79739 2012-02-27 20:23:24 -0800 Use IPP malloc for AudioArray 2012-07-10 11:54:26 -0700 1 1 1 Unclassified WebKit Web Audio 528+ (Nightly build) Unspecified Unspecified RESOLVED WONTFIX P2 Normal --- 0 xingnan.wang webkit-unassigned crogers jchaffraix oldest_to_newest 566323 0 xingnan.wang 2012-02-27 20:23:24 -0800 IPP library provides the ippmalloc which can malloc memory aligned with 32-byte. As AudioArray needs 16-byte alignment we can use ippmalloc instead. 566328 1 129177 xingnan.wang 2012-02-27 20:30:44 -0800 Created attachment 129177 Patch 566329 2 xingnan.wang 2012-02-27 20:31:41 -0800 (In reply to comment #1) > Created an attachment (id=129177) [details] > Patch Uploaded the patch, thanks. 566331 3 crogers 2012-02-27 20:32:00 -0800 Will 32-byte alignment get us any advantage over the 16-byte alignment? 566335 4 xingnan.wang 2012-02-27 20:42:13 -0800 (In reply to comment #3) > Will 32-byte alignment get us any advantage over the 16-byte alignment? I think 32-byte and 16-byte alignment are not too much different here, I have to use it because the IPP only provides 32-byte aligned ippmalloc function. The advantage we can get is the aligned memory is more easily allocated than original way. 567314 5 crogers 2012-02-28 18:37:51 -0800 (In reply to comment #4) > (In reply to comment #3) > > Will 32-byte alignment get us any advantage over the 16-byte alignment? > > I think 32-byte and 16-byte alignment are not too much different here, I have to use it because the IPP only provides 32-byte aligned ippmalloc function. The advantage we can get is the aligned memory is more easily allocated than original way. I'm not sure what you mean by "more easily allocated". I understand that the run-time code path in the IPP case would be slightly simpler in AudioArray.h, but is it actually a measurable performance win? The disadvantage of having this special-casing is that the code is somewhat harder to read with this patch. And this particular file is already somewhat tweaky even as it is right now, so it would be great to avoid adding additional complexity (harder to read through the code) unless there's a pretty good win. 606062 6 129177 jchaffraix 2012-04-19 15:59:55 -0700 Comment on attachment 129177 Patch I agree with Chris here. If the switch leads to performance benefits then we should consider this change but in the absence of hard evidence or more substancial explanations, this is a maintenance burden on the project with little benefits. 