PDEP (R64, R64, M64) - Port Usage

With blocking instructions for port '1':

Code:

   0:	44 8d 15 80 00 00 00 	lea    r10d,[rip+0x80]        # 0x87
   7:	44 8d 1d 80 00 00 00 	lea    r11d,[rip+0x80]        # 0x8e
   e:	44 8d 25 80 00 00 00 	lea    r12d,[rip+0x80]        # 0x95
  15:	8d 1d 80 00 00 00    	lea    ebx,[rip+0x80]        # 0x9b
  1b:	8d 0d 80 00 00 00    	lea    ecx,[rip+0x80]        # 0xa1
  21:	8d 15 80 00 00 00    	lea    edx,[rip+0x80]        # 0xa7
  27:	44 8d 15 80 00 00 00 	lea    r10d,[rip+0x80]        # 0xae
  2e:	44 8d 1d 80 00 00 00 	lea    r11d,[rip+0x80]        # 0xb5
  35:	44 8d 25 80 00 00 00 	lea    r12d,[rip+0x80]        # 0xbc
  3c:	8d 1d 80 00 00 00    	lea    ebx,[rip+0x80]        # 0xc2
  42:	c4 42 b3 f5 06       	pdep   r8,r9,QWORD 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: 11.09
- Reference cycles: 8.32
- UOPS_PORT_1: 11.0

⇨ One μop that can only use port '1'

With blocking instructions for port '23A':

Code:

   0:	45 8b 56 40          	mov    r10d,DWORD PTR [r14+0x40]
   4:	45 8b 5e 44          	mov    r11d,DWORD PTR [r14+0x44]
   8:	45 8b 66 48          	mov    r12d,DWORD PTR [r14+0x48]
   c:	41 8b 5e 4c          	mov    ebx,DWORD PTR [r14+0x4c]
  10:	41 8b 4e 50          	mov    ecx,DWORD PTR [r14+0x50]
  14:	41 8b 56 54          	mov    edx,DWORD PTR [r14+0x54]
  18:	45 8b 56 40          	mov    r10d,DWORD PTR [r14+0x40]
  1c:	45 8b 5e 44          	mov    r11d,DWORD PTR [r14+0x44]
  20:	45 8b 66 48          	mov    r12d,DWORD PTR [r14+0x48]
  24:	41 8b 5e 4c          	mov    ebx,DWORD PTR [r14+0x4c]
  28:	c4 42 b3 f5 06       	pdep   r8,r9,QWORD 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: 3.76
- Reference cycles: 2.88
- UOPS_PORT_23A: 11.0

⇨ One μop that can only use port '23A'

With an indexed addressing mode

With blocking instructions for port '1':

Code:

   0:	44 8d 15 80 00 00 00 	lea    r10d,[rip+0x80]        # 0x87
   7:	44 8d 1d 80 00 00 00 	lea    r11d,[rip+0x80]        # 0x8e
   e:	44 8d 25 80 00 00 00 	lea    r12d,[rip+0x80]        # 0x95
  15:	8d 1d 80 00 00 00    	lea    ebx,[rip+0x80]        # 0x9b
  1b:	8d 0d 80 00 00 00    	lea    ecx,[rip+0x80]        # 0xa1
  21:	8d 15 80 00 00 00    	lea    edx,[rip+0x80]        # 0xa7
  27:	44 8d 15 80 00 00 00 	lea    r10d,[rip+0x80]        # 0xae
  2e:	44 8d 1d 80 00 00 00 	lea    r11d,[rip+0x80]        # 0xb5
  35:	44 8d 25 80 00 00 00 	lea    r12d,[rip+0x80]        # 0xbc
  3c:	8d 1d 80 00 00 00    	lea    ebx,[rip+0x80]        # 0xc2
  42:	c4 02 b3 f5 04 2e    	pdep   r8,r9,QWORD 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: 11.09
- Reference cycles: 8.16
- UOPS_PORT_1: 11.0

⇨ One μop that can only use port '1'

With blocking instructions for port '23A':

Code:

   0:	45 8b 56 40          	mov    r10d,DWORD PTR [r14+0x40]
   4:	45 8b 5e 44          	mov    r11d,DWORD PTR [r14+0x44]
   8:	45 8b 66 48          	mov    r12d,DWORD PTR [r14+0x48]
   c:	41 8b 5e 4c          	mov    ebx,DWORD PTR [r14+0x4c]
  10:	41 8b 4e 50          	mov    ecx,DWORD PTR [r14+0x50]
  14:	41 8b 56 54          	mov    edx,DWORD PTR [r14+0x54]
  18:	45 8b 56 40          	mov    r10d,DWORD PTR [r14+0x40]
  1c:	45 8b 5e 44          	mov    r11d,DWORD PTR [r14+0x44]
  20:	45 8b 66 48          	mov    r12d,DWORD PTR [r14+0x48]
  24:	41 8b 5e 4c          	mov    ebx,DWORD PTR [r14+0x4c]
  28:	c4 02 b3 f5 04 2e    	pdep   r8,r9,QWORD 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: 3.75
- Reference cycles: 2.88
- UOPS_PORT_23A: 11.0

⇨ One μop that can only use port '23A'