Auto merge of #141 - rust-lang-nursery:aapcs, r=alexcrichton

use AAPCS calling convention on all aeabi intrinsics

also, on ARM, inline(always) the actual implementation of the intrinsics so we
end with code like this:

```
00000000 <__aeabi_dadd>:
    (implementation here)
```

instead of "trampolines" like this:

```
00000000 <__aeabi_dadd>:
    (shuffle registers)
    (call __adddf3)

00000000 <__adddf3>:
    (implementation here)
```

closes #116

cc #66
r? @alexcrichton
cc @mattico

src/arm.rs

@@ -62,44 +62,44 @@ pub unsafe fn __aeabi_ldivmod() {
 
 // TODO: These aeabi_* functions should be defined as aliases
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_dadd(a: f64, b: f64) -> f64 {
+pub extern "aapcs" fn __aeabi_dadd(a: f64, b: f64) -> f64 {
     ::float::add::__adddf3(a, b)
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_fadd(a: f32, b: f32) -> f32 {
+pub extern "aapcs" fn __aeabi_fadd(a: f32, b: f32) -> f32 {
     ::float::add::__addsf3(a, b)
 }
 
 #[cfg(not(all(feature = "c", target_arch = "arm", not(target_os = "ios"), not(thumbv6m))))]
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_idiv(a: i32, b: i32) -> i32 {
+pub extern "aapcs" fn __aeabi_idiv(a: i32, b: i32) -> i32 {
     ::int::sdiv::__divsi3(a, b)
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_lasr(a: i64, b: u32) -> i64 {
+pub extern "aapcs" fn __aeabi_lasr(a: i64, b: u32) -> i64 {
     ::int::shift::__ashrdi3(a, b)
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_llsl(a: u64, b: u32) -> u64 {
+pub extern "aapcs" fn __aeabi_llsl(a: u64, b: u32) -> u64 {
     ::int::shift::__ashldi3(a, b)
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_llsr(a: u64, b: u32) -> u64 {
+pub extern "aapcs" fn __aeabi_llsr(a: u64, b: u32) -> u64 {
     ::int::shift::__lshrdi3(a, b)
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_lmul(a: u64, b: u64) -> u64 {
+pub extern "aapcs" fn __aeabi_lmul(a: u64, b: u64) -> u64 {
     ::int::mul::__muldi3(a, b)
 }
 
 #[cfg(not(all(feature = "c", target_arch = "arm", not(target_os = "ios"), not(thumbv6m))))]
 #[cfg_attr(not(test), no_mangle)]
-pub extern "C" fn __aeabi_uidiv(a: u32, b: u32) -> u32 {
+pub extern "aapcs" fn __aeabi_uidiv(a: u32, b: u32) -> u32 {
     ::int::udiv::__udivsi3(a, b)
 }
 
@@ -113,55 +113,55 @@ extern "C" {
 // FIXME: The `*4` and `*8` variants should be defined as aliases.
 
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memcpy(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memcpy(dest: *mut u8, src: *const u8, n: usize) {
     memcpy(dest, src, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memcpy4(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memcpy4(dest: *mut u8, src: *const u8, n: usize) {
     memcpy(dest, src, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memcpy8(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memcpy8(dest: *mut u8, src: *const u8, n: usize) {
     memcpy(dest, src, n);
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memmove(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memmove(dest: *mut u8, src: *const u8, n: usize) {
     memmove(dest, src, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memmove4(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memmove4(dest: *mut u8, src: *const u8, n: usize) {
     memmove(dest, src, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memmove8(dest: *mut u8, src: *const u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memmove8(dest: *mut u8, src: *const u8, n: usize) {
     memmove(dest, src, n);
 }
 
 // Note the different argument order
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memset(dest: *mut u8, n: usize, c: i32) {
+pub unsafe extern "aapcs" fn __aeabi_memset(dest: *mut u8, n: usize, c: i32) {
     memset(dest, c, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memset4(dest: *mut u8, n: usize, c: i32) {
+pub unsafe extern "aapcs" fn __aeabi_memset4(dest: *mut u8, n: usize, c: i32) {
     memset(dest, c, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memset8(dest: *mut u8, n: usize, c: i32) {
+pub unsafe extern "aapcs" fn __aeabi_memset8(dest: *mut u8, n: usize, c: i32) {
     memset(dest, c, n);
 }
 
