Add mulHi to SkNx

Add mulHi to base SkNx, and specialize implementations for Sk4u for
neon and sse.

Add casts for converting from uint8_t by 4 to uint32_t by 4.

Cq-Include-Trybots: skia.primary:Test-Debian9-Clang-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD
Change-Id: I29a32e2ad9812a47fff841ceca334e562362836f
Reviewed-on: https://skia-review.googlesource.com/57960
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Herb Derby <herb@google.com>

2 files changed, 25 insertions, 4 deletions

diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index b906a02c82..32be78f66b 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -497,6 +497,13 @@ public:
     AI static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_u32(a.fVec, b.fVec); }
     // TODO as needed
 
+    AI SkNx mulHi(const SkNx& m) const {
+        uint64x2_t hi = vmull_u32(vget_high_u32(fVec), vget_high_u32(m.fVec));
+        uint64x2_t lo = vmull_u32( vget_low_u32(fVec),  vget_low_u32(m.fVec));
+
+        return { vcombine_u32(vshrn_n_u64(lo,32), vshrn_n_u64(hi,32)) };
+    }
+
     AI SkNx thenElse(const SkNx& t, const SkNx& e) const {
         return vbslq_u32(fVec, t.fVec, e.fVec);
     }
@@ -529,9 +536,13 @@ template<> AI /*static*/ Sk4b SkNx_cast<uint8_t, float>(const Sk4f& src) {
     return vqmovn_u16(vcombine_u16(_16, _16));
 }
 
-template<> AI /*static*/ Sk4i SkNx_cast<int32_t, uint8_t>(const Sk4b& src) {
+template<> AI /*static*/ Sk4u SkNx_cast<uint32_t, uint8_t>(const Sk4b& src) {
     uint16x8_t _16 = vmovl_u8(src.fVec);
-    return vreinterpretq_s32_u32(vmovl_u16(vget_low_u16(_16)));
+    return vmovl_u16(vget_low_u16(_16));
+}
+
+template<> AI /*static*/ Sk4i SkNx_cast<int32_t, uint8_t>(const Sk4b& src) {
+    return vreinterpretq_s32_u32(SkNx_cast<uint32_t>(src).fVec);
 }
 
 template<> AI /*static*/ Sk4f SkNx_cast<float, uint8_t>(const Sk4b& src) {
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index 469aefb145..d4d4781e5a 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -287,10 +287,16 @@ public:
     #endif
     }
 
+    AI SkNx mulHi(SkNx m) const {
+        SkNx v20{_mm_mul_epu32(m.fVec, fVec)};
+        SkNx v31{_mm_mul_epu32(_mm_srli_si128(m.fVec, 4), _mm_srli_si128(fVec, 4))};
+
+        return SkNx{v20[1], v31[1], v20[3], v31[3]};
+    }
+
     __m128i fVec;
 };
 
-
 template <>
 class SkNx<4, uint16_t> {
 public:
@@ -568,7 +574,7 @@ template<> AI /*static*/ Sk4b SkNx_cast<uint8_t, float>(const Sk4f& src) {
 #endif
 }
 
-template<> AI /*static*/ Sk4i SkNx_cast<int32_t, uint8_t>(const Sk4b& src) {
+template<> AI /*static*/ Sk4u SkNx_cast<uint32_t, uint8_t>(const Sk4b& src) {
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     const int _ = ~0;
     return _mm_shuffle_epi8(src.fVec, _mm_setr_epi8(0,_,_,_, 1,_,_,_, 2,_,_,_, 3,_,_,_));
@@ -578,6 +584,10 @@ template<> AI /*static*/ Sk4i SkNx_cast<int32_t, uint8_t>(const Sk4b& src) {
 #endif
 }
 
+template<> AI /*static*/ Sk4i SkNx_cast<int32_t, uint8_t>(const Sk4b& src) {
+    return SkNx_cast<uint32_t>(src).fVec;
+}
+
 template<> AI /*static*/ Sk4f SkNx_cast<float, uint8_t>(const Sk4b& src) {
     return _mm_cvtepi32_ps(SkNx_cast<int32_t>(src).fVec);
 }