JIT: Transform SELECT(relop, 1/0, 0/1) to relop

[jakobbotsch]

Fix#81479

Initially I generalized this to handle any power of two via LSH(cond, shift). Like GCC and Clang do here: https://godbolt.org/z/hxPf3Ka8e
However, it had worse performance than a cmov from my measurements (around 20% slower). It would use one less register in many situations though.

We could also do this directly in if-conversion, though then we wouldn't get potential future SELECTs created in different places.

Example:

[MethodImpl(MethodImplOptions.NoInlining)]publicboolIfElseCondition(boolinput){if(input){returnfalse;}else{returntrue;}}

Before:

G_M21872_IG02: ;; offset=0000H 84D2 testdl,dl7403je SHORT G_M21872_IG05 ;; size=4 bbWeight=1 PerfScore 1.25G_M21872_IG03: ;; offset=0004H 33C0 xoreax,eax ;; size=2 bbWeight=0.50 PerfScore 0.12G_M21872_IG04: ;; offset=0006H C3 ret ;; size=1 bbWeight=0.50 PerfScore 0.50G_M21872_IG05: ;; offset=0007H B801000000 moveax,1 ;; size=5 bbWeight=0.50 PerfScore 0.12G_M21872_IG06: ;; offset=000CH C3 ret ;; size=1 bbWeight=0.50 PerfScore 0.50

After #81267:

G_M21872_IG02: ;; offset=0000H B801000000 moveax,1 33C9 xorecx,ecx 84D2 testdl,dl 0F45C1 cmovne eax,ecx ;; size=12 bbWeight=1 PerfScore 1.00G_M21872_IG03: ;; offset=000CH C3 ret

After #81267 + this PR:

G_M21872_IG02: ;; offset=0000H 33C0 xoreax,eax 84D2 testdl,dl 0F94C0 sete al ;; size=7 bbWeight=1 PerfScore 1.50G_M21872_IG03: ;; offset=0007H C3 ret

[ghost]

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

Issue Details

---

Fix #81479

FWIW, initially I generalized this to handle any power of two via LSH(cond, shift). Like GCC and Clang do here: https://godbolt.org/z/hxPf3Ka8e
However, it had worse performance than a cmov from my measurements (around 20% slower). It would use one less register in many situations though.

We could also do this directly in if-conversion, though then we wouldn't get potential future SELECTs created in different places.

Author:	jakobbotsch
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[tannergooding]

This should help Roslyn/F# for this case: dotnet/roslyn#61483

[jakobbotsch]

DiffsNew diffs

cc @dotnet/jit-contrib PTAL @EgorBo

[EgorBo]

Why cast here is needed? if we know that we return either 0 or 1 - can we leave it TYP_INT (sort of legal after lowering)

[jakobbotsch]

Hmm, indeed it looks like compare nodes support being TYP_LONG typed in codegen, so I can just retype it. Will fix

[jakobbotsch]

New diffs, a bit better now, but I guess it was rare that we needed this cast

[cincuranet]

Possible regression in IfStatements.IfStatements in:

• [Perf] Windows/arm64: 13 Regressions on 2/9/2023 7:28:55 PM perf-autofiling-issues#13055
• [Perf] Linux/arm64: 3 Regressions on 2/9/2023 7:28:55 PM perf-autofiling-issues#13044

[AndyAyersMS]

Possible improvements:

• [Perf] Windows/arm64: 19 Improvements on 2/9/2023 7:28:55 PM perf-autofiling-issues#13063