-
Notifications
You must be signed in to change notification settings - Fork 15.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add support for solaris atomicops #32
Conversation
This patch adds support for atomic operations on Solaris, on any platform. It makes use of the atomic functions made available in Solaris' atomic.h header.
@@ -0,0 +1,189 @@ | |||
// Protocol Buffers - Google's data interchange format |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you update this copyright notice and remove lines that have "Protocol Buffers", "Google"?
Should be good now |
@@ -0,0 +1,188 @@ | |||
// Copyright 2014 Google Inc. All rights reserved. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I was suggesting putting yourself as the copyright owner and removing "Google Inc.", "Google Protocol Buffers" to avoid confusion. Note that putting "Google Inc." here does not affect your rights.
Let me know if you want to use your own name, or is fine with using "Google Inc.".
I'm fine with Google Inc. |
Can you also change CONTRIBUTORS.txt and add yourself to the "patches" section? Thanks! |
Added myself |
Add support for solaris atomicops
Moved lunit to third_party for Google compliance.
…um value is detected. This should never happen, so I don't think it matters much exactly what kind of exception we throw. We could even arguably return null, but this option saves a lot of space while still preserving some error checking. See https://godbolt.org/z/jKhcKs3x1 for code gen. This generates much tighter bytecode and ARM assembly than alternatives. As this code is generated many times over, small wins in code size here can reduce icache pressure, APK size, and OAT size. This java code: ```java Object uoe() { throw new UnsupportedOperationException(); } Object npe2() { throw null; } ``` Generates this dex code: ``` .method uoe()Ljava/lang/Object; new-instance v0, Ljava/lang/UnsupportedOperationException; invoke-direct {v0}, Ljava/lang/UnsupportedOperationException;-><init>()V throw v0 .end method .method npe2()Ljava/lang/Object; const/4 v0, 0x0 throw v0 .end method ``` Which generates this OAT code: ``` java.lang.Object SomeProto.uoe() [84 bytes] 0x000081c0 sub x16, sp, #0x2000 (8192) 0x000081c4 ldr wzr, [x16] StackMap[0] native_pc=0x41c8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000081c8 str x0, [sp, #-48]! 0x000081cc str x22, [sp, #24] 0x000081d0 stp x23, lr, [sp, #32] 0x000081d4 ldr x21, [x21] StackMap[1] native_pc=0x41d8, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000081d8 mov x22, x1 0x000081dc adrp x0, #+0x4000 (addr 0x0000c000) 0x000081e0 ldr w0, [x0, #4] 0x000081e4 ldr lr, [tr, #464] ; pAllocObjectInitialized 0x000081e8 blr lr StackMap[2] native_pc=0x41ec, dex_pc=0x0, register_mask=0x400000, stack_mask=0b 0x000081ec dmb ishst 0x000081f0 mov x1, x0 0x000081f4 mov x23, x1 0x000081f8 adrp x0, #+0x4000 (addr 0x0000c000) 0x000081fc ldr w0, [x0, #12] 0x00008200 ldr lr, [x0, #24] 0x00008204 blr lr StackMap[3] native_pc=0x4208, dex_pc=0x2, register_mask=0xc00000, stack_mask=0b 0x00008208 mov x0, x23 0x0000820c ldr lr, [tr, #1264] ; pDeliverException 0x00008210 blr lr StackMap[4] native_pc=0x4214, dex_pc=0x5, register_mask=0xc00000, stack_mask=0b java.lang.Object SomeProto.npe2() [36 bytes] 0x000080d0 sub x16, sp, #0x2000 (8192) 0x000080d4 ldr wzr, [x16] StackMap[0] native_pc=0x40d8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000080d8 str x0, [sp, #-32]! 0x000080dc stp x22, lr, [sp, #16] 0x000080e0 ldr x21, [x21] StackMap[1] native_pc=0x40e4, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000080e4 mov x22, x1 0x000080e8 mov w0, #0x0 0x000080ec ldr lr, [tr, #1264] ; pDeliverException 0x000080f0 blr lr StackMap[2] native_pc=0x40f4, dex_pc=0x1, register_mask=0x400000, stack_mask=0b ``` This saves 84-36 = 48 bytes of OAT per method. PiperOrigin-RevId: 684258075