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memclr(dest: *mut u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memclr(dest: *mut u8, n: usize) {
     memset(dest, 0, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memclr4(dest: *mut u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memclr4(dest: *mut u8, n: usize) {
     memset(dest, 0, n);
 }
 #[cfg_attr(not(test), no_mangle)]
-pub unsafe extern "C" fn __aeabi_memclr8(dest: *mut u8, n: usize) {
+pub unsafe extern "aapcs" fn __aeabi_memclr8(dest: *mut u8, n: usize) {
     memset(dest, 0, n);
 }
 

src/float/add.rs

@@ -4,11 +4,11 @@ use core::num::Wrapping;
 use float::Float;
 
 macro_rules! add {
-    ($intrinsic:ident: $ty:ty) => {
+    ($abi:tt, $intrinsic:ident: $ty:ty) => {
         /// Returns `a + b`
         #[allow(unused_parens)]
         #[cfg_attr(not(test), no_mangle)]
-        pub extern fn $intrinsic(a: $ty, b: $ty) -> $ty {
+        pub extern $abi fn $intrinsic(a: $ty, b: $ty) -> $ty {
             let one = Wrapping(1 as <$ty as Float>::Int);
             let zero = Wrapping(0 as <$ty as Float>::Int);
 
@@ -181,8 +181,17 @@ macro_rules! add {
     }
 }
 
-add!(__addsf3: f32);
-add!(__adddf3: f64);
+#[cfg(target_arch = "arm")]
+add!("aapcs", __addsf3: f32);
+
+#[cfg(not(target_arch = "arm"))]
+add!("C", __addsf3: f32);
+
+#[cfg(target_arch = "arm")]
+add!("aapcs", __adddf3: f64);
+
+#[cfg(not(target_arch = "arm"))]
+add!("C", __adddf3: f64);
 
 // NOTE(cfg) for some reason, on arm*-unknown-linux-gnueabi*, our implementation doesn't
 // match the output of its gcc_s or compiler-rt counterpart. Until we investigate further, we'll
@@ -194,14 +203,14 @@ mod tests {
     use qc::{F32, F64};
 
     check! {
-        fn __addsf3(f: extern fn(f32, f32) -> f32,
+        fn __addsf3(f: extern "C" fn(f32, f32) -> f32,
                     a: F32,
                     b: F32)
                     -> Option<F32> {
             Some(F32(f(a.0, b.0)))
         }
 
-        fn __adddf3(f: extern fn(f64, f64) -> f64,
+        fn __adddf3(f: extern "C" fn(f64, f64) -> f64,
                     a: F64,
                     b: F64) -> Option<F64> {
             Some(F64(f(a.0, b.0)))

src/float/pow.rs

@@ -34,13 +34,13 @@ mod tests {
     use qc::{I32, F32, F64};
 
     check! {
-        fn __powisf2(f: extern fn(f32, i32) -> f32, 
+        fn __powisf2(f: extern "C" fn(f32, i32) -> f32,
                      a: F32,
                      b: I32) -> Option<F32> {
             Some(F32(f(a.0, b.0)))
         }
 
-        fn __powidf2(f: extern fn(f64, i32) -> f64, 
+        fn __powidf2(f: extern "C" fn(f64, i32) -> f64,
                      a: F64,
                      b: I32) -> Option<F64> {
             Some(F64(f(a.0, b.0)))

src/int/mul.rs

@@ -2,10 +2,11 @@ use int::LargeInt;
 use int::Int;
 
 macro_rules! mul {
-    ($intrinsic:ident: $ty:ty) => {
+    ($(#[$attr:meta])+ |
+     $abi:tt, $intrinsic:ident: $ty:ty) => {
         /// Returns `a * b`
-        #[cfg_attr(not(test), no_mangle)]
-        pub extern "C" fn $intrinsic(a: $ty, b: $ty) -> $ty {
+        $(#[$attr])+
+        pub extern $abi fn $intrinsic(a: $ty, b: $ty) -> $ty {
             let half_bits = <$ty>::bits() / 4;
             let lower_mask = !0 >> half_bits;
             let mut low = (a.low() & lower_mask).wrapping_mul(b.low() & lower_mask);
@@ -73,9 +74,17 @@ macro_rules! mulo {
 }
 
