diff --git a/builtins.cpp b/builtins.cpp
index f3a0cf59..886eec15 100644
--- a/builtins.cpp
+++ b/builtins.cpp
@@ -581,6 +581,15 @@ lSetInternalFunctions(llvm::Module *module) {
         "__stdlib_sinf",
         "__stdlib_tan",
         "__stdlib_tanf",
+        "__svml_sin",
+        "__svml_cos",
+        "__svml_sincos",
+        "__svml_tan",
+        "__svml_atan",
+        "__svml_atan2",
+        "__svml_exp",
+        "__svml_log",
+        "__svml_pow",
         "__undef_uniform",
         "__undef_varying",
         "__vec4_add_float",
@@ -1050,6 +1059,8 @@ DefineStdlib(SymbolTable *symbolTable, llvm::LLVMContext *ctx, llvm::Module *mod
                        symbolTable);
     lDefineConstantInt("__math_lib_ispc_fast", (int)Globals::Math_ISPCFast,
                        module, symbolTable);
+    lDefineConstantInt("__math_lib_svml", (int)Globals::Math_SVML, module,
+                       symbolTable);
     lDefineConstantInt("__math_lib_system", (int)Globals::Math_System, module,
                        symbolTable);
     lDefineConstantIntFunc("__fast_masked_vload", (int)g->opt.fastMaskedVload,
diff --git a/builtins/target-avx-x2.ll b/builtins/target-avx-x2.ll
index 8fb2e427..d9e0322b 100644
--- a/builtins/target-avx-x2.ll
+++ b/builtins/target-avx-x2.ll
@@ -134,6 +134,23 @@ define <16 x float> @__sqrt_varying_float(<16 x float>) nounwind readonly always
   ret <16 x float> %call
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; svml
+
+; FIXME: need either to wire these up to the 8-wide SVML entrypoints,
+; or, use the macro to call the 4-wide ones 4x with our 16-wide
+; vectors...
+
+declare <16 x float> @__svml_sin(<16 x float>)
+declare <16 x float> @__svml_cos(<16 x float>)
+declare void @__svml_sincos(<16 x float>, <16 x float> *, <16 x float> *)
+declare <16 x float> @__svml_tan(<16 x float>)
+declare <16 x float> @__svml_atan(<16 x float>)
+declare <16 x float> @__svml_atan2(<16 x float>, <16 x float>)
+declare <16 x float> @__svml_exp(<16 x float>)
+declare <16 x float> @__svml_log(<16 x float>)
+declare <16 x float> @__svml_pow(<16 x float>, <16 x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-avx.ll b/builtins/target-avx.ll
index adaed9ba..90e2f3ac 100644
--- a/builtins/target-avx.ll
+++ b/builtins/target-avx.ll
@@ -134,6 +134,23 @@ define <8 x float> @__sqrt_varying_float(<8 x float>) nounwind readonly alwaysin
   ret <8 x float> %call
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; svml
+
+; FIXME: need either to wire these up to the 8-wide SVML entrypoints,
+; or, use the macro to call the 4-wide ones twice with our 8-wide
+; vectors...
+
+declare <8 x float> @__svml_sin(<8 x float>)
+declare <8 x float> @__svml_cos(<8 x float>)
+declare void @__svml_sincos(<8 x float>, <8 x float> *, <8 x float> *)
+declare <8 x float> @__svml_tan(<8 x float>)
+declare <8 x float> @__svml_atan(<8 x float>)
+declare <8 x float> @__svml_atan2(<8 x float>, <8 x float>)
+declare <8 x float> @__svml_exp(<8 x float>)
+declare <8 x float> @__svml_log(<8 x float>)
+declare <8 x float> @__svml_pow(<8 x float>, <8 x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-generic-1.ll b/builtins/target-generic-1.ll
index 3472c207..31ebcdd5 100644
--- a/builtins/target-generic-1.ll
+++ b/builtins/target-generic-1.ll
@@ -647,6 +647,104 @@ define  <1 x float> @__rsqrt_varying_float(<1 x float> %v) nounwind readonly alw
   
