Remove (most) LLVM intrinsics from the compiler

This commit does the following

 - Removes most LLVM intrinsics from the compiler
 - Adds functions that optimize down to specific LLVM intrinsics to
   libcore
 - Utilize the sqrt intrinsic in f32/f64

Author: auroranockert

src/libcore/num/f32.rs

@@ -16,7 +16,7 @@ use num::NumCast;
 use num::strconv;
 use num;
 use option::Option;
-use unstable::intrinsics::floorf32;
+use unstable::intrinsics;
 use from_str;
 use to_str;
 
@@ -93,7 +93,6 @@ delegate!(fn ldexp_radix(n: c_float, i: c_int) -> c_float =
     cmath::c_float_utils::ldexp_radix)
 delegate!(fn sin(n: c_float) -> c_float = cmath::c_float_utils::sin)
 delegate!(fn sinh(n: c_float) -> c_float = cmath::c_float_utils::sinh)
-delegate!(fn sqrt(n: c_float) -> c_float = cmath::c_float_utils::sqrt)
 delegate!(fn tan(n: c_float) -> c_float = cmath::c_float_utils::tan)
 delegate!(fn tanh(n: c_float) -> c_float = cmath::c_float_utils::tanh)
 delegate!(fn tgamma(n: c_float) -> c_float = cmath::c_float_utils::tgamma)
@@ -146,7 +145,10 @@ pub fn gt(x: f32, y: f32) -> bool { return x > y; }
 
 /// Returns `x` rounded down
 #[inline(always)]
-pub fn floor(x: f32) -> f32 { unsafe { floorf32(x) } }
+pub fn floor(x: f32) -> f32 { unsafe { intrinsics::llvm::floorf32(x) } }
+
+#[inline(always)]
+pub fn sqrt(x: f32) -> f32 { unsafe { intrinsics::llvm::sqrtf32(x) } }
 
 // FIXME (#1999): replace the predicates below with llvm intrinsics or
 // calls to the libmath macros in the rust runtime for performance.

src/libcore/num/f64.rs

@@ -16,7 +16,7 @@ use num::NumCast;
 use num::strconv;
 use num;
 use option::Option;
-use unstable::intrinsics::floorf64;
+use unstable::intrinsics;
 use to_str;
 use from_str;
 
@@ -94,7 +94,6 @@ delegate!(fn ldexp_radix(n: c_double, i: c_int) -> c_double =
     cmath::c_double_utils::ldexp_radix)
 delegate!(fn sin(n: c_double) -> c_double = cmath::c_double_utils::sin)
 delegate!(fn sinh(n: c_double) -> c_double = cmath::c_double_utils::sinh)
-delegate!(fn sqrt(n: c_double) -> c_double = cmath::c_double_utils::sqrt)
 delegate!(fn tan(n: c_double) -> c_double = cmath::c_double_utils::tan)
 delegate!(fn tanh(n: c_double) -> c_double = cmath::c_double_utils::tanh)
 delegate!(fn tgamma(n: c_double) -> c_double = cmath::c_double_utils::tgamma)
@@ -221,7 +220,10 @@ pub fn is_finite(x: f64) -> bool {
 
 /// Returns `x` rounded down
 #[inline(always)]
-pub fn floor(x: f64) -> f64 { unsafe { floorf64(x) } }
+pub fn floor(x: f64) -> f64 { unsafe { intrinsics::llvm::floorf64(x) } }
+
+#[inline(always)]
+pub fn sqrt(x: f64) -> f64 { unsafe { intrinsics::llvm::sqrtf64(x) } }
 
 // FIXME (#1999): add is_normal, is_subnormal, and fpclassify
 

src/libcore/unstable/intrinsics.rs

@@ -63,69 +63,258 @@ pub extern "rust-intrinsic" {
 
     pub fn memmove32(dst: *mut u8, src: *u8, size: u32);
     pub fn memmove64(dst: *mut u8, src: *u8, size: u64);
+}
 