129177 2012-02-27 20:30:44 -0800 2012-04-19 15:59:55 -0700 Patch bug-79739-20120228123332.patch text/plain 4302 xingnan.wang U3VidmVyc2lvbiBSZXZpc2lvbjogMTA5MDYyCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggNjc4NWM1MDE3ODJjNzY3 OTQ4MzU3Njk3OWUzZjIzZGE0NjI0ZTkzNS4uYjBmYTRlNWRjOGQ4N2MzYWMzODhkZGMyODUzNTUw NGRiZjZiNjMzOSAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDE4IEBACisyMDEyLTAyLTI3ICBYaW5n bmFuIFdhbmcgIDx4aW5nbmFuLndhbmdAaW50ZWwuY29tPgorCisgICAgICAgIFVzZSBJUFAgbWFs bG9jIGZvciBBdWRpb0FycmF5CisgICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3df YnVnLmNnaT9pZD03OTczOQorCisgICAgICAgIFJldmlld2VkIGJ5IE5PQk9EWSAoT09QUyEpLgor CisgICAgICAgIElQUCBsaWJyYXJ5IHByb3ZpZGVzIHRoZSBpcHBtYWxsb2Mgd2hpY2ggY2FuIG1h bGxvYyBtZW1vcnkgYWxpZ25lZCB3aXRoIDMyLWJ5dGUuCisgICAgICAgIEFzIEF1ZGlvQXJyYXkg bmVlZHMgMTYtYnl0ZSBhbGlnbm1lbnQgd2UgY2FuIHVzZSBpcHBtYWxsb2MgaW5zdGVhZC4KKwor ICAgICAgICAqIFdlYkNvcmUuZ3lwL1dlYkNvcmUuZ3lwOgorICAgICAgICAqIHBsYXRmb3JtL2F1 ZGlvL0F1ZGlvQXJyYXkuaDoKKyAgICAgICAgKFdlYkNvcmU6OkF1ZGlvQXJyYXk6On5BdWRpb0Fy cmF5KToKKyAgICAgICAgKFdlYkNvcmU6OkF1ZGlvQXJyYXk6OmFsbG9jYXRlKToKKwogMjAxMi0w Mi0yNyAgSHVhbmcgRG9uZ3N1bmcgIDxsdXh0ZWxsYUBjb21wYW55MTAwLm5ldD4KIAogICAgICAg ICBGaXhlZCBhIHR5cG8gaW4gQ2FudmFzUmVuZGVyaW5nQ29udGV4dDJEOjpkcmF3SW1hZ2UoSFRN TENhbnZhc0VsZW1lbnQpOwpkaWZmIC0tZ2l0IGEvU291cmNlL1dlYktpdC9jaHJvbWl1bS9DaGFu Z2VMb2cgYi9Tb3VyY2UvV2ViS2l0L2Nocm9taXVtL0NoYW5nZUxvZwppbmRleCBkN2ExZTFhMzY4 Y2ZmM2E5ZDc2MjA1ZTBlOGFmZjMyMzM3OWU1MmY4Li5jNDUxZWZjMjU3YWI3NGY2YmYxMTgyYzRj ZGQxOWYyZTgzZTIyZjJmIDEwMDY0NAotLS0gYS9Tb3VyY2UvV2ViS2l0L2Nocm9taXVtL0NoYW5n ZUxvZworKysgYi9Tb3VyY2UvV2ViS2l0L2Nocm9taXVtL0NoYW5nZUxvZwpAQCAtMSwzICsxLDEy IEBACisyMDEyLTAyLTI3ICBYaW5nbmFuIFdhbmcgIDx4aW5nbmFuLndhbmdAaW50ZWwuY29tPgor CisgICAgICAgIFVzZSBJUFAgbWFsbG9jIGZvciBBdWRpb0FycmF5CisgICAgICAgIGh0dHBzOi8v YnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD03OTczOQorCisgICAgICAgIFJldmlld2Vk IGJ5IE5PQk9EWSAoT09QUyEpLgorCisgICAgICAgICogV2ViS2l0Lmd5cDoKKwogMjAxMi0wMi0y NyAgQWRyaWVubmUgV2Fsa2VyICA8ZW5uZUBnb29nbGUuY29tPgogCiAgICAgICAgIFtjaHJvbWl1 bV0gVW5yZXZpZXdlZCBnYXJkZW5pbmcsIGRpc2FibGUgVHJhbnNwYXJlbmN5V2luVGVzdCB1bml0 IHRlc3RzCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9XZWJDb3JlLmd5cC9XZWJDb3JlLmd5 cCBiL1NvdXJjZS9XZWJDb3JlL1dlYkNvcmUuZ3lwL1dlYkNvcmUuZ3lwCmluZGV4IGFhOTUwZmQ0 MjMyMzQ4ZGNjNGEzYzQzNDc2ZDg3YTU4Y2Y1N2ViOGQuLjg0MjlkYWU1YzZhNjhhMTVhNzk2ODdj NTczNDk5YTVmNTJjZWQ4ZGMgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJDb3JlL1dlYkNvcmUuZ3lw L1dlYkNvcmUuZ3lwCisrKyBiL1NvdXJjZS9XZWJDb3JlL1dlYkNvcmUuZ3lwL1dlYkNvcmUuZ3lw CkBAIC0yMDQwLDYgKzIwNDAsMTMgQEAKICAgICAgICdzb3VyY2VzLyc6IFsKICAgICAgICAgWydl eGNsdWRlJywgJ3Rlc3RpbmcvanMnXSwKICAgICAgIF0sCisgICAgICAnY29uZGl0aW9ucyc6IFsK KyAgICAgICAgWydPUz09ImxpbnV4IiBhbmQgIldURl9VU0VfV0VCQVVESU9fSVBQPTEiIGluIGZl YXR1cmVfZGVmaW5lcycsIHsKKyAgICAgICAgICAnY2ZsYWdzJzogWworICAgICAgICAgICAgJzwh QChwa2ctY29uZmlnIC0tY2ZsYWdzLW9ubHktSSBpcHApJywKKyAgICAgICAgICBdLAorICAgICAg ICB9XSwKKyAgICAgIF0sCiAgICAgfSwKICAgXSwgICMgdGFyZ2V0cwogfQpkaWZmIC0tZ2l0IGEv U291cmNlL1dlYkNvcmUvcGxhdGZvcm0vYXVkaW8vQXVkaW9BcnJheS5oIGIvU291cmNlL1dlYkNv cmUvcGxhdGZvcm0vYXVkaW8vQXVkaW9BcnJheS5oCmluZGV4IDdhMjI1MWIzOGIyZjQyZTA1Yzhl