 #[cfg(not(all(feature = "c", target_arch = "x86")))]
-mul!(__muldi3: u64);
+mul!(#[cfg_attr(all(not(test), not(target_arch = "arm")), no_mangle)]
+     #[cfg_attr(all(not(test), target_arch = "arm"), inline(always))]
+     | "C", __muldi3: u64);
+
+#[cfg(not(target_arch = "arm"))]
+mul!(#[cfg_attr(not(test), no_mangle)]
+     | "C", __multi3: i128);
 
-mul!(__multi3: i128);
+#[cfg(target_arch = "arm")]
+mul!(#[cfg_attr(not(test), no_mangle)]
+     | "aapcs", __multi3: i128);
 
 mulo!(__mulosi4: i32);
 mulo!(__mulodi4: i64);
@@ -90,12 +99,12 @@ mod tests {
     use qc::{I32, I64, U64};
 
     check! {
-        fn __muldi3(f: extern fn(u64, u64) -> u64, a: U64, b: U64)
+        fn __muldi3(f: extern "C" fn(u64, u64) -> u64, a: U64, b: U64)
                     -> Option<u64> {
             Some(f(a.0, b.0))
         }
 
-        fn __mulosi4(f: extern fn(i32, i32, &mut i32) -> i32,
+        fn __mulosi4(f: extern "C" fn(i32, i32, &mut i32) -> i32,
                      a: I32,
                      b: I32) -> Option<(i32, i32)> {
             let (a, b) = (a.0, b.0);
@@ -107,7 +116,7 @@ mod tests {
             Some((r, overflow))
         }
 
-        fn __mulodi4(f: extern fn(i64, i64, &mut i32) -> i64,
+        fn __mulodi4(f: extern "C" fn(i64, i64, &mut i32) -> i64,
                      a: I64,
                      b: I64) -> Option<(i64, i32)> {
             let (a, b) = (a.0, b.0);
@@ -130,11 +139,11 @@ mod tests_i128 {
     use qc::I128;
 
     check! {
-        fn __multi3(f: extern fn(i128, i128) -> i128, a: I128, b: I128)
+        fn __multi3(f: extern "C" fn(i128, i128) -> i128, a: I128, b: I128)
                     -> Option<i128> {
             Some(f(a.0, b.0))
         }
-        fn __muloti4(f: extern fn(i128, i128, &mut i32) -> i128,
+        fn __muloti4(f: extern "C" fn(i128, i128, &mut i32) -> i128,
                      a: I128,
                      b: I128) -> Option<(i128, i32)> {
             let (a, b) = (a.0, b.0);

src/int/sdiv.rs

@@ -40,10 +40,10 @@ macro_rules! mod_ {
 }
 
 macro_rules! divmod {
-    ($intrinsic:ident, $div:ident: $ty:ty) => {
+    ($abi:tt, $intrinsic:ident, $div:ident: $ty:ty) => {
         /// Returns `a / b` and sets `*rem = n % d`
         #[cfg_attr(not(test), no_mangle)]
-        pub extern "C" fn $intrinsic(a: $ty, b: $ty, rem: &mut $ty) -> $ty {
+        pub extern $abi fn $intrinsic(a: $ty, b: $ty, rem: &mut $ty) -> $ty {
             #[cfg(all(feature = "c", any(target_arch = "x86")))]
             extern {
                 fn $div(a: $ty, b: $ty) -> $ty;
@@ -86,16 +86,20 @@ mod_!(__modti3: i128, u128);
 mod_!(__modti3: i128, u128, ::U64x2, ::sconv);
 
 #[cfg(not(all(feature = "c", target_arch = "arm", not(target_os = "ios"))))]
-divmod!(__divmodsi4, __divsi3: i32);
+divmod!("C", __divmodsi4, __divsi3: i32);
 
