<table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Issue</th>
<td>
<a href=https://github.com/llvm/llvm-project/issues/69530>69530</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>
Slow register allocation on arm64
</td>
</tr>
<tr>
<th>Labels</th>
<td>
new issue
</td>
</tr>
<tr>
<th>Assignees</th>
<td>
</td>
</tr>
<tr>
<th>Reporter</th>
<td>
etang-cw
</td>
</tr>
</table>
<pre>
To reproduce:
- Download [gfxreconstruct-slow.zip](https://github.com/llvm/llvm-project/files/13034364/gfxreconstruct-slow.zip) and unzip it
- Run `time clang++ -std=c++11 -O2 -arch arm64 -c generated_vulkan_replay_consumer.cpp -I include -DVK_NO_PROTOTYPES -DVK_ENABLE_BETA_EXTENSIONS -DNDEBUG -o generated_vulkan_replay_consumer.arm64.o -ftime-trace`
- View the time it took to compile
- View the time trace json file
- Do the same with `-arch x86_64`
Expected result:
Both take a similar amount of time
Actual result:
On an M1, the x86_64 build takes a bit over 1m, but the arm64 build takes over 13m
In the time trace json, each instantiation of `InitializeOutputStructPNext` takes ~23 seconds to `RunPass {"detail":"Greedy Register Allocator"}`. In the x86 builds, the same pass takes about 1 second per instantiation.
Reproduced with both Xcode 15's clang and a copy of clang-16 from MacPorts
I realize this file is terrible from a codegen time perspective, with a massive switch statement inside a templated function that gets instantiated many times, but even so, the 13x time difference between the backends seemed worth a bug report
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyUVMFu4zYQ_Rr6MpAhUZbiHHyIa2cRtGsHSbrYngyKGkncUKRAjmwnh357QSreeostil5siBzOvHnvzQjvVWsQV6xYs2IzEyN11q2QhGkTeZpVtn5bvVhwODhbjxJZfsfSDUvvEtjYk9FW1MCKdducHUprPLlRUuK1Pc3f1cCKDePLjmjw4SG_Z_y-VdSN1VzanvF7rY-Xv2Rw9htKYvy-URo94_dZnuaLvFyEV_9SgN-CMDWM5l0NoOiC7Wk0wMqUVI8gtTAt42vG15B4qlm-kdNnlkGy55AIJzsQri8XkEho0aAThPXhOOpXYQ4OBy3eDqH62KOby2GA5AGUkXqsEZLNl18Pu_3h8Wn_sn_543H7PB1td3fr37aH9fbl7rD9-rLdPT_sd-Fut9muf_8Eif3vUhHU3ELShFYSckIiK9NLm18UnoA6hNioIiBrX4EsSNsPSuPP42IW-OatgeYqaGNjiBc9wklRFwicqDkvy0O5-F53-t2eB5SENTj0o6bvxlhb6oDEK4IAr3qlhQPR29EQ2CYCuM5yJ2kU-p859gaEgc8Z479ETBMAqEal65jbg4BKEdgjOsj6EFeNFGMnHa9Dp6C8v677YH7GR8iDQnagjCdhSAlS1gTcrEwfjCIltHrH_UjDSM_Rio87PBMr049af_IcfDBq7YMOrEyfRvMovAd2s2ac10hCacZ5nAf-ySHWb_CErfKEDu60tlKQdSHiZsPKdA7wgfW8LKe2_IWWKNUQkn9wUtmRIPsAAAO6HxuZXzPwdBnpelK7Crp9lbZGyArGb_w0OHG8BEg7vAUe4lmSldA428NnIR-tI_8Ds-AwsgTUKR8dBsoDoXOq0jg9DAlrbNFMCgzofHCTOmLoLeIR0IftdETwJ0WyA0-CsEdDoSlVB38R9oMO8wPNaGTUijpB0CL5q9axhl6Yt1jLX7yCRzTg7YXLLD9PWGrVNOjQSIQK6YQ4sV8J-YpBVY_YB86siyCrsQ3b0Tqa1au8vs1vxQxXWXl7U5ZFcVPMulW9LKqsLqoKsWzkIhNc1M3iplmWTdrcYjVTK57yPEuzJc8WZVbMy2bZVKUQmKXNoswrtkixF0rPw6KcW9fOlPcjrsrbIk9nWlSofVzhnBs8QbwMDio2M7eKy7UaW88WqVae_N9ZSJHG1bO2J3AXC4rJgtH3Zpqm2ej06n9v8YgirPGI8q8AAAD__4V3IAU">