 }
 
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; svml stuff
+
+define  <1 x float> @__svml_sin(<1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_sinf4(<1 x float> %0)
+  ;ret <1 x float> %ret
+  ;%r = extractelement <1 x float> %0, i32 0
+  ;%s = call float @llvm.sin.f32(float %r)
+  ;%rv = insertelement <1 x float> undef, float %r, i32 0
+  ;ret <1 x float> %rv
+  unary1to1(float,@llvm.sin.f32)
+   
+}
+
+define  <1 x float> @__svml_cos(<1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_cosf4(<1 x float> %0)
+  ;ret <1 x float> %ret
+  ;%r = extractelement <1 x float> %0, i32 0
+  ;%s = call float @llvm.cos.f32(float %r)
+  ;%rv = insertelement <1 x float> undef, float %r, i32 0
+  ;ret <1 x float> %rv
+  unary1to1(float, @llvm.cos.f32)
+
+}
+
+define  void @__svml_sincos(<1 x float>, <1 x float> *, <1 x float> *) nounwind readnone alwaysinline {
+;  %s = call <1 x float> @__svml_sincosf4(<1 x float> * %2, <1 x float> %0)
+;  store <1 x float> %s, <1 x float> * %1
+;  ret void
+   %sin = call <1 x float> @__svml_sin (<1 x float> %0)
+   %cos = call <1 x float> @__svml_cos (<1 x float> %0)
+   store <1 x float> %sin, <1 x float> * %1
+   store <1 x float> %cos, <1 x float> * %2
+   ret void
+}
+
+define  <1 x float> @__svml_tan(<1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_tanf4(<1 x float> %0)
+  ;ret <1 x float> %ret
+  ;%r = extractelement <1 x float> %0, i32 0
+  ;%s = call float @llvm_tan_f32(float %r)
+  ;%rv = insertelement <1 x float> undef, float %r, i32 0
+  ;ret <1 x float> %rv
+  ;unasry1to1(float, @llvm.tan.f32)
+  ; UNSUPPORTED!
+  ret <1 x float > %0
+}
+
+define  <1 x float> @__svml_atan(<1 x float>) nounwind readnone alwaysinline {
+;  %ret = call <1 x float> @__svml_atanf4(<1 x float> %0)
+;  ret <1 x float> %ret
+  ;%r = extractelement <1 x float> %0, i32 0
+  ;%s = call float @llvm_atan_f32(float %r)
+  ;%rv = insertelement <1 x float> undef, float %r, i32 0
+  ;ret <1 x float> %rv
+  ;unsary1to1(float,@llvm.atan.f32)
+  ;UNSUPPORTED!
+  ret <1 x float > %0
+
+}
+
+define  <1 x float> @__svml_atan2(<1 x float>, <1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_atan2f4(<1 x float> %0, <1 x float> %1)
+  ;ret <1 x float> %ret
+  ;%y = extractelement <1 x float> %0, i32 0
+  ;%x = extractelement <1 x float> %1, i32 0
+  ;%q = fdiv float %y, %x
+  ;%a = call float @llvm.atan.f32 (float %q)
+  ;%rv = insertelement <1 x float> undef, float %a, i32 0
+  ;ret <1 x float> %rv
+  ; UNSUPPORTED!
+  ret <1 x float > %0
+}
+
+define  <1 x float> @__svml_exp(<1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_expf4(<1 x float> %0)
+  ;ret <1 x float> %ret
+  unary1to1(float, @llvm.exp.f32)
+}
+
+define  <1 x float> @__svml_log(<1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_logf4(<1 x float> %0)
+  ;ret <1 x float> %ret
+  unary1to1(float, @llvm.log.f32)
+}
+
+define  <1 x float> @__svml_pow(<1 x float>, <1 x float>) nounwind readnone alwaysinline {
+  ;%ret = call <1 x float> @__svml_powf4(<1 x float> %0, <1 x float> %1)
+  ;ret <1 x float> %ret
+  %r = extractelement <1 x float> %0, i32 0
+  %e  = extractelement <1 x float> %1, i32 0
+  %s = call float @llvm.pow.f32(float %r,float %e)
+  %rv = insertelement <1 x float> undef, float %s, i32 0
+  ret <1 x float> %rv
+
+}
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-generic-common.ll b/builtins/target-generic-common.ll
index c683ff45..2896c6b1 100644
--- a/builtins/target-generic-common.ll
+++ b/builtins/target-generic-common.ll
@@ -202,6 +202,22 @@ declare i64 @__count_trailing_zeros_i64(i64) nounwind readnone
 declare i32 @__count_leading_zeros_i32(i32) nounwind readnone
 declare i64 @__count_leading_zeros_i64(i64) nounwind readnone
 
