.Dictionary`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long]:TryInsert(MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,long,ubyte):bool:this]
        mov      eax, dword ptr [rbp-70H]
-						;; size=51 bbWeight=330.15 PerfScore 4456.99
+						;; size=67 bbWeight=330.15 PerfScore 6437.88
 G_M13415_IG34:
        cmp      r14d, dword ptr [rbx+44H]
        jne      G_M13415_IG47
@@ -1031,6 +1049,11 @@ G_M13415_IG48:
        int3     
 						;; size=6 bbWeight=0 PerfScore 0.00
 
-; Total bytes of code 1573, prolog size 84, PerfScore 71319.51, instruction count 350, allocated bytes for code 1573 (MethodHash=e316cb98) for method MartinCostello.AdventOfCode.Puzzles.Y2021.Day22:<Reboot>g__Reboot|8_1(System.Collections.Generic.List`1[System.ValueTuple`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,bool]]):long (Tier1-OSR)
+; Total bytes of code 1721, prolog size 84, PerfScore 82673.40, instruction count 368, allocated bytes for code 1721 (MethodHash=e316cb98) for method MartinCostello.AdventOfCode.Puzzles.Y2021.Day22:<Reboot>g__Reboot|8_1(System.Collections.Generic.List`1[System.ValueTuple`2[MartinCostello.AdventOfCode.Puzzles.Y2021.Day22+Cuboid,bool]]):long (Tier1-OSR)

src/coreclr/jit/emitxarch.h

EgorBo · 2023-07-15T19:26:22Z

SPMI is failing everywhere, real diffs were +24 few cases, e.g.:

@@ -163,7 +163,9 @@ G_M13885_IG02:        ; bbWeight=1, gcrefRegs=380000 {x19 x20 x21}, byrefRegs=00
             fmul    v0.4s, v1.4s, v0.s[0]
             fsub    v0.4s, v16.4s, v0.4s
             str     q0, [fp, #0x20]	// [V50 tmp41]
-            str     q0, [sp]	// [V09 OutArgs]
+            str     d0, [sp]	// [V09 OutArgs]
+            mov     w1, v0.s[2]
+            str     w1, [sp, #0x08]	// [V09 OutArgs+0x08]
             ldp     s0, s1, [fp, #0xA0]	// [V24 tmp15], [V24 tmp15+0x04]
             ldr     s2, [fp, #0xA8]	// [V24 tmp15+0x08]
             ldp     s3, s4, [fp, #0xC0]	// [V06 loc2], [V06 loc2+0x04]
@@ -189,7 +191,7 @@ G_M13885_IG02:        ; bbWeight=1, gcrefRegs=380000 {x19 x20 x21}, byrefRegs=00
             cmp     w22, #5
             mov     v8.d[1], v9.d[0]
             blt     G_M13885_IG05
-						;; size=248 bbWeight=1 PerfScore 109.00
+						;; size=256 bbWeight=1 PerfScore 111.00

Because previously we only ran that logic on osx-arm64

EgorBo added 2 commits July 14, 2023 18:41

Fix invalid simd12 store

41bec7c

remove ISMETHOD

4bf8aed

ghost added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jul 14, 2023

ghost assigned EgorBo Jul 14, 2023

EgorBo marked this pull request as ready for review July 14, 2023 19:34

jakobbotsch reviewed Jul 14, 2023

View reviewed changes

src/coreclr/jit/codegenlinear.cpp Outdated Show resolved Hide resolved

EgorBo added 2 commits July 14, 2023 22:39

Add a test

372912e

Merge branch 'main' of github.com:dotnet/runtime into fix-simd12-store

c4284ae

EgorBo added 2 commits July 15, 2023 12:28

Unify paths

c3ac997

Update Runtime_79118.cs

5bc6bab

jakobbotsch reviewed Jul 15, 2023

View reviewed changes

src/coreclr/jit/emitxarch.h Outdated Show resolved Hide resolved

jakobbotsch approved these changes Jul 15, 2023

View reviewed changes

build-analysis bot mentioned this pull request Jul 15, 2023

Timeout in System.Net.Quic.Functional.Tests #86019

Closed

EgorBo added 3 commits July 15, 2023 16:26

Address feedback

c5fd82e

Merge branch 'main' of github.com:dotnet/runtime into fix-simd12-store

5d37c9c

Update Runtime_79118.cs

0c3ad38

This was referenced Jul 15, 2023

Could not load file or assembly 'Microsoft.CodeAnalysis.NetAnalyzers #84995

Closed

Unable to load Analyzer assembly .../Microsoft.CodeAnalysis.Analyzers.dll : Not a valid assembly #85082

Closed

EgorBo merged commit b898a54 into dotnet:main Jul 15, 2023

EgorBo deleted the fix-simd12-store branch July 15, 2023 19:26

ghost locked as resolved and limited conversation to collaborators Aug 15, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix genPutArgStkFieldList for SIMD12 #88920

Fix genPutArgStkFieldList for SIMD12 #88920

Uh oh!

EgorBo commented Jul 14, 2023 •

edited

Loading

Uh oh!

ghost commented Jul 14, 2023

Uh oh!

filipnavara commented Jul 14, 2023 •

edited

Loading

Uh oh!

EgorBo commented Jul 14, 2023

Uh oh!

Uh oh!

jakobbotsch commented Jul 14, 2023

Uh oh!

Uh oh!

EgorBo commented Jul 15, 2023 •

edited

Loading

Uh oh!

Uh oh!

Fix genPutArgStkFieldList for SIMD12 #88920

Fix genPutArgStkFieldList for SIMD12 #88920

Uh oh!

Conversation

EgorBo commented Jul 14, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ghost commented Jul 14, 2023

Uh oh!

filipnavara commented Jul 14, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

EgorBo commented Jul 14, 2023

Uh oh!

Uh oh!

jakobbotsch commented Jul 14, 2023

Uh oh!

Uh oh!

EgorBo commented Jul 15, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

EgorBo commented Jul 14, 2023 •

edited

Loading

filipnavara commented Jul 14, 2023 •

edited

Loading

EgorBo commented Jul 15, 2023 •

edited

Loading