use LLVM intrinsic for sqrt. helps JuliaLang#6112

pwl · Mar 13, 2014 · 244ec92 · 244ec92
1 parent 9785017
commit 244ec92
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 4 deletions.
diff --git a/base/boot.jl b/base/boot.jl
@@ -152,7 +152,7 @@ export
  #checked_smul, checked_ssub, checked_uadd, checked_umul, checked_usub,
  #nan_dom_err, copysign_float, ctlz_int, ctpop_int, cttz_int,
  #div_float, eq_float, eq_int, eqfsi64, eqfui64, flipsign_int, select_value,
- #fpext64, fpiseq, fpislt, fpsiround, fpuiround, fptosi, fptoui,
+ #sqrt_llvm, fpext64, fpiseq, fpislt, fpsiround, fpuiround, fptosi, fptoui,
  #fptrunc32, le_float, lefsi64, lefui64, lesif64,
  #leuif64, lshr_int, lt_float, ltfsi64, ltfui64, ltsif64, ltuif64, mul_float,
  #mul_int, ne_float, ne_int, neg_float, neg_int, not_int, or_int, rem_float,

diff --git a/base/math.jl b/base/math.jl
@@ -22,7 +22,7 @@ import Base: log, exp, sin, cos, tan, sinh, cosh, tanh, asin,
  acos, atan, asinh, acosh, atanh, sqrt, log2, log10,
  max, min, minmax, ceil, floor, trunc, round, ^, exp2, exp10
 
-import Core.Intrinsics.nan_dom_err
+import Core.Intrinsics: nan_dom_err, sqrt_llvm, box, unbox
 
 # non-type specific math functions
 
@@ -272,7 +272,7 @@ exp10(x::Integer) = exp10(float(x))
 
 # functions that return NaN on non-NaN argument for domain error
 for f in (:sin, :cos, :tan, :asin, :acos, :acosh, :atanh, :log, :log2, :log10,
- :lgamma, :sqrt, :log1p)
+ :lgamma, :log1p)
  @eval begin
  ($f)(x::Float64) = nan_dom_err(ccall(($(string(f)),libm), Float64, (Float64,), x), x)
  ($f)(x::Float32) = nan_dom_err(ccall(($(string(f,"f")),libm), Float32, (Float32,), x), x)
@@ -281,6 +281,11 @@ for f in (:sin, :cos, :tan, :asin, :acos, :acosh, :atanh, :log, :log2, :log10,
  end
 end
 
+sqrt(x::Float64) = nan_dom_err(box(Float64,sqrt_llvm(unbox(Float64,x))), x)
+sqrt(x::Float32) = nan_dom_err(box(Float32,sqrt_llvm(unbox(Float32,x))), x)
+sqrt(x::Real) = sqrt(float(x))
+@vectorize_1arg Number sqrt
+
 for f in (:ceil, :trunc, :significand) # :rint, :nearbyint
  @eval begin
  ($f)(x::Float64) = ccall(($(string(f)),libm), Float64, (Float64,), x)

diff --git a/src/intrinsics.cpp b/src/intrinsics.cpp
@@ -34,6 +34,7 @@ namespace JL_I {
  nan_dom_err,
  // functions
  abs_float, copysign_float, flipsign_int, select_value,
+ sqrt_llvm,
  // pointer access
  pointerref, pointerset, pointertoref,
  // c interface
@@ -1243,6 +1244,12 @@ static Value *emit_intrinsic(intrinsic f, jl_value_t **args, size_t nargs,
  HANDLE(jl_alloca,1) {
  return builder.CreateAlloca(IntegerType::get(jl_LLVMContext, 8),JL_INT(x));
  }
+ HANDLE(sqrt_llvm,1) {
+ x = FP(x);
+ return builder.CreateCall(Intrinsic::getDeclaration(jl_Module, Intrinsic::sqrt,
+ ArrayRef<Type*>(x->getType())),
+ x);
+ }
  default:
  assert(false);
  }
@@ -1321,7 +1328,7 @@ extern "C" void jl_init_intrinsic_functions(void)
  ADD_I(uitofp); ADD_I(sitofp);
  ADD_I(fptrunc); ADD_I(fpext);
  ADD_I(abs_float); ADD_I(copysign_float);
- ADD_I(flipsign_int); ADD_I(select_value);
+ ADD_I(flipsign_int); ADD_I(select_value); ADD_I(sqrt_llvm);
  ADD_I(pointerref); ADD_I(pointerset); ADD_I(pointertoref);
  ADD_I(checked_sadd); ADD_I(checked_uadd);
  ADD_I(checked_ssub); ADD_I(checked_usub);