VGATHERQPD (YMM, VSIB_YMM, YMM) - Port Usage

With blocking instructions for port '0':

Code:

   0:	41 0f 7e c0          	movd   r8d,mm0
   4:	41 0f 7e c1          	movd   r9d,mm0
   8:	41 0f 7e c2          	movd   r10d,mm0
   c:	41 0f 7e c3          	movd   r11d,mm0
  10:	41 0f 7e c4          	movd   r12d,mm0
  14:	0f 7e c3             	movd   ebx,mm0
  17:	0f 7e c1             	movd   ecx,mm0
  1a:	0f 7e c2             	movd   edx,mm0
  1d:	41 0f 7e c0          	movd   r8d,mm0
  21:	41 0f 7e c1          	movd   r9d,mm0
  25:	c4 82 f5 93 04 36    	vgatherqpd ymm0,QWORD PTR [r14+ymm14*1],ymm1

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
PXOR MM0, MM0;
VZEROALL;
MOV RAX, 0x4000000040000000;
MOV [R14+0], RAX;
MOV [R14+8], RAX;
MOV [R14+16], RAX;
MOV [R14+24], RAX;
VMOVUPD YMM0, [R14];
VXORPS YMM1, YMM1, YMM1;
VXORPS YMM14, YMM14, YMM14

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 11.26
- Reference cycles: 9.31
- UOPS_PORT_0: 11.0

⇨ One μop that can only use port '0'

With blocking instructions for port '5':

Code:

   0:	41 0f 6e c0          	movd   mm0,r8d
   4:	41 0f 6e c8          	movd   mm1,r8d
   8:	41 0f 6e d0          	movd   mm2,r8d
   c:	41 0f 6e d8          	movd   mm3,r8d
  10:	41 0f 6e e0          	movd   mm4,r8d
  14:	41 0f 6e e8          	movd   mm5,r8d
  18:	41 0f 6e f0          	movd   mm6,r8d
  1c:	41 0f 6e f8          	movd   mm7,r8d
  20:	41 0f 6e c0          	movd   mm0,r8d
  24:	41 0f 6e c8          	movd   mm1,r8d
  28:	c4 82 f5 93 04 36    	vgatherqpd ymm0,QWORD PTR [r14+ymm14*1],ymm1

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
MOV RAX, 0x4000000040000000;
MOV [R14+0], RAX;
MOV [R14+8], RAX;
MOV [R14+16], RAX;
MOV [R14+24], RAX;
VMOVUPD YMM0, [R14];
VXORPS YMM1, YMM1, YMM1;
VXORPS YMM14, YMM14, YMM14

Show nanoBench command
Results:
- Instructions retired: 11.0
- Core cycles: 11.09
- Reference cycles: 9.31
- UOPS_PORT_5: 11.0

⇨ One μop that can only use port '5'

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

Code:

   0:	c5 e1 58 d4          	vaddpd xmm2,xmm3,xmm4
   4:	c5 e1 58 ec          	vaddpd xmm5,xmm3,xmm4
   8:	c5 e1 58 f4          	vaddpd xmm6,xmm3,xmm4
   c:	c5 e1 58 fc          	vaddpd xmm7,xmm3,xmm4
  10:	c5 61 58 c4          	vaddpd xmm8,xmm3,xmm4
  14:	c5 61 58 cc          	vaddpd xmm9,xmm3,xmm4
  18:	c5 61 58 d4          	vaddpd xmm10,xmm3,xmm4
  1c:	c5 61 58 dc          	vaddpd xmm11,xmm3,xmm4
  20:	c5 61 58 e4          	vaddpd xmm12,xmm3,xmm4
  24:	c5 e1 58 d4          	vaddpd xmm2,xmm3,xmm4
  28:	c5 e1 58 ec          	vaddpd xmm5,xmm3,xmm4
  2c:	c5 e1 58 f4          	vaddpd xmm6,xmm3,xmm4
  30:	c5 e1 58 fc          	vaddpd xmm7,xmm3,xmm4
  34:	c5 61 58 c4          	vaddpd xmm8,xmm3,xmm4
  38:	c5 61 58 cc          	vaddpd xmm9,xmm3,xmm4
  3c:	c5 61 58 d4          	vaddpd xmm10,xmm3,xmm4
  40:	c4 82 f5 93 04 36    	vgatherqpd ymm0,QWORD PTR [r14+ymm14*1],ymm1

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
MOV RAX, 0x4000000040000000;
MOV [R14+0], RAX;
MOV [R14+8], RAX;
VMOVUPD XMM3, [R14];
VMOVUPD XMM4, [R14];
MOV [R14+16], RAX;
MOV [R14+24], RAX;
VMOVUPD YMM0, [R14];
VXORPS YMM1, YMM1, YMM1;
VXORPS YMM14, YMM14, YMM14