+;; svml
+
+; FIXME: need either to wire these up to the 8-wide SVML entrypoints,
+; or, use the macro to call the 4-wide ones twice with our 8-wide
+; vectors...
+
+declare <WIDTH x float> @__svml_sin(<WIDTH x float>)
+declare <WIDTH x float> @__svml_cos(<WIDTH x float>)
+declare void @__svml_sincos(<WIDTH x float>, <WIDTH x float> *, <WIDTH x float> *)
+declare <WIDTH x float> @__svml_tan(<WIDTH x float>)
+declare <WIDTH x float> @__svml_atan(<WIDTH x float>)
+declare <WIDTH x float> @__svml_atan2(<WIDTH x float>, <WIDTH x float>)
+declare <WIDTH x float> @__svml_exp(<WIDTH x float>)
+declare <WIDTH x float> @__svml_log(<WIDTH x float>)
+declare <WIDTH x float> @__svml_pow(<WIDTH x float>, <WIDTH x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; reductions
 
diff --git a/builtins/target-neon-common.ll b/builtins/target-neon-common.ll
index f892a0a1..696b0748 100644
--- a/builtins/target-neon-common.ll
+++ b/builtins/target-neon-common.ll
@@ -313,6 +313,19 @@ define void @__masked_store_blend_i64(<WIDTH x i64>* nocapture %ptr,
   ret void
 }
 
+;; yuck.  We need declarations of these, even though we shouldnt ever
+;; actually generate calls to them for the NEON target...
+
+declare <WIDTH x float> @__svml_sin(<WIDTH x float>)
+declare <WIDTH x float> @__svml_cos(<WIDTH x float>)
+declare void @__svml_sincos(<WIDTH x float>, <WIDTH x float> *, <WIDTH x float> *)
+declare <WIDTH x float> @__svml_tan(<WIDTH x float>)
+declare <WIDTH x float> @__svml_atan(<WIDTH x float>)
+declare <WIDTH x float> @__svml_atan2(<WIDTH x float>, <WIDTH x float>)
+declare <WIDTH x float> @__svml_exp(<WIDTH x float>)
+declare <WIDTH x float> @__svml_log(<WIDTH x float>)
+declare <WIDTH x float> @__svml_pow(<WIDTH x float>, <WIDTH x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; gather
 
