[GN] Define USE_EABI_HARDFLOAT=1 when arm_float_abi=="hard".

BUILD.gn

Index: BUILD.gn
diff --git a/BUILD.gn b/BUILD.gn
index dbc02959db14cb330691c15771f6e07b8a6f33b2..8b5e35d6471a87d2b08a98116d9e502ce3930658 100644
--- a/BUILD.gn
+++ b/BUILD.gn
@@ -143,7 +143,6 @@ config("toolchain") {
   cflags = []
 
   # TODO(jochen): Add support for arm subarchs, mips, mipsel, mips64el.
-
   if (v8_target_arch == "arm") {
     defines += [ "V8_TARGET_ARCH_ARM" ]
     if (current_cpu == "arm") {
@@ -170,8 +169,13 @@ config("toolchain") {
         "CAN_USE_ARMV7_INSTRUCTIONS",
         "CAN_USE_VFP3_INSTRUCTIONS",
         "CAN_USE_VFP32DREGS",
-        "USE_EABI_HARDFLOAT=0",
       ]
+
+      if (arm_float_abi == "hard") {

[slan, 2016/03/29 17:03:23]

Although this passes all the tryjobs, I've just discovered that arm_float_abi is
only declared when current_cpu == "arm". This is going to be an issue, as we
need to check this value to build a tool on the *host* (when target_cpu is "x86"
or "x64". I did not discover this until now, because Chromecast sets this value
in args.gn, giving it global visibility due to a GN KI.

In this case, we need to understand the ABI for the *target* toolchain, not the
current toolchain to set this define correctly.

Dirk: How do you feel about removing the if (current_toolchain == "arm")
condition from the delcaration in arm.gni? This would remove a developer
safeguard, but it would enable us to use the args as if their intent is to
describe the *target* (which I think is always the case).

Alternatively, we could add another arg in here, like v8_target_arm_float_abi,
but that is less preferable in my mind.

Thoughts?

[Dirk Pranke, 2016/03/29 21:03:40]

I assume you mean "(when current_cpu is "x86" or "x64")", right? target_cpu is 
effectively a constant and should be == "arm" across all toolchains; I don't
know
why you would need the arm args if target_cpu != "arm" or "arm64".

As an aside, it's important to distinguish between the host toolchain (which
should
always be x64) and the snapshot toolchain (which should be x86 for arm and x64
for
arm64).
I think we could probably change (current_cpu == "arm") to (target_cpu == "arm")
instead, which is probably what the intent actually is.
I think we can probably avoid that.

Let me know if my suggestions above work.

[slan, 2016/03/29 23:45:08]

Yes, that is what I meant! Thanks for seeing through the typo :)
I like this idea:
https://codereview.chromium.org/1842833003/
Thanks for the review!

+        defines += [ "USE_EABI_HARDFLOAT=1" ]
+      } else if (arm_float_abi == "softfp") {
+        defines += [ "USE_EABI_HARDFLOAT=0" ]
+      }
     }
 
     # TODO(jochen): Add support for arm_test_noprobe.
@@ -187,7 +191,7 @@ config("toolchain") {
   }
   if (v8_target_arch == "s390") {
     defines += [ "V8_TARGET_ARCH_S390" ]
-  } 
+  }
   if (v8_target_arch == "s390x") {
     defines += [
       "V8_TARGET_ARCH_S390",
@@ -200,7 +204,7 @@ config("toolchain") {
   if (v8_target_arch == "x64") {
     defines += [ "V8_TARGET_ARCH_X64" ]
   }
-     
+
   if (is_win) {
     defines += [ "WIN32" ]
     # TODO(jochen): Support v8_enable_prof.