NDBhMTBiYjczNDJjNGIyZTY0ZDAuLmM5ZjdmODRjNzkxNDU2N2RhYjE2ODg2NmRmMTc4OTZkMmRl MDVhOGMgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJDb3JlL3BsYXRmb3JtL2F1ZGlvL0F1ZGlvQXJy YXkuaAorKysgYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9hdWRpby9BdWRpb0FycmF5LmgKQEAg LTMzLDYgKzMzLDEwIEBACiAjaW5jbHVkZSA8d3RmL0Zhc3RNYWxsb2MuaD4KICNpbmNsdWRlIDx3 dGYvVmVjdG9yLmg+CiAKKyNpZiBVU0UoV0VCQVVESU9fSVBQKQorI2luY2x1ZGUgPGlwcGNvcmUu aD4KKyNlbmRpZiAvLyBVU0UoV0VCQVVESU9fSVBQKQorCiBuYW1lc3BhY2UgV2ViQ29yZSB7CiAK IHRlbXBsYXRlPHR5cGVuYW1lIFQ+CkBAIC00Niw3ICs1MCwxMSBAQCBwdWJsaWM6CiAKICAgICB+ QXVkaW9BcnJheSgpCiAgICAgeworI2lmIFVTRShXRUJBVURJT19JUFApCisgICAgICAgIGlwcEZy ZWUocmVpbnRlcnByZXRfY2FzdDx2b2lkKj4obV9hbGxvY2F0aW9uKSk7CisjZWxzZQogICAgICAg ICBmYXN0RnJlZShtX2FsbG9jYXRpb24pOworI2VuZGlmCiAgICAgfQogCiAgICAgLy8gSXQncyBP SyB0byBjYWxsIGFsbG9jYXRlKCkgbXVsdGlwbGUgdGltZXMsIGJ1dCBkYXRhIHdpbGwgKm5vdCog YmUgY29waWVkIGZyb20gYW4gaW5pdGlhbCBhbGxvY2F0aW9uCkBAIC02MCw2ICs2OCwxNiBAQCBw dWJsaWM6CiAgICAgICAKICAgICAgICAgdW5zaWduZWQgaW5pdGlhbFNpemUgPSBzaXplb2YoVCkg KiBuOwogCisjaWYgVVNFKFdFQkFVRElPX0lQUCkKKyAgICAgICAgaWYgKG1fYWxsb2NhdGlvbikK KyAgICAgICAgICAgIGlwcEZyZWUocmVpbnRlcnByZXRfY2FzdDx2b2lkKj4obV9hbGxvY2F0aW9u KSk7CisKKyAgICAgICAgLy8gVGhlIGlwcE1hbGxvYyBhbGxvY2F0ZXMgbWVtb3J5IGFsaWduZWQg dG8gMzItYnl0ZSBib3VuZGFyeS4KKyAgICAgICAgbV9hbGxvY2F0aW9uID0gcmVpbnRlcnByZXRf Y2FzdDxUKj4oaXBwTWFsbG9jKGluaXRpYWxTaXplKSk7CisgICAgICAgIG1fYWxpZ25lZERhdGEg PSBtX2FsbG9jYXRpb247CisgICAgICAgIG1fc2l6ZSA9IG47CisgICAgICAgIHplcm8oKTsKKyNl bHNlCiAgICAgICAgIC8vIDE2LWJ5dGUgYWxpZ25tZW50IGZvciAxMjhiaXQgU0lNRC4KICAgICAg ICAgY29uc3Qgc2l6ZV90IGFsaWdubWVudCA9IDE2OwogCkBAIC05Myw2ICsxMTEsNyBAQCBwdWJs aWM6CiAgICAgICAgICAgICAgICAgZmFzdEZyZWUoYWxsb2NhdGlvbik7CiAgICAgICAgICAgICB9 CiAgICAgICAgIH0KKyNlbmRpZgogICAgIH0KIAogICAgIFQqIGRhdGEoKSB7IHJldHVybiBtX2Fs aWduZWREYXRhOyB9CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L2Nocm9taXVtL1dlYktpdC5n eXAgYi9Tb3VyY2UvV2ViS2l0L2Nocm9taXVtL1dlYktpdC5neXAKaW5kZXggM2RhMDQ3N2EwMTE0 ZGU5ZjhiNzYwNGRmZDFiYjUwYzhjN2M2NGZhMC4uZjJiZjFlNzVlZjJjYTA5OGQ1NDA1YmRkODNi MGNjZjkyN2U5ZjVkZSAxMDA2NDQKLS0tIGEvU291cmNlL1dlYktpdC9jaHJvbWl1bS9XZWJLaXQu Z3lwCisrKyBiL1NvdXJjZS9XZWJLaXQvY2hyb21pdW0vV2ViS2l0Lmd5cApAQCAtNjk0LDYgKzY5 NCwxMSBAQAogICAgICAgICAgICAgICAgICdzcmMvd2luL1dlYlNjcmVlbkluZm9GYWN0b3J5LmNw cCcsCiAgICAgICAgICAgICBdLAogICAgICAgICAgICAgJ2NvbmRpdGlvbnMnOiBbCisgICAgICAg ICAgICAgICAgWydPUz09ImxpbnV4IiBhbmQgIldURl9VU0VfV0VCQVVESU9fSVBQPTEiIGluIGZl YXR1cmVfZGVmaW5lcycsIHsKKyAgICAgICAgICAgICAgICAgICAgJ2NmbGFncyc6IFsKKyAgICAg ICAgICAgICAgICAgICAgICAgICc8IUAocGtnLWNvbmZpZyAtLWNmbGFncy1vbmx5LUkgaXBwKScs CisgICAgICAgICAgICAgICAgICAgIF0sCisgICAgICAgICAgICAgICAgfV0sCiAgICAgICAgICAg ICAgICAgWydpbnNpZGVfY2hyb21pdW1fYnVpbGQ9PTEnLCB7CiAgICAgICAgICAgICAgICAgICAg ICd0eXBlJzogJzwoY29tcG9uZW50KScsCiAK