fillUnsafeSimd

src/pixie/internal.nim

@@ -80,11 +80,7 @@ proc fillUnsafe*(
   ## Fills the image data with the color starting at index start and
   ## continuing for len indices.
   when allowSimd and compiles(fillUnsafeSimd):
-    fillUnsafeSimd(
-      cast[ptr UncheckedArray[ColorRGBX]](data[start].addr),
-      len,
-      color
-    )
+    fillUnsafeSimd(data, start, len, color)
     return
 
   let rgbx = color.asRgbx()

src/pixie/runtimechecked/avx.nim

@@ -7,25 +7,30 @@ when defined(release):
   {.push checks: off.}
 
 proc fillUnsafeAvx*(
-  data: ptr UncheckedArray[ColorRGBX],
-  len: int,
+  data: var seq[ColorRGBX],
+  start, len: int,
   color: SomeColor
 ) =
   let rgbx = color.asRgbx()
 
-  var i: int
-  while i < len and (cast[uint](data[i].addr) and 31) != 0: # Align to 32 bytes
+  var
+    i = start
+    p = cast[uint](data[i].addr)
+  # Align to 32 bytes
+  while i < (start + len) and (p and 31) != 0:
     data[i] = rgbx
     inc i
+    p += 4
 
   let
-    iterations = (len - i) div 8
+    iterations = (start + len - i) div 8
     colorVec = mm256_set1_epi32(cast[int32](rgbx))
   for _ in 0 ..< iterations:
-    mm256_store_si256(data[i].addr, colorVec)
-    i += 8
-  # Fill whatever is left the slow way
-  for i in i ..< len:
+    mm256_store_si256(cast[pointer](p), colorVec)
+    p += 32
+  i += 8 * iterations
+
+  for i in i ..< start + len:
     data[i] = rgbx
 
 when defined(release):

src/pixie/simd.nim

@@ -31,30 +31,36 @@ when defined(amd64):
     result = mm_unpacklo_epi8(mm_setzero_si128(), result)
 
   proc fillUnsafeSimd*(
-    data: ptr UncheckedArray[ColorRGBX],
-    len: int,
+    data: var seq[ColorRGBX],
+    start, len: int,
     color: SomeColor
   ) =
-    if cpuHasAvx and len >= 64:
-      fillUnsafeAvx(data, len, color)
-    else:
-      let rgbx = color.asRgbx()
+    if cpuHasAvx:
+      fillUnsafeAvx(data, start, len, color)
+      return
 
-      var i: int
-      while i < len and (cast[uint](data[i].addr) and 15) != 0: # Align to 16 bytes
-        data[i] = rgbx
-        inc i
+    let rgbx = color.asRgbx()
 
-      let
-        colorVec = mm_set1_epi32(cast[int32](rgbx))
-        iterations = (len - i) div 8
-      for _ in 0 ..< iterations:
-        mm_store_si128(data[i].addr, colorVec)
-        mm_store_si128(data[i + 4].addr, colorVec)
-        i += 8
+    var
+      i = start
+      p = cast[uint](data[i].addr)
+    # Align to 16 bytes
+    while i < (start + len) and (p and 15) != 0:
+      data[i] = rgbx
+      inc i
+      p += 4
 
-      for i in i ..< len:
-        data[i] = rgbx
+    let
+      colorVec = mm_set1_epi32(cast[int32](rgbx))
+      iterations = (start + len - i) div 8
+    for _ in 0 ..< iterations:
+      mm_store_si128(cast[pointer](p), colorVec)
+      mm_store_si128(cast[pointer](p + 16), colorVec)
+      p += 32
+    i += iterations * 8
+
+    for i in i ..< start + len:
+      data[i] = rgbx
 
   proc isOneColorSimd*(data: ptr UncheckedArray[ColorRGBX], len: int): bool =
     if cpuHasAvx2: