[webkit-dev] SunSpider: GCC 4.0 vs. 4.2 results
Mark Rowe
mrowe at apple.com
Sat Jul 19 21:13:23 PDT 2008
Jon,
Thanks for these results. I hadn't seen figures comparing GCC 4.0 to
4.2 with SunSpider on PowerPC until now. The results on Intel
machines are slightly less promising however. GCC 4.2 looks to be
around a 10% regression over 4.0 on SunSpider on my MacBook Pro.
- Mark
On 2008-07-19, at 20:27, Jon Shier wrote:
> Devs:
> Thought this may interest someone. Overall 4.2 gives SunSpider a
> small speedup, but there are several tests in which it actually slows
> it down. These tests were run using the command line sunspider
> application after clean builds on Quad G5, using both the default
> settings and manually switching the projects to use GCC 4.2. Here is
> the result comparison:
>
> TEST COMPARISON FROM
> TO DETAILS
>
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
>
> ** TOTAL **: 1.026x as fast 2969.7ms +/- 0.2% 2893.6ms
> +/- 0.3% significant
>
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
>
> 3d: 1.076x as fast 399.9ms +/- 0.7% 371.7ms
> +/- 0.6% significant
> cube: 1.116x as fast 127.1ms +/- 0.5% 113.9ms
> +/- 0.6% significant
> morph: 1.086x as fast 145.7ms +/- 1.0% 134.2ms
> +/- 0.9% significant
> raytrace: 1.028x as fast 127.1ms +/- 0.9% 123.6ms
> +/- 0.6% significant
>
> access: 1.006x as fast 355.5ms +/- 0.3% 353.4ms
> +/- 0.2% significant
> binary-trees: 1.122x as fast 54.3ms +/- 0.6% 48.4ms
> +/- 0.8% significant
> fannkuch: *1.047x as slow* 98.7ms +/- 0.3% 103.3ms
> +/- 0.5% significant
> nbody: 1.008x as fast 174.1ms +/- 0.4% 172.7ms
> +/- 0.3% significant
> nsieve: *1.021x as slow* 28.4ms +/- 1.3% 29.0ms
> +/- 0.0% significant
>
> bitops: *1.087x as slow* 244.6ms +/- 0.3% 265.8ms
> +/- 0.9% significant
> 3bit-bits-in-byte: *1.138x as slow* 35.6ms +/- 1.4% 40.5ms
> +/- 0.9% significant
> bits-in-byte: *1.099x as slow* 47.3ms +/- 0.7% 52.0ms
> +/- 0.0% significant
> bitwise-and: *1.25x as slow* 76.2ms +/- 0.4% 95.3ms
> +/- 2.5% significant
> nsieve-bits: 1.096x as fast 85.5ms +/- 0.4% 78.0ms
> +/- 0.4% significant
>
> controlflow: *1.175x as slow* 22.9ms +/- 1.0% 26.9ms
> +/- 0.8% significant
> recursive: *1.175x as slow* 22.9ms +/- 1.0% 26.9ms
> +/- 0.8% significant
>
> crypto: 1.027x as fast 179.0ms +/- 0.6% 174.3ms
> +/- 0.7% significant
> aes: *1.071x as slow* 52.3ms +/- 0.7% 56.0ms
> +/- 0.6% significant
> md5: 1.082x as fast 64.6ms +/- 1.3% 59.7ms
> +/- 0.6% significant
> sha1: 1.060x as fast 62.1ms +/- 1.1% 58.6ms
> +/- 1.3% significant
>
> date: - 252.1ms +/- 0.8% 252.1ms
> +/- 0.7%
> format-tofte: 1.040x as fast 139.0ms +/- 1.1% 133.7ms
> +/- 0.6% significant
> format-xparb: *1.047x as slow* 113.1ms +/- 0.7% 118.4ms
> +/- 1.0% significant
>
> math: 1.063x as fast 353.1ms +/- 1.1% 332.1ms
> +/- 0.5% significant
> cordic: 1.057x as fast 114.6ms +/- 2.1% 108.4ms
> +/- 0.3% significant
> partial-sums: 1.030x as fast 157.0ms +/- 0.8% 152.5ms
> +/- 0.7% significant
> spectral-norm: 1.145x as fast 81.5ms +/- 1.2% 71.2ms
> +/- 0.8% significant
>
> regexp: 1.110x as fast 403.6ms +/- 0.2% 363.5ms
> +/- 0.2% significant
> dna: 1.110x as fast 403.6ms +/- 0.2% 363.5ms
> +/- 0.2% significant
>
> string: 1.007x as fast 759.0ms +/- 0.3% 753.8ms
> +/- 0.5% significant
> base64: *1.060x as slow* 100.5ms +/- 1.0% 106.5ms
> +/- 2.3% significant
> fasta: 1.102x as fast 152.7ms +/- 0.8% 138.6ms
> +/- 0.8% significant
> tagcloud: *1.011x as slow* 179.5ms +/- 0.7% 181.5ms
> +/- 0.6% significant
> unpack-code: *1.014x as slow* 198.2ms +/- 0.3% 201.0ms
> +/- 0.5% significant
> validate-input: 1.015x as fast 128.1ms +/- 1.0% 126.2ms
> +/- 0.9% significant
>
>
> 4.0 is the From column, so 4.2 is the To. I don't know how useful
> these results will be, just thought I'd share.
>
>
>
> Jon Shier
> _______________________________________________
> webkit-dev mailing list
> webkit-dev at lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2413 bytes
Desc: not available
Url : http://lists.webkit.org/pipermail/webkit-dev/attachments/20080719/ab7fe029/attachment.bin
More information about the webkit-dev
mailing list