PAPI Count Issue

[ENash1]

Two questions:

1) I am using an AMD 64 Opteron processor and I am trying to count instructions. I've done all the proper setup code-wise but the instruction count I receive for the function I'm trying to count is double. The function uses assembly code:

# 4 4 4

.section .text
.align 2,0x90
.globl mxmul
mxmul:
movddup 0x0(%rsi), %xmm0
movdqu 0x0(%rdi), %xmm1
mulpd %xmm1, %xmm0
movddup 0x8(%rsi), %xmm2
movdqu 0x20(%rdi), %xmm3
mulpd %xmm3, %xmm2
addpd %xmm0, %xmm2
movddup 0x20(%rsi), %xmm4
mulpd %xmm1, %xmm4
movddup 0x28(%rsi), %xmm5
mulpd %xmm3, %xmm5
addpd %xmm4, %xmm5
movddup 0x40(%rsi), %xmm6
mulpd %xmm1, %xmm6
movddup 0x48(%rsi), %xmm7
mulpd %xmm3, %xmm7
addpd %xmm6, %xmm7
movddup 0x60(%rsi), %xmm8
mulpd %xmm1, %xmm8
movddup 0x68(%rsi), %xmm9
mulpd %xmm3, %xmm9
addpd %xmm8, %xmm9
movddup 0x10(%rsi), %xmm0
movdqu 0x40(%rdi), %xmm1
mulpd %xmm1, %xmm0
movddup 0x18(%rsi), %xmm10
movdqu 0x60(%rdi), %xmm3
mulpd %xmm3, %xmm10
addpd %xmm0, %xmm10
movddup 0x30(%rsi), %xmm4
mulpd %xmm1, %xmm4
movddup 0x38(%rsi), %xmm11
mulpd %xmm3, %xmm11
addpd %xmm4, %xmm11
movddup 0x50(%rsi), %xmm12
mulpd %xmm1, %xmm12
movddup 0x58(%rsi), %xmm13
mulpd %xmm3, %xmm13
addpd %xmm12, %xmm13
movddup 0x70(%rsi), %xmm14
mulpd %xmm1, %xmm14
movddup 0x78(%rsi), %xmm15
mulpd %xmm3, %xmm15
addpd %xmm14, %xmm15
addpd %xmm2, %xmm10
movapd %xmm10,0x0(%rdx)
addpd %xmm5, %xmm11
movapd %xmm11,0x20(%rdx)
addpd %xmm7, %xmm13
movapd %xmm13,0x40(%rdx)
addpd %xmm9, %xmm15
movapd %xmm15,0x60(%rdx)
movddup 0x0(%rsi), %xmm0
movdqu 0x10(%rdi), %xmm1
mulpd %xmm1, %xmm0
movddup 0x8(%rsi), %xmm2
movdqu 0x30(%rdi), %xmm3
mulpd %xmm3, %xmm2
addpd %xmm0, %xmm2
movddup 0x20(%rsi), %xmm4
mulpd %xmm1, %xmm4
movddup 0x28(%rsi), %xmm5
mulpd %xmm3, %xmm5
addpd %xmm4, %xmm5
movddup 0x40(%rsi), %xmm6
mulpd %xmm1, %xmm6
movddup 0x48(%rsi), %xmm7
mulpd %xmm3, %xmm7
addpd %xmm6, %xmm7
movddup 0x60(%rsi), %xmm8
mulpd %xmm1, %xmm8
movddup 0x68(%rsi), %xmm9
mulpd %xmm3, %xmm9
addpd %xmm8, %xmm9
movddup 0x10(%rsi), %xmm0
movdqu 0x50(%rdi), %xmm1
mulpd %xmm1, %xmm0
movddup 0x18(%rsi), %xmm10
movdqu 0x70(%rdi), %xmm3
mulpd %xmm3, %xmm10
addpd %xmm0, %xmm10
movddup 0x30(%rsi), %xmm4
mulpd %xmm1, %xmm4
movddup 0x38(%rsi), %xmm11
mulpd %xmm3, %xmm11
addpd %xmm4, %xmm11
movddup 0x50(%rsi), %xmm12
mulpd %xmm1, %xmm12
movddup 0x58(%rsi), %xmm13
mulpd %xmm3, %xmm13
addpd %xmm12, %xmm13
movddup 0x70(%rsi), %xmm14
mulpd %xmm1, %xmm14
movddup 0x78(%rsi), %xmm15
mulpd %xmm3, %xmm15
addpd %xmm14, %xmm15
addpd %xmm2, %xmm10
movapd %xmm10,0x10(%rdx)
addpd %xmm5, %xmm11
movapd %xmm11,0x30(%rdx)
addpd %xmm7, %xmm13
movapd %xmm13,0x50(%rdx)
addpd %xmm9, %xmm15
movapd %xmm15,0x70(%rdx)
retq
nop
.align 2, 0x90
.type mxmul,@function

Can you tell me why I get double?

2) Branching from the first question, I tried counting the Vector/SIMD instructions executed, hoping I'd get a more accurate number, and I receive 0 every time. Can you explain any of this?

[danterpstra]

What version of PAPI?
What event are you counting?
Is it exactly double the expected count?
Have you read the description of the native event(s) as provided by AMD and reported by uitls/papi_native_avail?
Same for the Vector/SIMD situation.
Doesn't look like your assembly has any explicit vector instructions...

[oddy1113]

PAPI is a portable hardware performance counter library developed by the Innovative Computing Labratory at the University of Tennessee .