JIT: Optimize *x = dblCns to *x = intCns

[EgorBo]

Just a small optimization for #52286 (comment) comment.

void Test(ref float x) => x = 0;

Current codegen:

; Method Program:Test(byref) G_M20809_IG01:  vzeroupper  ;; bbWeight=1 PerfScore 1.00 G_M20809_IG02:  vxorps xmm0, xmm0  vmovss dword ptr [rcx], xmm0 ;; bbWeight=1 PerfScore 0.83 G_M20809_IG03:  ret  ;; bbWeight=1 PerfScore 1.00 ; Total bytes of code: 12

New codegen:

; Method Program:Test(byref) G_M20809_IG02:  mov dword ptr [rcx], 0 ;; bbWeight=1 PerfScore 1.00 G_M20809_IG03:  ret  ;; bbWeight=1 PerfScore 1.00 ; Total bytes of code: 7

SuperPMI results:

asm.aspnet.run.windows.x64.checked.2 -86 (-0.21% of base) asm.benchmarks.run.windows.x64.checked -684 (-1.02% of base) asm.libraries.crossgen.windows.x64.checked.3 -641 (-0.78% of base) asm.libraries.crossgen2.windows.x64.checked.2 -600 (-0.77% of base) asm.libraries.pmi.windows.x64.checked -967 (-0.99% of base) asm.tests_libraries.pmi.windows.x64.checked -25294 (-0.72% of base) asm.tests.pmi.windows.x64.checked.1 -71306 (-0.35% of base) 

/cc @dotnet/jit-contrib @SingleAccretion

[tannergooding]

Changes LGTM.

[EgorBo]

@dotnet/jit-contrib @tannergooding anyone wants to approve? 🙂