Show nanoBench command
Results:
- Instructions retired: 17.0
- Core cycles: 8.95
- Reference cycles: 7.87
- UOPS_PORT_1: 8.48
- UOPS_PORT_0: 8.52

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:	45 8b 56 48          	mov    r10d,DWORD PTR [r14+0x48]
  2c:	45 8b 5e 4c          	mov    r11d,DWORD PTR [r14+0x4c]
  30:	45 8b 66 50          	mov    r12d,DWORD PTR [r14+0x50]
  34:	41 8b 5e 54          	mov    ebx,DWORD PTR [r14+0x54]
  38:	41 8b 4e 58          	mov    ecx,DWORD PTR [r14+0x58]
  3c:	41 8b 56 5c          	mov    edx,DWORD PTR [r14+0x5c]
  40:	c4 82 f5 93 04 36    	vgatherqpd ymm0,QWORD PTR [r14+ymm14*1],ymm1

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
MOV RAX, 0x4000000040000000;
MOV [R14+0], RAX;
MOV [R14+8], RAX;
MOV [R14+16], RAX;
MOV [R14+24], RAX;
VMOVUPD YMM0, [R14];
VXORPS YMM1, YMM1, YMM1;
VXORPS YMM14, YMM14, YMM14

Show nanoBench command
Results:
- Instructions retired: 17.0
- Core cycles: 11.29
- Reference cycles: 10.01
- UOPS_PORT_2: 10.0
- UOPS_PORT_3: 10.0

⇨ 4 μops that can only use ports {'2', '3'}

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

Code:

   0:	c5 e1 55 d4          	vandnpd xmm2,xmm3,xmm4
   4:	c5 e1 55 ec          	vandnpd xmm5,xmm3,xmm4
   8:	c5 e1 55 f4          	vandnpd xmm6,xmm3,xmm4
   c:	c5 e1 55 fc          	vandnpd xmm7,xmm3,xmm4
  10:	c5 61 55 c4          	vandnpd xmm8,xmm3,xmm4
  14:	c5 61 55 cc          	vandnpd xmm9,xmm3,xmm4
  18:	c5 61 55 d4          	vandnpd xmm10,xmm3,xmm4
  1c:	c5 61 55 dc          	vandnpd xmm11,xmm3,xmm4
  20:	c5 61 55 e4          	vandnpd xmm12,xmm3,xmm4
  24:	c5 e1 55 d4          	vandnpd xmm2,xmm3,xmm4
  28:	c5 e1 55 ec          	vandnpd xmm5,xmm3,xmm4
  2c:	c5 e1 55 f4          	vandnpd xmm6,xmm3,xmm4
  30:	c5 e1 55 fc          	vandnpd xmm7,xmm3,xmm4
  34:	c5 61 55 c4          	vandnpd xmm8,xmm3,xmm4
  38:	c5 61 55 cc          	vandnpd xmm9,xmm3,xmm4
  3c:	c5 61 55 d4          	vandnpd xmm10,xmm3,xmm4
  40:	c5 61 55 dc          	vandnpd xmm11,xmm3,xmm4
  44:	c5 61 55 e4          	vandnpd xmm12,xmm3,xmm4
  48:	c5 e1 55 d4          	vandnpd xmm2,xmm3,xmm4
  4c:	c5 e1 55 ec          	vandnpd xmm5,xmm3,xmm4
  50:	c5 e1 55 f4          	vandnpd xmm6,xmm3,xmm4
  54:	c5 e1 55 fc          	vandnpd xmm7,xmm3,xmm4
  58:	c5 61 55 c4          	vandnpd xmm8,xmm3,xmm4
  5c:	c5 61 55 cc          	vandnpd xmm9,xmm3,xmm4
  60:	c4 82 f5 93 04 36    	vgatherqpd ymm0,QWORD PTR [r14+ymm14*1],ymm1

Init:

MOV R15, 10000;
L: VADDPS YMM0, YMM1, YMM1;
VADDPS YMM0, YMM1, YMM1;
DEC R15;
JNZ L;
VZEROALL;
VXORPS XMM3, XMM3, XMM3;
VXORPS XMM4, XMM4, XMM4;
MOV RAX, 0x4000000040000000;
MOV [R14+0], RAX;
MOV [R14+8], RAX;
MOV [R14+16], RAX;
MOV [R14+24], RAX;
VMOVUPD YMM0, [R14];
VXORPS YMM1, YMM1, YMM1;
VXORPS YMM14, YMM14, YMM14

Show nanoBench command
Results:
- Instructions retired: 25.0
- Core cycles: 10.66
- Reference cycles: 9.73
- UOPS_PORT_5: 8.63
- UOPS_PORT_1: 9.23
- UOPS_PORT_0: 9.13

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