diff --git a/builtins/target-sse2-x2.ll b/builtins/target-sse2-x2.ll
index 057ea98f..da22a66c 100644
--- a/builtins/target-sse2-x2.ll
+++ b/builtins/target-sse2-x2.ll
@@ -102,6 +102,92 @@ define <8 x float> @__sqrt_varying_float(<8 x float>) nounwind readonly alwaysin
   ret <8 x float> %call
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; svml stuff
+
+declare <4 x float> @__svml_sinf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_cosf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_sincosf4(<4 x float> *, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_tanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atan2f4(<4 x float>, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_expf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_logf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_powf4(<4 x float>, <4 x float>) nounwind readnone
+
+
+define <8 x float> @__svml_sin(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_sinf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_cos(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_cosf4, %0)
+  ret <8 x float> %ret
+}
+
+define void @__svml_sincos(<8 x float>, <8 x float> *,
+                                    <8 x float> *) nounwind readnone alwaysinline {
+  ; call svml_sincosf4 two times with the two 4-wide sub-vectors
+  %a = shufflevector <8 x float> %0, <8 x float> undef,
+         <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %b = shufflevector <8 x float> %0, <8 x float> undef,
+         <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+
+  %cospa = alloca <4 x float>
+  %sa = call <4 x float> @__svml_sincosf4(<4 x float> * %cospa, <4 x float> %a)
+
+  %cospb = alloca <4 x float>
+  %sb = call <4 x float> @__svml_sincosf4(<4 x float> * %cospb, <4 x float> %b)
+
+  %sin = shufflevector <4 x float> %sa, <4 x float> %sb,
+         <8 x i32> <i32 0, i32 1, i32 2, i32 3,
+                    i32 4, i32 5, i32 6, i32 7>
+  store <8 x float> %sin, <8 x float> * %1
+
+  %cosa = load <4 x float> * %cospa
+  %cosb = load <4 x float> * %cospb
+  %cos = shufflevector <4 x float> %cosa, <4 x float> %cosb,
+         <8 x i32> <i32 0, i32 1, i32 2, i32 3,
+                    i32 4, i32 5, i32 6, i32 7>
+  store <8 x float> %cos, <8 x float> * %2
+
+  ret void
+}
+
+define <8 x float> @__svml_tan(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_tanf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_atan(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_atanf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_atan2(<8 x float>,
+                                          <8 x float>) nounwind readnone alwaysinline {
+  binary4to8(ret, float, @__svml_atan2f4, %0, %1)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_exp(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_expf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_log(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_logf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_pow(<8 x float>,
+                                        <8 x float>) nounwind readnone alwaysinline {
+  binary4to8(ret, float, @__svml_powf4, %0, %1)
+  ret <8 x float> %ret
+}
+
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-sse2.ll b/builtins/target-sse2.ll
index e0a5c3d5..a6b206b6 100644
--- a/builtins/target-sse2.ll
+++ b/builtins/target-sse2.ll
@@ -493,6 +493,66 @@ define <4 x float> @__sqrt_varying_float(<4 x float>) nounwind readonly alwaysin
   ret <4 x float> %call
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; svml stuff
+
+declare <4 x float> @__svml_sinf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_cosf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_sincosf4(<4 x float> *, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_tanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atan2f4(<4 x float>, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_expf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_logf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_powf4(<4 x float>, <4 x float>) nounwind readnone
+
+
+define <4 x float> @__svml_sin(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_sinf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_cos(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_cosf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define void @__svml_sincos(<4 x float>, <4 x float> *, <4 x float> *) nounwind readnone alwaysinline {
+  %s = call <4 x float> @__svml_sincosf4(<4 x float> * %2, <4 x float> %0)
+  store <4 x float> %s, <4 x float> * %1
+  ret void
+}
+
+define <4 x float> @__svml_tan(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_tanf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_atan(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_atanf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_atan2(<4 x float>, <4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_atan2f4(<4 x float> %0, <4 x float> %1)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_exp(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_expf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_log(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_logf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_pow(<4 x float>, <4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_powf4(<4 x float> %0, <4 x float> %1)
+  ret <4 x float> %ret
+}
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-sse4-16.ll b/builtins/target-sse4-16.ll
index b4772552..d7f3833d 100644
--- a/builtins/target-sse4-16.ll
+++ b/builtins/target-sse4-16.ll
@@ -205,6 +205,21 @@ define <8 x double> @__max_varying_double(<8 x double>, <8 x double>) nounwind r
   ret <8 x double> %ret
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; svml
+
+; FIXME
+
+declare <8 x float> @__svml_sin(<8 x float>)
+declare <8 x float> @__svml_cos(<8 x float>)
+declare void @__svml_sincos(<8 x float>, <8 x float> *, <8 x float> *)
+declare <8 x float> @__svml_tan(<8 x float>)
+declare <8 x float> @__svml_atan(<8 x float>)
+declare <8 x float> @__svml_atan2(<8 x float>, <8 x float>)
+declare <8 x float> @__svml_exp(<8 x float>)
+declare <8 x float> @__svml_log(<8 x float>)
+declare <8 x float> @__svml_pow(<8 x float>, <8 x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ; horizontal ops / reductions
 
diff --git a/builtins/target-sse4-8.ll b/builtins/target-sse4-8.ll
index a75d8e3a..fd4b74d7 100644
--- a/builtins/target-sse4-8.ll
+++ b/builtins/target-sse4-8.ll
@@ -217,6 +217,21 @@ define <16 x double> @__max_varying_double(<16 x double>, <16 x double>) nounwin
   ret <16 x double> %ret
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; svml
+
+; FIXME
+
+declare <16 x float> @__svml_sin(<16 x float>)
+declare <16 x float> @__svml_cos(<16 x float>)
+declare void @__svml_sincos(<16 x float>, <16 x float> *, <16 x float> *)
+declare <16 x float> @__svml_tan(<16 x float>)
+declare <16 x float> @__svml_atan(<16 x float>)
+declare <16 x float> @__svml_atan2(<16 x float>, <16 x float>)
+declare <16 x float> @__svml_exp(<16 x float>)
+declare <16 x float> @__svml_log(<16 x float>)
+declare <16 x float> @__svml_pow(<16 x float>, <16 x float>)
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ; horizontal ops / reductions
 
