Dot product of shorts and floats: C# vs. C++

dp.asm

; C#, RyuJIT, release, x64
00007ffe`fc6808ec 4c63c8          movsxd  r9,eax						          ; loop: r9 = i
00007ffe`fc6808ef 4e0fbf4c4910    movsx   r9,word ptr [rcx+r9*2+10h]	; r9 = a[i]
00007ffe`fc6808f5 f3410f2ac9      cvtsi2ss xmm1,r9d						        ; xmm1 = (float)r9d
00007ffe`fc6808fa 4c63c8          movsxd  r9,eax                      ; r9 = i
00007ffe`fc6808fd f3420f594c8a10  mulss   xmm1,dword ptr [rdx+r9*4+10h] ; xmm1 *= b[i]
00007ffe`fc680904 f30f58c1        addss   xmm0,xmm1						        ; sum += xmm1
00007ffe`fc680908 ffc0            inc     eax							            ; ++i
00007ffe`fc68090a 443bc0          cmp     r8d,eax						          ; if i < a.Length
00007ffe`fc68090d 7fdd            jg      00007ffe`fc6808ec				    ;   goto loop
00007ffe`fc68090f eb2c            jmp     00007ffe`fc68093d				    ; goto done

; VC++ 2015, release, x64
00007fff`4f501090 0fbf43fc        movsx   eax,word ptr [rbx-4]				; eax = &a[rbx-2]
00007fff`4f501094 660f6ec8        movd    xmm1,eax							      ; xmm1 = eax
00007fff`4f501098 0fbf43fe        movsx   eax,word ptr [rbx-2]				; eax = &a[rbx-1]
00007fff`4f50109c 0f5bc9          cvtdq2ps xmm1,xmm1							    ; xmm1 = (float)xmm1
00007fff`4f50109f 660f6ec0        movd    xmm0,eax							      ; xmm0 = eax
00007fff`4f5010a3 f3410f594bf8    mulss   xmm1,dword ptr [r11-8]			; xmm1 *= &b[r11-2]
00007fff`4f5010a9 0fbf03          movsx   eax,word ptr [rbx]					; eax = &a[rbx]
00007fff`4f5010ac 0f5bc0          cvtdq2ps xmm0,xmm0							    ; xmm0 = (float)xmm0
00007fff`4f5010af f30f58ca        addss   xmm1,xmm2							      ; xmm1 += sum
00007fff`4f5010b3 f3410f5943fc    mulss   xmm0,dword ptr [r11-4]		  ; xmm0 *= &b[r11-1]
00007fff`4f5010b9 0f28d1          movaps  xmm2,xmm1							      ; sum = xmm1
00007fff`4f5010bc 660f6ec8        movd    xmm1,eax							      ; xmm1 = eax
00007fff`4f5010c0 0fbf4302        movsx   eax,word ptr [rbx+2]				; eax = &a[rbx+1]
00007fff`4f5010c4 f30f58d0        addss   xmm2,xmm0							      ; sum += xmm0
00007fff`4f5010c8 0f5bc9          cvtdq2ps xmm1,xmm1							    ; xmm1 = (float)xmm1
00007fff`4f5010cb 4883c308        add     rbx,8								        ; rbx += 8
00007fff`4f5010cf 660f6ec0        movd    xmm0,eax							      ; xmm0 = eax
00007fff`4f5010d3 f3410f590b      mulss   xmm1,dword ptr [r11]				; xmm1 *= &b[r11]
00007fff`4f5010d8 0f5bc0          cvtdq2ps xmm0,xmm0							    ; xmm0 = (float)xmm0
00007fff`4f5010db f30f58d1        addss   xmm2,xmm1							      ; sum += xmm1
00007fff`4f5010df f3410f594304    mulss   xmm0,dword ptr [r11+4]		  ; xmm0 *= &b[r11+1]
00007fff`4f5010e5 4983c310        add     r11,10h								      ; r11 += 16
00007fff`4f5010e9 f30f58d0        addss   xmm2,xmm0							      ; sum += xmm0
00007fff`4f5010ed 4983e801        sub     r8,1								        ; --r8
00007fff`4f5010f1 759d            jne     00007fff`4f501090					  ; if not zero goto loop