…um value is detected. This should never happen, so I don't think it matters much exactly what kind of exception we throw. We could even arguably return null, but this option saves a lot of space while still preserving some error checking. See https://godbolt.org/z/jKhcKs3x1 for code gen. This generates much tighter bytecode and ARM assembly than alternatives. As this code is generated many times over, small wins in code size here can reduce icache pressure, APK size, and OAT size. This java code: ```java Object uoe() { throw new UnsupportedOperationException(); } Object npe2() { throw null; } ``` Generates this dex code: ``` .method uoe()Ljava/lang/Object; new-instance v0, Ljava/lang/UnsupportedOperationException; invoke-direct {v0}, Ljava/lang/UnsupportedOperationException;-><init>()V throw v0 .end method .method npe2()Ljava/lang/Object; const/4 v0, 0x0 throw v0 .end method ``` Which generates this OAT code: ``` java.lang.Object SomeProto.uoe() [84 bytes] 0x000081c0 sub x16, sp, #0x2000 (8192) 0x000081c4 ldr wzr, [x16] StackMap[0] native_pc=0x41c8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000081c8 str x0, [sp, #-48]! 0x000081cc str x22, [sp, #24] 0x000081d0 stp x23, lr, [sp, #32] 0x000081d4 ldr x21, [x21] StackMap[1] native_pc=0x41d8, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000081d8 mov x22, x1 0x000081dc adrp x0, #+0x4000 (addr 0x0000c000) 0x000081e0 ldr w0, [x0, #4] 0x000081e4 ldr lr, [tr, #464] ; pAllocObjectInitialized 0x000081e8 blr lr StackMap[2] native_pc=0x41ec, dex_pc=0x0, register_mask=0x400000, stack_mask=0b 0x000081ec dmb ishst 0x000081f0 mov x1, x0 0x000081f4 mov x23, x1 0x000081f8 adrp x0, #+0x4000 (addr 0x0000c000) 0x000081fc ldr w0, [x0, #12] 0x00008200 ldr lr, [x0, #24] 0x00008204 blr lr StackMap[3] native_pc=0x4208, dex_pc=0x2, register_mask=0xc00000, stack_mask=0b 0x00008208 mov x0, x23 0x0000820c ldr lr, [tr, #1264] ; pDeliverException 0x00008210 blr lr StackMap[4] native_pc=0x4214, dex_pc=0x5, register_mask=0xc00000, stack_mask=0b java.lang.Object SomeProto.npe2() [36 bytes] 0x000080d0 sub x16, sp, #0x2000 (8192) 0x000080d4 ldr wzr, [x16] StackMap[0] native_pc=0x40d8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000080d8 str x0, [sp, #-32]! 0x000080dc stp x22, lr, [sp, #16] 0x000080e0 ldr x21, [x21] StackMap[1] native_pc=0x40e4, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000080e4 mov x22, x1 0x000080e8 mov w0, #0x0 0x000080ec ldr lr, [tr, #1264] ; pDeliverException 0x000080f0 blr lr StackMap[2] native_pc=0x40f4, dex_pc=0x1, register_mask=0x400000, stack_mask=0b ``` This saves 84-36 = 48 bytes of OAT per method. PiperOrigin-RevId: 684258075
…um value is detected. This should never happen, so I don't think it matters much exactly what kind of exception we throw. We could even arguably return null, but this option saves a lot of space while still preserving some error checking. See https://godbolt.org/z/jKhcKs3x1 for code gen. This generates much tighter bytecode and ARM assembly than alternatives. As this code is generated many times over, small wins in code size here can reduce icache pressure, APK size, and OAT size. This java code: ```java Object uoe() { throw new UnsupportedOperationException(); } Object npe2() { throw null; } ``` Generates this dex code: ``` .method uoe()Ljava/lang/Object; new-instance v0, Ljava/lang/UnsupportedOperationException; invoke-direct {v0}, Ljava/lang/UnsupportedOperationException;-><init>()V throw v0 .end method .method npe2()Ljava/lang/Object; const/4 v0, 0x0 throw v0 .end method ``` Which generates this OAT code: ``` java.lang.Object SomeProto.uoe() [84 bytes] 0x000081c0 sub x16, sp, #0x2000 (8192) 0x000081c4 ldr wzr, [x16] StackMap[0] native_pc=0x41c8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000081c8 str x0, [sp, #-48]! 