PDEP (R64, R64, M64) - Port Usage

With blocking instructions for port '1':

Code:

   0:	47 8d 94 6e 80 00 00 	lea    r10d,[r14+r13*2+0x80]
   7:	00 
   8:	47 8d 9c 6e 80 00 00 	lea    r11d,[r14+r13*2+0x80]
   f:	00 
  10:	47 8d a4 6e 80 00 00 	lea    r12d,[r14+r13*2+0x80]
  17:	00 
  18:	43 8d 9c 6e 80 00 00 	lea    ebx,[r14+r13*2+0x80]
  1f:	00 
  20:	43 8d 8c 6e 80 00 00 	lea    ecx,[r14+r13*2+0x80]
  27:	00 
  28:	43 8d 94 6e 80 00 00 	lea    edx,[r14+r13*2+0x80]
  2f:	00 
  30:	47 8d 94 6e 80 00 00 	lea    r10d,[r14+r13*2+0x80]
  37:	00 
  38:	47 8d 9c 6e 80 00 00 	lea    r11d,[r14+r13*2+0x80]
  3f:	00 
  40:	47 8d a4 6e 80 00 00 	lea    r12d,[r14+r13*2+0x80]
  47:	00 
  48:	43 8d 9c 6e 80 00 00 	lea    ebx,[r14+r13*2+0x80]
  4f:	00 
  50:	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: 11.06
- Reference cycles: 11.6
- UOPS_PORT_1: 11.0

⇨ One μop that can only use port '1'

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

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: 5.61
- Reference cycles: 5.9
- UOPS_PORT_2: 5.5
- UOPS_PORT_3: 5.5

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

With an indexed addressing mode

With blocking instructions for port '1':

Code:

   0:	47 8d 94 6e 80 00 00 	lea    r10d,[r14+r13*2+0x80]
   7:	00 
   8:	47 8d 9c 6e 80 00 00 	lea    r11d,[r14+r13*2+0x80]
   f:	00 
  10:	47 8d a4 6e 80 00 00 	lea    r12d,[r14+r13*2+0x80]
  17:	00 
  18:	43 8d 9c 6e 80 00 00 	lea    ebx,[r14+r13*2+0x80]
  1f:	00 
  20:	43 8d 8c 6e 80 00 00 	lea    ecx,[r14+r13*2+0x80]
  27:	00 
  28:	43 8d 94 6e 80 00 00 	lea    edx,[r14+r13*2+0x80]
  2f:	00 
  30:	47 8d 94 6e 80 00 00 	lea    r10d,[r14+r13*2+0x80]
  37:	00 
  38:	47 8d 9c 6e 80 00 00 	lea    r11d,[r14+r13*2+0x80]
  3f:	00 
  40:	47 8d a4 6e 80 00 00 	lea    r12d,[r14+r13*2+0x80]
  47:	00 
  48:	43 8d 9c 6e 80 00 00 	lea    ebx,[r14+r13*2+0x80]
  4f:	00 
  50:	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: 11.06
- Reference cycles: 11.63
- UOPS_PORT_1: 11.0

⇨ One μop that can only use port '1'

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

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: 5.61
- Reference cycles: 6.0
- UOPS_PORT_2: 5.5
- UOPS_PORT_3: 5.5

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