+#[nolink]
+pub extern mod llvm {
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="3, i32, 1"]
+    #[constant_arg="4, i1, 0"]
+    #[link_name="llvm.memmove.p0i8.p0i8.i32"]
+    #[cfg(not(stage0))]
+    pub fn memmove32(dst: *mut u8, src: *u8, size: i32);
+
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="3, i32, 1"]
+    #[constant_arg="4, i1, 0"]
+    #[link_name="llvm.memmove.p0i8.p0i8.i64"]
+    #[cfg(not(stage0))]
+    pub fn memmove64(dst: *mut u8, src: *u8, size: i64);
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.sqrt.f32"]
     pub fn sqrtf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.sqrt.f64"]
     pub fn sqrtf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.powi.f32"]
     pub fn powif32(a: f32, x: i32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.powi.f64"]
     pub fn powif64(a: f64, x: i32) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.sin.f32"]
     pub fn sinf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.sin.f64"]
     pub fn sinf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.cos.f32"]
     pub fn cosf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.cos.f64"]
     pub fn cosf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.pow.f32"]
     pub fn powf32(a: f32, x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.pow.f64"]
     pub fn powf64(a: f64, x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.exp.f32"]
     pub fn expf32(x: f32) -> f32;
-    pub fn expf64(x: f64) -> f64;
 
-    pub fn exp2f32(x: f32) -> f32;
-    pub fn exp2f64(x: f64) -> f64;
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.exp.f64"]
+    pub fn expf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.log.f32"]
     pub fn logf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.log.f64"]
     pub fn logf64(x: f64) -> f64;
 
+    /* NOTE: Do these intrinsics even exist?
+    pub fn exp2f32(x: f32) -> f32;
+    pub fn exp2f64(x: f64) -> f64;
+
     pub fn log10f32(x: f32) -> f32;
     pub fn log10f64(x: f64) -> f64;
 
     pub fn log2f32(x: f32) -> f32;
     pub fn log2f64(x: f64) -> f64;
+    */
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fma.f32"]
     pub fn fmaf32(a: f32, b: f32, c: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fma.f64"]
     pub fn fmaf64(a: f64, b: f64, c: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fmuladd.f32"]
+    pub fn fmuladdf32(a: f32, b: f32, c: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fmuladd.f64"]
+    pub fn fmuladdf64(a: f64, b: f64, c: f64) -> f64;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fabs.f32"]
     pub fn fabsf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.fabs.f64"]
     pub fn fabsf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.floor.f32"]
     pub fn floorf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.floor.f64"]
     pub fn floorf64(x: f64) -> f64;
 
+    /* NOTE: Needs LLVM 3.3
+    #[rust_stack]
+    #[link_name="llvm.ceil.f32"]
     pub fn ceilf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[link_name="llvm.ceil.f64"]
     pub fn ceilf64(x: f64) -> f64;
 
+    #[rust_stack]
+    #[link_name="llvm.trunc.f32"]
     pub fn truncf32(x: f32) -> f32;
+
+    #[rust_stack]
+    #[link_name="llvm.trunc.f64"]
     pub fn truncf64(x: f64) -> f64;
+    */
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.ctpop.i8"]
     pub fn ctpop8(x: i8) -> i8;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.ctpop.i16"]
     pub fn ctpop16(x: i16) -> i16;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.ctpop.i32"]
     pub fn ctpop32(x: i32) -> i32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.ctpop.i64"]
     pub fn ctpop64(x: i64) -> i64;
 
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.ctlz.i8"]
     pub fn ctlz8(x: i8) -> i8;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.ctlz.i16"]
     pub fn ctlz16(x: i16) -> i16;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.ctlz.i32"]
     pub fn ctlz32(x: i32) -> i32;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.ctlz.i64"]
     pub fn ctlz64(x: i64) -> i64;
 
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.cttz.i8"]
     pub fn cttz8(x: i8) -> i8;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.cttz.i16"]
     pub fn cttz16(x: i16) -> i16;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.cttz.i32"]
     pub fn cttz32(x: i32) -> i32;
+
+    #[cfg(not(stage0))]
+    #[rust_stack]
+    #[inline(always)]
+    #[constant_arg="1, i1, 0"]
+    #[link_name="llvm.cttz.i64"]
     pub fn cttz64(x: i64) -> i64;
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.bswap.i16"]
     pub fn bswap16(x: i16) -> i16;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.bswap.i32"]
     pub fn bswap32(x: i32) -> i32;
+
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.bswap.i64"]
     pub fn bswap64(x: i64) -> i64;
-}
 
+    #[rust_stack]
+    #[inline(always)]
+    #[link_name="llvm.readcyclecounter"]
+    pub fn read_cyclecounter() -> i64;
+}