diff --git a/builtins/target-sse4-x2.ll b/builtins/target-sse4-x2.ll
index 897a09eb..a7faddb3 100644
--- a/builtins/target-sse4-x2.ll
+++ b/builtins/target-sse4-x2.ll
@@ -102,6 +102,92 @@ define <8 x float> @__sqrt_varying_float(<8 x float>) nounwind readonly alwaysin
   ret <8 x float> %call
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; svml stuff
+
+declare <4 x float> @__svml_sinf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_cosf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_sincosf4(<4 x float> *, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_tanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atan2f4(<4 x float>, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_expf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_logf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_powf4(<4 x float>, <4 x float>) nounwind readnone
+
+
+define <8 x float> @__svml_sin(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_sinf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_cos(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_cosf4, %0)
+  ret <8 x float> %ret
+}
+
+define void @__svml_sincos(<8 x float>, <8 x float> *,
+                                    <8 x float> *) nounwind readnone alwaysinline {
+  ; call svml_sincosf4 two times with the two 4-wide sub-vectors
+  %a = shufflevector <8 x float> %0, <8 x float> undef,
+         <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %b = shufflevector <8 x float> %0, <8 x float> undef,
+         <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+
+  %cospa = alloca <4 x float>
+  %sa = call <4 x float> @__svml_sincosf4(<4 x float> * %cospa, <4 x float> %a)
+
+  %cospb = alloca <4 x float>
+  %sb = call <4 x float> @__svml_sincosf4(<4 x float> * %cospb, <4 x float> %b)
+
+  %sin = shufflevector <4 x float> %sa, <4 x float> %sb,
+         <8 x i32> <i32 0, i32 1, i32 2, i32 3,
+                    i32 4, i32 5, i32 6, i32 7>
+  store <8 x float> %sin, <8 x float> * %1
+
+  %cosa = load <4 x float> * %cospa
+  %cosb = load <4 x float> * %cospb
+  %cos = shufflevector <4 x float> %cosa, <4 x float> %cosb,
+         <8 x i32> <i32 0, i32 1, i32 2, i32 3,
+                    i32 4, i32 5, i32 6, i32 7>
+  store <8 x float> %cos, <8 x float> * %2
+
+  ret void
+}
+
+define <8 x float> @__svml_tan(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_tanf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_atan(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_atanf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_atan2(<8 x float>,
+                                          <8 x float>) nounwind readnone alwaysinline {
+  binary4to8(ret, float, @__svml_atan2f4, %0, %1)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_exp(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_expf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_log(<8 x float>) nounwind readnone alwaysinline {
+  unary4to8(ret, float, @__svml_logf4, %0)
+  ret <8 x float> %ret
+}
+
+define <8 x float> @__svml_pow(<8 x float>,
+                                        <8 x float>) nounwind readnone alwaysinline {
+  binary4to8(ret, float, @__svml_powf4, %0, %1)
+  ret <8 x float> %ret
+}
+
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; float min/max
 
diff --git a/builtins/target-sse4.ll b/builtins/target-sse4.ll
index 5429b461..e05b865f 100644
--- a/builtins/target-sse4.ll
+++ b/builtins/target-sse4.ll
@@ -206,6 +206,66 @@ define <4 x double> @__max_varying_double(<4 x double>, <4 x double>) nounwind r
   ret <4 x double> %ret
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+; svml stuff
+
+declare <4 x float> @__svml_sinf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_cosf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_sincosf4(<4 x float> *, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_tanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atanf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_atan2f4(<4 x float>, <4 x float>) nounwind readnone
+declare <4 x float> @__svml_expf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_logf4(<4 x float>) nounwind readnone
+declare <4 x float> @__svml_powf4(<4 x float>, <4 x float>) nounwind readnone
+
+
+define <4 x float> @__svml_sin(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_sinf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_cos(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_cosf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define void @__svml_sincos(<4 x float>, <4 x float> *, <4 x float> *) nounwind readnone alwaysinline {
+  %s = call <4 x float> @__svml_sincosf4(<4 x float> * %2, <4 x float> %0)
+  store <4 x float> %s, <4 x float> * %1
+  ret void
+}
+
+define <4 x float> @__svml_tan(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_tanf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_atan(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_atanf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_atan2(<4 x float>, <4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_atan2f4(<4 x float> %0, <4 x float> %1)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_exp(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_expf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_log(<4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_logf4(<4 x float> %0)
+  ret <4 x float> %ret
+}
+
+define <4 x float> @__svml_pow(<4 x float>, <4 x float>) nounwind readnone alwaysinline {
+  %ret = call <4 x float> @__svml_powf4(<4 x float> %0, <4 x float> %1)
+  ret <4 x float> %ret
+}
+
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ; horizontal ops / reductions
 
