Reviewers: ulan, Benedikt Meurer,
Message:
I realise that you may prefer to wait for a Clang fix for this issue. In
that
case, it might be better to wait, since without this patch a fixed Clang
could
generate marginally better code.
What do you think?
Description:
ARM64: Work around Clang's inline assembly limitations.
Clang bug: https://llvm.org/bugs/show_bug.cgi?id=23575
This is the simplest workaround; just don't use the "J" constraint for
Clang. At worst, an extra `mov` will be generated to synthesize negative
values in a register.
BUG=
Please review this at https://codereview.chromium.org/1161553008/
Base URL: https://chromium.googlesource.com/v8/v8.git@master
Affected files (+20, -0 lines):
M src/base/atomicops_internals_arm64_gcc.h
Index: src/base/atomicops_internals_arm64_gcc.h
diff --git a/src/base/atomicops_internals_arm64_gcc.h
b/src/base/atomicops_internals_arm64_gcc.h
index
b01783e6a7e275cfe09e53576cfecea892ab41ff..c7de95d80d174d2740fc2b57d1e975a2aaf519fe
100644
--- a/src/base/atomicops_internals_arm64_gcc.h
+++ b/src/base/atomicops_internals_arm64_gcc.h
@@ -39,7 +39,12 @@ inline Atomic32 NoBarrier_CompareAndSwap(volatile
Atomic32* ptr,
: [prev]"=&r" (prev),
[temp]"=&r" (temp),
[ptr]"+Q" (*ptr)
+#ifdef __clang__
+ // Workaround for Clang (https://llvm.org/bugs/show_bug.cgi?id=23575).
+ : [old_value]"Ir" (old_value),
+#else
: [old_value]"IJr" (old_value),
+#endif
[new_value]"r" (new_value)
: "cc", "memory"
); // NOLINT
@@ -81,7 +86,12 @@ inline Atomic32 NoBarrier_AtomicIncrement(volatile
Atomic32* ptr,
: [result]"=&r" (result),
[temp]"=&r" (temp),
[ptr]"+Q" (*ptr)
+#ifdef __clang__
+ // Workaround for Clang (https://llvm.org/bugs/show_bug.cgi?id=23575).
+ : [increment]"Ir" (increment)
+#else
: [increment]"IJr" (increment)
+#endif
: "memory"
); // NOLINT
@@ -189,7 +199,12 @@ inline Atomic64 NoBarrier_CompareAndSwap(volatile
Atomic64* ptr,
: [prev]"=&r" (prev),
[temp]"=&r" (temp),
[ptr]"+Q" (*ptr)
+#ifdef __clang__
+ // Workaround for Clang (https://llvm.org/bugs/show_bug.cgi?id=23575).
+ : [old_value]"Ir" (old_value),
+#else
: [old_value]"IJr" (old_value),
+#endif
[new_value]"r" (new_value)
: "cc", "memory"
); // NOLINT
@@ -231,7 +246,12 @@ inline Atomic64 NoBarrier_AtomicIncrement(volatile
Atomic64* ptr,
: [result]"=&r" (result),
[temp]"=&r" (temp),
[ptr]"+Q" (*ptr)
+#ifdef __clang__
+ // Workaround for Clang (https://llvm.org/bugs/show_bug.cgi?id=23575).
+ : [increment]"Ir" (increment)
+#else
: [increment]"IJr" (increment)
+#endif
: "memory"
); // NOLINT
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
