VPERMILPS_EVEX (XMM, M32_1to4, I8) - Latency

Experiment 1

• Instruction: {evex} VPERMILPS XMM0, dword ptr [R14] {1to4}, 2
• Chain instructions: VMOVQ R12, XMM0;XOR R14, R12;XOR R14, R12;XOR R14, R12;XOR R14, R12
• Chain latency: ≥5
• Code:

     0:	62 d3 7d 18 04 06 02 	vpermilps xmm0,DWORD PTR [r14]{1to4},0x2
     7:	c4 c1 f9 7e c4       	vmovq  r12,xmm0
     c:	4d 31 e6             	xor    r14,r12
     f:	4d 31 e6             	xor    r14,r12
    12:	4d 31 e6             	xor    r14,r12
    15:	4d 31 e6             	xor    r14,r12
  

• 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

• Show nanoBench command
• Results:
  • Instructions retired: 6.0
  • Core cycles: 13.01
  • Reference cycles: 9.35
  • UOPS_EXECUTED.THREAD: 7.0

Experiment 1

• Instruction: {evex} VPERMILPS XMM0, dword ptr [R14+R13] {1to4}, 2
• Chain instructions: VMOVQ R12, XMM0;XOR R13, R12;XOR R13, R12;XOR R13, R12;XOR R13, R12
• Chain latency: ≥5
• Code:

     0:	62 93 7d 18 04 04 2e 	vpermilps xmm0,DWORD PTR [r14+r13*1]{1to4},0x2
     7:	02 
     8:	c4 c1 f9 7e c4       	vmovq  r12,xmm0
     d:	4d 31 e5             	xor    r13,r12
    10:	4d 31 e5             	xor    r13,r12
    13:	4d 31 e5             	xor    r13,r12
    16:	4d 31 e5             	xor    r13,r12
  

• 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;
  XOR R13, R13

• Show nanoBench command
• Results:
  • Instructions retired: 6.0
  • Core cycles: 13.0
  • Reference cycles: 9.2
  • UOPS_EXECUTED.THREAD: 7.0

Experiment 1

• Instruction: {evex} VPERMILPS XMM0, dword ptr [R14] {1to4}, 2
• Chain instructions: VSHUFPD XMM13, XMM0, XMM0, 0;VSHUFPD XMM13, XMM13, XMM13, 0;VSHUFPD XMM13, XMM13, XMM13, 0;VSHUFPD XMM13, XMM13, XMM13, 0;VSHUFPD XMM13, XMM13, XMM13, 0;VMOVSS [R14], XMM13
• Chain latency: ≥6
• Code:

     0:	62 d3 7d 18 04 06 02 	vpermilps xmm0,DWORD PTR [r14]{1to4},0x2
     7:	c5 79 c6 e8 00       	vshufpd xmm13,xmm0,xmm0,0x0
     c:	c4 41 11 c6 ed 00    	vshufpd xmm13,xmm13,xmm13,0x0
    12:	c4 41 11 c6 ed 00    	vshufpd xmm13,xmm13,xmm13,0x0
    18:	c4 41 11 c6 ed 00    	vshufpd xmm13,xmm13,xmm13,0x0
    1e:	c4 41 11 c6 ed 00    	vshufpd xmm13,xmm13,xmm13,0x0
    24:	c4 41 7a 11 2e       	vmovss DWORD PTR [r14],xmm13
  

• 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

• Show nanoBench command
• Results:
  • Instructions retired: 7.0
  • Core cycles: 11.0
  • Reference cycles: 8.13
  • UOPS_EXECUTED.THREAD: 9.0

Experiment 2

• Instruction: {evex} VPERMILPS XMM0, dword ptr [R14] {1to4}, 2
• Chain instructions: VPAND XMM13, XMM0, XMM0;VPAND XMM13, XMM13, XMM13;VPAND XMM13, XMM13, XMM13;VPAND XMM13, XMM13, XMM13;VPAND XMM13, XMM13, XMM13;VMOVD [R14], XMM13
• Chain latency: ≥6
• Code:

     0:	62 d3 7d 18 04 06 02 	vpermilps xmm0,DWORD PTR [r14]{1to4},0x2
     7:	c5 79 db e8          	vpand  xmm13,xmm0,xmm0
     b:	c4 41 11 db ed       	vpand  xmm13,xmm13,xmm13
    10:	c4 41 11 db ed       	vpand  xmm13,xmm13,xmm13
    15:	c4 41 11 db ed       	vpand  xmm13,xmm13,xmm13
    1a:	c4 41 11 db ed       	vpand  xmm13,xmm13,xmm13
    1f:	c4 41 79 7e 2e       	vmovd  DWORD PTR [r14],xmm13
  

• 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

• Show nanoBench command
• Results:
  • Instructions retired: 7.0
  • Core cycles: 11.11
  • Reference cycles: 8.28
  • UOPS_EXECUTED.THREAD: 9.0