no simd for tcc

src/pixie/blends.nim

@@ -1,8 +1,8 @@
 ## Blending modes.
 
-import chroma, common, math
+import chroma, common, internal, std/math
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 # See https://www.w3.org/TR/compositing-1/
@@ -274,7 +274,7 @@ proc blendSoftLight(backdrop, source: ColorRGBX): ColorRGBX =
     source = source.rgba()
 
   var rgba: ColorRGBA
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let
       vb = mm_setr_ps(
         backdrop.r.float32,
@@ -479,7 +479,7 @@ proc masker*(blendMode: BlendMode): Masker {.raises: [PixieError].} =
   else:
     raise newException(PixieError, "No masker for " & $blendMode)
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   type
     BlenderSimd* = proc(blackdrop, source: M128i): M128i {.gcsafe, raises: [].}
       ## Function signature returned by blenderSimd.

src/pixie/fileformats/jpeg.nim

@@ -1,7 +1,7 @@
-import chroma, flatty/binny, pixie/common, pixie/images, pixie/masks, sequtils,
-    std/decls, strutils
+import chroma, flatty/binny, pixie/common, pixie/images, pixie/internal,
+    pixie/masks, sequtils, std/decls, strutils
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 # This JPEG decoder is loosely based on stb_image which is public domain.
@@ -881,7 +881,7 @@ proc quantizationAndIDCTPass(state: var DecoderState) =
       for row in 0 ..< w:
         var data {.byaddr.} = state.components[comp].blocks[row][column]
 
-        when defined(amd64) and not defined(pixieNoSimd):
+        when defined(amd64) and allowSimd:
           for i in 0 ..< 8: # 8 per pass
             var q = mm_loadu_si128(state.quantizationTables[qTableId][i * 8].addr)
             q = mm_unpacklo_epi8(q, mm_setzero_si128())

src/pixie/images.nim

@@ -1,6 +1,6 @@
 import blends, bumpy, chroma, common, masks, pixie/internal, vmath
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 const h = 0.5.float32
@@ -29,7 +29,7 @@ proc newImage*(width, height: int): Image {.raises: [PixieError].} =
 proc newImage*(mask: Mask): Image {.raises: [PixieError].} =
   result = newImage(mask.width, mask.height)
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     for _ in 0 ..< mask.data.len div 16:
       var alphas = mm_loadu_si128(mask.data[i].addr)
       for j in 0 ..< 4:
@@ -106,7 +106,7 @@ proc isOneColor*(image: Image): bool {.raises: [].} =
   let color = image.data[0]
 
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let colorVec = mm_set1_epi32(cast[int32](color))
     for _ in 0 ..< image.data.len div 8:
       let
@@ -127,7 +127,7 @@ proc isTransparent*(image: Image): bool {.raises: [].} =
   result = true
 
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let vecZero = mm_setzero_si128()
     for _ in 0 ..< image.data.len div 16:
       let
@@ -254,7 +254,7 @@ proc minifyBy2*(image: Image, power = 1): Image {.raises: [PixieError].} =
     )
     for y in 0 ..< resultEvenHeight:
       var x: int
-      when defined(amd64) and not defined(pixieNoSimd):
+      when defined(amd64) and allowSimd:
         let
           oddMask = mm_set1_epi16(cast[int16](0xff00))
           first32 = cast[M128i]([uint32.high, 0, 0, 0])
@@ -348,7 +348,7 @@ proc magnifyBy2*(image: Image, power = 1): Image {.raises: [PixieError].} =
   for y in 0 ..< image.height:
     # Write one row of pixels duplicated by scale
     var x: int
-    when defined(amd64) and not defined(pixieNoSimd):
+    when defined(amd64) and allowSimd:
       if scale == 2:
         while x <= image.width - 4:
           let
@@ -391,7 +391,7 @@ proc applyOpacity*(target: Image | Mask, opacity: float32) {.raises: [].} =
     return
 