-divmod!(__divmoddi4, __divdi3: i64);
+#[cfg(target_arch = "arm")]
+divmod!("aapcs", __divmoddi4, __divdi3: i64);
+
+#[cfg(not(target_arch = "arm"))]
+divmod!("C", __divmoddi4, __divdi3: i64);
 
 #[cfg(test)]
 mod tests {
     use qc::{U32, U64};
 
     check! {
-        fn __divdi3(f: extern fn(i64, i64) -> i64, n: U64, d: U64) -> Option<i64> {
+        fn __divdi3(f: extern "C" fn(i64, i64) -> i64, n: U64, d: U64) -> Option<i64> {
             let (n, d) = (n.0 as i64, d.0 as i64);
             if d == 0 {
                 None
@@ -104,7 +108,7 @@ mod tests {
             }
         }
 
-        fn __moddi3(f: extern fn(i64, i64) -> i64, n: U64, d: U64) -> Option<i64> {
+        fn __moddi3(f: extern "C" fn(i64, i64) -> i64, n: U64, d: U64) -> Option<i64> {
             let (n, d) = (n.0 as i64, d.0 as i64);
             if d == 0 {
                 None
@@ -113,7 +117,22 @@ mod tests {
             }
         }
 
-        fn __divmoddi4(f: extern fn(i64, i64, &mut i64) -> i64,
+        #[cfg(target_arch = "arm")]
+        fn __divmoddi4(f: extern "aapcs" fn(i64, i64, &mut i64) -> i64,
+                       n: U64,
+                       d: U64) -> Option<(i64, i64)> {
+            let (n, d) = (n.0 as i64, d.0 as i64);
+            if d == 0 {
+                None
+            } else {
+                let mut r = 0;
+                let q = f(n, d, &mut r);
+                Some((q, r))
+            }
+        }
+
+        #[cfg(not(target_arch = "arm"))]
+        fn __divmoddi4(f: extern "C" fn(i64, i64, &mut i64) -> i64,
                        n: U64,
                        d: U64) -> Option<(i64, i64)> {
             let (n, d) = (n.0 as i64, d.0 as i64);
@@ -126,7 +145,7 @@ mod tests {
             }
         }
 
-        fn __divsi3(f: extern fn(i32, i32) -> i32,
+        fn __divsi3(f: extern "C" fn(i32, i32) -> i32,
                     n: U32,
                     d: U32) -> Option<i32> {
             let (n, d) = (n.0 as i32, d.0 as i32);
@@ -137,7 +156,7 @@ mod tests {
             }
         }
 
-        fn __modsi3(f: extern fn(i32, i32) -> i32,
+        fn __modsi3(f: extern "C" fn(i32, i32) -> i32,
                     n: U32,
                     d: U32) -> Option<i32> {
             let (n, d) = (n.0 as i32, d.0 as i32);
@@ -148,7 +167,7 @@ mod tests {
             }
         }
 
-        fn __divmodsi4(f: extern fn(i32, i32, &mut i32) -> i32,
+        fn __divmodsi4(f: extern "C" fn(i32, i32, &mut i32) -> i32,
                        n: U32,
                        d: U32) -> Option<(i32, i32)> {
             let (n, d) = (n.0 as i32, d.0 as i32);
@@ -172,7 +191,7 @@ mod tests_i128 {
     use qc::U128;
     check! {
 
-        fn __divti3(f: extern fn(i128, i128) -> i128, n: U128, d: U128) -> Option<i128> {
+        fn __divti3(f: extern "C" fn(i128, i128) -> i128, n: U128, d: U128) -> Option<i128> {
             let (n, d) = (n.0 as i128, d.0 as i128);
             if d == 0 {
                 None
@@ -181,7 +200,7 @@ mod tests_i128 {
             }
         }
 
-        fn __modti3(f: extern fn(i128, i128) -> i128, n: U128, d: U128) -> Option<i128> {
+        fn __modti3(f: extern "C" fn(i128, i128) -> i128, n: U128, d: U128) -> Option<i128> {
             let (n, d) = (n.0 as i128, d.0 as i128);
             if d == 0 {
                 None