VMOVNTDQA (YMM, M256) - Port Usage

With blocking instructions for ports {'2', '3'}:

Code:

   0:	45 8b 46 40          	mov    r8d,DWORD PTR [r14+0x40]
   4:	45 8b 4e 44          	mov    r9d,DWORD PTR [r14+0x44]
   8:	45 8b 56 48          	mov    r10d,DWORD PTR [r14+0x48]
   c:	45 8b 5e 4c          	mov    r11d,DWORD PTR [r14+0x4c]
  10:	45 8b 66 50          	mov    r12d,DWORD PTR [r14+0x50]
  14:	41 8b 5e 54          	mov    ebx,DWORD PTR [r14+0x54]
  18:	41 8b 4e 58          	mov    ecx,DWORD PTR [r14+0x58]
  1c:	41 8b 56 5c          	mov    edx,DWORD PTR [r14+0x5c]
  20:	45 8b 46 40          	mov    r8d,DWORD PTR [r14+0x40]
  24:	45 8b 4e 44          	mov    r9d,DWORD PTR [r14+0x44]
  28:	c4 c2 7d 2a 06       	vmovntdqa ymm0,YMMWORD PTR [r14]

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 5.6
- Reference cycles: 3.83
- UOPS_PORT_3: 5.5
- UOPS_PORT_2: 5.5

⇨ One μop that can only use ports {'2', '3'}

With blocking instructions for ports {'0', '1', '5'}:

Code:

   0:	c5 e9 55 cb          	vandnpd xmm1,xmm2,xmm3
   4:	c5 e9 55 e3          	vandnpd xmm4,xmm2,xmm3
   8:	c5 e9 55 eb          	vandnpd xmm5,xmm2,xmm3
   c:	c5 e9 55 f3          	vandnpd xmm6,xmm2,xmm3
  10:	c5 e9 55 fb          	vandnpd xmm7,xmm2,xmm3
  14:	c5 69 55 c3          	vandnpd xmm8,xmm2,xmm3
  18:	c5 69 55 cb          	vandnpd xmm9,xmm2,xmm3
  1c:	c5 69 55 d3          	vandnpd xmm10,xmm2,xmm3
  20:	c5 69 55 db          	vandnpd xmm11,xmm2,xmm3
  24:	c5 69 55 e3          	vandnpd xmm12,xmm2,xmm3
  28:	c4 c2 7d 2a 06       	vmovntdqa ymm0,YMMWORD PTR [r14]

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
VXORPS XMM2, XMM2, XMM2;
VXORPS XMM3, XMM3, XMM3

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 3.78
- Reference cycles: 2.76
- UOPS_PORT_1: 3.67
- UOPS_PORT_0: 3.66
- UOPS_PORT_5: 3.67

⇨ One μop that can only use ports {'0', '1', '5'}

With an indexed addressing mode

With blocking instructions for ports {'2', '3'}:

Code:

   0:	45 8b 46 40          	mov    r8d,DWORD PTR [r14+0x40]
   4:	45 8b 4e 44          	mov    r9d,DWORD PTR [r14+0x44]
   8:	45 8b 56 48          	mov    r10d,DWORD PTR [r14+0x48]
   c:	45 8b 5e 4c          	mov    r11d,DWORD PTR [r14+0x4c]
  10:	45 8b 66 50          	mov    r12d,DWORD PTR [r14+0x50]
  14:	41 8b 5e 54          	mov    ebx,DWORD PTR [r14+0x54]
  18:	41 8b 4e 58          	mov    ecx,DWORD PTR [r14+0x58]
  1c:	41 8b 56 5c          	mov    edx,DWORD PTR [r14+0x5c]
  20:	45 8b 46 40          	mov    r8d,DWORD PTR [r14+0x40]
  24:	45 8b 4e 44          	mov    r9d,DWORD PTR [r14+0x44]
  28:	c4 82 7d 2a 04 2e    	vmovntdqa ymm0,YMMWORD PTR [r14+r13*1]

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
XOR R13, R13

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 5.6
- Reference cycles: 3.99
- UOPS_PORT_3: 5.5
- UOPS_PORT_2: 5.5

⇨ One μop that can only use ports {'2', '3'}

With blocking instructions for ports {'0', '1', '5'}:

Code:

   0:	c5 e9 55 cb          	vandnpd xmm1,xmm2,xmm3
   4:	c5 e9 55 e3          	vandnpd xmm4,xmm2,xmm3
   8:	c5 e9 55 eb          	vandnpd xmm5,xmm2,xmm3
   c:	c5 e9 55 f3          	vandnpd xmm6,xmm2,xmm3
  10:	c5 e9 55 fb          	vandnpd xmm7,xmm2,xmm3
  14:	c5 69 55 c3          	vandnpd xmm8,xmm2,xmm3
  18:	c5 69 55 cb          	vandnpd xmm9,xmm2,xmm3
  1c:	c5 69 55 d3          	vandnpd xmm10,xmm2,xmm3
  20:	c5 69 55 db          	vandnpd xmm11,xmm2,xmm3
  24:	c5 69 55 e3          	vandnpd xmm12,xmm2,xmm3
  28:	c4 82 7d 2a 04 2e    	vmovntdqa ymm0,YMMWORD PTR [r14+r13*1]

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
VXORPS XMM2, XMM2, XMM2;
VXORPS XMM3, XMM3, XMM3;
XOR R13, R13

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 3.78
- Reference cycles: 2.76
- UOPS_PORT_1: 3.67
- UOPS_PORT_0: 3.66
- UOPS_PORT_5: 3.67

⇨ One μop that can only use ports {'0', '1', '5'}