Support Float32 output from SkColorSpaceXform

src/opts/SkNx_sse.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkNx_sse_DEFINED
 #define SkNx_sse_DEFINED
 
@@ skipping 460 matching lines @@
 static inline void Sk4h_store4(void* dst, const Sk4h& r, const Sk4h& g, const Sk4h& b,
                                const Sk4h& a) {
     __m128i rg = _mm_unpacklo_epi16(r.fVec, g.fVec);
     __m128i ba = _mm_unpacklo_epi16(b.fVec, a.fVec);
     __m128i lo = _mm_unpacklo_epi32(rg, ba);
     __m128i hi = _mm_unpackhi_epi32(rg, ba);
     _mm_storeu_si128(((__m128i*) dst) + 0, lo);
     _mm_storeu_si128(((__m128i*) dst) + 1, hi);
 }
 
+static inline void Sk4f_load4(const void* ptr, Sk4f* r, Sk4f* g, Sk4f* b, Sk4f* a) {

[mtklein, 2016/09/14 20:30:36]

May want to call _MM_TRANSPOSE_PS, or crib off it:

https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_MM_TRANS&...

[msarett, 2016/09/14 21:09:43]

Cool!  Done.

+    __m128 v0 = _mm_loadu_ps(((float*)ptr) +  0),
+           v1 = _mm_loadu_ps(((float*)ptr) +  4),
+           v2 = _mm_loadu_ps(((float*)ptr) +  8),
+           v3 = _mm_loadu_ps(((float*)ptr) + 12);
+    __m128 rg01 = _mm_unpacklo_ps(v0, v1),   // r0 r1 g0 g1
+           ba01 = _mm_unpackhi_ps(v0, v1),   // b0 b1 a0 a1
+           rg23 = _mm_unpacklo_ps(v2, v3),   // r2 r3 g2 g3
+           ba23 = _mm_unpacklo_ps(v2, v3);   // b2 b3 a2 a3
+    *r = _mm_shuffle_ps(rg01, rg23, 0x88);   // 00 01 00 01
+    *g = _mm_shuffle_ps(rg01, rg23, 0xEE);   // 10 11 10 11
+    *b = _mm_shuffle_ps(ba01, ba23, 0x88);   // 00 01 00 01
+    *a = _mm_shuffle_ps(ba01, ba23, 0xEE);   // 10 11 10 11
+}
+
+static inline void Sk4f_store4(void* dst, const Sk4f& r, const Sk4f& g, const Sk4f& b,
+                               const Sk4f& a) {
+    __m128 rg01 = _mm_unpacklo_ps(r.fVec, g.fVec);
+    __m128 rg23 = _mm_unpackhi_ps(r.fVec, g.fVec);
+    __m128 ba01 = _mm_unpacklo_ps(b.fVec, a.fVec);
+    __m128 ba23 = _mm_unpacklo_ps(b.fVec, a.fVec);
+    __m128 v0 = _mm_shuffle_ps(rg01, ba01, 0x88);
+    __m128 v1 = _mm_shuffle_ps(rg01, ba01, 0xEE);
+    __m128 v2 = _mm_shuffle_ps(rg23, ba23, 0x88);
+    __m128 v3 = _mm_shuffle_ps(rg23, ba23, 0xEE);
+    _mm_storeu_ps(((float*) dst) +  0, v0);
+    _mm_storeu_ps(((float*) dst) +  4, v1);
+    _mm_storeu_ps(((float*) dst) +  8, v2);
+    _mm_storeu_ps(((float*) dst) + 12, v3);
+}
+
 #endif//SkNx_sse_DEFINED