avx2 versions

src/pixie/simd/avx2.nim

@@ -1,4 +1,4 @@
-import avx, chroma, internal, nimsimd/avx2, pixie/common, vmath
+import avx, chroma, internal, nimsimd/avx2, pixie/blends, pixie/common, vmath
 
 when defined(gcc) or defined(clang):
   {.localPassc: "-mavx2".}
@@ -376,5 +376,92 @@ proc minifyBy2Avx2*(image: Image, power = 1): Image {.simd.} =
     # Set src as this result for if we do another power
     src = result
 
+proc blitLineNormalAvx2*(
+  a, b: ptr UncheckedArray[ColorRGBX], len: int
+) {.simd.} =
+  let
+    alphaMask = mm256_set1_epi32(cast[int32](0xff000000))
+    oddMask = mm256_set1_epi16(cast[int16](0xff00))
+    div255 = mm256_set1_epi16(cast[int16](0x8081))
+    vec255 = mm256_set1_epi8(255)
+    vecAlpha255 = mm256_set1_epi32(cast[int32]([0.uint8, 255, 0, 255]))
+
+  var i: int
+  while i < len - 8:
+    let
+      source = mm256_loadu_si256(b[i].addr)
+      eq255 = mm256_cmpeq_epi8(source, vec255)
+    if (mm256_movemask_epi8(eq255) and 0x88888888) == 0x88888888: # Opaque source
+      mm256_storeu_si256(a[i].addr, source)
+    else:
+      let backdrop = mm256_loadu_si256(a[i].addr)
+
+      var
+        sourceAlpha = mm256_and_si256(source, alphaMask)
+        backdropEven = mm256_slli_epi16(backdrop, 8)
+        backdropOdd = mm256_and_si256(backdrop, oddMask)
+
+      sourceAlpha = mm256_or_si256(sourceAlpha, mm256_srli_epi32(sourceAlpha, 16))
+
+      let multiplier = mm256_sub_epi32(vecAlpha255, sourceAlpha)
+
+      backdropEven = mm256_mulhi_epu16(backdropEven, multiplier)
+      backdropOdd = mm256_mulhi_epu16(backdropOdd, multiplier)
+      backdropEven = mm256_srli_epi16(mm256_mulhi_epu16(backdropEven, div255), 7)
+      backdropOdd = mm256_srli_epi16(mm256_mulhi_epu16(backdropOdd, div255), 7)
+
+      let added = mm256_add_epi8(
+        source,
+        mm256_or_si256(backdropEven, mm256_slli_epi16(backdropOdd, 8))
+      )
+
+      mm256_storeu_si256(a[i].addr, added)
+
+    i += 8
+
+  for i in i ..< len:
+    a[i] = blendNormal(a[i], b[i])
+
+proc blitLineMaskAvx2*(
+  a, b: ptr UncheckedArray[ColorRGBX], len: int
+) {.simd.} =
+  let
+    alphaMask = mm256_set1_epi32(cast[int32](0xff000000))
+    oddMask = mm256_set1_epi16(cast[int16](0xff00))
+    div255 = mm256_set1_epi16(cast[int16](0x8081))
+    vec255 = mm256_set1_epi8(255)
+
+  var i: int
+  while i < len - 8:
+    let
+      source = mm256_loadu_si256(b[i].addr)
+      eq255 = mm256_cmpeq_epi8(source, vec255)
+    if (mm256_movemask_epi8(eq255) and 0x88888888) == 0x88888888: # Opaque source
+      discard
+    else:
+      let backdrop = mm256_loadu_si256(a[i].addr)
+
+      var
+        sourceAlpha = mm256_and_si256(source, alphaMask)
+        backdropEven = mm256_slli_epi16(backdrop, 8)
+        backdropOdd = mm256_and_si256(backdrop, oddMask)
+
+      sourceAlpha = mm256_or_si256(sourceAlpha, mm256_srli_epi32(sourceAlpha, 16))
+
+      backdropEven = mm256_mulhi_epu16(backdropEven, sourceAlpha)
+      backdropOdd = mm256_mulhi_epu16(backdropOdd, sourceAlpha)
+      backdropEven = mm256_srli_epi16(mm256_mulhi_epu16(backdropEven, div255), 7)
+      backdropOdd = mm256_srli_epi16(mm256_mulhi_epu16(backdropOdd, div255), 7)
+
+      mm256_storeu_si256(
+        a[i].addr,
+        mm256_or_si256(backdropEven, mm256_slli_epi16(backdropOdd, 8))
+      )
+
+    i += 8
+
+  for i in i ..< len:
+    a[i] = blendMask(a[i], b[i])
+
 when defined(release):
   {.pop.}