diff --git a/include/private/SkVx.h b/include/private/SkVx.h
index f6fa8d5237..2b325e8c71 100644
--- a/include/private/SkVx.h
+++ b/include/private/SkVx.h
@@ -558,9 +558,6 @@ SIN Vec<N,int> lrint(const Vec<N,float>& x) {
                 lrint(x.hi));
 }
 
-// TODO: new-style platform specializations for rcp() / rsqrt()?
-SIN Vec<N,float>   rcp(const Vec<N,float>& x) { return 1/x; }
-SIN Vec<N,float> rsqrt(const Vec<N,float>& x) { return rcp(sqrt(x)); }
 SIN Vec<N,float> fract(const Vec<N,float>& x) { return x - floor(x); }
 
 // The default logic for to_half/from_half is borrowed from skcms,
@@ -673,37 +670,6 @@ SIN Vec<N,uint8_t> approx_scale(const Vec<N,uint8_t>& x, const Vec<N,uint8_t>& y
     }
 #endif
 
-#if !defined(SKNX_NO_SIMD)
-
-    // Platform-specific specializations and overloads can now drop in here.
-
-    #if defined(__AVX__)
-        SI Vec<8,float> rsqrt(const Vec<8,float>& x) {
-            return bit_pun<Vec<8,float>>(_mm256_rsqrt_ps(bit_pun<__m256>(x)));
-        }
-        SI Vec<8,float> rcp(const Vec<8,float>& x) {
-            return bit_pun<Vec<8,float>>(_mm256_rcp_ps(bit_pun<__m256>(x)));
-        }
-    #endif
-
-    #if defined(__SSE__)
-        SI Vec<4,float> rsqrt(const Vec<4,float>& x) {
-            return bit_pun<Vec<4,float>>(_mm_rsqrt_ps(bit_pun<__m128>(x)));
-        }
-        SI Vec<4,float> rcp(const Vec<4,float>& x) {
-            return bit_pun<Vec<4,float>>(_mm_rcp_ps(bit_pun<__m128>(x)));
-        }
-
-        SI Vec<2,float> rsqrt(const Vec<2,float>& x) {
-            return shuffle<0,1>(rsqrt(shuffle<0,1,0,1>(x)));
-        }
-        SI Vec<2,float>   rcp(const Vec<2,float>& x) {
-            return shuffle<0,1>(  rcp(shuffle<0,1,0,1>(x)));
-        }
-    #endif
-
-#endif // !defined(SKNX_NO_SIMD)
-
 }  // namespace skvx
 
 #undef SINTU
diff --git a/tests/SkVxTest.cpp b/tests/SkVxTest.cpp
index 3a07e023e1..60b815a484 100644
--- a/tests/SkVxTest.cpp
+++ b/tests/SkVxTest.cpp
@@ -29,24 +29,6 @@ using long2 = skvx::Vec<2,int64_t>;
 using long4 = skvx::Vec<4,int64_t>;
 using long8 = skvx::Vec<8,int64_t>;
 
-// These are unused, and just here so I can look at the disassembly.
-float2 Sqrt(float2 x) { return sqrt(x); }
-float4 Sqrt(float4 x) { return sqrt(x); }
-float8 Sqrt(float8 x) { return sqrt(x); }
-
-float4 RSqrt(float4 x) { return rsqrt(x); }
-float4   Rcp(float4 x) { return   rcp(x); }
-float4  Ceil(float4 x) { return  ceil(x); }
-float4 Floor(float4 x) { return floor(x); }
-float4 Trunc(float4 x) { return trunc(x); }
-float4 Round(float4 x) { return round(x); }
-float4   Abs(float4 x) { return   abs(x); }
-
-float4 Min(float4 x, float4 y) { return min(x,y); }
-float4 Max(float4 x, float4 y) { return max(x,y); }
-
-float4 IfThenElse(int4 c, float4 t, float4 e) { return if_then_else(c,t,e); }
-
 DEF_TEST(SkVx, r) {
     static_assert(sizeof(float2) ==  8, "");
     static_assert(sizeof(float4) == 16, "");
@@ -106,12 +88,7 @@ DEF_TEST(SkVx, r) {
 
     // TODO(mtklein): these tests could be made less loose.
     REPORTER_ASSERT(r, all( sqrt(float4{2,3,4,5}) < float4{2,2,3,3}));
-    REPORTER_ASSERT(r, all(  rcp(float4{2,3,4,5}) < float4{1.0f,0.5f,0.5f,0.3f}));
-    REPORTER_ASSERT(r, all(rsqrt(float4{2,3,4,5}) < float4{1.0f,1.0f,1.0f,0.5f}));
-
     REPORTER_ASSERT(r, all( sqrt(float2{2,3}) < float2{2,2}));
-    REPORTER_ASSERT(r, all(  rcp(float2{2,3}) < float2{1.0f,0.5f}));
-    REPORTER_ASSERT(r, all(rsqrt(float2{2,3}) < float2{1.0f,1.0f}));
 
     REPORTER_ASSERT(r, all(skvx::cast<int>(float4{-1.5f,0.5f,1.0f,1.5f}) == int4{-1,0,1,1}));