Add Sk4f_RoundToInt

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2134753006
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review-Url: https://codereview.chromium.org/2134753006

4 files changed, 35 insertions, 9 deletions

diff --git a/src/core/SkNx.h b/src/core/SkNx.h
index 374a0fc57c..dec63f8d89 100644
--- a/src/core/SkNx.h
+++ b/src/core/SkNx.h
@@ -300,6 +300,15 @@ typedef SkNx<4,       int> Sk4i;
     #include "../opts/SkNx_sse.h"
 #elif !defined(SKNX_NO_SIMD) && defined(SK_ARM_HAS_NEON)
     #include "../opts/SkNx_neon.h"
+#else
+
+SI Sk4i Sk4f_round(const Sk4f& x) {
+    return { (int) lrintf (x[0]),
+             (int) lrintf (x[1]),
+             (int) lrintf (x[2]),
+             (int) lrintf (x[3]), };
+}
+
 #endif
 
 SI void Sk4f_ToBytes(uint8_t p[16], const Sk4f& a, const Sk4f& b, const Sk4f& c, const Sk4f& d) {
diff --git a/src/opts/SkColorXform_opts.h b/src/opts/SkColorXform_opts.h
index 30bf164cbe..72ef35381e 100644
--- a/src/opts/SkColorXform_opts.h
+++ b/src/opts/SkColorXform_opts.h
@@ -84,19 +84,19 @@ static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
                 dstGreens = clamp_0_to_255(dstGreens);
                 dstBlues  = clamp_0_to_255(dstBlues);
 
-                auto rgba = (SkNx_cast<int>(dstReds)        )
-                          | (SkNx_cast<int>(dstGreens) <<  8)
-                          | (SkNx_cast<int>(dstBlues)  << 16)
-                          | (Sk4i{          0xFF       << 24});
+                auto rgba = (Sk4f_round(dstReds)        )
+                          | (Sk4f_round(dstGreens) <<  8)
+                          | (Sk4f_round(dstBlues)  << 16)
+                          | (Sk4i{      0xFF       << 24});
                 rgba.store(dst);
             } else {
                 Sk4f scaledReds   = Sk4f::Min(Sk4f::Max(1023.0f * dstReds,   0.0f), 1023.0f);
                 Sk4f scaledGreens = Sk4f::Min(Sk4f::Max(1023.0f * dstGreens, 0.0f), 1023.0f);
                 Sk4f scaledBlues  = Sk4f::Min(Sk4f::Max(1023.0f * dstBlues,  0.0f), 1023.0f);
 
-                Sk4i indicesReds   = SkNx_cast<int>(scaledReds + 0.5f);
-                Sk4i indicesGreens = SkNx_cast<int>(scaledGreens + 0.5f);
-                Sk4i indicesBlues  = SkNx_cast<int>(scaledBlues + 0.5f);
+                Sk4i indicesReds   = Sk4f_round(scaledReds);
+                Sk4i indicesGreens = Sk4f_round(scaledGreens);
+                Sk4i indicesBlues  = Sk4f_round(scaledBlues);
 
                 dst[0] = dstTables[0][indicesReds  [0]]
                        | dstTables[1][indicesGreens[0]] <<  8
@@ -149,13 +149,13 @@ static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
             dstPixel = clamp_0_to_255(dstPixel);
 
             uint32_t rgba;
-            SkNx_cast<uint8_t>(dstPixel).store(&rgba);
+            SkNx_cast<uint8_t>(Sk4f_round(dstPixel)).store(&rgba);
             rgba |= 0xFF000000;
             *dst = rgba;
         } else {
             Sk4f scaledPixel = Sk4f::Min(Sk4f::Max(1023.0f * dstPixel, 0.0f), 1023.0f);
 
-            Sk4i indices = SkNx_cast<int>(scaledPixel + 0.5f);
+            Sk4i indices = Sk4f_round(scaledPixel);
 
             *dst = dstTables[0][indices[0]]
                  | dstTables[1][indices[1]] <<  8
diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index bb81cded18..6d4cade12e 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -450,4 +450,13 @@ template<> inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) {
     return vmovn_u16(vcombine_u16(src.fVec, src.fVec));
 }
 
+template<> inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) {
+    uint16x4_t _16 = vqmovun_s32(src.fVec);
+    return vqmovn_u16(vcombine_u16(_16, _16));
+}
+
+static inline Sk4i Sk4f_round(const Sk4f& x) {
+    return vcvtq_s32_f32((x + 0.5f).fVec);
+}
+
 #endif//SkNx_neon_DEFINED
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index 65d9873c5c..8952ff77f7 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -371,4 +371,12 @@ template<> /*static*/ inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src)
     return _mm_packus_epi16(src.fVec, src.fVec);
 }
 
+template<> inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) {
+    return _mm_packus_epi16(_mm_packus_epi16(src.fVec, src.fVec), src.fVec);
+}
+
+static inline Sk4i Sk4f_round(const Sk4f& x) {
+    return _mm_cvtps_epi32(x.fVec);
+}
+
 #endif//SkNx_sse_DEFINED