0x000081cc str x22, [sp, #24] 0x000081d0 stp x23, lr, [sp, #32] 0x000081d4 ldr x21, [x21] StackMap[1] native_pc=0x41d8, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000081d8 mov x22, x1 0x000081dc adrp x0, #+0x4000 (addr 0x0000c000) 0x000081e0 ldr w0, [x0, #4] 0x000081e4 ldr lr, [tr, #464] ; pAllocObjectInitialized 0x000081e8 blr lr StackMap[2] native_pc=0x41ec, dex_pc=0x0, register_mask=0x400000, stack_mask=0b 0x000081ec dmb ishst 0x000081f0 mov x1, x0 0x000081f4 mov x23, x1 0x000081f8 adrp x0, #+0x4000 (addr 0x0000c000) 0x000081fc ldr w0, [x0, #12] 0x00008200 ldr lr, [x0, #24] 0x00008204 blr lr StackMap[3] native_pc=0x4208, dex_pc=0x2, register_mask=0xc00000, stack_mask=0b 0x00008208 mov x0, x23 0x0000820c ldr lr, [tr, #1264] ; pDeliverException 0x00008210 blr lr StackMap[4] native_pc=0x4214, dex_pc=0x5, register_mask=0xc00000, stack_mask=0b java.lang.Object SomeProto.npe2() [36 bytes] 0x000080d0 sub x16, sp, #0x2000 (8192) 0x000080d4 ldr wzr, [x16] StackMap[0] native_pc=0x40d8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000080d8 str x0, [sp, #-32]! 0x000080dc stp x22, lr, [sp, #16] 0x000080e0 ldr x21, [x21] StackMap[1] native_pc=0x40e4, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000080e4 mov x22, x1 0x000080e8 mov w0, #0x0 0x000080ec ldr lr, [tr, #1264] ; pDeliverException 0x000080f0 blr lr StackMap[2] native_pc=0x40f4, dex_pc=0x1, register_mask=0x400000, stack_mask=0b ``` This saves 84-36 = 48 bytes of OAT per method. PiperOrigin-RevId: 684258075
…um value is detected. This should never happen, so I don't think it matters much exactly what kind of exception we throw. We could even arguably return null, but this option saves a lot of space while still preserving some error checking. See https://godbolt.org/z/jKhcKs3x1 for code gen. This generates much tighter bytecode and ARM assembly than alternatives. As this code is generated many times over, small wins in code size here can reduce icache pressure, APK size, and OAT size. This java code: ```java Object uoe() { throw new UnsupportedOperationException(); } Object npe2() { throw null; } ``` Generates this dex code: ``` .method uoe()Ljava/lang/Object; new-instance v0, Ljava/lang/UnsupportedOperationException; invoke-direct {v0}, Ljava/lang/UnsupportedOperationException;-><init>()V throw v0 .end method .method npe2()Ljava/lang/Object; const/4 v0, 0x0 throw v0 .end method ``` Which generates this OAT code: ``` java.lang.Object SomeProto.uoe() [84 bytes] 0x000081c0 sub x16, sp, #0x2000 (8192) 0x000081c4 ldr wzr, [x16] StackMap[0] native_pc=0x41c8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000081c8 str x0, [sp, #-48]! 0x000081cc str x22, [sp, #24] 0x000081d0 stp x23, lr, [sp, #32] 0x000081d4 ldr x21, [x21] StackMap[1] native_pc=0x41d8, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000081d8 mov x22, x1 0x000081dc adrp x0, #+0x4000 (addr 0x0000c000) 0x000081e0 ldr w0, [x0, #4] 0x000081e4 ldr lr, [tr, #464] ; pAllocObjectInitialized 0x000081e8 blr lr StackMap[2] native_pc=0x41ec, dex_pc=0x0, register_mask=0x400000, stack_mask=0b 0x000081ec dmb ishst 0x000081f0 mov x1, x0 0x000081f4 mov x23, x1 0x000081f8 adrp x0, #+0x4000 (addr 0x0000c000) 0x000081fc ldr w0, [x0, #12] 0x00008200 ldr lr, [x0, #24] 0x00008204 blr lr StackMap[3] native_pc=0x4208, dex_pc=0x2, register_mask=0xc00000, stack_mask=0b 0x00008208 mov x0, x23 0x0000820c ldr lr, [tr, #1264] ; pDeliverException 0x00008210 blr lr StackMap[4] native_pc=0x4214, dex_pc=0x5, register_mask=0xc00000, stack_mask=0b java.lang.Object SomeProto.npe2() [36 bytes] 0x000080d0 sub x16, sp, #0x2000 (8192) 0x000080d4 ldr wzr, [x16] StackMap[0] native_pc=0x40d8, dex_pc=0x0, register_mask=0x0, stack_mask=0b 0x000080d8 str x0, [sp, #-32]! 0x000080dc stp x22, lr, [sp, #16] 0x000080e0 ldr x21, [x21] StackMap[1] native_pc=0x40e4, dex_pc=0x0, register_mask=0x2, stack_mask=0b 0x000080e4 mov x22, x1 0x000080e8 mov w0, #0x0 0x000080ec ldr lr, [tr, #1264] ; pDeliverException 0x000080f0 blr lr StackMap[2] native_pc=0x40f4, dex_pc=0x1, register_mask=0x400000, stack_mask=0b ``` This saves 84-36 = 48 bytes of OAT per method. PiperOrigin-RevId: 684620833
This patch adds support for atomic operations on Solaris, on any platform.
It makes use of the atomic functions made available in Solaris' atomic.h
header.