diff --git a/docs/ispc.rst b/docs/ispc.rst
index 476046e8..ff07f6d8 100644
--- a/docs/ispc.rst
+++ b/docs/ispc.rst
@@ -3333,6 +3333,9 @@ for this argument.
   approximately 1.45e-6 over the range -10pi to 10pi.)
 * ``fast``: more efficient but lower accuracy versions of the default ``ispc``
   implementations.
+* ``svml``: use Intel "Short Vector Math Library".  Use
+  ``icc`` to link your final executable so that the appropriate libraries
+  are linked.
 * ``system``: use the system's math library.  On many systems, these
   functions are more accurate than both of ``ispc``'s implementations.
   Using these functions may be quite
diff --git a/ispc.h b/ispc.h
index fc78e415..4804832f 100644
--- a/ispc.h
+++ b/ispc.h
@@ -488,7 +488,7 @@ struct Globals {
 
     /** There are a number of math libraries that can be used for
         transcendentals and the like during program compilation. */
-    enum MathLib { Math_ISPC, Math_ISPCFast, Math_System };
+    enum MathLib { Math_ISPC, Math_ISPCFast, Math_SVML, Math_System };
     MathLib mathLib;
 
     /** Records whether the ispc standard library should be made available
diff --git a/main.cpp b/main.cpp
index 61c62042..21a47de8 100644
--- a/main.cpp
+++ b/main.cpp
@@ -112,6 +112,7 @@ usage(int ret) {
     printf("    [--math-lib=<option>]\t\tSelect math library\n");
     printf("        default\t\t\t\tUse ispc's built-in math functions\n");
     printf("        fast\t\t\t\tUse high-performance but lower-accuracy math functions\n");
+    printf("        svml\t\t\t\tUse the Intel(r) SVML math libraries\n");
     printf("        system\t\t\t\tUse the system's math library (*may be quite slow*)\n");
     printf("    [-MMM <filename>\t\t\t\tWrite #include dependencies to given file.\n");
     printf("    [--nostdlib]\t\t\tDon't make the ispc standard library available\n");
@@ -408,6 +409,8 @@ int main(int Argc, char *Argv[]) {
                 g->mathLib = Globals::Math_ISPC;
             else if (!strcmp(lib, "fast"))
                 g->mathLib = Globals::Math_ISPCFast;
+            else if (!strcmp(lib, "svml"))
+                g->mathLib = Globals::Math_SVML;
             else if (!strcmp(lib, "system"))
                 g->mathLib = Globals::Math_System;
             else {
diff --git a/stdlib.ispc b/stdlib.ispc
index f7d135dd..e4f8844f 100644
--- a/stdlib.ispc
+++ b/stdlib.ispc
@@ -2179,7 +2179,10 @@ static inline uniform float frexp(uniform float x, uniform int * uniform pw2) {
 
 __declspec(safe)
 static inline float sin(float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        return __svml_sin(x_full);
+    }
+    else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_sinf(extract(x_full, i));
@@ -2238,7 +2241,8 @@ static inline float sin(float x_full) {
 
 __declspec(safe)
 static inline uniform float sin(uniform float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_sinf(x_full);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2309,7 +2313,8 @@ static inline float asin(float x) {
     bool isnan = (x > 1);
 
     float v;
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml ||
+        __math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_asinf(extract(x, i));
@@ -2365,7 +2370,8 @@ static inline uniform float asin(uniform float x) {
     uniform bool isnan = (x > 1);
 
     uniform float v;
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml ||
+        __math_lib == __math_lib_system) {
         return __stdlib_asinf(x);
     }
     else if (__math_lib == __math_lib_ispc)
@@ -2410,7 +2416,10 @@ static inline uniform float asin(uniform float x) {
 
 __declspec(safe)
 static inline float cos(float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        return __svml_cos(x_full);
+    }
+    else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_cosf(extract(x_full, i));
@@ -2468,7 +2477,8 @@ static inline float cos(float x_full) {
 