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     when type(target) is Image:
       let byteLen = target.data.len * 4
     else:
@@ -447,7 +447,7 @@ proc applyOpacity*(target: Image | Mask, opacity: float32) {.raises: [].} =
 proc invert*(target: Image) {.raises: [].} =
   ## Inverts all of the colors and alpha.
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let vec255 = mm_set1_epi8(cast[int8](255))
     let byteLen = target.data.len * 4
     for _ in 0 ..< byteLen div 16:
@@ -536,7 +536,7 @@ proc newMask*(image: Image): Mask {.raises: [PixieError].} =
   result = newMask(image.width, image.height)
 
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     for _ in 0 ..< image.data.len div 16:
       let
         a = mm_loadu_si128(image.data[i + 0].addr)
@@ -798,7 +798,7 @@ proc drawUber(
             )
             continue
 
-        when defined(amd64) and not defined(pixieNoSimd):
+        when defined(amd64) and allowSimd:
           case blendMode:
           of OverwriteBlend:
             for _ in 0 ..< (xStop - xStart) div 16:

src/pixie/internal.nim

@@ -1,6 +1,8 @@
 import chroma, system/memory, vmath
 
-when defined(amd64) and not defined(pixieNoSimd):
+const allowSimd* = not defined(pixieNoSimd) and not defined(tcc)
+
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 template currentExceptionAsPixieError*(): untyped =
@@ -59,7 +61,7 @@ proc fillUnsafe*(
     nimSetMem(data[start].addr, rgbx.r.cint, len * 4)
   else:
     var i = start
-    when defined(amd64) and not defined(pixieNoSimd):
+    when defined(amd64) and allowSimd:
       # When supported, SIMD fill until we run out of room
       let colorVec = mm_set1_epi32(cast[int32](rgbx))
       for _ in 0 ..< len div 8:
@@ -93,7 +95,7 @@ proc toStraightAlpha*(data: var seq[ColorRGBA | ColorRGBX]) {.raises: [].} =
 proc toPremultipliedAlpha*(data: var seq[ColorRGBA | ColorRGBX]) {.raises: [].} =
   ## Converts an image to premultiplied alpha from straight alpha.
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     # When supported, SIMD convert as much as possible
     let
       alphaMask = mm_set1_epi32(cast[int32](0xff000000))
@@ -140,7 +142,7 @@ proc isOpaque*(data: var seq[ColorRGBX], start, len: int): bool =
   result = true
 
   var i = start
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let
       vec255 = mm_set1_epi32(cast[int32](uint32.high))
       colorMask = mm_set1_epi32(cast[int32]([255.uint8, 255, 255, 0]))
@@ -161,7 +163,7 @@ proc isOpaque*(data: var seq[ColorRGBX], start, len: int): bool =
     if data[j].a != 255:
       return false
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   proc packAlphaValues*(v: M128i): M128i {.inline, raises: [].} =
     ## Shuffle the alpha values for these 4 colors to the first 4 bytes
     let mask = mm_set1_epi32(cast[int32](0xff000000))

src/pixie/masks.nim

@@ -1,6 +1,6 @@
 import common, internal, vmath
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 type
@@ -87,7 +87,7 @@ proc minifyBy2*(mask: Mask, power = 1): Mask {.raises: [PixieError].} =
     result = newMask(src.width div 2, src.height div 2)
     for y in 0 ..< result.height:
       var x: int
-      when defined(amd64) and not defined(pixieNoSimd):
+      when defined(amd64) and allowSimd:
         let
           oddMask = mm_set1_epi16(cast[int16](0xff00))
           firstByte = cast[M128i](
@@ -169,7 +169,7 @@ proc magnifyBy2*(mask: Mask, power = 1): Mask {.raises: [PixieError].} =
   for y in 0 ..< mask.height:
     # Write one row of values duplicated by scale
     var x: int
-    when defined(amd64) and not defined(pixieNoSimd):
+    when defined(amd64) and allowSimd:
       if scale == 2:
         while x <= mask.width - 16:
           let
@@ -236,7 +236,7 @@ proc getValueSmooth*(mask: Mask, x, y: float32): uint8 {.raises: [].} =
 proc invert*(mask: Mask) {.raises: [].} =
   ## Inverts all of the values - creates a negative of the mask.
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let vec255 = mm_set1_epi8(cast[int8](255))
     let byteLen = mask.data.len
     for _ in 0 ..< byteLen div 16:
@@ -312,7 +312,7 @@ proc spread*(mask: Mask, spread: float32) {.raises: [PixieError].} =
 proc ceil*(mask: Mask) {.raises: [].} =
   ## A value of 0 stays 0. Anything else turns into 255.
   var i: int
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     let
       zeroVec = mm_setzero_si128()
       vec255 = mm_set1_epi32(cast[int32](uint32.high))

