AVX for 32-bit Windows applications

[hutch--]

I only got this far with my old Core2 quad.

Code Select Expand


Calculating the sum of a float array in different ways.
That'll take a little while. Please be patient ...

Simple C implementation:
------------------------
sum1              = 8390656.00
Elapsed Time      = 16.44 Seconds

C implementation with 4 accumulators:
-------------------------------------
sum2              = 8390656.00
Elapsed Time      = 9.61 Seconds
Performance Boost = 171%

Assembly Language with 4 XMM accumulators:
------------------------------------------
sum3              = 8390656.00
Elapsed Time      = 1.42 Seconds
Performance Boost = 1156%

Your current CPU doesn't support the AVX instruction set.
The application terminates now.


[Gunther]

Thank you Hutch. It's clear that the Core2 doesn't support AVX. But the other timings are interesting.

Sinsi, you're using VMware and that allows AVX. Interesting point. Thank you, too.

Gunther

[Gunther]

I've installed yesterday VMware Player with Windows 7-64 as host and Windows 7-32 as guest. Here are the results for fsum.exe, which is under post #43:

Calculating the sum of a float array in different ways.
That'll take a little while. Please be patient ...

Simple C implementation:
------------------------
sum1              = 8390656.00
Elapsed Time      = 12.96 Seconds

C implementation with 4 accumulators:
-------------------------------------
sum2              = 8390656.00
Elapsed Time      = 6.46 Seconds
Performance Boost = 201%

Assembly Language with 4 XMM accumulators:
------------------------------------------
sum3              = 8390656.00
Elapsed Time      = 1.09 Seconds
Performance Boost = 1187%

Assembly Language with 4 YMM accumulators:
------------------------------------------
sum4              = 8390656.00
Elapsed Time      = 0.75 Seconds
Performance Boost = 1733%

Gunther

[habran]

I have found error for vsqrtps, vsqrtpd in JWasm

file instravx.h

line 30
was:

Code Select Expand


avxins (SQRTPD,   vsqrtpd,       P_AVX, VX_L ) /* L, s */
avxins (SQRTPS,   vsqrtps,       P_AVX, VX_L ) /* L, s */

change to:

Code Select Expand


avxins (SQRTPD,   vsqrtpd,       P_AVX, VX_L|VX_NND ) /* L, ns */
avxins (SQRTPS,   vsqrtps,       P_AVX, VX_L|VX_NND ) /* L, ns */


[Gunther]

Good catch, habran.  :t

Gunther

[habran]

Thanks mate