 __declspec(safe)
 static inline uniform float cos(uniform float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_cosf(x_full);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2545,7 +2555,10 @@ static inline uniform float acos(uniform float v) {
 __declspec(safe)
 static inline void sincos(float x_full, varying float * uniform sin_result, 
                           varying float * uniform cos_result) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        __svml_sincos(x_full, sin_result, cos_result);
+    }
+    else if (__math_lib == __math_lib_system) {
         foreach_active (i) {
             uniform float s, c;
             __stdlib_sincosf(extract(x_full, i), &s, &c);
@@ -2612,7 +2625,8 @@ static inline void sincos(float x_full, varying float * uniform sin_result,
 __declspec(safe)
 static inline void sincos(uniform float x_full, uniform float * uniform sin_result,
                           uniform float * uniform cos_result) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         __stdlib_sincosf(x_full, sin_result, cos_result);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2673,7 +2687,10 @@ static inline void sincos(uniform float x_full, uniform float * uniform sin_resu
 
 __declspec(safe)
 static inline float tan(float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        return __svml_tan(x_full);
+    }
+    else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_tanf(extract(x_full, i));
@@ -2749,7 +2766,8 @@ static inline float tan(float x_full) {
 
 __declspec(safe)
 static inline uniform float tan(uniform float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_tanf(x_full);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2820,7 +2838,10 @@ static inline uniform float tan(uniform float x_full) {
 
 __declspec(safe)
 static inline float atan(float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        return __svml_atan(x_full);
+    }
+    else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_atanf(extract(x_full, i));
@@ -2868,7 +2889,8 @@ static inline float atan(float x_full) {
 
 __declspec(safe)
 static inline uniform float atan(uniform float x_full) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_atanf(x_full);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2911,7 +2933,10 @@ static inline uniform float atan(uniform float x_full) {
 
 __declspec(safe)
 static inline float atan2(float y, float x) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_svml) {
+        return __svml_atan2(y, x);
+    }
+    else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
             uniform float r = __stdlib_atan2f(extract(y, i), extract(x, i));
@@ -2947,7 +2972,8 @@ static inline float atan2(float y, float x) {
 
 __declspec(safe)
 static inline uniform float atan2(uniform float y, uniform float x) {
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_atan2f(y, x);
     }
     else if (__math_lib == __math_lib_ispc || 
@@ -2970,6 +2996,9 @@ static inline float exp(float x_full) {
     if (__have_native_transcendentals) {
         return __exp_varying_float(x_full);
     }
+    else if (__math_lib == __math_lib_svml) {
+        return __svml_exp(x_full);
+    }
     else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
@@ -3049,7 +3078,8 @@ static inline uniform float exp(uniform float x_full) {
     if (__have_native_transcendentals) {
         return __exp_uniform_float(x_full);
     }
-    else if (__math_lib == __math_lib_system) {
+    else if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_expf(x_full);
     }
     else if (__math_lib == __math_lib_ispc_fast) {
@@ -3173,6 +3203,9 @@ static inline float log(float x_full) {
     if (__have_native_transcendentals) {
         return __log_varying_float(x_full);
     }
+    else if (__math_lib == __math_lib_svml) {
+        return __svml_log(x_full);
+    }
     else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
@@ -3261,7 +3294,8 @@ static inline uniform float log(uniform float x_full) {
     if (__have_native_transcendentals) {
         return __log_uniform_float(x_full);
     }
-    else if (__math_lib == __math_lib_system) {
+    else if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_logf(x_full);
     }
     else if (__math_lib == __math_lib_ispc_fast) {
@@ -3344,6 +3378,9 @@ static inline float pow(float a, float b) {
     if (__have_native_transcendentals) {
         return __pow_varying_float(a, b);
     }
+    else if (__math_lib == __math_lib_svml) {
+        return __svml_pow(a, b);
+    }
     else if (__math_lib == __math_lib_system) {
         float ret;
         foreach_active (i) {
@@ -3363,7 +3400,8 @@ static inline uniform float pow(uniform float a, uniform float b) {
     if (__have_native_transcendentals) {
         return __pow_uniform_float(a, b);
     }
-    if (__math_lib == __math_lib_system) {
+    if (__math_lib == __math_lib_system ||
+        __math_lib == __math_lib_svml) {
         return __stdlib_powf(a, b);
     }
     else if (__math_lib == __math_lib_ispc ||