src/pixie/paints.nim

@@ -1,6 +1,6 @@
-import chroma, common, images, vmath
+import chroma, common, images, internal, vmath
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 type
@@ -122,7 +122,7 @@ proc fillGradientLinear(image: Image, paint: Paint) =
   if at.y == to.y: # Horizontal gradient
     var x: int
     while x < image.width:
-      when defined(amd64) and not defined(pixieNoSimd):
+      when defined(amd64) and allowSimd:
         if x + 4 <= image.width:
           var colors: array[4, ColorRGBX]
           for i in 0 ..< 4:
@@ -153,7 +153,7 @@ proc fillGradientLinear(image: Image, paint: Paint) =
         t = toLineSpace(at, to, xy)
         rgbx = paint.gradientColor(t)
       var x: int
-      when defined(amd64) and not defined(pixieNoSimd):
+      when defined(amd64) and allowSimd:
         let colorVec = mm_set1_epi32(cast[int32](rgbx))
         for _ in 0 ..< image.width div 4:
           mm_storeu_si128(image.data[image.dataIndex(x, y)].addr, colorVec)

src/pixie/paths.nim

@@ -1,7 +1,7 @@
 import blends, bumpy, chroma, common, fenv, images, internal, masks, paints,
     strutils, vmath
 
-when defined(amd64) and not defined(pixieNoSimd):
+when defined(amd64) and allowSimd:
   import nimsimd/sse2
 
 type
@@ -1296,7 +1296,7 @@ proc computeCoverage(
         let fillLen = at.int - fillStart
         if fillLen > 0:
           var i = fillStart
-          when defined(amd64) and not defined(pixieNoSimd):
+          when defined(amd64) and allowSimd:
             let sampleCoverageVec = mm_set1_epi8(cast[int8](sampleCoverage))
             for _ in 0 ..< fillLen div 16:
               var coverageVec = mm_loadu_si128(coverages[i - startX].addr)
@@ -1326,7 +1326,7 @@ proc fillCoverage(
   blendMode: BlendMode
 ) =
   var x = startX
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     if blendMode.hasSimdBlender():
       # When supported, SIMD blend as much as possible
       let
@@ -1445,7 +1445,7 @@ proc fillCoverage(
   blendMode: BlendMode
 ) =
   var x = startX
-  when defined(amd64) and not defined(pixieNoSimd):
+  when defined(amd64) and allowSimd:
     if blendMode.hasSimdMasker():
       let
         maskerSimd = blendMode.maskerSimd()
@@ -1511,7 +1511,7 @@ proc fillHits(
       continue
 
     var x = fillStart
-    when defined(amd64) and not defined(pixieNoSimd):
+    when defined(amd64) and allowSimd:
       if blendMode.hasSimdBlender():
         # When supported, SIMD blend as much as possible
         let colorVec = mm_set1_epi32(cast[int32](rgbx))
@@ -1573,7 +1573,7 @@ proc fillHits(
       continue
 
     var x = fillStart
-    when defined(amd64) and not defined(pixieNoSimd):
+    when defined(amd64) and allowSimd:
       if blendMode.hasSimdMasker():
         let
           maskerSimd = blendMode.maskerSimd()