Span<> tests
A repro for coreclr issue
Benchmark results
BenchmarkDotNet=v0.10.12, OS=Windows 10 Redstone 3 [1709, Fall Creators Update] (10.0.16299.192)
Intel Core i7-4600U CPU 2.10GHz (Haswell), 1 CPU, 4 logical cores and 2 physical cores
Frequency=2630627 Hz, Resolution=380.1375 ns, Timer=TSC
.NET Core SDK=2.1.300-preview2-008042
[Host] : .NET Core 2.1.0-preview2-26130-06 (Framework 4.6.26130.05), 64bit RyuJIT
Core : .NET Core 2.1.0-preview2-26130-06 (Framework 4.6.26130.05), 64bit RyuJIT
Job=Core Runtime=Core
| Method | Mean | Error | StdDev | Scaled | ScaledSD |
|---|---|---|---|---|---|
| ArrayBenchmark | 15.46 ns | 0.0847 ns | 0.0792 ns | 1.00 | 0.00 |
| UnsafeBenchmark | 10.71 ns | 0.1411 ns | 0.1250 ns | 0.69 | 0.01 |
| SpanBenchmark | 27.84 ns | 0.2352 ns | 0.2200 ns | 1.80 | 0.02 |
| SpanUnsafeBenchmark | 19.39 ns | 0.1233 ns | 0.1153 ns | 1.25 | 0.01 |
Assembly code:
The assembly code Span generates:
00007ffc`fe39eaf0 SpanTest.SpanBenchmarks.SpanBenchmark()
Span x1 = stackalloc int[16];
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
00007ffc`fe39eb22 b804000000 mov eax,4
00007ffc`fe39eb27 6a00 push 0
00007ffc`fe39eb29 6a00 push 0
00007ffc`fe39eb2b 48ffc8 dec rax
00007ffc`fe39eb2e 75f7 jne 00007ffc`fe39eb27
00007ffc`fe39eb30 488d0424 lea rax,[rsp]
00007ffc`fe39eb34 48896538 mov qword ptr [rbp+38h],rsp
00007ffc`fe39eb38 33d2 xor edx,edx
00007ffc`fe39eb3a 48895510 mov qword ptr [rbp+10h],rdx
00007ffc`fe39eb3e 48894510 mov qword ptr [rbp+10h],rax
00007ffc`fe39eb42 488b4510 mov rax,qword ptr [rbp+10h]
00007ffc`fe39eb46 488d5518 lea rdx,[rbp+18h]
00007ffc`fe39eb4a 488902 mov qword ptr [rdx],rax
00007ffc`fe39eb4d c7452010000000 mov dword ptr [rbp+20h],10h
var sum = 0;
^^^^^^^^^^^^
00007ffc`fe39eb54 c4e17a6f4518 vmovdqu xmm0,xmmword ptr [rbp+18h]
00007ffc`fe39eb5a c4e17a7f4528 vmovdqu xmmword ptr [rbp+28h],xmm0
00007ffc`fe39eb60 33c0 xor eax,eax
for (int i = 0; i < x1.Length; i++)
^^^^^^^^^
00007ffc`fe39eb62 33d2 xor edx,edx
00007ffc`fe39eb64 8b4d30 mov ecx,dword ptr [rbp+30h]
00007ffc`fe39eb67 85c9 test ecx,ecx
00007ffc`fe39eb69 7e11 jle 00007ffc`fe39eb7c
sum += x1[i];
^^^^^^^^^^^^^
00007ffc`fe39eb6b 4c8b4528 mov r8,qword ptr [rbp+28h]
00007ffc`fe39eb6f 4c63ca movsxd r9,edx
00007ffc`fe39eb72 43030488 add eax,dword ptr [r8+r9*4]
for (int i = 0; i < x1.Length; i++)
^^^
00007ffc`fe39eb76 ffc2 inc edx
00007ffc`fe39eb78 3bd1 cmp edx,ecx
00007ffc`fe39eb7a 7cef jl 00007ffc`fe39eb6b
return sum;
^^^^^^^^^^^
00007ffc`fe39eb7c 48b9c368fccd3f200000 mov rcx,203FCDFC68C3hCompared to unsafe stack allocation:
00007ffc`fe3aeb50 SpanTest.SpanBenchmarks.UnsafeBenchmark()
int* x1 = stackalloc int[16];
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
00007ffc`fe3aeb74 b804000000 mov eax,4
00007ffc`fe3aeb79 6a00 push 0
00007ffc`fe3aeb7b 6a00 push 0
00007ffc`fe3aeb7d 48ffc8 dec rax
00007ffc`fe3aeb80 75f7 jne 00007ffc`fe3aeb79
00007ffc`fe3aeb82 488d0424 lea rax,[rsp]
00007ffc`fe3aeb86 48896508 mov qword ptr [rbp+8],rsp
var sum = 0;
^^^^^^^^^^^^
00007ffc`fe3aeb8a 33d2 xor edx,edx
for (int i = 0; i < 16; i++)
^^^^^^^^^
00007ffc`fe3aeb8c 33c9 xor ecx,ecx
sum += x1[i];
^^^^^^^^^^^^^
00007ffc`fe3aeb8e 4c63c1 movsxd r8,ecx
00007ffc`fe3aeb91 42031480 add edx,dword ptr [rax+r8*4]
for (int i = 0; i < 16; i++)
^^^
00007ffc`fe3aeb95 ffc1 inc ecx
for (int i = 0; i < 16; i++)
^^^^^^
00007ffc`fe3aeb97 83f910 cmp ecx,10h
00007ffc`fe3aeb9a 7cf2 jl 00007ffc`fe3aeb8e
return sum;
^^^^^^^^^^^
00007ffc`fe3aeb9c 8bc2 mov eax,edx